Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
8f18cf13 | 19 | #include <linux/kernel.h> |
065631f6 | 20 | #include <linux/bio.h> |
39279cc3 | 21 | #include <linux/buffer_head.h> |
f2eb0a24 | 22 | #include <linux/file.h> |
39279cc3 CM |
23 | #include <linux/fs.h> |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/highmem.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/init.h> | |
28 | #include <linux/string.h> | |
39279cc3 CM |
29 | #include <linux/backing-dev.h> |
30 | #include <linux/mpage.h> | |
31 | #include <linux/swap.h> | |
32 | #include <linux/writeback.h> | |
33 | #include <linux/statfs.h> | |
34 | #include <linux/compat.h> | |
9ebefb18 | 35 | #include <linux/bit_spinlock.h> |
5103e947 | 36 | #include <linux/xattr.h> |
33268eaf | 37 | #include <linux/posix_acl.h> |
d899e052 | 38 | #include <linux/falloc.h> |
5a0e3ad6 | 39 | #include <linux/slab.h> |
7a36ddec | 40 | #include <linux/ratelimit.h> |
22c44fe6 | 41 | #include <linux/mount.h> |
55e301fd | 42 | #include <linux/btrfs.h> |
53b381b3 | 43 | #include <linux/blkdev.h> |
f23b5a59 | 44 | #include <linux/posix_acl_xattr.h> |
e2e40f2c | 45 | #include <linux/uio.h> |
39279cc3 CM |
46 | #include "ctree.h" |
47 | #include "disk-io.h" | |
48 | #include "transaction.h" | |
49 | #include "btrfs_inode.h" | |
39279cc3 | 50 | #include "print-tree.h" |
e6dcd2dc | 51 | #include "ordered-data.h" |
95819c05 | 52 | #include "xattr.h" |
e02119d5 | 53 | #include "tree-log.h" |
4a54c8c1 | 54 | #include "volumes.h" |
c8b97818 | 55 | #include "compression.h" |
b4ce94de | 56 | #include "locking.h" |
dc89e982 | 57 | #include "free-space-cache.h" |
581bb050 | 58 | #include "inode-map.h" |
38c227d8 | 59 | #include "backref.h" |
f23b5a59 | 60 | #include "hash.h" |
63541927 | 61 | #include "props.h" |
31193213 | 62 | #include "qgroup.h" |
39279cc3 CM |
63 | |
64 | struct btrfs_iget_args { | |
90d3e592 | 65 | struct btrfs_key *location; |
39279cc3 CM |
66 | struct btrfs_root *root; |
67 | }; | |
68 | ||
f28a4928 FM |
69 | struct btrfs_dio_data { |
70 | u64 outstanding_extents; | |
71 | u64 reserve; | |
72 | u64 unsubmitted_oe_range_start; | |
73 | u64 unsubmitted_oe_range_end; | |
74 | }; | |
75 | ||
6e1d5dcc AD |
76 | static const struct inode_operations btrfs_dir_inode_operations; |
77 | static const struct inode_operations btrfs_symlink_inode_operations; | |
78 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
79 | static const struct inode_operations btrfs_special_inode_operations; | |
80 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
81 | static const struct address_space_operations btrfs_aops; |
82 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 83 | static const struct file_operations btrfs_dir_file_operations; |
20e5506b | 84 | static const struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
85 | |
86 | static struct kmem_cache *btrfs_inode_cachep; | |
87 | struct kmem_cache *btrfs_trans_handle_cachep; | |
88 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 89 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 90 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
91 | |
92 | #define S_SHIFT 12 | |
4d4ab6d6 | 93 | static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { |
39279cc3 CM |
94 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, |
95 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
96 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
97 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
98 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
99 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
100 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
101 | }; | |
102 | ||
3972f260 | 103 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
a41ad394 | 104 | static int btrfs_truncate(struct inode *inode); |
5fd02043 | 105 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
106 | static noinline int cow_file_range(struct inode *inode, |
107 | struct page *locked_page, | |
108 | u64 start, u64 end, int *page_started, | |
109 | unsigned long *nr_written, int unlock); | |
70c8a91c JB |
110 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
111 | u64 len, u64 orig_start, | |
112 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
113 | u64 orig_block_len, u64 ram_bytes, |
114 | int type); | |
7b128766 | 115 | |
48a3b636 | 116 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 117 | |
6a3891c5 JB |
118 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
119 | void btrfs_test_inode_set_ops(struct inode *inode) | |
120 | { | |
121 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
122 | } | |
123 | #endif | |
124 | ||
f34f57a3 | 125 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
126 | struct inode *inode, struct inode *dir, |
127 | const struct qstr *qstr) | |
0279b4cd JO |
128 | { |
129 | int err; | |
130 | ||
f34f57a3 | 131 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 132 | if (!err) |
2a7dba39 | 133 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
134 | return err; |
135 | } | |
136 | ||
c8b97818 CM |
137 | /* |
138 | * this does all the hard work for inserting an inline extent into | |
139 | * the btree. The caller should have done a btrfs_drop_extents so that | |
140 | * no overlapping inline items exist in the btree | |
141 | */ | |
40f76580 | 142 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
1acae57b | 143 | struct btrfs_path *path, int extent_inserted, |
c8b97818 CM |
144 | struct btrfs_root *root, struct inode *inode, |
145 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 146 | int compress_type, |
c8b97818 CM |
147 | struct page **compressed_pages) |
148 | { | |
c8b97818 CM |
149 | struct extent_buffer *leaf; |
150 | struct page *page = NULL; | |
151 | char *kaddr; | |
152 | unsigned long ptr; | |
153 | struct btrfs_file_extent_item *ei; | |
154 | int err = 0; | |
155 | int ret; | |
156 | size_t cur_size = size; | |
c8b97818 | 157 | unsigned long offset; |
c8b97818 | 158 | |
fe3f566c | 159 | if (compressed_size && compressed_pages) |
c8b97818 | 160 | cur_size = compressed_size; |
c8b97818 | 161 | |
1acae57b | 162 | inode_add_bytes(inode, size); |
c8b97818 | 163 | |
1acae57b FDBM |
164 | if (!extent_inserted) { |
165 | struct btrfs_key key; | |
166 | size_t datasize; | |
c8b97818 | 167 | |
1acae57b FDBM |
168 | key.objectid = btrfs_ino(inode); |
169 | key.offset = start; | |
962a298f | 170 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 171 | |
1acae57b FDBM |
172 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
173 | path->leave_spinning = 1; | |
174 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
175 | datasize); | |
176 | if (ret) { | |
177 | err = ret; | |
178 | goto fail; | |
179 | } | |
c8b97818 CM |
180 | } |
181 | leaf = path->nodes[0]; | |
182 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
183 | struct btrfs_file_extent_item); | |
184 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
185 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
186 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
187 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
188 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
189 | ptr = btrfs_file_extent_inline_start(ei); | |
190 | ||
261507a0 | 191 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
192 | struct page *cpage; |
193 | int i = 0; | |
d397712b | 194 | while (compressed_size > 0) { |
c8b97818 | 195 | cpage = compressed_pages[i]; |
5b050f04 | 196 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 197 | PAGE_SIZE); |
c8b97818 | 198 | |
7ac687d9 | 199 | kaddr = kmap_atomic(cpage); |
c8b97818 | 200 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 201 | kunmap_atomic(kaddr); |
c8b97818 CM |
202 | |
203 | i++; | |
204 | ptr += cur_size; | |
205 | compressed_size -= cur_size; | |
206 | } | |
207 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 208 | compress_type); |
c8b97818 CM |
209 | } else { |
210 | page = find_get_page(inode->i_mapping, | |
09cbfeaf | 211 | start >> PAGE_SHIFT); |
c8b97818 | 212 | btrfs_set_file_extent_compression(leaf, ei, 0); |
7ac687d9 | 213 | kaddr = kmap_atomic(page); |
09cbfeaf | 214 | offset = start & (PAGE_SIZE - 1); |
c8b97818 | 215 | write_extent_buffer(leaf, kaddr + offset, ptr, size); |
7ac687d9 | 216 | kunmap_atomic(kaddr); |
09cbfeaf | 217 | put_page(page); |
c8b97818 CM |
218 | } |
219 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 220 | btrfs_release_path(path); |
c8b97818 | 221 | |
c2167754 YZ |
222 | /* |
223 | * we're an inline extent, so nobody can | |
224 | * extend the file past i_size without locking | |
225 | * a page we already have locked. | |
226 | * | |
227 | * We must do any isize and inode updates | |
228 | * before we unlock the pages. Otherwise we | |
229 | * could end up racing with unlink. | |
230 | */ | |
c8b97818 | 231 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 232 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 233 | |
79787eaa | 234 | return ret; |
c8b97818 | 235 | fail: |
c8b97818 CM |
236 | return err; |
237 | } | |
238 | ||
239 | ||
240 | /* | |
241 | * conditionally insert an inline extent into the file. This | |
242 | * does the checks required to make sure the data is small enough | |
243 | * to fit as an inline extent. | |
244 | */ | |
00361589 JB |
245 | static noinline int cow_file_range_inline(struct btrfs_root *root, |
246 | struct inode *inode, u64 start, | |
247 | u64 end, size_t compressed_size, | |
248 | int compress_type, | |
249 | struct page **compressed_pages) | |
c8b97818 | 250 | { |
00361589 | 251 | struct btrfs_trans_handle *trans; |
c8b97818 CM |
252 | u64 isize = i_size_read(inode); |
253 | u64 actual_end = min(end + 1, isize); | |
254 | u64 inline_len = actual_end - start; | |
fda2832f | 255 | u64 aligned_end = ALIGN(end, root->sectorsize); |
c8b97818 CM |
256 | u64 data_len = inline_len; |
257 | int ret; | |
1acae57b FDBM |
258 | struct btrfs_path *path; |
259 | int extent_inserted = 0; | |
260 | u32 extent_item_size; | |
c8b97818 CM |
261 | |
262 | if (compressed_size) | |
263 | data_len = compressed_size; | |
264 | ||
265 | if (start > 0 || | |
0c29ba99 | 266 | actual_end > root->sectorsize || |
354877be | 267 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) || |
c8b97818 CM |
268 | (!compressed_size && |
269 | (actual_end & (root->sectorsize - 1)) == 0) || | |
270 | end + 1 < isize || | |
271 | data_len > root->fs_info->max_inline) { | |
272 | return 1; | |
273 | } | |
274 | ||
1acae57b FDBM |
275 | path = btrfs_alloc_path(); |
276 | if (!path) | |
277 | return -ENOMEM; | |
278 | ||
00361589 | 279 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
280 | if (IS_ERR(trans)) { |
281 | btrfs_free_path(path); | |
00361589 | 282 | return PTR_ERR(trans); |
1acae57b | 283 | } |
00361589 JB |
284 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
285 | ||
1acae57b FDBM |
286 | if (compressed_size && compressed_pages) |
287 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
288 | compressed_size); | |
289 | else | |
290 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
291 | inline_len); | |
292 | ||
293 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
294 | start, aligned_end, NULL, | |
295 | 1, 1, extent_item_size, &extent_inserted); | |
00361589 JB |
296 | if (ret) { |
297 | btrfs_abort_transaction(trans, root, ret); | |
298 | goto out; | |
299 | } | |
c8b97818 CM |
300 | |
301 | if (isize > actual_end) | |
302 | inline_len = min_t(u64, isize, actual_end); | |
1acae57b FDBM |
303 | ret = insert_inline_extent(trans, path, extent_inserted, |
304 | root, inode, start, | |
c8b97818 | 305 | inline_len, compressed_size, |
fe3f566c | 306 | compress_type, compressed_pages); |
2adcac1a | 307 | if (ret && ret != -ENOSPC) { |
79787eaa | 308 | btrfs_abort_transaction(trans, root, ret); |
00361589 | 309 | goto out; |
2adcac1a | 310 | } else if (ret == -ENOSPC) { |
00361589 JB |
311 | ret = 1; |
312 | goto out; | |
79787eaa | 313 | } |
2adcac1a | 314 | |
bdc20e67 | 315 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
0ca1f7ce | 316 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 317 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
00361589 | 318 | out: |
94ed938a QW |
319 | /* |
320 | * Don't forget to free the reserved space, as for inlined extent | |
321 | * it won't count as data extent, free them directly here. | |
322 | * And at reserve time, it's always aligned to page size, so | |
323 | * just free one page here. | |
324 | */ | |
09cbfeaf | 325 | btrfs_qgroup_free_data(inode, 0, PAGE_SIZE); |
1acae57b | 326 | btrfs_free_path(path); |
00361589 JB |
327 | btrfs_end_transaction(trans, root); |
328 | return ret; | |
c8b97818 CM |
329 | } |
330 | ||
771ed689 CM |
331 | struct async_extent { |
332 | u64 start; | |
333 | u64 ram_size; | |
334 | u64 compressed_size; | |
335 | struct page **pages; | |
336 | unsigned long nr_pages; | |
261507a0 | 337 | int compress_type; |
771ed689 CM |
338 | struct list_head list; |
339 | }; | |
340 | ||
341 | struct async_cow { | |
342 | struct inode *inode; | |
343 | struct btrfs_root *root; | |
344 | struct page *locked_page; | |
345 | u64 start; | |
346 | u64 end; | |
347 | struct list_head extents; | |
348 | struct btrfs_work work; | |
349 | }; | |
350 | ||
351 | static noinline int add_async_extent(struct async_cow *cow, | |
352 | u64 start, u64 ram_size, | |
353 | u64 compressed_size, | |
354 | struct page **pages, | |
261507a0 LZ |
355 | unsigned long nr_pages, |
356 | int compress_type) | |
771ed689 CM |
357 | { |
358 | struct async_extent *async_extent; | |
359 | ||
360 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 361 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
362 | async_extent->start = start; |
363 | async_extent->ram_size = ram_size; | |
364 | async_extent->compressed_size = compressed_size; | |
365 | async_extent->pages = pages; | |
366 | async_extent->nr_pages = nr_pages; | |
261507a0 | 367 | async_extent->compress_type = compress_type; |
771ed689 CM |
368 | list_add_tail(&async_extent->list, &cow->extents); |
369 | return 0; | |
370 | } | |
371 | ||
f79707b0 WS |
372 | static inline int inode_need_compress(struct inode *inode) |
373 | { | |
374 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
375 | ||
376 | /* force compress */ | |
377 | if (btrfs_test_opt(root, FORCE_COMPRESS)) | |
378 | return 1; | |
379 | /* bad compression ratios */ | |
380 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) | |
381 | return 0; | |
382 | if (btrfs_test_opt(root, COMPRESS) || | |
383 | BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS || | |
384 | BTRFS_I(inode)->force_compress) | |
385 | return 1; | |
386 | return 0; | |
387 | } | |
388 | ||
d352ac68 | 389 | /* |
771ed689 CM |
390 | * we create compressed extents in two phases. The first |
391 | * phase compresses a range of pages that have already been | |
392 | * locked (both pages and state bits are locked). | |
c8b97818 | 393 | * |
771ed689 CM |
394 | * This is done inside an ordered work queue, and the compression |
395 | * is spread across many cpus. The actual IO submission is step | |
396 | * two, and the ordered work queue takes care of making sure that | |
397 | * happens in the same order things were put onto the queue by | |
398 | * writepages and friends. | |
c8b97818 | 399 | * |
771ed689 CM |
400 | * If this code finds it can't get good compression, it puts an |
401 | * entry onto the work queue to write the uncompressed bytes. This | |
402 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
403 | * are written in the same order that the flusher thread sent them |
404 | * down. | |
d352ac68 | 405 | */ |
c44f649e | 406 | static noinline void compress_file_range(struct inode *inode, |
771ed689 CM |
407 | struct page *locked_page, |
408 | u64 start, u64 end, | |
409 | struct async_cow *async_cow, | |
410 | int *num_added) | |
b888db2b CM |
411 | { |
412 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
db94535d | 413 | u64 num_bytes; |
db94535d | 414 | u64 blocksize = root->sectorsize; |
c8b97818 | 415 | u64 actual_end; |
42dc7bab | 416 | u64 isize = i_size_read(inode); |
e6dcd2dc | 417 | int ret = 0; |
c8b97818 CM |
418 | struct page **pages = NULL; |
419 | unsigned long nr_pages; | |
420 | unsigned long nr_pages_ret = 0; | |
421 | unsigned long total_compressed = 0; | |
422 | unsigned long total_in = 0; | |
ee22184b BL |
423 | unsigned long max_compressed = SZ_128K; |
424 | unsigned long max_uncompressed = SZ_128K; | |
c8b97818 CM |
425 | int i; |
426 | int will_compress; | |
261507a0 | 427 | int compress_type = root->fs_info->compress_type; |
4adaa611 | 428 | int redirty = 0; |
b888db2b | 429 | |
4cb13e5d | 430 | /* if this is a small write inside eof, kick off a defrag */ |
ee22184b | 431 | if ((end - start + 1) < SZ_16K && |
4cb13e5d | 432 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) |
4cb5300b CM |
433 | btrfs_add_inode_defrag(NULL, inode); |
434 | ||
42dc7bab | 435 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
436 | again: |
437 | will_compress = 0; | |
09cbfeaf KS |
438 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
439 | nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_SIZE); | |
be20aa9d | 440 | |
f03d9301 CM |
441 | /* |
442 | * we don't want to send crud past the end of i_size through | |
443 | * compression, that's just a waste of CPU time. So, if the | |
444 | * end of the file is before the start of our current | |
445 | * requested range of bytes, we bail out to the uncompressed | |
446 | * cleanup code that can deal with all of this. | |
447 | * | |
448 | * It isn't really the fastest way to fix things, but this is a | |
449 | * very uncommon corner. | |
450 | */ | |
451 | if (actual_end <= start) | |
452 | goto cleanup_and_bail_uncompressed; | |
453 | ||
c8b97818 CM |
454 | total_compressed = actual_end - start; |
455 | ||
4bcbb332 SW |
456 | /* |
457 | * skip compression for a small file range(<=blocksize) that | |
01327610 | 458 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
459 | */ |
460 | if (total_compressed <= blocksize && | |
461 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
462 | goto cleanup_and_bail_uncompressed; | |
463 | ||
c8b97818 CM |
464 | /* we want to make sure that amount of ram required to uncompress |
465 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
466 | * of a compressed extent to 128k. This is a crucial number |
467 | * because it also controls how easily we can spread reads across | |
468 | * cpus for decompression. | |
469 | * | |
470 | * We also want to make sure the amount of IO required to do | |
471 | * a random read is reasonably small, so we limit the size of | |
472 | * a compressed extent to 128k. | |
c8b97818 CM |
473 | */ |
474 | total_compressed = min(total_compressed, max_uncompressed); | |
fda2832f | 475 | num_bytes = ALIGN(end - start + 1, blocksize); |
be20aa9d | 476 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
477 | total_in = 0; |
478 | ret = 0; | |
db94535d | 479 | |
771ed689 CM |
480 | /* |
481 | * we do compression for mount -o compress and when the | |
482 | * inode has not been flagged as nocompress. This flag can | |
483 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 484 | */ |
f79707b0 | 485 | if (inode_need_compress(inode)) { |
c8b97818 | 486 | WARN_ON(pages); |
31e818fe | 487 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
488 | if (!pages) { |
489 | /* just bail out to the uncompressed code */ | |
490 | goto cont; | |
491 | } | |
c8b97818 | 492 | |
261507a0 LZ |
493 | if (BTRFS_I(inode)->force_compress) |
494 | compress_type = BTRFS_I(inode)->force_compress; | |
495 | ||
4adaa611 CM |
496 | /* |
497 | * we need to call clear_page_dirty_for_io on each | |
498 | * page in the range. Otherwise applications with the file | |
499 | * mmap'd can wander in and change the page contents while | |
500 | * we are compressing them. | |
501 | * | |
502 | * If the compression fails for any reason, we set the pages | |
503 | * dirty again later on. | |
504 | */ | |
505 | extent_range_clear_dirty_for_io(inode, start, end); | |
506 | redirty = 1; | |
261507a0 LZ |
507 | ret = btrfs_compress_pages(compress_type, |
508 | inode->i_mapping, start, | |
509 | total_compressed, pages, | |
510 | nr_pages, &nr_pages_ret, | |
511 | &total_in, | |
512 | &total_compressed, | |
513 | max_compressed); | |
c8b97818 CM |
514 | |
515 | if (!ret) { | |
516 | unsigned long offset = total_compressed & | |
09cbfeaf | 517 | (PAGE_SIZE - 1); |
c8b97818 CM |
518 | struct page *page = pages[nr_pages_ret - 1]; |
519 | char *kaddr; | |
520 | ||
521 | /* zero the tail end of the last page, we might be | |
522 | * sending it down to disk | |
523 | */ | |
524 | if (offset) { | |
7ac687d9 | 525 | kaddr = kmap_atomic(page); |
c8b97818 | 526 | memset(kaddr + offset, 0, |
09cbfeaf | 527 | PAGE_SIZE - offset); |
7ac687d9 | 528 | kunmap_atomic(kaddr); |
c8b97818 CM |
529 | } |
530 | will_compress = 1; | |
531 | } | |
532 | } | |
560f7d75 | 533 | cont: |
c8b97818 CM |
534 | if (start == 0) { |
535 | /* lets try to make an inline extent */ | |
771ed689 | 536 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 537 | /* we didn't compress the entire range, try |
771ed689 | 538 | * to make an uncompressed inline extent. |
c8b97818 | 539 | */ |
00361589 JB |
540 | ret = cow_file_range_inline(root, inode, start, end, |
541 | 0, 0, NULL); | |
c8b97818 | 542 | } else { |
771ed689 | 543 | /* try making a compressed inline extent */ |
00361589 | 544 | ret = cow_file_range_inline(root, inode, start, end, |
fe3f566c LZ |
545 | total_compressed, |
546 | compress_type, pages); | |
c8b97818 | 547 | } |
79787eaa | 548 | if (ret <= 0) { |
151a41bc JB |
549 | unsigned long clear_flags = EXTENT_DELALLOC | |
550 | EXTENT_DEFRAG; | |
e6eb4314 FM |
551 | unsigned long page_error_op; |
552 | ||
151a41bc | 553 | clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0; |
e6eb4314 | 554 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 555 | |
771ed689 | 556 | /* |
79787eaa JM |
557 | * inline extent creation worked or returned error, |
558 | * we don't need to create any more async work items. | |
559 | * Unlock and free up our temp pages. | |
771ed689 | 560 | */ |
c2790a2e | 561 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
151a41bc | 562 | clear_flags, PAGE_UNLOCK | |
c2790a2e JB |
563 | PAGE_CLEAR_DIRTY | |
564 | PAGE_SET_WRITEBACK | | |
e6eb4314 | 565 | page_error_op | |
c2790a2e | 566 | PAGE_END_WRITEBACK); |
c8b97818 CM |
567 | goto free_pages_out; |
568 | } | |
569 | } | |
570 | ||
571 | if (will_compress) { | |
572 | /* | |
573 | * we aren't doing an inline extent round the compressed size | |
574 | * up to a block size boundary so the allocator does sane | |
575 | * things | |
576 | */ | |
fda2832f | 577 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
578 | |
579 | /* | |
580 | * one last check to make sure the compression is really a | |
581 | * win, compare the page count read with the blocks on disk | |
582 | */ | |
09cbfeaf | 583 | total_in = ALIGN(total_in, PAGE_SIZE); |
c8b97818 CM |
584 | if (total_compressed >= total_in) { |
585 | will_compress = 0; | |
586 | } else { | |
c8b97818 CM |
587 | num_bytes = total_in; |
588 | } | |
589 | } | |
590 | if (!will_compress && pages) { | |
591 | /* | |
592 | * the compression code ran but failed to make things smaller, | |
593 | * free any pages it allocated and our page pointer array | |
594 | */ | |
595 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 596 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 597 | put_page(pages[i]); |
c8b97818 CM |
598 | } |
599 | kfree(pages); | |
600 | pages = NULL; | |
601 | total_compressed = 0; | |
602 | nr_pages_ret = 0; | |
603 | ||
604 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
605 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
606 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 607 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 608 | } |
c8b97818 | 609 | } |
771ed689 CM |
610 | if (will_compress) { |
611 | *num_added += 1; | |
c8b97818 | 612 | |
771ed689 CM |
613 | /* the async work queues will take care of doing actual |
614 | * allocation on disk for these compressed pages, | |
615 | * and will submit them to the elevator. | |
616 | */ | |
617 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
618 | total_compressed, pages, nr_pages_ret, |
619 | compress_type); | |
179e29e4 | 620 | |
24ae6365 | 621 | if (start + num_bytes < end) { |
771ed689 CM |
622 | start += num_bytes; |
623 | pages = NULL; | |
624 | cond_resched(); | |
625 | goto again; | |
626 | } | |
627 | } else { | |
f03d9301 | 628 | cleanup_and_bail_uncompressed: |
771ed689 CM |
629 | /* |
630 | * No compression, but we still need to write the pages in | |
631 | * the file we've been given so far. redirty the locked | |
632 | * page if it corresponds to our extent and set things up | |
633 | * for the async work queue to run cow_file_range to do | |
634 | * the normal delalloc dance | |
635 | */ | |
636 | if (page_offset(locked_page) >= start && | |
637 | page_offset(locked_page) <= end) { | |
638 | __set_page_dirty_nobuffers(locked_page); | |
639 | /* unlocked later on in the async handlers */ | |
640 | } | |
4adaa611 CM |
641 | if (redirty) |
642 | extent_range_redirty_for_io(inode, start, end); | |
261507a0 LZ |
643 | add_async_extent(async_cow, start, end - start + 1, |
644 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
645 | *num_added += 1; |
646 | } | |
3b951516 | 647 | |
c44f649e | 648 | return; |
771ed689 CM |
649 | |
650 | free_pages_out: | |
651 | for (i = 0; i < nr_pages_ret; i++) { | |
652 | WARN_ON(pages[i]->mapping); | |
09cbfeaf | 653 | put_page(pages[i]); |
771ed689 | 654 | } |
d397712b | 655 | kfree(pages); |
771ed689 | 656 | } |
771ed689 | 657 | |
40ae837b FM |
658 | static void free_async_extent_pages(struct async_extent *async_extent) |
659 | { | |
660 | int i; | |
661 | ||
662 | if (!async_extent->pages) | |
663 | return; | |
664 | ||
665 | for (i = 0; i < async_extent->nr_pages; i++) { | |
666 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 667 | put_page(async_extent->pages[i]); |
40ae837b FM |
668 | } |
669 | kfree(async_extent->pages); | |
670 | async_extent->nr_pages = 0; | |
671 | async_extent->pages = NULL; | |
771ed689 CM |
672 | } |
673 | ||
674 | /* | |
675 | * phase two of compressed writeback. This is the ordered portion | |
676 | * of the code, which only gets called in the order the work was | |
677 | * queued. We walk all the async extents created by compress_file_range | |
678 | * and send them down to the disk. | |
679 | */ | |
dec8f175 | 680 | static noinline void submit_compressed_extents(struct inode *inode, |
771ed689 CM |
681 | struct async_cow *async_cow) |
682 | { | |
683 | struct async_extent *async_extent; | |
684 | u64 alloc_hint = 0; | |
771ed689 CM |
685 | struct btrfs_key ins; |
686 | struct extent_map *em; | |
687 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
688 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
689 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 690 | int ret = 0; |
771ed689 | 691 | |
3e04e7f1 | 692 | again: |
d397712b | 693 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
694 | async_extent = list_entry(async_cow->extents.next, |
695 | struct async_extent, list); | |
696 | list_del(&async_extent->list); | |
c8b97818 | 697 | |
771ed689 CM |
698 | io_tree = &BTRFS_I(inode)->io_tree; |
699 | ||
f5a84ee3 | 700 | retry: |
771ed689 CM |
701 | /* did the compression code fall back to uncompressed IO? */ |
702 | if (!async_extent->pages) { | |
703 | int page_started = 0; | |
704 | unsigned long nr_written = 0; | |
705 | ||
706 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 707 | async_extent->start + |
d0082371 | 708 | async_extent->ram_size - 1); |
771ed689 CM |
709 | |
710 | /* allocate blocks */ | |
f5a84ee3 JB |
711 | ret = cow_file_range(inode, async_cow->locked_page, |
712 | async_extent->start, | |
713 | async_extent->start + | |
714 | async_extent->ram_size - 1, | |
715 | &page_started, &nr_written, 0); | |
771ed689 | 716 | |
79787eaa JM |
717 | /* JDM XXX */ |
718 | ||
771ed689 CM |
719 | /* |
720 | * if page_started, cow_file_range inserted an | |
721 | * inline extent and took care of all the unlocking | |
722 | * and IO for us. Otherwise, we need to submit | |
723 | * all those pages down to the drive. | |
724 | */ | |
f5a84ee3 | 725 | if (!page_started && !ret) |
771ed689 CM |
726 | extent_write_locked_range(io_tree, |
727 | inode, async_extent->start, | |
d397712b | 728 | async_extent->start + |
771ed689 CM |
729 | async_extent->ram_size - 1, |
730 | btrfs_get_extent, | |
731 | WB_SYNC_ALL); | |
3e04e7f1 JB |
732 | else if (ret) |
733 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
734 | kfree(async_extent); |
735 | cond_resched(); | |
736 | continue; | |
737 | } | |
738 | ||
739 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 740 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 741 | |
00361589 | 742 | ret = btrfs_reserve_extent(root, |
771ed689 CM |
743 | async_extent->compressed_size, |
744 | async_extent->compressed_size, | |
e570fd27 | 745 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 746 | if (ret) { |
40ae837b | 747 | free_async_extent_pages(async_extent); |
3e04e7f1 | 748 | |
fdf8e2ea JB |
749 | if (ret == -ENOSPC) { |
750 | unlock_extent(io_tree, async_extent->start, | |
751 | async_extent->start + | |
752 | async_extent->ram_size - 1); | |
ce62003f LB |
753 | |
754 | /* | |
755 | * we need to redirty the pages if we decide to | |
756 | * fallback to uncompressed IO, otherwise we | |
757 | * will not submit these pages down to lower | |
758 | * layers. | |
759 | */ | |
760 | extent_range_redirty_for_io(inode, | |
761 | async_extent->start, | |
762 | async_extent->start + | |
763 | async_extent->ram_size - 1); | |
764 | ||
79787eaa | 765 | goto retry; |
fdf8e2ea | 766 | } |
3e04e7f1 | 767 | goto out_free; |
f5a84ee3 | 768 | } |
c2167754 YZ |
769 | /* |
770 | * here we're doing allocation and writeback of the | |
771 | * compressed pages | |
772 | */ | |
773 | btrfs_drop_extent_cache(inode, async_extent->start, | |
774 | async_extent->start + | |
775 | async_extent->ram_size - 1, 0); | |
776 | ||
172ddd60 | 777 | em = alloc_extent_map(); |
b9aa55be LB |
778 | if (!em) { |
779 | ret = -ENOMEM; | |
3e04e7f1 | 780 | goto out_free_reserve; |
b9aa55be | 781 | } |
771ed689 CM |
782 | em->start = async_extent->start; |
783 | em->len = async_extent->ram_size; | |
445a6944 | 784 | em->orig_start = em->start; |
2ab28f32 JB |
785 | em->mod_start = em->start; |
786 | em->mod_len = em->len; | |
c8b97818 | 787 | |
771ed689 CM |
788 | em->block_start = ins.objectid; |
789 | em->block_len = ins.offset; | |
b4939680 | 790 | em->orig_block_len = ins.offset; |
cc95bef6 | 791 | em->ram_bytes = async_extent->ram_size; |
771ed689 | 792 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
261507a0 | 793 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
794 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
795 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
70c8a91c | 796 | em->generation = -1; |
771ed689 | 797 | |
d397712b | 798 | while (1) { |
890871be | 799 | write_lock(&em_tree->lock); |
09a2a8f9 | 800 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 801 | write_unlock(&em_tree->lock); |
771ed689 CM |
802 | if (ret != -EEXIST) { |
803 | free_extent_map(em); | |
804 | break; | |
805 | } | |
806 | btrfs_drop_extent_cache(inode, async_extent->start, | |
807 | async_extent->start + | |
808 | async_extent->ram_size - 1, 0); | |
809 | } | |
810 | ||
3e04e7f1 JB |
811 | if (ret) |
812 | goto out_free_reserve; | |
813 | ||
261507a0 LZ |
814 | ret = btrfs_add_ordered_extent_compress(inode, |
815 | async_extent->start, | |
816 | ins.objectid, | |
817 | async_extent->ram_size, | |
818 | ins.offset, | |
819 | BTRFS_ORDERED_COMPRESSED, | |
820 | async_extent->compress_type); | |
d9f85963 FM |
821 | if (ret) { |
822 | btrfs_drop_extent_cache(inode, async_extent->start, | |
823 | async_extent->start + | |
824 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 825 | goto out_free_reserve; |
d9f85963 | 826 | } |
9cfa3e34 | 827 | btrfs_dec_block_group_reservations(root->fs_info, ins.objectid); |
771ed689 | 828 | |
771ed689 CM |
829 | /* |
830 | * clear dirty, set writeback and unlock the pages. | |
831 | */ | |
c2790a2e | 832 | extent_clear_unlock_delalloc(inode, async_extent->start, |
a791e35e CM |
833 | async_extent->start + |
834 | async_extent->ram_size - 1, | |
151a41bc JB |
835 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
836 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 837 | PAGE_SET_WRITEBACK); |
771ed689 | 838 | ret = btrfs_submit_compressed_write(inode, |
d397712b CM |
839 | async_extent->start, |
840 | async_extent->ram_size, | |
841 | ins.objectid, | |
842 | ins.offset, async_extent->pages, | |
843 | async_extent->nr_pages); | |
fce2a4e6 FM |
844 | if (ret) { |
845 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
846 | struct page *p = async_extent->pages[0]; | |
847 | const u64 start = async_extent->start; | |
848 | const u64 end = start + async_extent->ram_size - 1; | |
849 | ||
850 | p->mapping = inode->i_mapping; | |
851 | tree->ops->writepage_end_io_hook(p, start, end, | |
852 | NULL, 0); | |
853 | p->mapping = NULL; | |
854 | extent_clear_unlock_delalloc(inode, start, end, NULL, 0, | |
855 | PAGE_END_WRITEBACK | | |
856 | PAGE_SET_ERROR); | |
40ae837b | 857 | free_async_extent_pages(async_extent); |
fce2a4e6 | 858 | } |
771ed689 CM |
859 | alloc_hint = ins.objectid + ins.offset; |
860 | kfree(async_extent); | |
861 | cond_resched(); | |
862 | } | |
dec8f175 | 863 | return; |
3e04e7f1 | 864 | out_free_reserve: |
9cfa3e34 | 865 | btrfs_dec_block_group_reservations(root->fs_info, ins.objectid); |
e570fd27 | 866 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
79787eaa | 867 | out_free: |
c2790a2e | 868 | extent_clear_unlock_delalloc(inode, async_extent->start, |
3e04e7f1 JB |
869 | async_extent->start + |
870 | async_extent->ram_size - 1, | |
c2790a2e | 871 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
872 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
873 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
874 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
875 | PAGE_SET_ERROR); | |
40ae837b | 876 | free_async_extent_pages(async_extent); |
79787eaa | 877 | kfree(async_extent); |
3e04e7f1 | 878 | goto again; |
771ed689 CM |
879 | } |
880 | ||
4b46fce2 JB |
881 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
882 | u64 num_bytes) | |
883 | { | |
884 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
885 | struct extent_map *em; | |
886 | u64 alloc_hint = 0; | |
887 | ||
888 | read_lock(&em_tree->lock); | |
889 | em = search_extent_mapping(em_tree, start, num_bytes); | |
890 | if (em) { | |
891 | /* | |
892 | * if block start isn't an actual block number then find the | |
893 | * first block in this inode and use that as a hint. If that | |
894 | * block is also bogus then just don't worry about it. | |
895 | */ | |
896 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
897 | free_extent_map(em); | |
898 | em = search_extent_mapping(em_tree, 0, 0); | |
899 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
900 | alloc_hint = em->block_start; | |
901 | if (em) | |
902 | free_extent_map(em); | |
903 | } else { | |
904 | alloc_hint = em->block_start; | |
905 | free_extent_map(em); | |
906 | } | |
907 | } | |
908 | read_unlock(&em_tree->lock); | |
909 | ||
910 | return alloc_hint; | |
911 | } | |
912 | ||
771ed689 CM |
913 | /* |
914 | * when extent_io.c finds a delayed allocation range in the file, | |
915 | * the call backs end up in this code. The basic idea is to | |
916 | * allocate extents on disk for the range, and create ordered data structs | |
917 | * in ram to track those extents. | |
918 | * | |
919 | * locked_page is the page that writepage had locked already. We use | |
920 | * it to make sure we don't do extra locks or unlocks. | |
921 | * | |
922 | * *page_started is set to one if we unlock locked_page and do everything | |
923 | * required to start IO on it. It may be clean and already done with | |
924 | * IO when we return. | |
925 | */ | |
00361589 JB |
926 | static noinline int cow_file_range(struct inode *inode, |
927 | struct page *locked_page, | |
928 | u64 start, u64 end, int *page_started, | |
929 | unsigned long *nr_written, | |
930 | int unlock) | |
771ed689 | 931 | { |
00361589 | 932 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
933 | u64 alloc_hint = 0; |
934 | u64 num_bytes; | |
935 | unsigned long ram_size; | |
936 | u64 disk_num_bytes; | |
937 | u64 cur_alloc_size; | |
938 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
939 | struct btrfs_key ins; |
940 | struct extent_map *em; | |
941 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
942 | int ret = 0; | |
943 | ||
02ecd2c2 JB |
944 | if (btrfs_is_free_space_inode(inode)) { |
945 | WARN_ON_ONCE(1); | |
29bce2f3 JB |
946 | ret = -EINVAL; |
947 | goto out_unlock; | |
02ecd2c2 | 948 | } |
771ed689 | 949 | |
fda2832f | 950 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 CM |
951 | num_bytes = max(blocksize, num_bytes); |
952 | disk_num_bytes = num_bytes; | |
771ed689 | 953 | |
4cb5300b | 954 | /* if this is a small write inside eof, kick off defrag */ |
ee22184b | 955 | if (num_bytes < SZ_64K && |
4cb13e5d | 956 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) |
00361589 | 957 | btrfs_add_inode_defrag(NULL, inode); |
4cb5300b | 958 | |
771ed689 CM |
959 | if (start == 0) { |
960 | /* lets try to make an inline extent */ | |
00361589 JB |
961 | ret = cow_file_range_inline(root, inode, start, end, 0, 0, |
962 | NULL); | |
771ed689 | 963 | if (ret == 0) { |
c2790a2e JB |
964 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
965 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc | 966 | EXTENT_DEFRAG, PAGE_UNLOCK | |
c2790a2e JB |
967 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
968 | PAGE_END_WRITEBACK); | |
c2167754 | 969 | |
771ed689 | 970 | *nr_written = *nr_written + |
09cbfeaf | 971 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 972 | *page_started = 1; |
771ed689 | 973 | goto out; |
79787eaa | 974 | } else if (ret < 0) { |
79787eaa | 975 | goto out_unlock; |
771ed689 CM |
976 | } |
977 | } | |
978 | ||
979 | BUG_ON(disk_num_bytes > | |
6c41761f | 980 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 981 | |
4b46fce2 | 982 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
983 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
984 | ||
d397712b | 985 | while (disk_num_bytes > 0) { |
a791e35e CM |
986 | unsigned long op; |
987 | ||
287a0ab9 | 988 | cur_alloc_size = disk_num_bytes; |
00361589 | 989 | ret = btrfs_reserve_extent(root, cur_alloc_size, |
771ed689 | 990 | root->sectorsize, 0, alloc_hint, |
e570fd27 | 991 | &ins, 1, 1); |
00361589 | 992 | if (ret < 0) |
79787eaa | 993 | goto out_unlock; |
d397712b | 994 | |
172ddd60 | 995 | em = alloc_extent_map(); |
b9aa55be LB |
996 | if (!em) { |
997 | ret = -ENOMEM; | |
ace68bac | 998 | goto out_reserve; |
b9aa55be | 999 | } |
e6dcd2dc | 1000 | em->start = start; |
445a6944 | 1001 | em->orig_start = em->start; |
771ed689 CM |
1002 | ram_size = ins.offset; |
1003 | em->len = ins.offset; | |
2ab28f32 JB |
1004 | em->mod_start = em->start; |
1005 | em->mod_len = em->len; | |
c8b97818 | 1006 | |
e6dcd2dc | 1007 | em->block_start = ins.objectid; |
c8b97818 | 1008 | em->block_len = ins.offset; |
b4939680 | 1009 | em->orig_block_len = ins.offset; |
cc95bef6 | 1010 | em->ram_bytes = ram_size; |
e6dcd2dc | 1011 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 1012 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
70c8a91c | 1013 | em->generation = -1; |
c8b97818 | 1014 | |
d397712b | 1015 | while (1) { |
890871be | 1016 | write_lock(&em_tree->lock); |
09a2a8f9 | 1017 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1018 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
1019 | if (ret != -EEXIST) { |
1020 | free_extent_map(em); | |
1021 | break; | |
1022 | } | |
1023 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 1024 | start + ram_size - 1, 0); |
e6dcd2dc | 1025 | } |
ace68bac LB |
1026 | if (ret) |
1027 | goto out_reserve; | |
e6dcd2dc | 1028 | |
98d20f67 | 1029 | cur_alloc_size = ins.offset; |
e6dcd2dc | 1030 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 1031 | ram_size, cur_alloc_size, 0); |
ace68bac | 1032 | if (ret) |
d9f85963 | 1033 | goto out_drop_extent_cache; |
c8b97818 | 1034 | |
17d217fe YZ |
1035 | if (root->root_key.objectid == |
1036 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1037 | ret = btrfs_reloc_clone_csums(inode, start, | |
1038 | cur_alloc_size); | |
00361589 | 1039 | if (ret) |
d9f85963 | 1040 | goto out_drop_extent_cache; |
17d217fe YZ |
1041 | } |
1042 | ||
9cfa3e34 FM |
1043 | btrfs_dec_block_group_reservations(root->fs_info, ins.objectid); |
1044 | ||
d397712b | 1045 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 1046 | break; |
d397712b | 1047 | |
c8b97818 CM |
1048 | /* we're not doing compressed IO, don't unlock the first |
1049 | * page (which the caller expects to stay locked), don't | |
1050 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1051 | * |
1052 | * Do set the Private2 bit so we know this page was properly | |
1053 | * setup for writepage | |
c8b97818 | 1054 | */ |
c2790a2e JB |
1055 | op = unlock ? PAGE_UNLOCK : 0; |
1056 | op |= PAGE_SET_PRIVATE2; | |
a791e35e | 1057 | |
c2790a2e JB |
1058 | extent_clear_unlock_delalloc(inode, start, |
1059 | start + ram_size - 1, locked_page, | |
1060 | EXTENT_LOCKED | EXTENT_DELALLOC, | |
1061 | op); | |
c8b97818 | 1062 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
1063 | num_bytes -= cur_alloc_size; |
1064 | alloc_hint = ins.objectid + ins.offset; | |
1065 | start += cur_alloc_size; | |
b888db2b | 1066 | } |
79787eaa | 1067 | out: |
be20aa9d | 1068 | return ret; |
b7d5b0a8 | 1069 | |
d9f85963 FM |
1070 | out_drop_extent_cache: |
1071 | btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0); | |
ace68bac | 1072 | out_reserve: |
9cfa3e34 | 1073 | btrfs_dec_block_group_reservations(root->fs_info, ins.objectid); |
e570fd27 | 1074 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
79787eaa | 1075 | out_unlock: |
c2790a2e | 1076 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
151a41bc JB |
1077 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
1078 | EXTENT_DELALLOC | EXTENT_DEFRAG, | |
1079 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
1080 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 1081 | goto out; |
771ed689 | 1082 | } |
c8b97818 | 1083 | |
771ed689 CM |
1084 | /* |
1085 | * work queue call back to started compression on a file and pages | |
1086 | */ | |
1087 | static noinline void async_cow_start(struct btrfs_work *work) | |
1088 | { | |
1089 | struct async_cow *async_cow; | |
1090 | int num_added = 0; | |
1091 | async_cow = container_of(work, struct async_cow, work); | |
1092 | ||
1093 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1094 | async_cow->start, async_cow->end, async_cow, | |
1095 | &num_added); | |
8180ef88 | 1096 | if (num_added == 0) { |
cb77fcd8 | 1097 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1098 | async_cow->inode = NULL; |
8180ef88 | 1099 | } |
771ed689 CM |
1100 | } |
1101 | ||
1102 | /* | |
1103 | * work queue call back to submit previously compressed pages | |
1104 | */ | |
1105 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1106 | { | |
1107 | struct async_cow *async_cow; | |
1108 | struct btrfs_root *root; | |
1109 | unsigned long nr_pages; | |
1110 | ||
1111 | async_cow = container_of(work, struct async_cow, work); | |
1112 | ||
1113 | root = async_cow->root; | |
09cbfeaf KS |
1114 | nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >> |
1115 | PAGE_SHIFT; | |
771ed689 | 1116 | |
ee863954 DS |
1117 | /* |
1118 | * atomic_sub_return implies a barrier for waitqueue_active | |
1119 | */ | |
66657b31 | 1120 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
ee22184b | 1121 | 5 * SZ_1M && |
771ed689 CM |
1122 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1123 | wake_up(&root->fs_info->async_submit_wait); | |
1124 | ||
d397712b | 1125 | if (async_cow->inode) |
771ed689 | 1126 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1127 | } |
c8b97818 | 1128 | |
771ed689 CM |
1129 | static noinline void async_cow_free(struct btrfs_work *work) |
1130 | { | |
1131 | struct async_cow *async_cow; | |
1132 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1133 | if (async_cow->inode) |
cb77fcd8 | 1134 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1135 | kfree(async_cow); |
1136 | } | |
1137 | ||
1138 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1139 | u64 start, u64 end, int *page_started, | |
1140 | unsigned long *nr_written) | |
1141 | { | |
1142 | struct async_cow *async_cow; | |
1143 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1144 | unsigned long nr_pages; | |
1145 | u64 cur_end; | |
ee22184b | 1146 | int limit = 10 * SZ_1M; |
771ed689 | 1147 | |
a3429ab7 CM |
1148 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1149 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1150 | while (start < end) { |
771ed689 | 1151 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1152 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1153 | async_cow->inode = igrab(inode); |
771ed689 CM |
1154 | async_cow->root = root; |
1155 | async_cow->locked_page = locked_page; | |
1156 | async_cow->start = start; | |
1157 | ||
f79707b0 WS |
1158 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS && |
1159 | !btrfs_test_opt(root, FORCE_COMPRESS)) | |
771ed689 CM |
1160 | cur_end = end; |
1161 | else | |
ee22184b | 1162 | cur_end = min(end, start + SZ_512K - 1); |
771ed689 CM |
1163 | |
1164 | async_cow->end = cur_end; | |
1165 | INIT_LIST_HEAD(&async_cow->extents); | |
1166 | ||
9e0af237 LB |
1167 | btrfs_init_work(&async_cow->work, |
1168 | btrfs_delalloc_helper, | |
1169 | async_cow_start, async_cow_submit, | |
1170 | async_cow_free); | |
771ed689 | 1171 | |
09cbfeaf KS |
1172 | nr_pages = (cur_end - start + PAGE_SIZE) >> |
1173 | PAGE_SHIFT; | |
771ed689 CM |
1174 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); |
1175 | ||
afe3d242 QW |
1176 | btrfs_queue_work(root->fs_info->delalloc_workers, |
1177 | &async_cow->work); | |
771ed689 CM |
1178 | |
1179 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1180 | wait_event(root->fs_info->async_submit_wait, | |
1181 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1182 | limit)); | |
1183 | } | |
1184 | ||
d397712b | 1185 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1186 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1187 | wait_event(root->fs_info->async_submit_wait, | |
1188 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1189 | 0)); | |
1190 | } | |
1191 | ||
1192 | *nr_written += nr_pages; | |
1193 | start = cur_end + 1; | |
1194 | } | |
1195 | *page_started = 1; | |
1196 | return 0; | |
be20aa9d CM |
1197 | } |
1198 | ||
d397712b | 1199 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1200 | u64 bytenr, u64 num_bytes) |
1201 | { | |
1202 | int ret; | |
1203 | struct btrfs_ordered_sum *sums; | |
1204 | LIST_HEAD(list); | |
1205 | ||
07d400a6 | 1206 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1207 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1208 | if (ret == 0 && list_empty(&list)) |
1209 | return 0; | |
1210 | ||
1211 | while (!list_empty(&list)) { | |
1212 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1213 | list_del(&sums->list); | |
1214 | kfree(sums); | |
1215 | } | |
1216 | return 1; | |
1217 | } | |
1218 | ||
d352ac68 CM |
1219 | /* |
1220 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1221 | * of the extents that exist in the file, and COWs the file as required. | |
1222 | * | |
1223 | * If no cow copies or snapshots exist, we write directly to the existing | |
1224 | * blocks on disk | |
1225 | */ | |
7f366cfe CM |
1226 | static noinline int run_delalloc_nocow(struct inode *inode, |
1227 | struct page *locked_page, | |
771ed689 CM |
1228 | u64 start, u64 end, int *page_started, int force, |
1229 | unsigned long *nr_written) | |
be20aa9d | 1230 | { |
be20aa9d | 1231 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1232 | struct btrfs_trans_handle *trans; |
be20aa9d | 1233 | struct extent_buffer *leaf; |
be20aa9d | 1234 | struct btrfs_path *path; |
80ff3856 | 1235 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1236 | struct btrfs_key found_key; |
80ff3856 YZ |
1237 | u64 cow_start; |
1238 | u64 cur_offset; | |
1239 | u64 extent_end; | |
5d4f98a2 | 1240 | u64 extent_offset; |
80ff3856 YZ |
1241 | u64 disk_bytenr; |
1242 | u64 num_bytes; | |
b4939680 | 1243 | u64 disk_num_bytes; |
cc95bef6 | 1244 | u64 ram_bytes; |
80ff3856 | 1245 | int extent_type; |
79787eaa | 1246 | int ret, err; |
d899e052 | 1247 | int type; |
80ff3856 YZ |
1248 | int nocow; |
1249 | int check_prev = 1; | |
82d5902d | 1250 | bool nolock; |
33345d01 | 1251 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1252 | |
1253 | path = btrfs_alloc_path(); | |
17ca04af | 1254 | if (!path) { |
c2790a2e JB |
1255 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1256 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1257 | EXTENT_DO_ACCOUNTING | |
1258 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1259 | PAGE_CLEAR_DIRTY | |
1260 | PAGE_SET_WRITEBACK | | |
1261 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1262 | return -ENOMEM; |
17ca04af | 1263 | } |
82d5902d | 1264 | |
83eea1f1 | 1265 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1266 | |
1267 | if (nolock) | |
7a7eaa40 | 1268 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1269 | else |
7a7eaa40 | 1270 | trans = btrfs_join_transaction(root); |
ff5714cc | 1271 | |
79787eaa | 1272 | if (IS_ERR(trans)) { |
c2790a2e JB |
1273 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1274 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1275 | EXTENT_DO_ACCOUNTING | |
1276 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1277 | PAGE_CLEAR_DIRTY | |
1278 | PAGE_SET_WRITEBACK | | |
1279 | PAGE_END_WRITEBACK); | |
79787eaa JM |
1280 | btrfs_free_path(path); |
1281 | return PTR_ERR(trans); | |
1282 | } | |
1283 | ||
74b21075 | 1284 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1285 | |
80ff3856 YZ |
1286 | cow_start = (u64)-1; |
1287 | cur_offset = start; | |
1288 | while (1) { | |
33345d01 | 1289 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1290 | cur_offset, 0); |
d788a349 | 1291 | if (ret < 0) |
79787eaa | 1292 | goto error; |
80ff3856 YZ |
1293 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1294 | leaf = path->nodes[0]; | |
1295 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1296 | path->slots[0] - 1); | |
33345d01 | 1297 | if (found_key.objectid == ino && |
80ff3856 YZ |
1298 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1299 | path->slots[0]--; | |
1300 | } | |
1301 | check_prev = 0; | |
1302 | next_slot: | |
1303 | leaf = path->nodes[0]; | |
1304 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1305 | ret = btrfs_next_leaf(root, path); | |
d788a349 | 1306 | if (ret < 0) |
79787eaa | 1307 | goto error; |
80ff3856 YZ |
1308 | if (ret > 0) |
1309 | break; | |
1310 | leaf = path->nodes[0]; | |
1311 | } | |
be20aa9d | 1312 | |
80ff3856 YZ |
1313 | nocow = 0; |
1314 | disk_bytenr = 0; | |
17d217fe | 1315 | num_bytes = 0; |
80ff3856 YZ |
1316 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1317 | ||
1d512cb7 FM |
1318 | if (found_key.objectid > ino) |
1319 | break; | |
1320 | if (WARN_ON_ONCE(found_key.objectid < ino) || | |
1321 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1322 | path->slots[0]++; | |
1323 | goto next_slot; | |
1324 | } | |
1325 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || | |
80ff3856 YZ |
1326 | found_key.offset > end) |
1327 | break; | |
1328 | ||
1329 | if (found_key.offset > cur_offset) { | |
1330 | extent_end = found_key.offset; | |
e9061e21 | 1331 | extent_type = 0; |
80ff3856 YZ |
1332 | goto out_check; |
1333 | } | |
1334 | ||
1335 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1336 | struct btrfs_file_extent_item); | |
1337 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1338 | ||
cc95bef6 | 1339 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1340 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1341 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1342 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1343 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1344 | extent_end = found_key.offset + |
1345 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1346 | disk_num_bytes = |
1347 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1348 | if (extent_end <= start) { |
1349 | path->slots[0]++; | |
1350 | goto next_slot; | |
1351 | } | |
17d217fe YZ |
1352 | if (disk_bytenr == 0) |
1353 | goto out_check; | |
80ff3856 YZ |
1354 | if (btrfs_file_extent_compression(leaf, fi) || |
1355 | btrfs_file_extent_encryption(leaf, fi) || | |
1356 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1357 | goto out_check; | |
d899e052 YZ |
1358 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1359 | goto out_check; | |
d2fb3437 | 1360 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1361 | goto out_check; |
33345d01 | 1362 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1363 | found_key.offset - |
1364 | extent_offset, disk_bytenr)) | |
17d217fe | 1365 | goto out_check; |
5d4f98a2 | 1366 | disk_bytenr += extent_offset; |
17d217fe YZ |
1367 | disk_bytenr += cur_offset - found_key.offset; |
1368 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 WS |
1369 | /* |
1370 | * if there are pending snapshots for this root, | |
1371 | * we fall into common COW way. | |
1372 | */ | |
1373 | if (!nolock) { | |
9ea24bbe | 1374 | err = btrfs_start_write_no_snapshoting(root); |
e9894fd3 WS |
1375 | if (!err) |
1376 | goto out_check; | |
1377 | } | |
17d217fe YZ |
1378 | /* |
1379 | * force cow if csum exists in the range. | |
1380 | * this ensure that csum for a given extent are | |
1381 | * either valid or do not exist. | |
1382 | */ | |
1383 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1384 | goto out_check; | |
f78c436c FM |
1385 | if (!btrfs_inc_nocow_writers(root->fs_info, |
1386 | disk_bytenr)) | |
1387 | goto out_check; | |
80ff3856 YZ |
1388 | nocow = 1; |
1389 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1390 | extent_end = found_key.offset + | |
514ac8ad CM |
1391 | btrfs_file_extent_inline_len(leaf, |
1392 | path->slots[0], fi); | |
80ff3856 YZ |
1393 | extent_end = ALIGN(extent_end, root->sectorsize); |
1394 | } else { | |
1395 | BUG_ON(1); | |
1396 | } | |
1397 | out_check: | |
1398 | if (extent_end <= start) { | |
1399 | path->slots[0]++; | |
e9894fd3 | 1400 | if (!nolock && nocow) |
9ea24bbe | 1401 | btrfs_end_write_no_snapshoting(root); |
f78c436c FM |
1402 | if (nocow) |
1403 | btrfs_dec_nocow_writers(root->fs_info, | |
1404 | disk_bytenr); | |
80ff3856 YZ |
1405 | goto next_slot; |
1406 | } | |
1407 | if (!nocow) { | |
1408 | if (cow_start == (u64)-1) | |
1409 | cow_start = cur_offset; | |
1410 | cur_offset = extent_end; | |
1411 | if (cur_offset > end) | |
1412 | break; | |
1413 | path->slots[0]++; | |
1414 | goto next_slot; | |
7ea394f1 YZ |
1415 | } |
1416 | ||
b3b4aa74 | 1417 | btrfs_release_path(path); |
80ff3856 | 1418 | if (cow_start != (u64)-1) { |
00361589 JB |
1419 | ret = cow_file_range(inode, locked_page, |
1420 | cow_start, found_key.offset - 1, | |
1421 | page_started, nr_written, 1); | |
e9894fd3 WS |
1422 | if (ret) { |
1423 | if (!nolock && nocow) | |
9ea24bbe | 1424 | btrfs_end_write_no_snapshoting(root); |
f78c436c FM |
1425 | if (nocow) |
1426 | btrfs_dec_nocow_writers(root->fs_info, | |
1427 | disk_bytenr); | |
79787eaa | 1428 | goto error; |
e9894fd3 | 1429 | } |
80ff3856 | 1430 | cow_start = (u64)-1; |
7ea394f1 | 1431 | } |
80ff3856 | 1432 | |
d899e052 YZ |
1433 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1434 | struct extent_map *em; | |
1435 | struct extent_map_tree *em_tree; | |
1436 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1437 | em = alloc_extent_map(); |
79787eaa | 1438 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1439 | em->start = cur_offset; |
70c8a91c | 1440 | em->orig_start = found_key.offset - extent_offset; |
d899e052 YZ |
1441 | em->len = num_bytes; |
1442 | em->block_len = num_bytes; | |
1443 | em->block_start = disk_bytenr; | |
b4939680 | 1444 | em->orig_block_len = disk_num_bytes; |
cc95bef6 | 1445 | em->ram_bytes = ram_bytes; |
d899e052 | 1446 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
2ab28f32 JB |
1447 | em->mod_start = em->start; |
1448 | em->mod_len = em->len; | |
d899e052 | 1449 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
b11e234d | 1450 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
70c8a91c | 1451 | em->generation = -1; |
d899e052 | 1452 | while (1) { |
890871be | 1453 | write_lock(&em_tree->lock); |
09a2a8f9 | 1454 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1455 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1456 | if (ret != -EEXIST) { |
1457 | free_extent_map(em); | |
1458 | break; | |
1459 | } | |
1460 | btrfs_drop_extent_cache(inode, em->start, | |
1461 | em->start + em->len - 1, 0); | |
1462 | } | |
1463 | type = BTRFS_ORDERED_PREALLOC; | |
1464 | } else { | |
1465 | type = BTRFS_ORDERED_NOCOW; | |
1466 | } | |
80ff3856 YZ |
1467 | |
1468 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1469 | num_bytes, num_bytes, type); |
f78c436c FM |
1470 | if (nocow) |
1471 | btrfs_dec_nocow_writers(root->fs_info, disk_bytenr); | |
79787eaa | 1472 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1473 | |
efa56464 YZ |
1474 | if (root->root_key.objectid == |
1475 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1476 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1477 | num_bytes); | |
e9894fd3 WS |
1478 | if (ret) { |
1479 | if (!nolock && nocow) | |
9ea24bbe | 1480 | btrfs_end_write_no_snapshoting(root); |
79787eaa | 1481 | goto error; |
e9894fd3 | 1482 | } |
efa56464 YZ |
1483 | } |
1484 | ||
c2790a2e JB |
1485 | extent_clear_unlock_delalloc(inode, cur_offset, |
1486 | cur_offset + num_bytes - 1, | |
1487 | locked_page, EXTENT_LOCKED | | |
1488 | EXTENT_DELALLOC, PAGE_UNLOCK | | |
1489 | PAGE_SET_PRIVATE2); | |
e9894fd3 | 1490 | if (!nolock && nocow) |
9ea24bbe | 1491 | btrfs_end_write_no_snapshoting(root); |
80ff3856 YZ |
1492 | cur_offset = extent_end; |
1493 | if (cur_offset > end) | |
1494 | break; | |
be20aa9d | 1495 | } |
b3b4aa74 | 1496 | btrfs_release_path(path); |
80ff3856 | 1497 | |
17ca04af | 1498 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1499 | cow_start = cur_offset; |
17ca04af JB |
1500 | cur_offset = end; |
1501 | } | |
1502 | ||
80ff3856 | 1503 | if (cow_start != (u64)-1) { |
00361589 JB |
1504 | ret = cow_file_range(inode, locked_page, cow_start, end, |
1505 | page_started, nr_written, 1); | |
d788a349 | 1506 | if (ret) |
79787eaa | 1507 | goto error; |
80ff3856 YZ |
1508 | } |
1509 | ||
79787eaa | 1510 | error: |
a698d075 | 1511 | err = btrfs_end_transaction(trans, root); |
79787eaa JM |
1512 | if (!ret) |
1513 | ret = err; | |
1514 | ||
17ca04af | 1515 | if (ret && cur_offset < end) |
c2790a2e JB |
1516 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
1517 | locked_page, EXTENT_LOCKED | | |
151a41bc JB |
1518 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1519 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1520 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1521 | PAGE_SET_WRITEBACK | |
1522 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1523 | btrfs_free_path(path); |
79787eaa | 1524 | return ret; |
be20aa9d CM |
1525 | } |
1526 | ||
47059d93 WS |
1527 | static inline int need_force_cow(struct inode *inode, u64 start, u64 end) |
1528 | { | |
1529 | ||
1530 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
1531 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) | |
1532 | return 0; | |
1533 | ||
1534 | /* | |
1535 | * @defrag_bytes is a hint value, no spinlock held here, | |
1536 | * if is not zero, it means the file is defragging. | |
1537 | * Force cow if given extent needs to be defragged. | |
1538 | */ | |
1539 | if (BTRFS_I(inode)->defrag_bytes && | |
1540 | test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1541 | EXTENT_DEFRAG, 0, NULL)) | |
1542 | return 1; | |
1543 | ||
1544 | return 0; | |
1545 | } | |
1546 | ||
d352ac68 CM |
1547 | /* |
1548 | * extent_io.c call back to do delayed allocation processing | |
1549 | */ | |
c8b97818 | 1550 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1551 | u64 start, u64 end, int *page_started, |
1552 | unsigned long *nr_written) | |
be20aa9d | 1553 | { |
be20aa9d | 1554 | int ret; |
47059d93 | 1555 | int force_cow = need_force_cow(inode, start, end); |
a2135011 | 1556 | |
47059d93 | 1557 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
c8b97818 | 1558 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1559 | page_started, 1, nr_written); |
47059d93 | 1560 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
d899e052 | 1561 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1562 | page_started, 0, nr_written); |
7816030e | 1563 | } else if (!inode_need_compress(inode)) { |
7f366cfe CM |
1564 | ret = cow_file_range(inode, locked_page, start, end, |
1565 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1566 | } else { |
1567 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1568 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1569 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1570 | page_started, nr_written); |
7ddf5a42 | 1571 | } |
b888db2b CM |
1572 | return ret; |
1573 | } | |
1574 | ||
1bf85046 JM |
1575 | static void btrfs_split_extent_hook(struct inode *inode, |
1576 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1577 | { |
dcab6a3b JB |
1578 | u64 size; |
1579 | ||
0ca1f7ce | 1580 | /* not delalloc, ignore it */ |
9ed74f2d | 1581 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1582 | return; |
9ed74f2d | 1583 | |
dcab6a3b JB |
1584 | size = orig->end - orig->start + 1; |
1585 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
1586 | u64 num_extents; | |
1587 | u64 new_size; | |
1588 | ||
1589 | /* | |
ba117213 JB |
1590 | * See the explanation in btrfs_merge_extent_hook, the same |
1591 | * applies here, just in reverse. | |
dcab6a3b JB |
1592 | */ |
1593 | new_size = orig->end - split + 1; | |
ba117213 | 1594 | num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1, |
dcab6a3b | 1595 | BTRFS_MAX_EXTENT_SIZE); |
ba117213 JB |
1596 | new_size = split - orig->start; |
1597 | num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1598 | BTRFS_MAX_EXTENT_SIZE); | |
1599 | if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1600 | BTRFS_MAX_EXTENT_SIZE) >= num_extents) | |
dcab6a3b JB |
1601 | return; |
1602 | } | |
1603 | ||
9e0baf60 JB |
1604 | spin_lock(&BTRFS_I(inode)->lock); |
1605 | BTRFS_I(inode)->outstanding_extents++; | |
1606 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1607 | } |
1608 | ||
1609 | /* | |
1610 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1611 | * extents so we can keep track of new extents that are just merged onto old | |
1612 | * extents, such as when we are doing sequential writes, so we can properly | |
1613 | * account for the metadata space we'll need. | |
1614 | */ | |
1bf85046 JM |
1615 | static void btrfs_merge_extent_hook(struct inode *inode, |
1616 | struct extent_state *new, | |
1617 | struct extent_state *other) | |
9ed74f2d | 1618 | { |
dcab6a3b JB |
1619 | u64 new_size, old_size; |
1620 | u64 num_extents; | |
1621 | ||
9ed74f2d JB |
1622 | /* not delalloc, ignore it */ |
1623 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1624 | return; |
9ed74f2d | 1625 | |
8461a3de JB |
1626 | if (new->start > other->start) |
1627 | new_size = new->end - other->start + 1; | |
1628 | else | |
1629 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
1630 | |
1631 | /* we're not bigger than the max, unreserve the space and go */ | |
1632 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
1633 | spin_lock(&BTRFS_I(inode)->lock); | |
1634 | BTRFS_I(inode)->outstanding_extents--; | |
1635 | spin_unlock(&BTRFS_I(inode)->lock); | |
1636 | return; | |
1637 | } | |
1638 | ||
1639 | /* | |
ba117213 JB |
1640 | * We have to add up either side to figure out how many extents were |
1641 | * accounted for before we merged into one big extent. If the number of | |
1642 | * extents we accounted for is <= the amount we need for the new range | |
1643 | * then we can return, otherwise drop. Think of it like this | |
1644 | * | |
1645 | * [ 4k][MAX_SIZE] | |
1646 | * | |
1647 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
1648 | * need 2 outstanding extents, on one side we have 1 and the other side | |
1649 | * we have 1 so they are == and we can return. But in this case | |
1650 | * | |
1651 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
1652 | * | |
1653 | * Each range on their own accounts for 2 extents, but merged together | |
1654 | * they are only 3 extents worth of accounting, so we need to drop in | |
1655 | * this case. | |
dcab6a3b | 1656 | */ |
ba117213 | 1657 | old_size = other->end - other->start + 1; |
dcab6a3b JB |
1658 | num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1, |
1659 | BTRFS_MAX_EXTENT_SIZE); | |
ba117213 JB |
1660 | old_size = new->end - new->start + 1; |
1661 | num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1662 | BTRFS_MAX_EXTENT_SIZE); | |
1663 | ||
dcab6a3b | 1664 | if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1, |
ba117213 | 1665 | BTRFS_MAX_EXTENT_SIZE) >= num_extents) |
dcab6a3b JB |
1666 | return; |
1667 | ||
9e0baf60 JB |
1668 | spin_lock(&BTRFS_I(inode)->lock); |
1669 | BTRFS_I(inode)->outstanding_extents--; | |
1670 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1671 | } |
1672 | ||
eb73c1b7 MX |
1673 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1674 | struct inode *inode) | |
1675 | { | |
1676 | spin_lock(&root->delalloc_lock); | |
1677 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1678 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1679 | &root->delalloc_inodes); | |
1680 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1681 | &BTRFS_I(inode)->runtime_flags); | |
1682 | root->nr_delalloc_inodes++; | |
1683 | if (root->nr_delalloc_inodes == 1) { | |
1684 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1685 | BUG_ON(!list_empty(&root->delalloc_root)); | |
1686 | list_add_tail(&root->delalloc_root, | |
1687 | &root->fs_info->delalloc_roots); | |
1688 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1689 | } | |
1690 | } | |
1691 | spin_unlock(&root->delalloc_lock); | |
1692 | } | |
1693 | ||
1694 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1695 | struct inode *inode) | |
1696 | { | |
1697 | spin_lock(&root->delalloc_lock); | |
1698 | if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1699 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1700 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1701 | &BTRFS_I(inode)->runtime_flags); | |
1702 | root->nr_delalloc_inodes--; | |
1703 | if (!root->nr_delalloc_inodes) { | |
1704 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1705 | BUG_ON(list_empty(&root->delalloc_root)); | |
1706 | list_del_init(&root->delalloc_root); | |
1707 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1708 | } | |
1709 | } | |
1710 | spin_unlock(&root->delalloc_lock); | |
1711 | } | |
1712 | ||
d352ac68 CM |
1713 | /* |
1714 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1715 | * bytes in this file, and to maintain the list of inodes that | |
1716 | * have pending delalloc work to be done. | |
1717 | */ | |
1bf85046 | 1718 | static void btrfs_set_bit_hook(struct inode *inode, |
9ee49a04 | 1719 | struct extent_state *state, unsigned *bits) |
291d673e | 1720 | { |
9ed74f2d | 1721 | |
47059d93 WS |
1722 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1723 | WARN_ON(1); | |
75eff68e CM |
1724 | /* |
1725 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1726 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1727 | * bit, which is only set or cleared with irqs on |
1728 | */ | |
0ca1f7ce | 1729 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1730 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1731 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1732 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1733 | |
9e0baf60 | 1734 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1735 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1736 | } else { |
1737 | spin_lock(&BTRFS_I(inode)->lock); | |
1738 | BTRFS_I(inode)->outstanding_extents++; | |
1739 | spin_unlock(&BTRFS_I(inode)->lock); | |
1740 | } | |
287a0ab9 | 1741 | |
6a3891c5 JB |
1742 | /* For sanity tests */ |
1743 | if (btrfs_test_is_dummy_root(root)) | |
1744 | return; | |
1745 | ||
963d678b MX |
1746 | __percpu_counter_add(&root->fs_info->delalloc_bytes, len, |
1747 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1748 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1749 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
1750 | if (*bits & EXTENT_DEFRAG) |
1751 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 1752 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1753 | &BTRFS_I(inode)->runtime_flags)) |
1754 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1755 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1756 | } |
291d673e CM |
1757 | } |
1758 | ||
d352ac68 CM |
1759 | /* |
1760 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1761 | */ | |
1bf85046 | 1762 | static void btrfs_clear_bit_hook(struct inode *inode, |
41074888 | 1763 | struct extent_state *state, |
9ee49a04 | 1764 | unsigned *bits) |
291d673e | 1765 | { |
47059d93 | 1766 | u64 len = state->end + 1 - state->start; |
dcab6a3b JB |
1767 | u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1, |
1768 | BTRFS_MAX_EXTENT_SIZE); | |
47059d93 WS |
1769 | |
1770 | spin_lock(&BTRFS_I(inode)->lock); | |
1771 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) | |
1772 | BTRFS_I(inode)->defrag_bytes -= len; | |
1773 | spin_unlock(&BTRFS_I(inode)->lock); | |
1774 | ||
75eff68e CM |
1775 | /* |
1776 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1777 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1778 | * bit, which is only set or cleared with irqs on |
1779 | */ | |
0ca1f7ce | 1780 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1781 | struct btrfs_root *root = BTRFS_I(inode)->root; |
83eea1f1 | 1782 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1783 | |
9e0baf60 | 1784 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1785 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1786 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1787 | spin_lock(&BTRFS_I(inode)->lock); | |
dcab6a3b | 1788 | BTRFS_I(inode)->outstanding_extents -= num_extents; |
9e0baf60 JB |
1789 | spin_unlock(&BTRFS_I(inode)->lock); |
1790 | } | |
0ca1f7ce | 1791 | |
b6d08f06 JB |
1792 | /* |
1793 | * We don't reserve metadata space for space cache inodes so we | |
1794 | * don't need to call dellalloc_release_metadata if there is an | |
1795 | * error. | |
1796 | */ | |
1797 | if (*bits & EXTENT_DO_ACCOUNTING && | |
1798 | root != root->fs_info->tree_root) | |
0ca1f7ce YZ |
1799 | btrfs_delalloc_release_metadata(inode, len); |
1800 | ||
6a3891c5 JB |
1801 | /* For sanity tests. */ |
1802 | if (btrfs_test_is_dummy_root(root)) | |
1803 | return; | |
1804 | ||
0cb59c99 | 1805 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
7ee9e440 | 1806 | && do_list && !(state->state & EXTENT_NORESERVE)) |
51773bec QW |
1807 | btrfs_free_reserved_data_space_noquota(inode, |
1808 | state->start, len); | |
9ed74f2d | 1809 | |
963d678b MX |
1810 | __percpu_counter_add(&root->fs_info->delalloc_bytes, -len, |
1811 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1812 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1813 | BTRFS_I(inode)->delalloc_bytes -= len; |
0cb59c99 | 1814 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
df0af1a5 | 1815 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1816 | &BTRFS_I(inode)->runtime_flags)) |
1817 | btrfs_del_delalloc_inode(root, inode); | |
df0af1a5 | 1818 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1819 | } |
291d673e CM |
1820 | } |
1821 | ||
d352ac68 CM |
1822 | /* |
1823 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1824 | * we don't create bios that span stripes or chunks | |
1825 | */ | |
64a16701 | 1826 | int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset, |
c8b97818 CM |
1827 | size_t size, struct bio *bio, |
1828 | unsigned long bio_flags) | |
239b14b3 CM |
1829 | { |
1830 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
4f024f37 | 1831 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
1832 | u64 length = 0; |
1833 | u64 map_length; | |
239b14b3 CM |
1834 | int ret; |
1835 | ||
771ed689 CM |
1836 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1837 | return 0; | |
1838 | ||
4f024f37 | 1839 | length = bio->bi_iter.bi_size; |
239b14b3 | 1840 | map_length = length; |
64a16701 | 1841 | ret = btrfs_map_block(root->fs_info, rw, logical, |
f188591e | 1842 | &map_length, NULL, 0); |
3ec706c8 | 1843 | /* Will always return 0 with map_multi == NULL */ |
3444a972 | 1844 | BUG_ON(ret < 0); |
d397712b | 1845 | if (map_length < length + size) |
239b14b3 | 1846 | return 1; |
3444a972 | 1847 | return 0; |
239b14b3 CM |
1848 | } |
1849 | ||
d352ac68 CM |
1850 | /* |
1851 | * in order to insert checksums into the metadata in large chunks, | |
1852 | * we wait until bio submission time. All the pages in the bio are | |
1853 | * checksummed and sums are attached onto the ordered extent record. | |
1854 | * | |
1855 | * At IO completion time the cums attached on the ordered extent record | |
1856 | * are inserted into the btree | |
1857 | */ | |
d397712b CM |
1858 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1859 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1860 | unsigned long bio_flags, |
1861 | u64 bio_offset) | |
065631f6 | 1862 | { |
065631f6 | 1863 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1864 | int ret = 0; |
e015640f | 1865 | |
d20f7043 | 1866 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1867 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1868 | return 0; |
1869 | } | |
e015640f | 1870 | |
4a69a410 CM |
1871 | /* |
1872 | * in order to insert checksums into the metadata in large chunks, | |
1873 | * we wait until bio submission time. All the pages in the bio are | |
1874 | * checksummed and sums are attached onto the ordered extent record. | |
1875 | * | |
1876 | * At IO completion time the cums attached on the ordered extent record | |
1877 | * are inserted into the btree | |
1878 | */ | |
b2950863 | 1879 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1880 | int mirror_num, unsigned long bio_flags, |
1881 | u64 bio_offset) | |
4a69a410 CM |
1882 | { |
1883 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
61891923 SB |
1884 | int ret; |
1885 | ||
1886 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 1); | |
4246a0b6 CH |
1887 | if (ret) { |
1888 | bio->bi_error = ret; | |
1889 | bio_endio(bio); | |
1890 | } | |
61891923 | 1891 | return ret; |
44b8bd7e CM |
1892 | } |
1893 | ||
d352ac68 | 1894 | /* |
cad321ad CM |
1895 | * extent_io.c submission hook. This does the right thing for csum calculation |
1896 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1897 | */ |
b2950863 | 1898 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1899 | int mirror_num, unsigned long bio_flags, |
1900 | u64 bio_offset) | |
44b8bd7e CM |
1901 | { |
1902 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
0d51e28a | 1903 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
44b8bd7e | 1904 | int ret = 0; |
19b9bdb0 | 1905 | int skip_sum; |
b812ce28 | 1906 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1907 | |
6cbff00f | 1908 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1909 | |
83eea1f1 | 1910 | if (btrfs_is_free_space_inode(inode)) |
0d51e28a | 1911 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 1912 | |
7b6d91da | 1913 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1914 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1915 | if (ret) | |
61891923 | 1916 | goto out; |
5fd02043 | 1917 | |
d20f7043 | 1918 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1919 | ret = btrfs_submit_compressed_read(inode, bio, |
1920 | mirror_num, | |
1921 | bio_flags); | |
1922 | goto out; | |
c2db1073 TI |
1923 | } else if (!skip_sum) { |
1924 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1925 | if (ret) | |
61891923 | 1926 | goto out; |
c2db1073 | 1927 | } |
4d1b5fb4 | 1928 | goto mapit; |
b812ce28 | 1929 | } else if (async && !skip_sum) { |
17d217fe YZ |
1930 | /* csum items have already been cloned */ |
1931 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1932 | goto mapit; | |
19b9bdb0 | 1933 | /* we're doing a write, do the async checksumming */ |
61891923 | 1934 | ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, |
44b8bd7e | 1935 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1936 | bio_flags, bio_offset, |
1937 | __btrfs_submit_bio_start, | |
4a69a410 | 1938 | __btrfs_submit_bio_done); |
61891923 | 1939 | goto out; |
b812ce28 JB |
1940 | } else if (!skip_sum) { |
1941 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); | |
1942 | if (ret) | |
1943 | goto out; | |
19b9bdb0 CM |
1944 | } |
1945 | ||
0b86a832 | 1946 | mapit: |
61891923 SB |
1947 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); |
1948 | ||
1949 | out: | |
4246a0b6 CH |
1950 | if (ret < 0) { |
1951 | bio->bi_error = ret; | |
1952 | bio_endio(bio); | |
1953 | } | |
61891923 | 1954 | return ret; |
065631f6 | 1955 | } |
6885f308 | 1956 | |
d352ac68 CM |
1957 | /* |
1958 | * given a list of ordered sums record them in the inode. This happens | |
1959 | * at IO completion time based on sums calculated at bio submission time. | |
1960 | */ | |
ba1da2f4 | 1961 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1962 | struct inode *inode, u64 file_offset, |
1963 | struct list_head *list) | |
1964 | { | |
e6dcd2dc CM |
1965 | struct btrfs_ordered_sum *sum; |
1966 | ||
c6e30871 | 1967 | list_for_each_entry(sum, list, list) { |
39847c4d | 1968 | trans->adding_csums = 1; |
d20f7043 CM |
1969 | btrfs_csum_file_blocks(trans, |
1970 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 1971 | trans->adding_csums = 0; |
e6dcd2dc CM |
1972 | } |
1973 | return 0; | |
1974 | } | |
1975 | ||
2ac55d41 JB |
1976 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1977 | struct extent_state **cached_state) | |
ea8c2819 | 1978 | { |
09cbfeaf | 1979 | WARN_ON((end & (PAGE_SIZE - 1)) == 0); |
ea8c2819 | 1980 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
7cd8c752 | 1981 | cached_state); |
ea8c2819 CM |
1982 | } |
1983 | ||
d352ac68 | 1984 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1985 | struct btrfs_writepage_fixup { |
1986 | struct page *page; | |
1987 | struct btrfs_work work; | |
1988 | }; | |
1989 | ||
b2950863 | 1990 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1991 | { |
1992 | struct btrfs_writepage_fixup *fixup; | |
1993 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1994 | struct extent_state *cached_state = NULL; |
247e743c CM |
1995 | struct page *page; |
1996 | struct inode *inode; | |
1997 | u64 page_start; | |
1998 | u64 page_end; | |
87826df0 | 1999 | int ret; |
247e743c CM |
2000 | |
2001 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2002 | page = fixup->page; | |
4a096752 | 2003 | again: |
247e743c CM |
2004 | lock_page(page); |
2005 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
2006 | ClearPageChecked(page); | |
2007 | goto out_page; | |
2008 | } | |
2009 | ||
2010 | inode = page->mapping->host; | |
2011 | page_start = page_offset(page); | |
09cbfeaf | 2012 | page_end = page_offset(page) + PAGE_SIZE - 1; |
247e743c | 2013 | |
ff13db41 | 2014 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
d0082371 | 2015 | &cached_state); |
4a096752 CM |
2016 | |
2017 | /* already ordered? We're done */ | |
8b62b72b | 2018 | if (PagePrivate2(page)) |
247e743c | 2019 | goto out; |
4a096752 | 2020 | |
dbfdb6d1 | 2021 | ordered = btrfs_lookup_ordered_range(inode, page_start, |
09cbfeaf | 2022 | PAGE_SIZE); |
4a096752 | 2023 | if (ordered) { |
2ac55d41 JB |
2024 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
2025 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
2026 | unlock_page(page); |
2027 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 2028 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2029 | goto again; |
2030 | } | |
247e743c | 2031 | |
7cf5b976 | 2032 | ret = btrfs_delalloc_reserve_space(inode, page_start, |
09cbfeaf | 2033 | PAGE_SIZE); |
87826df0 JM |
2034 | if (ret) { |
2035 | mapping_set_error(page->mapping, ret); | |
2036 | end_extent_writepage(page, ret, page_start, page_end); | |
2037 | ClearPageChecked(page); | |
2038 | goto out; | |
2039 | } | |
2040 | ||
2ac55d41 | 2041 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 2042 | ClearPageChecked(page); |
87826df0 | 2043 | set_page_dirty(page); |
247e743c | 2044 | out: |
2ac55d41 JB |
2045 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
2046 | &cached_state, GFP_NOFS); | |
247e743c CM |
2047 | out_page: |
2048 | unlock_page(page); | |
09cbfeaf | 2049 | put_page(page); |
b897abec | 2050 | kfree(fixup); |
247e743c CM |
2051 | } |
2052 | ||
2053 | /* | |
2054 | * There are a few paths in the higher layers of the kernel that directly | |
2055 | * set the page dirty bit without asking the filesystem if it is a | |
2056 | * good idea. This causes problems because we want to make sure COW | |
2057 | * properly happens and the data=ordered rules are followed. | |
2058 | * | |
c8b97818 | 2059 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2060 | * hasn't been properly setup for IO. We kick off an async process |
2061 | * to fix it up. The async helper will wait for ordered extents, set | |
2062 | * the delalloc bit and make it safe to write the page. | |
2063 | */ | |
b2950863 | 2064 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
2065 | { |
2066 | struct inode *inode = page->mapping->host; | |
2067 | struct btrfs_writepage_fixup *fixup; | |
2068 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 2069 | |
8b62b72b CM |
2070 | /* this page is properly in the ordered list */ |
2071 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2072 | return 0; |
2073 | ||
2074 | if (PageChecked(page)) | |
2075 | return -EAGAIN; | |
2076 | ||
2077 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2078 | if (!fixup) | |
2079 | return -EAGAIN; | |
f421950f | 2080 | |
247e743c | 2081 | SetPageChecked(page); |
09cbfeaf | 2082 | get_page(page); |
9e0af237 LB |
2083 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
2084 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 2085 | fixup->page = page; |
dc6e3209 | 2086 | btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work); |
87826df0 | 2087 | return -EBUSY; |
247e743c CM |
2088 | } |
2089 | ||
d899e052 YZ |
2090 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
2091 | struct inode *inode, u64 file_pos, | |
2092 | u64 disk_bytenr, u64 disk_num_bytes, | |
2093 | u64 num_bytes, u64 ram_bytes, | |
2094 | u8 compression, u8 encryption, | |
2095 | u16 other_encoding, int extent_type) | |
2096 | { | |
2097 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2098 | struct btrfs_file_extent_item *fi; | |
2099 | struct btrfs_path *path; | |
2100 | struct extent_buffer *leaf; | |
2101 | struct btrfs_key ins; | |
1acae57b | 2102 | int extent_inserted = 0; |
d899e052 YZ |
2103 | int ret; |
2104 | ||
2105 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2106 | if (!path) |
2107 | return -ENOMEM; | |
d899e052 | 2108 | |
a1ed835e CM |
2109 | /* |
2110 | * we may be replacing one extent in the tree with another. | |
2111 | * The new extent is pinned in the extent map, and we don't want | |
2112 | * to drop it from the cache until it is completely in the btree. | |
2113 | * | |
2114 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2115 | * the caller is expected to unpin it and allow it to be merged | |
2116 | * with the others. | |
2117 | */ | |
1acae57b FDBM |
2118 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
2119 | file_pos + num_bytes, NULL, 0, | |
2120 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
2121 | if (ret) |
2122 | goto out; | |
d899e052 | 2123 | |
1acae57b FDBM |
2124 | if (!extent_inserted) { |
2125 | ins.objectid = btrfs_ino(inode); | |
2126 | ins.offset = file_pos; | |
2127 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2128 | ||
2129 | path->leave_spinning = 1; | |
2130 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
2131 | sizeof(*fi)); | |
2132 | if (ret) | |
2133 | goto out; | |
2134 | } | |
d899e052 YZ |
2135 | leaf = path->nodes[0]; |
2136 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2137 | struct btrfs_file_extent_item); | |
2138 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
2139 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
2140 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
2141 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
2142 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2143 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2144 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2145 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2146 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2147 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2148 | |
d899e052 | 2149 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2150 | btrfs_release_path(path); |
d899e052 YZ |
2151 | |
2152 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2153 | |
2154 | ins.objectid = disk_bytenr; | |
2155 | ins.offset = disk_num_bytes; | |
2156 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
2157 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
2158 | root->root_key.objectid, | |
5846a3c2 QW |
2159 | btrfs_ino(inode), file_pos, |
2160 | ram_bytes, &ins); | |
297d750b | 2161 | /* |
5846a3c2 QW |
2162 | * Release the reserved range from inode dirty range map, as it is |
2163 | * already moved into delayed_ref_head | |
297d750b QW |
2164 | */ |
2165 | btrfs_qgroup_release_data(inode, file_pos, ram_bytes); | |
79787eaa | 2166 | out: |
d899e052 | 2167 | btrfs_free_path(path); |
b9473439 | 2168 | |
79787eaa | 2169 | return ret; |
d899e052 YZ |
2170 | } |
2171 | ||
38c227d8 LB |
2172 | /* snapshot-aware defrag */ |
2173 | struct sa_defrag_extent_backref { | |
2174 | struct rb_node node; | |
2175 | struct old_sa_defrag_extent *old; | |
2176 | u64 root_id; | |
2177 | u64 inum; | |
2178 | u64 file_pos; | |
2179 | u64 extent_offset; | |
2180 | u64 num_bytes; | |
2181 | u64 generation; | |
2182 | }; | |
2183 | ||
2184 | struct old_sa_defrag_extent { | |
2185 | struct list_head list; | |
2186 | struct new_sa_defrag_extent *new; | |
2187 | ||
2188 | u64 extent_offset; | |
2189 | u64 bytenr; | |
2190 | u64 offset; | |
2191 | u64 len; | |
2192 | int count; | |
2193 | }; | |
2194 | ||
2195 | struct new_sa_defrag_extent { | |
2196 | struct rb_root root; | |
2197 | struct list_head head; | |
2198 | struct btrfs_path *path; | |
2199 | struct inode *inode; | |
2200 | u64 file_pos; | |
2201 | u64 len; | |
2202 | u64 bytenr; | |
2203 | u64 disk_len; | |
2204 | u8 compress_type; | |
2205 | }; | |
2206 | ||
2207 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2208 | struct sa_defrag_extent_backref *b2) | |
2209 | { | |
2210 | if (b1->root_id < b2->root_id) | |
2211 | return -1; | |
2212 | else if (b1->root_id > b2->root_id) | |
2213 | return 1; | |
2214 | ||
2215 | if (b1->inum < b2->inum) | |
2216 | return -1; | |
2217 | else if (b1->inum > b2->inum) | |
2218 | return 1; | |
2219 | ||
2220 | if (b1->file_pos < b2->file_pos) | |
2221 | return -1; | |
2222 | else if (b1->file_pos > b2->file_pos) | |
2223 | return 1; | |
2224 | ||
2225 | /* | |
2226 | * [------------------------------] ===> (a range of space) | |
2227 | * |<--->| |<---->| =============> (fs/file tree A) | |
2228 | * |<---------------------------->| ===> (fs/file tree B) | |
2229 | * | |
2230 | * A range of space can refer to two file extents in one tree while | |
2231 | * refer to only one file extent in another tree. | |
2232 | * | |
2233 | * So we may process a disk offset more than one time(two extents in A) | |
2234 | * and locate at the same extent(one extent in B), then insert two same | |
2235 | * backrefs(both refer to the extent in B). | |
2236 | */ | |
2237 | return 0; | |
2238 | } | |
2239 | ||
2240 | static void backref_insert(struct rb_root *root, | |
2241 | struct sa_defrag_extent_backref *backref) | |
2242 | { | |
2243 | struct rb_node **p = &root->rb_node; | |
2244 | struct rb_node *parent = NULL; | |
2245 | struct sa_defrag_extent_backref *entry; | |
2246 | int ret; | |
2247 | ||
2248 | while (*p) { | |
2249 | parent = *p; | |
2250 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2251 | ||
2252 | ret = backref_comp(backref, entry); | |
2253 | if (ret < 0) | |
2254 | p = &(*p)->rb_left; | |
2255 | else | |
2256 | p = &(*p)->rb_right; | |
2257 | } | |
2258 | ||
2259 | rb_link_node(&backref->node, parent, p); | |
2260 | rb_insert_color(&backref->node, root); | |
2261 | } | |
2262 | ||
2263 | /* | |
2264 | * Note the backref might has changed, and in this case we just return 0. | |
2265 | */ | |
2266 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2267 | void *ctx) | |
2268 | { | |
2269 | struct btrfs_file_extent_item *extent; | |
2270 | struct btrfs_fs_info *fs_info; | |
2271 | struct old_sa_defrag_extent *old = ctx; | |
2272 | struct new_sa_defrag_extent *new = old->new; | |
2273 | struct btrfs_path *path = new->path; | |
2274 | struct btrfs_key key; | |
2275 | struct btrfs_root *root; | |
2276 | struct sa_defrag_extent_backref *backref; | |
2277 | struct extent_buffer *leaf; | |
2278 | struct inode *inode = new->inode; | |
2279 | int slot; | |
2280 | int ret; | |
2281 | u64 extent_offset; | |
2282 | u64 num_bytes; | |
2283 | ||
2284 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
2285 | inum == btrfs_ino(inode)) | |
2286 | return 0; | |
2287 | ||
2288 | key.objectid = root_id; | |
2289 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2290 | key.offset = (u64)-1; | |
2291 | ||
2292 | fs_info = BTRFS_I(inode)->root->fs_info; | |
2293 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2294 | if (IS_ERR(root)) { | |
2295 | if (PTR_ERR(root) == -ENOENT) | |
2296 | return 0; | |
2297 | WARN_ON(1); | |
2298 | pr_debug("inum=%llu, offset=%llu, root_id=%llu\n", | |
2299 | inum, offset, root_id); | |
2300 | return PTR_ERR(root); | |
2301 | } | |
2302 | ||
2303 | key.objectid = inum; | |
2304 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2305 | if (offset > (u64)-1 << 32) | |
2306 | key.offset = 0; | |
2307 | else | |
2308 | key.offset = offset; | |
2309 | ||
2310 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2311 | if (WARN_ON(ret < 0)) |
38c227d8 | 2312 | return ret; |
50f1319c | 2313 | ret = 0; |
38c227d8 LB |
2314 | |
2315 | while (1) { | |
2316 | cond_resched(); | |
2317 | ||
2318 | leaf = path->nodes[0]; | |
2319 | slot = path->slots[0]; | |
2320 | ||
2321 | if (slot >= btrfs_header_nritems(leaf)) { | |
2322 | ret = btrfs_next_leaf(root, path); | |
2323 | if (ret < 0) { | |
2324 | goto out; | |
2325 | } else if (ret > 0) { | |
2326 | ret = 0; | |
2327 | goto out; | |
2328 | } | |
2329 | continue; | |
2330 | } | |
2331 | ||
2332 | path->slots[0]++; | |
2333 | ||
2334 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2335 | ||
2336 | if (key.objectid > inum) | |
2337 | goto out; | |
2338 | ||
2339 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2340 | continue; | |
2341 | ||
2342 | extent = btrfs_item_ptr(leaf, slot, | |
2343 | struct btrfs_file_extent_item); | |
2344 | ||
2345 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2346 | continue; | |
2347 | ||
e68afa49 LB |
2348 | /* |
2349 | * 'offset' refers to the exact key.offset, | |
2350 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2351 | * (key.offset - extent_offset). | |
2352 | */ | |
2353 | if (key.offset != offset) | |
38c227d8 LB |
2354 | continue; |
2355 | ||
e68afa49 | 2356 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2357 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2358 | |
38c227d8 LB |
2359 | if (extent_offset >= old->extent_offset + old->offset + |
2360 | old->len || extent_offset + num_bytes <= | |
2361 | old->extent_offset + old->offset) | |
2362 | continue; | |
38c227d8 LB |
2363 | break; |
2364 | } | |
2365 | ||
2366 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2367 | if (!backref) { | |
2368 | ret = -ENOENT; | |
2369 | goto out; | |
2370 | } | |
2371 | ||
2372 | backref->root_id = root_id; | |
2373 | backref->inum = inum; | |
e68afa49 | 2374 | backref->file_pos = offset; |
38c227d8 LB |
2375 | backref->num_bytes = num_bytes; |
2376 | backref->extent_offset = extent_offset; | |
2377 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2378 | backref->old = old; | |
2379 | backref_insert(&new->root, backref); | |
2380 | old->count++; | |
2381 | out: | |
2382 | btrfs_release_path(path); | |
2383 | WARN_ON(ret); | |
2384 | return ret; | |
2385 | } | |
2386 | ||
2387 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2388 | struct new_sa_defrag_extent *new) | |
2389 | { | |
2390 | struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info; | |
2391 | struct old_sa_defrag_extent *old, *tmp; | |
2392 | int ret; | |
2393 | ||
2394 | new->path = path; | |
2395 | ||
2396 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2397 | ret = iterate_inodes_from_logical(old->bytenr + |
2398 | old->extent_offset, fs_info, | |
38c227d8 LB |
2399 | path, record_one_backref, |
2400 | old); | |
4724b106 JB |
2401 | if (ret < 0 && ret != -ENOENT) |
2402 | return false; | |
38c227d8 LB |
2403 | |
2404 | /* no backref to be processed for this extent */ | |
2405 | if (!old->count) { | |
2406 | list_del(&old->list); | |
2407 | kfree(old); | |
2408 | } | |
2409 | } | |
2410 | ||
2411 | if (list_empty(&new->head)) | |
2412 | return false; | |
2413 | ||
2414 | return true; | |
2415 | } | |
2416 | ||
2417 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2418 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2419 | struct new_sa_defrag_extent *new) |
38c227d8 | 2420 | { |
116e0024 | 2421 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2422 | return 0; |
2423 | ||
2424 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2425 | return 0; | |
2426 | ||
116e0024 LB |
2427 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2428 | return 0; | |
2429 | ||
2430 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2431 | btrfs_file_extent_other_encoding(leaf, fi)) |
2432 | return 0; | |
2433 | ||
2434 | return 1; | |
2435 | } | |
2436 | ||
2437 | /* | |
2438 | * Note the backref might has changed, and in this case we just return 0. | |
2439 | */ | |
2440 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2441 | struct sa_defrag_extent_backref *prev, | |
2442 | struct sa_defrag_extent_backref *backref) | |
2443 | { | |
2444 | struct btrfs_file_extent_item *extent; | |
2445 | struct btrfs_file_extent_item *item; | |
2446 | struct btrfs_ordered_extent *ordered; | |
2447 | struct btrfs_trans_handle *trans; | |
2448 | struct btrfs_fs_info *fs_info; | |
2449 | struct btrfs_root *root; | |
2450 | struct btrfs_key key; | |
2451 | struct extent_buffer *leaf; | |
2452 | struct old_sa_defrag_extent *old = backref->old; | |
2453 | struct new_sa_defrag_extent *new = old->new; | |
2454 | struct inode *src_inode = new->inode; | |
2455 | struct inode *inode; | |
2456 | struct extent_state *cached = NULL; | |
2457 | int ret = 0; | |
2458 | u64 start; | |
2459 | u64 len; | |
2460 | u64 lock_start; | |
2461 | u64 lock_end; | |
2462 | bool merge = false; | |
2463 | int index; | |
2464 | ||
2465 | if (prev && prev->root_id == backref->root_id && | |
2466 | prev->inum == backref->inum && | |
2467 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2468 | merge = true; | |
2469 | ||
2470 | /* step 1: get root */ | |
2471 | key.objectid = backref->root_id; | |
2472 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2473 | key.offset = (u64)-1; | |
2474 | ||
2475 | fs_info = BTRFS_I(src_inode)->root->fs_info; | |
2476 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
2477 | ||
2478 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2479 | if (IS_ERR(root)) { | |
2480 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2481 | if (PTR_ERR(root) == -ENOENT) | |
2482 | return 0; | |
2483 | return PTR_ERR(root); | |
2484 | } | |
38c227d8 | 2485 | |
bcbba5e6 WS |
2486 | if (btrfs_root_readonly(root)) { |
2487 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2488 | return 0; | |
2489 | } | |
2490 | ||
38c227d8 LB |
2491 | /* step 2: get inode */ |
2492 | key.objectid = backref->inum; | |
2493 | key.type = BTRFS_INODE_ITEM_KEY; | |
2494 | key.offset = 0; | |
2495 | ||
2496 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2497 | if (IS_ERR(inode)) { | |
2498 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2499 | return 0; | |
2500 | } | |
2501 | ||
2502 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2503 | ||
2504 | /* step 3: relink backref */ | |
2505 | lock_start = backref->file_pos; | |
2506 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2507 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
ff13db41 | 2508 | &cached); |
38c227d8 LB |
2509 | |
2510 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2511 | if (ordered) { | |
2512 | btrfs_put_ordered_extent(ordered); | |
2513 | goto out_unlock; | |
2514 | } | |
2515 | ||
2516 | trans = btrfs_join_transaction(root); | |
2517 | if (IS_ERR(trans)) { | |
2518 | ret = PTR_ERR(trans); | |
2519 | goto out_unlock; | |
2520 | } | |
2521 | ||
2522 | key.objectid = backref->inum; | |
2523 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2524 | key.offset = backref->file_pos; | |
2525 | ||
2526 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2527 | if (ret < 0) { | |
2528 | goto out_free_path; | |
2529 | } else if (ret > 0) { | |
2530 | ret = 0; | |
2531 | goto out_free_path; | |
2532 | } | |
2533 | ||
2534 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2535 | struct btrfs_file_extent_item); | |
2536 | ||
2537 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2538 | backref->generation) | |
2539 | goto out_free_path; | |
2540 | ||
2541 | btrfs_release_path(path); | |
2542 | ||
2543 | start = backref->file_pos; | |
2544 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2545 | start += old->extent_offset + old->offset - | |
2546 | backref->extent_offset; | |
2547 | ||
2548 | len = min(backref->extent_offset + backref->num_bytes, | |
2549 | old->extent_offset + old->offset + old->len); | |
2550 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2551 | ||
2552 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2553 | start + len, 1); | |
2554 | if (ret) | |
2555 | goto out_free_path; | |
2556 | again: | |
2557 | key.objectid = btrfs_ino(inode); | |
2558 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2559 | key.offset = start; | |
2560 | ||
a09a0a70 | 2561 | path->leave_spinning = 1; |
38c227d8 LB |
2562 | if (merge) { |
2563 | struct btrfs_file_extent_item *fi; | |
2564 | u64 extent_len; | |
2565 | struct btrfs_key found_key; | |
2566 | ||
3c9665df | 2567 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2568 | if (ret < 0) |
2569 | goto out_free_path; | |
2570 | ||
2571 | path->slots[0]--; | |
2572 | leaf = path->nodes[0]; | |
2573 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2574 | ||
2575 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2576 | struct btrfs_file_extent_item); | |
2577 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2578 | ||
116e0024 LB |
2579 | if (extent_len + found_key.offset == start && |
2580 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2581 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2582 | extent_len + len); | |
2583 | btrfs_mark_buffer_dirty(leaf); | |
2584 | inode_add_bytes(inode, len); | |
2585 | ||
2586 | ret = 1; | |
2587 | goto out_free_path; | |
2588 | } else { | |
2589 | merge = false; | |
2590 | btrfs_release_path(path); | |
2591 | goto again; | |
2592 | } | |
2593 | } | |
2594 | ||
2595 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2596 | sizeof(*extent)); | |
2597 | if (ret) { | |
2598 | btrfs_abort_transaction(trans, root, ret); | |
2599 | goto out_free_path; | |
2600 | } | |
2601 | ||
2602 | leaf = path->nodes[0]; | |
2603 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2604 | struct btrfs_file_extent_item); | |
2605 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2606 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2607 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2608 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2609 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2610 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2611 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2612 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2613 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2614 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2615 | ||
2616 | btrfs_mark_buffer_dirty(leaf); | |
2617 | inode_add_bytes(inode, len); | |
a09a0a70 | 2618 | btrfs_release_path(path); |
38c227d8 LB |
2619 | |
2620 | ret = btrfs_inc_extent_ref(trans, root, new->bytenr, | |
2621 | new->disk_len, 0, | |
2622 | backref->root_id, backref->inum, | |
b06c4bf5 | 2623 | new->file_pos); /* start - extent_offset */ |
38c227d8 LB |
2624 | if (ret) { |
2625 | btrfs_abort_transaction(trans, root, ret); | |
2626 | goto out_free_path; | |
2627 | } | |
2628 | ||
2629 | ret = 1; | |
2630 | out_free_path: | |
2631 | btrfs_release_path(path); | |
a09a0a70 | 2632 | path->leave_spinning = 0; |
38c227d8 LB |
2633 | btrfs_end_transaction(trans, root); |
2634 | out_unlock: | |
2635 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2636 | &cached, GFP_NOFS); | |
2637 | iput(inode); | |
2638 | return ret; | |
2639 | } | |
2640 | ||
6f519564 LB |
2641 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2642 | { | |
2643 | struct old_sa_defrag_extent *old, *tmp; | |
2644 | ||
2645 | if (!new) | |
2646 | return; | |
2647 | ||
2648 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
6f519564 LB |
2649 | kfree(old); |
2650 | } | |
2651 | kfree(new); | |
2652 | } | |
2653 | ||
38c227d8 LB |
2654 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2655 | { | |
2656 | struct btrfs_path *path; | |
38c227d8 LB |
2657 | struct sa_defrag_extent_backref *backref; |
2658 | struct sa_defrag_extent_backref *prev = NULL; | |
2659 | struct inode *inode; | |
2660 | struct btrfs_root *root; | |
2661 | struct rb_node *node; | |
2662 | int ret; | |
2663 | ||
2664 | inode = new->inode; | |
2665 | root = BTRFS_I(inode)->root; | |
2666 | ||
2667 | path = btrfs_alloc_path(); | |
2668 | if (!path) | |
2669 | return; | |
2670 | ||
2671 | if (!record_extent_backrefs(path, new)) { | |
2672 | btrfs_free_path(path); | |
2673 | goto out; | |
2674 | } | |
2675 | btrfs_release_path(path); | |
2676 | ||
2677 | while (1) { | |
2678 | node = rb_first(&new->root); | |
2679 | if (!node) | |
2680 | break; | |
2681 | rb_erase(node, &new->root); | |
2682 | ||
2683 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2684 | ||
2685 | ret = relink_extent_backref(path, prev, backref); | |
2686 | WARN_ON(ret < 0); | |
2687 | ||
2688 | kfree(prev); | |
2689 | ||
2690 | if (ret == 1) | |
2691 | prev = backref; | |
2692 | else | |
2693 | prev = NULL; | |
2694 | cond_resched(); | |
2695 | } | |
2696 | kfree(prev); | |
2697 | ||
2698 | btrfs_free_path(path); | |
38c227d8 | 2699 | out: |
6f519564 LB |
2700 | free_sa_defrag_extent(new); |
2701 | ||
38c227d8 LB |
2702 | atomic_dec(&root->fs_info->defrag_running); |
2703 | wake_up(&root->fs_info->transaction_wait); | |
38c227d8 LB |
2704 | } |
2705 | ||
2706 | static struct new_sa_defrag_extent * | |
2707 | record_old_file_extents(struct inode *inode, | |
2708 | struct btrfs_ordered_extent *ordered) | |
2709 | { | |
2710 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2711 | struct btrfs_path *path; | |
2712 | struct btrfs_key key; | |
6f519564 | 2713 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2714 | struct new_sa_defrag_extent *new; |
2715 | int ret; | |
2716 | ||
2717 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2718 | if (!new) | |
2719 | return NULL; | |
2720 | ||
2721 | new->inode = inode; | |
2722 | new->file_pos = ordered->file_offset; | |
2723 | new->len = ordered->len; | |
2724 | new->bytenr = ordered->start; | |
2725 | new->disk_len = ordered->disk_len; | |
2726 | new->compress_type = ordered->compress_type; | |
2727 | new->root = RB_ROOT; | |
2728 | INIT_LIST_HEAD(&new->head); | |
2729 | ||
2730 | path = btrfs_alloc_path(); | |
2731 | if (!path) | |
2732 | goto out_kfree; | |
2733 | ||
2734 | key.objectid = btrfs_ino(inode); | |
2735 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2736 | key.offset = new->file_pos; | |
2737 | ||
2738 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2739 | if (ret < 0) | |
2740 | goto out_free_path; | |
2741 | if (ret > 0 && path->slots[0] > 0) | |
2742 | path->slots[0]--; | |
2743 | ||
2744 | /* find out all the old extents for the file range */ | |
2745 | while (1) { | |
2746 | struct btrfs_file_extent_item *extent; | |
2747 | struct extent_buffer *l; | |
2748 | int slot; | |
2749 | u64 num_bytes; | |
2750 | u64 offset; | |
2751 | u64 end; | |
2752 | u64 disk_bytenr; | |
2753 | u64 extent_offset; | |
2754 | ||
2755 | l = path->nodes[0]; | |
2756 | slot = path->slots[0]; | |
2757 | ||
2758 | if (slot >= btrfs_header_nritems(l)) { | |
2759 | ret = btrfs_next_leaf(root, path); | |
2760 | if (ret < 0) | |
6f519564 | 2761 | goto out_free_path; |
38c227d8 LB |
2762 | else if (ret > 0) |
2763 | break; | |
2764 | continue; | |
2765 | } | |
2766 | ||
2767 | btrfs_item_key_to_cpu(l, &key, slot); | |
2768 | ||
2769 | if (key.objectid != btrfs_ino(inode)) | |
2770 | break; | |
2771 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2772 | break; | |
2773 | if (key.offset >= new->file_pos + new->len) | |
2774 | break; | |
2775 | ||
2776 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2777 | ||
2778 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2779 | if (key.offset + num_bytes < new->file_pos) | |
2780 | goto next; | |
2781 | ||
2782 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2783 | if (!disk_bytenr) | |
2784 | goto next; | |
2785 | ||
2786 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2787 | ||
2788 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2789 | if (!old) | |
6f519564 | 2790 | goto out_free_path; |
38c227d8 LB |
2791 | |
2792 | offset = max(new->file_pos, key.offset); | |
2793 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2794 | ||
2795 | old->bytenr = disk_bytenr; | |
2796 | old->extent_offset = extent_offset; | |
2797 | old->offset = offset - key.offset; | |
2798 | old->len = end - offset; | |
2799 | old->new = new; | |
2800 | old->count = 0; | |
2801 | list_add_tail(&old->list, &new->head); | |
2802 | next: | |
2803 | path->slots[0]++; | |
2804 | cond_resched(); | |
2805 | } | |
2806 | ||
2807 | btrfs_free_path(path); | |
2808 | atomic_inc(&root->fs_info->defrag_running); | |
2809 | ||
2810 | return new; | |
2811 | ||
38c227d8 LB |
2812 | out_free_path: |
2813 | btrfs_free_path(path); | |
2814 | out_kfree: | |
6f519564 | 2815 | free_sa_defrag_extent(new); |
38c227d8 LB |
2816 | return NULL; |
2817 | } | |
2818 | ||
e570fd27 MX |
2819 | static void btrfs_release_delalloc_bytes(struct btrfs_root *root, |
2820 | u64 start, u64 len) | |
2821 | { | |
2822 | struct btrfs_block_group_cache *cache; | |
2823 | ||
2824 | cache = btrfs_lookup_block_group(root->fs_info, start); | |
2825 | ASSERT(cache); | |
2826 | ||
2827 | spin_lock(&cache->lock); | |
2828 | cache->delalloc_bytes -= len; | |
2829 | spin_unlock(&cache->lock); | |
2830 | ||
2831 | btrfs_put_block_group(cache); | |
2832 | } | |
2833 | ||
d352ac68 CM |
2834 | /* as ordered data IO finishes, this gets called so we can finish |
2835 | * an ordered extent if the range of bytes in the file it covers are | |
2836 | * fully written. | |
2837 | */ | |
5fd02043 | 2838 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2839 | { |
5fd02043 | 2840 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 2841 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2842 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2843 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2844 | struct extent_state *cached_state = NULL; |
38c227d8 | 2845 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2846 | int compress_type = 0; |
77cef2ec JB |
2847 | int ret = 0; |
2848 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2849 | bool nolock; |
77cef2ec | 2850 | bool truncated = false; |
e6dcd2dc | 2851 | |
83eea1f1 | 2852 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 2853 | |
5fd02043 JB |
2854 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2855 | ret = -EIO; | |
2856 | goto out; | |
2857 | } | |
2858 | ||
f612496b MX |
2859 | btrfs_free_io_failure_record(inode, ordered_extent->file_offset, |
2860 | ordered_extent->file_offset + | |
2861 | ordered_extent->len - 1); | |
2862 | ||
77cef2ec JB |
2863 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2864 | truncated = true; | |
2865 | logical_len = ordered_extent->truncated_len; | |
2866 | /* Truncated the entire extent, don't bother adding */ | |
2867 | if (!logical_len) | |
2868 | goto out; | |
2869 | } | |
2870 | ||
c2167754 | 2871 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2872 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a QW |
2873 | |
2874 | /* | |
2875 | * For mwrite(mmap + memset to write) case, we still reserve | |
2876 | * space for NOCOW range. | |
2877 | * As NOCOW won't cause a new delayed ref, just free the space | |
2878 | */ | |
2879 | btrfs_qgroup_free_data(inode, ordered_extent->file_offset, | |
2880 | ordered_extent->len); | |
6c760c07 JB |
2881 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2882 | if (nolock) | |
2883 | trans = btrfs_join_transaction_nolock(root); | |
2884 | else | |
2885 | trans = btrfs_join_transaction(root); | |
2886 | if (IS_ERR(trans)) { | |
2887 | ret = PTR_ERR(trans); | |
2888 | trans = NULL; | |
2889 | goto out; | |
c2167754 | 2890 | } |
6c760c07 JB |
2891 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2892 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2893 | if (ret) /* -ENOMEM or corruption */ | |
2894 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
2895 | goto out; |
2896 | } | |
e6dcd2dc | 2897 | |
2ac55d41 JB |
2898 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2899 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ff13db41 | 2900 | &cached_state); |
e6dcd2dc | 2901 | |
38c227d8 LB |
2902 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2903 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2904 | EXTENT_DEFRAG, 1, cached_state); | |
2905 | if (ret) { | |
2906 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 2907 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
2908 | /* the inode is shared */ |
2909 | new = record_old_file_extents(inode, ordered_extent); | |
2910 | ||
2911 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
2912 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2913 | EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS); | |
2914 | } | |
2915 | ||
0cb59c99 | 2916 | if (nolock) |
7a7eaa40 | 2917 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 2918 | else |
7a7eaa40 | 2919 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2920 | if (IS_ERR(trans)) { |
2921 | ret = PTR_ERR(trans); | |
2922 | trans = NULL; | |
2923 | goto out_unlock; | |
2924 | } | |
a79b7d4b | 2925 | |
0ca1f7ce | 2926 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 2927 | |
c8b97818 | 2928 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 2929 | compress_type = ordered_extent->compress_type; |
d899e052 | 2930 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 2931 | BUG_ON(compress_type); |
920bbbfb | 2932 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
2933 | ordered_extent->file_offset, |
2934 | ordered_extent->file_offset + | |
77cef2ec | 2935 | logical_len); |
d899e052 | 2936 | } else { |
0af3d00b | 2937 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
2938 | ret = insert_reserved_file_extent(trans, inode, |
2939 | ordered_extent->file_offset, | |
2940 | ordered_extent->start, | |
2941 | ordered_extent->disk_len, | |
77cef2ec | 2942 | logical_len, logical_len, |
261507a0 | 2943 | compress_type, 0, 0, |
d899e052 | 2944 | BTRFS_FILE_EXTENT_REG); |
e570fd27 MX |
2945 | if (!ret) |
2946 | btrfs_release_delalloc_bytes(root, | |
2947 | ordered_extent->start, | |
2948 | ordered_extent->disk_len); | |
d899e052 | 2949 | } |
5dc562c5 JB |
2950 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
2951 | ordered_extent->file_offset, ordered_extent->len, | |
2952 | trans->transid); | |
79787eaa JM |
2953 | if (ret < 0) { |
2954 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 2955 | goto out_unlock; |
79787eaa | 2956 | } |
2ac55d41 | 2957 | |
e6dcd2dc CM |
2958 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
2959 | &ordered_extent->list); | |
2960 | ||
6c760c07 JB |
2961 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2962 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2963 | if (ret) { /* -ENOMEM or corruption */ | |
2964 | btrfs_abort_transaction(trans, root, ret); | |
2965 | goto out_unlock; | |
1ef30be1 JB |
2966 | } |
2967 | ret = 0; | |
5fd02043 JB |
2968 | out_unlock: |
2969 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
2970 | ordered_extent->file_offset + | |
2971 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 2972 | out: |
5b0e95bf | 2973 | if (root != root->fs_info->tree_root) |
0cb59c99 | 2974 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 MX |
2975 | if (trans) |
2976 | btrfs_end_transaction(trans, root); | |
0cb59c99 | 2977 | |
77cef2ec JB |
2978 | if (ret || truncated) { |
2979 | u64 start, end; | |
2980 | ||
2981 | if (truncated) | |
2982 | start = ordered_extent->file_offset + logical_len; | |
2983 | else | |
2984 | start = ordered_extent->file_offset; | |
2985 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
2986 | clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS); | |
2987 | ||
2988 | /* Drop the cache for the part of the extent we didn't write. */ | |
2989 | btrfs_drop_extent_cache(inode, start, end, 0); | |
5fd02043 | 2990 | |
0bec9ef5 JB |
2991 | /* |
2992 | * If the ordered extent had an IOERR or something else went | |
2993 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
2994 | * back to the allocator. We only free the extent in the |
2995 | * truncated case if we didn't write out the extent at all. | |
0bec9ef5 | 2996 | */ |
77cef2ec JB |
2997 | if ((ret || !logical_len) && |
2998 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
0bec9ef5 JB |
2999 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
3000 | btrfs_free_reserved_extent(root, ordered_extent->start, | |
e570fd27 | 3001 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
3002 | } |
3003 | ||
3004 | ||
5fd02043 | 3005 | /* |
8bad3c02 LB |
3006 | * This needs to be done to make sure anybody waiting knows we are done |
3007 | * updating everything for this ordered extent. | |
5fd02043 JB |
3008 | */ |
3009 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
3010 | ||
38c227d8 | 3011 | /* for snapshot-aware defrag */ |
6f519564 LB |
3012 | if (new) { |
3013 | if (ret) { | |
3014 | free_sa_defrag_extent(new); | |
3015 | atomic_dec(&root->fs_info->defrag_running); | |
3016 | } else { | |
3017 | relink_file_extents(new); | |
3018 | } | |
3019 | } | |
38c227d8 | 3020 | |
e6dcd2dc CM |
3021 | /* once for us */ |
3022 | btrfs_put_ordered_extent(ordered_extent); | |
3023 | /* once for the tree */ | |
3024 | btrfs_put_ordered_extent(ordered_extent); | |
3025 | ||
5fd02043 JB |
3026 | return ret; |
3027 | } | |
3028 | ||
3029 | static void finish_ordered_fn(struct btrfs_work *work) | |
3030 | { | |
3031 | struct btrfs_ordered_extent *ordered_extent; | |
3032 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
3033 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
3034 | } |
3035 | ||
b2950863 | 3036 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
3037 | struct extent_state *state, int uptodate) |
3038 | { | |
5fd02043 JB |
3039 | struct inode *inode = page->mapping->host; |
3040 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3041 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
9e0af237 LB |
3042 | struct btrfs_workqueue *wq; |
3043 | btrfs_work_func_t func; | |
5fd02043 | 3044 | |
1abe9b8a | 3045 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
3046 | ||
8b62b72b | 3047 | ClearPagePrivate2(page); |
5fd02043 JB |
3048 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
3049 | end - start + 1, uptodate)) | |
3050 | return 0; | |
3051 | ||
9e0af237 LB |
3052 | if (btrfs_is_free_space_inode(inode)) { |
3053 | wq = root->fs_info->endio_freespace_worker; | |
3054 | func = btrfs_freespace_write_helper; | |
3055 | } else { | |
3056 | wq = root->fs_info->endio_write_workers; | |
3057 | func = btrfs_endio_write_helper; | |
3058 | } | |
5fd02043 | 3059 | |
9e0af237 LB |
3060 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
3061 | NULL); | |
3062 | btrfs_queue_work(wq, &ordered_extent->work); | |
5fd02043 JB |
3063 | |
3064 | return 0; | |
211f90e6 CM |
3065 | } |
3066 | ||
dc380aea MX |
3067 | static int __readpage_endio_check(struct inode *inode, |
3068 | struct btrfs_io_bio *io_bio, | |
3069 | int icsum, struct page *page, | |
3070 | int pgoff, u64 start, size_t len) | |
3071 | { | |
3072 | char *kaddr; | |
3073 | u32 csum_expected; | |
3074 | u32 csum = ~(u32)0; | |
dc380aea MX |
3075 | |
3076 | csum_expected = *(((u32 *)io_bio->csum) + icsum); | |
3077 | ||
3078 | kaddr = kmap_atomic(page); | |
3079 | csum = btrfs_csum_data(kaddr + pgoff, csum, len); | |
3080 | btrfs_csum_final(csum, (char *)&csum); | |
3081 | if (csum != csum_expected) | |
3082 | goto zeroit; | |
3083 | ||
3084 | kunmap_atomic(kaddr); | |
3085 | return 0; | |
3086 | zeroit: | |
94647322 DS |
3087 | btrfs_warn_rl(BTRFS_I(inode)->root->fs_info, |
3088 | "csum failed ino %llu off %llu csum %u expected csum %u", | |
dc380aea MX |
3089 | btrfs_ino(inode), start, csum, csum_expected); |
3090 | memset(kaddr + pgoff, 1, len); | |
3091 | flush_dcache_page(page); | |
3092 | kunmap_atomic(kaddr); | |
3093 | if (csum_expected == 0) | |
3094 | return 0; | |
3095 | return -EIO; | |
3096 | } | |
3097 | ||
d352ac68 CM |
3098 | /* |
3099 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
3100 | * if there's a match, we allow the bio to finish. If not, the code in |
3101 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 3102 | */ |
facc8a22 MX |
3103 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
3104 | u64 phy_offset, struct page *page, | |
3105 | u64 start, u64 end, int mirror) | |
07157aac | 3106 | { |
4eee4fa4 | 3107 | size_t offset = start - page_offset(page); |
07157aac | 3108 | struct inode *inode = page->mapping->host; |
d1310b2e | 3109 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3110 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 3111 | |
d20f7043 CM |
3112 | if (PageChecked(page)) { |
3113 | ClearPageChecked(page); | |
dc380aea | 3114 | return 0; |
d20f7043 | 3115 | } |
6cbff00f CH |
3116 | |
3117 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 3118 | return 0; |
17d217fe YZ |
3119 | |
3120 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 3121 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
91166212 | 3122 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM); |
b6cda9bc | 3123 | return 0; |
17d217fe | 3124 | } |
d20f7043 | 3125 | |
facc8a22 | 3126 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
dc380aea MX |
3127 | return __readpage_endio_check(inode, io_bio, phy_offset, page, offset, |
3128 | start, (size_t)(end - start + 1)); | |
07157aac | 3129 | } |
b888db2b | 3130 | |
24bbcf04 YZ |
3131 | void btrfs_add_delayed_iput(struct inode *inode) |
3132 | { | |
3133 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
8089fe62 | 3134 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3135 | |
3136 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3137 | return; | |
3138 | ||
24bbcf04 | 3139 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3140 | if (binode->delayed_iput_count == 0) { |
3141 | ASSERT(list_empty(&binode->delayed_iput)); | |
3142 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
3143 | } else { | |
3144 | binode->delayed_iput_count++; | |
3145 | } | |
24bbcf04 YZ |
3146 | spin_unlock(&fs_info->delayed_iput_lock); |
3147 | } | |
3148 | ||
3149 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
3150 | { | |
24bbcf04 | 3151 | struct btrfs_fs_info *fs_info = root->fs_info; |
24bbcf04 | 3152 | |
24bbcf04 | 3153 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3154 | while (!list_empty(&fs_info->delayed_iputs)) { |
3155 | struct btrfs_inode *inode; | |
3156 | ||
3157 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3158 | struct btrfs_inode, delayed_iput); | |
3159 | if (inode->delayed_iput_count) { | |
3160 | inode->delayed_iput_count--; | |
3161 | list_move_tail(&inode->delayed_iput, | |
3162 | &fs_info->delayed_iputs); | |
3163 | } else { | |
3164 | list_del_init(&inode->delayed_iput); | |
3165 | } | |
3166 | spin_unlock(&fs_info->delayed_iput_lock); | |
3167 | iput(&inode->vfs_inode); | |
3168 | spin_lock(&fs_info->delayed_iput_lock); | |
24bbcf04 | 3169 | } |
8089fe62 | 3170 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3171 | } |
3172 | ||
d68fc57b | 3173 | /* |
42b2aa86 | 3174 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
3175 | * files in the subvolume, it removes orphan item and frees block_rsv |
3176 | * structure. | |
3177 | */ | |
3178 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
3179 | struct btrfs_root *root) | |
3180 | { | |
90290e19 | 3181 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
3182 | int ret; |
3183 | ||
8a35d95f | 3184 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
3185 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
3186 | return; | |
3187 | ||
90290e19 | 3188 | spin_lock(&root->orphan_lock); |
8a35d95f | 3189 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
3190 | spin_unlock(&root->orphan_lock); |
3191 | return; | |
3192 | } | |
3193 | ||
3194 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
3195 | spin_unlock(&root->orphan_lock); | |
3196 | return; | |
3197 | } | |
3198 | ||
3199 | block_rsv = root->orphan_block_rsv; | |
3200 | root->orphan_block_rsv = NULL; | |
3201 | spin_unlock(&root->orphan_lock); | |
3202 | ||
27cdeb70 | 3203 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) && |
d68fc57b YZ |
3204 | btrfs_root_refs(&root->root_item) > 0) { |
3205 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
3206 | root->root_key.objectid); | |
4ef31a45 JB |
3207 | if (ret) |
3208 | btrfs_abort_transaction(trans, root, ret); | |
3209 | else | |
27cdeb70 MX |
3210 | clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, |
3211 | &root->state); | |
d68fc57b YZ |
3212 | } |
3213 | ||
90290e19 JB |
3214 | if (block_rsv) { |
3215 | WARN_ON(block_rsv->size > 0); | |
3216 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
3217 | } |
3218 | } | |
3219 | ||
7b128766 JB |
3220 | /* |
3221 | * This creates an orphan entry for the given inode in case something goes | |
3222 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
3223 | * |
3224 | * NOTE: caller of this function should reserve 5 units of metadata for | |
3225 | * this function. | |
7b128766 JB |
3226 | */ |
3227 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
3228 | { | |
3229 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3230 | struct btrfs_block_rsv *block_rsv = NULL; |
3231 | int reserve = 0; | |
3232 | int insert = 0; | |
3233 | int ret; | |
7b128766 | 3234 | |
d68fc57b | 3235 | if (!root->orphan_block_rsv) { |
66d8f3dd | 3236 | block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
b532402e TI |
3237 | if (!block_rsv) |
3238 | return -ENOMEM; | |
d68fc57b | 3239 | } |
7b128766 | 3240 | |
d68fc57b YZ |
3241 | spin_lock(&root->orphan_lock); |
3242 | if (!root->orphan_block_rsv) { | |
3243 | root->orphan_block_rsv = block_rsv; | |
3244 | } else if (block_rsv) { | |
3245 | btrfs_free_block_rsv(root, block_rsv); | |
3246 | block_rsv = NULL; | |
7b128766 | 3247 | } |
7b128766 | 3248 | |
8a35d95f JB |
3249 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3250 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
3251 | #if 0 |
3252 | /* | |
3253 | * For proper ENOSPC handling, we should do orphan | |
3254 | * cleanup when mounting. But this introduces backward | |
3255 | * compatibility issue. | |
3256 | */ | |
3257 | if (!xchg(&root->orphan_item_inserted, 1)) | |
3258 | insert = 2; | |
3259 | else | |
3260 | insert = 1; | |
3261 | #endif | |
3262 | insert = 1; | |
321f0e70 | 3263 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
3264 | } |
3265 | ||
72ac3c0d JB |
3266 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3267 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3268 | reserve = 1; |
d68fc57b | 3269 | spin_unlock(&root->orphan_lock); |
7b128766 | 3270 | |
d68fc57b YZ |
3271 | /* grab metadata reservation from transaction handle */ |
3272 | if (reserve) { | |
3273 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
3b6571c1 JB |
3274 | ASSERT(!ret); |
3275 | if (ret) { | |
3276 | atomic_dec(&root->orphan_inodes); | |
3277 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
3278 | &BTRFS_I(inode)->runtime_flags); | |
3279 | if (insert) | |
3280 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, | |
3281 | &BTRFS_I(inode)->runtime_flags); | |
3282 | return ret; | |
3283 | } | |
d68fc57b | 3284 | } |
7b128766 | 3285 | |
d68fc57b YZ |
3286 | /* insert an orphan item to track this unlinked/truncated file */ |
3287 | if (insert >= 1) { | |
33345d01 | 3288 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
4ef31a45 | 3289 | if (ret) { |
703c88e0 | 3290 | atomic_dec(&root->orphan_inodes); |
4ef31a45 JB |
3291 | if (reserve) { |
3292 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
3293 | &BTRFS_I(inode)->runtime_flags); | |
3294 | btrfs_orphan_release_metadata(inode); | |
3295 | } | |
3296 | if (ret != -EEXIST) { | |
e8e7cff6 JB |
3297 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3298 | &BTRFS_I(inode)->runtime_flags); | |
4ef31a45 JB |
3299 | btrfs_abort_transaction(trans, root, ret); |
3300 | return ret; | |
3301 | } | |
79787eaa JM |
3302 | } |
3303 | ret = 0; | |
d68fc57b YZ |
3304 | } |
3305 | ||
3306 | /* insert an orphan item to track subvolume contains orphan files */ | |
3307 | if (insert >= 2) { | |
3308 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
3309 | root->root_key.objectid); | |
79787eaa JM |
3310 | if (ret && ret != -EEXIST) { |
3311 | btrfs_abort_transaction(trans, root, ret); | |
3312 | return ret; | |
3313 | } | |
d68fc57b YZ |
3314 | } |
3315 | return 0; | |
7b128766 JB |
3316 | } |
3317 | ||
3318 | /* | |
3319 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
3320 | * item for this particular inode. | |
3321 | */ | |
48a3b636 ES |
3322 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3323 | struct inode *inode) | |
7b128766 JB |
3324 | { |
3325 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3326 | int delete_item = 0; |
3327 | int release_rsv = 0; | |
7b128766 JB |
3328 | int ret = 0; |
3329 | ||
d68fc57b | 3330 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
3331 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3332 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3333 | delete_item = 1; |
7b128766 | 3334 | |
72ac3c0d JB |
3335 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3336 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3337 | release_rsv = 1; |
d68fc57b | 3338 | spin_unlock(&root->orphan_lock); |
7b128766 | 3339 | |
703c88e0 | 3340 | if (delete_item) { |
8a35d95f | 3341 | atomic_dec(&root->orphan_inodes); |
703c88e0 FDBM |
3342 | if (trans) |
3343 | ret = btrfs_del_orphan_item(trans, root, | |
3344 | btrfs_ino(inode)); | |
8a35d95f | 3345 | } |
7b128766 | 3346 | |
703c88e0 FDBM |
3347 | if (release_rsv) |
3348 | btrfs_orphan_release_metadata(inode); | |
3349 | ||
4ef31a45 | 3350 | return ret; |
7b128766 JB |
3351 | } |
3352 | ||
3353 | /* | |
3354 | * this cleans up any orphans that may be left on the list from the last use | |
3355 | * of this root. | |
3356 | */ | |
66b4ffd1 | 3357 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
3358 | { |
3359 | struct btrfs_path *path; | |
3360 | struct extent_buffer *leaf; | |
7b128766 JB |
3361 | struct btrfs_key key, found_key; |
3362 | struct btrfs_trans_handle *trans; | |
3363 | struct inode *inode; | |
8f6d7f4f | 3364 | u64 last_objectid = 0; |
7b128766 JB |
3365 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3366 | ||
d68fc57b | 3367 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3368 | return 0; |
c71bf099 YZ |
3369 | |
3370 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3371 | if (!path) { |
3372 | ret = -ENOMEM; | |
3373 | goto out; | |
3374 | } | |
e4058b54 | 3375 | path->reada = READA_BACK; |
7b128766 JB |
3376 | |
3377 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3378 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3379 | key.offset = (u64)-1; |
3380 | ||
7b128766 JB |
3381 | while (1) { |
3382 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3383 | if (ret < 0) |
3384 | goto out; | |
7b128766 JB |
3385 | |
3386 | /* | |
3387 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3388 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3389 | * find the key and see if we have stuff that matches |
3390 | */ | |
3391 | if (ret > 0) { | |
66b4ffd1 | 3392 | ret = 0; |
7b128766 JB |
3393 | if (path->slots[0] == 0) |
3394 | break; | |
3395 | path->slots[0]--; | |
3396 | } | |
3397 | ||
3398 | /* pull out the item */ | |
3399 | leaf = path->nodes[0]; | |
7b128766 JB |
3400 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3401 | ||
3402 | /* make sure the item matches what we want */ | |
3403 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3404 | break; | |
962a298f | 3405 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3406 | break; |
3407 | ||
3408 | /* release the path since we're done with it */ | |
b3b4aa74 | 3409 | btrfs_release_path(path); |
7b128766 JB |
3410 | |
3411 | /* | |
3412 | * this is where we are basically btrfs_lookup, without the | |
3413 | * crossing root thing. we store the inode number in the | |
3414 | * offset of the orphan item. | |
3415 | */ | |
8f6d7f4f JB |
3416 | |
3417 | if (found_key.offset == last_objectid) { | |
c2cf52eb SK |
3418 | btrfs_err(root->fs_info, |
3419 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3420 | ret = -EINVAL; |
3421 | goto out; | |
3422 | } | |
3423 | ||
3424 | last_objectid = found_key.offset; | |
3425 | ||
5d4f98a2 YZ |
3426 | found_key.objectid = found_key.offset; |
3427 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3428 | found_key.offset = 0; | |
73f73415 | 3429 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3430 | ret = PTR_ERR_OR_ZERO(inode); |
a8c9e576 | 3431 | if (ret && ret != -ESTALE) |
66b4ffd1 | 3432 | goto out; |
7b128766 | 3433 | |
f8e9e0b0 AJ |
3434 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
3435 | struct btrfs_root *dead_root; | |
3436 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3437 | int is_dead_root = 0; | |
3438 | ||
3439 | /* | |
3440 | * this is an orphan in the tree root. Currently these | |
3441 | * could come from 2 sources: | |
3442 | * a) a snapshot deletion in progress | |
3443 | * b) a free space cache inode | |
3444 | * We need to distinguish those two, as the snapshot | |
3445 | * orphan must not get deleted. | |
3446 | * find_dead_roots already ran before us, so if this | |
3447 | * is a snapshot deletion, we should find the root | |
3448 | * in the dead_roots list | |
3449 | */ | |
3450 | spin_lock(&fs_info->trans_lock); | |
3451 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3452 | root_list) { | |
3453 | if (dead_root->root_key.objectid == | |
3454 | found_key.objectid) { | |
3455 | is_dead_root = 1; | |
3456 | break; | |
3457 | } | |
3458 | } | |
3459 | spin_unlock(&fs_info->trans_lock); | |
3460 | if (is_dead_root) { | |
3461 | /* prevent this orphan from being found again */ | |
3462 | key.offset = found_key.objectid - 1; | |
3463 | continue; | |
3464 | } | |
3465 | } | |
7b128766 | 3466 | /* |
a8c9e576 JB |
3467 | * Inode is already gone but the orphan item is still there, |
3468 | * kill the orphan item. | |
7b128766 | 3469 | */ |
a8c9e576 JB |
3470 | if (ret == -ESTALE) { |
3471 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
3472 | if (IS_ERR(trans)) { |
3473 | ret = PTR_ERR(trans); | |
3474 | goto out; | |
3475 | } | |
c2cf52eb SK |
3476 | btrfs_debug(root->fs_info, "auto deleting %Lu", |
3477 | found_key.objectid); | |
a8c9e576 JB |
3478 | ret = btrfs_del_orphan_item(trans, root, |
3479 | found_key.objectid); | |
5b21f2ed | 3480 | btrfs_end_transaction(trans, root); |
4ef31a45 JB |
3481 | if (ret) |
3482 | goto out; | |
7b128766 JB |
3483 | continue; |
3484 | } | |
3485 | ||
a8c9e576 JB |
3486 | /* |
3487 | * add this inode to the orphan list so btrfs_orphan_del does | |
3488 | * the proper thing when we hit it | |
3489 | */ | |
8a35d95f JB |
3490 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3491 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3492 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3493 | |
7b128766 JB |
3494 | /* if we have links, this was a truncate, lets do that */ |
3495 | if (inode->i_nlink) { | |
fae7f21c | 3496 | if (WARN_ON(!S_ISREG(inode->i_mode))) { |
a41ad394 JB |
3497 | iput(inode); |
3498 | continue; | |
3499 | } | |
7b128766 | 3500 | nr_truncate++; |
f3fe820c JB |
3501 | |
3502 | /* 1 for the orphan item deletion. */ | |
3503 | trans = btrfs_start_transaction(root, 1); | |
3504 | if (IS_ERR(trans)) { | |
c69b26b0 | 3505 | iput(inode); |
f3fe820c JB |
3506 | ret = PTR_ERR(trans); |
3507 | goto out; | |
3508 | } | |
3509 | ret = btrfs_orphan_add(trans, inode); | |
3510 | btrfs_end_transaction(trans, root); | |
c69b26b0 JB |
3511 | if (ret) { |
3512 | iput(inode); | |
f3fe820c | 3513 | goto out; |
c69b26b0 | 3514 | } |
f3fe820c | 3515 | |
66b4ffd1 | 3516 | ret = btrfs_truncate(inode); |
4a7d0f68 JB |
3517 | if (ret) |
3518 | btrfs_orphan_del(NULL, inode); | |
7b128766 JB |
3519 | } else { |
3520 | nr_unlink++; | |
3521 | } | |
3522 | ||
3523 | /* this will do delete_inode and everything for us */ | |
3524 | iput(inode); | |
66b4ffd1 JB |
3525 | if (ret) |
3526 | goto out; | |
7b128766 | 3527 | } |
3254c876 MX |
3528 | /* release the path since we're done with it */ |
3529 | btrfs_release_path(path); | |
3530 | ||
d68fc57b YZ |
3531 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3532 | ||
3533 | if (root->orphan_block_rsv) | |
3534 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
3535 | (u64)-1); | |
3536 | ||
27cdeb70 MX |
3537 | if (root->orphan_block_rsv || |
3538 | test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { | |
7a7eaa40 | 3539 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
3540 | if (!IS_ERR(trans)) |
3541 | btrfs_end_transaction(trans, root); | |
d68fc57b | 3542 | } |
7b128766 JB |
3543 | |
3544 | if (nr_unlink) | |
4884b476 | 3545 | btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3546 | if (nr_truncate) |
4884b476 | 3547 | btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3548 | |
3549 | out: | |
3550 | if (ret) | |
68b663d1 | 3551 | btrfs_err(root->fs_info, |
c2cf52eb | 3552 | "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3553 | btrfs_free_path(path); |
3554 | return ret; | |
7b128766 JB |
3555 | } |
3556 | ||
46a53cca CM |
3557 | /* |
3558 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3559 | * don't find any xattrs, we know there can't be any acls. | |
3560 | * | |
3561 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3562 | */ | |
3563 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3564 | int slot, u64 objectid, |
3565 | int *first_xattr_slot) | |
46a53cca CM |
3566 | { |
3567 | u32 nritems = btrfs_header_nritems(leaf); | |
3568 | struct btrfs_key found_key; | |
f23b5a59 JB |
3569 | static u64 xattr_access = 0; |
3570 | static u64 xattr_default = 0; | |
46a53cca CM |
3571 | int scanned = 0; |
3572 | ||
f23b5a59 | 3573 | if (!xattr_access) { |
97d79299 AG |
3574 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3575 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3576 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3577 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3578 | } |
3579 | ||
46a53cca | 3580 | slot++; |
63541927 | 3581 | *first_xattr_slot = -1; |
46a53cca CM |
3582 | while (slot < nritems) { |
3583 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3584 | ||
3585 | /* we found a different objectid, there must not be acls */ | |
3586 | if (found_key.objectid != objectid) | |
3587 | return 0; | |
3588 | ||
3589 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3590 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3591 | if (*first_xattr_slot == -1) |
3592 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3593 | if (found_key.offset == xattr_access || |
3594 | found_key.offset == xattr_default) | |
3595 | return 1; | |
3596 | } | |
46a53cca CM |
3597 | |
3598 | /* | |
3599 | * we found a key greater than an xattr key, there can't | |
3600 | * be any acls later on | |
3601 | */ | |
3602 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3603 | return 0; | |
3604 | ||
3605 | slot++; | |
3606 | scanned++; | |
3607 | ||
3608 | /* | |
3609 | * it goes inode, inode backrefs, xattrs, extents, | |
3610 | * so if there are a ton of hard links to an inode there can | |
3611 | * be a lot of backrefs. Don't waste time searching too hard, | |
3612 | * this is just an optimization | |
3613 | */ | |
3614 | if (scanned >= 8) | |
3615 | break; | |
3616 | } | |
3617 | /* we hit the end of the leaf before we found an xattr or | |
3618 | * something larger than an xattr. We have to assume the inode | |
3619 | * has acls | |
3620 | */ | |
63541927 FDBM |
3621 | if (*first_xattr_slot == -1) |
3622 | *first_xattr_slot = slot; | |
46a53cca CM |
3623 | return 1; |
3624 | } | |
3625 | ||
d352ac68 CM |
3626 | /* |
3627 | * read an inode from the btree into the in-memory inode | |
3628 | */ | |
5d4f98a2 | 3629 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
3630 | { |
3631 | struct btrfs_path *path; | |
5f39d397 | 3632 | struct extent_buffer *leaf; |
39279cc3 CM |
3633 | struct btrfs_inode_item *inode_item; |
3634 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3635 | struct btrfs_key location; | |
67de1176 | 3636 | unsigned long ptr; |
46a53cca | 3637 | int maybe_acls; |
618e21d5 | 3638 | u32 rdev; |
39279cc3 | 3639 | int ret; |
2f7e33d4 | 3640 | bool filled = false; |
63541927 | 3641 | int first_xattr_slot; |
2f7e33d4 MX |
3642 | |
3643 | ret = btrfs_fill_inode(inode, &rdev); | |
3644 | if (!ret) | |
3645 | filled = true; | |
39279cc3 CM |
3646 | |
3647 | path = btrfs_alloc_path(); | |
1748f843 MF |
3648 | if (!path) |
3649 | goto make_bad; | |
3650 | ||
39279cc3 | 3651 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3652 | |
39279cc3 | 3653 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 3654 | if (ret) |
39279cc3 | 3655 | goto make_bad; |
39279cc3 | 3656 | |
5f39d397 | 3657 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3658 | |
3659 | if (filled) | |
67de1176 | 3660 | goto cache_index; |
2f7e33d4 | 3661 | |
5f39d397 CM |
3662 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3663 | struct btrfs_inode_item); | |
5f39d397 | 3664 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3665 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3666 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3667 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
dbe674a9 | 3668 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 | 3669 | |
a937b979 DS |
3670 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3671 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3672 | |
a937b979 DS |
3673 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3674 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3675 | |
a937b979 DS |
3676 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3677 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3678 | |
9cc97d64 | 3679 | BTRFS_I(inode)->i_otime.tv_sec = |
3680 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3681 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3682 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3683 | |
a76a3cd4 | 3684 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3685 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3686 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3687 | ||
6e17d30b YD |
3688 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
3689 | inode->i_generation = BTRFS_I(inode)->generation; | |
3690 | inode->i_rdev = 0; | |
3691 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3692 | ||
3693 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3694 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3695 | ||
3696 | cache_index: | |
5dc562c5 JB |
3697 | /* |
3698 | * If we were modified in the current generation and evicted from memory | |
3699 | * and then re-read we need to do a full sync since we don't have any | |
3700 | * idea about which extents were modified before we were evicted from | |
3701 | * cache. | |
6e17d30b YD |
3702 | * |
3703 | * This is required for both inode re-read from disk and delayed inode | |
3704 | * in delayed_nodes_tree. | |
5dc562c5 JB |
3705 | */ |
3706 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
3707 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3708 | &BTRFS_I(inode)->runtime_flags); | |
3709 | ||
bde6c242 FM |
3710 | /* |
3711 | * We don't persist the id of the transaction where an unlink operation | |
3712 | * against the inode was last made. So here we assume the inode might | |
3713 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3714 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3715 | * between the inode and its parent if the inode is fsync'ed and the log | |
3716 | * replayed. For example, in the scenario: | |
3717 | * | |
3718 | * touch mydir/foo | |
3719 | * ln mydir/foo mydir/bar | |
3720 | * sync | |
3721 | * unlink mydir/bar | |
3722 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3723 | * xfs_io -c fsync mydir/foo | |
3724 | * <power failure> | |
3725 | * mount fs, triggers fsync log replay | |
3726 | * | |
3727 | * We must make sure that when we fsync our inode foo we also log its | |
3728 | * parent inode, otherwise after log replay the parent still has the | |
3729 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3730 | * and doesn't have an inode ref with the name "bar" anymore. | |
3731 | * | |
3732 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3733 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3734 | * transaction commits on fsync if our inode is a directory, or if our |
3735 | * inode is not a directory, logging its parent unnecessarily. | |
3736 | */ | |
3737 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3738 | ||
67de1176 MX |
3739 | path->slots[0]++; |
3740 | if (inode->i_nlink != 1 || | |
3741 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3742 | goto cache_acl; | |
3743 | ||
3744 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
3745 | if (location.objectid != btrfs_ino(inode)) | |
3746 | goto cache_acl; | |
3747 | ||
3748 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3749 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3750 | struct btrfs_inode_ref *ref; | |
3751 | ||
3752 | ref = (struct btrfs_inode_ref *)ptr; | |
3753 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3754 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3755 | struct btrfs_inode_extref *extref; | |
3756 | ||
3757 | extref = (struct btrfs_inode_extref *)ptr; | |
3758 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3759 | extref); | |
3760 | } | |
2f7e33d4 | 3761 | cache_acl: |
46a53cca CM |
3762 | /* |
3763 | * try to precache a NULL acl entry for files that don't have | |
3764 | * any xattrs or acls | |
3765 | */ | |
33345d01 | 3766 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
63541927 FDBM |
3767 | btrfs_ino(inode), &first_xattr_slot); |
3768 | if (first_xattr_slot != -1) { | |
3769 | path->slots[0] = first_xattr_slot; | |
3770 | ret = btrfs_load_inode_props(inode, path); | |
3771 | if (ret) | |
3772 | btrfs_err(root->fs_info, | |
351fd353 | 3773 | "error loading props for ino %llu (root %llu): %d", |
63541927 FDBM |
3774 | btrfs_ino(inode), |
3775 | root->root_key.objectid, ret); | |
3776 | } | |
3777 | btrfs_free_path(path); | |
3778 | ||
72c04902 AV |
3779 | if (!maybe_acls) |
3780 | cache_no_acl(inode); | |
46a53cca | 3781 | |
39279cc3 | 3782 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3783 | case S_IFREG: |
3784 | inode->i_mapping->a_ops = &btrfs_aops; | |
d1310b2e | 3785 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3786 | inode->i_fop = &btrfs_file_operations; |
3787 | inode->i_op = &btrfs_file_inode_operations; | |
3788 | break; | |
3789 | case S_IFDIR: | |
3790 | inode->i_fop = &btrfs_dir_file_operations; | |
3791 | if (root == root->fs_info->tree_root) | |
3792 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
3793 | else | |
3794 | inode->i_op = &btrfs_dir_inode_operations; | |
3795 | break; | |
3796 | case S_IFLNK: | |
3797 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3798 | inode_nohighmem(inode); |
39279cc3 CM |
3799 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
3800 | break; | |
618e21d5 | 3801 | default: |
0279b4cd | 3802 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3803 | init_special_inode(inode, inode->i_mode, rdev); |
3804 | break; | |
39279cc3 | 3805 | } |
6cbff00f CH |
3806 | |
3807 | btrfs_update_iflags(inode); | |
39279cc3 CM |
3808 | return; |
3809 | ||
3810 | make_bad: | |
39279cc3 | 3811 | btrfs_free_path(path); |
39279cc3 CM |
3812 | make_bad_inode(inode); |
3813 | } | |
3814 | ||
d352ac68 CM |
3815 | /* |
3816 | * given a leaf and an inode, copy the inode fields into the leaf | |
3817 | */ | |
e02119d5 CM |
3818 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3819 | struct extent_buffer *leaf, | |
5f39d397 | 3820 | struct btrfs_inode_item *item, |
39279cc3 CM |
3821 | struct inode *inode) |
3822 | { | |
51fab693 LB |
3823 | struct btrfs_map_token token; |
3824 | ||
3825 | btrfs_init_map_token(&token); | |
5f39d397 | 3826 | |
51fab693 LB |
3827 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3828 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3829 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3830 | &token); | |
3831 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3832 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3833 | |
a937b979 | 3834 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
51fab693 | 3835 | inode->i_atime.tv_sec, &token); |
a937b979 | 3836 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
51fab693 | 3837 | inode->i_atime.tv_nsec, &token); |
5f39d397 | 3838 | |
a937b979 | 3839 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
51fab693 | 3840 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3841 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
51fab693 | 3842 | inode->i_mtime.tv_nsec, &token); |
5f39d397 | 3843 | |
a937b979 | 3844 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
51fab693 | 3845 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3846 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
51fab693 | 3847 | inode->i_ctime.tv_nsec, &token); |
5f39d397 | 3848 | |
9cc97d64 | 3849 | btrfs_set_token_timespec_sec(leaf, &item->otime, |
3850 | BTRFS_I(inode)->i_otime.tv_sec, &token); | |
3851 | btrfs_set_token_timespec_nsec(leaf, &item->otime, | |
3852 | BTRFS_I(inode)->i_otime.tv_nsec, &token); | |
3853 | ||
51fab693 LB |
3854 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3855 | &token); | |
3856 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3857 | &token); | |
3858 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3859 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3860 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3861 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3862 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3863 | } |
3864 | ||
d352ac68 CM |
3865 | /* |
3866 | * copy everything in the in-memory inode into the btree. | |
3867 | */ | |
2115133f | 3868 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3869 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3870 | { |
3871 | struct btrfs_inode_item *inode_item; | |
3872 | struct btrfs_path *path; | |
5f39d397 | 3873 | struct extent_buffer *leaf; |
39279cc3 CM |
3874 | int ret; |
3875 | ||
3876 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3877 | if (!path) |
3878 | return -ENOMEM; | |
3879 | ||
b9473439 | 3880 | path->leave_spinning = 1; |
16cdcec7 MX |
3881 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3882 | 1); | |
39279cc3 CM |
3883 | if (ret) { |
3884 | if (ret > 0) | |
3885 | ret = -ENOENT; | |
3886 | goto failed; | |
3887 | } | |
3888 | ||
5f39d397 CM |
3889 | leaf = path->nodes[0]; |
3890 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3891 | struct btrfs_inode_item); |
39279cc3 | 3892 | |
e02119d5 | 3893 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3894 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3895 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3896 | ret = 0; |
3897 | failed: | |
39279cc3 CM |
3898 | btrfs_free_path(path); |
3899 | return ret; | |
3900 | } | |
3901 | ||
2115133f CM |
3902 | /* |
3903 | * copy everything in the in-memory inode into the btree. | |
3904 | */ | |
3905 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3906 | struct btrfs_root *root, struct inode *inode) | |
3907 | { | |
3908 | int ret; | |
3909 | ||
3910 | /* | |
3911 | * If the inode is a free space inode, we can deadlock during commit | |
3912 | * if we put it into the delayed code. | |
3913 | * | |
3914 | * The data relocation inode should also be directly updated | |
3915 | * without delay | |
3916 | */ | |
83eea1f1 | 3917 | if (!btrfs_is_free_space_inode(inode) |
1d52c78a JB |
3918 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
3919 | && !root->fs_info->log_root_recovering) { | |
8ea05e3a AB |
3920 | btrfs_update_root_times(trans, root); |
3921 | ||
2115133f CM |
3922 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3923 | if (!ret) | |
3924 | btrfs_set_inode_last_trans(trans, inode); | |
3925 | return ret; | |
3926 | } | |
3927 | ||
3928 | return btrfs_update_inode_item(trans, root, inode); | |
3929 | } | |
3930 | ||
be6aef60 JB |
3931 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3932 | struct btrfs_root *root, | |
3933 | struct inode *inode) | |
2115133f CM |
3934 | { |
3935 | int ret; | |
3936 | ||
3937 | ret = btrfs_update_inode(trans, root, inode); | |
3938 | if (ret == -ENOSPC) | |
3939 | return btrfs_update_inode_item(trans, root, inode); | |
3940 | return ret; | |
3941 | } | |
3942 | ||
d352ac68 CM |
3943 | /* |
3944 | * unlink helper that gets used here in inode.c and in the tree logging | |
3945 | * recovery code. It remove a link in a directory with a given name, and | |
3946 | * also drops the back refs in the inode to the directory | |
3947 | */ | |
92986796 AV |
3948 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3949 | struct btrfs_root *root, | |
3950 | struct inode *dir, struct inode *inode, | |
3951 | const char *name, int name_len) | |
39279cc3 CM |
3952 | { |
3953 | struct btrfs_path *path; | |
39279cc3 | 3954 | int ret = 0; |
5f39d397 | 3955 | struct extent_buffer *leaf; |
39279cc3 | 3956 | struct btrfs_dir_item *di; |
5f39d397 | 3957 | struct btrfs_key key; |
aec7477b | 3958 | u64 index; |
33345d01 LZ |
3959 | u64 ino = btrfs_ino(inode); |
3960 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3961 | |
3962 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3963 | if (!path) { |
3964 | ret = -ENOMEM; | |
554233a6 | 3965 | goto out; |
54aa1f4d CM |
3966 | } |
3967 | ||
b9473439 | 3968 | path->leave_spinning = 1; |
33345d01 | 3969 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
3970 | name, name_len, -1); |
3971 | if (IS_ERR(di)) { | |
3972 | ret = PTR_ERR(di); | |
3973 | goto err; | |
3974 | } | |
3975 | if (!di) { | |
3976 | ret = -ENOENT; | |
3977 | goto err; | |
3978 | } | |
5f39d397 CM |
3979 | leaf = path->nodes[0]; |
3980 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 3981 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3982 | if (ret) |
3983 | goto err; | |
b3b4aa74 | 3984 | btrfs_release_path(path); |
39279cc3 | 3985 | |
67de1176 MX |
3986 | /* |
3987 | * If we don't have dir index, we have to get it by looking up | |
3988 | * the inode ref, since we get the inode ref, remove it directly, | |
3989 | * it is unnecessary to do delayed deletion. | |
3990 | * | |
3991 | * But if we have dir index, needn't search inode ref to get it. | |
3992 | * Since the inode ref is close to the inode item, it is better | |
3993 | * that we delay to delete it, and just do this deletion when | |
3994 | * we update the inode item. | |
3995 | */ | |
3996 | if (BTRFS_I(inode)->dir_index) { | |
3997 | ret = btrfs_delayed_delete_inode_ref(inode); | |
3998 | if (!ret) { | |
3999 | index = BTRFS_I(inode)->dir_index; | |
4000 | goto skip_backref; | |
4001 | } | |
4002 | } | |
4003 | ||
33345d01 LZ |
4004 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
4005 | dir_ino, &index); | |
aec7477b | 4006 | if (ret) { |
c2cf52eb SK |
4007 | btrfs_info(root->fs_info, |
4008 | "failed to delete reference to %.*s, inode %llu parent %llu", | |
c1c9ff7c | 4009 | name_len, name, ino, dir_ino); |
79787eaa | 4010 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
4011 | goto err; |
4012 | } | |
67de1176 | 4013 | skip_backref: |
16cdcec7 | 4014 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
4015 | if (ret) { |
4016 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 4017 | goto err; |
79787eaa | 4018 | } |
39279cc3 | 4019 | |
e02119d5 | 4020 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 4021 | inode, dir_ino); |
79787eaa JM |
4022 | if (ret != 0 && ret != -ENOENT) { |
4023 | btrfs_abort_transaction(trans, root, ret); | |
4024 | goto err; | |
4025 | } | |
e02119d5 CM |
4026 | |
4027 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
4028 | dir, index); | |
6418c961 CM |
4029 | if (ret == -ENOENT) |
4030 | ret = 0; | |
d4e3991b ZB |
4031 | else if (ret) |
4032 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 CM |
4033 | err: |
4034 | btrfs_free_path(path); | |
e02119d5 CM |
4035 | if (ret) |
4036 | goto out; | |
4037 | ||
4038 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
4039 | inode_inc_iversion(inode); |
4040 | inode_inc_iversion(dir); | |
04b285f3 DD |
4041 | inode->i_ctime = dir->i_mtime = |
4042 | dir->i_ctime = current_fs_time(inode->i_sb); | |
b9959295 | 4043 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 4044 | out: |
39279cc3 CM |
4045 | return ret; |
4046 | } | |
4047 | ||
92986796 AV |
4048 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4049 | struct btrfs_root *root, | |
4050 | struct inode *dir, struct inode *inode, | |
4051 | const char *name, int name_len) | |
4052 | { | |
4053 | int ret; | |
4054 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
4055 | if (!ret) { | |
8b558c5f | 4056 | drop_nlink(inode); |
92986796 AV |
4057 | ret = btrfs_update_inode(trans, root, inode); |
4058 | } | |
4059 | return ret; | |
4060 | } | |
39279cc3 | 4061 | |
a22285a6 YZ |
4062 | /* |
4063 | * helper to start transaction for unlink and rmdir. | |
4064 | * | |
d52be818 JB |
4065 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4066 | * if we cannot make our reservations the normal way try and see if there is | |
4067 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4068 | * allow the unlink to occur. | |
a22285a6 | 4069 | */ |
d52be818 | 4070 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4071 | { |
a22285a6 | 4072 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 4073 | |
e70bea5f JB |
4074 | /* |
4075 | * 1 for the possible orphan item | |
4076 | * 1 for the dir item | |
4077 | * 1 for the dir index | |
4078 | * 1 for the inode ref | |
e70bea5f JB |
4079 | * 1 for the inode |
4080 | */ | |
8eab77ff | 4081 | return btrfs_start_transaction_fallback_global_rsv(root, 5, 5); |
a22285a6 YZ |
4082 | } |
4083 | ||
4084 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4085 | { | |
4086 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4087 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 4088 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4089 | int ret; |
a22285a6 | 4090 | |
d52be818 | 4091 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4092 | if (IS_ERR(trans)) |
4093 | return PTR_ERR(trans); | |
5f39d397 | 4094 | |
2b0143b5 | 4095 | btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0); |
12fcfd22 | 4096 | |
2b0143b5 | 4097 | ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry), |
e02119d5 | 4098 | dentry->d_name.name, dentry->d_name.len); |
b532402e TI |
4099 | if (ret) |
4100 | goto out; | |
7b128766 | 4101 | |
a22285a6 | 4102 | if (inode->i_nlink == 0) { |
7b128766 | 4103 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
4104 | if (ret) |
4105 | goto out; | |
a22285a6 | 4106 | } |
7b128766 | 4107 | |
b532402e | 4108 | out: |
d52be818 | 4109 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4110 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
4111 | return ret; |
4112 | } | |
4113 | ||
4df27c4d YZ |
4114 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
4115 | struct btrfs_root *root, | |
4116 | struct inode *dir, u64 objectid, | |
4117 | const char *name, int name_len) | |
4118 | { | |
4119 | struct btrfs_path *path; | |
4120 | struct extent_buffer *leaf; | |
4121 | struct btrfs_dir_item *di; | |
4122 | struct btrfs_key key; | |
4123 | u64 index; | |
4124 | int ret; | |
33345d01 | 4125 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
4126 | |
4127 | path = btrfs_alloc_path(); | |
4128 | if (!path) | |
4129 | return -ENOMEM; | |
4130 | ||
33345d01 | 4131 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4132 | name, name_len, -1); |
79787eaa JM |
4133 | if (IS_ERR_OR_NULL(di)) { |
4134 | if (!di) | |
4135 | ret = -ENOENT; | |
4136 | else | |
4137 | ret = PTR_ERR(di); | |
4138 | goto out; | |
4139 | } | |
4df27c4d YZ |
4140 | |
4141 | leaf = path->nodes[0]; | |
4142 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4143 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4144 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
4145 | if (ret) { |
4146 | btrfs_abort_transaction(trans, root, ret); | |
4147 | goto out; | |
4148 | } | |
b3b4aa74 | 4149 | btrfs_release_path(path); |
4df27c4d YZ |
4150 | |
4151 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
4152 | objectid, root->root_key.objectid, | |
33345d01 | 4153 | dir_ino, &index, name, name_len); |
4df27c4d | 4154 | if (ret < 0) { |
79787eaa JM |
4155 | if (ret != -ENOENT) { |
4156 | btrfs_abort_transaction(trans, root, ret); | |
4157 | goto out; | |
4158 | } | |
33345d01 | 4159 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4160 | name, name_len); |
79787eaa JM |
4161 | if (IS_ERR_OR_NULL(di)) { |
4162 | if (!di) | |
4163 | ret = -ENOENT; | |
4164 | else | |
4165 | ret = PTR_ERR(di); | |
4166 | btrfs_abort_transaction(trans, root, ret); | |
4167 | goto out; | |
4168 | } | |
4df27c4d YZ |
4169 | |
4170 | leaf = path->nodes[0]; | |
4171 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 4172 | btrfs_release_path(path); |
4df27c4d YZ |
4173 | index = key.offset; |
4174 | } | |
945d8962 | 4175 | btrfs_release_path(path); |
4df27c4d | 4176 | |
16cdcec7 | 4177 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
4178 | if (ret) { |
4179 | btrfs_abort_transaction(trans, root, ret); | |
4180 | goto out; | |
4181 | } | |
4df27c4d YZ |
4182 | |
4183 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 4184 | inode_inc_iversion(dir); |
04b285f3 | 4185 | dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb); |
5a24e84c | 4186 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
4187 | if (ret) |
4188 | btrfs_abort_transaction(trans, root, ret); | |
4189 | out: | |
71d7aed0 | 4190 | btrfs_free_path(path); |
79787eaa | 4191 | return ret; |
4df27c4d YZ |
4192 | } |
4193 | ||
39279cc3 CM |
4194 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4195 | { | |
2b0143b5 | 4196 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4197 | int err = 0; |
39279cc3 | 4198 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4199 | struct btrfs_trans_handle *trans; |
39279cc3 | 4200 | |
b3ae244e | 4201 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4202 | return -ENOTEMPTY; |
b3ae244e DS |
4203 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
4204 | return -EPERM; | |
134d4512 | 4205 | |
d52be818 | 4206 | trans = __unlink_start_trans(dir); |
a22285a6 | 4207 | if (IS_ERR(trans)) |
5df6a9f6 | 4208 | return PTR_ERR(trans); |
5df6a9f6 | 4209 | |
33345d01 | 4210 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
4211 | err = btrfs_unlink_subvol(trans, root, dir, |
4212 | BTRFS_I(inode)->location.objectid, | |
4213 | dentry->d_name.name, | |
4214 | dentry->d_name.len); | |
4215 | goto out; | |
4216 | } | |
4217 | ||
7b128766 JB |
4218 | err = btrfs_orphan_add(trans, inode); |
4219 | if (err) | |
4df27c4d | 4220 | goto out; |
7b128766 | 4221 | |
39279cc3 | 4222 | /* now the directory is empty */ |
2b0143b5 | 4223 | err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry), |
e02119d5 | 4224 | dentry->d_name.name, dentry->d_name.len); |
d397712b | 4225 | if (!err) |
dbe674a9 | 4226 | btrfs_i_size_write(inode, 0); |
4df27c4d | 4227 | out: |
d52be818 | 4228 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4229 | btrfs_btree_balance_dirty(root); |
3954401f | 4230 | |
39279cc3 CM |
4231 | return err; |
4232 | } | |
4233 | ||
28f75a0e CM |
4234 | static int truncate_space_check(struct btrfs_trans_handle *trans, |
4235 | struct btrfs_root *root, | |
4236 | u64 bytes_deleted) | |
4237 | { | |
4238 | int ret; | |
4239 | ||
dc95f7bf JB |
4240 | /* |
4241 | * This is only used to apply pressure to the enospc system, we don't | |
4242 | * intend to use this reservation at all. | |
4243 | */ | |
28f75a0e | 4244 | bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted); |
dc95f7bf | 4245 | bytes_deleted *= root->nodesize; |
28f75a0e CM |
4246 | ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv, |
4247 | bytes_deleted, BTRFS_RESERVE_NO_FLUSH); | |
dc95f7bf JB |
4248 | if (!ret) { |
4249 | trace_btrfs_space_reservation(root->fs_info, "transaction", | |
4250 | trans->transid, | |
4251 | bytes_deleted, 1); | |
28f75a0e | 4252 | trans->bytes_reserved += bytes_deleted; |
dc95f7bf | 4253 | } |
28f75a0e CM |
4254 | return ret; |
4255 | ||
4256 | } | |
4257 | ||
0305cd5f FM |
4258 | static int truncate_inline_extent(struct inode *inode, |
4259 | struct btrfs_path *path, | |
4260 | struct btrfs_key *found_key, | |
4261 | const u64 item_end, | |
4262 | const u64 new_size) | |
4263 | { | |
4264 | struct extent_buffer *leaf = path->nodes[0]; | |
4265 | int slot = path->slots[0]; | |
4266 | struct btrfs_file_extent_item *fi; | |
4267 | u32 size = (u32)(new_size - found_key->offset); | |
4268 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4269 | ||
4270 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
4271 | ||
4272 | if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) { | |
4273 | loff_t offset = new_size; | |
09cbfeaf | 4274 | loff_t page_end = ALIGN(offset, PAGE_SIZE); |
0305cd5f FM |
4275 | |
4276 | /* | |
4277 | * Zero out the remaining of the last page of our inline extent, | |
4278 | * instead of directly truncating our inline extent here - that | |
4279 | * would be much more complex (decompressing all the data, then | |
4280 | * compressing the truncated data, which might be bigger than | |
4281 | * the size of the inline extent, resize the extent, etc). | |
4282 | * We release the path because to get the page we might need to | |
4283 | * read the extent item from disk (data not in the page cache). | |
4284 | */ | |
4285 | btrfs_release_path(path); | |
9703fefe CR |
4286 | return btrfs_truncate_block(inode, offset, page_end - offset, |
4287 | 0); | |
0305cd5f FM |
4288 | } |
4289 | ||
4290 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4291 | size = btrfs_file_extent_calc_inline_size(size); | |
4292 | btrfs_truncate_item(root, path, size, 1); | |
4293 | ||
4294 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) | |
4295 | inode_sub_bytes(inode, item_end + 1 - new_size); | |
4296 | ||
4297 | return 0; | |
4298 | } | |
4299 | ||
39279cc3 CM |
4300 | /* |
4301 | * this can truncate away extent items, csum items and directory items. | |
4302 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4303 | * any higher than new_size |
39279cc3 CM |
4304 | * |
4305 | * csum items that cross the new i_size are truncated to the new size | |
4306 | * as well. | |
7b128766 JB |
4307 | * |
4308 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4309 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4310 | */ |
8082510e YZ |
4311 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4312 | struct btrfs_root *root, | |
4313 | struct inode *inode, | |
4314 | u64 new_size, u32 min_type) | |
39279cc3 | 4315 | { |
39279cc3 | 4316 | struct btrfs_path *path; |
5f39d397 | 4317 | struct extent_buffer *leaf; |
39279cc3 | 4318 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4319 | struct btrfs_key key; |
4320 | struct btrfs_key found_key; | |
39279cc3 | 4321 | u64 extent_start = 0; |
db94535d | 4322 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4323 | u64 extent_offset = 0; |
39279cc3 | 4324 | u64 item_end = 0; |
c1aa4575 | 4325 | u64 last_size = new_size; |
8082510e | 4326 | u32 found_type = (u8)-1; |
39279cc3 CM |
4327 | int found_extent; |
4328 | int del_item; | |
85e21bac CM |
4329 | int pending_del_nr = 0; |
4330 | int pending_del_slot = 0; | |
179e29e4 | 4331 | int extent_type = -1; |
8082510e YZ |
4332 | int ret; |
4333 | int err = 0; | |
33345d01 | 4334 | u64 ino = btrfs_ino(inode); |
28ed1345 | 4335 | u64 bytes_deleted = 0; |
1262133b JB |
4336 | bool be_nice = 0; |
4337 | bool should_throttle = 0; | |
28f75a0e | 4338 | bool should_end = 0; |
8082510e YZ |
4339 | |
4340 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4341 | |
28ed1345 CM |
4342 | /* |
4343 | * for non-free space inodes and ref cows, we want to back off from | |
4344 | * time to time | |
4345 | */ | |
4346 | if (!btrfs_is_free_space_inode(inode) && | |
4347 | test_bit(BTRFS_ROOT_REF_COWS, &root->state)) | |
4348 | be_nice = 1; | |
4349 | ||
0eb0e19c MF |
4350 | path = btrfs_alloc_path(); |
4351 | if (!path) | |
4352 | return -ENOMEM; | |
e4058b54 | 4353 | path->reada = READA_BACK; |
0eb0e19c | 4354 | |
5dc562c5 JB |
4355 | /* |
4356 | * We want to drop from the next block forward in case this new size is | |
4357 | * not block aligned since we will be keeping the last block of the | |
4358 | * extent just the way it is. | |
4359 | */ | |
27cdeb70 MX |
4360 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
4361 | root == root->fs_info->tree_root) | |
fda2832f QW |
4362 | btrfs_drop_extent_cache(inode, ALIGN(new_size, |
4363 | root->sectorsize), (u64)-1, 0); | |
8082510e | 4364 | |
16cdcec7 MX |
4365 | /* |
4366 | * This function is also used to drop the items in the log tree before | |
4367 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
4368 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
4369 | * items. | |
4370 | */ | |
4371 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4372 | btrfs_kill_delayed_inode_items(inode); | |
4373 | ||
33345d01 | 4374 | key.objectid = ino; |
39279cc3 | 4375 | key.offset = (u64)-1; |
5f39d397 CM |
4376 | key.type = (u8)-1; |
4377 | ||
85e21bac | 4378 | search_again: |
28ed1345 CM |
4379 | /* |
4380 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4381 | * up a huge file in a single leaf. Most of the time that | |
4382 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4383 | */ | |
ee22184b | 4384 | if (be_nice && bytes_deleted > SZ_32M) { |
28ed1345 CM |
4385 | if (btrfs_should_end_transaction(trans, root)) { |
4386 | err = -EAGAIN; | |
4387 | goto error; | |
4388 | } | |
4389 | } | |
4390 | ||
4391 | ||
b9473439 | 4392 | path->leave_spinning = 1; |
85e21bac | 4393 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
4394 | if (ret < 0) { |
4395 | err = ret; | |
4396 | goto out; | |
4397 | } | |
d397712b | 4398 | |
85e21bac | 4399 | if (ret > 0) { |
e02119d5 CM |
4400 | /* there are no items in the tree for us to truncate, we're |
4401 | * done | |
4402 | */ | |
8082510e YZ |
4403 | if (path->slots[0] == 0) |
4404 | goto out; | |
85e21bac CM |
4405 | path->slots[0]--; |
4406 | } | |
4407 | ||
d397712b | 4408 | while (1) { |
39279cc3 | 4409 | fi = NULL; |
5f39d397 CM |
4410 | leaf = path->nodes[0]; |
4411 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4412 | found_type = found_key.type; |
39279cc3 | 4413 | |
33345d01 | 4414 | if (found_key.objectid != ino) |
39279cc3 | 4415 | break; |
5f39d397 | 4416 | |
85e21bac | 4417 | if (found_type < min_type) |
39279cc3 CM |
4418 | break; |
4419 | ||
5f39d397 | 4420 | item_end = found_key.offset; |
39279cc3 | 4421 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4422 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4423 | struct btrfs_file_extent_item); |
179e29e4 CM |
4424 | extent_type = btrfs_file_extent_type(leaf, fi); |
4425 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4426 | item_end += |
db94535d | 4427 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 4428 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 4429 | item_end += btrfs_file_extent_inline_len(leaf, |
514ac8ad | 4430 | path->slots[0], fi); |
39279cc3 | 4431 | } |
008630c1 | 4432 | item_end--; |
39279cc3 | 4433 | } |
8082510e YZ |
4434 | if (found_type > min_type) { |
4435 | del_item = 1; | |
4436 | } else { | |
4437 | if (item_end < new_size) | |
b888db2b | 4438 | break; |
8082510e YZ |
4439 | if (found_key.offset >= new_size) |
4440 | del_item = 1; | |
4441 | else | |
4442 | del_item = 0; | |
39279cc3 | 4443 | } |
39279cc3 | 4444 | found_extent = 0; |
39279cc3 | 4445 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4446 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4447 | goto delete; | |
4448 | ||
7f4f6e0a JB |
4449 | if (del_item) |
4450 | last_size = found_key.offset; | |
4451 | else | |
4452 | last_size = new_size; | |
4453 | ||
179e29e4 | 4454 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
39279cc3 | 4455 | u64 num_dec; |
db94535d | 4456 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4457 | if (!del_item) { |
db94535d CM |
4458 | u64 orig_num_bytes = |
4459 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4460 | extent_num_bytes = ALIGN(new_size - |
4461 | found_key.offset, | |
4462 | root->sectorsize); | |
db94535d CM |
4463 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4464 | extent_num_bytes); | |
4465 | num_dec = (orig_num_bytes - | |
9069218d | 4466 | extent_num_bytes); |
27cdeb70 MX |
4467 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4468 | &root->state) && | |
4469 | extent_start != 0) | |
a76a3cd4 | 4470 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4471 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4472 | } else { |
db94535d CM |
4473 | extent_num_bytes = |
4474 | btrfs_file_extent_disk_num_bytes(leaf, | |
4475 | fi); | |
5d4f98a2 YZ |
4476 | extent_offset = found_key.offset - |
4477 | btrfs_file_extent_offset(leaf, fi); | |
4478 | ||
39279cc3 | 4479 | /* FIXME blocksize != 4096 */ |
9069218d | 4480 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4481 | if (extent_start != 0) { |
4482 | found_extent = 1; | |
27cdeb70 MX |
4483 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4484 | &root->state)) | |
a76a3cd4 | 4485 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4486 | } |
39279cc3 | 4487 | } |
9069218d | 4488 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4489 | /* |
4490 | * we can't truncate inline items that have had | |
4491 | * special encodings | |
4492 | */ | |
4493 | if (!del_item && | |
c8b97818 CM |
4494 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
4495 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
514ac8ad CM |
4496 | |
4497 | /* | |
0305cd5f FM |
4498 | * Need to release path in order to truncate a |
4499 | * compressed extent. So delete any accumulated | |
4500 | * extent items so far. | |
514ac8ad | 4501 | */ |
0305cd5f FM |
4502 | if (btrfs_file_extent_compression(leaf, fi) != |
4503 | BTRFS_COMPRESS_NONE && pending_del_nr) { | |
4504 | err = btrfs_del_items(trans, root, path, | |
4505 | pending_del_slot, | |
4506 | pending_del_nr); | |
4507 | if (err) { | |
4508 | btrfs_abort_transaction(trans, | |
4509 | root, | |
4510 | err); | |
4511 | goto error; | |
4512 | } | |
4513 | pending_del_nr = 0; | |
4514 | } | |
4515 | ||
4516 | err = truncate_inline_extent(inode, path, | |
4517 | &found_key, | |
4518 | item_end, | |
4519 | new_size); | |
4520 | if (err) { | |
4521 | btrfs_abort_transaction(trans, | |
4522 | root, err); | |
4523 | goto error; | |
4524 | } | |
27cdeb70 MX |
4525 | } else if (test_bit(BTRFS_ROOT_REF_COWS, |
4526 | &root->state)) { | |
0305cd5f | 4527 | inode_sub_bytes(inode, item_end + 1 - new_size); |
9069218d | 4528 | } |
39279cc3 | 4529 | } |
179e29e4 | 4530 | delete: |
39279cc3 | 4531 | if (del_item) { |
85e21bac CM |
4532 | if (!pending_del_nr) { |
4533 | /* no pending yet, add ourselves */ | |
4534 | pending_del_slot = path->slots[0]; | |
4535 | pending_del_nr = 1; | |
4536 | } else if (pending_del_nr && | |
4537 | path->slots[0] + 1 == pending_del_slot) { | |
4538 | /* hop on the pending chunk */ | |
4539 | pending_del_nr++; | |
4540 | pending_del_slot = path->slots[0]; | |
4541 | } else { | |
d397712b | 4542 | BUG(); |
85e21bac | 4543 | } |
39279cc3 CM |
4544 | } else { |
4545 | break; | |
4546 | } | |
28f75a0e CM |
4547 | should_throttle = 0; |
4548 | ||
27cdeb70 MX |
4549 | if (found_extent && |
4550 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
4551 | root == root->fs_info->tree_root)) { | |
b9473439 | 4552 | btrfs_set_path_blocking(path); |
28ed1345 | 4553 | bytes_deleted += extent_num_bytes; |
39279cc3 | 4554 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
4555 | extent_num_bytes, 0, |
4556 | btrfs_header_owner(leaf), | |
b06c4bf5 | 4557 | ino, extent_offset); |
39279cc3 | 4558 | BUG_ON(ret); |
1262133b | 4559 | if (btrfs_should_throttle_delayed_refs(trans, root)) |
28ed1345 | 4560 | btrfs_async_run_delayed_refs(root, |
31b9655f | 4561 | trans->transid, |
28ed1345 | 4562 | trans->delayed_ref_updates * 2, 0); |
28f75a0e CM |
4563 | if (be_nice) { |
4564 | if (truncate_space_check(trans, root, | |
4565 | extent_num_bytes)) { | |
4566 | should_end = 1; | |
4567 | } | |
4568 | if (btrfs_should_throttle_delayed_refs(trans, | |
4569 | root)) { | |
4570 | should_throttle = 1; | |
4571 | } | |
4572 | } | |
39279cc3 | 4573 | } |
85e21bac | 4574 | |
8082510e YZ |
4575 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4576 | break; | |
4577 | ||
4578 | if (path->slots[0] == 0 || | |
1262133b | 4579 | path->slots[0] != pending_del_slot || |
28f75a0e | 4580 | should_throttle || should_end) { |
8082510e YZ |
4581 | if (pending_del_nr) { |
4582 | ret = btrfs_del_items(trans, root, path, | |
4583 | pending_del_slot, | |
4584 | pending_del_nr); | |
79787eaa JM |
4585 | if (ret) { |
4586 | btrfs_abort_transaction(trans, | |
4587 | root, ret); | |
4588 | goto error; | |
4589 | } | |
8082510e YZ |
4590 | pending_del_nr = 0; |
4591 | } | |
b3b4aa74 | 4592 | btrfs_release_path(path); |
28f75a0e | 4593 | if (should_throttle) { |
1262133b JB |
4594 | unsigned long updates = trans->delayed_ref_updates; |
4595 | if (updates) { | |
4596 | trans->delayed_ref_updates = 0; | |
4597 | ret = btrfs_run_delayed_refs(trans, root, updates * 2); | |
4598 | if (ret && !err) | |
4599 | err = ret; | |
4600 | } | |
4601 | } | |
28f75a0e CM |
4602 | /* |
4603 | * if we failed to refill our space rsv, bail out | |
4604 | * and let the transaction restart | |
4605 | */ | |
4606 | if (should_end) { | |
4607 | err = -EAGAIN; | |
4608 | goto error; | |
4609 | } | |
85e21bac | 4610 | goto search_again; |
8082510e YZ |
4611 | } else { |
4612 | path->slots[0]--; | |
85e21bac | 4613 | } |
39279cc3 | 4614 | } |
8082510e | 4615 | out: |
85e21bac CM |
4616 | if (pending_del_nr) { |
4617 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4618 | pending_del_nr); | |
79787eaa JM |
4619 | if (ret) |
4620 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 4621 | } |
79787eaa | 4622 | error: |
c1aa4575 | 4623 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
7f4f6e0a | 4624 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
28ed1345 | 4625 | |
39279cc3 | 4626 | btrfs_free_path(path); |
28ed1345 | 4627 | |
ee22184b | 4628 | if (be_nice && bytes_deleted > SZ_32M) { |
28ed1345 CM |
4629 | unsigned long updates = trans->delayed_ref_updates; |
4630 | if (updates) { | |
4631 | trans->delayed_ref_updates = 0; | |
4632 | ret = btrfs_run_delayed_refs(trans, root, updates * 2); | |
4633 | if (ret && !err) | |
4634 | err = ret; | |
4635 | } | |
4636 | } | |
8082510e | 4637 | return err; |
39279cc3 CM |
4638 | } |
4639 | ||
4640 | /* | |
9703fefe | 4641 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4642 | * @inode - inode that we're zeroing |
4643 | * @from - the offset to start zeroing | |
4644 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4645 | * offset | |
4646 | * @front - zero up to the offset instead of from the offset on | |
4647 | * | |
9703fefe | 4648 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4649 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4650 | */ |
9703fefe | 4651 | int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len, |
2aaa6655 | 4652 | int front) |
39279cc3 | 4653 | { |
2aaa6655 | 4654 | struct address_space *mapping = inode->i_mapping; |
db94535d | 4655 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4656 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4657 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4658 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 4659 | char *kaddr; |
db94535d | 4660 | u32 blocksize = root->sectorsize; |
09cbfeaf | 4661 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4662 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4663 | struct page *page; |
3b16a4e3 | 4664 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4665 | int ret = 0; |
9703fefe CR |
4666 | u64 block_start; |
4667 | u64 block_end; | |
39279cc3 | 4668 | |
2aaa6655 JB |
4669 | if ((offset & (blocksize - 1)) == 0 && |
4670 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4671 | goto out; |
9703fefe | 4672 | |
7cf5b976 | 4673 | ret = btrfs_delalloc_reserve_space(inode, |
9703fefe | 4674 | round_down(from, blocksize), blocksize); |
5d5e103a JB |
4675 | if (ret) |
4676 | goto out; | |
39279cc3 | 4677 | |
211c17f5 | 4678 | again: |
3b16a4e3 | 4679 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4680 | if (!page) { |
7cf5b976 | 4681 | btrfs_delalloc_release_space(inode, |
9703fefe CR |
4682 | round_down(from, blocksize), |
4683 | blocksize); | |
ac6a2b36 | 4684 | ret = -ENOMEM; |
39279cc3 | 4685 | goto out; |
5d5e103a | 4686 | } |
e6dcd2dc | 4687 | |
9703fefe CR |
4688 | block_start = round_down(from, blocksize); |
4689 | block_end = block_start + blocksize - 1; | |
e6dcd2dc | 4690 | |
39279cc3 | 4691 | if (!PageUptodate(page)) { |
9ebefb18 | 4692 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4693 | lock_page(page); |
211c17f5 CM |
4694 | if (page->mapping != mapping) { |
4695 | unlock_page(page); | |
09cbfeaf | 4696 | put_page(page); |
211c17f5 CM |
4697 | goto again; |
4698 | } | |
39279cc3 CM |
4699 | if (!PageUptodate(page)) { |
4700 | ret = -EIO; | |
89642229 | 4701 | goto out_unlock; |
39279cc3 CM |
4702 | } |
4703 | } | |
211c17f5 | 4704 | wait_on_page_writeback(page); |
e6dcd2dc | 4705 | |
9703fefe | 4706 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc CM |
4707 | set_page_extent_mapped(page); |
4708 | ||
9703fefe | 4709 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4710 | if (ordered) { |
9703fefe | 4711 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4712 | &cached_state, GFP_NOFS); |
e6dcd2dc | 4713 | unlock_page(page); |
09cbfeaf | 4714 | put_page(page); |
eb84ae03 | 4715 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4716 | btrfs_put_ordered_extent(ordered); |
4717 | goto again; | |
4718 | } | |
4719 | ||
9703fefe | 4720 | clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end, |
9e8a4a8b LB |
4721 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4722 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 4723 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 4724 | |
9703fefe | 4725 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, |
2ac55d41 | 4726 | &cached_state); |
9ed74f2d | 4727 | if (ret) { |
9703fefe | 4728 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4729 | &cached_state, GFP_NOFS); |
9ed74f2d JB |
4730 | goto out_unlock; |
4731 | } | |
4732 | ||
9703fefe | 4733 | if (offset != blocksize) { |
2aaa6655 | 4734 | if (!len) |
9703fefe | 4735 | len = blocksize - offset; |
e6dcd2dc | 4736 | kaddr = kmap(page); |
2aaa6655 | 4737 | if (front) |
9703fefe CR |
4738 | memset(kaddr + (block_start - page_offset(page)), |
4739 | 0, offset); | |
2aaa6655 | 4740 | else |
9703fefe CR |
4741 | memset(kaddr + (block_start - page_offset(page)) + offset, |
4742 | 0, len); | |
e6dcd2dc CM |
4743 | flush_dcache_page(page); |
4744 | kunmap(page); | |
4745 | } | |
247e743c | 4746 | ClearPageChecked(page); |
e6dcd2dc | 4747 | set_page_dirty(page); |
9703fefe | 4748 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state, |
2ac55d41 | 4749 | GFP_NOFS); |
39279cc3 | 4750 | |
89642229 | 4751 | out_unlock: |
5d5e103a | 4752 | if (ret) |
9703fefe CR |
4753 | btrfs_delalloc_release_space(inode, block_start, |
4754 | blocksize); | |
39279cc3 | 4755 | unlock_page(page); |
09cbfeaf | 4756 | put_page(page); |
39279cc3 CM |
4757 | out: |
4758 | return ret; | |
4759 | } | |
4760 | ||
16e7549f JB |
4761 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4762 | u64 offset, u64 len) | |
4763 | { | |
4764 | struct btrfs_trans_handle *trans; | |
4765 | int ret; | |
4766 | ||
4767 | /* | |
4768 | * Still need to make sure the inode looks like it's been updated so | |
4769 | * that any holes get logged if we fsync. | |
4770 | */ | |
4771 | if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) { | |
4772 | BTRFS_I(inode)->last_trans = root->fs_info->generation; | |
4773 | BTRFS_I(inode)->last_sub_trans = root->log_transid; | |
4774 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4775 | return 0; | |
4776 | } | |
4777 | ||
4778 | /* | |
4779 | * 1 - for the one we're dropping | |
4780 | * 1 - for the one we're adding | |
4781 | * 1 - for updating the inode. | |
4782 | */ | |
4783 | trans = btrfs_start_transaction(root, 3); | |
4784 | if (IS_ERR(trans)) | |
4785 | return PTR_ERR(trans); | |
4786 | ||
4787 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4788 | if (ret) { | |
4789 | btrfs_abort_transaction(trans, root, ret); | |
4790 | btrfs_end_transaction(trans, root); | |
4791 | return ret; | |
4792 | } | |
4793 | ||
4794 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset, | |
4795 | 0, 0, len, 0, len, 0, 0, 0); | |
4796 | if (ret) | |
4797 | btrfs_abort_transaction(trans, root, ret); | |
4798 | else | |
4799 | btrfs_update_inode(trans, root, inode); | |
4800 | btrfs_end_transaction(trans, root); | |
4801 | return ret; | |
4802 | } | |
4803 | ||
695a0d0d JB |
4804 | /* |
4805 | * This function puts in dummy file extents for the area we're creating a hole | |
4806 | * for. So if we are truncating this file to a larger size we need to insert | |
4807 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4808 | * the range between oldsize and size | |
4809 | */ | |
a41ad394 | 4810 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4811 | { |
9036c102 YZ |
4812 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4813 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4814 | struct extent_map *em = NULL; |
2ac55d41 | 4815 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4816 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
fda2832f QW |
4817 | u64 hole_start = ALIGN(oldsize, root->sectorsize); |
4818 | u64 block_end = ALIGN(size, root->sectorsize); | |
9036c102 YZ |
4819 | u64 last_byte; |
4820 | u64 cur_offset; | |
4821 | u64 hole_size; | |
9ed74f2d | 4822 | int err = 0; |
39279cc3 | 4823 | |
a71754fc | 4824 | /* |
9703fefe CR |
4825 | * If our size started in the middle of a block we need to zero out the |
4826 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
4827 | * expose stale data. |
4828 | */ | |
9703fefe | 4829 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
4830 | if (err) |
4831 | return err; | |
4832 | ||
9036c102 YZ |
4833 | if (size <= hole_start) |
4834 | return 0; | |
4835 | ||
9036c102 YZ |
4836 | while (1) { |
4837 | struct btrfs_ordered_extent *ordered; | |
fa7c1494 | 4838 | |
ff13db41 | 4839 | lock_extent_bits(io_tree, hole_start, block_end - 1, |
d0082371 | 4840 | &cached_state); |
fa7c1494 MX |
4841 | ordered = btrfs_lookup_ordered_range(inode, hole_start, |
4842 | block_end - hole_start); | |
9036c102 YZ |
4843 | if (!ordered) |
4844 | break; | |
2ac55d41 JB |
4845 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
4846 | &cached_state, GFP_NOFS); | |
fa7c1494 | 4847 | btrfs_start_ordered_extent(inode, ordered, 1); |
9036c102 YZ |
4848 | btrfs_put_ordered_extent(ordered); |
4849 | } | |
39279cc3 | 4850 | |
9036c102 YZ |
4851 | cur_offset = hole_start; |
4852 | while (1) { | |
4853 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
4854 | block_end - cur_offset, 0); | |
79787eaa JM |
4855 | if (IS_ERR(em)) { |
4856 | err = PTR_ERR(em); | |
f2767956 | 4857 | em = NULL; |
79787eaa JM |
4858 | break; |
4859 | } | |
9036c102 | 4860 | last_byte = min(extent_map_end(em), block_end); |
fda2832f | 4861 | last_byte = ALIGN(last_byte , root->sectorsize); |
8082510e | 4862 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4863 | struct extent_map *hole_em; |
9036c102 | 4864 | hole_size = last_byte - cur_offset; |
9ed74f2d | 4865 | |
16e7549f JB |
4866 | err = maybe_insert_hole(root, inode, cur_offset, |
4867 | hole_size); | |
4868 | if (err) | |
3893e33b | 4869 | break; |
5dc562c5 JB |
4870 | btrfs_drop_extent_cache(inode, cur_offset, |
4871 | cur_offset + hole_size - 1, 0); | |
4872 | hole_em = alloc_extent_map(); | |
4873 | if (!hole_em) { | |
4874 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4875 | &BTRFS_I(inode)->runtime_flags); | |
4876 | goto next; | |
4877 | } | |
4878 | hole_em->start = cur_offset; | |
4879 | hole_em->len = hole_size; | |
4880 | hole_em->orig_start = cur_offset; | |
8082510e | 4881 | |
5dc562c5 JB |
4882 | hole_em->block_start = EXTENT_MAP_HOLE; |
4883 | hole_em->block_len = 0; | |
b4939680 | 4884 | hole_em->orig_block_len = 0; |
cc95bef6 | 4885 | hole_em->ram_bytes = hole_size; |
5dc562c5 JB |
4886 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
4887 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
16e7549f | 4888 | hole_em->generation = root->fs_info->generation; |
8082510e | 4889 | |
5dc562c5 JB |
4890 | while (1) { |
4891 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4892 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4893 | write_unlock(&em_tree->lock); |
4894 | if (err != -EEXIST) | |
4895 | break; | |
4896 | btrfs_drop_extent_cache(inode, cur_offset, | |
4897 | cur_offset + | |
4898 | hole_size - 1, 0); | |
4899 | } | |
4900 | free_extent_map(hole_em); | |
9036c102 | 4901 | } |
16e7549f | 4902 | next: |
9036c102 | 4903 | free_extent_map(em); |
a22285a6 | 4904 | em = NULL; |
9036c102 | 4905 | cur_offset = last_byte; |
8082510e | 4906 | if (cur_offset >= block_end) |
9036c102 YZ |
4907 | break; |
4908 | } | |
a22285a6 | 4909 | free_extent_map(em); |
2ac55d41 JB |
4910 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
4911 | GFP_NOFS); | |
9036c102 YZ |
4912 | return err; |
4913 | } | |
39279cc3 | 4914 | |
3972f260 | 4915 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4916 | { |
f4a2f4c5 MX |
4917 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4918 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4919 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4920 | loff_t newsize = attr->ia_size; |
4921 | int mask = attr->ia_valid; | |
8082510e YZ |
4922 | int ret; |
4923 | ||
3972f260 ES |
4924 | /* |
4925 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4926 | * special case where we need to update the times despite not having | |
4927 | * these flags set. For all other operations the VFS set these flags | |
4928 | * explicitly if it wants a timestamp update. | |
4929 | */ | |
dff6efc3 CH |
4930 | if (newsize != oldsize) { |
4931 | inode_inc_iversion(inode); | |
4932 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
4933 | inode->i_ctime = inode->i_mtime = | |
4934 | current_fs_time(inode->i_sb); | |
4935 | } | |
3972f260 | 4936 | |
a41ad394 | 4937 | if (newsize > oldsize) { |
9ea24bbe FM |
4938 | /* |
4939 | * Don't do an expanding truncate while snapshoting is ongoing. | |
4940 | * This is to ensure the snapshot captures a fully consistent | |
4941 | * state of this file - if the snapshot captures this expanding | |
4942 | * truncation, it must capture all writes that happened before | |
4943 | * this truncation. | |
4944 | */ | |
0bc19f90 | 4945 | btrfs_wait_for_snapshot_creation(root); |
a41ad394 | 4946 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
9ea24bbe FM |
4947 | if (ret) { |
4948 | btrfs_end_write_no_snapshoting(root); | |
8082510e | 4949 | return ret; |
9ea24bbe | 4950 | } |
8082510e | 4951 | |
f4a2f4c5 | 4952 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe FM |
4953 | if (IS_ERR(trans)) { |
4954 | btrfs_end_write_no_snapshoting(root); | |
f4a2f4c5 | 4955 | return PTR_ERR(trans); |
9ea24bbe | 4956 | } |
f4a2f4c5 MX |
4957 | |
4958 | i_size_write(inode, newsize); | |
4959 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
27772b68 | 4960 | pagecache_isize_extended(inode, oldsize, newsize); |
f4a2f4c5 | 4961 | ret = btrfs_update_inode(trans, root, inode); |
9ea24bbe | 4962 | btrfs_end_write_no_snapshoting(root); |
7ad85bb7 | 4963 | btrfs_end_transaction(trans, root); |
a41ad394 | 4964 | } else { |
8082510e | 4965 | |
a41ad394 JB |
4966 | /* |
4967 | * We're truncating a file that used to have good data down to | |
4968 | * zero. Make sure it gets into the ordered flush list so that | |
4969 | * any new writes get down to disk quickly. | |
4970 | */ | |
4971 | if (newsize == 0) | |
72ac3c0d JB |
4972 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
4973 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 4974 | |
f3fe820c JB |
4975 | /* |
4976 | * 1 for the orphan item we're going to add | |
4977 | * 1 for the orphan item deletion. | |
4978 | */ | |
4979 | trans = btrfs_start_transaction(root, 2); | |
4980 | if (IS_ERR(trans)) | |
4981 | return PTR_ERR(trans); | |
4982 | ||
4983 | /* | |
4984 | * We need to do this in case we fail at _any_ point during the | |
4985 | * actual truncate. Once we do the truncate_setsize we could | |
4986 | * invalidate pages which forces any outstanding ordered io to | |
4987 | * be instantly completed which will give us extents that need | |
4988 | * to be truncated. If we fail to get an orphan inode down we | |
4989 | * could have left over extents that were never meant to live, | |
01327610 | 4990 | * so we need to guarantee from this point on that everything |
f3fe820c JB |
4991 | * will be consistent. |
4992 | */ | |
4993 | ret = btrfs_orphan_add(trans, inode); | |
4994 | btrfs_end_transaction(trans, root); | |
4995 | if (ret) | |
4996 | return ret; | |
4997 | ||
a41ad394 JB |
4998 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
4999 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
5000 | |
5001 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
5002 | btrfs_inode_block_unlocked_dio(inode); | |
5003 | inode_dio_wait(inode); | |
5004 | btrfs_inode_resume_unlocked_dio(inode); | |
5005 | ||
a41ad394 | 5006 | ret = btrfs_truncate(inode); |
7f4f6e0a JB |
5007 | if (ret && inode->i_nlink) { |
5008 | int err; | |
5009 | ||
5010 | /* | |
5011 | * failed to truncate, disk_i_size is only adjusted down | |
5012 | * as we remove extents, so it should represent the true | |
5013 | * size of the inode, so reset the in memory size and | |
5014 | * delete our orphan entry. | |
5015 | */ | |
5016 | trans = btrfs_join_transaction(root); | |
5017 | if (IS_ERR(trans)) { | |
5018 | btrfs_orphan_del(NULL, inode); | |
5019 | return ret; | |
5020 | } | |
5021 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
5022 | err = btrfs_orphan_del(trans, inode); | |
5023 | if (err) | |
5024 | btrfs_abort_transaction(trans, root, err); | |
5025 | btrfs_end_transaction(trans, root); | |
5026 | } | |
8082510e YZ |
5027 | } |
5028 | ||
a41ad394 | 5029 | return ret; |
8082510e YZ |
5030 | } |
5031 | ||
9036c102 YZ |
5032 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
5033 | { | |
2b0143b5 | 5034 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5035 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5036 | int err; |
39279cc3 | 5037 | |
b83cc969 LZ |
5038 | if (btrfs_root_readonly(root)) |
5039 | return -EROFS; | |
5040 | ||
9036c102 YZ |
5041 | err = inode_change_ok(inode, attr); |
5042 | if (err) | |
5043 | return err; | |
2bf5a725 | 5044 | |
5a3f23d5 | 5045 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5046 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5047 | if (err) |
5048 | return err; | |
39279cc3 | 5049 | } |
9036c102 | 5050 | |
1025774c CH |
5051 | if (attr->ia_valid) { |
5052 | setattr_copy(inode, attr); | |
0c4d2d95 | 5053 | inode_inc_iversion(inode); |
22c44fe6 | 5054 | err = btrfs_dirty_inode(inode); |
1025774c | 5055 | |
22c44fe6 | 5056 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 5057 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 5058 | } |
33268eaf | 5059 | |
39279cc3 CM |
5060 | return err; |
5061 | } | |
61295eb8 | 5062 | |
131e404a FDBM |
5063 | /* |
5064 | * While truncating the inode pages during eviction, we get the VFS calling | |
5065 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
5066 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
5067 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
5068 | * extent_state structures over and over, wasting lots of time. | |
5069 | * | |
5070 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
5071 | * those expensive operations on a per page basis and do only the ordered io | |
5072 | * finishing, while we release here the extent_map and extent_state structures, | |
5073 | * without the excessive merging and splitting. | |
5074 | */ | |
5075 | static void evict_inode_truncate_pages(struct inode *inode) | |
5076 | { | |
5077 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5078 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5079 | struct rb_node *node; | |
5080 | ||
5081 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5082 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5083 | |
5084 | write_lock(&map_tree->lock); | |
5085 | while (!RB_EMPTY_ROOT(&map_tree->map)) { | |
5086 | struct extent_map *em; | |
5087 | ||
5088 | node = rb_first(&map_tree->map); | |
5089 | em = rb_entry(node, struct extent_map, rb_node); | |
180589ef WS |
5090 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5091 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5092 | remove_extent_mapping(map_tree, em); |
5093 | free_extent_map(em); | |
7064dd5c FM |
5094 | if (need_resched()) { |
5095 | write_unlock(&map_tree->lock); | |
5096 | cond_resched(); | |
5097 | write_lock(&map_tree->lock); | |
5098 | } | |
131e404a FDBM |
5099 | } |
5100 | write_unlock(&map_tree->lock); | |
5101 | ||
6ca07097 FM |
5102 | /* |
5103 | * Keep looping until we have no more ranges in the io tree. | |
5104 | * We can have ongoing bios started by readpages (called from readahead) | |
9c6429d9 FM |
5105 | * that have their endio callback (extent_io.c:end_bio_extent_readpage) |
5106 | * still in progress (unlocked the pages in the bio but did not yet | |
5107 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5108 | * ranges can still be locked and eviction started because before |
5109 | * submitting those bios, which are executed by a separate task (work | |
5110 | * queue kthread), inode references (inode->i_count) were not taken | |
5111 | * (which would be dropped in the end io callback of each bio). | |
5112 | * Therefore here we effectively end up waiting for those bios and | |
5113 | * anyone else holding locked ranges without having bumped the inode's | |
5114 | * reference count - if we don't do it, when they access the inode's | |
5115 | * io_tree to unlock a range it may be too late, leading to an | |
5116 | * use-after-free issue. | |
5117 | */ | |
131e404a FDBM |
5118 | spin_lock(&io_tree->lock); |
5119 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5120 | struct extent_state *state; | |
5121 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5122 | u64 start; |
5123 | u64 end; | |
131e404a FDBM |
5124 | |
5125 | node = rb_first(&io_tree->state); | |
5126 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5127 | start = state->start; |
5128 | end = state->end; | |
131e404a FDBM |
5129 | spin_unlock(&io_tree->lock); |
5130 | ||
ff13db41 | 5131 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5132 | |
5133 | /* | |
5134 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5135 | * and its reserved space won't be freed by delayed_ref. | |
5136 | * So we need to free its reserved space here. | |
5137 | * (Refer to comment in btrfs_invalidatepage, case 2) | |
5138 | * | |
5139 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5140 | */ | |
5141 | if (state->state & EXTENT_DELALLOC) | |
5142 | btrfs_qgroup_free_data(inode, start, end - start + 1); | |
5143 | ||
6ca07097 | 5144 | clear_extent_bit(io_tree, start, end, |
131e404a FDBM |
5145 | EXTENT_LOCKED | EXTENT_DIRTY | |
5146 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
5147 | EXTENT_DEFRAG, 1, 1, | |
5148 | &cached_state, GFP_NOFS); | |
131e404a | 5149 | |
7064dd5c | 5150 | cond_resched(); |
131e404a FDBM |
5151 | spin_lock(&io_tree->lock); |
5152 | } | |
5153 | spin_unlock(&io_tree->lock); | |
5154 | } | |
5155 | ||
bd555975 | 5156 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
5157 | { |
5158 | struct btrfs_trans_handle *trans; | |
5159 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 5160 | struct btrfs_block_rsv *rsv, *global_rsv; |
3bce876f | 5161 | int steal_from_global = 0; |
07127184 | 5162 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 CM |
5163 | int ret; |
5164 | ||
1abe9b8a | 5165 | trace_btrfs_inode_evict(inode); |
5166 | ||
131e404a FDBM |
5167 | evict_inode_truncate_pages(inode); |
5168 | ||
69e9c6c6 SB |
5169 | if (inode->i_nlink && |
5170 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5171 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
5172 | btrfs_is_free_space_inode(inode))) | |
bd555975 AV |
5173 | goto no_delete; |
5174 | ||
39279cc3 | 5175 | if (is_bad_inode(inode)) { |
7b128766 | 5176 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
5177 | goto no_delete; |
5178 | } | |
bd555975 | 5179 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
a30e577c JM |
5180 | if (!special_file(inode->i_mode)) |
5181 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
5f39d397 | 5182 | |
f612496b MX |
5183 | btrfs_free_io_failure_record(inode, 0, (u64)-1); |
5184 | ||
c71bf099 | 5185 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 5186 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 5187 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
5188 | goto no_delete; |
5189 | } | |
5190 | ||
76dda93c | 5191 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5192 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5193 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5194 | goto no_delete; |
5195 | } | |
5196 | ||
0e8c36a9 MX |
5197 | ret = btrfs_commit_inode_delayed_inode(inode); |
5198 | if (ret) { | |
5199 | btrfs_orphan_del(NULL, inode); | |
5200 | goto no_delete; | |
5201 | } | |
5202 | ||
66d8f3dd | 5203 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
5204 | if (!rsv) { |
5205 | btrfs_orphan_del(NULL, inode); | |
5206 | goto no_delete; | |
5207 | } | |
4a338542 | 5208 | rsv->size = min_size; |
ca7e70f5 | 5209 | rsv->failfast = 1; |
726c35fa | 5210 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 5211 | |
dbe674a9 | 5212 | btrfs_i_size_write(inode, 0); |
5f39d397 | 5213 | |
4289a667 | 5214 | /* |
8407aa46 MX |
5215 | * This is a bit simpler than btrfs_truncate since we've already |
5216 | * reserved our space for our orphan item in the unlink, so we just | |
5217 | * need to reserve some slack space in case we add bytes and update | |
5218 | * inode item when doing the truncate. | |
4289a667 | 5219 | */ |
8082510e | 5220 | while (1) { |
08e007d2 MX |
5221 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
5222 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
5223 | |
5224 | /* | |
5225 | * Try and steal from the global reserve since we will | |
5226 | * likely not use this space anyway, we want to try as | |
5227 | * hard as possible to get this to work. | |
5228 | */ | |
5229 | if (ret) | |
3bce876f JB |
5230 | steal_from_global++; |
5231 | else | |
5232 | steal_from_global = 0; | |
5233 | ret = 0; | |
d68fc57b | 5234 | |
3bce876f JB |
5235 | /* |
5236 | * steal_from_global == 0: we reserved stuff, hooray! | |
5237 | * steal_from_global == 1: we didn't reserve stuff, boo! | |
5238 | * steal_from_global == 2: we've committed, still not a lot of | |
5239 | * room but maybe we'll have room in the global reserve this | |
5240 | * time. | |
5241 | * steal_from_global == 3: abandon all hope! | |
5242 | */ | |
5243 | if (steal_from_global > 2) { | |
c2cf52eb SK |
5244 | btrfs_warn(root->fs_info, |
5245 | "Could not get space for a delete, will truncate on mount %d", | |
5246 | ret); | |
4289a667 JB |
5247 | btrfs_orphan_del(NULL, inode); |
5248 | btrfs_free_block_rsv(root, rsv); | |
5249 | goto no_delete; | |
d68fc57b | 5250 | } |
7b128766 | 5251 | |
0e8c36a9 | 5252 | trans = btrfs_join_transaction(root); |
4289a667 JB |
5253 | if (IS_ERR(trans)) { |
5254 | btrfs_orphan_del(NULL, inode); | |
5255 | btrfs_free_block_rsv(root, rsv); | |
5256 | goto no_delete; | |
d68fc57b | 5257 | } |
7b128766 | 5258 | |
3bce876f | 5259 | /* |
01327610 | 5260 | * We can't just steal from the global reserve, we need to make |
3bce876f JB |
5261 | * sure there is room to do it, if not we need to commit and try |
5262 | * again. | |
5263 | */ | |
5264 | if (steal_from_global) { | |
5265 | if (!btrfs_check_space_for_delayed_refs(trans, root)) | |
5266 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, | |
25d609f8 | 5267 | min_size, 0); |
3bce876f JB |
5268 | else |
5269 | ret = -ENOSPC; | |
5270 | } | |
5271 | ||
5272 | /* | |
5273 | * Couldn't steal from the global reserve, we have too much | |
5274 | * pending stuff built up, commit the transaction and try it | |
5275 | * again. | |
5276 | */ | |
5277 | if (ret) { | |
5278 | ret = btrfs_commit_transaction(trans, root); | |
5279 | if (ret) { | |
5280 | btrfs_orphan_del(NULL, inode); | |
5281 | btrfs_free_block_rsv(root, rsv); | |
5282 | goto no_delete; | |
5283 | } | |
5284 | continue; | |
5285 | } else { | |
5286 | steal_from_global = 0; | |
5287 | } | |
5288 | ||
4289a667 JB |
5289 | trans->block_rsv = rsv; |
5290 | ||
d68fc57b | 5291 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
28ed1345 | 5292 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 5293 | break; |
85e21bac | 5294 | |
8407aa46 | 5295 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e YZ |
5296 | btrfs_end_transaction(trans, root); |
5297 | trans = NULL; | |
b53d3f5d | 5298 | btrfs_btree_balance_dirty(root); |
8082510e | 5299 | } |
5f39d397 | 5300 | |
4289a667 JB |
5301 | btrfs_free_block_rsv(root, rsv); |
5302 | ||
4ef31a45 JB |
5303 | /* |
5304 | * Errors here aren't a big deal, it just means we leave orphan items | |
5305 | * in the tree. They will be cleaned up on the next mount. | |
5306 | */ | |
8082510e | 5307 | if (ret == 0) { |
4289a667 | 5308 | trans->block_rsv = root->orphan_block_rsv; |
4ef31a45 JB |
5309 | btrfs_orphan_del(trans, inode); |
5310 | } else { | |
5311 | btrfs_orphan_del(NULL, inode); | |
8082510e | 5312 | } |
54aa1f4d | 5313 | |
4289a667 | 5314 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
5315 | if (!(root == root->fs_info->tree_root || |
5316 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 5317 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 5318 | |
54aa1f4d | 5319 | btrfs_end_transaction(trans, root); |
b53d3f5d | 5320 | btrfs_btree_balance_dirty(root); |
39279cc3 | 5321 | no_delete: |
89042e5a | 5322 | btrfs_remove_delayed_node(inode); |
dbd5768f | 5323 | clear_inode(inode); |
39279cc3 CM |
5324 | } |
5325 | ||
5326 | /* | |
5327 | * this returns the key found in the dir entry in the location pointer. | |
5328 | * If no dir entries were found, location->objectid is 0. | |
5329 | */ | |
5330 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
5331 | struct btrfs_key *location) | |
5332 | { | |
5333 | const char *name = dentry->d_name.name; | |
5334 | int namelen = dentry->d_name.len; | |
5335 | struct btrfs_dir_item *di; | |
5336 | struct btrfs_path *path; | |
5337 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5338 | int ret = 0; |
39279cc3 CM |
5339 | |
5340 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5341 | if (!path) |
5342 | return -ENOMEM; | |
3954401f | 5343 | |
33345d01 | 5344 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 5345 | namelen, 0); |
0d9f7f3e Y |
5346 | if (IS_ERR(di)) |
5347 | ret = PTR_ERR(di); | |
d397712b | 5348 | |
c704005d | 5349 | if (IS_ERR_OR_NULL(di)) |
3954401f | 5350 | goto out_err; |
d397712b | 5351 | |
5f39d397 | 5352 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 5353 | out: |
39279cc3 CM |
5354 | btrfs_free_path(path); |
5355 | return ret; | |
3954401f CM |
5356 | out_err: |
5357 | location->objectid = 0; | |
5358 | goto out; | |
39279cc3 CM |
5359 | } |
5360 | ||
5361 | /* | |
5362 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5363 | * needs to be changed to reflect the root directory of the tree root. This | |
5364 | * is kind of like crossing a mount point. | |
5365 | */ | |
5366 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
5367 | struct inode *dir, |
5368 | struct dentry *dentry, | |
5369 | struct btrfs_key *location, | |
5370 | struct btrfs_root **sub_root) | |
39279cc3 | 5371 | { |
4df27c4d YZ |
5372 | struct btrfs_path *path; |
5373 | struct btrfs_root *new_root; | |
5374 | struct btrfs_root_ref *ref; | |
5375 | struct extent_buffer *leaf; | |
1d4c08e0 | 5376 | struct btrfs_key key; |
4df27c4d YZ |
5377 | int ret; |
5378 | int err = 0; | |
39279cc3 | 5379 | |
4df27c4d YZ |
5380 | path = btrfs_alloc_path(); |
5381 | if (!path) { | |
5382 | err = -ENOMEM; | |
5383 | goto out; | |
5384 | } | |
39279cc3 | 5385 | |
4df27c4d | 5386 | err = -ENOENT; |
1d4c08e0 DS |
5387 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5388 | key.type = BTRFS_ROOT_REF_KEY; | |
5389 | key.offset = location->objectid; | |
5390 | ||
5391 | ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path, | |
5392 | 0, 0); | |
4df27c4d YZ |
5393 | if (ret) { |
5394 | if (ret < 0) | |
5395 | err = ret; | |
5396 | goto out; | |
5397 | } | |
39279cc3 | 5398 | |
4df27c4d YZ |
5399 | leaf = path->nodes[0]; |
5400 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 5401 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
5402 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5403 | goto out; | |
39279cc3 | 5404 | |
4df27c4d YZ |
5405 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5406 | (unsigned long)(ref + 1), | |
5407 | dentry->d_name.len); | |
5408 | if (ret) | |
5409 | goto out; | |
5410 | ||
b3b4aa74 | 5411 | btrfs_release_path(path); |
4df27c4d YZ |
5412 | |
5413 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
5414 | if (IS_ERR(new_root)) { | |
5415 | err = PTR_ERR(new_root); | |
5416 | goto out; | |
5417 | } | |
5418 | ||
4df27c4d YZ |
5419 | *sub_root = new_root; |
5420 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5421 | location->type = BTRFS_INODE_ITEM_KEY; | |
5422 | location->offset = 0; | |
5423 | err = 0; | |
5424 | out: | |
5425 | btrfs_free_path(path); | |
5426 | return err; | |
39279cc3 CM |
5427 | } |
5428 | ||
5d4f98a2 YZ |
5429 | static void inode_tree_add(struct inode *inode) |
5430 | { | |
5431 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5432 | struct btrfs_inode *entry; | |
03e860bd FNP |
5433 | struct rb_node **p; |
5434 | struct rb_node *parent; | |
cef21937 | 5435 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
33345d01 | 5436 | u64 ino = btrfs_ino(inode); |
5d4f98a2 | 5437 | |
1d3382cb | 5438 | if (inode_unhashed(inode)) |
76dda93c | 5439 | return; |
e1409cef | 5440 | parent = NULL; |
5d4f98a2 | 5441 | spin_lock(&root->inode_lock); |
e1409cef | 5442 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5443 | while (*p) { |
5444 | parent = *p; | |
5445 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5446 | ||
33345d01 | 5447 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 5448 | p = &parent->rb_left; |
33345d01 | 5449 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 5450 | p = &parent->rb_right; |
5d4f98a2 YZ |
5451 | else { |
5452 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5453 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5454 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5455 | RB_CLEAR_NODE(parent); |
5456 | spin_unlock(&root->inode_lock); | |
cef21937 | 5457 | return; |
5d4f98a2 YZ |
5458 | } |
5459 | } | |
cef21937 FDBM |
5460 | rb_link_node(new, parent, p); |
5461 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5462 | spin_unlock(&root->inode_lock); |
5463 | } | |
5464 | ||
5465 | static void inode_tree_del(struct inode *inode) | |
5466 | { | |
5467 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 5468 | int empty = 0; |
5d4f98a2 | 5469 | |
03e860bd | 5470 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5471 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5472 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5473 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5474 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5475 | } |
03e860bd | 5476 | spin_unlock(&root->inode_lock); |
76dda93c | 5477 | |
69e9c6c6 | 5478 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5479 | synchronize_srcu(&root->fs_info->subvol_srcu); |
5480 | spin_lock(&root->inode_lock); | |
5481 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5482 | spin_unlock(&root->inode_lock); | |
5483 | if (empty) | |
5484 | btrfs_add_dead_root(root); | |
5485 | } | |
5486 | } | |
5487 | ||
143bede5 | 5488 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
5489 | { |
5490 | struct rb_node *node; | |
5491 | struct rb_node *prev; | |
5492 | struct btrfs_inode *entry; | |
5493 | struct inode *inode; | |
5494 | u64 objectid = 0; | |
5495 | ||
7813b3db LB |
5496 | if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
5497 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
76dda93c YZ |
5498 | |
5499 | spin_lock(&root->inode_lock); | |
5500 | again: | |
5501 | node = root->inode_tree.rb_node; | |
5502 | prev = NULL; | |
5503 | while (node) { | |
5504 | prev = node; | |
5505 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
5506 | ||
33345d01 | 5507 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 5508 | node = node->rb_left; |
33345d01 | 5509 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
5510 | node = node->rb_right; |
5511 | else | |
5512 | break; | |
5513 | } | |
5514 | if (!node) { | |
5515 | while (prev) { | |
5516 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 5517 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
5518 | node = prev; |
5519 | break; | |
5520 | } | |
5521 | prev = rb_next(prev); | |
5522 | } | |
5523 | } | |
5524 | while (node) { | |
5525 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 5526 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
5527 | inode = igrab(&entry->vfs_inode); |
5528 | if (inode) { | |
5529 | spin_unlock(&root->inode_lock); | |
5530 | if (atomic_read(&inode->i_count) > 1) | |
5531 | d_prune_aliases(inode); | |
5532 | /* | |
45321ac5 | 5533 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
5534 | * the inode cache when its usage count |
5535 | * hits zero. | |
5536 | */ | |
5537 | iput(inode); | |
5538 | cond_resched(); | |
5539 | spin_lock(&root->inode_lock); | |
5540 | goto again; | |
5541 | } | |
5542 | ||
5543 | if (cond_resched_lock(&root->inode_lock)) | |
5544 | goto again; | |
5545 | ||
5546 | node = rb_next(node); | |
5547 | } | |
5548 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
5549 | } |
5550 | ||
e02119d5 CM |
5551 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5552 | { | |
5553 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5554 | inode->i_ino = args->location->objectid; |
5555 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5556 | sizeof(*args->location)); | |
e02119d5 | 5557 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5558 | return 0; |
5559 | } | |
5560 | ||
5561 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5562 | { | |
5563 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5564 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5565 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5566 | } |
5567 | ||
5d4f98a2 | 5568 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5569 | struct btrfs_key *location, |
5d4f98a2 | 5570 | struct btrfs_root *root) |
39279cc3 CM |
5571 | { |
5572 | struct inode *inode; | |
5573 | struct btrfs_iget_args args; | |
90d3e592 | 5574 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5575 | |
90d3e592 | 5576 | args.location = location; |
39279cc3 CM |
5577 | args.root = root; |
5578 | ||
778ba82b | 5579 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5580 | btrfs_init_locked_inode, |
5581 | (void *)&args); | |
5582 | return inode; | |
5583 | } | |
5584 | ||
1a54ef8c BR |
5585 | /* Get an inode object given its location and corresponding root. |
5586 | * Returns in *is_new if the inode was read from disk | |
5587 | */ | |
5588 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 5589 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
5590 | { |
5591 | struct inode *inode; | |
5592 | ||
90d3e592 | 5593 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5594 | if (!inode) |
5d4f98a2 | 5595 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5596 | |
5597 | if (inode->i_state & I_NEW) { | |
1a54ef8c | 5598 | btrfs_read_locked_inode(inode); |
1748f843 MF |
5599 | if (!is_bad_inode(inode)) { |
5600 | inode_tree_add(inode); | |
5601 | unlock_new_inode(inode); | |
5602 | if (new) | |
5603 | *new = 1; | |
5604 | } else { | |
e0b6d65b ST |
5605 | unlock_new_inode(inode); |
5606 | iput(inode); | |
5607 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
5608 | } |
5609 | } | |
5610 | ||
1a54ef8c BR |
5611 | return inode; |
5612 | } | |
5613 | ||
4df27c4d YZ |
5614 | static struct inode *new_simple_dir(struct super_block *s, |
5615 | struct btrfs_key *key, | |
5616 | struct btrfs_root *root) | |
5617 | { | |
5618 | struct inode *inode = new_inode(s); | |
5619 | ||
5620 | if (!inode) | |
5621 | return ERR_PTR(-ENOMEM); | |
5622 | ||
4df27c4d YZ |
5623 | BTRFS_I(inode)->root = root; |
5624 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5625 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5626 | |
5627 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5628 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
5629 | inode->i_fop = &simple_dir_operations; |
5630 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
04b285f3 | 5631 | inode->i_mtime = current_fs_time(inode->i_sb); |
9cc97d64 | 5632 | inode->i_atime = inode->i_mtime; |
5633 | inode->i_ctime = inode->i_mtime; | |
5634 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
4df27c4d YZ |
5635 | |
5636 | return inode; | |
5637 | } | |
5638 | ||
3de4586c | 5639 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5640 | { |
d397712b | 5641 | struct inode *inode; |
4df27c4d | 5642 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5643 | struct btrfs_root *sub_root = root; |
5644 | struct btrfs_key location; | |
76dda93c | 5645 | int index; |
b4aff1f8 | 5646 | int ret = 0; |
39279cc3 CM |
5647 | |
5648 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5649 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5650 | |
39e3c955 | 5651 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
5652 | if (ret < 0) |
5653 | return ERR_PTR(ret); | |
5f39d397 | 5654 | |
4df27c4d | 5655 | if (location.objectid == 0) |
5662344b | 5656 | return ERR_PTR(-ENOENT); |
4df27c4d YZ |
5657 | |
5658 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 5659 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
5660 | return inode; |
5661 | } | |
5662 | ||
5663 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
5664 | ||
76dda93c | 5665 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
5666 | ret = fixup_tree_root_location(root, dir, dentry, |
5667 | &location, &sub_root); | |
5668 | if (ret < 0) { | |
5669 | if (ret != -ENOENT) | |
5670 | inode = ERR_PTR(ret); | |
5671 | else | |
5672 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5673 | } else { | |
73f73415 | 5674 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5675 | } |
76dda93c YZ |
5676 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
5677 | ||
34d19bad | 5678 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
5679 | down_read(&root->fs_info->cleanup_work_sem); |
5680 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 5681 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 5682 | up_read(&root->fs_info->cleanup_work_sem); |
01cd3367 JB |
5683 | if (ret) { |
5684 | iput(inode); | |
66b4ffd1 | 5685 | inode = ERR_PTR(ret); |
01cd3367 | 5686 | } |
c71bf099 YZ |
5687 | } |
5688 | ||
3de4586c CM |
5689 | return inode; |
5690 | } | |
5691 | ||
fe15ce44 | 5692 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5693 | { |
5694 | struct btrfs_root *root; | |
2b0143b5 | 5695 | struct inode *inode = d_inode(dentry); |
76dda93c | 5696 | |
848cce0d | 5697 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5698 | inode = d_inode(dentry->d_parent); |
76dda93c | 5699 | |
848cce0d LZ |
5700 | if (inode) { |
5701 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5702 | if (btrfs_root_refs(&root->root_item) == 0) |
5703 | return 1; | |
848cce0d LZ |
5704 | |
5705 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
5706 | return 1; | |
efefb143 | 5707 | } |
76dda93c YZ |
5708 | return 0; |
5709 | } | |
5710 | ||
b4aff1f8 JB |
5711 | static void btrfs_dentry_release(struct dentry *dentry) |
5712 | { | |
944a4515 | 5713 | kfree(dentry->d_fsdata); |
b4aff1f8 JB |
5714 | } |
5715 | ||
3de4586c | 5716 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5717 | unsigned int flags) |
3de4586c | 5718 | { |
5662344b | 5719 | struct inode *inode; |
a66e7cc6 | 5720 | |
5662344b TI |
5721 | inode = btrfs_lookup_dentry(dir, dentry); |
5722 | if (IS_ERR(inode)) { | |
5723 | if (PTR_ERR(inode) == -ENOENT) | |
5724 | inode = NULL; | |
5725 | else | |
5726 | return ERR_CAST(inode); | |
5727 | } | |
5728 | ||
41d28bca | 5729 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5730 | } |
5731 | ||
16cdcec7 | 5732 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
5733 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
5734 | }; | |
5735 | ||
9cdda8d3 | 5736 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5737 | { |
9cdda8d3 | 5738 | struct inode *inode = file_inode(file); |
39279cc3 CM |
5739 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5740 | struct btrfs_item *item; | |
5741 | struct btrfs_dir_item *di; | |
5742 | struct btrfs_key key; | |
5f39d397 | 5743 | struct btrfs_key found_key; |
39279cc3 | 5744 | struct btrfs_path *path; |
16cdcec7 MX |
5745 | struct list_head ins_list; |
5746 | struct list_head del_list; | |
39279cc3 | 5747 | int ret; |
5f39d397 | 5748 | struct extent_buffer *leaf; |
39279cc3 | 5749 | int slot; |
39279cc3 CM |
5750 | unsigned char d_type; |
5751 | int over = 0; | |
5752 | u32 di_cur; | |
5753 | u32 di_total; | |
5754 | u32 di_len; | |
5755 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
5756 | char tmp_name[32]; |
5757 | char *name_ptr; | |
5758 | int name_len; | |
9cdda8d3 | 5759 | int is_curr = 0; /* ctx->pos points to the current index? */ |
bc4ef759 | 5760 | bool emitted; |
02dbfc99 | 5761 | bool put = false; |
39279cc3 CM |
5762 | |
5763 | /* FIXME, use a real flag for deciding about the key type */ | |
5764 | if (root->fs_info->tree_root == root) | |
5765 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 5766 | |
9cdda8d3 AV |
5767 | if (!dir_emit_dots(file, ctx)) |
5768 | return 0; | |
5769 | ||
49593bfa | 5770 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5771 | if (!path) |
5772 | return -ENOMEM; | |
ff5714cc | 5773 | |
e4058b54 | 5774 | path->reada = READA_FORWARD; |
49593bfa | 5775 | |
16cdcec7 MX |
5776 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5777 | INIT_LIST_HEAD(&ins_list); | |
5778 | INIT_LIST_HEAD(&del_list); | |
02dbfc99 OS |
5779 | put = btrfs_readdir_get_delayed_items(inode, &ins_list, |
5780 | &del_list); | |
16cdcec7 MX |
5781 | } |
5782 | ||
962a298f | 5783 | key.type = key_type; |
9cdda8d3 | 5784 | key.offset = ctx->pos; |
33345d01 | 5785 | key.objectid = btrfs_ino(inode); |
5f39d397 | 5786 | |
39279cc3 CM |
5787 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5788 | if (ret < 0) | |
5789 | goto err; | |
49593bfa | 5790 | |
bc4ef759 | 5791 | emitted = false; |
49593bfa | 5792 | while (1) { |
5f39d397 | 5793 | leaf = path->nodes[0]; |
39279cc3 | 5794 | slot = path->slots[0]; |
b9e03af0 LZ |
5795 | if (slot >= btrfs_header_nritems(leaf)) { |
5796 | ret = btrfs_next_leaf(root, path); | |
5797 | if (ret < 0) | |
5798 | goto err; | |
5799 | else if (ret > 0) | |
5800 | break; | |
5801 | continue; | |
39279cc3 | 5802 | } |
3de4586c | 5803 | |
dd3cc16b | 5804 | item = btrfs_item_nr(slot); |
5f39d397 CM |
5805 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5806 | ||
5807 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5808 | break; |
962a298f | 5809 | if (found_key.type != key_type) |
39279cc3 | 5810 | break; |
9cdda8d3 | 5811 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5812 | goto next; |
16cdcec7 MX |
5813 | if (key_type == BTRFS_DIR_INDEX_KEY && |
5814 | btrfs_should_delete_dir_index(&del_list, | |
5815 | found_key.offset)) | |
5816 | goto next; | |
5f39d397 | 5817 | |
9cdda8d3 | 5818 | ctx->pos = found_key.offset; |
16cdcec7 | 5819 | is_curr = 1; |
49593bfa | 5820 | |
39279cc3 CM |
5821 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
5822 | di_cur = 0; | |
5f39d397 | 5823 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
5824 | |
5825 | while (di_cur < di_total) { | |
5f39d397 CM |
5826 | struct btrfs_key location; |
5827 | ||
22a94d44 JB |
5828 | if (verify_dir_item(root, leaf, di)) |
5829 | break; | |
5830 | ||
5f39d397 | 5831 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 5832 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
5833 | name_ptr = tmp_name; |
5834 | } else { | |
49e350a4 | 5835 | name_ptr = kmalloc(name_len, GFP_KERNEL); |
49593bfa DW |
5836 | if (!name_ptr) { |
5837 | ret = -ENOMEM; | |
5838 | goto err; | |
5839 | } | |
5f39d397 CM |
5840 | } |
5841 | read_extent_buffer(leaf, name_ptr, | |
5842 | (unsigned long)(di + 1), name_len); | |
5843 | ||
5844 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
5845 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 5846 | |
fede766f | 5847 | |
3de4586c | 5848 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
5849 | * skip it. |
5850 | * | |
5851 | * In contrast to old kernels, we insert the snapshot's | |
5852 | * dir item and dir index after it has been created, so | |
5853 | * we won't find a reference to our own snapshot. We | |
5854 | * still keep the following code for backward | |
5855 | * compatibility. | |
3de4586c CM |
5856 | */ |
5857 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
5858 | location.objectid == root->root_key.objectid) { | |
5859 | over = 0; | |
5860 | goto skip; | |
5861 | } | |
9cdda8d3 AV |
5862 | over = !dir_emit(ctx, name_ptr, name_len, |
5863 | location.objectid, d_type); | |
5f39d397 | 5864 | |
3de4586c | 5865 | skip: |
5f39d397 CM |
5866 | if (name_ptr != tmp_name) |
5867 | kfree(name_ptr); | |
5868 | ||
39279cc3 CM |
5869 | if (over) |
5870 | goto nopos; | |
bc4ef759 | 5871 | emitted = true; |
5103e947 | 5872 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 5873 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
5874 | di_cur += di_len; |
5875 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
5876 | } | |
b9e03af0 LZ |
5877 | next: |
5878 | path->slots[0]++; | |
39279cc3 | 5879 | } |
49593bfa | 5880 | |
16cdcec7 MX |
5881 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5882 | if (is_curr) | |
9cdda8d3 | 5883 | ctx->pos++; |
bc4ef759 | 5884 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted); |
16cdcec7 MX |
5885 | if (ret) |
5886 | goto nopos; | |
5887 | } | |
5888 | ||
bc4ef759 DS |
5889 | /* |
5890 | * If we haven't emitted any dir entry, we must not touch ctx->pos as | |
5891 | * it was was set to the termination value in previous call. We assume | |
5892 | * that "." and ".." were emitted if we reach this point and set the | |
5893 | * termination value as well for an empty directory. | |
5894 | */ | |
5895 | if (ctx->pos > 2 && !emitted) | |
5896 | goto nopos; | |
5897 | ||
49593bfa | 5898 | /* Reached end of directory/root. Bump pos past the last item. */ |
db62efbb ZB |
5899 | ctx->pos++; |
5900 | ||
5901 | /* | |
5902 | * Stop new entries from being returned after we return the last | |
5903 | * entry. | |
5904 | * | |
5905 | * New directory entries are assigned a strictly increasing | |
5906 | * offset. This means that new entries created during readdir | |
5907 | * are *guaranteed* to be seen in the future by that readdir. | |
5908 | * This has broken buggy programs which operate on names as | |
5909 | * they're returned by readdir. Until we re-use freed offsets | |
5910 | * we have this hack to stop new entries from being returned | |
5911 | * under the assumption that they'll never reach this huge | |
5912 | * offset. | |
5913 | * | |
5914 | * This is being careful not to overflow 32bit loff_t unless the | |
5915 | * last entry requires it because doing so has broken 32bit apps | |
5916 | * in the past. | |
5917 | */ | |
5918 | if (key_type == BTRFS_DIR_INDEX_KEY) { | |
5919 | if (ctx->pos >= INT_MAX) | |
5920 | ctx->pos = LLONG_MAX; | |
5921 | else | |
5922 | ctx->pos = INT_MAX; | |
5923 | } | |
39279cc3 CM |
5924 | nopos: |
5925 | ret = 0; | |
5926 | err: | |
02dbfc99 OS |
5927 | if (put) |
5928 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 5929 | btrfs_free_path(path); |
39279cc3 CM |
5930 | return ret; |
5931 | } | |
5932 | ||
a9185b41 | 5933 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
5934 | { |
5935 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5936 | struct btrfs_trans_handle *trans; | |
5937 | int ret = 0; | |
0af3d00b | 5938 | bool nolock = false; |
39279cc3 | 5939 | |
72ac3c0d | 5940 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
5941 | return 0; |
5942 | ||
83eea1f1 | 5943 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 5944 | nolock = true; |
0af3d00b | 5945 | |
a9185b41 | 5946 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 5947 | if (nolock) |
7a7eaa40 | 5948 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 5949 | else |
7a7eaa40 | 5950 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5951 | if (IS_ERR(trans)) |
5952 | return PTR_ERR(trans); | |
a698d075 | 5953 | ret = btrfs_commit_transaction(trans, root); |
39279cc3 CM |
5954 | } |
5955 | return ret; | |
5956 | } | |
5957 | ||
5958 | /* | |
54aa1f4d | 5959 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5960 | * inode changes. But, it is most likely to find the inode in cache. |
5961 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5962 | * to keep or drop this code. | |
5963 | */ | |
48a3b636 | 5964 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
5965 | { |
5966 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5967 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5968 | int ret; |
5969 | ||
72ac3c0d | 5970 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5971 | return 0; |
39279cc3 | 5972 | |
7a7eaa40 | 5973 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5974 | if (IS_ERR(trans)) |
5975 | return PTR_ERR(trans); | |
8929ecfa YZ |
5976 | |
5977 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
5978 | if (ret && ret == -ENOSPC) { |
5979 | /* whoops, lets try again with the full transaction */ | |
5980 | btrfs_end_transaction(trans, root); | |
5981 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
5982 | if (IS_ERR(trans)) |
5983 | return PTR_ERR(trans); | |
8929ecfa | 5984 | |
94b60442 | 5985 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 5986 | } |
39279cc3 | 5987 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
5988 | if (BTRFS_I(inode)->delayed_node) |
5989 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
5990 | |
5991 | return ret; | |
5992 | } | |
5993 | ||
5994 | /* | |
5995 | * This is a copy of file_update_time. We need this so we can return error on | |
5996 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
5997 | */ | |
e41f941a JB |
5998 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
5999 | int flags) | |
22c44fe6 | 6000 | { |
2bc55652 AB |
6001 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6002 | ||
6003 | if (btrfs_root_readonly(root)) | |
6004 | return -EROFS; | |
6005 | ||
e41f941a | 6006 | if (flags & S_VERSION) |
22c44fe6 | 6007 | inode_inc_iversion(inode); |
e41f941a JB |
6008 | if (flags & S_CTIME) |
6009 | inode->i_ctime = *now; | |
6010 | if (flags & S_MTIME) | |
6011 | inode->i_mtime = *now; | |
6012 | if (flags & S_ATIME) | |
6013 | inode->i_atime = *now; | |
6014 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
6015 | } |
6016 | ||
d352ac68 CM |
6017 | /* |
6018 | * find the highest existing sequence number in a directory | |
6019 | * and then set the in-memory index_cnt variable to reflect | |
6020 | * free sequence numbers | |
6021 | */ | |
aec7477b JB |
6022 | static int btrfs_set_inode_index_count(struct inode *inode) |
6023 | { | |
6024 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6025 | struct btrfs_key key, found_key; | |
6026 | struct btrfs_path *path; | |
6027 | struct extent_buffer *leaf; | |
6028 | int ret; | |
6029 | ||
33345d01 | 6030 | key.objectid = btrfs_ino(inode); |
962a298f | 6031 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
6032 | key.offset = (u64)-1; |
6033 | ||
6034 | path = btrfs_alloc_path(); | |
6035 | if (!path) | |
6036 | return -ENOMEM; | |
6037 | ||
6038 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6039 | if (ret < 0) | |
6040 | goto out; | |
6041 | /* FIXME: we should be able to handle this */ | |
6042 | if (ret == 0) | |
6043 | goto out; | |
6044 | ret = 0; | |
6045 | ||
6046 | /* | |
6047 | * MAGIC NUMBER EXPLANATION: | |
6048 | * since we search a directory based on f_pos we have to start at 2 | |
6049 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
6050 | * else has to start at 2 | |
6051 | */ | |
6052 | if (path->slots[0] == 0) { | |
6053 | BTRFS_I(inode)->index_cnt = 2; | |
6054 | goto out; | |
6055 | } | |
6056 | ||
6057 | path->slots[0]--; | |
6058 | ||
6059 | leaf = path->nodes[0]; | |
6060 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6061 | ||
33345d01 | 6062 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 6063 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
aec7477b JB |
6064 | BTRFS_I(inode)->index_cnt = 2; |
6065 | goto out; | |
6066 | } | |
6067 | ||
6068 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
6069 | out: | |
6070 | btrfs_free_path(path); | |
6071 | return ret; | |
6072 | } | |
6073 | ||
d352ac68 CM |
6074 | /* |
6075 | * helper to find a free sequence number in a given directory. This current | |
6076 | * code is very simple, later versions will do smarter things in the btree | |
6077 | */ | |
3de4586c | 6078 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
6079 | { |
6080 | int ret = 0; | |
6081 | ||
6082 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
6083 | ret = btrfs_inode_delayed_dir_index_count(dir); |
6084 | if (ret) { | |
6085 | ret = btrfs_set_inode_index_count(dir); | |
6086 | if (ret) | |
6087 | return ret; | |
6088 | } | |
aec7477b JB |
6089 | } |
6090 | ||
00e4e6b3 | 6091 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
6092 | BTRFS_I(dir)->index_cnt++; |
6093 | ||
6094 | return ret; | |
6095 | } | |
6096 | ||
b0d5d10f CM |
6097 | static int btrfs_insert_inode_locked(struct inode *inode) |
6098 | { | |
6099 | struct btrfs_iget_args args; | |
6100 | args.location = &BTRFS_I(inode)->location; | |
6101 | args.root = BTRFS_I(inode)->root; | |
6102 | ||
6103 | return insert_inode_locked4(inode, | |
6104 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6105 | btrfs_find_actor, &args); | |
6106 | } | |
6107 | ||
39279cc3 CM |
6108 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6109 | struct btrfs_root *root, | |
aec7477b | 6110 | struct inode *dir, |
9c58309d | 6111 | const char *name, int name_len, |
175a4eb7 AV |
6112 | u64 ref_objectid, u64 objectid, |
6113 | umode_t mode, u64 *index) | |
39279cc3 CM |
6114 | { |
6115 | struct inode *inode; | |
5f39d397 | 6116 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6117 | struct btrfs_key *location; |
5f39d397 | 6118 | struct btrfs_path *path; |
9c58309d CM |
6119 | struct btrfs_inode_ref *ref; |
6120 | struct btrfs_key key[2]; | |
6121 | u32 sizes[2]; | |
ef3b9af5 | 6122 | int nitems = name ? 2 : 1; |
9c58309d | 6123 | unsigned long ptr; |
39279cc3 | 6124 | int ret; |
39279cc3 | 6125 | |
5f39d397 | 6126 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6127 | if (!path) |
6128 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6129 | |
39279cc3 | 6130 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
6131 | if (!inode) { |
6132 | btrfs_free_path(path); | |
39279cc3 | 6133 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6134 | } |
39279cc3 | 6135 | |
5762b5c9 FM |
6136 | /* |
6137 | * O_TMPFILE, set link count to 0, so that after this point, | |
6138 | * we fill in an inode item with the correct link count. | |
6139 | */ | |
6140 | if (!name) | |
6141 | set_nlink(inode, 0); | |
6142 | ||
581bb050 LZ |
6143 | /* |
6144 | * we have to initialize this early, so we can reclaim the inode | |
6145 | * number if we fail afterwards in this function. | |
6146 | */ | |
6147 | inode->i_ino = objectid; | |
6148 | ||
ef3b9af5 | 6149 | if (dir && name) { |
1abe9b8a | 6150 | trace_btrfs_inode_request(dir); |
6151 | ||
3de4586c | 6152 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 6153 | if (ret) { |
8fb27640 | 6154 | btrfs_free_path(path); |
09771430 | 6155 | iput(inode); |
aec7477b | 6156 | return ERR_PTR(ret); |
09771430 | 6157 | } |
ef3b9af5 FM |
6158 | } else if (dir) { |
6159 | *index = 0; | |
aec7477b JB |
6160 | } |
6161 | /* | |
6162 | * index_cnt is ignored for everything but a dir, | |
6163 | * btrfs_get_inode_index_count has an explanation for the magic | |
6164 | * number | |
6165 | */ | |
6166 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6167 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 6168 | BTRFS_I(inode)->root = root; |
e02119d5 | 6169 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6170 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6171 | |
5dc562c5 JB |
6172 | /* |
6173 | * We could have gotten an inode number from somebody who was fsynced | |
6174 | * and then removed in this same transaction, so let's just set full | |
6175 | * sync since it will be a full sync anyway and this will blow away the | |
6176 | * old info in the log. | |
6177 | */ | |
6178 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
6179 | ||
9c58309d | 6180 | key[0].objectid = objectid; |
962a298f | 6181 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6182 | key[0].offset = 0; |
6183 | ||
9c58309d | 6184 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6185 | |
6186 | if (name) { | |
6187 | /* | |
6188 | * Start new inodes with an inode_ref. This is slightly more | |
6189 | * efficient for small numbers of hard links since they will | |
6190 | * be packed into one item. Extended refs will kick in if we | |
6191 | * add more hard links than can fit in the ref item. | |
6192 | */ | |
6193 | key[1].objectid = objectid; | |
962a298f | 6194 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
6195 | key[1].offset = ref_objectid; |
6196 | ||
6197 | sizes[1] = name_len + sizeof(*ref); | |
6198 | } | |
9c58309d | 6199 | |
b0d5d10f CM |
6200 | location = &BTRFS_I(inode)->location; |
6201 | location->objectid = objectid; | |
6202 | location->offset = 0; | |
962a298f | 6203 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6204 | |
6205 | ret = btrfs_insert_inode_locked(inode); | |
6206 | if (ret < 0) | |
6207 | goto fail; | |
6208 | ||
b9473439 | 6209 | path->leave_spinning = 1; |
ef3b9af5 | 6210 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6211 | if (ret != 0) |
b0d5d10f | 6212 | goto fail_unlock; |
5f39d397 | 6213 | |
ecc11fab | 6214 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 6215 | inode_set_bytes(inode, 0); |
9cc97d64 | 6216 | |
04b285f3 | 6217 | inode->i_mtime = current_fs_time(inode->i_sb); |
9cc97d64 | 6218 | inode->i_atime = inode->i_mtime; |
6219 | inode->i_ctime = inode->i_mtime; | |
6220 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
6221 | ||
5f39d397 CM |
6222 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6223 | struct btrfs_inode_item); | |
293f7e07 LZ |
6224 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
6225 | sizeof(*inode_item)); | |
e02119d5 | 6226 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6227 | |
ef3b9af5 FM |
6228 | if (name) { |
6229 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6230 | struct btrfs_inode_ref); | |
6231 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6232 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6233 | ptr = (unsigned long)(ref + 1); | |
6234 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6235 | } | |
9c58309d | 6236 | |
5f39d397 CM |
6237 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6238 | btrfs_free_path(path); | |
6239 | ||
6cbff00f CH |
6240 | btrfs_inherit_iflags(inode, dir); |
6241 | ||
569254b0 | 6242 | if (S_ISREG(mode)) { |
94272164 CM |
6243 | if (btrfs_test_opt(root, NODATASUM)) |
6244 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
213490b3 | 6245 | if (btrfs_test_opt(root, NODATACOW)) |
f2bdf9a8 JB |
6246 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6247 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6248 | } |
6249 | ||
5d4f98a2 | 6250 | inode_tree_add(inode); |
1abe9b8a | 6251 | |
6252 | trace_btrfs_inode_new(inode); | |
1973f0fa | 6253 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 6254 | |
8ea05e3a AB |
6255 | btrfs_update_root_times(trans, root); |
6256 | ||
63541927 FDBM |
6257 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6258 | if (ret) | |
6259 | btrfs_err(root->fs_info, | |
6260 | "error inheriting props for ino %llu (root %llu): %d", | |
6261 | btrfs_ino(inode), root->root_key.objectid, ret); | |
6262 | ||
39279cc3 | 6263 | return inode; |
b0d5d10f CM |
6264 | |
6265 | fail_unlock: | |
6266 | unlock_new_inode(inode); | |
5f39d397 | 6267 | fail: |
ef3b9af5 | 6268 | if (dir && name) |
aec7477b | 6269 | BTRFS_I(dir)->index_cnt--; |
5f39d397 | 6270 | btrfs_free_path(path); |
09771430 | 6271 | iput(inode); |
5f39d397 | 6272 | return ERR_PTR(ret); |
39279cc3 CM |
6273 | } |
6274 | ||
6275 | static inline u8 btrfs_inode_type(struct inode *inode) | |
6276 | { | |
6277 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
6278 | } | |
6279 | ||
d352ac68 CM |
6280 | /* |
6281 | * utility function to add 'inode' into 'parent_inode' with | |
6282 | * a give name and a given sequence number. | |
6283 | * if 'add_backref' is true, also insert a backref from the | |
6284 | * inode to the parent directory. | |
6285 | */ | |
e02119d5 CM |
6286 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
6287 | struct inode *parent_inode, struct inode *inode, | |
6288 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 6289 | { |
4df27c4d | 6290 | int ret = 0; |
39279cc3 | 6291 | struct btrfs_key key; |
e02119d5 | 6292 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
6293 | u64 ino = btrfs_ino(inode); |
6294 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6295 | |
33345d01 | 6296 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
6297 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
6298 | } else { | |
33345d01 | 6299 | key.objectid = ino; |
962a298f | 6300 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6301 | key.offset = 0; |
6302 | } | |
6303 | ||
33345d01 | 6304 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
6305 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
6306 | key.objectid, root->root_key.objectid, | |
33345d01 | 6307 | parent_ino, index, name, name_len); |
4df27c4d | 6308 | } else if (add_backref) { |
33345d01 LZ |
6309 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6310 | parent_ino, index); | |
4df27c4d | 6311 | } |
39279cc3 | 6312 | |
79787eaa JM |
6313 | /* Nothing to clean up yet */ |
6314 | if (ret) | |
6315 | return ret; | |
4df27c4d | 6316 | |
79787eaa JM |
6317 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
6318 | parent_inode, &key, | |
6319 | btrfs_inode_type(inode), index); | |
9c52057c | 6320 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6321 | goto fail_dir_item; |
6322 | else if (ret) { | |
6323 | btrfs_abort_transaction(trans, root, ret); | |
6324 | return ret; | |
39279cc3 | 6325 | } |
79787eaa JM |
6326 | |
6327 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
6328 | name_len * 2); | |
0c4d2d95 | 6329 | inode_inc_iversion(parent_inode); |
04b285f3 DD |
6330 | parent_inode->i_mtime = parent_inode->i_ctime = |
6331 | current_fs_time(parent_inode->i_sb); | |
79787eaa JM |
6332 | ret = btrfs_update_inode(trans, root, parent_inode); |
6333 | if (ret) | |
6334 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 6335 | return ret; |
fe66a05a CM |
6336 | |
6337 | fail_dir_item: | |
6338 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6339 | u64 local_index; | |
6340 | int err; | |
6341 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
6342 | key.objectid, root->root_key.objectid, | |
6343 | parent_ino, &local_index, name, name_len); | |
6344 | ||
6345 | } else if (add_backref) { | |
6346 | u64 local_index; | |
6347 | int err; | |
6348 | ||
6349 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6350 | ino, parent_ino, &local_index); | |
6351 | } | |
6352 | return ret; | |
39279cc3 CM |
6353 | } |
6354 | ||
6355 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
6356 | struct inode *dir, struct dentry *dentry, |
6357 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 6358 | { |
a1b075d2 JB |
6359 | int err = btrfs_add_link(trans, dir, inode, |
6360 | dentry->d_name.name, dentry->d_name.len, | |
6361 | backref, index); | |
39279cc3 CM |
6362 | if (err > 0) |
6363 | err = -EEXIST; | |
6364 | return err; | |
6365 | } | |
6366 | ||
618e21d5 | 6367 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6368 | umode_t mode, dev_t rdev) |
618e21d5 JB |
6369 | { |
6370 | struct btrfs_trans_handle *trans; | |
6371 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6372 | struct inode *inode = NULL; |
618e21d5 JB |
6373 | int err; |
6374 | int drop_inode = 0; | |
6375 | u64 objectid; | |
00e4e6b3 | 6376 | u64 index = 0; |
618e21d5 | 6377 | |
9ed74f2d JB |
6378 | /* |
6379 | * 2 for inode item and ref | |
6380 | * 2 for dir items | |
6381 | * 1 for xattr if selinux is on | |
6382 | */ | |
a22285a6 YZ |
6383 | trans = btrfs_start_transaction(root, 5); |
6384 | if (IS_ERR(trans)) | |
6385 | return PTR_ERR(trans); | |
1832a6d5 | 6386 | |
581bb050 LZ |
6387 | err = btrfs_find_free_ino(root, &objectid); |
6388 | if (err) | |
6389 | goto out_unlock; | |
6390 | ||
aec7477b | 6391 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6392 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6393 | mode, &index); |
7cf96da3 TI |
6394 | if (IS_ERR(inode)) { |
6395 | err = PTR_ERR(inode); | |
618e21d5 | 6396 | goto out_unlock; |
7cf96da3 | 6397 | } |
618e21d5 | 6398 | |
ad19db71 CS |
6399 | /* |
6400 | * If the active LSM wants to access the inode during | |
6401 | * d_instantiate it needs these. Smack checks to see | |
6402 | * if the filesystem supports xattrs by looking at the | |
6403 | * ops vector. | |
6404 | */ | |
ad19db71 | 6405 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6406 | init_special_inode(inode, inode->i_mode, rdev); |
6407 | ||
6408 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6409 | if (err) |
b0d5d10f CM |
6410 | goto out_unlock_inode; |
6411 | ||
6412 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); | |
6413 | if (err) { | |
6414 | goto out_unlock_inode; | |
6415 | } else { | |
1b4ab1bb | 6416 | btrfs_update_inode(trans, root, inode); |
b0d5d10f | 6417 | unlock_new_inode(inode); |
08c422c2 | 6418 | d_instantiate(dentry, inode); |
618e21d5 | 6419 | } |
b0d5d10f | 6420 | |
618e21d5 | 6421 | out_unlock: |
7ad85bb7 | 6422 | btrfs_end_transaction(trans, root); |
c581afc8 | 6423 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6424 | btrfs_btree_balance_dirty(root); |
618e21d5 JB |
6425 | if (drop_inode) { |
6426 | inode_dec_link_count(inode); | |
6427 | iput(inode); | |
6428 | } | |
618e21d5 | 6429 | return err; |
b0d5d10f CM |
6430 | |
6431 | out_unlock_inode: | |
6432 | drop_inode = 1; | |
6433 | unlock_new_inode(inode); | |
6434 | goto out_unlock; | |
6435 | ||
618e21d5 JB |
6436 | } |
6437 | ||
39279cc3 | 6438 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6439 | umode_t mode, bool excl) |
39279cc3 CM |
6440 | { |
6441 | struct btrfs_trans_handle *trans; | |
6442 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6443 | struct inode *inode = NULL; |
43baa579 | 6444 | int drop_inode_on_err = 0; |
a22285a6 | 6445 | int err; |
39279cc3 | 6446 | u64 objectid; |
00e4e6b3 | 6447 | u64 index = 0; |
39279cc3 | 6448 | |
9ed74f2d JB |
6449 | /* |
6450 | * 2 for inode item and ref | |
6451 | * 2 for dir items | |
6452 | * 1 for xattr if selinux is on | |
6453 | */ | |
a22285a6 YZ |
6454 | trans = btrfs_start_transaction(root, 5); |
6455 | if (IS_ERR(trans)) | |
6456 | return PTR_ERR(trans); | |
9ed74f2d | 6457 | |
581bb050 LZ |
6458 | err = btrfs_find_free_ino(root, &objectid); |
6459 | if (err) | |
6460 | goto out_unlock; | |
6461 | ||
aec7477b | 6462 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6463 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6464 | mode, &index); |
7cf96da3 TI |
6465 | if (IS_ERR(inode)) { |
6466 | err = PTR_ERR(inode); | |
39279cc3 | 6467 | goto out_unlock; |
7cf96da3 | 6468 | } |
43baa579 | 6469 | drop_inode_on_err = 1; |
ad19db71 CS |
6470 | /* |
6471 | * If the active LSM wants to access the inode during | |
6472 | * d_instantiate it needs these. Smack checks to see | |
6473 | * if the filesystem supports xattrs by looking at the | |
6474 | * ops vector. | |
6475 | */ | |
6476 | inode->i_fop = &btrfs_file_operations; | |
6477 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6478 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6479 | |
6480 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6481 | if (err) | |
6482 | goto out_unlock_inode; | |
6483 | ||
6484 | err = btrfs_update_inode(trans, root, inode); | |
6485 | if (err) | |
6486 | goto out_unlock_inode; | |
ad19db71 | 6487 | |
a1b075d2 | 6488 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 6489 | if (err) |
b0d5d10f | 6490 | goto out_unlock_inode; |
43baa579 | 6491 | |
43baa579 | 6492 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
b0d5d10f | 6493 | unlock_new_inode(inode); |
43baa579 FB |
6494 | d_instantiate(dentry, inode); |
6495 | ||
39279cc3 | 6496 | out_unlock: |
7ad85bb7 | 6497 | btrfs_end_transaction(trans, root); |
43baa579 | 6498 | if (err && drop_inode_on_err) { |
39279cc3 CM |
6499 | inode_dec_link_count(inode); |
6500 | iput(inode); | |
6501 | } | |
c581afc8 | 6502 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6503 | btrfs_btree_balance_dirty(root); |
39279cc3 | 6504 | return err; |
b0d5d10f CM |
6505 | |
6506 | out_unlock_inode: | |
6507 | unlock_new_inode(inode); | |
6508 | goto out_unlock; | |
6509 | ||
39279cc3 CM |
6510 | } |
6511 | ||
6512 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6513 | struct dentry *dentry) | |
6514 | { | |
271dba45 | 6515 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6516 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6517 | struct inode *inode = d_inode(old_dentry); |
00e4e6b3 | 6518 | u64 index; |
39279cc3 CM |
6519 | int err; |
6520 | int drop_inode = 0; | |
6521 | ||
4a8be425 TH |
6522 | /* do not allow sys_link's with other subvols of the same device */ |
6523 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 6524 | return -EXDEV; |
4a8be425 | 6525 | |
f186373f | 6526 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6527 | return -EMLINK; |
4a8be425 | 6528 | |
3de4586c | 6529 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
6530 | if (err) |
6531 | goto fail; | |
6532 | ||
a22285a6 | 6533 | /* |
7e6b6465 | 6534 | * 2 items for inode and inode ref |
a22285a6 | 6535 | * 2 items for dir items |
7e6b6465 | 6536 | * 1 item for parent inode |
a22285a6 | 6537 | */ |
7e6b6465 | 6538 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
6539 | if (IS_ERR(trans)) { |
6540 | err = PTR_ERR(trans); | |
271dba45 | 6541 | trans = NULL; |
a22285a6 YZ |
6542 | goto fail; |
6543 | } | |
5f39d397 | 6544 | |
67de1176 MX |
6545 | /* There are several dir indexes for this inode, clear the cache. */ |
6546 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6547 | inc_nlink(inode); |
0c4d2d95 | 6548 | inode_inc_iversion(inode); |
04b285f3 | 6549 | inode->i_ctime = current_fs_time(inode->i_sb); |
7de9c6ee | 6550 | ihold(inode); |
e9976151 | 6551 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6552 | |
a1b075d2 | 6553 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 6554 | |
a5719521 | 6555 | if (err) { |
54aa1f4d | 6556 | drop_inode = 1; |
a5719521 | 6557 | } else { |
10d9f309 | 6558 | struct dentry *parent = dentry->d_parent; |
a5719521 | 6559 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6560 | if (err) |
6561 | goto fail; | |
ef3b9af5 FM |
6562 | if (inode->i_nlink == 1) { |
6563 | /* | |
6564 | * If new hard link count is 1, it's a file created | |
6565 | * with open(2) O_TMPFILE flag. | |
6566 | */ | |
6567 | err = btrfs_orphan_del(trans, inode); | |
6568 | if (err) | |
6569 | goto fail; | |
6570 | } | |
08c422c2 | 6571 | d_instantiate(dentry, inode); |
6a912213 | 6572 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 6573 | } |
39279cc3 | 6574 | |
c581afc8 | 6575 | btrfs_balance_delayed_items(root); |
1832a6d5 | 6576 | fail: |
271dba45 FM |
6577 | if (trans) |
6578 | btrfs_end_transaction(trans, root); | |
39279cc3 CM |
6579 | if (drop_inode) { |
6580 | inode_dec_link_count(inode); | |
6581 | iput(inode); | |
6582 | } | |
b53d3f5d | 6583 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
6584 | return err; |
6585 | } | |
6586 | ||
18bb1db3 | 6587 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6588 | { |
b9d86667 | 6589 | struct inode *inode = NULL; |
39279cc3 CM |
6590 | struct btrfs_trans_handle *trans; |
6591 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6592 | int err = 0; | |
6593 | int drop_on_err = 0; | |
b9d86667 | 6594 | u64 objectid = 0; |
00e4e6b3 | 6595 | u64 index = 0; |
39279cc3 | 6596 | |
9ed74f2d JB |
6597 | /* |
6598 | * 2 items for inode and ref | |
6599 | * 2 items for dir items | |
6600 | * 1 for xattr if selinux is on | |
6601 | */ | |
a22285a6 YZ |
6602 | trans = btrfs_start_transaction(root, 5); |
6603 | if (IS_ERR(trans)) | |
6604 | return PTR_ERR(trans); | |
39279cc3 | 6605 | |
581bb050 LZ |
6606 | err = btrfs_find_free_ino(root, &objectid); |
6607 | if (err) | |
6608 | goto out_fail; | |
6609 | ||
aec7477b | 6610 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6611 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6612 | S_IFDIR | mode, &index); |
39279cc3 CM |
6613 | if (IS_ERR(inode)) { |
6614 | err = PTR_ERR(inode); | |
6615 | goto out_fail; | |
6616 | } | |
5f39d397 | 6617 | |
39279cc3 | 6618 | drop_on_err = 1; |
b0d5d10f CM |
6619 | /* these must be set before we unlock the inode */ |
6620 | inode->i_op = &btrfs_dir_inode_operations; | |
6621 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6622 | |
2a7dba39 | 6623 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6624 | if (err) |
b0d5d10f | 6625 | goto out_fail_inode; |
39279cc3 | 6626 | |
dbe674a9 | 6627 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
6628 | err = btrfs_update_inode(trans, root, inode); |
6629 | if (err) | |
b0d5d10f | 6630 | goto out_fail_inode; |
5f39d397 | 6631 | |
a1b075d2 JB |
6632 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
6633 | dentry->d_name.len, 0, index); | |
39279cc3 | 6634 | if (err) |
b0d5d10f | 6635 | goto out_fail_inode; |
5f39d397 | 6636 | |
39279cc3 | 6637 | d_instantiate(dentry, inode); |
b0d5d10f CM |
6638 | /* |
6639 | * mkdir is special. We're unlocking after we call d_instantiate | |
6640 | * to avoid a race with nfsd calling d_instantiate. | |
6641 | */ | |
6642 | unlock_new_inode(inode); | |
39279cc3 | 6643 | drop_on_err = 0; |
39279cc3 CM |
6644 | |
6645 | out_fail: | |
7ad85bb7 | 6646 | btrfs_end_transaction(trans, root); |
c7cfb8a5 WS |
6647 | if (drop_on_err) { |
6648 | inode_dec_link_count(inode); | |
39279cc3 | 6649 | iput(inode); |
c7cfb8a5 | 6650 | } |
c581afc8 | 6651 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6652 | btrfs_btree_balance_dirty(root); |
39279cc3 | 6653 | return err; |
b0d5d10f CM |
6654 | |
6655 | out_fail_inode: | |
6656 | unlock_new_inode(inode); | |
6657 | goto out_fail; | |
39279cc3 CM |
6658 | } |
6659 | ||
e6c4efd8 QW |
6660 | /* Find next extent map of a given extent map, caller needs to ensure locks */ |
6661 | static struct extent_map *next_extent_map(struct extent_map *em) | |
6662 | { | |
6663 | struct rb_node *next; | |
6664 | ||
6665 | next = rb_next(&em->rb_node); | |
6666 | if (!next) | |
6667 | return NULL; | |
6668 | return container_of(next, struct extent_map, rb_node); | |
6669 | } | |
6670 | ||
6671 | static struct extent_map *prev_extent_map(struct extent_map *em) | |
6672 | { | |
6673 | struct rb_node *prev; | |
6674 | ||
6675 | prev = rb_prev(&em->rb_node); | |
6676 | if (!prev) | |
6677 | return NULL; | |
6678 | return container_of(prev, struct extent_map, rb_node); | |
6679 | } | |
6680 | ||
d352ac68 | 6681 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
e6c4efd8 | 6682 | * the existing extent is the nearest extent to map_start, |
d352ac68 | 6683 | * and an extent that you want to insert, deal with overlap and insert |
e6c4efd8 | 6684 | * the best fitted new extent into the tree. |
d352ac68 | 6685 | */ |
3b951516 CM |
6686 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
6687 | struct extent_map *existing, | |
e6dcd2dc | 6688 | struct extent_map *em, |
51f395ad | 6689 | u64 map_start) |
3b951516 | 6690 | { |
e6c4efd8 QW |
6691 | struct extent_map *prev; |
6692 | struct extent_map *next; | |
6693 | u64 start; | |
6694 | u64 end; | |
3b951516 | 6695 | u64 start_diff; |
3b951516 | 6696 | |
e6dcd2dc | 6697 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
e6c4efd8 QW |
6698 | |
6699 | if (existing->start > map_start) { | |
6700 | next = existing; | |
6701 | prev = prev_extent_map(next); | |
6702 | } else { | |
6703 | prev = existing; | |
6704 | next = next_extent_map(prev); | |
6705 | } | |
6706 | ||
6707 | start = prev ? extent_map_end(prev) : em->start; | |
6708 | start = max_t(u64, start, em->start); | |
6709 | end = next ? next->start : extent_map_end(em); | |
6710 | end = min_t(u64, end, extent_map_end(em)); | |
6711 | start_diff = start - em->start; | |
6712 | em->start = start; | |
6713 | em->len = end - start; | |
c8b97818 CM |
6714 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
6715 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 6716 | em->block_start += start_diff; |
c8b97818 CM |
6717 | em->block_len -= start_diff; |
6718 | } | |
09a2a8f9 | 6719 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6720 | } |
6721 | ||
c8b97818 | 6722 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6723 | struct page *page, |
c8b97818 CM |
6724 | size_t pg_offset, u64 extent_offset, |
6725 | struct btrfs_file_extent_item *item) | |
6726 | { | |
6727 | int ret; | |
6728 | struct extent_buffer *leaf = path->nodes[0]; | |
6729 | char *tmp; | |
6730 | size_t max_size; | |
6731 | unsigned long inline_size; | |
6732 | unsigned long ptr; | |
261507a0 | 6733 | int compress_type; |
c8b97818 CM |
6734 | |
6735 | WARN_ON(pg_offset != 0); | |
261507a0 | 6736 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6737 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6738 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6739 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6740 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6741 | if (!tmp) |
6742 | return -ENOMEM; | |
c8b97818 CM |
6743 | ptr = btrfs_file_extent_inline_start(item); |
6744 | ||
6745 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6746 | ||
09cbfeaf | 6747 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6748 | ret = btrfs_decompress(compress_type, tmp, page, |
6749 | extent_offset, inline_size, max_size); | |
c8b97818 | 6750 | kfree(tmp); |
166ae5a4 | 6751 | return ret; |
c8b97818 CM |
6752 | } |
6753 | ||
d352ac68 CM |
6754 | /* |
6755 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
6756 | * the ugly parts come from merging extents from the disk with the in-ram |
6757 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
6758 | * where the in-ram extents might be locked pending data=ordered completion. |
6759 | * | |
6760 | * This also copies inline extents directly into the page. | |
6761 | */ | |
d397712b | 6762 | |
a52d9a80 | 6763 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 6764 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
6765 | int create) |
6766 | { | |
6767 | int ret; | |
6768 | int err = 0; | |
a52d9a80 CM |
6769 | u64 extent_start = 0; |
6770 | u64 extent_end = 0; | |
33345d01 | 6771 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 6772 | u32 found_type; |
f421950f | 6773 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
6774 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6775 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
6776 | struct extent_buffer *leaf; |
6777 | struct btrfs_key found_key; | |
a52d9a80 CM |
6778 | struct extent_map *em = NULL; |
6779 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 6780 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 6781 | struct btrfs_trans_handle *trans = NULL; |
7ffbb598 | 6782 | const bool new_inline = !page || create; |
a52d9a80 | 6783 | |
a52d9a80 | 6784 | again: |
890871be | 6785 | read_lock(&em_tree->lock); |
d1310b2e | 6786 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
6787 | if (em) |
6788 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 6789 | read_unlock(&em_tree->lock); |
d1310b2e | 6790 | |
a52d9a80 | 6791 | if (em) { |
e1c4b745 CM |
6792 | if (em->start > start || em->start + em->len <= start) |
6793 | free_extent_map(em); | |
6794 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6795 | free_extent_map(em); |
6796 | else | |
6797 | goto out; | |
a52d9a80 | 6798 | } |
172ddd60 | 6799 | em = alloc_extent_map(); |
a52d9a80 | 6800 | if (!em) { |
d1310b2e CM |
6801 | err = -ENOMEM; |
6802 | goto out; | |
a52d9a80 | 6803 | } |
e6dcd2dc | 6804 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 6805 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6806 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6807 | em->len = (u64)-1; |
c8b97818 | 6808 | em->block_len = (u64)-1; |
f421950f CM |
6809 | |
6810 | if (!path) { | |
6811 | path = btrfs_alloc_path(); | |
026fd317 JB |
6812 | if (!path) { |
6813 | err = -ENOMEM; | |
6814 | goto out; | |
6815 | } | |
6816 | /* | |
6817 | * Chances are we'll be called again, so go ahead and do | |
6818 | * readahead | |
6819 | */ | |
e4058b54 | 6820 | path->reada = READA_FORWARD; |
f421950f CM |
6821 | } |
6822 | ||
179e29e4 CM |
6823 | ret = btrfs_lookup_file_extent(trans, root, path, |
6824 | objectid, start, trans != NULL); | |
a52d9a80 CM |
6825 | if (ret < 0) { |
6826 | err = ret; | |
6827 | goto out; | |
6828 | } | |
6829 | ||
6830 | if (ret != 0) { | |
6831 | if (path->slots[0] == 0) | |
6832 | goto not_found; | |
6833 | path->slots[0]--; | |
6834 | } | |
6835 | ||
5f39d397 CM |
6836 | leaf = path->nodes[0]; |
6837 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6838 | struct btrfs_file_extent_item); |
a52d9a80 | 6839 | /* are we inside the extent that was found? */ |
5f39d397 | 6840 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
962a298f | 6841 | found_type = found_key.type; |
5f39d397 | 6842 | if (found_key.objectid != objectid || |
a52d9a80 | 6843 | found_type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6844 | /* |
6845 | * If we backup past the first extent we want to move forward | |
6846 | * and see if there is an extent in front of us, otherwise we'll | |
6847 | * say there is a hole for our whole search range which can | |
6848 | * cause problems. | |
6849 | */ | |
6850 | extent_end = start; | |
6851 | goto next; | |
a52d9a80 CM |
6852 | } |
6853 | ||
5f39d397 CM |
6854 | found_type = btrfs_file_extent_type(leaf, item); |
6855 | extent_start = found_key.offset; | |
d899e052 YZ |
6856 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6857 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6858 | extent_end = extent_start + |
db94535d | 6859 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
6860 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
6861 | size_t size; | |
514ac8ad | 6862 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
fda2832f | 6863 | extent_end = ALIGN(extent_start + size, root->sectorsize); |
9036c102 | 6864 | } |
25a50341 | 6865 | next: |
9036c102 YZ |
6866 | if (start >= extent_end) { |
6867 | path->slots[0]++; | |
6868 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6869 | ret = btrfs_next_leaf(root, path); | |
6870 | if (ret < 0) { | |
6871 | err = ret; | |
6872 | goto out; | |
a52d9a80 | 6873 | } |
9036c102 YZ |
6874 | if (ret > 0) |
6875 | goto not_found; | |
6876 | leaf = path->nodes[0]; | |
a52d9a80 | 6877 | } |
9036c102 YZ |
6878 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6879 | if (found_key.objectid != objectid || | |
6880 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6881 | goto not_found; | |
6882 | if (start + len <= found_key.offset) | |
6883 | goto not_found; | |
e2eca69d WS |
6884 | if (start > found_key.offset) |
6885 | goto next; | |
9036c102 | 6886 | em->start = start; |
70c8a91c | 6887 | em->orig_start = start; |
9036c102 YZ |
6888 | em->len = found_key.offset - start; |
6889 | goto not_found_em; | |
6890 | } | |
6891 | ||
7ffbb598 FM |
6892 | btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em); |
6893 | ||
d899e052 YZ |
6894 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6895 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 CM |
6896 | goto insert; |
6897 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 6898 | unsigned long ptr; |
a52d9a80 | 6899 | char *map; |
3326d1b0 CM |
6900 | size_t size; |
6901 | size_t extent_offset; | |
6902 | size_t copy_size; | |
a52d9a80 | 6903 | |
7ffbb598 | 6904 | if (new_inline) |
689f9346 | 6905 | goto out; |
5f39d397 | 6906 | |
514ac8ad | 6907 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
9036c102 | 6908 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
6909 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
6910 | size - extent_offset); | |
3326d1b0 | 6911 | em->start = extent_start + extent_offset; |
fda2832f | 6912 | em->len = ALIGN(copy_size, root->sectorsize); |
b4939680 | 6913 | em->orig_block_len = em->len; |
70c8a91c | 6914 | em->orig_start = em->start; |
689f9346 | 6915 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 6916 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
6917 | if (btrfs_file_extent_compression(leaf, item) != |
6918 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 6919 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 6920 | extent_offset, item); |
166ae5a4 ZB |
6921 | if (ret) { |
6922 | err = ret; | |
6923 | goto out; | |
6924 | } | |
c8b97818 CM |
6925 | } else { |
6926 | map = kmap(page); | |
6927 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6928 | copy_size); | |
09cbfeaf | 6929 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 6930 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 6931 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
6932 | copy_size); |
6933 | } | |
c8b97818 CM |
6934 | kunmap(page); |
6935 | } | |
179e29e4 CM |
6936 | flush_dcache_page(page); |
6937 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 6938 | BUG(); |
179e29e4 CM |
6939 | if (!trans) { |
6940 | kunmap(page); | |
6941 | free_extent_map(em); | |
6942 | em = NULL; | |
ff5714cc | 6943 | |
b3b4aa74 | 6944 | btrfs_release_path(path); |
7a7eaa40 | 6945 | trans = btrfs_join_transaction(root); |
ff5714cc | 6946 | |
3612b495 TI |
6947 | if (IS_ERR(trans)) |
6948 | return ERR_CAST(trans); | |
179e29e4 CM |
6949 | goto again; |
6950 | } | |
c8b97818 | 6951 | map = kmap(page); |
70dec807 | 6952 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 6953 | copy_size); |
c8b97818 | 6954 | kunmap(page); |
179e29e4 | 6955 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 6956 | } |
d1310b2e | 6957 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6958 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 6959 | goto insert; |
a52d9a80 CM |
6960 | } |
6961 | not_found: | |
6962 | em->start = start; | |
70c8a91c | 6963 | em->orig_start = start; |
d1310b2e | 6964 | em->len = len; |
a52d9a80 | 6965 | not_found_em: |
5f39d397 | 6966 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 6967 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 6968 | insert: |
b3b4aa74 | 6969 | btrfs_release_path(path); |
d1310b2e | 6970 | if (em->start > start || extent_map_end(em) <= start) { |
c2cf52eb | 6971 | btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]", |
c1c9ff7c | 6972 | em->start, em->len, start, len); |
a52d9a80 CM |
6973 | err = -EIO; |
6974 | goto out; | |
6975 | } | |
d1310b2e CM |
6976 | |
6977 | err = 0; | |
890871be | 6978 | write_lock(&em_tree->lock); |
09a2a8f9 | 6979 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6980 | /* it is possible that someone inserted the extent into the tree |
6981 | * while we had the lock dropped. It is also possible that | |
6982 | * an overlapping map exists in the tree | |
6983 | */ | |
a52d9a80 | 6984 | if (ret == -EEXIST) { |
3b951516 | 6985 | struct extent_map *existing; |
e6dcd2dc CM |
6986 | |
6987 | ret = 0; | |
6988 | ||
e6c4efd8 QW |
6989 | existing = search_extent_mapping(em_tree, start, len); |
6990 | /* | |
6991 | * existing will always be non-NULL, since there must be | |
6992 | * extent causing the -EEXIST. | |
6993 | */ | |
8dff9c85 CM |
6994 | if (existing->start == em->start && |
6995 | extent_map_end(existing) == extent_map_end(em) && | |
6996 | em->block_start == existing->block_start) { | |
6997 | /* | |
6998 | * these two extents are the same, it happens | |
6999 | * with inlines especially | |
7000 | */ | |
7001 | free_extent_map(em); | |
7002 | em = existing; | |
7003 | err = 0; | |
7004 | ||
7005 | } else if (start >= extent_map_end(existing) || | |
32be3a1a | 7006 | start <= existing->start) { |
e6c4efd8 QW |
7007 | /* |
7008 | * The existing extent map is the one nearest to | |
7009 | * the [start, start + len) range which overlaps | |
7010 | */ | |
7011 | err = merge_extent_mapping(em_tree, existing, | |
7012 | em, start); | |
e1c4b745 | 7013 | free_extent_map(existing); |
e6c4efd8 | 7014 | if (err) { |
3b951516 CM |
7015 | free_extent_map(em); |
7016 | em = NULL; | |
7017 | } | |
7018 | } else { | |
7019 | free_extent_map(em); | |
7020 | em = existing; | |
e6dcd2dc | 7021 | err = 0; |
a52d9a80 | 7022 | } |
a52d9a80 | 7023 | } |
890871be | 7024 | write_unlock(&em_tree->lock); |
a52d9a80 | 7025 | out: |
1abe9b8a | 7026 | |
4cd8587c | 7027 | trace_btrfs_get_extent(root, em); |
1abe9b8a | 7028 | |
527afb44 | 7029 | btrfs_free_path(path); |
a52d9a80 CM |
7030 | if (trans) { |
7031 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 7032 | if (!err) |
a52d9a80 CM |
7033 | err = ret; |
7034 | } | |
a52d9a80 CM |
7035 | if (err) { |
7036 | free_extent_map(em); | |
a52d9a80 CM |
7037 | return ERR_PTR(err); |
7038 | } | |
79787eaa | 7039 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
7040 | return em; |
7041 | } | |
7042 | ||
ec29ed5b CM |
7043 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
7044 | size_t pg_offset, u64 start, u64 len, | |
7045 | int create) | |
7046 | { | |
7047 | struct extent_map *em; | |
7048 | struct extent_map *hole_em = NULL; | |
7049 | u64 range_start = start; | |
7050 | u64 end; | |
7051 | u64 found; | |
7052 | u64 found_end; | |
7053 | int err = 0; | |
7054 | ||
7055 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
7056 | if (IS_ERR(em)) | |
7057 | return em; | |
7058 | if (em) { | |
7059 | /* | |
f9e4fb53 LB |
7060 | * if our em maps to |
7061 | * - a hole or | |
7062 | * - a pre-alloc extent, | |
7063 | * there might actually be delalloc bytes behind it. | |
ec29ed5b | 7064 | */ |
f9e4fb53 LB |
7065 | if (em->block_start != EXTENT_MAP_HOLE && |
7066 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
ec29ed5b CM |
7067 | return em; |
7068 | else | |
7069 | hole_em = em; | |
7070 | } | |
7071 | ||
7072 | /* check to see if we've wrapped (len == -1 or similar) */ | |
7073 | end = start + len; | |
7074 | if (end < start) | |
7075 | end = (u64)-1; | |
7076 | else | |
7077 | end -= 1; | |
7078 | ||
7079 | em = NULL; | |
7080 | ||
7081 | /* ok, we didn't find anything, lets look for delalloc */ | |
7082 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
7083 | end, len, EXTENT_DELALLOC, 1); | |
7084 | found_end = range_start + found; | |
7085 | if (found_end < range_start) | |
7086 | found_end = (u64)-1; | |
7087 | ||
7088 | /* | |
7089 | * we didn't find anything useful, return | |
7090 | * the original results from get_extent() | |
7091 | */ | |
7092 | if (range_start > end || found_end <= start) { | |
7093 | em = hole_em; | |
7094 | hole_em = NULL; | |
7095 | goto out; | |
7096 | } | |
7097 | ||
7098 | /* adjust the range_start to make sure it doesn't | |
7099 | * go backwards from the start they passed in | |
7100 | */ | |
67871254 | 7101 | range_start = max(start, range_start); |
ec29ed5b CM |
7102 | found = found_end - range_start; |
7103 | ||
7104 | if (found > 0) { | |
7105 | u64 hole_start = start; | |
7106 | u64 hole_len = len; | |
7107 | ||
172ddd60 | 7108 | em = alloc_extent_map(); |
ec29ed5b CM |
7109 | if (!em) { |
7110 | err = -ENOMEM; | |
7111 | goto out; | |
7112 | } | |
7113 | /* | |
7114 | * when btrfs_get_extent can't find anything it | |
7115 | * returns one huge hole | |
7116 | * | |
7117 | * make sure what it found really fits our range, and | |
7118 | * adjust to make sure it is based on the start from | |
7119 | * the caller | |
7120 | */ | |
7121 | if (hole_em) { | |
7122 | u64 calc_end = extent_map_end(hole_em); | |
7123 | ||
7124 | if (calc_end <= start || (hole_em->start > end)) { | |
7125 | free_extent_map(hole_em); | |
7126 | hole_em = NULL; | |
7127 | } else { | |
7128 | hole_start = max(hole_em->start, start); | |
7129 | hole_len = calc_end - hole_start; | |
7130 | } | |
7131 | } | |
7132 | em->bdev = NULL; | |
7133 | if (hole_em && range_start > hole_start) { | |
7134 | /* our hole starts before our delalloc, so we | |
7135 | * have to return just the parts of the hole | |
7136 | * that go until the delalloc starts | |
7137 | */ | |
7138 | em->len = min(hole_len, | |
7139 | range_start - hole_start); | |
7140 | em->start = hole_start; | |
7141 | em->orig_start = hole_start; | |
7142 | /* | |
7143 | * don't adjust block start at all, | |
7144 | * it is fixed at EXTENT_MAP_HOLE | |
7145 | */ | |
7146 | em->block_start = hole_em->block_start; | |
7147 | em->block_len = hole_len; | |
f9e4fb53 LB |
7148 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
7149 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
7150 | } else { |
7151 | em->start = range_start; | |
7152 | em->len = found; | |
7153 | em->orig_start = range_start; | |
7154 | em->block_start = EXTENT_MAP_DELALLOC; | |
7155 | em->block_len = found; | |
7156 | } | |
7157 | } else if (hole_em) { | |
7158 | return hole_em; | |
7159 | } | |
7160 | out: | |
7161 | ||
7162 | free_extent_map(hole_em); | |
7163 | if (err) { | |
7164 | free_extent_map(em); | |
7165 | return ERR_PTR(err); | |
7166 | } | |
7167 | return em; | |
7168 | } | |
7169 | ||
5f9a8a51 FM |
7170 | static struct extent_map *btrfs_create_dio_extent(struct inode *inode, |
7171 | const u64 start, | |
7172 | const u64 len, | |
7173 | const u64 orig_start, | |
7174 | const u64 block_start, | |
7175 | const u64 block_len, | |
7176 | const u64 orig_block_len, | |
7177 | const u64 ram_bytes, | |
7178 | const int type) | |
7179 | { | |
7180 | struct extent_map *em = NULL; | |
7181 | int ret; | |
7182 | ||
7183 | down_read(&BTRFS_I(inode)->dio_sem); | |
7184 | if (type != BTRFS_ORDERED_NOCOW) { | |
7185 | em = create_pinned_em(inode, start, len, orig_start, | |
7186 | block_start, block_len, orig_block_len, | |
7187 | ram_bytes, type); | |
7188 | if (IS_ERR(em)) | |
7189 | goto out; | |
7190 | } | |
7191 | ret = btrfs_add_ordered_extent_dio(inode, start, block_start, | |
7192 | len, block_len, type); | |
7193 | if (ret) { | |
7194 | if (em) { | |
7195 | free_extent_map(em); | |
7196 | btrfs_drop_extent_cache(inode, start, | |
7197 | start + len - 1, 0); | |
7198 | } | |
7199 | em = ERR_PTR(ret); | |
7200 | } | |
7201 | out: | |
7202 | up_read(&BTRFS_I(inode)->dio_sem); | |
7203 | ||
7204 | return em; | |
7205 | } | |
7206 | ||
4b46fce2 JB |
7207 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
7208 | u64 start, u64 len) | |
7209 | { | |
7210 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
70c8a91c | 7211 | struct extent_map *em; |
4b46fce2 JB |
7212 | struct btrfs_key ins; |
7213 | u64 alloc_hint; | |
7214 | int ret; | |
4b46fce2 | 7215 | |
4b46fce2 | 7216 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
00361589 | 7217 | ret = btrfs_reserve_extent(root, len, root->sectorsize, 0, |
e570fd27 | 7218 | alloc_hint, &ins, 1, 1); |
00361589 JB |
7219 | if (ret) |
7220 | return ERR_PTR(ret); | |
4b46fce2 | 7221 | |
5f9a8a51 FM |
7222 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
7223 | ins.objectid, ins.offset, ins.offset, | |
7224 | ins.offset, 0); | |
9cfa3e34 | 7225 | btrfs_dec_block_group_reservations(root->fs_info, ins.objectid); |
5f9a8a51 | 7226 | if (IS_ERR(em)) |
e570fd27 | 7227 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
de0ee0ed | 7228 | |
4b46fce2 JB |
7229 | return em; |
7230 | } | |
7231 | ||
46bfbb5c CM |
7232 | /* |
7233 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
7234 | * block must be cow'd | |
7235 | */ | |
00361589 | 7236 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
7237 | u64 *orig_start, u64 *orig_block_len, |
7238 | u64 *ram_bytes) | |
46bfbb5c | 7239 | { |
00361589 | 7240 | struct btrfs_trans_handle *trans; |
46bfbb5c CM |
7241 | struct btrfs_path *path; |
7242 | int ret; | |
7243 | struct extent_buffer *leaf; | |
7244 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7245 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7246 | struct btrfs_file_extent_item *fi; |
7247 | struct btrfs_key key; | |
7248 | u64 disk_bytenr; | |
7249 | u64 backref_offset; | |
7250 | u64 extent_end; | |
7251 | u64 num_bytes; | |
7252 | int slot; | |
7253 | int found_type; | |
7ee9e440 | 7254 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7255 | |
46bfbb5c CM |
7256 | path = btrfs_alloc_path(); |
7257 | if (!path) | |
7258 | return -ENOMEM; | |
7259 | ||
00361589 | 7260 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), |
46bfbb5c CM |
7261 | offset, 0); |
7262 | if (ret < 0) | |
7263 | goto out; | |
7264 | ||
7265 | slot = path->slots[0]; | |
7266 | if (ret == 1) { | |
7267 | if (slot == 0) { | |
7268 | /* can't find the item, must cow */ | |
7269 | ret = 0; | |
7270 | goto out; | |
7271 | } | |
7272 | slot--; | |
7273 | } | |
7274 | ret = 0; | |
7275 | leaf = path->nodes[0]; | |
7276 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 7277 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
7278 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7279 | /* not our file or wrong item type, must cow */ | |
7280 | goto out; | |
7281 | } | |
7282 | ||
7283 | if (key.offset > offset) { | |
7284 | /* Wrong offset, must cow */ | |
7285 | goto out; | |
7286 | } | |
7287 | ||
7288 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7289 | found_type = btrfs_file_extent_type(leaf, fi); | |
7290 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7291 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7292 | /* not a regular extent, must cow */ | |
7293 | goto out; | |
7294 | } | |
7ee9e440 JB |
7295 | |
7296 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7297 | goto out; | |
7298 | ||
e77751aa MX |
7299 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7300 | if (extent_end <= offset) | |
7301 | goto out; | |
7302 | ||
46bfbb5c | 7303 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7304 | if (disk_bytenr == 0) |
7305 | goto out; | |
7306 | ||
7307 | if (btrfs_file_extent_compression(leaf, fi) || | |
7308 | btrfs_file_extent_encryption(leaf, fi) || | |
7309 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7310 | goto out; | |
7311 | ||
46bfbb5c CM |
7312 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7313 | ||
7ee9e440 JB |
7314 | if (orig_start) { |
7315 | *orig_start = key.offset - backref_offset; | |
7316 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7317 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7318 | } | |
eb384b55 | 7319 | |
46bfbb5c CM |
7320 | if (btrfs_extent_readonly(root, disk_bytenr)) |
7321 | goto out; | |
7b2b7085 MX |
7322 | |
7323 | num_bytes = min(offset + *len, extent_end) - offset; | |
7324 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7325 | u64 range_end; | |
7326 | ||
7327 | range_end = round_up(offset + num_bytes, root->sectorsize) - 1; | |
7328 | ret = test_range_bit(io_tree, offset, range_end, | |
7329 | EXTENT_DELALLOC, 0, NULL); | |
7330 | if (ret) { | |
7331 | ret = -EAGAIN; | |
7332 | goto out; | |
7333 | } | |
7334 | } | |
7335 | ||
1bda19eb | 7336 | btrfs_release_path(path); |
46bfbb5c CM |
7337 | |
7338 | /* | |
7339 | * look for other files referencing this extent, if we | |
7340 | * find any we must cow | |
7341 | */ | |
00361589 JB |
7342 | trans = btrfs_join_transaction(root); |
7343 | if (IS_ERR(trans)) { | |
7344 | ret = 0; | |
46bfbb5c | 7345 | goto out; |
00361589 JB |
7346 | } |
7347 | ||
7348 | ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), | |
7349 | key.offset - backref_offset, disk_bytenr); | |
7350 | btrfs_end_transaction(trans, root); | |
7351 | if (ret) { | |
7352 | ret = 0; | |
7353 | goto out; | |
7354 | } | |
46bfbb5c CM |
7355 | |
7356 | /* | |
7357 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7358 | * in this extent we are about to write. If there | |
7359 | * are any csums in that range we have to cow in order | |
7360 | * to keep the csums correct | |
7361 | */ | |
7362 | disk_bytenr += backref_offset; | |
7363 | disk_bytenr += offset - key.offset; | |
46bfbb5c CM |
7364 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) |
7365 | goto out; | |
7366 | /* | |
7367 | * all of the above have passed, it is safe to overwrite this extent | |
7368 | * without cow | |
7369 | */ | |
eb384b55 | 7370 | *len = num_bytes; |
46bfbb5c CM |
7371 | ret = 1; |
7372 | out: | |
7373 | btrfs_free_path(path); | |
7374 | return ret; | |
7375 | } | |
7376 | ||
fc4adbff AG |
7377 | bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end) |
7378 | { | |
7379 | struct radix_tree_root *root = &inode->i_mapping->page_tree; | |
7380 | int found = false; | |
7381 | void **pagep = NULL; | |
7382 | struct page *page = NULL; | |
7383 | int start_idx; | |
7384 | int end_idx; | |
7385 | ||
09cbfeaf | 7386 | start_idx = start >> PAGE_SHIFT; |
fc4adbff AG |
7387 | |
7388 | /* | |
7389 | * end is the last byte in the last page. end == start is legal | |
7390 | */ | |
09cbfeaf | 7391 | end_idx = end >> PAGE_SHIFT; |
fc4adbff AG |
7392 | |
7393 | rcu_read_lock(); | |
7394 | ||
7395 | /* Most of the code in this while loop is lifted from | |
7396 | * find_get_page. It's been modified to begin searching from a | |
7397 | * page and return just the first page found in that range. If the | |
7398 | * found idx is less than or equal to the end idx then we know that | |
7399 | * a page exists. If no pages are found or if those pages are | |
7400 | * outside of the range then we're fine (yay!) */ | |
7401 | while (page == NULL && | |
7402 | radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) { | |
7403 | page = radix_tree_deref_slot(pagep); | |
7404 | if (unlikely(!page)) | |
7405 | break; | |
7406 | ||
7407 | if (radix_tree_exception(page)) { | |
809f9016 FM |
7408 | if (radix_tree_deref_retry(page)) { |
7409 | page = NULL; | |
fc4adbff | 7410 | continue; |
809f9016 | 7411 | } |
fc4adbff AG |
7412 | /* |
7413 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
7414 | * here as an exceptional entry: so return it without | |
7415 | * attempting to raise page count. | |
7416 | */ | |
6fdef6d4 | 7417 | page = NULL; |
fc4adbff AG |
7418 | break; /* TODO: Is this relevant for this use case? */ |
7419 | } | |
7420 | ||
91405151 FM |
7421 | if (!page_cache_get_speculative(page)) { |
7422 | page = NULL; | |
fc4adbff | 7423 | continue; |
91405151 | 7424 | } |
fc4adbff AG |
7425 | |
7426 | /* | |
7427 | * Has the page moved? | |
7428 | * This is part of the lockless pagecache protocol. See | |
7429 | * include/linux/pagemap.h for details. | |
7430 | */ | |
7431 | if (unlikely(page != *pagep)) { | |
09cbfeaf | 7432 | put_page(page); |
fc4adbff AG |
7433 | page = NULL; |
7434 | } | |
7435 | } | |
7436 | ||
7437 | if (page) { | |
7438 | if (page->index <= end_idx) | |
7439 | found = true; | |
09cbfeaf | 7440 | put_page(page); |
fc4adbff AG |
7441 | } |
7442 | ||
7443 | rcu_read_unlock(); | |
7444 | return found; | |
7445 | } | |
7446 | ||
eb838e73 JB |
7447 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
7448 | struct extent_state **cached_state, int writing) | |
7449 | { | |
7450 | struct btrfs_ordered_extent *ordered; | |
7451 | int ret = 0; | |
7452 | ||
7453 | while (1) { | |
7454 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7455 | cached_state); |
eb838e73 JB |
7456 | /* |
7457 | * We're concerned with the entire range that we're going to be | |
01327610 | 7458 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7459 | * extents in this range. |
7460 | */ | |
7461 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
7462 | lockend - lockstart + 1); | |
7463 | ||
7464 | /* | |
7465 | * We need to make sure there are no buffered pages in this | |
7466 | * range either, we could have raced between the invalidate in | |
7467 | * generic_file_direct_write and locking the extent. The | |
7468 | * invalidate needs to happen so that reads after a write do not | |
7469 | * get stale data. | |
7470 | */ | |
fc4adbff AG |
7471 | if (!ordered && |
7472 | (!writing || | |
7473 | !btrfs_page_exists_in_range(inode, lockstart, lockend))) | |
eb838e73 JB |
7474 | break; |
7475 | ||
7476 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7477 | cached_state, GFP_NOFS); | |
7478 | ||
7479 | if (ordered) { | |
ade77029 FM |
7480 | /* |
7481 | * If we are doing a DIO read and the ordered extent we | |
7482 | * found is for a buffered write, we can not wait for it | |
7483 | * to complete and retry, because if we do so we can | |
7484 | * deadlock with concurrent buffered writes on page | |
7485 | * locks. This happens only if our DIO read covers more | |
7486 | * than one extent map, if at this point has already | |
7487 | * created an ordered extent for a previous extent map | |
7488 | * and locked its range in the inode's io tree, and a | |
7489 | * concurrent write against that previous extent map's | |
7490 | * range and this range started (we unlock the ranges | |
7491 | * in the io tree only when the bios complete and | |
7492 | * buffered writes always lock pages before attempting | |
7493 | * to lock range in the io tree). | |
7494 | */ | |
7495 | if (writing || | |
7496 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
7497 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7498 | else | |
7499 | ret = -ENOTBLK; | |
eb838e73 JB |
7500 | btrfs_put_ordered_extent(ordered); |
7501 | } else { | |
eb838e73 | 7502 | /* |
b850ae14 FM |
7503 | * We could trigger writeback for this range (and wait |
7504 | * for it to complete) and then invalidate the pages for | |
7505 | * this range (through invalidate_inode_pages2_range()), | |
7506 | * but that can lead us to a deadlock with a concurrent | |
7507 | * call to readpages() (a buffered read or a defrag call | |
7508 | * triggered a readahead) on a page lock due to an | |
7509 | * ordered dio extent we created before but did not have | |
7510 | * yet a corresponding bio submitted (whence it can not | |
7511 | * complete), which makes readpages() wait for that | |
7512 | * ordered extent to complete while holding a lock on | |
7513 | * that page. | |
eb838e73 | 7514 | */ |
b850ae14 | 7515 | ret = -ENOTBLK; |
eb838e73 JB |
7516 | } |
7517 | ||
ade77029 FM |
7518 | if (ret) |
7519 | break; | |
7520 | ||
eb838e73 JB |
7521 | cond_resched(); |
7522 | } | |
7523 | ||
7524 | return ret; | |
7525 | } | |
7526 | ||
69ffb543 JB |
7527 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
7528 | u64 len, u64 orig_start, | |
7529 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
7530 | u64 orig_block_len, u64 ram_bytes, |
7531 | int type) | |
69ffb543 JB |
7532 | { |
7533 | struct extent_map_tree *em_tree; | |
7534 | struct extent_map *em; | |
7535 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7536 | int ret; | |
7537 | ||
7538 | em_tree = &BTRFS_I(inode)->extent_tree; | |
7539 | em = alloc_extent_map(); | |
7540 | if (!em) | |
7541 | return ERR_PTR(-ENOMEM); | |
7542 | ||
7543 | em->start = start; | |
7544 | em->orig_start = orig_start; | |
2ab28f32 JB |
7545 | em->mod_start = start; |
7546 | em->mod_len = len; | |
69ffb543 JB |
7547 | em->len = len; |
7548 | em->block_len = block_len; | |
7549 | em->block_start = block_start; | |
7550 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7551 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7552 | em->ram_bytes = ram_bytes; |
70c8a91c | 7553 | em->generation = -1; |
69ffb543 JB |
7554 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
7555 | if (type == BTRFS_ORDERED_PREALLOC) | |
b11e234d | 7556 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
69ffb543 JB |
7557 | |
7558 | do { | |
7559 | btrfs_drop_extent_cache(inode, em->start, | |
7560 | em->start + em->len - 1, 0); | |
7561 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7562 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 JB |
7563 | write_unlock(&em_tree->lock); |
7564 | } while (ret == -EEXIST); | |
7565 | ||
7566 | if (ret) { | |
7567 | free_extent_map(em); | |
7568 | return ERR_PTR(ret); | |
7569 | } | |
7570 | ||
7571 | return em; | |
7572 | } | |
7573 | ||
9c9464cc FM |
7574 | static void adjust_dio_outstanding_extents(struct inode *inode, |
7575 | struct btrfs_dio_data *dio_data, | |
7576 | const u64 len) | |
7577 | { | |
7578 | unsigned num_extents; | |
7579 | ||
7580 | num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1, | |
7581 | BTRFS_MAX_EXTENT_SIZE); | |
7582 | /* | |
7583 | * If we have an outstanding_extents count still set then we're | |
7584 | * within our reservation, otherwise we need to adjust our inode | |
7585 | * counter appropriately. | |
7586 | */ | |
7587 | if (dio_data->outstanding_extents) { | |
7588 | dio_data->outstanding_extents -= num_extents; | |
7589 | } else { | |
7590 | spin_lock(&BTRFS_I(inode)->lock); | |
7591 | BTRFS_I(inode)->outstanding_extents += num_extents; | |
7592 | spin_unlock(&BTRFS_I(inode)->lock); | |
7593 | } | |
7594 | } | |
7595 | ||
4b46fce2 JB |
7596 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7597 | struct buffer_head *bh_result, int create) | |
7598 | { | |
7599 | struct extent_map *em; | |
7600 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 7601 | struct extent_state *cached_state = NULL; |
50745b0a | 7602 | struct btrfs_dio_data *dio_data = NULL; |
4b46fce2 | 7603 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7604 | u64 lockstart, lockend; |
4b46fce2 | 7605 | u64 len = bh_result->b_size; |
eb838e73 | 7606 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7607 | int ret = 0; |
eb838e73 | 7608 | |
172a5049 | 7609 | if (create) |
3266789f | 7610 | unlock_bits |= EXTENT_DIRTY; |
172a5049 | 7611 | else |
c329861d | 7612 | len = min_t(u64, len, root->sectorsize); |
eb838e73 | 7613 | |
c329861d JB |
7614 | lockstart = start; |
7615 | lockend = start + len - 1; | |
7616 | ||
e1cbbfa5 JB |
7617 | if (current->journal_info) { |
7618 | /* | |
7619 | * Need to pull our outstanding extents and set journal_info to NULL so | |
01327610 | 7620 | * that anything that needs to check if there's a transaction doesn't get |
e1cbbfa5 JB |
7621 | * confused. |
7622 | */ | |
50745b0a | 7623 | dio_data = current->journal_info; |
e1cbbfa5 JB |
7624 | current->journal_info = NULL; |
7625 | } | |
7626 | ||
eb838e73 JB |
7627 | /* |
7628 | * If this errors out it's because we couldn't invalidate pagecache for | |
7629 | * this range and we need to fallback to buffered. | |
7630 | */ | |
9c9464cc FM |
7631 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, |
7632 | create)) { | |
7633 | ret = -ENOTBLK; | |
7634 | goto err; | |
7635 | } | |
eb838e73 | 7636 | |
4b46fce2 | 7637 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); |
eb838e73 JB |
7638 | if (IS_ERR(em)) { |
7639 | ret = PTR_ERR(em); | |
7640 | goto unlock_err; | |
7641 | } | |
4b46fce2 JB |
7642 | |
7643 | /* | |
7644 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7645 | * io. INLINE is special, and we could probably kludge it in here, but | |
7646 | * it's still buffered so for safety lets just fall back to the generic | |
7647 | * buffered path. | |
7648 | * | |
7649 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7650 | * decompress it, so there will be buffering required no matter what we | |
7651 | * do, so go ahead and fallback to buffered. | |
7652 | * | |
01327610 | 7653 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7654 | * to buffered IO. Don't blame me, this is the price we pay for using |
7655 | * the generic code. | |
7656 | */ | |
7657 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7658 | em->block_start == EXTENT_MAP_INLINE) { | |
7659 | free_extent_map(em); | |
eb838e73 JB |
7660 | ret = -ENOTBLK; |
7661 | goto unlock_err; | |
4b46fce2 JB |
7662 | } |
7663 | ||
7664 | /* Just a good old fashioned hole, return */ | |
7665 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
7666 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
7667 | free_extent_map(em); | |
eb838e73 | 7668 | goto unlock_err; |
4b46fce2 JB |
7669 | } |
7670 | ||
7671 | /* | |
7672 | * We don't allocate a new extent in the following cases | |
7673 | * | |
7674 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7675 | * existing extent. | |
7676 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7677 | * just use the extent. | |
7678 | * | |
7679 | */ | |
46bfbb5c | 7680 | if (!create) { |
eb838e73 JB |
7681 | len = min(len, em->len - (start - em->start)); |
7682 | lockstart = start + len; | |
7683 | goto unlock; | |
46bfbb5c | 7684 | } |
4b46fce2 JB |
7685 | |
7686 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7687 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7688 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 | 7689 | int type; |
eb384b55 | 7690 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
7691 | |
7692 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7693 | type = BTRFS_ORDERED_PREALLOC; | |
7694 | else | |
7695 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 7696 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 7697 | block_start = em->block_start + (start - em->start); |
46bfbb5c | 7698 | |
00361589 | 7699 | if (can_nocow_extent(inode, start, &len, &orig_start, |
f78c436c FM |
7700 | &orig_block_len, &ram_bytes) == 1 && |
7701 | btrfs_inc_nocow_writers(root->fs_info, block_start)) { | |
5f9a8a51 | 7702 | struct extent_map *em2; |
0b901916 | 7703 | |
5f9a8a51 FM |
7704 | em2 = btrfs_create_dio_extent(inode, start, len, |
7705 | orig_start, block_start, | |
7706 | len, orig_block_len, | |
7707 | ram_bytes, type); | |
f78c436c | 7708 | btrfs_dec_nocow_writers(root->fs_info, block_start); |
69ffb543 JB |
7709 | if (type == BTRFS_ORDERED_PREALLOC) { |
7710 | free_extent_map(em); | |
5f9a8a51 | 7711 | em = em2; |
69ffb543 | 7712 | } |
5f9a8a51 FM |
7713 | if (em2 && IS_ERR(em2)) { |
7714 | ret = PTR_ERR(em2); | |
eb838e73 | 7715 | goto unlock_err; |
46bfbb5c CM |
7716 | } |
7717 | goto unlock; | |
4b46fce2 | 7718 | } |
4b46fce2 | 7719 | } |
00361589 | 7720 | |
46bfbb5c CM |
7721 | /* |
7722 | * this will cow the extent, reset the len in case we changed | |
7723 | * it above | |
7724 | */ | |
7725 | len = bh_result->b_size; | |
70c8a91c JB |
7726 | free_extent_map(em); |
7727 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
7728 | if (IS_ERR(em)) { |
7729 | ret = PTR_ERR(em); | |
7730 | goto unlock_err; | |
7731 | } | |
46bfbb5c CM |
7732 | len = min(len, em->len - (start - em->start)); |
7733 | unlock: | |
4b46fce2 JB |
7734 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
7735 | inode->i_blkbits; | |
46bfbb5c | 7736 | bh_result->b_size = len; |
4b46fce2 JB |
7737 | bh_result->b_bdev = em->bdev; |
7738 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
7739 | if (create) { |
7740 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7741 | set_buffer_new(bh_result); | |
7742 | ||
7743 | /* | |
7744 | * Need to update the i_size under the extent lock so buffered | |
7745 | * readers will get the updated i_size when we unlock. | |
7746 | */ | |
7747 | if (start + len > i_size_read(inode)) | |
7748 | i_size_write(inode, start + len); | |
0934856d | 7749 | |
9c9464cc | 7750 | adjust_dio_outstanding_extents(inode, dio_data, len); |
7cf5b976 | 7751 | btrfs_free_reserved_data_space(inode, start, len); |
50745b0a | 7752 | WARN_ON(dio_data->reserve < len); |
7753 | dio_data->reserve -= len; | |
f28a4928 | 7754 | dio_data->unsubmitted_oe_range_end = start + len; |
50745b0a | 7755 | current->journal_info = dio_data; |
c3473e83 | 7756 | } |
4b46fce2 | 7757 | |
eb838e73 JB |
7758 | /* |
7759 | * In the case of write we need to clear and unlock the entire range, | |
7760 | * in the case of read we need to unlock only the end area that we | |
7761 | * aren't using if there is any left over space. | |
7762 | */ | |
24c03fa5 | 7763 | if (lockstart < lockend) { |
0934856d MX |
7764 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
7765 | lockend, unlock_bits, 1, 0, | |
7766 | &cached_state, GFP_NOFS); | |
24c03fa5 | 7767 | } else { |
eb838e73 | 7768 | free_extent_state(cached_state); |
24c03fa5 | 7769 | } |
eb838e73 | 7770 | |
4b46fce2 JB |
7771 | free_extent_map(em); |
7772 | ||
7773 | return 0; | |
eb838e73 JB |
7774 | |
7775 | unlock_err: | |
eb838e73 JB |
7776 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7777 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
9c9464cc | 7778 | err: |
50745b0a | 7779 | if (dio_data) |
7780 | current->journal_info = dio_data; | |
9c9464cc FM |
7781 | /* |
7782 | * Compensate the delalloc release we do in btrfs_direct_IO() when we | |
7783 | * write less data then expected, so that we don't underflow our inode's | |
7784 | * outstanding extents counter. | |
7785 | */ | |
7786 | if (create && dio_data) | |
7787 | adjust_dio_outstanding_extents(inode, dio_data, len); | |
7788 | ||
eb838e73 | 7789 | return ret; |
4b46fce2 JB |
7790 | } |
7791 | ||
8b110e39 MX |
7792 | static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio, |
7793 | int rw, int mirror_num) | |
7794 | { | |
7795 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7796 | int ret; | |
7797 | ||
7798 | BUG_ON(rw & REQ_WRITE); | |
7799 | ||
7800 | bio_get(bio); | |
7801 | ||
7802 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, | |
7803 | BTRFS_WQ_ENDIO_DIO_REPAIR); | |
7804 | if (ret) | |
7805 | goto err; | |
7806 | ||
7807 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); | |
7808 | err: | |
7809 | bio_put(bio); | |
7810 | return ret; | |
7811 | } | |
7812 | ||
7813 | static int btrfs_check_dio_repairable(struct inode *inode, | |
7814 | struct bio *failed_bio, | |
7815 | struct io_failure_record *failrec, | |
7816 | int failed_mirror) | |
7817 | { | |
7818 | int num_copies; | |
7819 | ||
7820 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, | |
7821 | failrec->logical, failrec->len); | |
7822 | if (num_copies == 1) { | |
7823 | /* | |
7824 | * we only have a single copy of the data, so don't bother with | |
7825 | * all the retry and error correction code that follows. no | |
7826 | * matter what the error is, it is very likely to persist. | |
7827 | */ | |
7828 | pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
7829 | num_copies, failrec->this_mirror, failed_mirror); | |
7830 | return 0; | |
7831 | } | |
7832 | ||
7833 | failrec->failed_mirror = failed_mirror; | |
7834 | failrec->this_mirror++; | |
7835 | if (failrec->this_mirror == failed_mirror) | |
7836 | failrec->this_mirror++; | |
7837 | ||
7838 | if (failrec->this_mirror > num_copies) { | |
7839 | pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
7840 | num_copies, failrec->this_mirror, failed_mirror); | |
7841 | return 0; | |
7842 | } | |
7843 | ||
7844 | return 1; | |
7845 | } | |
7846 | ||
7847 | static int dio_read_error(struct inode *inode, struct bio *failed_bio, | |
2dabb324 CR |
7848 | struct page *page, unsigned int pgoff, |
7849 | u64 start, u64 end, int failed_mirror, | |
7850 | bio_end_io_t *repair_endio, void *repair_arg) | |
8b110e39 MX |
7851 | { |
7852 | struct io_failure_record *failrec; | |
7853 | struct bio *bio; | |
7854 | int isector; | |
7855 | int read_mode; | |
7856 | int ret; | |
7857 | ||
7858 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
7859 | ||
7860 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
7861 | if (ret) | |
7862 | return ret; | |
7863 | ||
7864 | ret = btrfs_check_dio_repairable(inode, failed_bio, failrec, | |
7865 | failed_mirror); | |
7866 | if (!ret) { | |
7867 | free_io_failure(inode, failrec); | |
7868 | return -EIO; | |
7869 | } | |
7870 | ||
2dabb324 CR |
7871 | if ((failed_bio->bi_vcnt > 1) |
7872 | || (failed_bio->bi_io_vec->bv_len | |
7873 | > BTRFS_I(inode)->root->sectorsize)) | |
8b110e39 MX |
7874 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; |
7875 | else | |
7876 | read_mode = READ_SYNC; | |
7877 | ||
7878 | isector = start - btrfs_io_bio(failed_bio)->logical; | |
7879 | isector >>= inode->i_sb->s_blocksize_bits; | |
7880 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2dabb324 | 7881 | pgoff, isector, repair_endio, repair_arg); |
8b110e39 MX |
7882 | if (!bio) { |
7883 | free_io_failure(inode, failrec); | |
7884 | return -EIO; | |
7885 | } | |
7886 | ||
7887 | btrfs_debug(BTRFS_I(inode)->root->fs_info, | |
7888 | "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n", | |
7889 | read_mode, failrec->this_mirror, failrec->in_validation); | |
7890 | ||
7891 | ret = submit_dio_repair_bio(inode, bio, read_mode, | |
7892 | failrec->this_mirror); | |
7893 | if (ret) { | |
7894 | free_io_failure(inode, failrec); | |
7895 | bio_put(bio); | |
7896 | } | |
7897 | ||
7898 | return ret; | |
7899 | } | |
7900 | ||
7901 | struct btrfs_retry_complete { | |
7902 | struct completion done; | |
7903 | struct inode *inode; | |
7904 | u64 start; | |
7905 | int uptodate; | |
7906 | }; | |
7907 | ||
4246a0b6 | 7908 | static void btrfs_retry_endio_nocsum(struct bio *bio) |
8b110e39 MX |
7909 | { |
7910 | struct btrfs_retry_complete *done = bio->bi_private; | |
2dabb324 | 7911 | struct inode *inode; |
8b110e39 MX |
7912 | struct bio_vec *bvec; |
7913 | int i; | |
7914 | ||
4246a0b6 | 7915 | if (bio->bi_error) |
8b110e39 MX |
7916 | goto end; |
7917 | ||
2dabb324 CR |
7918 | ASSERT(bio->bi_vcnt == 1); |
7919 | inode = bio->bi_io_vec->bv_page->mapping->host; | |
7920 | ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize); | |
7921 | ||
8b110e39 MX |
7922 | done->uptodate = 1; |
7923 | bio_for_each_segment_all(bvec, bio, i) | |
7924 | clean_io_failure(done->inode, done->start, bvec->bv_page, 0); | |
7925 | end: | |
7926 | complete(&done->done); | |
7927 | bio_put(bio); | |
7928 | } | |
7929 | ||
7930 | static int __btrfs_correct_data_nocsum(struct inode *inode, | |
7931 | struct btrfs_io_bio *io_bio) | |
4b46fce2 | 7932 | { |
2dabb324 | 7933 | struct btrfs_fs_info *fs_info; |
2c30c71b | 7934 | struct bio_vec *bvec; |
8b110e39 | 7935 | struct btrfs_retry_complete done; |
4b46fce2 | 7936 | u64 start; |
2dabb324 CR |
7937 | unsigned int pgoff; |
7938 | u32 sectorsize; | |
7939 | int nr_sectors; | |
2c30c71b | 7940 | int i; |
c1dc0896 | 7941 | int ret; |
4b46fce2 | 7942 | |
2dabb324 CR |
7943 | fs_info = BTRFS_I(inode)->root->fs_info; |
7944 | sectorsize = BTRFS_I(inode)->root->sectorsize; | |
7945 | ||
8b110e39 MX |
7946 | start = io_bio->logical; |
7947 | done.inode = inode; | |
7948 | ||
7949 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { | |
2dabb324 CR |
7950 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len); |
7951 | pgoff = bvec->bv_offset; | |
7952 | ||
7953 | next_block_or_try_again: | |
8b110e39 MX |
7954 | done.uptodate = 0; |
7955 | done.start = start; | |
7956 | init_completion(&done.done); | |
7957 | ||
2dabb324 CR |
7958 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, |
7959 | pgoff, start, start + sectorsize - 1, | |
7960 | io_bio->mirror_num, | |
7961 | btrfs_retry_endio_nocsum, &done); | |
8b110e39 MX |
7962 | if (ret) |
7963 | return ret; | |
7964 | ||
7965 | wait_for_completion(&done.done); | |
7966 | ||
7967 | if (!done.uptodate) { | |
7968 | /* We might have another mirror, so try again */ | |
2dabb324 | 7969 | goto next_block_or_try_again; |
8b110e39 MX |
7970 | } |
7971 | ||
2dabb324 CR |
7972 | start += sectorsize; |
7973 | ||
7974 | if (nr_sectors--) { | |
7975 | pgoff += sectorsize; | |
7976 | goto next_block_or_try_again; | |
7977 | } | |
8b110e39 MX |
7978 | } |
7979 | ||
7980 | return 0; | |
7981 | } | |
7982 | ||
4246a0b6 | 7983 | static void btrfs_retry_endio(struct bio *bio) |
8b110e39 MX |
7984 | { |
7985 | struct btrfs_retry_complete *done = bio->bi_private; | |
7986 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
2dabb324 | 7987 | struct inode *inode; |
8b110e39 | 7988 | struct bio_vec *bvec; |
2dabb324 | 7989 | u64 start; |
8b110e39 MX |
7990 | int uptodate; |
7991 | int ret; | |
7992 | int i; | |
7993 | ||
4246a0b6 | 7994 | if (bio->bi_error) |
8b110e39 MX |
7995 | goto end; |
7996 | ||
7997 | uptodate = 1; | |
2dabb324 CR |
7998 | |
7999 | start = done->start; | |
8000 | ||
8001 | ASSERT(bio->bi_vcnt == 1); | |
8002 | inode = bio->bi_io_vec->bv_page->mapping->host; | |
8003 | ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize); | |
8004 | ||
8b110e39 MX |
8005 | bio_for_each_segment_all(bvec, bio, i) { |
8006 | ret = __readpage_endio_check(done->inode, io_bio, i, | |
2dabb324 CR |
8007 | bvec->bv_page, bvec->bv_offset, |
8008 | done->start, bvec->bv_len); | |
8b110e39 MX |
8009 | if (!ret) |
8010 | clean_io_failure(done->inode, done->start, | |
2dabb324 | 8011 | bvec->bv_page, bvec->bv_offset); |
8b110e39 MX |
8012 | else |
8013 | uptodate = 0; | |
8014 | } | |
8015 | ||
8016 | done->uptodate = uptodate; | |
8017 | end: | |
8018 | complete(&done->done); | |
8019 | bio_put(bio); | |
8020 | } | |
8021 | ||
8022 | static int __btrfs_subio_endio_read(struct inode *inode, | |
8023 | struct btrfs_io_bio *io_bio, int err) | |
8024 | { | |
2dabb324 | 8025 | struct btrfs_fs_info *fs_info; |
8b110e39 MX |
8026 | struct bio_vec *bvec; |
8027 | struct btrfs_retry_complete done; | |
8028 | u64 start; | |
8029 | u64 offset = 0; | |
2dabb324 CR |
8030 | u32 sectorsize; |
8031 | int nr_sectors; | |
8032 | unsigned int pgoff; | |
8033 | int csum_pos; | |
8b110e39 MX |
8034 | int i; |
8035 | int ret; | |
dc380aea | 8036 | |
2dabb324 CR |
8037 | fs_info = BTRFS_I(inode)->root->fs_info; |
8038 | sectorsize = BTRFS_I(inode)->root->sectorsize; | |
8039 | ||
8b110e39 | 8040 | err = 0; |
c1dc0896 | 8041 | start = io_bio->logical; |
8b110e39 MX |
8042 | done.inode = inode; |
8043 | ||
c1dc0896 | 8044 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { |
2dabb324 CR |
8045 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len); |
8046 | ||
8047 | pgoff = bvec->bv_offset; | |
8048 | next_block: | |
8049 | csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset); | |
8050 | ret = __readpage_endio_check(inode, io_bio, csum_pos, | |
8051 | bvec->bv_page, pgoff, start, | |
8052 | sectorsize); | |
8b110e39 MX |
8053 | if (likely(!ret)) |
8054 | goto next; | |
8055 | try_again: | |
8056 | done.uptodate = 0; | |
8057 | done.start = start; | |
8058 | init_completion(&done.done); | |
8059 | ||
2dabb324 CR |
8060 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, |
8061 | pgoff, start, start + sectorsize - 1, | |
8062 | io_bio->mirror_num, | |
8063 | btrfs_retry_endio, &done); | |
8b110e39 MX |
8064 | if (ret) { |
8065 | err = ret; | |
8066 | goto next; | |
8067 | } | |
8068 | ||
8069 | wait_for_completion(&done.done); | |
8070 | ||
8071 | if (!done.uptodate) { | |
8072 | /* We might have another mirror, so try again */ | |
8073 | goto try_again; | |
8074 | } | |
8075 | next: | |
2dabb324 CR |
8076 | offset += sectorsize; |
8077 | start += sectorsize; | |
8078 | ||
8079 | ASSERT(nr_sectors); | |
8080 | ||
8081 | if (--nr_sectors) { | |
8082 | pgoff += sectorsize; | |
8083 | goto next_block; | |
8084 | } | |
2c30c71b | 8085 | } |
c1dc0896 MX |
8086 | |
8087 | return err; | |
8088 | } | |
8089 | ||
8b110e39 MX |
8090 | static int btrfs_subio_endio_read(struct inode *inode, |
8091 | struct btrfs_io_bio *io_bio, int err) | |
8092 | { | |
8093 | bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8094 | ||
8095 | if (skip_csum) { | |
8096 | if (unlikely(err)) | |
8097 | return __btrfs_correct_data_nocsum(inode, io_bio); | |
8098 | else | |
8099 | return 0; | |
8100 | } else { | |
8101 | return __btrfs_subio_endio_read(inode, io_bio, err); | |
8102 | } | |
8103 | } | |
8104 | ||
4246a0b6 | 8105 | static void btrfs_endio_direct_read(struct bio *bio) |
c1dc0896 MX |
8106 | { |
8107 | struct btrfs_dio_private *dip = bio->bi_private; | |
8108 | struct inode *inode = dip->inode; | |
8109 | struct bio *dio_bio; | |
8110 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
4246a0b6 | 8111 | int err = bio->bi_error; |
c1dc0896 | 8112 | |
8b110e39 MX |
8113 | if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) |
8114 | err = btrfs_subio_endio_read(inode, io_bio, err); | |
c1dc0896 | 8115 | |
4b46fce2 | 8116 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, |
d0082371 | 8117 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 8118 | dio_bio = dip->dio_bio; |
4b46fce2 | 8119 | |
4b46fce2 | 8120 | kfree(dip); |
c0da7aa1 | 8121 | |
1636d1d7 | 8122 | dio_bio->bi_error = bio->bi_error; |
4246a0b6 | 8123 | dio_end_io(dio_bio, bio->bi_error); |
23ea8e5a MX |
8124 | |
8125 | if (io_bio->end_io) | |
8126 | io_bio->end_io(io_bio, err); | |
9be3395b | 8127 | bio_put(bio); |
4b46fce2 JB |
8128 | } |
8129 | ||
14543774 FM |
8130 | static void btrfs_endio_direct_write_update_ordered(struct inode *inode, |
8131 | const u64 offset, | |
8132 | const u64 bytes, | |
8133 | const int uptodate) | |
4b46fce2 | 8134 | { |
4b46fce2 | 8135 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4b46fce2 | 8136 | struct btrfs_ordered_extent *ordered = NULL; |
14543774 FM |
8137 | u64 ordered_offset = offset; |
8138 | u64 ordered_bytes = bytes; | |
4b46fce2 JB |
8139 | int ret; |
8140 | ||
163cf09c CM |
8141 | again: |
8142 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
8143 | &ordered_offset, | |
4246a0b6 | 8144 | ordered_bytes, |
14543774 | 8145 | uptodate); |
4b46fce2 | 8146 | if (!ret) |
163cf09c | 8147 | goto out_test; |
4b46fce2 | 8148 | |
9e0af237 LB |
8149 | btrfs_init_work(&ordered->work, btrfs_endio_write_helper, |
8150 | finish_ordered_fn, NULL, NULL); | |
fccb5d86 QW |
8151 | btrfs_queue_work(root->fs_info->endio_write_workers, |
8152 | &ordered->work); | |
163cf09c CM |
8153 | out_test: |
8154 | /* | |
8155 | * our bio might span multiple ordered extents. If we haven't | |
8156 | * completed the accounting for the whole dio, go back and try again | |
8157 | */ | |
14543774 FM |
8158 | if (ordered_offset < offset + bytes) { |
8159 | ordered_bytes = offset + bytes - ordered_offset; | |
5fd02043 | 8160 | ordered = NULL; |
163cf09c CM |
8161 | goto again; |
8162 | } | |
14543774 FM |
8163 | } |
8164 | ||
8165 | static void btrfs_endio_direct_write(struct bio *bio) | |
8166 | { | |
8167 | struct btrfs_dio_private *dip = bio->bi_private; | |
8168 | struct bio *dio_bio = dip->dio_bio; | |
8169 | ||
8170 | btrfs_endio_direct_write_update_ordered(dip->inode, | |
8171 | dip->logical_offset, | |
8172 | dip->bytes, | |
8173 | !bio->bi_error); | |
4b46fce2 | 8174 | |
4b46fce2 | 8175 | kfree(dip); |
c0da7aa1 | 8176 | |
1636d1d7 | 8177 | dio_bio->bi_error = bio->bi_error; |
4246a0b6 | 8178 | dio_end_io(dio_bio, bio->bi_error); |
9be3395b | 8179 | bio_put(bio); |
4b46fce2 JB |
8180 | } |
8181 | ||
eaf25d93 CM |
8182 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
8183 | struct bio *bio, int mirror_num, | |
8184 | unsigned long bio_flags, u64 offset) | |
8185 | { | |
8186 | int ret; | |
8187 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8188 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 8189 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
8190 | return 0; |
8191 | } | |
8192 | ||
4246a0b6 | 8193 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
8194 | { |
8195 | struct btrfs_dio_private *dip = bio->bi_private; | |
4246a0b6 | 8196 | int err = bio->bi_error; |
e65e1535 | 8197 | |
8b110e39 MX |
8198 | if (err) |
8199 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
8200 | "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d", | |
8201 | btrfs_ino(dip->inode), bio->bi_rw, | |
8202 | (unsigned long long)bio->bi_iter.bi_sector, | |
8203 | bio->bi_iter.bi_size, err); | |
8204 | ||
8205 | if (dip->subio_endio) | |
8206 | err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err); | |
c1dc0896 MX |
8207 | |
8208 | if (err) { | |
e65e1535 MX |
8209 | dip->errors = 1; |
8210 | ||
8211 | /* | |
8212 | * before atomic variable goto zero, we must make sure | |
8213 | * dip->errors is perceived to be set. | |
8214 | */ | |
4e857c58 | 8215 | smp_mb__before_atomic(); |
e65e1535 MX |
8216 | } |
8217 | ||
8218 | /* if there are more bios still pending for this dio, just exit */ | |
8219 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
8220 | goto out; | |
8221 | ||
9be3395b | 8222 | if (dip->errors) { |
e65e1535 | 8223 | bio_io_error(dip->orig_bio); |
9be3395b | 8224 | } else { |
4246a0b6 CH |
8225 | dip->dio_bio->bi_error = 0; |
8226 | bio_endio(dip->orig_bio); | |
e65e1535 MX |
8227 | } |
8228 | out: | |
8229 | bio_put(bio); | |
8230 | } | |
8231 | ||
8232 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
8233 | u64 first_sector, gfp_t gfp_flags) | |
8234 | { | |
da2f0f74 | 8235 | struct bio *bio; |
22365979 | 8236 | bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags); |
da2f0f74 CM |
8237 | if (bio) |
8238 | bio_associate_current(bio); | |
8239 | return bio; | |
e65e1535 MX |
8240 | } |
8241 | ||
c1dc0896 MX |
8242 | static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root, |
8243 | struct inode *inode, | |
8244 | struct btrfs_dio_private *dip, | |
8245 | struct bio *bio, | |
8246 | u64 file_offset) | |
8247 | { | |
8248 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
8249 | struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio); | |
8250 | int ret; | |
8251 | ||
8252 | /* | |
8253 | * We load all the csum data we need when we submit | |
8254 | * the first bio to reduce the csum tree search and | |
8255 | * contention. | |
8256 | */ | |
8257 | if (dip->logical_offset == file_offset) { | |
8258 | ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio, | |
8259 | file_offset); | |
8260 | if (ret) | |
8261 | return ret; | |
8262 | } | |
8263 | ||
8264 | if (bio == dip->orig_bio) | |
8265 | return 0; | |
8266 | ||
8267 | file_offset -= dip->logical_offset; | |
8268 | file_offset >>= inode->i_sb->s_blocksize_bits; | |
8269 | io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset); | |
8270 | ||
8271 | return 0; | |
8272 | } | |
8273 | ||
e65e1535 MX |
8274 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, |
8275 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 8276 | int async_submit) |
e65e1535 | 8277 | { |
facc8a22 | 8278 | struct btrfs_dio_private *dip = bio->bi_private; |
e65e1535 MX |
8279 | int write = rw & REQ_WRITE; |
8280 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8281 | int ret; | |
8282 | ||
b812ce28 JB |
8283 | if (async_submit) |
8284 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
8285 | ||
e65e1535 | 8286 | bio_get(bio); |
5fd02043 JB |
8287 | |
8288 | if (!write) { | |
bfebd8b5 DS |
8289 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, |
8290 | BTRFS_WQ_ENDIO_DATA); | |
5fd02043 JB |
8291 | if (ret) |
8292 | goto err; | |
8293 | } | |
e65e1535 | 8294 | |
1ae39938 JB |
8295 | if (skip_sum) |
8296 | goto map; | |
8297 | ||
8298 | if (write && async_submit) { | |
e65e1535 MX |
8299 | ret = btrfs_wq_submit_bio(root->fs_info, |
8300 | inode, rw, bio, 0, 0, | |
8301 | file_offset, | |
8302 | __btrfs_submit_bio_start_direct_io, | |
8303 | __btrfs_submit_bio_done); | |
8304 | goto err; | |
1ae39938 JB |
8305 | } else if (write) { |
8306 | /* | |
8307 | * If we aren't doing async submit, calculate the csum of the | |
8308 | * bio now. | |
8309 | */ | |
8310 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
8311 | if (ret) | |
8312 | goto err; | |
23ea8e5a | 8313 | } else { |
c1dc0896 MX |
8314 | ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio, |
8315 | file_offset); | |
c2db1073 TI |
8316 | if (ret) |
8317 | goto err; | |
8318 | } | |
1ae39938 JB |
8319 | map: |
8320 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
8321 | err: |
8322 | bio_put(bio); | |
8323 | return ret; | |
8324 | } | |
8325 | ||
8326 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
8327 | int skip_sum) | |
8328 | { | |
8329 | struct inode *inode = dip->inode; | |
8330 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
e65e1535 MX |
8331 | struct bio *bio; |
8332 | struct bio *orig_bio = dip->orig_bio; | |
8333 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
4f024f37 | 8334 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 MX |
8335 | u64 file_offset = dip->logical_offset; |
8336 | u64 submit_len = 0; | |
8337 | u64 map_length; | |
5f4dc8fc | 8338 | u32 blocksize = root->sectorsize; |
1ae39938 | 8339 | int async_submit = 0; |
5f4dc8fc CR |
8340 | int nr_sectors; |
8341 | int ret; | |
8342 | int i; | |
e65e1535 | 8343 | |
4f024f37 | 8344 | map_length = orig_bio->bi_iter.bi_size; |
53b381b3 | 8345 | ret = btrfs_map_block(root->fs_info, rw, start_sector << 9, |
e65e1535 | 8346 | &map_length, NULL, 0); |
7a5c3c9b | 8347 | if (ret) |
e65e1535 | 8348 | return -EIO; |
facc8a22 | 8349 | |
4f024f37 | 8350 | if (map_length >= orig_bio->bi_iter.bi_size) { |
02f57c7a | 8351 | bio = orig_bio; |
c1dc0896 | 8352 | dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED; |
02f57c7a JB |
8353 | goto submit; |
8354 | } | |
8355 | ||
53b381b3 | 8356 | /* async crcs make it difficult to collect full stripe writes. */ |
ffe2d203 | 8357 | if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
8358 | async_submit = 0; |
8359 | else | |
8360 | async_submit = 1; | |
8361 | ||
02f57c7a JB |
8362 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
8363 | if (!bio) | |
8364 | return -ENOMEM; | |
7a5c3c9b | 8365 | |
02f57c7a JB |
8366 | bio->bi_private = dip; |
8367 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 8368 | btrfs_io_bio(bio)->logical = file_offset; |
02f57c7a JB |
8369 | atomic_inc(&dip->pending_bios); |
8370 | ||
e65e1535 | 8371 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
5f4dc8fc CR |
8372 | nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len); |
8373 | i = 0; | |
8374 | next_block: | |
8375 | if (unlikely(map_length < submit_len + blocksize || | |
8376 | bio_add_page(bio, bvec->bv_page, blocksize, | |
8377 | bvec->bv_offset + (i * blocksize)) < blocksize)) { | |
e65e1535 MX |
8378 | /* |
8379 | * inc the count before we submit the bio so | |
8380 | * we know the end IO handler won't happen before | |
8381 | * we inc the count. Otherwise, the dip might get freed | |
8382 | * before we're done setting it up | |
8383 | */ | |
8384 | atomic_inc(&dip->pending_bios); | |
8385 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
8386 | file_offset, skip_sum, | |
c329861d | 8387 | async_submit); |
e65e1535 MX |
8388 | if (ret) { |
8389 | bio_put(bio); | |
8390 | atomic_dec(&dip->pending_bios); | |
8391 | goto out_err; | |
8392 | } | |
8393 | ||
e65e1535 MX |
8394 | start_sector += submit_len >> 9; |
8395 | file_offset += submit_len; | |
8396 | ||
8397 | submit_len = 0; | |
e65e1535 MX |
8398 | |
8399 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
8400 | start_sector, GFP_NOFS); | |
8401 | if (!bio) | |
8402 | goto out_err; | |
8403 | bio->bi_private = dip; | |
8404 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 8405 | btrfs_io_bio(bio)->logical = file_offset; |
e65e1535 | 8406 | |
4f024f37 | 8407 | map_length = orig_bio->bi_iter.bi_size; |
53b381b3 | 8408 | ret = btrfs_map_block(root->fs_info, rw, |
3ec706c8 | 8409 | start_sector << 9, |
e65e1535 MX |
8410 | &map_length, NULL, 0); |
8411 | if (ret) { | |
8412 | bio_put(bio); | |
8413 | goto out_err; | |
8414 | } | |
5f4dc8fc CR |
8415 | |
8416 | goto next_block; | |
e65e1535 | 8417 | } else { |
5f4dc8fc CR |
8418 | submit_len += blocksize; |
8419 | if (--nr_sectors) { | |
8420 | i++; | |
8421 | goto next_block; | |
8422 | } | |
e65e1535 MX |
8423 | bvec++; |
8424 | } | |
8425 | } | |
8426 | ||
02f57c7a | 8427 | submit: |
e65e1535 | 8428 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 8429 | async_submit); |
e65e1535 MX |
8430 | if (!ret) |
8431 | return 0; | |
8432 | ||
8433 | bio_put(bio); | |
8434 | out_err: | |
8435 | dip->errors = 1; | |
8436 | /* | |
8437 | * before atomic variable goto zero, we must | |
8438 | * make sure dip->errors is perceived to be set. | |
8439 | */ | |
4e857c58 | 8440 | smp_mb__before_atomic(); |
e65e1535 MX |
8441 | if (atomic_dec_and_test(&dip->pending_bios)) |
8442 | bio_io_error(dip->orig_bio); | |
8443 | ||
8444 | /* bio_end_io() will handle error, so we needn't return it */ | |
8445 | return 0; | |
8446 | } | |
8447 | ||
9be3395b CM |
8448 | static void btrfs_submit_direct(int rw, struct bio *dio_bio, |
8449 | struct inode *inode, loff_t file_offset) | |
4b46fce2 | 8450 | { |
61de718f FM |
8451 | struct btrfs_dio_private *dip = NULL; |
8452 | struct bio *io_bio = NULL; | |
23ea8e5a | 8453 | struct btrfs_io_bio *btrfs_bio; |
4b46fce2 | 8454 | int skip_sum; |
7b6d91da | 8455 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
8456 | int ret = 0; |
8457 | ||
8458 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8459 | ||
9be3395b | 8460 | io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS); |
9be3395b CM |
8461 | if (!io_bio) { |
8462 | ret = -ENOMEM; | |
8463 | goto free_ordered; | |
8464 | } | |
8465 | ||
c1dc0896 | 8466 | dip = kzalloc(sizeof(*dip), GFP_NOFS); |
4b46fce2 JB |
8467 | if (!dip) { |
8468 | ret = -ENOMEM; | |
61de718f | 8469 | goto free_ordered; |
4b46fce2 | 8470 | } |
4b46fce2 | 8471 | |
9be3395b | 8472 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
8473 | dip->inode = inode; |
8474 | dip->logical_offset = file_offset; | |
4f024f37 KO |
8475 | dip->bytes = dio_bio->bi_iter.bi_size; |
8476 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
9be3395b | 8477 | io_bio->bi_private = dip; |
9be3395b CM |
8478 | dip->orig_bio = io_bio; |
8479 | dip->dio_bio = dio_bio; | |
e65e1535 | 8480 | atomic_set(&dip->pending_bios, 0); |
c1dc0896 MX |
8481 | btrfs_bio = btrfs_io_bio(io_bio); |
8482 | btrfs_bio->logical = file_offset; | |
4b46fce2 | 8483 | |
c1dc0896 | 8484 | if (write) { |
9be3395b | 8485 | io_bio->bi_end_io = btrfs_endio_direct_write; |
c1dc0896 | 8486 | } else { |
9be3395b | 8487 | io_bio->bi_end_io = btrfs_endio_direct_read; |
c1dc0896 MX |
8488 | dip->subio_endio = btrfs_subio_endio_read; |
8489 | } | |
4b46fce2 | 8490 | |
f28a4928 FM |
8491 | /* |
8492 | * Reset the range for unsubmitted ordered extents (to a 0 length range) | |
8493 | * even if we fail to submit a bio, because in such case we do the | |
8494 | * corresponding error handling below and it must not be done a second | |
8495 | * time by btrfs_direct_IO(). | |
8496 | */ | |
8497 | if (write) { | |
8498 | struct btrfs_dio_data *dio_data = current->journal_info; | |
8499 | ||
8500 | dio_data->unsubmitted_oe_range_end = dip->logical_offset + | |
8501 | dip->bytes; | |
8502 | dio_data->unsubmitted_oe_range_start = | |
8503 | dio_data->unsubmitted_oe_range_end; | |
8504 | } | |
8505 | ||
e65e1535 MX |
8506 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
8507 | if (!ret) | |
eaf25d93 | 8508 | return; |
9be3395b | 8509 | |
23ea8e5a MX |
8510 | if (btrfs_bio->end_io) |
8511 | btrfs_bio->end_io(btrfs_bio, ret); | |
9be3395b | 8512 | |
4b46fce2 JB |
8513 | free_ordered: |
8514 | /* | |
61de718f FM |
8515 | * If we arrived here it means either we failed to submit the dip |
8516 | * or we either failed to clone the dio_bio or failed to allocate the | |
8517 | * dip. If we cloned the dio_bio and allocated the dip, we can just | |
8518 | * call bio_endio against our io_bio so that we get proper resource | |
8519 | * cleanup if we fail to submit the dip, otherwise, we must do the | |
8520 | * same as btrfs_endio_direct_[write|read] because we can't call these | |
8521 | * callbacks - they require an allocated dip and a clone of dio_bio. | |
4b46fce2 | 8522 | */ |
61de718f | 8523 | if (io_bio && dip) { |
4246a0b6 CH |
8524 | io_bio->bi_error = -EIO; |
8525 | bio_endio(io_bio); | |
61de718f FM |
8526 | /* |
8527 | * The end io callbacks free our dip, do the final put on io_bio | |
8528 | * and all the cleanup and final put for dio_bio (through | |
8529 | * dio_end_io()). | |
8530 | */ | |
8531 | dip = NULL; | |
8532 | io_bio = NULL; | |
8533 | } else { | |
14543774 FM |
8534 | if (write) |
8535 | btrfs_endio_direct_write_update_ordered(inode, | |
8536 | file_offset, | |
8537 | dio_bio->bi_iter.bi_size, | |
8538 | 0); | |
8539 | else | |
61de718f FM |
8540 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, |
8541 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
14543774 | 8542 | |
4246a0b6 | 8543 | dio_bio->bi_error = -EIO; |
61de718f FM |
8544 | /* |
8545 | * Releases and cleans up our dio_bio, no need to bio_put() | |
8546 | * nor bio_endio()/bio_io_error() against dio_bio. | |
8547 | */ | |
8548 | dio_end_io(dio_bio, ret); | |
4b46fce2 | 8549 | } |
61de718f FM |
8550 | if (io_bio) |
8551 | bio_put(io_bio); | |
8552 | kfree(dip); | |
4b46fce2 JB |
8553 | } |
8554 | ||
6f673763 | 8555 | static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb, |
28060d5d | 8556 | const struct iov_iter *iter, loff_t offset) |
5a5f79b5 CM |
8557 | { |
8558 | int seg; | |
a1b75f7d | 8559 | int i; |
5a5f79b5 CM |
8560 | unsigned blocksize_mask = root->sectorsize - 1; |
8561 | ssize_t retval = -EINVAL; | |
5a5f79b5 CM |
8562 | |
8563 | if (offset & blocksize_mask) | |
8564 | goto out; | |
8565 | ||
28060d5d AV |
8566 | if (iov_iter_alignment(iter) & blocksize_mask) |
8567 | goto out; | |
a1b75f7d | 8568 | |
28060d5d | 8569 | /* If this is a write we don't need to check anymore */ |
6f673763 | 8570 | if (iov_iter_rw(iter) == WRITE) |
28060d5d AV |
8571 | return 0; |
8572 | /* | |
8573 | * Check to make sure we don't have duplicate iov_base's in this | |
8574 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
8575 | * when reading back. | |
8576 | */ | |
8577 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
8578 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
8579 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
8580 | goto out; |
8581 | } | |
5a5f79b5 CM |
8582 | } |
8583 | retval = 0; | |
8584 | out: | |
8585 | return retval; | |
8586 | } | |
eb838e73 | 8587 | |
c8b8e32d | 8588 | static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
16432985 | 8589 | { |
4b46fce2 JB |
8590 | struct file *file = iocb->ki_filp; |
8591 | struct inode *inode = file->f_mapping->host; | |
50745b0a | 8592 | struct btrfs_root *root = BTRFS_I(inode)->root; |
8593 | struct btrfs_dio_data dio_data = { 0 }; | |
c8b8e32d | 8594 | loff_t offset = iocb->ki_pos; |
0934856d | 8595 | size_t count = 0; |
2e60a51e | 8596 | int flags = 0; |
38851cc1 MX |
8597 | bool wakeup = true; |
8598 | bool relock = false; | |
0934856d | 8599 | ssize_t ret; |
4b46fce2 | 8600 | |
6f673763 | 8601 | if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset)) |
5a5f79b5 | 8602 | return 0; |
3f7c579c | 8603 | |
fe0f07d0 | 8604 | inode_dio_begin(inode); |
4e857c58 | 8605 | smp_mb__after_atomic(); |
38851cc1 | 8606 | |
0e267c44 | 8607 | /* |
41bd9ca4 MX |
8608 | * The generic stuff only does filemap_write_and_wait_range, which |
8609 | * isn't enough if we've written compressed pages to this area, so | |
8610 | * we need to flush the dirty pages again to make absolutely sure | |
8611 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 8612 | */ |
a6cbcd4a | 8613 | count = iov_iter_count(iter); |
41bd9ca4 MX |
8614 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
8615 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
8616 | filemap_fdatawrite_range(inode->i_mapping, offset, |
8617 | offset + count - 1); | |
0e267c44 | 8618 | |
6f673763 | 8619 | if (iov_iter_rw(iter) == WRITE) { |
38851cc1 MX |
8620 | /* |
8621 | * If the write DIO is beyond the EOF, we need update | |
8622 | * the isize, but it is protected by i_mutex. So we can | |
8623 | * not unlock the i_mutex at this case. | |
8624 | */ | |
8625 | if (offset + count <= inode->i_size) { | |
5955102c | 8626 | inode_unlock(inode); |
38851cc1 MX |
8627 | relock = true; |
8628 | } | |
7cf5b976 | 8629 | ret = btrfs_delalloc_reserve_space(inode, offset, count); |
0934856d | 8630 | if (ret) |
38851cc1 | 8631 | goto out; |
50745b0a | 8632 | dio_data.outstanding_extents = div64_u64(count + |
e1cbbfa5 JB |
8633 | BTRFS_MAX_EXTENT_SIZE - 1, |
8634 | BTRFS_MAX_EXTENT_SIZE); | |
8635 | ||
8636 | /* | |
8637 | * We need to know how many extents we reserved so that we can | |
8638 | * do the accounting properly if we go over the number we | |
8639 | * originally calculated. Abuse current->journal_info for this. | |
8640 | */ | |
50745b0a | 8641 | dio_data.reserve = round_up(count, root->sectorsize); |
f28a4928 FM |
8642 | dio_data.unsubmitted_oe_range_start = (u64)offset; |
8643 | dio_data.unsubmitted_oe_range_end = (u64)offset; | |
50745b0a | 8644 | current->journal_info = &dio_data; |
ee39b432 DS |
8645 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
8646 | &BTRFS_I(inode)->runtime_flags)) { | |
fe0f07d0 | 8647 | inode_dio_end(inode); |
38851cc1 MX |
8648 | flags = DIO_LOCKING | DIO_SKIP_HOLES; |
8649 | wakeup = false; | |
0934856d MX |
8650 | } |
8651 | ||
17f8c842 OS |
8652 | ret = __blockdev_direct_IO(iocb, inode, |
8653 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
c8b8e32d | 8654 | iter, btrfs_get_blocks_direct, NULL, |
17f8c842 | 8655 | btrfs_submit_direct, flags); |
6f673763 | 8656 | if (iov_iter_rw(iter) == WRITE) { |
e1cbbfa5 | 8657 | current->journal_info = NULL; |
ddba1bfc | 8658 | if (ret < 0 && ret != -EIOCBQUEUED) { |
50745b0a | 8659 | if (dio_data.reserve) |
7cf5b976 QW |
8660 | btrfs_delalloc_release_space(inode, offset, |
8661 | dio_data.reserve); | |
f28a4928 FM |
8662 | /* |
8663 | * On error we might have left some ordered extents | |
8664 | * without submitting corresponding bios for them, so | |
8665 | * cleanup them up to avoid other tasks getting them | |
8666 | * and waiting for them to complete forever. | |
8667 | */ | |
8668 | if (dio_data.unsubmitted_oe_range_start < | |
8669 | dio_data.unsubmitted_oe_range_end) | |
8670 | btrfs_endio_direct_write_update_ordered(inode, | |
8671 | dio_data.unsubmitted_oe_range_start, | |
8672 | dio_data.unsubmitted_oe_range_end - | |
8673 | dio_data.unsubmitted_oe_range_start, | |
8674 | 0); | |
ddba1bfc | 8675 | } else if (ret >= 0 && (size_t)ret < count) |
7cf5b976 QW |
8676 | btrfs_delalloc_release_space(inode, offset, |
8677 | count - (size_t)ret); | |
0934856d | 8678 | } |
38851cc1 | 8679 | out: |
2e60a51e | 8680 | if (wakeup) |
fe0f07d0 | 8681 | inode_dio_end(inode); |
38851cc1 | 8682 | if (relock) |
5955102c | 8683 | inode_lock(inode); |
0934856d MX |
8684 | |
8685 | return ret; | |
16432985 CM |
8686 | } |
8687 | ||
05dadc09 TI |
8688 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
8689 | ||
1506fcc8 YS |
8690 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
8691 | __u64 start, __u64 len) | |
8692 | { | |
05dadc09 TI |
8693 | int ret; |
8694 | ||
8695 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
8696 | if (ret) | |
8697 | return ret; | |
8698 | ||
ec29ed5b | 8699 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
8700 | } |
8701 | ||
a52d9a80 | 8702 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8703 | { |
d1310b2e CM |
8704 | struct extent_io_tree *tree; |
8705 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 8706 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 8707 | } |
1832a6d5 | 8708 | |
a52d9a80 | 8709 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8710 | { |
d1310b2e | 8711 | struct extent_io_tree *tree; |
be7bd730 JB |
8712 | struct inode *inode = page->mapping->host; |
8713 | int ret; | |
b888db2b CM |
8714 | |
8715 | if (current->flags & PF_MEMALLOC) { | |
8716 | redirty_page_for_writepage(wbc, page); | |
8717 | unlock_page(page); | |
8718 | return 0; | |
8719 | } | |
be7bd730 JB |
8720 | |
8721 | /* | |
8722 | * If we are under memory pressure we will call this directly from the | |
8723 | * VM, we need to make sure we have the inode referenced for the ordered | |
8724 | * extent. If not just return like we didn't do anything. | |
8725 | */ | |
8726 | if (!igrab(inode)) { | |
8727 | redirty_page_for_writepage(wbc, page); | |
8728 | return AOP_WRITEPAGE_ACTIVATE; | |
8729 | } | |
d1310b2e | 8730 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
be7bd730 JB |
8731 | ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
8732 | btrfs_add_delayed_iput(inode); | |
8733 | return ret; | |
9ebefb18 CM |
8734 | } |
8735 | ||
48a3b636 ES |
8736 | static int btrfs_writepages(struct address_space *mapping, |
8737 | struct writeback_control *wbc) | |
b293f02e | 8738 | { |
d1310b2e | 8739 | struct extent_io_tree *tree; |
771ed689 | 8740 | |
d1310b2e | 8741 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
8742 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
8743 | } | |
8744 | ||
3ab2fb5a CM |
8745 | static int |
8746 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
8747 | struct list_head *pages, unsigned nr_pages) | |
8748 | { | |
d1310b2e CM |
8749 | struct extent_io_tree *tree; |
8750 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
8751 | return extent_readpages(tree, mapping, pages, nr_pages, |
8752 | btrfs_get_extent); | |
8753 | } | |
e6dcd2dc | 8754 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8755 | { |
d1310b2e CM |
8756 | struct extent_io_tree *tree; |
8757 | struct extent_map_tree *map; | |
a52d9a80 | 8758 | int ret; |
8c2383c3 | 8759 | |
d1310b2e CM |
8760 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
8761 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 8762 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
8763 | if (ret == 1) { |
8764 | ClearPagePrivate(page); | |
8765 | set_page_private(page, 0); | |
09cbfeaf | 8766 | put_page(page); |
39279cc3 | 8767 | } |
a52d9a80 | 8768 | return ret; |
39279cc3 CM |
8769 | } |
8770 | ||
e6dcd2dc CM |
8771 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8772 | { | |
98509cfc CM |
8773 | if (PageWriteback(page) || PageDirty(page)) |
8774 | return 0; | |
b335b003 | 8775 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
8776 | } |
8777 | ||
d47992f8 LC |
8778 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8779 | unsigned int length) | |
39279cc3 | 8780 | { |
5fd02043 | 8781 | struct inode *inode = page->mapping->host; |
d1310b2e | 8782 | struct extent_io_tree *tree; |
e6dcd2dc | 8783 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8784 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 8785 | u64 page_start = page_offset(page); |
09cbfeaf | 8786 | u64 page_end = page_start + PAGE_SIZE - 1; |
dbfdb6d1 CR |
8787 | u64 start; |
8788 | u64 end; | |
131e404a | 8789 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 8790 | |
8b62b72b CM |
8791 | /* |
8792 | * we have the page locked, so new writeback can't start, | |
8793 | * and the dirty bit won't be cleared while we are here. | |
8794 | * | |
8795 | * Wait for IO on this page so that we can safely clear | |
8796 | * the PagePrivate2 bit and do ordered accounting | |
8797 | */ | |
e6dcd2dc | 8798 | wait_on_page_writeback(page); |
8b62b72b | 8799 | |
5fd02043 | 8800 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
8801 | if (offset) { |
8802 | btrfs_releasepage(page, GFP_NOFS); | |
8803 | return; | |
8804 | } | |
131e404a FDBM |
8805 | |
8806 | if (!inode_evicting) | |
ff13db41 | 8807 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
dbfdb6d1 CR |
8808 | again: |
8809 | start = page_start; | |
8810 | ordered = btrfs_lookup_ordered_range(inode, start, | |
8811 | page_end - start + 1); | |
e6dcd2dc | 8812 | if (ordered) { |
dbfdb6d1 | 8813 | end = min(page_end, ordered->file_offset + ordered->len - 1); |
eb84ae03 CM |
8814 | /* |
8815 | * IO on this page will never be started, so we need | |
8816 | * to account for any ordered extents now | |
8817 | */ | |
131e404a | 8818 | if (!inode_evicting) |
dbfdb6d1 | 8819 | clear_extent_bit(tree, start, end, |
131e404a FDBM |
8820 | EXTENT_DIRTY | EXTENT_DELALLOC | |
8821 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | | |
8822 | EXTENT_DEFRAG, 1, 0, &cached_state, | |
8823 | GFP_NOFS); | |
8b62b72b CM |
8824 | /* |
8825 | * whoever cleared the private bit is responsible | |
8826 | * for the finish_ordered_io | |
8827 | */ | |
77cef2ec JB |
8828 | if (TestClearPagePrivate2(page)) { |
8829 | struct btrfs_ordered_inode_tree *tree; | |
8830 | u64 new_len; | |
8831 | ||
8832 | tree = &BTRFS_I(inode)->ordered_tree; | |
8833 | ||
8834 | spin_lock_irq(&tree->lock); | |
8835 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
dbfdb6d1 | 8836 | new_len = start - ordered->file_offset; |
77cef2ec JB |
8837 | if (new_len < ordered->truncated_len) |
8838 | ordered->truncated_len = new_len; | |
8839 | spin_unlock_irq(&tree->lock); | |
8840 | ||
8841 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
dbfdb6d1 CR |
8842 | start, |
8843 | end - start + 1, 1)) | |
77cef2ec | 8844 | btrfs_finish_ordered_io(ordered); |
8b62b72b | 8845 | } |
e6dcd2dc | 8846 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
8847 | if (!inode_evicting) { |
8848 | cached_state = NULL; | |
dbfdb6d1 | 8849 | lock_extent_bits(tree, start, end, |
131e404a FDBM |
8850 | &cached_state); |
8851 | } | |
dbfdb6d1 CR |
8852 | |
8853 | start = end + 1; | |
8854 | if (start < page_end) | |
8855 | goto again; | |
131e404a FDBM |
8856 | } |
8857 | ||
b9d0b389 QW |
8858 | /* |
8859 | * Qgroup reserved space handler | |
8860 | * Page here will be either | |
8861 | * 1) Already written to disk | |
8862 | * In this case, its reserved space is released from data rsv map | |
8863 | * and will be freed by delayed_ref handler finally. | |
8864 | * So even we call qgroup_free_data(), it won't decrease reserved | |
8865 | * space. | |
8866 | * 2) Not written to disk | |
8867 | * This means the reserved space should be freed here. | |
8868 | */ | |
09cbfeaf | 8869 | btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE); |
131e404a FDBM |
8870 | if (!inode_evicting) { |
8871 | clear_extent_bit(tree, page_start, page_end, | |
8872 | EXTENT_LOCKED | EXTENT_DIRTY | | |
8873 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
8874 | EXTENT_DEFRAG, 1, 1, | |
8875 | &cached_state, GFP_NOFS); | |
8876 | ||
8877 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 8878 | } |
e6dcd2dc | 8879 | |
4a096752 | 8880 | ClearPageChecked(page); |
9ad6b7bc | 8881 | if (PagePrivate(page)) { |
9ad6b7bc CM |
8882 | ClearPagePrivate(page); |
8883 | set_page_private(page, 0); | |
09cbfeaf | 8884 | put_page(page); |
9ad6b7bc | 8885 | } |
39279cc3 CM |
8886 | } |
8887 | ||
9ebefb18 CM |
8888 | /* |
8889 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8890 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8891 | * be careful to check for EOF conditions here. We set the page up correctly | |
8892 | * for a written page which means we get ENOSPC checking when writing into | |
8893 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8894 | * support these features. | |
8895 | * | |
8896 | * We are not allowed to take the i_mutex here so we have to play games to | |
8897 | * protect against truncate races as the page could now be beyond EOF. Because | |
8898 | * vmtruncate() writes the inode size before removing pages, once we have the | |
8899 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
8900 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
8901 | * unlock the page. | |
8902 | */ | |
c2ec175c | 8903 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 8904 | { |
c2ec175c | 8905 | struct page *page = vmf->page; |
496ad9aa | 8906 | struct inode *inode = file_inode(vma->vm_file); |
1832a6d5 | 8907 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
8908 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8909 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8910 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
8911 | char *kaddr; |
8912 | unsigned long zero_start; | |
9ebefb18 | 8913 | loff_t size; |
1832a6d5 | 8914 | int ret; |
9998eb70 | 8915 | int reserved = 0; |
d0b7da88 | 8916 | u64 reserved_space; |
a52d9a80 | 8917 | u64 page_start; |
e6dcd2dc | 8918 | u64 page_end; |
d0b7da88 CR |
8919 | u64 end; |
8920 | ||
09cbfeaf | 8921 | reserved_space = PAGE_SIZE; |
9ebefb18 | 8922 | |
b2b5ef5c | 8923 | sb_start_pagefault(inode->i_sb); |
df480633 | 8924 | page_start = page_offset(page); |
09cbfeaf | 8925 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 8926 | end = page_end; |
df480633 | 8927 | |
d0b7da88 CR |
8928 | /* |
8929 | * Reserving delalloc space after obtaining the page lock can lead to | |
8930 | * deadlock. For example, if a dirty page is locked by this function | |
8931 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
8932 | * dirty page write out, then the btrfs_writepage() function could | |
8933 | * end up waiting indefinitely to get a lock on the page currently | |
8934 | * being processed by btrfs_page_mkwrite() function. | |
8935 | */ | |
7cf5b976 | 8936 | ret = btrfs_delalloc_reserve_space(inode, page_start, |
d0b7da88 | 8937 | reserved_space); |
9998eb70 | 8938 | if (!ret) { |
e41f941a | 8939 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
8940 | reserved = 1; |
8941 | } | |
56a76f82 NP |
8942 | if (ret) { |
8943 | if (ret == -ENOMEM) | |
8944 | ret = VM_FAULT_OOM; | |
8945 | else /* -ENOSPC, -EIO, etc */ | |
8946 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
8947 | if (reserved) |
8948 | goto out; | |
8949 | goto out_noreserve; | |
56a76f82 | 8950 | } |
1832a6d5 | 8951 | |
56a76f82 | 8952 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8953 | again: |
9ebefb18 | 8954 | lock_page(page); |
9ebefb18 | 8955 | size = i_size_read(inode); |
a52d9a80 | 8956 | |
9ebefb18 | 8957 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8958 | (page_start >= size)) { |
9ebefb18 CM |
8959 | /* page got truncated out from underneath us */ |
8960 | goto out_unlock; | |
8961 | } | |
e6dcd2dc CM |
8962 | wait_on_page_writeback(page); |
8963 | ||
ff13db41 | 8964 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc CM |
8965 | set_page_extent_mapped(page); |
8966 | ||
eb84ae03 CM |
8967 | /* |
8968 | * we can't set the delalloc bits if there are pending ordered | |
8969 | * extents. Drop our locks and wait for them to finish | |
8970 | */ | |
d0b7da88 | 8971 | ordered = btrfs_lookup_ordered_range(inode, page_start, page_end); |
e6dcd2dc | 8972 | if (ordered) { |
2ac55d41 JB |
8973 | unlock_extent_cached(io_tree, page_start, page_end, |
8974 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 8975 | unlock_page(page); |
eb84ae03 | 8976 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
8977 | btrfs_put_ordered_extent(ordered); |
8978 | goto again; | |
8979 | } | |
8980 | ||
09cbfeaf | 8981 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
d0b7da88 | 8982 | reserved_space = round_up(size - page_start, root->sectorsize); |
09cbfeaf | 8983 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 CR |
8984 | end = page_start + reserved_space - 1; |
8985 | spin_lock(&BTRFS_I(inode)->lock); | |
8986 | BTRFS_I(inode)->outstanding_extents++; | |
8987 | spin_unlock(&BTRFS_I(inode)->lock); | |
8988 | btrfs_delalloc_release_space(inode, page_start, | |
09cbfeaf | 8989 | PAGE_SIZE - reserved_space); |
d0b7da88 CR |
8990 | } |
8991 | } | |
8992 | ||
fbf19087 JB |
8993 | /* |
8994 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
8995 | * if it was already dirty, so for space accounting reasons we need to | |
8996 | * clear any delalloc bits for the range we are fixing to save. There | |
8997 | * is probably a better way to do this, but for now keep consistent with | |
8998 | * prepare_pages in the normal write path. | |
8999 | */ | |
d0b7da88 | 9000 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
9e8a4a8b LB |
9001 | EXTENT_DIRTY | EXTENT_DELALLOC | |
9002 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 9003 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 9004 | |
d0b7da88 | 9005 | ret = btrfs_set_extent_delalloc(inode, page_start, end, |
2ac55d41 | 9006 | &cached_state); |
9ed74f2d | 9007 | if (ret) { |
2ac55d41 JB |
9008 | unlock_extent_cached(io_tree, page_start, page_end, |
9009 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
9010 | ret = VM_FAULT_SIGBUS; |
9011 | goto out_unlock; | |
9012 | } | |
e6dcd2dc | 9013 | ret = 0; |
9ebefb18 CM |
9014 | |
9015 | /* page is wholly or partially inside EOF */ | |
09cbfeaf KS |
9016 | if (page_start + PAGE_SIZE > size) |
9017 | zero_start = size & ~PAGE_MASK; | |
9ebefb18 | 9018 | else |
09cbfeaf | 9019 | zero_start = PAGE_SIZE; |
9ebefb18 | 9020 | |
09cbfeaf | 9021 | if (zero_start != PAGE_SIZE) { |
e6dcd2dc | 9022 | kaddr = kmap(page); |
09cbfeaf | 9023 | memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start); |
e6dcd2dc CM |
9024 | flush_dcache_page(page); |
9025 | kunmap(page); | |
9026 | } | |
247e743c | 9027 | ClearPageChecked(page); |
e6dcd2dc | 9028 | set_page_dirty(page); |
50a9b214 | 9029 | SetPageUptodate(page); |
5a3f23d5 | 9030 | |
257c62e1 CM |
9031 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
9032 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 9033 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 9034 | |
2ac55d41 | 9035 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
9036 | |
9037 | out_unlock: | |
b2b5ef5c JK |
9038 | if (!ret) { |
9039 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 9040 | return VM_FAULT_LOCKED; |
b2b5ef5c | 9041 | } |
9ebefb18 | 9042 | unlock_page(page); |
1832a6d5 | 9043 | out: |
d0b7da88 | 9044 | btrfs_delalloc_release_space(inode, page_start, reserved_space); |
9998eb70 | 9045 | out_noreserve: |
b2b5ef5c | 9046 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
9047 | return ret; |
9048 | } | |
9049 | ||
a41ad394 | 9050 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
9051 | { |
9052 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 9053 | struct btrfs_block_rsv *rsv; |
a71754fc | 9054 | int ret = 0; |
3893e33b | 9055 | int err = 0; |
39279cc3 | 9056 | struct btrfs_trans_handle *trans; |
dbe674a9 | 9057 | u64 mask = root->sectorsize - 1; |
07127184 | 9058 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 9059 | |
0ef8b726 JB |
9060 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), |
9061 | (u64)-1); | |
9062 | if (ret) | |
9063 | return ret; | |
39279cc3 | 9064 | |
fcb80c2a | 9065 | /* |
01327610 | 9066 | * Yes ladies and gentlemen, this is indeed ugly. The fact is we have |
fcb80c2a JB |
9067 | * 3 things going on here |
9068 | * | |
9069 | * 1) We need to reserve space for our orphan item and the space to | |
9070 | * delete our orphan item. Lord knows we don't want to have a dangling | |
9071 | * orphan item because we didn't reserve space to remove it. | |
9072 | * | |
9073 | * 2) We need to reserve space to update our inode. | |
9074 | * | |
9075 | * 3) We need to have something to cache all the space that is going to | |
9076 | * be free'd up by the truncate operation, but also have some slack | |
9077 | * space reserved in case it uses space during the truncate (thank you | |
9078 | * very much snapshotting). | |
9079 | * | |
01327610 | 9080 | * And we need these to all be separate. The fact is we can use a lot of |
fcb80c2a | 9081 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 9082 | * we will use, so we need the truncate reservation to be separate so it |
fcb80c2a JB |
9083 | * doesn't end up using space reserved for updating the inode or |
9084 | * removing the orphan item. We also need to be able to stop the | |
9085 | * transaction and start a new one, which means we need to be able to | |
9086 | * update the inode several times, and we have no idea of knowing how | |
9087 | * many times that will be, so we can't just reserve 1 item for the | |
01327610 | 9088 | * entirety of the operation, so that has to be done separately as well. |
fcb80c2a JB |
9089 | * Then there is the orphan item, which does indeed need to be held on |
9090 | * to for the whole operation, and we need nobody to touch this reserved | |
9091 | * space except the orphan code. | |
9092 | * | |
9093 | * So that leaves us with | |
9094 | * | |
9095 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
9096 | * 2) rsv - for the truncate reservation, which we will steal from the | |
9097 | * transaction reservation. | |
9098 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
9099 | * updating the inode. | |
9100 | */ | |
66d8f3dd | 9101 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
9102 | if (!rsv) |
9103 | return -ENOMEM; | |
4a338542 | 9104 | rsv->size = min_size; |
ca7e70f5 | 9105 | rsv->failfast = 1; |
f0cd846e | 9106 | |
907cbceb | 9107 | /* |
07127184 | 9108 | * 1 for the truncate slack space |
907cbceb JB |
9109 | * 1 for updating the inode. |
9110 | */ | |
f3fe820c | 9111 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
9112 | if (IS_ERR(trans)) { |
9113 | err = PTR_ERR(trans); | |
9114 | goto out; | |
9115 | } | |
f0cd846e | 9116 | |
907cbceb JB |
9117 | /* Migrate the slack space for the truncate to our reserve */ |
9118 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
25d609f8 | 9119 | min_size, 0); |
fcb80c2a | 9120 | BUG_ON(ret); |
f0cd846e | 9121 | |
5dc562c5 JB |
9122 | /* |
9123 | * So if we truncate and then write and fsync we normally would just | |
9124 | * write the extents that changed, which is a problem if we need to | |
9125 | * first truncate that entire inode. So set this flag so we write out | |
9126 | * all of the extents in the inode to the sync log so we're completely | |
9127 | * safe. | |
9128 | */ | |
9129 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 9130 | trans->block_rsv = rsv; |
907cbceb | 9131 | |
8082510e YZ |
9132 | while (1) { |
9133 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
9134 | inode->i_size, | |
9135 | BTRFS_EXTENT_DATA_KEY); | |
28ed1345 | 9136 | if (ret != -ENOSPC && ret != -EAGAIN) { |
3893e33b | 9137 | err = ret; |
8082510e | 9138 | break; |
3893e33b | 9139 | } |
39279cc3 | 9140 | |
fcb80c2a | 9141 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 9142 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
9143 | if (ret) { |
9144 | err = ret; | |
9145 | break; | |
9146 | } | |
ca7e70f5 | 9147 | |
8082510e | 9148 | btrfs_end_transaction(trans, root); |
b53d3f5d | 9149 | btrfs_btree_balance_dirty(root); |
ca7e70f5 JB |
9150 | |
9151 | trans = btrfs_start_transaction(root, 2); | |
9152 | if (IS_ERR(trans)) { | |
9153 | ret = err = PTR_ERR(trans); | |
9154 | trans = NULL; | |
9155 | break; | |
9156 | } | |
9157 | ||
9158 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
25d609f8 | 9159 | rsv, min_size, 0); |
ca7e70f5 JB |
9160 | BUG_ON(ret); /* shouldn't happen */ |
9161 | trans->block_rsv = rsv; | |
8082510e YZ |
9162 | } |
9163 | ||
9164 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 9165 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 9166 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
9167 | if (ret) |
9168 | err = ret; | |
8082510e YZ |
9169 | } |
9170 | ||
917c16b2 CM |
9171 | if (trans) { |
9172 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
9173 | ret = btrfs_update_inode(trans, root, inode); | |
9174 | if (ret && !err) | |
9175 | err = ret; | |
7b128766 | 9176 | |
7ad85bb7 | 9177 | ret = btrfs_end_transaction(trans, root); |
b53d3f5d | 9178 | btrfs_btree_balance_dirty(root); |
917c16b2 | 9179 | } |
fcb80c2a JB |
9180 | out: |
9181 | btrfs_free_block_rsv(root, rsv); | |
9182 | ||
3893e33b JB |
9183 | if (ret && !err) |
9184 | err = ret; | |
a41ad394 | 9185 | |
3893e33b | 9186 | return err; |
39279cc3 CM |
9187 | } |
9188 | ||
d352ac68 CM |
9189 | /* |
9190 | * create a new subvolume directory/inode (helper for the ioctl). | |
9191 | */ | |
d2fb3437 | 9192 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
9193 | struct btrfs_root *new_root, |
9194 | struct btrfs_root *parent_root, | |
9195 | u64 new_dirid) | |
39279cc3 | 9196 | { |
39279cc3 | 9197 | struct inode *inode; |
76dda93c | 9198 | int err; |
00e4e6b3 | 9199 | u64 index = 0; |
39279cc3 | 9200 | |
12fc9d09 FA |
9201 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
9202 | new_dirid, new_dirid, | |
9203 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
9204 | &index); | |
54aa1f4d | 9205 | if (IS_ERR(inode)) |
f46b5a66 | 9206 | return PTR_ERR(inode); |
39279cc3 CM |
9207 | inode->i_op = &btrfs_dir_inode_operations; |
9208 | inode->i_fop = &btrfs_dir_file_operations; | |
9209 | ||
bfe86848 | 9210 | set_nlink(inode, 1); |
dbe674a9 | 9211 | btrfs_i_size_write(inode, 0); |
b0d5d10f | 9212 | unlock_new_inode(inode); |
3b96362c | 9213 | |
63541927 FDBM |
9214 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
9215 | if (err) | |
9216 | btrfs_err(new_root->fs_info, | |
351fd353 | 9217 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
9218 | new_root->root_key.objectid, err); |
9219 | ||
76dda93c | 9220 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 9221 | |
76dda93c | 9222 | iput(inode); |
ce598979 | 9223 | return err; |
39279cc3 CM |
9224 | } |
9225 | ||
39279cc3 CM |
9226 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
9227 | { | |
9228 | struct btrfs_inode *ei; | |
2ead6ae7 | 9229 | struct inode *inode; |
39279cc3 CM |
9230 | |
9231 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
9232 | if (!ei) | |
9233 | return NULL; | |
2ead6ae7 YZ |
9234 | |
9235 | ei->root = NULL; | |
2ead6ae7 | 9236 | ei->generation = 0; |
15ee9bc7 | 9237 | ei->last_trans = 0; |
257c62e1 | 9238 | ei->last_sub_trans = 0; |
e02119d5 | 9239 | ei->logged_trans = 0; |
2ead6ae7 | 9240 | ei->delalloc_bytes = 0; |
47059d93 | 9241 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
9242 | ei->disk_i_size = 0; |
9243 | ei->flags = 0; | |
7709cde3 | 9244 | ei->csum_bytes = 0; |
2ead6ae7 | 9245 | ei->index_cnt = (u64)-1; |
67de1176 | 9246 | ei->dir_index = 0; |
2ead6ae7 | 9247 | ei->last_unlink_trans = 0; |
46d8bc34 | 9248 | ei->last_log_commit = 0; |
8089fe62 | 9249 | ei->delayed_iput_count = 0; |
2ead6ae7 | 9250 | |
9e0baf60 JB |
9251 | spin_lock_init(&ei->lock); |
9252 | ei->outstanding_extents = 0; | |
9253 | ei->reserved_extents = 0; | |
2ead6ae7 | 9254 | |
72ac3c0d | 9255 | ei->runtime_flags = 0; |
261507a0 | 9256 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 9257 | |
16cdcec7 MX |
9258 | ei->delayed_node = NULL; |
9259 | ||
9cc97d64 | 9260 | ei->i_otime.tv_sec = 0; |
9261 | ei->i_otime.tv_nsec = 0; | |
9262 | ||
2ead6ae7 | 9263 | inode = &ei->vfs_inode; |
a8067e02 | 9264 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
9265 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
9266 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
9267 | ei->io_tree.track_uptodate = 1; |
9268 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 9269 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 9270 | mutex_init(&ei->log_mutex); |
f248679e | 9271 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 9272 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 9273 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 9274 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 9275 | RB_CLEAR_NODE(&ei->rb_node); |
5f9a8a51 | 9276 | init_rwsem(&ei->dio_sem); |
2ead6ae7 YZ |
9277 | |
9278 | return inode; | |
39279cc3 CM |
9279 | } |
9280 | ||
aaedb55b JB |
9281 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
9282 | void btrfs_test_destroy_inode(struct inode *inode) | |
9283 | { | |
9284 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); | |
9285 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
9286 | } | |
9287 | #endif | |
9288 | ||
fa0d7e3d NP |
9289 | static void btrfs_i_callback(struct rcu_head *head) |
9290 | { | |
9291 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
9292 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9293 | } | |
9294 | ||
39279cc3 CM |
9295 | void btrfs_destroy_inode(struct inode *inode) |
9296 | { | |
e6dcd2dc | 9297 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
9298 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9299 | ||
b3d9b7a3 | 9300 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 9301 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
9302 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
9303 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
9304 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
9305 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
47059d93 | 9306 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 9307 | |
a6dbd429 JB |
9308 | /* |
9309 | * This can happen where we create an inode, but somebody else also | |
9310 | * created the same inode and we need to destroy the one we already | |
9311 | * created. | |
9312 | */ | |
9313 | if (!root) | |
9314 | goto free; | |
9315 | ||
8a35d95f JB |
9316 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
9317 | &BTRFS_I(inode)->runtime_flags)) { | |
c2cf52eb | 9318 | btrfs_info(root->fs_info, "inode %llu still on the orphan list", |
c1c9ff7c | 9319 | btrfs_ino(inode)); |
8a35d95f | 9320 | atomic_dec(&root->orphan_inodes); |
7b128766 | 9321 | } |
7b128766 | 9322 | |
d397712b | 9323 | while (1) { |
e6dcd2dc CM |
9324 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
9325 | if (!ordered) | |
9326 | break; | |
9327 | else { | |
c2cf52eb | 9328 | btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup", |
c1c9ff7c | 9329 | ordered->file_offset, ordered->len); |
e6dcd2dc CM |
9330 | btrfs_remove_ordered_extent(inode, ordered); |
9331 | btrfs_put_ordered_extent(ordered); | |
9332 | btrfs_put_ordered_extent(ordered); | |
9333 | } | |
9334 | } | |
56fa9d07 | 9335 | btrfs_qgroup_check_reserved_leak(inode); |
5d4f98a2 | 9336 | inode_tree_del(inode); |
5b21f2ed | 9337 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 9338 | free: |
fa0d7e3d | 9339 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
9340 | } |
9341 | ||
45321ac5 | 9342 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
9343 | { |
9344 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 9345 | |
6379ef9f NA |
9346 | if (root == NULL) |
9347 | return 1; | |
9348 | ||
fa6ac876 | 9349 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 9350 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 9351 | return 1; |
76dda93c | 9352 | else |
45321ac5 | 9353 | return generic_drop_inode(inode); |
76dda93c YZ |
9354 | } |
9355 | ||
0ee0fda0 | 9356 | static void init_once(void *foo) |
39279cc3 CM |
9357 | { |
9358 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
9359 | ||
9360 | inode_init_once(&ei->vfs_inode); | |
9361 | } | |
9362 | ||
9363 | void btrfs_destroy_cachep(void) | |
9364 | { | |
8c0a8537 KS |
9365 | /* |
9366 | * Make sure all delayed rcu free inodes are flushed before we | |
9367 | * destroy cache. | |
9368 | */ | |
9369 | rcu_barrier(); | |
5598e900 KM |
9370 | kmem_cache_destroy(btrfs_inode_cachep); |
9371 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
9372 | kmem_cache_destroy(btrfs_transaction_cachep); | |
9373 | kmem_cache_destroy(btrfs_path_cachep); | |
9374 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
9375 | } |
9376 | ||
9377 | int btrfs_init_cachep(void) | |
9378 | { | |
837e1972 | 9379 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 9380 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
9381 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
9382 | init_once); | |
39279cc3 CM |
9383 | if (!btrfs_inode_cachep) |
9384 | goto fail; | |
9601e3f6 | 9385 | |
837e1972 | 9386 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 CH |
9387 | sizeof(struct btrfs_trans_handle), 0, |
9388 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
9389 | if (!btrfs_trans_handle_cachep) |
9390 | goto fail; | |
9601e3f6 | 9391 | |
837e1972 | 9392 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 CH |
9393 | sizeof(struct btrfs_transaction), 0, |
9394 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
9395 | if (!btrfs_transaction_cachep) |
9396 | goto fail; | |
9601e3f6 | 9397 | |
837e1972 | 9398 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 CH |
9399 | sizeof(struct btrfs_path), 0, |
9400 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
9401 | if (!btrfs_path_cachep) |
9402 | goto fail; | |
9601e3f6 | 9403 | |
837e1972 | 9404 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 JB |
9405 | sizeof(struct btrfs_free_space), 0, |
9406 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
9407 | if (!btrfs_free_space_cachep) | |
9408 | goto fail; | |
9409 | ||
39279cc3 CM |
9410 | return 0; |
9411 | fail: | |
9412 | btrfs_destroy_cachep(); | |
9413 | return -ENOMEM; | |
9414 | } | |
9415 | ||
9416 | static int btrfs_getattr(struct vfsmount *mnt, | |
9417 | struct dentry *dentry, struct kstat *stat) | |
9418 | { | |
df0af1a5 | 9419 | u64 delalloc_bytes; |
2b0143b5 | 9420 | struct inode *inode = d_inode(dentry); |
fadc0d8b DS |
9421 | u32 blocksize = inode->i_sb->s_blocksize; |
9422 | ||
39279cc3 | 9423 | generic_fillattr(inode, stat); |
0ee5dc67 | 9424 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
9425 | |
9426 | spin_lock(&BTRFS_I(inode)->lock); | |
9427 | delalloc_bytes = BTRFS_I(inode)->delalloc_bytes; | |
9428 | spin_unlock(&BTRFS_I(inode)->lock); | |
fadc0d8b | 9429 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 9430 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
9431 | return 0; |
9432 | } | |
9433 | ||
cdd1fedf DF |
9434 | static int btrfs_rename_exchange(struct inode *old_dir, |
9435 | struct dentry *old_dentry, | |
9436 | struct inode *new_dir, | |
9437 | struct dentry *new_dentry) | |
9438 | { | |
9439 | struct btrfs_trans_handle *trans; | |
9440 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
9441 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
9442 | struct inode *new_inode = new_dentry->d_inode; | |
9443 | struct inode *old_inode = old_dentry->d_inode; | |
9444 | struct timespec ctime = CURRENT_TIME; | |
9445 | struct dentry *parent; | |
9446 | u64 old_ino = btrfs_ino(old_inode); | |
9447 | u64 new_ino = btrfs_ino(new_inode); | |
9448 | u64 old_idx = 0; | |
9449 | u64 new_idx = 0; | |
9450 | u64 root_objectid; | |
9451 | int ret; | |
86e8aa0e FM |
9452 | bool root_log_pinned = false; |
9453 | bool dest_log_pinned = false; | |
cdd1fedf DF |
9454 | |
9455 | /* we only allow rename subvolume link between subvolumes */ | |
9456 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
9457 | return -EXDEV; | |
9458 | ||
9459 | /* close the race window with snapshot create/destroy ioctl */ | |
9460 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
9461 | down_read(&root->fs_info->subvol_sem); | |
9462 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
9463 | down_read(&dest->fs_info->subvol_sem); | |
9464 | ||
9465 | /* | |
9466 | * We want to reserve the absolute worst case amount of items. So if | |
9467 | * both inodes are subvols and we need to unlink them then that would | |
9468 | * require 4 item modifications, but if they are both normal inodes it | |
9469 | * would require 5 item modifications, so we'll assume their normal | |
9470 | * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items | |
9471 | * should cover the worst case number of items we'll modify. | |
9472 | */ | |
9473 | trans = btrfs_start_transaction(root, 12); | |
9474 | if (IS_ERR(trans)) { | |
9475 | ret = PTR_ERR(trans); | |
9476 | goto out_notrans; | |
9477 | } | |
9478 | ||
9479 | /* | |
9480 | * We need to find a free sequence number both in the source and | |
9481 | * in the destination directory for the exchange. | |
9482 | */ | |
9483 | ret = btrfs_set_inode_index(new_dir, &old_idx); | |
9484 | if (ret) | |
9485 | goto out_fail; | |
9486 | ret = btrfs_set_inode_index(old_dir, &new_idx); | |
9487 | if (ret) | |
9488 | goto out_fail; | |
9489 | ||
9490 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
9491 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
9492 | ||
9493 | /* Reference for the source. */ | |
9494 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9495 | /* force full log commit if subvolume involved. */ | |
9496 | btrfs_set_log_full_commit(root->fs_info, trans); | |
9497 | } else { | |
376e5a57 FM |
9498 | btrfs_pin_log_trans(root); |
9499 | root_log_pinned = true; | |
cdd1fedf DF |
9500 | ret = btrfs_insert_inode_ref(trans, dest, |
9501 | new_dentry->d_name.name, | |
9502 | new_dentry->d_name.len, | |
9503 | old_ino, | |
9504 | btrfs_ino(new_dir), old_idx); | |
9505 | if (ret) | |
9506 | goto out_fail; | |
cdd1fedf DF |
9507 | } |
9508 | ||
9509 | /* And now for the dest. */ | |
9510 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9511 | /* force full log commit if subvolume involved. */ | |
9512 | btrfs_set_log_full_commit(dest->fs_info, trans); | |
9513 | } else { | |
376e5a57 FM |
9514 | btrfs_pin_log_trans(dest); |
9515 | dest_log_pinned = true; | |
cdd1fedf DF |
9516 | ret = btrfs_insert_inode_ref(trans, root, |
9517 | old_dentry->d_name.name, | |
9518 | old_dentry->d_name.len, | |
9519 | new_ino, | |
9520 | btrfs_ino(old_dir), new_idx); | |
9521 | if (ret) | |
9522 | goto out_fail; | |
cdd1fedf DF |
9523 | } |
9524 | ||
9525 | /* Update inode version and ctime/mtime. */ | |
9526 | inode_inc_iversion(old_dir); | |
9527 | inode_inc_iversion(new_dir); | |
9528 | inode_inc_iversion(old_inode); | |
9529 | inode_inc_iversion(new_inode); | |
9530 | old_dir->i_ctime = old_dir->i_mtime = ctime; | |
9531 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
9532 | old_inode->i_ctime = ctime; | |
9533 | new_inode->i_ctime = ctime; | |
9534 | ||
9535 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
9536 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
9537 | btrfs_record_unlink_dir(trans, new_dir, new_inode, 1); | |
9538 | } | |
9539 | ||
9540 | /* src is a subvolume */ | |
9541 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9542 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
9543 | ret = btrfs_unlink_subvol(trans, root, old_dir, | |
9544 | root_objectid, | |
9545 | old_dentry->d_name.name, | |
9546 | old_dentry->d_name.len); | |
9547 | } else { /* src is an inode */ | |
9548 | ret = __btrfs_unlink_inode(trans, root, old_dir, | |
9549 | old_dentry->d_inode, | |
9550 | old_dentry->d_name.name, | |
9551 | old_dentry->d_name.len); | |
9552 | if (!ret) | |
9553 | ret = btrfs_update_inode(trans, root, old_inode); | |
9554 | } | |
9555 | if (ret) { | |
9556 | btrfs_abort_transaction(trans, root, ret); | |
9557 | goto out_fail; | |
9558 | } | |
9559 | ||
9560 | /* dest is a subvolume */ | |
9561 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9562 | root_objectid = BTRFS_I(new_inode)->root->root_key.objectid; | |
9563 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
9564 | root_objectid, | |
9565 | new_dentry->d_name.name, | |
9566 | new_dentry->d_name.len); | |
9567 | } else { /* dest is an inode */ | |
9568 | ret = __btrfs_unlink_inode(trans, dest, new_dir, | |
9569 | new_dentry->d_inode, | |
9570 | new_dentry->d_name.name, | |
9571 | new_dentry->d_name.len); | |
9572 | if (!ret) | |
9573 | ret = btrfs_update_inode(trans, dest, new_inode); | |
9574 | } | |
9575 | if (ret) { | |
9576 | btrfs_abort_transaction(trans, root, ret); | |
9577 | goto out_fail; | |
9578 | } | |
9579 | ||
9580 | ret = btrfs_add_link(trans, new_dir, old_inode, | |
9581 | new_dentry->d_name.name, | |
9582 | new_dentry->d_name.len, 0, old_idx); | |
9583 | if (ret) { | |
9584 | btrfs_abort_transaction(trans, root, ret); | |
9585 | goto out_fail; | |
9586 | } | |
9587 | ||
9588 | ret = btrfs_add_link(trans, old_dir, new_inode, | |
9589 | old_dentry->d_name.name, | |
9590 | old_dentry->d_name.len, 0, new_idx); | |
9591 | if (ret) { | |
9592 | btrfs_abort_transaction(trans, root, ret); | |
9593 | goto out_fail; | |
9594 | } | |
9595 | ||
9596 | if (old_inode->i_nlink == 1) | |
9597 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9598 | if (new_inode->i_nlink == 1) | |
9599 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9600 | ||
86e8aa0e | 9601 | if (root_log_pinned) { |
cdd1fedf DF |
9602 | parent = new_dentry->d_parent; |
9603 | btrfs_log_new_name(trans, old_inode, old_dir, parent); | |
9604 | btrfs_end_log_trans(root); | |
86e8aa0e | 9605 | root_log_pinned = false; |
cdd1fedf | 9606 | } |
86e8aa0e | 9607 | if (dest_log_pinned) { |
cdd1fedf DF |
9608 | parent = old_dentry->d_parent; |
9609 | btrfs_log_new_name(trans, new_inode, new_dir, parent); | |
9610 | btrfs_end_log_trans(dest); | |
86e8aa0e | 9611 | dest_log_pinned = false; |
cdd1fedf DF |
9612 | } |
9613 | out_fail: | |
86e8aa0e FM |
9614 | /* |
9615 | * If we have pinned a log and an error happened, we unpin tasks | |
9616 | * trying to sync the log and force them to fallback to a transaction | |
9617 | * commit if the log currently contains any of the inodes involved in | |
9618 | * this rename operation (to ensure we do not persist a log with an | |
9619 | * inconsistent state for any of these inodes or leading to any | |
9620 | * inconsistencies when replayed). If the transaction was aborted, the | |
9621 | * abortion reason is propagated to userspace when attempting to commit | |
9622 | * the transaction. If the log does not contain any of these inodes, we | |
9623 | * allow the tasks to sync it. | |
9624 | */ | |
9625 | if (ret && (root_log_pinned || dest_log_pinned)) { | |
9626 | if (btrfs_inode_in_log(old_dir, root->fs_info->generation) || | |
9627 | btrfs_inode_in_log(new_dir, root->fs_info->generation) || | |
9628 | btrfs_inode_in_log(old_inode, root->fs_info->generation) || | |
9629 | (new_inode && | |
9630 | btrfs_inode_in_log(new_inode, root->fs_info->generation))) | |
9631 | btrfs_set_log_full_commit(root->fs_info, trans); | |
9632 | ||
9633 | if (root_log_pinned) { | |
9634 | btrfs_end_log_trans(root); | |
9635 | root_log_pinned = false; | |
9636 | } | |
9637 | if (dest_log_pinned) { | |
9638 | btrfs_end_log_trans(dest); | |
9639 | dest_log_pinned = false; | |
9640 | } | |
9641 | } | |
cdd1fedf DF |
9642 | ret = btrfs_end_transaction(trans, root); |
9643 | out_notrans: | |
9644 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
9645 | up_read(&dest->fs_info->subvol_sem); | |
9646 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
9647 | up_read(&root->fs_info->subvol_sem); | |
9648 | ||
9649 | return ret; | |
9650 | } | |
9651 | ||
9652 | static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans, | |
9653 | struct btrfs_root *root, | |
9654 | struct inode *dir, | |
9655 | struct dentry *dentry) | |
9656 | { | |
9657 | int ret; | |
9658 | struct inode *inode; | |
9659 | u64 objectid; | |
9660 | u64 index; | |
9661 | ||
9662 | ret = btrfs_find_free_ino(root, &objectid); | |
9663 | if (ret) | |
9664 | return ret; | |
9665 | ||
9666 | inode = btrfs_new_inode(trans, root, dir, | |
9667 | dentry->d_name.name, | |
9668 | dentry->d_name.len, | |
9669 | btrfs_ino(dir), | |
9670 | objectid, | |
9671 | S_IFCHR | WHITEOUT_MODE, | |
9672 | &index); | |
9673 | ||
9674 | if (IS_ERR(inode)) { | |
9675 | ret = PTR_ERR(inode); | |
9676 | return ret; | |
9677 | } | |
9678 | ||
9679 | inode->i_op = &btrfs_special_inode_operations; | |
9680 | init_special_inode(inode, inode->i_mode, | |
9681 | WHITEOUT_DEV); | |
9682 | ||
9683 | ret = btrfs_init_inode_security(trans, inode, dir, | |
9684 | &dentry->d_name); | |
9685 | if (ret) | |
c9901618 | 9686 | goto out; |
cdd1fedf DF |
9687 | |
9688 | ret = btrfs_add_nondir(trans, dir, dentry, | |
9689 | inode, 0, index); | |
9690 | if (ret) | |
c9901618 | 9691 | goto out; |
cdd1fedf DF |
9692 | |
9693 | ret = btrfs_update_inode(trans, root, inode); | |
c9901618 | 9694 | out: |
cdd1fedf | 9695 | unlock_new_inode(inode); |
c9901618 FM |
9696 | if (ret) |
9697 | inode_dec_link_count(inode); | |
cdd1fedf DF |
9698 | iput(inode); |
9699 | ||
c9901618 | 9700 | return ret; |
cdd1fedf DF |
9701 | } |
9702 | ||
d397712b | 9703 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
cdd1fedf DF |
9704 | struct inode *new_dir, struct dentry *new_dentry, |
9705 | unsigned int flags) | |
39279cc3 CM |
9706 | { |
9707 | struct btrfs_trans_handle *trans; | |
5062af35 | 9708 | unsigned int trans_num_items; |
39279cc3 | 9709 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9710 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9711 | struct inode *new_inode = d_inode(new_dentry); |
9712 | struct inode *old_inode = d_inode(old_dentry); | |
00e4e6b3 | 9713 | u64 index = 0; |
4df27c4d | 9714 | u64 root_objectid; |
39279cc3 | 9715 | int ret; |
33345d01 | 9716 | u64 old_ino = btrfs_ino(old_inode); |
3dc9e8f7 | 9717 | bool log_pinned = false; |
39279cc3 | 9718 | |
33345d01 | 9719 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9720 | return -EPERM; |
9721 | ||
4df27c4d | 9722 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9723 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9724 | return -EXDEV; |
9725 | ||
33345d01 LZ |
9726 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
9727 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 9728 | return -ENOTEMPTY; |
5f39d397 | 9729 | |
4df27c4d YZ |
9730 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9731 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9732 | return -ENOTEMPTY; | |
9c52057c CM |
9733 | |
9734 | ||
9735 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9736 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9737 | new_dentry->d_name.name, |
9738 | new_dentry->d_name.len); | |
9739 | ||
9740 | if (ret) { | |
9741 | if (ret == -EEXIST) { | |
9742 | /* we shouldn't get | |
9743 | * eexist without a new_inode */ | |
fae7f21c | 9744 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9745 | return ret; |
9746 | } | |
9747 | } else { | |
9748 | /* maybe -EOVERFLOW */ | |
9749 | return ret; | |
9750 | } | |
9751 | } | |
9752 | ret = 0; | |
9753 | ||
5a3f23d5 | 9754 | /* |
8d875f95 CM |
9755 | * we're using rename to replace one file with another. Start IO on it |
9756 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9757 | */ |
8d875f95 | 9758 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9759 | filemap_flush(old_inode->i_mapping); |
9760 | ||
76dda93c | 9761 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 9762 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 9763 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
9764 | /* |
9765 | * We want to reserve the absolute worst case amount of items. So if | |
9766 | * both inodes are subvols and we need to unlink them then that would | |
9767 | * require 4 item modifications, but if they are both normal inodes it | |
cdd1fedf | 9768 | * would require 5 item modifications, so we'll assume they are normal |
a22285a6 YZ |
9769 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items |
9770 | * should cover the worst case number of items we'll modify. | |
5062af35 FM |
9771 | * If our rename has the whiteout flag, we need more 5 units for the |
9772 | * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item | |
9773 | * when selinux is enabled). | |
a22285a6 | 9774 | */ |
5062af35 FM |
9775 | trans_num_items = 11; |
9776 | if (flags & RENAME_WHITEOUT) | |
9777 | trans_num_items += 5; | |
9778 | trans = btrfs_start_transaction(root, trans_num_items); | |
b44c59a8 | 9779 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9780 | ret = PTR_ERR(trans); |
9781 | goto out_notrans; | |
9782 | } | |
76dda93c | 9783 | |
4df27c4d YZ |
9784 | if (dest != root) |
9785 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 9786 | |
a5719521 YZ |
9787 | ret = btrfs_set_inode_index(new_dir, &index); |
9788 | if (ret) | |
9789 | goto out_fail; | |
5a3f23d5 | 9790 | |
67de1176 | 9791 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9792 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9793 | /* force full log commit if subvolume involved. */ |
995946dd | 9794 | btrfs_set_log_full_commit(root->fs_info, trans); |
4df27c4d | 9795 | } else { |
c4aba954 FM |
9796 | btrfs_pin_log_trans(root); |
9797 | log_pinned = true; | |
a5719521 YZ |
9798 | ret = btrfs_insert_inode_ref(trans, dest, |
9799 | new_dentry->d_name.name, | |
9800 | new_dentry->d_name.len, | |
33345d01 LZ |
9801 | old_ino, |
9802 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
9803 | if (ret) |
9804 | goto out_fail; | |
4df27c4d | 9805 | } |
5a3f23d5 | 9806 | |
0c4d2d95 JB |
9807 | inode_inc_iversion(old_dir); |
9808 | inode_inc_iversion(new_dir); | |
9809 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
9810 | old_dir->i_ctime = old_dir->i_mtime = |
9811 | new_dir->i_ctime = new_dir->i_mtime = | |
9812 | old_inode->i_ctime = current_fs_time(old_dir->i_sb); | |
5f39d397 | 9813 | |
12fcfd22 CM |
9814 | if (old_dentry->d_parent != new_dentry->d_parent) |
9815 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
9816 | ||
33345d01 | 9817 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
9818 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
9819 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
9820 | old_dentry->d_name.name, | |
9821 | old_dentry->d_name.len); | |
9822 | } else { | |
92986796 | 9823 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
2b0143b5 | 9824 | d_inode(old_dentry), |
92986796 AV |
9825 | old_dentry->d_name.name, |
9826 | old_dentry->d_name.len); | |
9827 | if (!ret) | |
9828 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 9829 | } |
79787eaa JM |
9830 | if (ret) { |
9831 | btrfs_abort_transaction(trans, root, ret); | |
9832 | goto out_fail; | |
9833 | } | |
39279cc3 CM |
9834 | |
9835 | if (new_inode) { | |
0c4d2d95 | 9836 | inode_inc_iversion(new_inode); |
04b285f3 | 9837 | new_inode->i_ctime = current_fs_time(new_inode->i_sb); |
33345d01 | 9838 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
9839 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
9840 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
9841 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
9842 | root_objectid, | |
9843 | new_dentry->d_name.name, | |
9844 | new_dentry->d_name.len); | |
9845 | BUG_ON(new_inode->i_nlink == 0); | |
9846 | } else { | |
9847 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
2b0143b5 | 9848 | d_inode(new_dentry), |
4df27c4d YZ |
9849 | new_dentry->d_name.name, |
9850 | new_dentry->d_name.len); | |
9851 | } | |
4ef31a45 | 9852 | if (!ret && new_inode->i_nlink == 0) |
2b0143b5 | 9853 | ret = btrfs_orphan_add(trans, d_inode(new_dentry)); |
79787eaa JM |
9854 | if (ret) { |
9855 | btrfs_abort_transaction(trans, root, ret); | |
9856 | goto out_fail; | |
9857 | } | |
39279cc3 | 9858 | } |
aec7477b | 9859 | |
4df27c4d YZ |
9860 | ret = btrfs_add_link(trans, new_dir, old_inode, |
9861 | new_dentry->d_name.name, | |
a5719521 | 9862 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
9863 | if (ret) { |
9864 | btrfs_abort_transaction(trans, root, ret); | |
9865 | goto out_fail; | |
9866 | } | |
39279cc3 | 9867 | |
67de1176 MX |
9868 | if (old_inode->i_nlink == 1) |
9869 | BTRFS_I(old_inode)->dir_index = index; | |
9870 | ||
3dc9e8f7 | 9871 | if (log_pinned) { |
10d9f309 | 9872 | struct dentry *parent = new_dentry->d_parent; |
3dc9e8f7 | 9873 | |
6a912213 | 9874 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d | 9875 | btrfs_end_log_trans(root); |
3dc9e8f7 | 9876 | log_pinned = false; |
4df27c4d | 9877 | } |
cdd1fedf DF |
9878 | |
9879 | if (flags & RENAME_WHITEOUT) { | |
9880 | ret = btrfs_whiteout_for_rename(trans, root, old_dir, | |
9881 | old_dentry); | |
9882 | ||
9883 | if (ret) { | |
9884 | btrfs_abort_transaction(trans, root, ret); | |
9885 | goto out_fail; | |
9886 | } | |
4df27c4d | 9887 | } |
39279cc3 | 9888 | out_fail: |
3dc9e8f7 FM |
9889 | /* |
9890 | * If we have pinned the log and an error happened, we unpin tasks | |
9891 | * trying to sync the log and force them to fallback to a transaction | |
9892 | * commit if the log currently contains any of the inodes involved in | |
9893 | * this rename operation (to ensure we do not persist a log with an | |
9894 | * inconsistent state for any of these inodes or leading to any | |
9895 | * inconsistencies when replayed). If the transaction was aborted, the | |
9896 | * abortion reason is propagated to userspace when attempting to commit | |
9897 | * the transaction. If the log does not contain any of these inodes, we | |
9898 | * allow the tasks to sync it. | |
9899 | */ | |
9900 | if (ret && log_pinned) { | |
9901 | if (btrfs_inode_in_log(old_dir, root->fs_info->generation) || | |
9902 | btrfs_inode_in_log(new_dir, root->fs_info->generation) || | |
9903 | btrfs_inode_in_log(old_inode, root->fs_info->generation) || | |
9904 | (new_inode && | |
9905 | btrfs_inode_in_log(new_inode, root->fs_info->generation))) | |
9906 | btrfs_set_log_full_commit(root->fs_info, trans); | |
9907 | ||
9908 | btrfs_end_log_trans(root); | |
9909 | log_pinned = false; | |
9910 | } | |
7ad85bb7 | 9911 | btrfs_end_transaction(trans, root); |
b44c59a8 | 9912 | out_notrans: |
33345d01 | 9913 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 9914 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 9915 | |
39279cc3 CM |
9916 | return ret; |
9917 | } | |
9918 | ||
80ace85c MS |
9919 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
9920 | struct inode *new_dir, struct dentry *new_dentry, | |
9921 | unsigned int flags) | |
9922 | { | |
cdd1fedf | 9923 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
9924 | return -EINVAL; |
9925 | ||
cdd1fedf DF |
9926 | if (flags & RENAME_EXCHANGE) |
9927 | return btrfs_rename_exchange(old_dir, old_dentry, new_dir, | |
9928 | new_dentry); | |
9929 | ||
9930 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags); | |
80ace85c MS |
9931 | } |
9932 | ||
8ccf6f19 MX |
9933 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
9934 | { | |
9935 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 9936 | struct inode *inode; |
8ccf6f19 MX |
9937 | |
9938 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
9939 | work); | |
9f23e289 | 9940 | inode = delalloc_work->inode; |
30424601 DS |
9941 | filemap_flush(inode->i_mapping); |
9942 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9943 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 9944 | filemap_flush(inode->i_mapping); |
8ccf6f19 MX |
9945 | |
9946 | if (delalloc_work->delay_iput) | |
9f23e289 | 9947 | btrfs_add_delayed_iput(inode); |
8ccf6f19 | 9948 | else |
9f23e289 | 9949 | iput(inode); |
8ccf6f19 MX |
9950 | complete(&delalloc_work->completion); |
9951 | } | |
9952 | ||
9953 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
651d494a | 9954 | int delay_iput) |
8ccf6f19 MX |
9955 | { |
9956 | struct btrfs_delalloc_work *work; | |
9957 | ||
100d5702 | 9958 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
9959 | if (!work) |
9960 | return NULL; | |
9961 | ||
9962 | init_completion(&work->completion); | |
9963 | INIT_LIST_HEAD(&work->list); | |
9964 | work->inode = inode; | |
8ccf6f19 | 9965 | work->delay_iput = delay_iput; |
9e0af237 LB |
9966 | WARN_ON_ONCE(!inode); |
9967 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, | |
9968 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
9969 | |
9970 | return work; | |
9971 | } | |
9972 | ||
9973 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
9974 | { | |
9975 | wait_for_completion(&work->completion); | |
100d5702 | 9976 | kfree(work); |
8ccf6f19 MX |
9977 | } |
9978 | ||
d352ac68 CM |
9979 | /* |
9980 | * some fairly slow code that needs optimization. This walks the list | |
9981 | * of all the inodes with pending delalloc and forces them to disk. | |
9982 | */ | |
6c255e67 MX |
9983 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput, |
9984 | int nr) | |
ea8c2819 | 9985 | { |
ea8c2819 | 9986 | struct btrfs_inode *binode; |
5b21f2ed | 9987 | struct inode *inode; |
8ccf6f19 MX |
9988 | struct btrfs_delalloc_work *work, *next; |
9989 | struct list_head works; | |
1eafa6c7 | 9990 | struct list_head splice; |
8ccf6f19 | 9991 | int ret = 0; |
ea8c2819 | 9992 | |
8ccf6f19 | 9993 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 9994 | INIT_LIST_HEAD(&splice); |
63607cc8 | 9995 | |
573bfb72 | 9996 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
9997 | spin_lock(&root->delalloc_lock); |
9998 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
9999 | while (!list_empty(&splice)) { |
10000 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 10001 | delalloc_inodes); |
1eafa6c7 | 10002 | |
eb73c1b7 MX |
10003 | list_move_tail(&binode->delalloc_inodes, |
10004 | &root->delalloc_inodes); | |
5b21f2ed | 10005 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 10006 | if (!inode) { |
eb73c1b7 | 10007 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 10008 | continue; |
df0af1a5 | 10009 | } |
eb73c1b7 | 10010 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 10011 | |
651d494a | 10012 | work = btrfs_alloc_delalloc_work(inode, delay_iput); |
5d99a998 | 10013 | if (!work) { |
f4ab9ea7 JB |
10014 | if (delay_iput) |
10015 | btrfs_add_delayed_iput(inode); | |
10016 | else | |
10017 | iput(inode); | |
1eafa6c7 | 10018 | ret = -ENOMEM; |
a1ecaabb | 10019 | goto out; |
5b21f2ed | 10020 | } |
1eafa6c7 | 10021 | list_add_tail(&work->list, &works); |
a44903ab QW |
10022 | btrfs_queue_work(root->fs_info->flush_workers, |
10023 | &work->work); | |
6c255e67 MX |
10024 | ret++; |
10025 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 10026 | goto out; |
5b21f2ed | 10027 | cond_resched(); |
eb73c1b7 | 10028 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 10029 | } |
eb73c1b7 | 10030 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 10031 | |
a1ecaabb | 10032 | out: |
eb73c1b7 MX |
10033 | list_for_each_entry_safe(work, next, &works, list) { |
10034 | list_del_init(&work->list); | |
10035 | btrfs_wait_and_free_delalloc_work(work); | |
10036 | } | |
10037 | ||
10038 | if (!list_empty_careful(&splice)) { | |
10039 | spin_lock(&root->delalloc_lock); | |
10040 | list_splice_tail(&splice, &root->delalloc_inodes); | |
10041 | spin_unlock(&root->delalloc_lock); | |
10042 | } | |
573bfb72 | 10043 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
10044 | return ret; |
10045 | } | |
1eafa6c7 | 10046 | |
eb73c1b7 MX |
10047 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
10048 | { | |
10049 | int ret; | |
1eafa6c7 | 10050 | |
2c21b4d7 | 10051 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
eb73c1b7 MX |
10052 | return -EROFS; |
10053 | ||
6c255e67 MX |
10054 | ret = __start_delalloc_inodes(root, delay_iput, -1); |
10055 | if (ret > 0) | |
10056 | ret = 0; | |
eb73c1b7 MX |
10057 | /* |
10058 | * the filemap_flush will queue IO into the worker threads, but | |
8c8bee1d CM |
10059 | * we have to make sure the IO is actually started and that |
10060 | * ordered extents get created before we return | |
10061 | */ | |
10062 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 10063 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 10064 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 10065 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
10066 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
10067 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
10068 | } |
10069 | atomic_dec(&root->fs_info->async_submit_draining); | |
eb73c1b7 MX |
10070 | return ret; |
10071 | } | |
10072 | ||
6c255e67 MX |
10073 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput, |
10074 | int nr) | |
eb73c1b7 MX |
10075 | { |
10076 | struct btrfs_root *root; | |
10077 | struct list_head splice; | |
10078 | int ret; | |
10079 | ||
2c21b4d7 | 10080 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10081 | return -EROFS; |
10082 | ||
10083 | INIT_LIST_HEAD(&splice); | |
10084 | ||
573bfb72 | 10085 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
10086 | spin_lock(&fs_info->delalloc_root_lock); |
10087 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 10088 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
10089 | root = list_first_entry(&splice, struct btrfs_root, |
10090 | delalloc_root); | |
10091 | root = btrfs_grab_fs_root(root); | |
10092 | BUG_ON(!root); | |
10093 | list_move_tail(&root->delalloc_root, | |
10094 | &fs_info->delalloc_roots); | |
10095 | spin_unlock(&fs_info->delalloc_root_lock); | |
10096 | ||
6c255e67 | 10097 | ret = __start_delalloc_inodes(root, delay_iput, nr); |
eb73c1b7 | 10098 | btrfs_put_fs_root(root); |
6c255e67 | 10099 | if (ret < 0) |
eb73c1b7 MX |
10100 | goto out; |
10101 | ||
6c255e67 MX |
10102 | if (nr != -1) { |
10103 | nr -= ret; | |
10104 | WARN_ON(nr < 0); | |
10105 | } | |
eb73c1b7 | 10106 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 10107 | } |
eb73c1b7 | 10108 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 10109 | |
6c255e67 | 10110 | ret = 0; |
eb73c1b7 MX |
10111 | atomic_inc(&fs_info->async_submit_draining); |
10112 | while (atomic_read(&fs_info->nr_async_submits) || | |
10113 | atomic_read(&fs_info->async_delalloc_pages)) { | |
10114 | wait_event(fs_info->async_submit_wait, | |
10115 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
10116 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
10117 | } | |
10118 | atomic_dec(&fs_info->async_submit_draining); | |
eb73c1b7 | 10119 | out: |
1eafa6c7 | 10120 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
10121 | spin_lock(&fs_info->delalloc_root_lock); |
10122 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
10123 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 10124 | } |
573bfb72 | 10125 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 10126 | return ret; |
ea8c2819 CM |
10127 | } |
10128 | ||
39279cc3 CM |
10129 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
10130 | const char *symname) | |
10131 | { | |
10132 | struct btrfs_trans_handle *trans; | |
10133 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10134 | struct btrfs_path *path; | |
10135 | struct btrfs_key key; | |
1832a6d5 | 10136 | struct inode *inode = NULL; |
39279cc3 CM |
10137 | int err; |
10138 | int drop_inode = 0; | |
10139 | u64 objectid; | |
67871254 | 10140 | u64 index = 0; |
39279cc3 CM |
10141 | int name_len; |
10142 | int datasize; | |
5f39d397 | 10143 | unsigned long ptr; |
39279cc3 | 10144 | struct btrfs_file_extent_item *ei; |
5f39d397 | 10145 | struct extent_buffer *leaf; |
39279cc3 | 10146 | |
f06becc4 | 10147 | name_len = strlen(symname); |
39279cc3 CM |
10148 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) |
10149 | return -ENAMETOOLONG; | |
1832a6d5 | 10150 | |
9ed74f2d JB |
10151 | /* |
10152 | * 2 items for inode item and ref | |
10153 | * 2 items for dir items | |
9269d12b FM |
10154 | * 1 item for updating parent inode item |
10155 | * 1 item for the inline extent item | |
9ed74f2d JB |
10156 | * 1 item for xattr if selinux is on |
10157 | */ | |
9269d12b | 10158 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
10159 | if (IS_ERR(trans)) |
10160 | return PTR_ERR(trans); | |
1832a6d5 | 10161 | |
581bb050 LZ |
10162 | err = btrfs_find_free_ino(root, &objectid); |
10163 | if (err) | |
10164 | goto out_unlock; | |
10165 | ||
aec7477b | 10166 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 10167 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 10168 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
10169 | if (IS_ERR(inode)) { |
10170 | err = PTR_ERR(inode); | |
39279cc3 | 10171 | goto out_unlock; |
7cf96da3 | 10172 | } |
39279cc3 | 10173 | |
ad19db71 CS |
10174 | /* |
10175 | * If the active LSM wants to access the inode during | |
10176 | * d_instantiate it needs these. Smack checks to see | |
10177 | * if the filesystem supports xattrs by looking at the | |
10178 | * ops vector. | |
10179 | */ | |
10180 | inode->i_fop = &btrfs_file_operations; | |
10181 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 10182 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
10183 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10184 | ||
10185 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
10186 | if (err) | |
10187 | goto out_unlock_inode; | |
ad19db71 | 10188 | |
39279cc3 | 10189 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10190 | if (!path) { |
10191 | err = -ENOMEM; | |
b0d5d10f | 10192 | goto out_unlock_inode; |
d8926bb3 | 10193 | } |
33345d01 | 10194 | key.objectid = btrfs_ino(inode); |
39279cc3 | 10195 | key.offset = 0; |
962a298f | 10196 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
10197 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
10198 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
10199 | datasize); | |
54aa1f4d | 10200 | if (err) { |
b0839166 | 10201 | btrfs_free_path(path); |
b0d5d10f | 10202 | goto out_unlock_inode; |
54aa1f4d | 10203 | } |
5f39d397 CM |
10204 | leaf = path->nodes[0]; |
10205 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
10206 | struct btrfs_file_extent_item); | |
10207 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
10208 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 10209 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
10210 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
10211 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
10212 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
10213 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
10214 | ||
39279cc3 | 10215 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
10216 | write_extent_buffer(leaf, symname, ptr, name_len); |
10217 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 10218 | btrfs_free_path(path); |
5f39d397 | 10219 | |
39279cc3 | 10220 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 10221 | inode_nohighmem(inode); |
39279cc3 | 10222 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
d899e052 | 10223 | inode_set_bytes(inode, name_len); |
f06becc4 | 10224 | btrfs_i_size_write(inode, name_len); |
54aa1f4d | 10225 | err = btrfs_update_inode(trans, root, inode); |
d50866d0 FM |
10226 | /* |
10227 | * Last step, add directory indexes for our symlink inode. This is the | |
10228 | * last step to avoid extra cleanup of these indexes if an error happens | |
10229 | * elsewhere above. | |
10230 | */ | |
10231 | if (!err) | |
10232 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); | |
b0d5d10f | 10233 | if (err) { |
54aa1f4d | 10234 | drop_inode = 1; |
b0d5d10f CM |
10235 | goto out_unlock_inode; |
10236 | } | |
10237 | ||
10238 | unlock_new_inode(inode); | |
10239 | d_instantiate(dentry, inode); | |
39279cc3 CM |
10240 | |
10241 | out_unlock: | |
7ad85bb7 | 10242 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
10243 | if (drop_inode) { |
10244 | inode_dec_link_count(inode); | |
10245 | iput(inode); | |
10246 | } | |
b53d3f5d | 10247 | btrfs_btree_balance_dirty(root); |
39279cc3 | 10248 | return err; |
b0d5d10f CM |
10249 | |
10250 | out_unlock_inode: | |
10251 | drop_inode = 1; | |
10252 | unlock_new_inode(inode); | |
10253 | goto out_unlock; | |
39279cc3 | 10254 | } |
16432985 | 10255 | |
0af3d00b JB |
10256 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
10257 | u64 start, u64 num_bytes, u64 min_size, | |
10258 | loff_t actual_len, u64 *alloc_hint, | |
10259 | struct btrfs_trans_handle *trans) | |
d899e052 | 10260 | { |
5dc562c5 JB |
10261 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
10262 | struct extent_map *em; | |
d899e052 YZ |
10263 | struct btrfs_root *root = BTRFS_I(inode)->root; |
10264 | struct btrfs_key ins; | |
d899e052 | 10265 | u64 cur_offset = start; |
55a61d1d | 10266 | u64 i_size; |
154ea289 | 10267 | u64 cur_bytes; |
0b670dc4 | 10268 | u64 last_alloc = (u64)-1; |
d899e052 | 10269 | int ret = 0; |
0af3d00b | 10270 | bool own_trans = true; |
d899e052 | 10271 | |
0af3d00b JB |
10272 | if (trans) |
10273 | own_trans = false; | |
d899e052 | 10274 | while (num_bytes > 0) { |
0af3d00b JB |
10275 | if (own_trans) { |
10276 | trans = btrfs_start_transaction(root, 3); | |
10277 | if (IS_ERR(trans)) { | |
10278 | ret = PTR_ERR(trans); | |
10279 | break; | |
10280 | } | |
5a303d5d YZ |
10281 | } |
10282 | ||
ee22184b | 10283 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 10284 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
10285 | /* |
10286 | * If we are severely fragmented we could end up with really | |
10287 | * small allocations, so if the allocator is returning small | |
10288 | * chunks lets make its job easier by only searching for those | |
10289 | * sized chunks. | |
10290 | */ | |
10291 | cur_bytes = min(cur_bytes, last_alloc); | |
00361589 | 10292 | ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0, |
e570fd27 | 10293 | *alloc_hint, &ins, 1, 0); |
5a303d5d | 10294 | if (ret) { |
0af3d00b JB |
10295 | if (own_trans) |
10296 | btrfs_end_transaction(trans, root); | |
a22285a6 | 10297 | break; |
d899e052 | 10298 | } |
9cfa3e34 | 10299 | btrfs_dec_block_group_reservations(root->fs_info, ins.objectid); |
5a303d5d | 10300 | |
0b670dc4 | 10301 | last_alloc = ins.offset; |
d899e052 YZ |
10302 | ret = insert_reserved_file_extent(trans, inode, |
10303 | cur_offset, ins.objectid, | |
10304 | ins.offset, ins.offset, | |
920bbbfb | 10305 | ins.offset, 0, 0, 0, |
d899e052 | 10306 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 10307 | if (ret) { |
857cc2fc | 10308 | btrfs_free_reserved_extent(root, ins.objectid, |
e570fd27 | 10309 | ins.offset, 0); |
79787eaa JM |
10310 | btrfs_abort_transaction(trans, root, ret); |
10311 | if (own_trans) | |
10312 | btrfs_end_transaction(trans, root); | |
10313 | break; | |
10314 | } | |
31193213 | 10315 | |
a1ed835e CM |
10316 | btrfs_drop_extent_cache(inode, cur_offset, |
10317 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 10318 | |
5dc562c5 JB |
10319 | em = alloc_extent_map(); |
10320 | if (!em) { | |
10321 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
10322 | &BTRFS_I(inode)->runtime_flags); | |
10323 | goto next; | |
10324 | } | |
10325 | ||
10326 | em->start = cur_offset; | |
10327 | em->orig_start = cur_offset; | |
10328 | em->len = ins.offset; | |
10329 | em->block_start = ins.objectid; | |
10330 | em->block_len = ins.offset; | |
b4939680 | 10331 | em->orig_block_len = ins.offset; |
cc95bef6 | 10332 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
10333 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
10334 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
10335 | em->generation = trans->transid; | |
10336 | ||
10337 | while (1) { | |
10338 | write_lock(&em_tree->lock); | |
09a2a8f9 | 10339 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
10340 | write_unlock(&em_tree->lock); |
10341 | if (ret != -EEXIST) | |
10342 | break; | |
10343 | btrfs_drop_extent_cache(inode, cur_offset, | |
10344 | cur_offset + ins.offset - 1, | |
10345 | 0); | |
10346 | } | |
10347 | free_extent_map(em); | |
10348 | next: | |
d899e052 YZ |
10349 | num_bytes -= ins.offset; |
10350 | cur_offset += ins.offset; | |
efa56464 | 10351 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 10352 | |
0c4d2d95 | 10353 | inode_inc_iversion(inode); |
04b285f3 | 10354 | inode->i_ctime = current_fs_time(inode->i_sb); |
6cbff00f | 10355 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 10356 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
10357 | (actual_len > inode->i_size) && |
10358 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 10359 | if (cur_offset > actual_len) |
55a61d1d | 10360 | i_size = actual_len; |
d1ea6a61 | 10361 | else |
55a61d1d JB |
10362 | i_size = cur_offset; |
10363 | i_size_write(inode, i_size); | |
10364 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
10365 | } |
10366 | ||
d899e052 | 10367 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
10368 | |
10369 | if (ret) { | |
10370 | btrfs_abort_transaction(trans, root, ret); | |
10371 | if (own_trans) | |
10372 | btrfs_end_transaction(trans, root); | |
10373 | break; | |
10374 | } | |
d899e052 | 10375 | |
0af3d00b JB |
10376 | if (own_trans) |
10377 | btrfs_end_transaction(trans, root); | |
5a303d5d | 10378 | } |
d899e052 YZ |
10379 | return ret; |
10380 | } | |
10381 | ||
0af3d00b JB |
10382 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
10383 | u64 start, u64 num_bytes, u64 min_size, | |
10384 | loff_t actual_len, u64 *alloc_hint) | |
10385 | { | |
10386 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10387 | min_size, actual_len, alloc_hint, | |
10388 | NULL); | |
10389 | } | |
10390 | ||
10391 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
10392 | struct btrfs_trans_handle *trans, int mode, | |
10393 | u64 start, u64 num_bytes, u64 min_size, | |
10394 | loff_t actual_len, u64 *alloc_hint) | |
10395 | { | |
10396 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10397 | min_size, actual_len, alloc_hint, trans); | |
10398 | } | |
10399 | ||
e6dcd2dc CM |
10400 | static int btrfs_set_page_dirty(struct page *page) |
10401 | { | |
e6dcd2dc CM |
10402 | return __set_page_dirty_nobuffers(page); |
10403 | } | |
10404 | ||
10556cb2 | 10405 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 10406 | { |
b83cc969 | 10407 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 10408 | umode_t mode = inode->i_mode; |
b83cc969 | 10409 | |
cb6db4e5 JM |
10410 | if (mask & MAY_WRITE && |
10411 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
10412 | if (btrfs_root_readonly(root)) | |
10413 | return -EROFS; | |
10414 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
10415 | return -EACCES; | |
10416 | } | |
2830ba7f | 10417 | return generic_permission(inode, mask); |
fdebe2bd | 10418 | } |
39279cc3 | 10419 | |
ef3b9af5 FM |
10420 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
10421 | { | |
10422 | struct btrfs_trans_handle *trans; | |
10423 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10424 | struct inode *inode = NULL; | |
10425 | u64 objectid; | |
10426 | u64 index; | |
10427 | int ret = 0; | |
10428 | ||
10429 | /* | |
10430 | * 5 units required for adding orphan entry | |
10431 | */ | |
10432 | trans = btrfs_start_transaction(root, 5); | |
10433 | if (IS_ERR(trans)) | |
10434 | return PTR_ERR(trans); | |
10435 | ||
10436 | ret = btrfs_find_free_ino(root, &objectid); | |
10437 | if (ret) | |
10438 | goto out; | |
10439 | ||
10440 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
10441 | btrfs_ino(dir), objectid, mode, &index); | |
10442 | if (IS_ERR(inode)) { | |
10443 | ret = PTR_ERR(inode); | |
10444 | inode = NULL; | |
10445 | goto out; | |
10446 | } | |
10447 | ||
ef3b9af5 FM |
10448 | inode->i_fop = &btrfs_file_operations; |
10449 | inode->i_op = &btrfs_file_inode_operations; | |
10450 | ||
10451 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 FM |
10452 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10453 | ||
b0d5d10f CM |
10454 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
10455 | if (ret) | |
10456 | goto out_inode; | |
10457 | ||
10458 | ret = btrfs_update_inode(trans, root, inode); | |
10459 | if (ret) | |
10460 | goto out_inode; | |
ef3b9af5 FM |
10461 | ret = btrfs_orphan_add(trans, inode); |
10462 | if (ret) | |
b0d5d10f | 10463 | goto out_inode; |
ef3b9af5 | 10464 | |
5762b5c9 FM |
10465 | /* |
10466 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
10467 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
10468 | * through: | |
10469 | * | |
10470 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
10471 | */ | |
10472 | set_nlink(inode, 1); | |
b0d5d10f | 10473 | unlock_new_inode(inode); |
ef3b9af5 FM |
10474 | d_tmpfile(dentry, inode); |
10475 | mark_inode_dirty(inode); | |
10476 | ||
10477 | out: | |
10478 | btrfs_end_transaction(trans, root); | |
10479 | if (ret) | |
10480 | iput(inode); | |
10481 | btrfs_balance_delayed_items(root); | |
10482 | btrfs_btree_balance_dirty(root); | |
ef3b9af5 | 10483 | return ret; |
b0d5d10f CM |
10484 | |
10485 | out_inode: | |
10486 | unlock_new_inode(inode); | |
10487 | goto out; | |
10488 | ||
ef3b9af5 FM |
10489 | } |
10490 | ||
b38ef71c FM |
10491 | /* Inspired by filemap_check_errors() */ |
10492 | int btrfs_inode_check_errors(struct inode *inode) | |
10493 | { | |
10494 | int ret = 0; | |
10495 | ||
10496 | if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) && | |
10497 | test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags)) | |
10498 | ret = -ENOSPC; | |
10499 | if (test_bit(AS_EIO, &inode->i_mapping->flags) && | |
10500 | test_and_clear_bit(AS_EIO, &inode->i_mapping->flags)) | |
10501 | ret = -EIO; | |
10502 | ||
10503 | return ret; | |
10504 | } | |
10505 | ||
6e1d5dcc | 10506 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10507 | .getattr = btrfs_getattr, |
39279cc3 CM |
10508 | .lookup = btrfs_lookup, |
10509 | .create = btrfs_create, | |
10510 | .unlink = btrfs_unlink, | |
10511 | .link = btrfs_link, | |
10512 | .mkdir = btrfs_mkdir, | |
10513 | .rmdir = btrfs_rmdir, | |
80ace85c | 10514 | .rename2 = btrfs_rename2, |
39279cc3 CM |
10515 | .symlink = btrfs_symlink, |
10516 | .setattr = btrfs_setattr, | |
618e21d5 | 10517 | .mknod = btrfs_mknod, |
e0d46f5c | 10518 | .setxattr = generic_setxattr, |
9172abbc | 10519 | .getxattr = generic_getxattr, |
5103e947 | 10520 | .listxattr = btrfs_listxattr, |
e0d46f5c | 10521 | .removexattr = generic_removexattr, |
fdebe2bd | 10522 | .permission = btrfs_permission, |
4e34e719 | 10523 | .get_acl = btrfs_get_acl, |
996a710d | 10524 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10525 | .update_time = btrfs_update_time, |
ef3b9af5 | 10526 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 10527 | }; |
6e1d5dcc | 10528 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 10529 | .lookup = btrfs_lookup, |
fdebe2bd | 10530 | .permission = btrfs_permission, |
4e34e719 | 10531 | .get_acl = btrfs_get_acl, |
996a710d | 10532 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10533 | .update_time = btrfs_update_time, |
39279cc3 | 10534 | }; |
76dda93c | 10535 | |
828c0950 | 10536 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10537 | .llseek = generic_file_llseek, |
10538 | .read = generic_read_dir, | |
02dbfc99 | 10539 | .iterate_shared = btrfs_real_readdir, |
34287aa3 | 10540 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10541 | #ifdef CONFIG_COMPAT |
4c63c245 | 10542 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 10543 | #endif |
6bf13c0c | 10544 | .release = btrfs_release_file, |
e02119d5 | 10545 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10546 | }; |
10547 | ||
20e5506b | 10548 | static const struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 10549 | .fill_delalloc = run_delalloc_range, |
065631f6 | 10550 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 10551 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 10552 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 10553 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 10554 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
10555 | .set_bit_hook = btrfs_set_bit_hook, |
10556 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
10557 | .merge_extent_hook = btrfs_merge_extent_hook, |
10558 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
10559 | }; |
10560 | ||
35054394 CM |
10561 | /* |
10562 | * btrfs doesn't support the bmap operation because swapfiles | |
10563 | * use bmap to make a mapping of extents in the file. They assume | |
10564 | * these extents won't change over the life of the file and they | |
10565 | * use the bmap result to do IO directly to the drive. | |
10566 | * | |
10567 | * the btrfs bmap call would return logical addresses that aren't | |
10568 | * suitable for IO and they also will change frequently as COW | |
10569 | * operations happen. So, swapfile + btrfs == corruption. | |
10570 | * | |
10571 | * For now we're avoiding this by dropping bmap. | |
10572 | */ | |
7f09410b | 10573 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
10574 | .readpage = btrfs_readpage, |
10575 | .writepage = btrfs_writepage, | |
b293f02e | 10576 | .writepages = btrfs_writepages, |
3ab2fb5a | 10577 | .readpages = btrfs_readpages, |
16432985 | 10578 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
10579 | .invalidatepage = btrfs_invalidatepage, |
10580 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 10581 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 10582 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
10583 | }; |
10584 | ||
7f09410b | 10585 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
10586 | .readpage = btrfs_readpage, |
10587 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
10588 | .invalidatepage = btrfs_invalidatepage, |
10589 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
10590 | }; |
10591 | ||
6e1d5dcc | 10592 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10593 | .getattr = btrfs_getattr, |
10594 | .setattr = btrfs_setattr, | |
e0d46f5c | 10595 | .setxattr = generic_setxattr, |
9172abbc | 10596 | .getxattr = generic_getxattr, |
5103e947 | 10597 | .listxattr = btrfs_listxattr, |
e0d46f5c | 10598 | .removexattr = generic_removexattr, |
fdebe2bd | 10599 | .permission = btrfs_permission, |
1506fcc8 | 10600 | .fiemap = btrfs_fiemap, |
4e34e719 | 10601 | .get_acl = btrfs_get_acl, |
996a710d | 10602 | .set_acl = btrfs_set_acl, |
e41f941a | 10603 | .update_time = btrfs_update_time, |
39279cc3 | 10604 | }; |
6e1d5dcc | 10605 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10606 | .getattr = btrfs_getattr, |
10607 | .setattr = btrfs_setattr, | |
fdebe2bd | 10608 | .permission = btrfs_permission, |
e0d46f5c | 10609 | .setxattr = generic_setxattr, |
9172abbc | 10610 | .getxattr = generic_getxattr, |
33268eaf | 10611 | .listxattr = btrfs_listxattr, |
e0d46f5c | 10612 | .removexattr = generic_removexattr, |
4e34e719 | 10613 | .get_acl = btrfs_get_acl, |
996a710d | 10614 | .set_acl = btrfs_set_acl, |
e41f941a | 10615 | .update_time = btrfs_update_time, |
618e21d5 | 10616 | }; |
6e1d5dcc | 10617 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 | 10618 | .readlink = generic_readlink, |
6b255391 | 10619 | .get_link = page_get_link, |
f209561a | 10620 | .getattr = btrfs_getattr, |
22c44fe6 | 10621 | .setattr = btrfs_setattr, |
fdebe2bd | 10622 | .permission = btrfs_permission, |
e0d46f5c | 10623 | .setxattr = generic_setxattr, |
9172abbc | 10624 | .getxattr = generic_getxattr, |
0279b4cd | 10625 | .listxattr = btrfs_listxattr, |
e0d46f5c | 10626 | .removexattr = generic_removexattr, |
e41f941a | 10627 | .update_time = btrfs_update_time, |
39279cc3 | 10628 | }; |
76dda93c | 10629 | |
82d339d9 | 10630 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 10631 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 10632 | .d_release = btrfs_dentry_release, |
76dda93c | 10633 | }; |