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 | |
458 | * isn't an inline extent, since it dosen't save disk space at all. | |
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 | } |
771ed689 | 827 | |
771ed689 CM |
828 | /* |
829 | * clear dirty, set writeback and unlock the pages. | |
830 | */ | |
c2790a2e | 831 | extent_clear_unlock_delalloc(inode, async_extent->start, |
a791e35e CM |
832 | async_extent->start + |
833 | async_extent->ram_size - 1, | |
151a41bc JB |
834 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
835 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 836 | PAGE_SET_WRITEBACK); |
771ed689 | 837 | ret = btrfs_submit_compressed_write(inode, |
d397712b CM |
838 | async_extent->start, |
839 | async_extent->ram_size, | |
840 | ins.objectid, | |
841 | ins.offset, async_extent->pages, | |
842 | async_extent->nr_pages); | |
fce2a4e6 FM |
843 | if (ret) { |
844 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
845 | struct page *p = async_extent->pages[0]; | |
846 | const u64 start = async_extent->start; | |
847 | const u64 end = start + async_extent->ram_size - 1; | |
848 | ||
849 | p->mapping = inode->i_mapping; | |
850 | tree->ops->writepage_end_io_hook(p, start, end, | |
851 | NULL, 0); | |
852 | p->mapping = NULL; | |
853 | extent_clear_unlock_delalloc(inode, start, end, NULL, 0, | |
854 | PAGE_END_WRITEBACK | | |
855 | PAGE_SET_ERROR); | |
40ae837b | 856 | free_async_extent_pages(async_extent); |
fce2a4e6 | 857 | } |
771ed689 CM |
858 | alloc_hint = ins.objectid + ins.offset; |
859 | kfree(async_extent); | |
860 | cond_resched(); | |
861 | } | |
dec8f175 | 862 | return; |
3e04e7f1 | 863 | out_free_reserve: |
e570fd27 | 864 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
79787eaa | 865 | out_free: |
c2790a2e | 866 | extent_clear_unlock_delalloc(inode, async_extent->start, |
3e04e7f1 JB |
867 | async_extent->start + |
868 | async_extent->ram_size - 1, | |
c2790a2e | 869 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
870 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
871 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
872 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
873 | PAGE_SET_ERROR); | |
40ae837b | 874 | free_async_extent_pages(async_extent); |
79787eaa | 875 | kfree(async_extent); |
3e04e7f1 | 876 | goto again; |
771ed689 CM |
877 | } |
878 | ||
4b46fce2 JB |
879 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
880 | u64 num_bytes) | |
881 | { | |
882 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
883 | struct extent_map *em; | |
884 | u64 alloc_hint = 0; | |
885 | ||
886 | read_lock(&em_tree->lock); | |
887 | em = search_extent_mapping(em_tree, start, num_bytes); | |
888 | if (em) { | |
889 | /* | |
890 | * if block start isn't an actual block number then find the | |
891 | * first block in this inode and use that as a hint. If that | |
892 | * block is also bogus then just don't worry about it. | |
893 | */ | |
894 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
895 | free_extent_map(em); | |
896 | em = search_extent_mapping(em_tree, 0, 0); | |
897 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
898 | alloc_hint = em->block_start; | |
899 | if (em) | |
900 | free_extent_map(em); | |
901 | } else { | |
902 | alloc_hint = em->block_start; | |
903 | free_extent_map(em); | |
904 | } | |
905 | } | |
906 | read_unlock(&em_tree->lock); | |
907 | ||
908 | return alloc_hint; | |
909 | } | |
910 | ||
771ed689 CM |
911 | /* |
912 | * when extent_io.c finds a delayed allocation range in the file, | |
913 | * the call backs end up in this code. The basic idea is to | |
914 | * allocate extents on disk for the range, and create ordered data structs | |
915 | * in ram to track those extents. | |
916 | * | |
917 | * locked_page is the page that writepage had locked already. We use | |
918 | * it to make sure we don't do extra locks or unlocks. | |
919 | * | |
920 | * *page_started is set to one if we unlock locked_page and do everything | |
921 | * required to start IO on it. It may be clean and already done with | |
922 | * IO when we return. | |
923 | */ | |
00361589 JB |
924 | static noinline int cow_file_range(struct inode *inode, |
925 | struct page *locked_page, | |
926 | u64 start, u64 end, int *page_started, | |
927 | unsigned long *nr_written, | |
928 | int unlock) | |
771ed689 | 929 | { |
00361589 | 930 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
931 | u64 alloc_hint = 0; |
932 | u64 num_bytes; | |
933 | unsigned long ram_size; | |
934 | u64 disk_num_bytes; | |
935 | u64 cur_alloc_size; | |
936 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
937 | struct btrfs_key ins; |
938 | struct extent_map *em; | |
939 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
940 | int ret = 0; | |
941 | ||
02ecd2c2 JB |
942 | if (btrfs_is_free_space_inode(inode)) { |
943 | WARN_ON_ONCE(1); | |
29bce2f3 JB |
944 | ret = -EINVAL; |
945 | goto out_unlock; | |
02ecd2c2 | 946 | } |
771ed689 | 947 | |
fda2832f | 948 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 CM |
949 | num_bytes = max(blocksize, num_bytes); |
950 | disk_num_bytes = num_bytes; | |
771ed689 | 951 | |
4cb5300b | 952 | /* if this is a small write inside eof, kick off defrag */ |
ee22184b | 953 | if (num_bytes < SZ_64K && |
4cb13e5d | 954 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) |
00361589 | 955 | btrfs_add_inode_defrag(NULL, inode); |
4cb5300b | 956 | |
771ed689 CM |
957 | if (start == 0) { |
958 | /* lets try to make an inline extent */ | |
00361589 JB |
959 | ret = cow_file_range_inline(root, inode, start, end, 0, 0, |
960 | NULL); | |
771ed689 | 961 | if (ret == 0) { |
c2790a2e JB |
962 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
963 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc | 964 | EXTENT_DEFRAG, PAGE_UNLOCK | |
c2790a2e JB |
965 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
966 | PAGE_END_WRITEBACK); | |
c2167754 | 967 | |
771ed689 | 968 | *nr_written = *nr_written + |
09cbfeaf | 969 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 970 | *page_started = 1; |
771ed689 | 971 | goto out; |
79787eaa | 972 | } else if (ret < 0) { |
79787eaa | 973 | goto out_unlock; |
771ed689 CM |
974 | } |
975 | } | |
976 | ||
977 | BUG_ON(disk_num_bytes > | |
6c41761f | 978 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 979 | |
4b46fce2 | 980 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
981 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
982 | ||
d397712b | 983 | while (disk_num_bytes > 0) { |
a791e35e CM |
984 | unsigned long op; |
985 | ||
287a0ab9 | 986 | cur_alloc_size = disk_num_bytes; |
00361589 | 987 | ret = btrfs_reserve_extent(root, cur_alloc_size, |
771ed689 | 988 | root->sectorsize, 0, alloc_hint, |
e570fd27 | 989 | &ins, 1, 1); |
00361589 | 990 | if (ret < 0) |
79787eaa | 991 | goto out_unlock; |
d397712b | 992 | |
172ddd60 | 993 | em = alloc_extent_map(); |
b9aa55be LB |
994 | if (!em) { |
995 | ret = -ENOMEM; | |
ace68bac | 996 | goto out_reserve; |
b9aa55be | 997 | } |
e6dcd2dc | 998 | em->start = start; |
445a6944 | 999 | em->orig_start = em->start; |
771ed689 CM |
1000 | ram_size = ins.offset; |
1001 | em->len = ins.offset; | |
2ab28f32 JB |
1002 | em->mod_start = em->start; |
1003 | em->mod_len = em->len; | |
c8b97818 | 1004 | |
e6dcd2dc | 1005 | em->block_start = ins.objectid; |
c8b97818 | 1006 | em->block_len = ins.offset; |
b4939680 | 1007 | em->orig_block_len = ins.offset; |
cc95bef6 | 1008 | em->ram_bytes = ram_size; |
e6dcd2dc | 1009 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 1010 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
70c8a91c | 1011 | em->generation = -1; |
c8b97818 | 1012 | |
d397712b | 1013 | while (1) { |
890871be | 1014 | write_lock(&em_tree->lock); |
09a2a8f9 | 1015 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1016 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
1017 | if (ret != -EEXIST) { |
1018 | free_extent_map(em); | |
1019 | break; | |
1020 | } | |
1021 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 1022 | start + ram_size - 1, 0); |
e6dcd2dc | 1023 | } |
ace68bac LB |
1024 | if (ret) |
1025 | goto out_reserve; | |
e6dcd2dc | 1026 | |
98d20f67 | 1027 | cur_alloc_size = ins.offset; |
e6dcd2dc | 1028 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 1029 | ram_size, cur_alloc_size, 0); |
ace68bac | 1030 | if (ret) |
d9f85963 | 1031 | goto out_drop_extent_cache; |
c8b97818 | 1032 | |
17d217fe YZ |
1033 | if (root->root_key.objectid == |
1034 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1035 | ret = btrfs_reloc_clone_csums(inode, start, | |
1036 | cur_alloc_size); | |
00361589 | 1037 | if (ret) |
d9f85963 | 1038 | goto out_drop_extent_cache; |
17d217fe YZ |
1039 | } |
1040 | ||
d397712b | 1041 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 1042 | break; |
d397712b | 1043 | |
c8b97818 CM |
1044 | /* we're not doing compressed IO, don't unlock the first |
1045 | * page (which the caller expects to stay locked), don't | |
1046 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1047 | * |
1048 | * Do set the Private2 bit so we know this page was properly | |
1049 | * setup for writepage | |
c8b97818 | 1050 | */ |
c2790a2e JB |
1051 | op = unlock ? PAGE_UNLOCK : 0; |
1052 | op |= PAGE_SET_PRIVATE2; | |
a791e35e | 1053 | |
c2790a2e JB |
1054 | extent_clear_unlock_delalloc(inode, start, |
1055 | start + ram_size - 1, locked_page, | |
1056 | EXTENT_LOCKED | EXTENT_DELALLOC, | |
1057 | op); | |
c8b97818 | 1058 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
1059 | num_bytes -= cur_alloc_size; |
1060 | alloc_hint = ins.objectid + ins.offset; | |
1061 | start += cur_alloc_size; | |
b888db2b | 1062 | } |
79787eaa | 1063 | out: |
be20aa9d | 1064 | return ret; |
b7d5b0a8 | 1065 | |
d9f85963 FM |
1066 | out_drop_extent_cache: |
1067 | btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0); | |
ace68bac | 1068 | out_reserve: |
e570fd27 | 1069 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
79787eaa | 1070 | out_unlock: |
c2790a2e | 1071 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
151a41bc JB |
1072 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
1073 | EXTENT_DELALLOC | EXTENT_DEFRAG, | |
1074 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
1075 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 1076 | goto out; |
771ed689 | 1077 | } |
c8b97818 | 1078 | |
771ed689 CM |
1079 | /* |
1080 | * work queue call back to started compression on a file and pages | |
1081 | */ | |
1082 | static noinline void async_cow_start(struct btrfs_work *work) | |
1083 | { | |
1084 | struct async_cow *async_cow; | |
1085 | int num_added = 0; | |
1086 | async_cow = container_of(work, struct async_cow, work); | |
1087 | ||
1088 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1089 | async_cow->start, async_cow->end, async_cow, | |
1090 | &num_added); | |
8180ef88 | 1091 | if (num_added == 0) { |
cb77fcd8 | 1092 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1093 | async_cow->inode = NULL; |
8180ef88 | 1094 | } |
771ed689 CM |
1095 | } |
1096 | ||
1097 | /* | |
1098 | * work queue call back to submit previously compressed pages | |
1099 | */ | |
1100 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1101 | { | |
1102 | struct async_cow *async_cow; | |
1103 | struct btrfs_root *root; | |
1104 | unsigned long nr_pages; | |
1105 | ||
1106 | async_cow = container_of(work, struct async_cow, work); | |
1107 | ||
1108 | root = async_cow->root; | |
09cbfeaf KS |
1109 | nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >> |
1110 | PAGE_SHIFT; | |
771ed689 | 1111 | |
ee863954 DS |
1112 | /* |
1113 | * atomic_sub_return implies a barrier for waitqueue_active | |
1114 | */ | |
66657b31 | 1115 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
ee22184b | 1116 | 5 * SZ_1M && |
771ed689 CM |
1117 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1118 | wake_up(&root->fs_info->async_submit_wait); | |
1119 | ||
d397712b | 1120 | if (async_cow->inode) |
771ed689 | 1121 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1122 | } |
c8b97818 | 1123 | |
771ed689 CM |
1124 | static noinline void async_cow_free(struct btrfs_work *work) |
1125 | { | |
1126 | struct async_cow *async_cow; | |
1127 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1128 | if (async_cow->inode) |
cb77fcd8 | 1129 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1130 | kfree(async_cow); |
1131 | } | |
1132 | ||
1133 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1134 | u64 start, u64 end, int *page_started, | |
1135 | unsigned long *nr_written) | |
1136 | { | |
1137 | struct async_cow *async_cow; | |
1138 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1139 | unsigned long nr_pages; | |
1140 | u64 cur_end; | |
ee22184b | 1141 | int limit = 10 * SZ_1M; |
771ed689 | 1142 | |
a3429ab7 CM |
1143 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1144 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1145 | while (start < end) { |
771ed689 | 1146 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1147 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1148 | async_cow->inode = igrab(inode); |
771ed689 CM |
1149 | async_cow->root = root; |
1150 | async_cow->locked_page = locked_page; | |
1151 | async_cow->start = start; | |
1152 | ||
f79707b0 WS |
1153 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS && |
1154 | !btrfs_test_opt(root, FORCE_COMPRESS)) | |
771ed689 CM |
1155 | cur_end = end; |
1156 | else | |
ee22184b | 1157 | cur_end = min(end, start + SZ_512K - 1); |
771ed689 CM |
1158 | |
1159 | async_cow->end = cur_end; | |
1160 | INIT_LIST_HEAD(&async_cow->extents); | |
1161 | ||
9e0af237 LB |
1162 | btrfs_init_work(&async_cow->work, |
1163 | btrfs_delalloc_helper, | |
1164 | async_cow_start, async_cow_submit, | |
1165 | async_cow_free); | |
771ed689 | 1166 | |
09cbfeaf KS |
1167 | nr_pages = (cur_end - start + PAGE_SIZE) >> |
1168 | PAGE_SHIFT; | |
771ed689 CM |
1169 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); |
1170 | ||
afe3d242 QW |
1171 | btrfs_queue_work(root->fs_info->delalloc_workers, |
1172 | &async_cow->work); | |
771ed689 CM |
1173 | |
1174 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1175 | wait_event(root->fs_info->async_submit_wait, | |
1176 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1177 | limit)); | |
1178 | } | |
1179 | ||
d397712b | 1180 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1181 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1182 | wait_event(root->fs_info->async_submit_wait, | |
1183 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1184 | 0)); | |
1185 | } | |
1186 | ||
1187 | *nr_written += nr_pages; | |
1188 | start = cur_end + 1; | |
1189 | } | |
1190 | *page_started = 1; | |
1191 | return 0; | |
be20aa9d CM |
1192 | } |
1193 | ||
d397712b | 1194 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1195 | u64 bytenr, u64 num_bytes) |
1196 | { | |
1197 | int ret; | |
1198 | struct btrfs_ordered_sum *sums; | |
1199 | LIST_HEAD(list); | |
1200 | ||
07d400a6 | 1201 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1202 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1203 | if (ret == 0 && list_empty(&list)) |
1204 | return 0; | |
1205 | ||
1206 | while (!list_empty(&list)) { | |
1207 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1208 | list_del(&sums->list); | |
1209 | kfree(sums); | |
1210 | } | |
1211 | return 1; | |
1212 | } | |
1213 | ||
d352ac68 CM |
1214 | /* |
1215 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1216 | * of the extents that exist in the file, and COWs the file as required. | |
1217 | * | |
1218 | * If no cow copies or snapshots exist, we write directly to the existing | |
1219 | * blocks on disk | |
1220 | */ | |
7f366cfe CM |
1221 | static noinline int run_delalloc_nocow(struct inode *inode, |
1222 | struct page *locked_page, | |
771ed689 CM |
1223 | u64 start, u64 end, int *page_started, int force, |
1224 | unsigned long *nr_written) | |
be20aa9d | 1225 | { |
be20aa9d | 1226 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1227 | struct btrfs_trans_handle *trans; |
be20aa9d | 1228 | struct extent_buffer *leaf; |
be20aa9d | 1229 | struct btrfs_path *path; |
80ff3856 | 1230 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1231 | struct btrfs_key found_key; |
80ff3856 YZ |
1232 | u64 cow_start; |
1233 | u64 cur_offset; | |
1234 | u64 extent_end; | |
5d4f98a2 | 1235 | u64 extent_offset; |
80ff3856 YZ |
1236 | u64 disk_bytenr; |
1237 | u64 num_bytes; | |
b4939680 | 1238 | u64 disk_num_bytes; |
cc95bef6 | 1239 | u64 ram_bytes; |
80ff3856 | 1240 | int extent_type; |
79787eaa | 1241 | int ret, err; |
d899e052 | 1242 | int type; |
80ff3856 YZ |
1243 | int nocow; |
1244 | int check_prev = 1; | |
82d5902d | 1245 | bool nolock; |
33345d01 | 1246 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1247 | |
1248 | path = btrfs_alloc_path(); | |
17ca04af | 1249 | if (!path) { |
c2790a2e JB |
1250 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1251 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1252 | EXTENT_DO_ACCOUNTING | |
1253 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1254 | PAGE_CLEAR_DIRTY | |
1255 | PAGE_SET_WRITEBACK | | |
1256 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1257 | return -ENOMEM; |
17ca04af | 1258 | } |
82d5902d | 1259 | |
83eea1f1 | 1260 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1261 | |
1262 | if (nolock) | |
7a7eaa40 | 1263 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1264 | else |
7a7eaa40 | 1265 | trans = btrfs_join_transaction(root); |
ff5714cc | 1266 | |
79787eaa | 1267 | if (IS_ERR(trans)) { |
c2790a2e JB |
1268 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1269 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1270 | EXTENT_DO_ACCOUNTING | |
1271 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1272 | PAGE_CLEAR_DIRTY | |
1273 | PAGE_SET_WRITEBACK | | |
1274 | PAGE_END_WRITEBACK); | |
79787eaa JM |
1275 | btrfs_free_path(path); |
1276 | return PTR_ERR(trans); | |
1277 | } | |
1278 | ||
74b21075 | 1279 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1280 | |
80ff3856 YZ |
1281 | cow_start = (u64)-1; |
1282 | cur_offset = start; | |
1283 | while (1) { | |
33345d01 | 1284 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1285 | cur_offset, 0); |
d788a349 | 1286 | if (ret < 0) |
79787eaa | 1287 | goto error; |
80ff3856 YZ |
1288 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1289 | leaf = path->nodes[0]; | |
1290 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1291 | path->slots[0] - 1); | |
33345d01 | 1292 | if (found_key.objectid == ino && |
80ff3856 YZ |
1293 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1294 | path->slots[0]--; | |
1295 | } | |
1296 | check_prev = 0; | |
1297 | next_slot: | |
1298 | leaf = path->nodes[0]; | |
1299 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1300 | ret = btrfs_next_leaf(root, path); | |
d788a349 | 1301 | if (ret < 0) |
79787eaa | 1302 | goto error; |
80ff3856 YZ |
1303 | if (ret > 0) |
1304 | break; | |
1305 | leaf = path->nodes[0]; | |
1306 | } | |
be20aa9d | 1307 | |
80ff3856 YZ |
1308 | nocow = 0; |
1309 | disk_bytenr = 0; | |
17d217fe | 1310 | num_bytes = 0; |
80ff3856 YZ |
1311 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1312 | ||
1d512cb7 FM |
1313 | if (found_key.objectid > ino) |
1314 | break; | |
1315 | if (WARN_ON_ONCE(found_key.objectid < ino) || | |
1316 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1317 | path->slots[0]++; | |
1318 | goto next_slot; | |
1319 | } | |
1320 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || | |
80ff3856 YZ |
1321 | found_key.offset > end) |
1322 | break; | |
1323 | ||
1324 | if (found_key.offset > cur_offset) { | |
1325 | extent_end = found_key.offset; | |
e9061e21 | 1326 | extent_type = 0; |
80ff3856 YZ |
1327 | goto out_check; |
1328 | } | |
1329 | ||
1330 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1331 | struct btrfs_file_extent_item); | |
1332 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1333 | ||
cc95bef6 | 1334 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1335 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1336 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1337 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1338 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1339 | extent_end = found_key.offset + |
1340 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1341 | disk_num_bytes = |
1342 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1343 | if (extent_end <= start) { |
1344 | path->slots[0]++; | |
1345 | goto next_slot; | |
1346 | } | |
17d217fe YZ |
1347 | if (disk_bytenr == 0) |
1348 | goto out_check; | |
80ff3856 YZ |
1349 | if (btrfs_file_extent_compression(leaf, fi) || |
1350 | btrfs_file_extent_encryption(leaf, fi) || | |
1351 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1352 | goto out_check; | |
d899e052 YZ |
1353 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1354 | goto out_check; | |
d2fb3437 | 1355 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1356 | goto out_check; |
33345d01 | 1357 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1358 | found_key.offset - |
1359 | extent_offset, disk_bytenr)) | |
17d217fe | 1360 | goto out_check; |
5d4f98a2 | 1361 | disk_bytenr += extent_offset; |
17d217fe YZ |
1362 | disk_bytenr += cur_offset - found_key.offset; |
1363 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 WS |
1364 | /* |
1365 | * if there are pending snapshots for this root, | |
1366 | * we fall into common COW way. | |
1367 | */ | |
1368 | if (!nolock) { | |
9ea24bbe | 1369 | err = btrfs_start_write_no_snapshoting(root); |
e9894fd3 WS |
1370 | if (!err) |
1371 | goto out_check; | |
1372 | } | |
17d217fe YZ |
1373 | /* |
1374 | * force cow if csum exists in the range. | |
1375 | * this ensure that csum for a given extent are | |
1376 | * either valid or do not exist. | |
1377 | */ | |
1378 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1379 | goto out_check; | |
80ff3856 YZ |
1380 | nocow = 1; |
1381 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1382 | extent_end = found_key.offset + | |
514ac8ad CM |
1383 | btrfs_file_extent_inline_len(leaf, |
1384 | path->slots[0], fi); | |
80ff3856 YZ |
1385 | extent_end = ALIGN(extent_end, root->sectorsize); |
1386 | } else { | |
1387 | BUG_ON(1); | |
1388 | } | |
1389 | out_check: | |
1390 | if (extent_end <= start) { | |
1391 | path->slots[0]++; | |
e9894fd3 | 1392 | if (!nolock && nocow) |
9ea24bbe | 1393 | btrfs_end_write_no_snapshoting(root); |
80ff3856 YZ |
1394 | goto next_slot; |
1395 | } | |
1396 | if (!nocow) { | |
1397 | if (cow_start == (u64)-1) | |
1398 | cow_start = cur_offset; | |
1399 | cur_offset = extent_end; | |
1400 | if (cur_offset > end) | |
1401 | break; | |
1402 | path->slots[0]++; | |
1403 | goto next_slot; | |
7ea394f1 YZ |
1404 | } |
1405 | ||
b3b4aa74 | 1406 | btrfs_release_path(path); |
80ff3856 | 1407 | if (cow_start != (u64)-1) { |
00361589 JB |
1408 | ret = cow_file_range(inode, locked_page, |
1409 | cow_start, found_key.offset - 1, | |
1410 | page_started, nr_written, 1); | |
e9894fd3 WS |
1411 | if (ret) { |
1412 | if (!nolock && nocow) | |
9ea24bbe | 1413 | btrfs_end_write_no_snapshoting(root); |
79787eaa | 1414 | goto error; |
e9894fd3 | 1415 | } |
80ff3856 | 1416 | cow_start = (u64)-1; |
7ea394f1 | 1417 | } |
80ff3856 | 1418 | |
d899e052 YZ |
1419 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1420 | struct extent_map *em; | |
1421 | struct extent_map_tree *em_tree; | |
1422 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1423 | em = alloc_extent_map(); |
79787eaa | 1424 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1425 | em->start = cur_offset; |
70c8a91c | 1426 | em->orig_start = found_key.offset - extent_offset; |
d899e052 YZ |
1427 | em->len = num_bytes; |
1428 | em->block_len = num_bytes; | |
1429 | em->block_start = disk_bytenr; | |
b4939680 | 1430 | em->orig_block_len = disk_num_bytes; |
cc95bef6 | 1431 | em->ram_bytes = ram_bytes; |
d899e052 | 1432 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
2ab28f32 JB |
1433 | em->mod_start = em->start; |
1434 | em->mod_len = em->len; | |
d899e052 | 1435 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
b11e234d | 1436 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
70c8a91c | 1437 | em->generation = -1; |
d899e052 | 1438 | while (1) { |
890871be | 1439 | write_lock(&em_tree->lock); |
09a2a8f9 | 1440 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1441 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1442 | if (ret != -EEXIST) { |
1443 | free_extent_map(em); | |
1444 | break; | |
1445 | } | |
1446 | btrfs_drop_extent_cache(inode, em->start, | |
1447 | em->start + em->len - 1, 0); | |
1448 | } | |
1449 | type = BTRFS_ORDERED_PREALLOC; | |
1450 | } else { | |
1451 | type = BTRFS_ORDERED_NOCOW; | |
1452 | } | |
80ff3856 YZ |
1453 | |
1454 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1455 | num_bytes, num_bytes, type); |
79787eaa | 1456 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1457 | |
efa56464 YZ |
1458 | if (root->root_key.objectid == |
1459 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1460 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1461 | num_bytes); | |
e9894fd3 WS |
1462 | if (ret) { |
1463 | if (!nolock && nocow) | |
9ea24bbe | 1464 | btrfs_end_write_no_snapshoting(root); |
79787eaa | 1465 | goto error; |
e9894fd3 | 1466 | } |
efa56464 YZ |
1467 | } |
1468 | ||
c2790a2e JB |
1469 | extent_clear_unlock_delalloc(inode, cur_offset, |
1470 | cur_offset + num_bytes - 1, | |
1471 | locked_page, EXTENT_LOCKED | | |
1472 | EXTENT_DELALLOC, PAGE_UNLOCK | | |
1473 | PAGE_SET_PRIVATE2); | |
e9894fd3 | 1474 | if (!nolock && nocow) |
9ea24bbe | 1475 | btrfs_end_write_no_snapshoting(root); |
80ff3856 YZ |
1476 | cur_offset = extent_end; |
1477 | if (cur_offset > end) | |
1478 | break; | |
be20aa9d | 1479 | } |
b3b4aa74 | 1480 | btrfs_release_path(path); |
80ff3856 | 1481 | |
17ca04af | 1482 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1483 | cow_start = cur_offset; |
17ca04af JB |
1484 | cur_offset = end; |
1485 | } | |
1486 | ||
80ff3856 | 1487 | if (cow_start != (u64)-1) { |
00361589 JB |
1488 | ret = cow_file_range(inode, locked_page, cow_start, end, |
1489 | page_started, nr_written, 1); | |
d788a349 | 1490 | if (ret) |
79787eaa | 1491 | goto error; |
80ff3856 YZ |
1492 | } |
1493 | ||
79787eaa | 1494 | error: |
a698d075 | 1495 | err = btrfs_end_transaction(trans, root); |
79787eaa JM |
1496 | if (!ret) |
1497 | ret = err; | |
1498 | ||
17ca04af | 1499 | if (ret && cur_offset < end) |
c2790a2e JB |
1500 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
1501 | locked_page, EXTENT_LOCKED | | |
151a41bc JB |
1502 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1503 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1504 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1505 | PAGE_SET_WRITEBACK | |
1506 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1507 | btrfs_free_path(path); |
79787eaa | 1508 | return ret; |
be20aa9d CM |
1509 | } |
1510 | ||
47059d93 WS |
1511 | static inline int need_force_cow(struct inode *inode, u64 start, u64 end) |
1512 | { | |
1513 | ||
1514 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
1515 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) | |
1516 | return 0; | |
1517 | ||
1518 | /* | |
1519 | * @defrag_bytes is a hint value, no spinlock held here, | |
1520 | * if is not zero, it means the file is defragging. | |
1521 | * Force cow if given extent needs to be defragged. | |
1522 | */ | |
1523 | if (BTRFS_I(inode)->defrag_bytes && | |
1524 | test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1525 | EXTENT_DEFRAG, 0, NULL)) | |
1526 | return 1; | |
1527 | ||
1528 | return 0; | |
1529 | } | |
1530 | ||
d352ac68 CM |
1531 | /* |
1532 | * extent_io.c call back to do delayed allocation processing | |
1533 | */ | |
c8b97818 | 1534 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1535 | u64 start, u64 end, int *page_started, |
1536 | unsigned long *nr_written) | |
be20aa9d | 1537 | { |
be20aa9d | 1538 | int ret; |
47059d93 | 1539 | int force_cow = need_force_cow(inode, start, end); |
a2135011 | 1540 | |
47059d93 | 1541 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
c8b97818 | 1542 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1543 | page_started, 1, nr_written); |
47059d93 | 1544 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
d899e052 | 1545 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1546 | page_started, 0, nr_written); |
7816030e | 1547 | } else if (!inode_need_compress(inode)) { |
7f366cfe CM |
1548 | ret = cow_file_range(inode, locked_page, start, end, |
1549 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1550 | } else { |
1551 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1552 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1553 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1554 | page_started, nr_written); |
7ddf5a42 | 1555 | } |
b888db2b CM |
1556 | return ret; |
1557 | } | |
1558 | ||
1bf85046 JM |
1559 | static void btrfs_split_extent_hook(struct inode *inode, |
1560 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1561 | { |
dcab6a3b JB |
1562 | u64 size; |
1563 | ||
0ca1f7ce | 1564 | /* not delalloc, ignore it */ |
9ed74f2d | 1565 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1566 | return; |
9ed74f2d | 1567 | |
dcab6a3b JB |
1568 | size = orig->end - orig->start + 1; |
1569 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
1570 | u64 num_extents; | |
1571 | u64 new_size; | |
1572 | ||
1573 | /* | |
ba117213 JB |
1574 | * See the explanation in btrfs_merge_extent_hook, the same |
1575 | * applies here, just in reverse. | |
dcab6a3b JB |
1576 | */ |
1577 | new_size = orig->end - split + 1; | |
ba117213 | 1578 | num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1, |
dcab6a3b | 1579 | BTRFS_MAX_EXTENT_SIZE); |
ba117213 JB |
1580 | new_size = split - orig->start; |
1581 | num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1582 | BTRFS_MAX_EXTENT_SIZE); | |
1583 | if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1584 | BTRFS_MAX_EXTENT_SIZE) >= num_extents) | |
dcab6a3b JB |
1585 | return; |
1586 | } | |
1587 | ||
9e0baf60 JB |
1588 | spin_lock(&BTRFS_I(inode)->lock); |
1589 | BTRFS_I(inode)->outstanding_extents++; | |
1590 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1591 | } |
1592 | ||
1593 | /* | |
1594 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1595 | * extents so we can keep track of new extents that are just merged onto old | |
1596 | * extents, such as when we are doing sequential writes, so we can properly | |
1597 | * account for the metadata space we'll need. | |
1598 | */ | |
1bf85046 JM |
1599 | static void btrfs_merge_extent_hook(struct inode *inode, |
1600 | struct extent_state *new, | |
1601 | struct extent_state *other) | |
9ed74f2d | 1602 | { |
dcab6a3b JB |
1603 | u64 new_size, old_size; |
1604 | u64 num_extents; | |
1605 | ||
9ed74f2d JB |
1606 | /* not delalloc, ignore it */ |
1607 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1608 | return; |
9ed74f2d | 1609 | |
8461a3de JB |
1610 | if (new->start > other->start) |
1611 | new_size = new->end - other->start + 1; | |
1612 | else | |
1613 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
1614 | |
1615 | /* we're not bigger than the max, unreserve the space and go */ | |
1616 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
1617 | spin_lock(&BTRFS_I(inode)->lock); | |
1618 | BTRFS_I(inode)->outstanding_extents--; | |
1619 | spin_unlock(&BTRFS_I(inode)->lock); | |
1620 | return; | |
1621 | } | |
1622 | ||
1623 | /* | |
ba117213 JB |
1624 | * We have to add up either side to figure out how many extents were |
1625 | * accounted for before we merged into one big extent. If the number of | |
1626 | * extents we accounted for is <= the amount we need for the new range | |
1627 | * then we can return, otherwise drop. Think of it like this | |
1628 | * | |
1629 | * [ 4k][MAX_SIZE] | |
1630 | * | |
1631 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
1632 | * need 2 outstanding extents, on one side we have 1 and the other side | |
1633 | * we have 1 so they are == and we can return. But in this case | |
1634 | * | |
1635 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
1636 | * | |
1637 | * Each range on their own accounts for 2 extents, but merged together | |
1638 | * they are only 3 extents worth of accounting, so we need to drop in | |
1639 | * this case. | |
dcab6a3b | 1640 | */ |
ba117213 | 1641 | old_size = other->end - other->start + 1; |
dcab6a3b JB |
1642 | num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1, |
1643 | BTRFS_MAX_EXTENT_SIZE); | |
ba117213 JB |
1644 | old_size = new->end - new->start + 1; |
1645 | num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1646 | BTRFS_MAX_EXTENT_SIZE); | |
1647 | ||
dcab6a3b | 1648 | if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1, |
ba117213 | 1649 | BTRFS_MAX_EXTENT_SIZE) >= num_extents) |
dcab6a3b JB |
1650 | return; |
1651 | ||
9e0baf60 JB |
1652 | spin_lock(&BTRFS_I(inode)->lock); |
1653 | BTRFS_I(inode)->outstanding_extents--; | |
1654 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1655 | } |
1656 | ||
eb73c1b7 MX |
1657 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1658 | struct inode *inode) | |
1659 | { | |
1660 | spin_lock(&root->delalloc_lock); | |
1661 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1662 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1663 | &root->delalloc_inodes); | |
1664 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1665 | &BTRFS_I(inode)->runtime_flags); | |
1666 | root->nr_delalloc_inodes++; | |
1667 | if (root->nr_delalloc_inodes == 1) { | |
1668 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1669 | BUG_ON(!list_empty(&root->delalloc_root)); | |
1670 | list_add_tail(&root->delalloc_root, | |
1671 | &root->fs_info->delalloc_roots); | |
1672 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1673 | } | |
1674 | } | |
1675 | spin_unlock(&root->delalloc_lock); | |
1676 | } | |
1677 | ||
1678 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1679 | struct inode *inode) | |
1680 | { | |
1681 | spin_lock(&root->delalloc_lock); | |
1682 | if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1683 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1684 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1685 | &BTRFS_I(inode)->runtime_flags); | |
1686 | root->nr_delalloc_inodes--; | |
1687 | if (!root->nr_delalloc_inodes) { | |
1688 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1689 | BUG_ON(list_empty(&root->delalloc_root)); | |
1690 | list_del_init(&root->delalloc_root); | |
1691 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1692 | } | |
1693 | } | |
1694 | spin_unlock(&root->delalloc_lock); | |
1695 | } | |
1696 | ||
d352ac68 CM |
1697 | /* |
1698 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1699 | * bytes in this file, and to maintain the list of inodes that | |
1700 | * have pending delalloc work to be done. | |
1701 | */ | |
1bf85046 | 1702 | static void btrfs_set_bit_hook(struct inode *inode, |
9ee49a04 | 1703 | struct extent_state *state, unsigned *bits) |
291d673e | 1704 | { |
9ed74f2d | 1705 | |
47059d93 WS |
1706 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1707 | WARN_ON(1); | |
75eff68e CM |
1708 | /* |
1709 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1710 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1711 | * bit, which is only set or cleared with irqs on |
1712 | */ | |
0ca1f7ce | 1713 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1714 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1715 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1716 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1717 | |
9e0baf60 | 1718 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1719 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1720 | } else { |
1721 | spin_lock(&BTRFS_I(inode)->lock); | |
1722 | BTRFS_I(inode)->outstanding_extents++; | |
1723 | spin_unlock(&BTRFS_I(inode)->lock); | |
1724 | } | |
287a0ab9 | 1725 | |
6a3891c5 JB |
1726 | /* For sanity tests */ |
1727 | if (btrfs_test_is_dummy_root(root)) | |
1728 | return; | |
1729 | ||
963d678b MX |
1730 | __percpu_counter_add(&root->fs_info->delalloc_bytes, len, |
1731 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1732 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1733 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
1734 | if (*bits & EXTENT_DEFRAG) |
1735 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 1736 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1737 | &BTRFS_I(inode)->runtime_flags)) |
1738 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1739 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1740 | } |
291d673e CM |
1741 | } |
1742 | ||
d352ac68 CM |
1743 | /* |
1744 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1745 | */ | |
1bf85046 | 1746 | static void btrfs_clear_bit_hook(struct inode *inode, |
41074888 | 1747 | struct extent_state *state, |
9ee49a04 | 1748 | unsigned *bits) |
291d673e | 1749 | { |
47059d93 | 1750 | u64 len = state->end + 1 - state->start; |
dcab6a3b JB |
1751 | u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1, |
1752 | BTRFS_MAX_EXTENT_SIZE); | |
47059d93 WS |
1753 | |
1754 | spin_lock(&BTRFS_I(inode)->lock); | |
1755 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) | |
1756 | BTRFS_I(inode)->defrag_bytes -= len; | |
1757 | spin_unlock(&BTRFS_I(inode)->lock); | |
1758 | ||
75eff68e CM |
1759 | /* |
1760 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1761 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1762 | * bit, which is only set or cleared with irqs on |
1763 | */ | |
0ca1f7ce | 1764 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1765 | struct btrfs_root *root = BTRFS_I(inode)->root; |
83eea1f1 | 1766 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1767 | |
9e0baf60 | 1768 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1769 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1770 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1771 | spin_lock(&BTRFS_I(inode)->lock); | |
dcab6a3b | 1772 | BTRFS_I(inode)->outstanding_extents -= num_extents; |
9e0baf60 JB |
1773 | spin_unlock(&BTRFS_I(inode)->lock); |
1774 | } | |
0ca1f7ce | 1775 | |
b6d08f06 JB |
1776 | /* |
1777 | * We don't reserve metadata space for space cache inodes so we | |
1778 | * don't need to call dellalloc_release_metadata if there is an | |
1779 | * error. | |
1780 | */ | |
1781 | if (*bits & EXTENT_DO_ACCOUNTING && | |
1782 | root != root->fs_info->tree_root) | |
0ca1f7ce YZ |
1783 | btrfs_delalloc_release_metadata(inode, len); |
1784 | ||
6a3891c5 JB |
1785 | /* For sanity tests. */ |
1786 | if (btrfs_test_is_dummy_root(root)) | |
1787 | return; | |
1788 | ||
0cb59c99 | 1789 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
7ee9e440 | 1790 | && do_list && !(state->state & EXTENT_NORESERVE)) |
51773bec QW |
1791 | btrfs_free_reserved_data_space_noquota(inode, |
1792 | state->start, len); | |
9ed74f2d | 1793 | |
963d678b MX |
1794 | __percpu_counter_add(&root->fs_info->delalloc_bytes, -len, |
1795 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1796 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1797 | BTRFS_I(inode)->delalloc_bytes -= len; |
0cb59c99 | 1798 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
df0af1a5 | 1799 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1800 | &BTRFS_I(inode)->runtime_flags)) |
1801 | btrfs_del_delalloc_inode(root, inode); | |
df0af1a5 | 1802 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1803 | } |
291d673e CM |
1804 | } |
1805 | ||
d352ac68 CM |
1806 | /* |
1807 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1808 | * we don't create bios that span stripes or chunks | |
1809 | */ | |
64a16701 | 1810 | int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset, |
c8b97818 CM |
1811 | size_t size, struct bio *bio, |
1812 | unsigned long bio_flags) | |
239b14b3 CM |
1813 | { |
1814 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
4f024f37 | 1815 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
1816 | u64 length = 0; |
1817 | u64 map_length; | |
239b14b3 CM |
1818 | int ret; |
1819 | ||
771ed689 CM |
1820 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1821 | return 0; | |
1822 | ||
4f024f37 | 1823 | length = bio->bi_iter.bi_size; |
239b14b3 | 1824 | map_length = length; |
64a16701 | 1825 | ret = btrfs_map_block(root->fs_info, rw, logical, |
f188591e | 1826 | &map_length, NULL, 0); |
3ec706c8 | 1827 | /* Will always return 0 with map_multi == NULL */ |
3444a972 | 1828 | BUG_ON(ret < 0); |
d397712b | 1829 | if (map_length < length + size) |
239b14b3 | 1830 | return 1; |
3444a972 | 1831 | return 0; |
239b14b3 CM |
1832 | } |
1833 | ||
d352ac68 CM |
1834 | /* |
1835 | * in order to insert checksums into the metadata in large chunks, | |
1836 | * we wait until bio submission time. All the pages in the bio are | |
1837 | * checksummed and sums are attached onto the ordered extent record. | |
1838 | * | |
1839 | * At IO completion time the cums attached on the ordered extent record | |
1840 | * are inserted into the btree | |
1841 | */ | |
d397712b CM |
1842 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1843 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1844 | unsigned long bio_flags, |
1845 | u64 bio_offset) | |
065631f6 | 1846 | { |
065631f6 | 1847 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1848 | int ret = 0; |
e015640f | 1849 | |
d20f7043 | 1850 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1851 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1852 | return 0; |
1853 | } | |
e015640f | 1854 | |
4a69a410 CM |
1855 | /* |
1856 | * in order to insert checksums into the metadata in large chunks, | |
1857 | * we wait until bio submission time. All the pages in the bio are | |
1858 | * checksummed and sums are attached onto the ordered extent record. | |
1859 | * | |
1860 | * At IO completion time the cums attached on the ordered extent record | |
1861 | * are inserted into the btree | |
1862 | */ | |
b2950863 | 1863 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1864 | int mirror_num, unsigned long bio_flags, |
1865 | u64 bio_offset) | |
4a69a410 CM |
1866 | { |
1867 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
61891923 SB |
1868 | int ret; |
1869 | ||
1870 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 1); | |
4246a0b6 CH |
1871 | if (ret) { |
1872 | bio->bi_error = ret; | |
1873 | bio_endio(bio); | |
1874 | } | |
61891923 | 1875 | return ret; |
44b8bd7e CM |
1876 | } |
1877 | ||
d352ac68 | 1878 | /* |
cad321ad CM |
1879 | * extent_io.c submission hook. This does the right thing for csum calculation |
1880 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1881 | */ |
b2950863 | 1882 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1883 | int mirror_num, unsigned long bio_flags, |
1884 | u64 bio_offset) | |
44b8bd7e CM |
1885 | { |
1886 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
0d51e28a | 1887 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
44b8bd7e | 1888 | int ret = 0; |
19b9bdb0 | 1889 | int skip_sum; |
b812ce28 | 1890 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1891 | |
6cbff00f | 1892 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1893 | |
83eea1f1 | 1894 | if (btrfs_is_free_space_inode(inode)) |
0d51e28a | 1895 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 1896 | |
7b6d91da | 1897 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1898 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1899 | if (ret) | |
61891923 | 1900 | goto out; |
5fd02043 | 1901 | |
d20f7043 | 1902 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1903 | ret = btrfs_submit_compressed_read(inode, bio, |
1904 | mirror_num, | |
1905 | bio_flags); | |
1906 | goto out; | |
c2db1073 TI |
1907 | } else if (!skip_sum) { |
1908 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1909 | if (ret) | |
61891923 | 1910 | goto out; |
c2db1073 | 1911 | } |
4d1b5fb4 | 1912 | goto mapit; |
b812ce28 | 1913 | } else if (async && !skip_sum) { |
17d217fe YZ |
1914 | /* csum items have already been cloned */ |
1915 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1916 | goto mapit; | |
19b9bdb0 | 1917 | /* we're doing a write, do the async checksumming */ |
61891923 | 1918 | ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, |
44b8bd7e | 1919 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1920 | bio_flags, bio_offset, |
1921 | __btrfs_submit_bio_start, | |
4a69a410 | 1922 | __btrfs_submit_bio_done); |
61891923 | 1923 | goto out; |
b812ce28 JB |
1924 | } else if (!skip_sum) { |
1925 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); | |
1926 | if (ret) | |
1927 | goto out; | |
19b9bdb0 CM |
1928 | } |
1929 | ||
0b86a832 | 1930 | mapit: |
61891923 SB |
1931 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); |
1932 | ||
1933 | out: | |
4246a0b6 CH |
1934 | if (ret < 0) { |
1935 | bio->bi_error = ret; | |
1936 | bio_endio(bio); | |
1937 | } | |
61891923 | 1938 | return ret; |
065631f6 | 1939 | } |
6885f308 | 1940 | |
d352ac68 CM |
1941 | /* |
1942 | * given a list of ordered sums record them in the inode. This happens | |
1943 | * at IO completion time based on sums calculated at bio submission time. | |
1944 | */ | |
ba1da2f4 | 1945 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1946 | struct inode *inode, u64 file_offset, |
1947 | struct list_head *list) | |
1948 | { | |
e6dcd2dc CM |
1949 | struct btrfs_ordered_sum *sum; |
1950 | ||
c6e30871 | 1951 | list_for_each_entry(sum, list, list) { |
39847c4d | 1952 | trans->adding_csums = 1; |
d20f7043 CM |
1953 | btrfs_csum_file_blocks(trans, |
1954 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 1955 | trans->adding_csums = 0; |
e6dcd2dc CM |
1956 | } |
1957 | return 0; | |
1958 | } | |
1959 | ||
2ac55d41 JB |
1960 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1961 | struct extent_state **cached_state) | |
ea8c2819 | 1962 | { |
09cbfeaf | 1963 | WARN_ON((end & (PAGE_SIZE - 1)) == 0); |
ea8c2819 | 1964 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1965 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1966 | } |
1967 | ||
d352ac68 | 1968 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1969 | struct btrfs_writepage_fixup { |
1970 | struct page *page; | |
1971 | struct btrfs_work work; | |
1972 | }; | |
1973 | ||
b2950863 | 1974 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1975 | { |
1976 | struct btrfs_writepage_fixup *fixup; | |
1977 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1978 | struct extent_state *cached_state = NULL; |
247e743c CM |
1979 | struct page *page; |
1980 | struct inode *inode; | |
1981 | u64 page_start; | |
1982 | u64 page_end; | |
87826df0 | 1983 | int ret; |
247e743c CM |
1984 | |
1985 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1986 | page = fixup->page; | |
4a096752 | 1987 | again: |
247e743c CM |
1988 | lock_page(page); |
1989 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1990 | ClearPageChecked(page); | |
1991 | goto out_page; | |
1992 | } | |
1993 | ||
1994 | inode = page->mapping->host; | |
1995 | page_start = page_offset(page); | |
09cbfeaf | 1996 | page_end = page_offset(page) + PAGE_SIZE - 1; |
247e743c | 1997 | |
ff13db41 | 1998 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
d0082371 | 1999 | &cached_state); |
4a096752 CM |
2000 | |
2001 | /* already ordered? We're done */ | |
8b62b72b | 2002 | if (PagePrivate2(page)) |
247e743c | 2003 | goto out; |
4a096752 | 2004 | |
dbfdb6d1 | 2005 | ordered = btrfs_lookup_ordered_range(inode, page_start, |
09cbfeaf | 2006 | PAGE_SIZE); |
4a096752 | 2007 | if (ordered) { |
2ac55d41 JB |
2008 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
2009 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
2010 | unlock_page(page); |
2011 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 2012 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2013 | goto again; |
2014 | } | |
247e743c | 2015 | |
7cf5b976 | 2016 | ret = btrfs_delalloc_reserve_space(inode, page_start, |
09cbfeaf | 2017 | PAGE_SIZE); |
87826df0 JM |
2018 | if (ret) { |
2019 | mapping_set_error(page->mapping, ret); | |
2020 | end_extent_writepage(page, ret, page_start, page_end); | |
2021 | ClearPageChecked(page); | |
2022 | goto out; | |
2023 | } | |
2024 | ||
2ac55d41 | 2025 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 2026 | ClearPageChecked(page); |
87826df0 | 2027 | set_page_dirty(page); |
247e743c | 2028 | out: |
2ac55d41 JB |
2029 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
2030 | &cached_state, GFP_NOFS); | |
247e743c CM |
2031 | out_page: |
2032 | unlock_page(page); | |
09cbfeaf | 2033 | put_page(page); |
b897abec | 2034 | kfree(fixup); |
247e743c CM |
2035 | } |
2036 | ||
2037 | /* | |
2038 | * There are a few paths in the higher layers of the kernel that directly | |
2039 | * set the page dirty bit without asking the filesystem if it is a | |
2040 | * good idea. This causes problems because we want to make sure COW | |
2041 | * properly happens and the data=ordered rules are followed. | |
2042 | * | |
c8b97818 | 2043 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2044 | * hasn't been properly setup for IO. We kick off an async process |
2045 | * to fix it up. The async helper will wait for ordered extents, set | |
2046 | * the delalloc bit and make it safe to write the page. | |
2047 | */ | |
b2950863 | 2048 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
2049 | { |
2050 | struct inode *inode = page->mapping->host; | |
2051 | struct btrfs_writepage_fixup *fixup; | |
2052 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 2053 | |
8b62b72b CM |
2054 | /* this page is properly in the ordered list */ |
2055 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2056 | return 0; |
2057 | ||
2058 | if (PageChecked(page)) | |
2059 | return -EAGAIN; | |
2060 | ||
2061 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2062 | if (!fixup) | |
2063 | return -EAGAIN; | |
f421950f | 2064 | |
247e743c | 2065 | SetPageChecked(page); |
09cbfeaf | 2066 | get_page(page); |
9e0af237 LB |
2067 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
2068 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 2069 | fixup->page = page; |
dc6e3209 | 2070 | btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work); |
87826df0 | 2071 | return -EBUSY; |
247e743c CM |
2072 | } |
2073 | ||
d899e052 YZ |
2074 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
2075 | struct inode *inode, u64 file_pos, | |
2076 | u64 disk_bytenr, u64 disk_num_bytes, | |
2077 | u64 num_bytes, u64 ram_bytes, | |
2078 | u8 compression, u8 encryption, | |
2079 | u16 other_encoding, int extent_type) | |
2080 | { | |
2081 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2082 | struct btrfs_file_extent_item *fi; | |
2083 | struct btrfs_path *path; | |
2084 | struct extent_buffer *leaf; | |
2085 | struct btrfs_key ins; | |
1acae57b | 2086 | int extent_inserted = 0; |
d899e052 YZ |
2087 | int ret; |
2088 | ||
2089 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2090 | if (!path) |
2091 | return -ENOMEM; | |
d899e052 | 2092 | |
a1ed835e CM |
2093 | /* |
2094 | * we may be replacing one extent in the tree with another. | |
2095 | * The new extent is pinned in the extent map, and we don't want | |
2096 | * to drop it from the cache until it is completely in the btree. | |
2097 | * | |
2098 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2099 | * the caller is expected to unpin it and allow it to be merged | |
2100 | * with the others. | |
2101 | */ | |
1acae57b FDBM |
2102 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
2103 | file_pos + num_bytes, NULL, 0, | |
2104 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
2105 | if (ret) |
2106 | goto out; | |
d899e052 | 2107 | |
1acae57b FDBM |
2108 | if (!extent_inserted) { |
2109 | ins.objectid = btrfs_ino(inode); | |
2110 | ins.offset = file_pos; | |
2111 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2112 | ||
2113 | path->leave_spinning = 1; | |
2114 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
2115 | sizeof(*fi)); | |
2116 | if (ret) | |
2117 | goto out; | |
2118 | } | |
d899e052 YZ |
2119 | leaf = path->nodes[0]; |
2120 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2121 | struct btrfs_file_extent_item); | |
2122 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
2123 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
2124 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
2125 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
2126 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2127 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2128 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2129 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2130 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2131 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2132 | |
d899e052 | 2133 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2134 | btrfs_release_path(path); |
d899e052 YZ |
2135 | |
2136 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2137 | |
2138 | ins.objectid = disk_bytenr; | |
2139 | ins.offset = disk_num_bytes; | |
2140 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
2141 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
2142 | root->root_key.objectid, | |
5846a3c2 QW |
2143 | btrfs_ino(inode), file_pos, |
2144 | ram_bytes, &ins); | |
297d750b | 2145 | /* |
5846a3c2 QW |
2146 | * Release the reserved range from inode dirty range map, as it is |
2147 | * already moved into delayed_ref_head | |
297d750b QW |
2148 | */ |
2149 | btrfs_qgroup_release_data(inode, file_pos, ram_bytes); | |
79787eaa | 2150 | out: |
d899e052 | 2151 | btrfs_free_path(path); |
b9473439 | 2152 | |
79787eaa | 2153 | return ret; |
d899e052 YZ |
2154 | } |
2155 | ||
38c227d8 LB |
2156 | /* snapshot-aware defrag */ |
2157 | struct sa_defrag_extent_backref { | |
2158 | struct rb_node node; | |
2159 | struct old_sa_defrag_extent *old; | |
2160 | u64 root_id; | |
2161 | u64 inum; | |
2162 | u64 file_pos; | |
2163 | u64 extent_offset; | |
2164 | u64 num_bytes; | |
2165 | u64 generation; | |
2166 | }; | |
2167 | ||
2168 | struct old_sa_defrag_extent { | |
2169 | struct list_head list; | |
2170 | struct new_sa_defrag_extent *new; | |
2171 | ||
2172 | u64 extent_offset; | |
2173 | u64 bytenr; | |
2174 | u64 offset; | |
2175 | u64 len; | |
2176 | int count; | |
2177 | }; | |
2178 | ||
2179 | struct new_sa_defrag_extent { | |
2180 | struct rb_root root; | |
2181 | struct list_head head; | |
2182 | struct btrfs_path *path; | |
2183 | struct inode *inode; | |
2184 | u64 file_pos; | |
2185 | u64 len; | |
2186 | u64 bytenr; | |
2187 | u64 disk_len; | |
2188 | u8 compress_type; | |
2189 | }; | |
2190 | ||
2191 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2192 | struct sa_defrag_extent_backref *b2) | |
2193 | { | |
2194 | if (b1->root_id < b2->root_id) | |
2195 | return -1; | |
2196 | else if (b1->root_id > b2->root_id) | |
2197 | return 1; | |
2198 | ||
2199 | if (b1->inum < b2->inum) | |
2200 | return -1; | |
2201 | else if (b1->inum > b2->inum) | |
2202 | return 1; | |
2203 | ||
2204 | if (b1->file_pos < b2->file_pos) | |
2205 | return -1; | |
2206 | else if (b1->file_pos > b2->file_pos) | |
2207 | return 1; | |
2208 | ||
2209 | /* | |
2210 | * [------------------------------] ===> (a range of space) | |
2211 | * |<--->| |<---->| =============> (fs/file tree A) | |
2212 | * |<---------------------------->| ===> (fs/file tree B) | |
2213 | * | |
2214 | * A range of space can refer to two file extents in one tree while | |
2215 | * refer to only one file extent in another tree. | |
2216 | * | |
2217 | * So we may process a disk offset more than one time(two extents in A) | |
2218 | * and locate at the same extent(one extent in B), then insert two same | |
2219 | * backrefs(both refer to the extent in B). | |
2220 | */ | |
2221 | return 0; | |
2222 | } | |
2223 | ||
2224 | static void backref_insert(struct rb_root *root, | |
2225 | struct sa_defrag_extent_backref *backref) | |
2226 | { | |
2227 | struct rb_node **p = &root->rb_node; | |
2228 | struct rb_node *parent = NULL; | |
2229 | struct sa_defrag_extent_backref *entry; | |
2230 | int ret; | |
2231 | ||
2232 | while (*p) { | |
2233 | parent = *p; | |
2234 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2235 | ||
2236 | ret = backref_comp(backref, entry); | |
2237 | if (ret < 0) | |
2238 | p = &(*p)->rb_left; | |
2239 | else | |
2240 | p = &(*p)->rb_right; | |
2241 | } | |
2242 | ||
2243 | rb_link_node(&backref->node, parent, p); | |
2244 | rb_insert_color(&backref->node, root); | |
2245 | } | |
2246 | ||
2247 | /* | |
2248 | * Note the backref might has changed, and in this case we just return 0. | |
2249 | */ | |
2250 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2251 | void *ctx) | |
2252 | { | |
2253 | struct btrfs_file_extent_item *extent; | |
2254 | struct btrfs_fs_info *fs_info; | |
2255 | struct old_sa_defrag_extent *old = ctx; | |
2256 | struct new_sa_defrag_extent *new = old->new; | |
2257 | struct btrfs_path *path = new->path; | |
2258 | struct btrfs_key key; | |
2259 | struct btrfs_root *root; | |
2260 | struct sa_defrag_extent_backref *backref; | |
2261 | struct extent_buffer *leaf; | |
2262 | struct inode *inode = new->inode; | |
2263 | int slot; | |
2264 | int ret; | |
2265 | u64 extent_offset; | |
2266 | u64 num_bytes; | |
2267 | ||
2268 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
2269 | inum == btrfs_ino(inode)) | |
2270 | return 0; | |
2271 | ||
2272 | key.objectid = root_id; | |
2273 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2274 | key.offset = (u64)-1; | |
2275 | ||
2276 | fs_info = BTRFS_I(inode)->root->fs_info; | |
2277 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2278 | if (IS_ERR(root)) { | |
2279 | if (PTR_ERR(root) == -ENOENT) | |
2280 | return 0; | |
2281 | WARN_ON(1); | |
2282 | pr_debug("inum=%llu, offset=%llu, root_id=%llu\n", | |
2283 | inum, offset, root_id); | |
2284 | return PTR_ERR(root); | |
2285 | } | |
2286 | ||
2287 | key.objectid = inum; | |
2288 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2289 | if (offset > (u64)-1 << 32) | |
2290 | key.offset = 0; | |
2291 | else | |
2292 | key.offset = offset; | |
2293 | ||
2294 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2295 | if (WARN_ON(ret < 0)) |
38c227d8 | 2296 | return ret; |
50f1319c | 2297 | ret = 0; |
38c227d8 LB |
2298 | |
2299 | while (1) { | |
2300 | cond_resched(); | |
2301 | ||
2302 | leaf = path->nodes[0]; | |
2303 | slot = path->slots[0]; | |
2304 | ||
2305 | if (slot >= btrfs_header_nritems(leaf)) { | |
2306 | ret = btrfs_next_leaf(root, path); | |
2307 | if (ret < 0) { | |
2308 | goto out; | |
2309 | } else if (ret > 0) { | |
2310 | ret = 0; | |
2311 | goto out; | |
2312 | } | |
2313 | continue; | |
2314 | } | |
2315 | ||
2316 | path->slots[0]++; | |
2317 | ||
2318 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2319 | ||
2320 | if (key.objectid > inum) | |
2321 | goto out; | |
2322 | ||
2323 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2324 | continue; | |
2325 | ||
2326 | extent = btrfs_item_ptr(leaf, slot, | |
2327 | struct btrfs_file_extent_item); | |
2328 | ||
2329 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2330 | continue; | |
2331 | ||
e68afa49 LB |
2332 | /* |
2333 | * 'offset' refers to the exact key.offset, | |
2334 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2335 | * (key.offset - extent_offset). | |
2336 | */ | |
2337 | if (key.offset != offset) | |
38c227d8 LB |
2338 | continue; |
2339 | ||
e68afa49 | 2340 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2341 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2342 | |
38c227d8 LB |
2343 | if (extent_offset >= old->extent_offset + old->offset + |
2344 | old->len || extent_offset + num_bytes <= | |
2345 | old->extent_offset + old->offset) | |
2346 | continue; | |
38c227d8 LB |
2347 | break; |
2348 | } | |
2349 | ||
2350 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2351 | if (!backref) { | |
2352 | ret = -ENOENT; | |
2353 | goto out; | |
2354 | } | |
2355 | ||
2356 | backref->root_id = root_id; | |
2357 | backref->inum = inum; | |
e68afa49 | 2358 | backref->file_pos = offset; |
38c227d8 LB |
2359 | backref->num_bytes = num_bytes; |
2360 | backref->extent_offset = extent_offset; | |
2361 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2362 | backref->old = old; | |
2363 | backref_insert(&new->root, backref); | |
2364 | old->count++; | |
2365 | out: | |
2366 | btrfs_release_path(path); | |
2367 | WARN_ON(ret); | |
2368 | return ret; | |
2369 | } | |
2370 | ||
2371 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2372 | struct new_sa_defrag_extent *new) | |
2373 | { | |
2374 | struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info; | |
2375 | struct old_sa_defrag_extent *old, *tmp; | |
2376 | int ret; | |
2377 | ||
2378 | new->path = path; | |
2379 | ||
2380 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2381 | ret = iterate_inodes_from_logical(old->bytenr + |
2382 | old->extent_offset, fs_info, | |
38c227d8 LB |
2383 | path, record_one_backref, |
2384 | old); | |
4724b106 JB |
2385 | if (ret < 0 && ret != -ENOENT) |
2386 | return false; | |
38c227d8 LB |
2387 | |
2388 | /* no backref to be processed for this extent */ | |
2389 | if (!old->count) { | |
2390 | list_del(&old->list); | |
2391 | kfree(old); | |
2392 | } | |
2393 | } | |
2394 | ||
2395 | if (list_empty(&new->head)) | |
2396 | return false; | |
2397 | ||
2398 | return true; | |
2399 | } | |
2400 | ||
2401 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2402 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2403 | struct new_sa_defrag_extent *new) |
38c227d8 | 2404 | { |
116e0024 | 2405 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2406 | return 0; |
2407 | ||
2408 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2409 | return 0; | |
2410 | ||
116e0024 LB |
2411 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2412 | return 0; | |
2413 | ||
2414 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2415 | btrfs_file_extent_other_encoding(leaf, fi)) |
2416 | return 0; | |
2417 | ||
2418 | return 1; | |
2419 | } | |
2420 | ||
2421 | /* | |
2422 | * Note the backref might has changed, and in this case we just return 0. | |
2423 | */ | |
2424 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2425 | struct sa_defrag_extent_backref *prev, | |
2426 | struct sa_defrag_extent_backref *backref) | |
2427 | { | |
2428 | struct btrfs_file_extent_item *extent; | |
2429 | struct btrfs_file_extent_item *item; | |
2430 | struct btrfs_ordered_extent *ordered; | |
2431 | struct btrfs_trans_handle *trans; | |
2432 | struct btrfs_fs_info *fs_info; | |
2433 | struct btrfs_root *root; | |
2434 | struct btrfs_key key; | |
2435 | struct extent_buffer *leaf; | |
2436 | struct old_sa_defrag_extent *old = backref->old; | |
2437 | struct new_sa_defrag_extent *new = old->new; | |
2438 | struct inode *src_inode = new->inode; | |
2439 | struct inode *inode; | |
2440 | struct extent_state *cached = NULL; | |
2441 | int ret = 0; | |
2442 | u64 start; | |
2443 | u64 len; | |
2444 | u64 lock_start; | |
2445 | u64 lock_end; | |
2446 | bool merge = false; | |
2447 | int index; | |
2448 | ||
2449 | if (prev && prev->root_id == backref->root_id && | |
2450 | prev->inum == backref->inum && | |
2451 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2452 | merge = true; | |
2453 | ||
2454 | /* step 1: get root */ | |
2455 | key.objectid = backref->root_id; | |
2456 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2457 | key.offset = (u64)-1; | |
2458 | ||
2459 | fs_info = BTRFS_I(src_inode)->root->fs_info; | |
2460 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
2461 | ||
2462 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2463 | if (IS_ERR(root)) { | |
2464 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2465 | if (PTR_ERR(root) == -ENOENT) | |
2466 | return 0; | |
2467 | return PTR_ERR(root); | |
2468 | } | |
38c227d8 | 2469 | |
bcbba5e6 WS |
2470 | if (btrfs_root_readonly(root)) { |
2471 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2472 | return 0; | |
2473 | } | |
2474 | ||
38c227d8 LB |
2475 | /* step 2: get inode */ |
2476 | key.objectid = backref->inum; | |
2477 | key.type = BTRFS_INODE_ITEM_KEY; | |
2478 | key.offset = 0; | |
2479 | ||
2480 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2481 | if (IS_ERR(inode)) { | |
2482 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2483 | return 0; | |
2484 | } | |
2485 | ||
2486 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2487 | ||
2488 | /* step 3: relink backref */ | |
2489 | lock_start = backref->file_pos; | |
2490 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2491 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
ff13db41 | 2492 | &cached); |
38c227d8 LB |
2493 | |
2494 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2495 | if (ordered) { | |
2496 | btrfs_put_ordered_extent(ordered); | |
2497 | goto out_unlock; | |
2498 | } | |
2499 | ||
2500 | trans = btrfs_join_transaction(root); | |
2501 | if (IS_ERR(trans)) { | |
2502 | ret = PTR_ERR(trans); | |
2503 | goto out_unlock; | |
2504 | } | |
2505 | ||
2506 | key.objectid = backref->inum; | |
2507 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2508 | key.offset = backref->file_pos; | |
2509 | ||
2510 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2511 | if (ret < 0) { | |
2512 | goto out_free_path; | |
2513 | } else if (ret > 0) { | |
2514 | ret = 0; | |
2515 | goto out_free_path; | |
2516 | } | |
2517 | ||
2518 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2519 | struct btrfs_file_extent_item); | |
2520 | ||
2521 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2522 | backref->generation) | |
2523 | goto out_free_path; | |
2524 | ||
2525 | btrfs_release_path(path); | |
2526 | ||
2527 | start = backref->file_pos; | |
2528 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2529 | start += old->extent_offset + old->offset - | |
2530 | backref->extent_offset; | |
2531 | ||
2532 | len = min(backref->extent_offset + backref->num_bytes, | |
2533 | old->extent_offset + old->offset + old->len); | |
2534 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2535 | ||
2536 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2537 | start + len, 1); | |
2538 | if (ret) | |
2539 | goto out_free_path; | |
2540 | again: | |
2541 | key.objectid = btrfs_ino(inode); | |
2542 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2543 | key.offset = start; | |
2544 | ||
a09a0a70 | 2545 | path->leave_spinning = 1; |
38c227d8 LB |
2546 | if (merge) { |
2547 | struct btrfs_file_extent_item *fi; | |
2548 | u64 extent_len; | |
2549 | struct btrfs_key found_key; | |
2550 | ||
3c9665df | 2551 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2552 | if (ret < 0) |
2553 | goto out_free_path; | |
2554 | ||
2555 | path->slots[0]--; | |
2556 | leaf = path->nodes[0]; | |
2557 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2558 | ||
2559 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2560 | struct btrfs_file_extent_item); | |
2561 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2562 | ||
116e0024 LB |
2563 | if (extent_len + found_key.offset == start && |
2564 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2565 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2566 | extent_len + len); | |
2567 | btrfs_mark_buffer_dirty(leaf); | |
2568 | inode_add_bytes(inode, len); | |
2569 | ||
2570 | ret = 1; | |
2571 | goto out_free_path; | |
2572 | } else { | |
2573 | merge = false; | |
2574 | btrfs_release_path(path); | |
2575 | goto again; | |
2576 | } | |
2577 | } | |
2578 | ||
2579 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2580 | sizeof(*extent)); | |
2581 | if (ret) { | |
2582 | btrfs_abort_transaction(trans, root, ret); | |
2583 | goto out_free_path; | |
2584 | } | |
2585 | ||
2586 | leaf = path->nodes[0]; | |
2587 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2588 | struct btrfs_file_extent_item); | |
2589 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2590 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2591 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2592 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2593 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2594 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2595 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2596 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2597 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2598 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2599 | ||
2600 | btrfs_mark_buffer_dirty(leaf); | |
2601 | inode_add_bytes(inode, len); | |
a09a0a70 | 2602 | btrfs_release_path(path); |
38c227d8 LB |
2603 | |
2604 | ret = btrfs_inc_extent_ref(trans, root, new->bytenr, | |
2605 | new->disk_len, 0, | |
2606 | backref->root_id, backref->inum, | |
b06c4bf5 | 2607 | new->file_pos); /* start - extent_offset */ |
38c227d8 LB |
2608 | if (ret) { |
2609 | btrfs_abort_transaction(trans, root, ret); | |
2610 | goto out_free_path; | |
2611 | } | |
2612 | ||
2613 | ret = 1; | |
2614 | out_free_path: | |
2615 | btrfs_release_path(path); | |
a09a0a70 | 2616 | path->leave_spinning = 0; |
38c227d8 LB |
2617 | btrfs_end_transaction(trans, root); |
2618 | out_unlock: | |
2619 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2620 | &cached, GFP_NOFS); | |
2621 | iput(inode); | |
2622 | return ret; | |
2623 | } | |
2624 | ||
6f519564 LB |
2625 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2626 | { | |
2627 | struct old_sa_defrag_extent *old, *tmp; | |
2628 | ||
2629 | if (!new) | |
2630 | return; | |
2631 | ||
2632 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
6f519564 LB |
2633 | kfree(old); |
2634 | } | |
2635 | kfree(new); | |
2636 | } | |
2637 | ||
38c227d8 LB |
2638 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2639 | { | |
2640 | struct btrfs_path *path; | |
38c227d8 LB |
2641 | struct sa_defrag_extent_backref *backref; |
2642 | struct sa_defrag_extent_backref *prev = NULL; | |
2643 | struct inode *inode; | |
2644 | struct btrfs_root *root; | |
2645 | struct rb_node *node; | |
2646 | int ret; | |
2647 | ||
2648 | inode = new->inode; | |
2649 | root = BTRFS_I(inode)->root; | |
2650 | ||
2651 | path = btrfs_alloc_path(); | |
2652 | if (!path) | |
2653 | return; | |
2654 | ||
2655 | if (!record_extent_backrefs(path, new)) { | |
2656 | btrfs_free_path(path); | |
2657 | goto out; | |
2658 | } | |
2659 | btrfs_release_path(path); | |
2660 | ||
2661 | while (1) { | |
2662 | node = rb_first(&new->root); | |
2663 | if (!node) | |
2664 | break; | |
2665 | rb_erase(node, &new->root); | |
2666 | ||
2667 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2668 | ||
2669 | ret = relink_extent_backref(path, prev, backref); | |
2670 | WARN_ON(ret < 0); | |
2671 | ||
2672 | kfree(prev); | |
2673 | ||
2674 | if (ret == 1) | |
2675 | prev = backref; | |
2676 | else | |
2677 | prev = NULL; | |
2678 | cond_resched(); | |
2679 | } | |
2680 | kfree(prev); | |
2681 | ||
2682 | btrfs_free_path(path); | |
38c227d8 | 2683 | out: |
6f519564 LB |
2684 | free_sa_defrag_extent(new); |
2685 | ||
38c227d8 LB |
2686 | atomic_dec(&root->fs_info->defrag_running); |
2687 | wake_up(&root->fs_info->transaction_wait); | |
38c227d8 LB |
2688 | } |
2689 | ||
2690 | static struct new_sa_defrag_extent * | |
2691 | record_old_file_extents(struct inode *inode, | |
2692 | struct btrfs_ordered_extent *ordered) | |
2693 | { | |
2694 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2695 | struct btrfs_path *path; | |
2696 | struct btrfs_key key; | |
6f519564 | 2697 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2698 | struct new_sa_defrag_extent *new; |
2699 | int ret; | |
2700 | ||
2701 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2702 | if (!new) | |
2703 | return NULL; | |
2704 | ||
2705 | new->inode = inode; | |
2706 | new->file_pos = ordered->file_offset; | |
2707 | new->len = ordered->len; | |
2708 | new->bytenr = ordered->start; | |
2709 | new->disk_len = ordered->disk_len; | |
2710 | new->compress_type = ordered->compress_type; | |
2711 | new->root = RB_ROOT; | |
2712 | INIT_LIST_HEAD(&new->head); | |
2713 | ||
2714 | path = btrfs_alloc_path(); | |
2715 | if (!path) | |
2716 | goto out_kfree; | |
2717 | ||
2718 | key.objectid = btrfs_ino(inode); | |
2719 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2720 | key.offset = new->file_pos; | |
2721 | ||
2722 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2723 | if (ret < 0) | |
2724 | goto out_free_path; | |
2725 | if (ret > 0 && path->slots[0] > 0) | |
2726 | path->slots[0]--; | |
2727 | ||
2728 | /* find out all the old extents for the file range */ | |
2729 | while (1) { | |
2730 | struct btrfs_file_extent_item *extent; | |
2731 | struct extent_buffer *l; | |
2732 | int slot; | |
2733 | u64 num_bytes; | |
2734 | u64 offset; | |
2735 | u64 end; | |
2736 | u64 disk_bytenr; | |
2737 | u64 extent_offset; | |
2738 | ||
2739 | l = path->nodes[0]; | |
2740 | slot = path->slots[0]; | |
2741 | ||
2742 | if (slot >= btrfs_header_nritems(l)) { | |
2743 | ret = btrfs_next_leaf(root, path); | |
2744 | if (ret < 0) | |
6f519564 | 2745 | goto out_free_path; |
38c227d8 LB |
2746 | else if (ret > 0) |
2747 | break; | |
2748 | continue; | |
2749 | } | |
2750 | ||
2751 | btrfs_item_key_to_cpu(l, &key, slot); | |
2752 | ||
2753 | if (key.objectid != btrfs_ino(inode)) | |
2754 | break; | |
2755 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2756 | break; | |
2757 | if (key.offset >= new->file_pos + new->len) | |
2758 | break; | |
2759 | ||
2760 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2761 | ||
2762 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2763 | if (key.offset + num_bytes < new->file_pos) | |
2764 | goto next; | |
2765 | ||
2766 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2767 | if (!disk_bytenr) | |
2768 | goto next; | |
2769 | ||
2770 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2771 | ||
2772 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2773 | if (!old) | |
6f519564 | 2774 | goto out_free_path; |
38c227d8 LB |
2775 | |
2776 | offset = max(new->file_pos, key.offset); | |
2777 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2778 | ||
2779 | old->bytenr = disk_bytenr; | |
2780 | old->extent_offset = extent_offset; | |
2781 | old->offset = offset - key.offset; | |
2782 | old->len = end - offset; | |
2783 | old->new = new; | |
2784 | old->count = 0; | |
2785 | list_add_tail(&old->list, &new->head); | |
2786 | next: | |
2787 | path->slots[0]++; | |
2788 | cond_resched(); | |
2789 | } | |
2790 | ||
2791 | btrfs_free_path(path); | |
2792 | atomic_inc(&root->fs_info->defrag_running); | |
2793 | ||
2794 | return new; | |
2795 | ||
38c227d8 LB |
2796 | out_free_path: |
2797 | btrfs_free_path(path); | |
2798 | out_kfree: | |
6f519564 | 2799 | free_sa_defrag_extent(new); |
38c227d8 LB |
2800 | return NULL; |
2801 | } | |
2802 | ||
e570fd27 MX |
2803 | static void btrfs_release_delalloc_bytes(struct btrfs_root *root, |
2804 | u64 start, u64 len) | |
2805 | { | |
2806 | struct btrfs_block_group_cache *cache; | |
2807 | ||
2808 | cache = btrfs_lookup_block_group(root->fs_info, start); | |
2809 | ASSERT(cache); | |
2810 | ||
2811 | spin_lock(&cache->lock); | |
2812 | cache->delalloc_bytes -= len; | |
2813 | spin_unlock(&cache->lock); | |
2814 | ||
2815 | btrfs_put_block_group(cache); | |
2816 | } | |
2817 | ||
d352ac68 CM |
2818 | /* as ordered data IO finishes, this gets called so we can finish |
2819 | * an ordered extent if the range of bytes in the file it covers are | |
2820 | * fully written. | |
2821 | */ | |
5fd02043 | 2822 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2823 | { |
5fd02043 | 2824 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 2825 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2826 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2827 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2828 | struct extent_state *cached_state = NULL; |
38c227d8 | 2829 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2830 | int compress_type = 0; |
77cef2ec JB |
2831 | int ret = 0; |
2832 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2833 | bool nolock; |
77cef2ec | 2834 | bool truncated = false; |
e6dcd2dc | 2835 | |
83eea1f1 | 2836 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 2837 | |
5fd02043 JB |
2838 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2839 | ret = -EIO; | |
2840 | goto out; | |
2841 | } | |
2842 | ||
f612496b MX |
2843 | btrfs_free_io_failure_record(inode, ordered_extent->file_offset, |
2844 | ordered_extent->file_offset + | |
2845 | ordered_extent->len - 1); | |
2846 | ||
77cef2ec JB |
2847 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2848 | truncated = true; | |
2849 | logical_len = ordered_extent->truncated_len; | |
2850 | /* Truncated the entire extent, don't bother adding */ | |
2851 | if (!logical_len) | |
2852 | goto out; | |
2853 | } | |
2854 | ||
c2167754 | 2855 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2856 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a QW |
2857 | |
2858 | /* | |
2859 | * For mwrite(mmap + memset to write) case, we still reserve | |
2860 | * space for NOCOW range. | |
2861 | * As NOCOW won't cause a new delayed ref, just free the space | |
2862 | */ | |
2863 | btrfs_qgroup_free_data(inode, ordered_extent->file_offset, | |
2864 | ordered_extent->len); | |
6c760c07 JB |
2865 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2866 | if (nolock) | |
2867 | trans = btrfs_join_transaction_nolock(root); | |
2868 | else | |
2869 | trans = btrfs_join_transaction(root); | |
2870 | if (IS_ERR(trans)) { | |
2871 | ret = PTR_ERR(trans); | |
2872 | trans = NULL; | |
2873 | goto out; | |
c2167754 | 2874 | } |
6c760c07 JB |
2875 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2876 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2877 | if (ret) /* -ENOMEM or corruption */ | |
2878 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
2879 | goto out; |
2880 | } | |
e6dcd2dc | 2881 | |
2ac55d41 JB |
2882 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2883 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ff13db41 | 2884 | &cached_state); |
e6dcd2dc | 2885 | |
38c227d8 LB |
2886 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2887 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2888 | EXTENT_DEFRAG, 1, cached_state); | |
2889 | if (ret) { | |
2890 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 2891 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
2892 | /* the inode is shared */ |
2893 | new = record_old_file_extents(inode, ordered_extent); | |
2894 | ||
2895 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
2896 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2897 | EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS); | |
2898 | } | |
2899 | ||
0cb59c99 | 2900 | if (nolock) |
7a7eaa40 | 2901 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 2902 | else |
7a7eaa40 | 2903 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2904 | if (IS_ERR(trans)) { |
2905 | ret = PTR_ERR(trans); | |
2906 | trans = NULL; | |
2907 | goto out_unlock; | |
2908 | } | |
a79b7d4b | 2909 | |
0ca1f7ce | 2910 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 2911 | |
c8b97818 | 2912 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 2913 | compress_type = ordered_extent->compress_type; |
d899e052 | 2914 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 2915 | BUG_ON(compress_type); |
920bbbfb | 2916 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
2917 | ordered_extent->file_offset, |
2918 | ordered_extent->file_offset + | |
77cef2ec | 2919 | logical_len); |
d899e052 | 2920 | } else { |
0af3d00b | 2921 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
2922 | ret = insert_reserved_file_extent(trans, inode, |
2923 | ordered_extent->file_offset, | |
2924 | ordered_extent->start, | |
2925 | ordered_extent->disk_len, | |
77cef2ec | 2926 | logical_len, logical_len, |
261507a0 | 2927 | compress_type, 0, 0, |
d899e052 | 2928 | BTRFS_FILE_EXTENT_REG); |
e570fd27 MX |
2929 | if (!ret) |
2930 | btrfs_release_delalloc_bytes(root, | |
2931 | ordered_extent->start, | |
2932 | ordered_extent->disk_len); | |
d899e052 | 2933 | } |
5dc562c5 JB |
2934 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
2935 | ordered_extent->file_offset, ordered_extent->len, | |
2936 | trans->transid); | |
79787eaa JM |
2937 | if (ret < 0) { |
2938 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 2939 | goto out_unlock; |
79787eaa | 2940 | } |
2ac55d41 | 2941 | |
e6dcd2dc CM |
2942 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
2943 | &ordered_extent->list); | |
2944 | ||
6c760c07 JB |
2945 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2946 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2947 | if (ret) { /* -ENOMEM or corruption */ | |
2948 | btrfs_abort_transaction(trans, root, ret); | |
2949 | goto out_unlock; | |
1ef30be1 JB |
2950 | } |
2951 | ret = 0; | |
5fd02043 JB |
2952 | out_unlock: |
2953 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
2954 | ordered_extent->file_offset + | |
2955 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 2956 | out: |
5b0e95bf | 2957 | if (root != root->fs_info->tree_root) |
0cb59c99 | 2958 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 MX |
2959 | if (trans) |
2960 | btrfs_end_transaction(trans, root); | |
0cb59c99 | 2961 | |
77cef2ec JB |
2962 | if (ret || truncated) { |
2963 | u64 start, end; | |
2964 | ||
2965 | if (truncated) | |
2966 | start = ordered_extent->file_offset + logical_len; | |
2967 | else | |
2968 | start = ordered_extent->file_offset; | |
2969 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
2970 | clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS); | |
2971 | ||
2972 | /* Drop the cache for the part of the extent we didn't write. */ | |
2973 | btrfs_drop_extent_cache(inode, start, end, 0); | |
5fd02043 | 2974 | |
0bec9ef5 JB |
2975 | /* |
2976 | * If the ordered extent had an IOERR or something else went | |
2977 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
2978 | * back to the allocator. We only free the extent in the |
2979 | * truncated case if we didn't write out the extent at all. | |
0bec9ef5 | 2980 | */ |
77cef2ec JB |
2981 | if ((ret || !logical_len) && |
2982 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
0bec9ef5 JB |
2983 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
2984 | btrfs_free_reserved_extent(root, ordered_extent->start, | |
e570fd27 | 2985 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
2986 | } |
2987 | ||
2988 | ||
5fd02043 | 2989 | /* |
8bad3c02 LB |
2990 | * This needs to be done to make sure anybody waiting knows we are done |
2991 | * updating everything for this ordered extent. | |
5fd02043 JB |
2992 | */ |
2993 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
2994 | ||
38c227d8 | 2995 | /* for snapshot-aware defrag */ |
6f519564 LB |
2996 | if (new) { |
2997 | if (ret) { | |
2998 | free_sa_defrag_extent(new); | |
2999 | atomic_dec(&root->fs_info->defrag_running); | |
3000 | } else { | |
3001 | relink_file_extents(new); | |
3002 | } | |
3003 | } | |
38c227d8 | 3004 | |
e6dcd2dc CM |
3005 | /* once for us */ |
3006 | btrfs_put_ordered_extent(ordered_extent); | |
3007 | /* once for the tree */ | |
3008 | btrfs_put_ordered_extent(ordered_extent); | |
3009 | ||
5fd02043 JB |
3010 | return ret; |
3011 | } | |
3012 | ||
3013 | static void finish_ordered_fn(struct btrfs_work *work) | |
3014 | { | |
3015 | struct btrfs_ordered_extent *ordered_extent; | |
3016 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
3017 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
3018 | } |
3019 | ||
b2950863 | 3020 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
3021 | struct extent_state *state, int uptodate) |
3022 | { | |
5fd02043 JB |
3023 | struct inode *inode = page->mapping->host; |
3024 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3025 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
9e0af237 LB |
3026 | struct btrfs_workqueue *wq; |
3027 | btrfs_work_func_t func; | |
5fd02043 | 3028 | |
1abe9b8a | 3029 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
3030 | ||
8b62b72b | 3031 | ClearPagePrivate2(page); |
5fd02043 JB |
3032 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
3033 | end - start + 1, uptodate)) | |
3034 | return 0; | |
3035 | ||
9e0af237 LB |
3036 | if (btrfs_is_free_space_inode(inode)) { |
3037 | wq = root->fs_info->endio_freespace_worker; | |
3038 | func = btrfs_freespace_write_helper; | |
3039 | } else { | |
3040 | wq = root->fs_info->endio_write_workers; | |
3041 | func = btrfs_endio_write_helper; | |
3042 | } | |
5fd02043 | 3043 | |
9e0af237 LB |
3044 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
3045 | NULL); | |
3046 | btrfs_queue_work(wq, &ordered_extent->work); | |
5fd02043 JB |
3047 | |
3048 | return 0; | |
211f90e6 CM |
3049 | } |
3050 | ||
dc380aea MX |
3051 | static int __readpage_endio_check(struct inode *inode, |
3052 | struct btrfs_io_bio *io_bio, | |
3053 | int icsum, struct page *page, | |
3054 | int pgoff, u64 start, size_t len) | |
3055 | { | |
3056 | char *kaddr; | |
3057 | u32 csum_expected; | |
3058 | u32 csum = ~(u32)0; | |
dc380aea MX |
3059 | |
3060 | csum_expected = *(((u32 *)io_bio->csum) + icsum); | |
3061 | ||
3062 | kaddr = kmap_atomic(page); | |
3063 | csum = btrfs_csum_data(kaddr + pgoff, csum, len); | |
3064 | btrfs_csum_final(csum, (char *)&csum); | |
3065 | if (csum != csum_expected) | |
3066 | goto zeroit; | |
3067 | ||
3068 | kunmap_atomic(kaddr); | |
3069 | return 0; | |
3070 | zeroit: | |
94647322 DS |
3071 | btrfs_warn_rl(BTRFS_I(inode)->root->fs_info, |
3072 | "csum failed ino %llu off %llu csum %u expected csum %u", | |
dc380aea MX |
3073 | btrfs_ino(inode), start, csum, csum_expected); |
3074 | memset(kaddr + pgoff, 1, len); | |
3075 | flush_dcache_page(page); | |
3076 | kunmap_atomic(kaddr); | |
3077 | if (csum_expected == 0) | |
3078 | return 0; | |
3079 | return -EIO; | |
3080 | } | |
3081 | ||
d352ac68 CM |
3082 | /* |
3083 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
3084 | * if there's a match, we allow the bio to finish. If not, the code in |
3085 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 3086 | */ |
facc8a22 MX |
3087 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
3088 | u64 phy_offset, struct page *page, | |
3089 | u64 start, u64 end, int mirror) | |
07157aac | 3090 | { |
4eee4fa4 | 3091 | size_t offset = start - page_offset(page); |
07157aac | 3092 | struct inode *inode = page->mapping->host; |
d1310b2e | 3093 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3094 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 3095 | |
d20f7043 CM |
3096 | if (PageChecked(page)) { |
3097 | ClearPageChecked(page); | |
dc380aea | 3098 | return 0; |
d20f7043 | 3099 | } |
6cbff00f CH |
3100 | |
3101 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 3102 | return 0; |
17d217fe YZ |
3103 | |
3104 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 3105 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
3106 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
3107 | GFP_NOFS); | |
b6cda9bc | 3108 | return 0; |
17d217fe | 3109 | } |
d20f7043 | 3110 | |
facc8a22 | 3111 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
dc380aea MX |
3112 | return __readpage_endio_check(inode, io_bio, phy_offset, page, offset, |
3113 | start, (size_t)(end - start + 1)); | |
07157aac | 3114 | } |
b888db2b | 3115 | |
24bbcf04 YZ |
3116 | void btrfs_add_delayed_iput(struct inode *inode) |
3117 | { | |
3118 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
8089fe62 | 3119 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3120 | |
3121 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3122 | return; | |
3123 | ||
24bbcf04 | 3124 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3125 | if (binode->delayed_iput_count == 0) { |
3126 | ASSERT(list_empty(&binode->delayed_iput)); | |
3127 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
3128 | } else { | |
3129 | binode->delayed_iput_count++; | |
3130 | } | |
24bbcf04 YZ |
3131 | spin_unlock(&fs_info->delayed_iput_lock); |
3132 | } | |
3133 | ||
3134 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
3135 | { | |
24bbcf04 | 3136 | struct btrfs_fs_info *fs_info = root->fs_info; |
24bbcf04 | 3137 | |
24bbcf04 | 3138 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3139 | while (!list_empty(&fs_info->delayed_iputs)) { |
3140 | struct btrfs_inode *inode; | |
3141 | ||
3142 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3143 | struct btrfs_inode, delayed_iput); | |
3144 | if (inode->delayed_iput_count) { | |
3145 | inode->delayed_iput_count--; | |
3146 | list_move_tail(&inode->delayed_iput, | |
3147 | &fs_info->delayed_iputs); | |
3148 | } else { | |
3149 | list_del_init(&inode->delayed_iput); | |
3150 | } | |
3151 | spin_unlock(&fs_info->delayed_iput_lock); | |
3152 | iput(&inode->vfs_inode); | |
3153 | spin_lock(&fs_info->delayed_iput_lock); | |
24bbcf04 | 3154 | } |
8089fe62 | 3155 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3156 | } |
3157 | ||
d68fc57b | 3158 | /* |
42b2aa86 | 3159 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
3160 | * files in the subvolume, it removes orphan item and frees block_rsv |
3161 | * structure. | |
3162 | */ | |
3163 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
3164 | struct btrfs_root *root) | |
3165 | { | |
90290e19 | 3166 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
3167 | int ret; |
3168 | ||
8a35d95f | 3169 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
3170 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
3171 | return; | |
3172 | ||
90290e19 | 3173 | spin_lock(&root->orphan_lock); |
8a35d95f | 3174 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
3175 | spin_unlock(&root->orphan_lock); |
3176 | return; | |
3177 | } | |
3178 | ||
3179 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
3180 | spin_unlock(&root->orphan_lock); | |
3181 | return; | |
3182 | } | |
3183 | ||
3184 | block_rsv = root->orphan_block_rsv; | |
3185 | root->orphan_block_rsv = NULL; | |
3186 | spin_unlock(&root->orphan_lock); | |
3187 | ||
27cdeb70 | 3188 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) && |
d68fc57b YZ |
3189 | btrfs_root_refs(&root->root_item) > 0) { |
3190 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
3191 | root->root_key.objectid); | |
4ef31a45 JB |
3192 | if (ret) |
3193 | btrfs_abort_transaction(trans, root, ret); | |
3194 | else | |
27cdeb70 MX |
3195 | clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, |
3196 | &root->state); | |
d68fc57b YZ |
3197 | } |
3198 | ||
90290e19 JB |
3199 | if (block_rsv) { |
3200 | WARN_ON(block_rsv->size > 0); | |
3201 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
3202 | } |
3203 | } | |
3204 | ||
7b128766 JB |
3205 | /* |
3206 | * This creates an orphan entry for the given inode in case something goes | |
3207 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
3208 | * |
3209 | * NOTE: caller of this function should reserve 5 units of metadata for | |
3210 | * this function. | |
7b128766 JB |
3211 | */ |
3212 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
3213 | { | |
3214 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3215 | struct btrfs_block_rsv *block_rsv = NULL; |
3216 | int reserve = 0; | |
3217 | int insert = 0; | |
3218 | int ret; | |
7b128766 | 3219 | |
d68fc57b | 3220 | if (!root->orphan_block_rsv) { |
66d8f3dd | 3221 | block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
b532402e TI |
3222 | if (!block_rsv) |
3223 | return -ENOMEM; | |
d68fc57b | 3224 | } |
7b128766 | 3225 | |
d68fc57b YZ |
3226 | spin_lock(&root->orphan_lock); |
3227 | if (!root->orphan_block_rsv) { | |
3228 | root->orphan_block_rsv = block_rsv; | |
3229 | } else if (block_rsv) { | |
3230 | btrfs_free_block_rsv(root, block_rsv); | |
3231 | block_rsv = NULL; | |
7b128766 | 3232 | } |
7b128766 | 3233 | |
8a35d95f JB |
3234 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3235 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
3236 | #if 0 |
3237 | /* | |
3238 | * For proper ENOSPC handling, we should do orphan | |
3239 | * cleanup when mounting. But this introduces backward | |
3240 | * compatibility issue. | |
3241 | */ | |
3242 | if (!xchg(&root->orphan_item_inserted, 1)) | |
3243 | insert = 2; | |
3244 | else | |
3245 | insert = 1; | |
3246 | #endif | |
3247 | insert = 1; | |
321f0e70 | 3248 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
3249 | } |
3250 | ||
72ac3c0d JB |
3251 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3252 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3253 | reserve = 1; |
d68fc57b | 3254 | spin_unlock(&root->orphan_lock); |
7b128766 | 3255 | |
d68fc57b YZ |
3256 | /* grab metadata reservation from transaction handle */ |
3257 | if (reserve) { | |
3258 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 3259 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 3260 | } |
7b128766 | 3261 | |
d68fc57b YZ |
3262 | /* insert an orphan item to track this unlinked/truncated file */ |
3263 | if (insert >= 1) { | |
33345d01 | 3264 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
4ef31a45 | 3265 | if (ret) { |
703c88e0 | 3266 | atomic_dec(&root->orphan_inodes); |
4ef31a45 JB |
3267 | if (reserve) { |
3268 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
3269 | &BTRFS_I(inode)->runtime_flags); | |
3270 | btrfs_orphan_release_metadata(inode); | |
3271 | } | |
3272 | if (ret != -EEXIST) { | |
e8e7cff6 JB |
3273 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3274 | &BTRFS_I(inode)->runtime_flags); | |
4ef31a45 JB |
3275 | btrfs_abort_transaction(trans, root, ret); |
3276 | return ret; | |
3277 | } | |
79787eaa JM |
3278 | } |
3279 | ret = 0; | |
d68fc57b YZ |
3280 | } |
3281 | ||
3282 | /* insert an orphan item to track subvolume contains orphan files */ | |
3283 | if (insert >= 2) { | |
3284 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
3285 | root->root_key.objectid); | |
79787eaa JM |
3286 | if (ret && ret != -EEXIST) { |
3287 | btrfs_abort_transaction(trans, root, ret); | |
3288 | return ret; | |
3289 | } | |
d68fc57b YZ |
3290 | } |
3291 | return 0; | |
7b128766 JB |
3292 | } |
3293 | ||
3294 | /* | |
3295 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
3296 | * item for this particular inode. | |
3297 | */ | |
48a3b636 ES |
3298 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3299 | struct inode *inode) | |
7b128766 JB |
3300 | { |
3301 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3302 | int delete_item = 0; |
3303 | int release_rsv = 0; | |
7b128766 JB |
3304 | int ret = 0; |
3305 | ||
d68fc57b | 3306 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
3307 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3308 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3309 | delete_item = 1; |
7b128766 | 3310 | |
72ac3c0d JB |
3311 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3312 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3313 | release_rsv = 1; |
d68fc57b | 3314 | spin_unlock(&root->orphan_lock); |
7b128766 | 3315 | |
703c88e0 | 3316 | if (delete_item) { |
8a35d95f | 3317 | atomic_dec(&root->orphan_inodes); |
703c88e0 FDBM |
3318 | if (trans) |
3319 | ret = btrfs_del_orphan_item(trans, root, | |
3320 | btrfs_ino(inode)); | |
8a35d95f | 3321 | } |
7b128766 | 3322 | |
703c88e0 FDBM |
3323 | if (release_rsv) |
3324 | btrfs_orphan_release_metadata(inode); | |
3325 | ||
4ef31a45 | 3326 | return ret; |
7b128766 JB |
3327 | } |
3328 | ||
3329 | /* | |
3330 | * this cleans up any orphans that may be left on the list from the last use | |
3331 | * of this root. | |
3332 | */ | |
66b4ffd1 | 3333 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
3334 | { |
3335 | struct btrfs_path *path; | |
3336 | struct extent_buffer *leaf; | |
7b128766 JB |
3337 | struct btrfs_key key, found_key; |
3338 | struct btrfs_trans_handle *trans; | |
3339 | struct inode *inode; | |
8f6d7f4f | 3340 | u64 last_objectid = 0; |
7b128766 JB |
3341 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3342 | ||
d68fc57b | 3343 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3344 | return 0; |
c71bf099 YZ |
3345 | |
3346 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3347 | if (!path) { |
3348 | ret = -ENOMEM; | |
3349 | goto out; | |
3350 | } | |
e4058b54 | 3351 | path->reada = READA_BACK; |
7b128766 JB |
3352 | |
3353 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3354 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3355 | key.offset = (u64)-1; |
3356 | ||
7b128766 JB |
3357 | while (1) { |
3358 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3359 | if (ret < 0) |
3360 | goto out; | |
7b128766 JB |
3361 | |
3362 | /* | |
3363 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3364 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3365 | * find the key and see if we have stuff that matches |
3366 | */ | |
3367 | if (ret > 0) { | |
66b4ffd1 | 3368 | ret = 0; |
7b128766 JB |
3369 | if (path->slots[0] == 0) |
3370 | break; | |
3371 | path->slots[0]--; | |
3372 | } | |
3373 | ||
3374 | /* pull out the item */ | |
3375 | leaf = path->nodes[0]; | |
7b128766 JB |
3376 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3377 | ||
3378 | /* make sure the item matches what we want */ | |
3379 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3380 | break; | |
962a298f | 3381 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3382 | break; |
3383 | ||
3384 | /* release the path since we're done with it */ | |
b3b4aa74 | 3385 | btrfs_release_path(path); |
7b128766 JB |
3386 | |
3387 | /* | |
3388 | * this is where we are basically btrfs_lookup, without the | |
3389 | * crossing root thing. we store the inode number in the | |
3390 | * offset of the orphan item. | |
3391 | */ | |
8f6d7f4f JB |
3392 | |
3393 | if (found_key.offset == last_objectid) { | |
c2cf52eb SK |
3394 | btrfs_err(root->fs_info, |
3395 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3396 | ret = -EINVAL; |
3397 | goto out; | |
3398 | } | |
3399 | ||
3400 | last_objectid = found_key.offset; | |
3401 | ||
5d4f98a2 YZ |
3402 | found_key.objectid = found_key.offset; |
3403 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3404 | found_key.offset = 0; | |
73f73415 | 3405 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3406 | ret = PTR_ERR_OR_ZERO(inode); |
a8c9e576 | 3407 | if (ret && ret != -ESTALE) |
66b4ffd1 | 3408 | goto out; |
7b128766 | 3409 | |
f8e9e0b0 AJ |
3410 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
3411 | struct btrfs_root *dead_root; | |
3412 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3413 | int is_dead_root = 0; | |
3414 | ||
3415 | /* | |
3416 | * this is an orphan in the tree root. Currently these | |
3417 | * could come from 2 sources: | |
3418 | * a) a snapshot deletion in progress | |
3419 | * b) a free space cache inode | |
3420 | * We need to distinguish those two, as the snapshot | |
3421 | * orphan must not get deleted. | |
3422 | * find_dead_roots already ran before us, so if this | |
3423 | * is a snapshot deletion, we should find the root | |
3424 | * in the dead_roots list | |
3425 | */ | |
3426 | spin_lock(&fs_info->trans_lock); | |
3427 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3428 | root_list) { | |
3429 | if (dead_root->root_key.objectid == | |
3430 | found_key.objectid) { | |
3431 | is_dead_root = 1; | |
3432 | break; | |
3433 | } | |
3434 | } | |
3435 | spin_unlock(&fs_info->trans_lock); | |
3436 | if (is_dead_root) { | |
3437 | /* prevent this orphan from being found again */ | |
3438 | key.offset = found_key.objectid - 1; | |
3439 | continue; | |
3440 | } | |
3441 | } | |
7b128766 | 3442 | /* |
a8c9e576 JB |
3443 | * Inode is already gone but the orphan item is still there, |
3444 | * kill the orphan item. | |
7b128766 | 3445 | */ |
a8c9e576 JB |
3446 | if (ret == -ESTALE) { |
3447 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
3448 | if (IS_ERR(trans)) { |
3449 | ret = PTR_ERR(trans); | |
3450 | goto out; | |
3451 | } | |
c2cf52eb SK |
3452 | btrfs_debug(root->fs_info, "auto deleting %Lu", |
3453 | found_key.objectid); | |
a8c9e576 JB |
3454 | ret = btrfs_del_orphan_item(trans, root, |
3455 | found_key.objectid); | |
5b21f2ed | 3456 | btrfs_end_transaction(trans, root); |
4ef31a45 JB |
3457 | if (ret) |
3458 | goto out; | |
7b128766 JB |
3459 | continue; |
3460 | } | |
3461 | ||
a8c9e576 JB |
3462 | /* |
3463 | * add this inode to the orphan list so btrfs_orphan_del does | |
3464 | * the proper thing when we hit it | |
3465 | */ | |
8a35d95f JB |
3466 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3467 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3468 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3469 | |
7b128766 JB |
3470 | /* if we have links, this was a truncate, lets do that */ |
3471 | if (inode->i_nlink) { | |
fae7f21c | 3472 | if (WARN_ON(!S_ISREG(inode->i_mode))) { |
a41ad394 JB |
3473 | iput(inode); |
3474 | continue; | |
3475 | } | |
7b128766 | 3476 | nr_truncate++; |
f3fe820c JB |
3477 | |
3478 | /* 1 for the orphan item deletion. */ | |
3479 | trans = btrfs_start_transaction(root, 1); | |
3480 | if (IS_ERR(trans)) { | |
c69b26b0 | 3481 | iput(inode); |
f3fe820c JB |
3482 | ret = PTR_ERR(trans); |
3483 | goto out; | |
3484 | } | |
3485 | ret = btrfs_orphan_add(trans, inode); | |
3486 | btrfs_end_transaction(trans, root); | |
c69b26b0 JB |
3487 | if (ret) { |
3488 | iput(inode); | |
f3fe820c | 3489 | goto out; |
c69b26b0 | 3490 | } |
f3fe820c | 3491 | |
66b4ffd1 | 3492 | ret = btrfs_truncate(inode); |
4a7d0f68 JB |
3493 | if (ret) |
3494 | btrfs_orphan_del(NULL, inode); | |
7b128766 JB |
3495 | } else { |
3496 | nr_unlink++; | |
3497 | } | |
3498 | ||
3499 | /* this will do delete_inode and everything for us */ | |
3500 | iput(inode); | |
66b4ffd1 JB |
3501 | if (ret) |
3502 | goto out; | |
7b128766 | 3503 | } |
3254c876 MX |
3504 | /* release the path since we're done with it */ |
3505 | btrfs_release_path(path); | |
3506 | ||
d68fc57b YZ |
3507 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3508 | ||
3509 | if (root->orphan_block_rsv) | |
3510 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
3511 | (u64)-1); | |
3512 | ||
27cdeb70 MX |
3513 | if (root->orphan_block_rsv || |
3514 | test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { | |
7a7eaa40 | 3515 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
3516 | if (!IS_ERR(trans)) |
3517 | btrfs_end_transaction(trans, root); | |
d68fc57b | 3518 | } |
7b128766 JB |
3519 | |
3520 | if (nr_unlink) | |
4884b476 | 3521 | btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3522 | if (nr_truncate) |
4884b476 | 3523 | btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3524 | |
3525 | out: | |
3526 | if (ret) | |
68b663d1 | 3527 | btrfs_err(root->fs_info, |
c2cf52eb | 3528 | "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3529 | btrfs_free_path(path); |
3530 | return ret; | |
7b128766 JB |
3531 | } |
3532 | ||
46a53cca CM |
3533 | /* |
3534 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3535 | * don't find any xattrs, we know there can't be any acls. | |
3536 | * | |
3537 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3538 | */ | |
3539 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3540 | int slot, u64 objectid, |
3541 | int *first_xattr_slot) | |
46a53cca CM |
3542 | { |
3543 | u32 nritems = btrfs_header_nritems(leaf); | |
3544 | struct btrfs_key found_key; | |
f23b5a59 JB |
3545 | static u64 xattr_access = 0; |
3546 | static u64 xattr_default = 0; | |
46a53cca CM |
3547 | int scanned = 0; |
3548 | ||
f23b5a59 | 3549 | if (!xattr_access) { |
97d79299 AG |
3550 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3551 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3552 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3553 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3554 | } |
3555 | ||
46a53cca | 3556 | slot++; |
63541927 | 3557 | *first_xattr_slot = -1; |
46a53cca CM |
3558 | while (slot < nritems) { |
3559 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3560 | ||
3561 | /* we found a different objectid, there must not be acls */ | |
3562 | if (found_key.objectid != objectid) | |
3563 | return 0; | |
3564 | ||
3565 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3566 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3567 | if (*first_xattr_slot == -1) |
3568 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3569 | if (found_key.offset == xattr_access || |
3570 | found_key.offset == xattr_default) | |
3571 | return 1; | |
3572 | } | |
46a53cca CM |
3573 | |
3574 | /* | |
3575 | * we found a key greater than an xattr key, there can't | |
3576 | * be any acls later on | |
3577 | */ | |
3578 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3579 | return 0; | |
3580 | ||
3581 | slot++; | |
3582 | scanned++; | |
3583 | ||
3584 | /* | |
3585 | * it goes inode, inode backrefs, xattrs, extents, | |
3586 | * so if there are a ton of hard links to an inode there can | |
3587 | * be a lot of backrefs. Don't waste time searching too hard, | |
3588 | * this is just an optimization | |
3589 | */ | |
3590 | if (scanned >= 8) | |
3591 | break; | |
3592 | } | |
3593 | /* we hit the end of the leaf before we found an xattr or | |
3594 | * something larger than an xattr. We have to assume the inode | |
3595 | * has acls | |
3596 | */ | |
63541927 FDBM |
3597 | if (*first_xattr_slot == -1) |
3598 | *first_xattr_slot = slot; | |
46a53cca CM |
3599 | return 1; |
3600 | } | |
3601 | ||
d352ac68 CM |
3602 | /* |
3603 | * read an inode from the btree into the in-memory inode | |
3604 | */ | |
5d4f98a2 | 3605 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
3606 | { |
3607 | struct btrfs_path *path; | |
5f39d397 | 3608 | struct extent_buffer *leaf; |
39279cc3 CM |
3609 | struct btrfs_inode_item *inode_item; |
3610 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3611 | struct btrfs_key location; | |
67de1176 | 3612 | unsigned long ptr; |
46a53cca | 3613 | int maybe_acls; |
618e21d5 | 3614 | u32 rdev; |
39279cc3 | 3615 | int ret; |
2f7e33d4 | 3616 | bool filled = false; |
63541927 | 3617 | int first_xattr_slot; |
2f7e33d4 MX |
3618 | |
3619 | ret = btrfs_fill_inode(inode, &rdev); | |
3620 | if (!ret) | |
3621 | filled = true; | |
39279cc3 CM |
3622 | |
3623 | path = btrfs_alloc_path(); | |
1748f843 MF |
3624 | if (!path) |
3625 | goto make_bad; | |
3626 | ||
39279cc3 | 3627 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3628 | |
39279cc3 | 3629 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 3630 | if (ret) |
39279cc3 | 3631 | goto make_bad; |
39279cc3 | 3632 | |
5f39d397 | 3633 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3634 | |
3635 | if (filled) | |
67de1176 | 3636 | goto cache_index; |
2f7e33d4 | 3637 | |
5f39d397 CM |
3638 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3639 | struct btrfs_inode_item); | |
5f39d397 | 3640 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3641 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3642 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3643 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
dbe674a9 | 3644 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 | 3645 | |
a937b979 DS |
3646 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3647 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3648 | |
a937b979 DS |
3649 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3650 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3651 | |
a937b979 DS |
3652 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3653 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3654 | |
9cc97d64 | 3655 | BTRFS_I(inode)->i_otime.tv_sec = |
3656 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3657 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3658 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3659 | |
a76a3cd4 | 3660 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3661 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3662 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3663 | ||
6e17d30b YD |
3664 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
3665 | inode->i_generation = BTRFS_I(inode)->generation; | |
3666 | inode->i_rdev = 0; | |
3667 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3668 | ||
3669 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3670 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3671 | ||
3672 | cache_index: | |
5dc562c5 JB |
3673 | /* |
3674 | * If we were modified in the current generation and evicted from memory | |
3675 | * and then re-read we need to do a full sync since we don't have any | |
3676 | * idea about which extents were modified before we were evicted from | |
3677 | * cache. | |
6e17d30b YD |
3678 | * |
3679 | * This is required for both inode re-read from disk and delayed inode | |
3680 | * in delayed_nodes_tree. | |
5dc562c5 JB |
3681 | */ |
3682 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
3683 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3684 | &BTRFS_I(inode)->runtime_flags); | |
3685 | ||
bde6c242 FM |
3686 | /* |
3687 | * We don't persist the id of the transaction where an unlink operation | |
3688 | * against the inode was last made. So here we assume the inode might | |
3689 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3690 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3691 | * between the inode and its parent if the inode is fsync'ed and the log | |
3692 | * replayed. For example, in the scenario: | |
3693 | * | |
3694 | * touch mydir/foo | |
3695 | * ln mydir/foo mydir/bar | |
3696 | * sync | |
3697 | * unlink mydir/bar | |
3698 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3699 | * xfs_io -c fsync mydir/foo | |
3700 | * <power failure> | |
3701 | * mount fs, triggers fsync log replay | |
3702 | * | |
3703 | * We must make sure that when we fsync our inode foo we also log its | |
3704 | * parent inode, otherwise after log replay the parent still has the | |
3705 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3706 | * and doesn't have an inode ref with the name "bar" anymore. | |
3707 | * | |
3708 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
3709 | * but it guarantees correctness at the expense of ocassional full | |
3710 | * transaction commits on fsync if our inode is a directory, or if our | |
3711 | * inode is not a directory, logging its parent unnecessarily. | |
3712 | */ | |
3713 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3714 | ||
67de1176 MX |
3715 | path->slots[0]++; |
3716 | if (inode->i_nlink != 1 || | |
3717 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3718 | goto cache_acl; | |
3719 | ||
3720 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
3721 | if (location.objectid != btrfs_ino(inode)) | |
3722 | goto cache_acl; | |
3723 | ||
3724 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3725 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3726 | struct btrfs_inode_ref *ref; | |
3727 | ||
3728 | ref = (struct btrfs_inode_ref *)ptr; | |
3729 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3730 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3731 | struct btrfs_inode_extref *extref; | |
3732 | ||
3733 | extref = (struct btrfs_inode_extref *)ptr; | |
3734 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3735 | extref); | |
3736 | } | |
2f7e33d4 | 3737 | cache_acl: |
46a53cca CM |
3738 | /* |
3739 | * try to precache a NULL acl entry for files that don't have | |
3740 | * any xattrs or acls | |
3741 | */ | |
33345d01 | 3742 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
63541927 FDBM |
3743 | btrfs_ino(inode), &first_xattr_slot); |
3744 | if (first_xattr_slot != -1) { | |
3745 | path->slots[0] = first_xattr_slot; | |
3746 | ret = btrfs_load_inode_props(inode, path); | |
3747 | if (ret) | |
3748 | btrfs_err(root->fs_info, | |
351fd353 | 3749 | "error loading props for ino %llu (root %llu): %d", |
63541927 FDBM |
3750 | btrfs_ino(inode), |
3751 | root->root_key.objectid, ret); | |
3752 | } | |
3753 | btrfs_free_path(path); | |
3754 | ||
72c04902 AV |
3755 | if (!maybe_acls) |
3756 | cache_no_acl(inode); | |
46a53cca | 3757 | |
39279cc3 | 3758 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3759 | case S_IFREG: |
3760 | inode->i_mapping->a_ops = &btrfs_aops; | |
d1310b2e | 3761 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3762 | inode->i_fop = &btrfs_file_operations; |
3763 | inode->i_op = &btrfs_file_inode_operations; | |
3764 | break; | |
3765 | case S_IFDIR: | |
3766 | inode->i_fop = &btrfs_dir_file_operations; | |
3767 | if (root == root->fs_info->tree_root) | |
3768 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
3769 | else | |
3770 | inode->i_op = &btrfs_dir_inode_operations; | |
3771 | break; | |
3772 | case S_IFLNK: | |
3773 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3774 | inode_nohighmem(inode); |
39279cc3 CM |
3775 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
3776 | break; | |
618e21d5 | 3777 | default: |
0279b4cd | 3778 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3779 | init_special_inode(inode, inode->i_mode, rdev); |
3780 | break; | |
39279cc3 | 3781 | } |
6cbff00f CH |
3782 | |
3783 | btrfs_update_iflags(inode); | |
39279cc3 CM |
3784 | return; |
3785 | ||
3786 | make_bad: | |
39279cc3 | 3787 | btrfs_free_path(path); |
39279cc3 CM |
3788 | make_bad_inode(inode); |
3789 | } | |
3790 | ||
d352ac68 CM |
3791 | /* |
3792 | * given a leaf and an inode, copy the inode fields into the leaf | |
3793 | */ | |
e02119d5 CM |
3794 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3795 | struct extent_buffer *leaf, | |
5f39d397 | 3796 | struct btrfs_inode_item *item, |
39279cc3 CM |
3797 | struct inode *inode) |
3798 | { | |
51fab693 LB |
3799 | struct btrfs_map_token token; |
3800 | ||
3801 | btrfs_init_map_token(&token); | |
5f39d397 | 3802 | |
51fab693 LB |
3803 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3804 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3805 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3806 | &token); | |
3807 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3808 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3809 | |
a937b979 | 3810 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
51fab693 | 3811 | inode->i_atime.tv_sec, &token); |
a937b979 | 3812 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
51fab693 | 3813 | inode->i_atime.tv_nsec, &token); |
5f39d397 | 3814 | |
a937b979 | 3815 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
51fab693 | 3816 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3817 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
51fab693 | 3818 | inode->i_mtime.tv_nsec, &token); |
5f39d397 | 3819 | |
a937b979 | 3820 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
51fab693 | 3821 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3822 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
51fab693 | 3823 | inode->i_ctime.tv_nsec, &token); |
5f39d397 | 3824 | |
9cc97d64 | 3825 | btrfs_set_token_timespec_sec(leaf, &item->otime, |
3826 | BTRFS_I(inode)->i_otime.tv_sec, &token); | |
3827 | btrfs_set_token_timespec_nsec(leaf, &item->otime, | |
3828 | BTRFS_I(inode)->i_otime.tv_nsec, &token); | |
3829 | ||
51fab693 LB |
3830 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3831 | &token); | |
3832 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3833 | &token); | |
3834 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3835 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3836 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3837 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3838 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3839 | } |
3840 | ||
d352ac68 CM |
3841 | /* |
3842 | * copy everything in the in-memory inode into the btree. | |
3843 | */ | |
2115133f | 3844 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3845 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3846 | { |
3847 | struct btrfs_inode_item *inode_item; | |
3848 | struct btrfs_path *path; | |
5f39d397 | 3849 | struct extent_buffer *leaf; |
39279cc3 CM |
3850 | int ret; |
3851 | ||
3852 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3853 | if (!path) |
3854 | return -ENOMEM; | |
3855 | ||
b9473439 | 3856 | path->leave_spinning = 1; |
16cdcec7 MX |
3857 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3858 | 1); | |
39279cc3 CM |
3859 | if (ret) { |
3860 | if (ret > 0) | |
3861 | ret = -ENOENT; | |
3862 | goto failed; | |
3863 | } | |
3864 | ||
5f39d397 CM |
3865 | leaf = path->nodes[0]; |
3866 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3867 | struct btrfs_inode_item); |
39279cc3 | 3868 | |
e02119d5 | 3869 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3870 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3871 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3872 | ret = 0; |
3873 | failed: | |
39279cc3 CM |
3874 | btrfs_free_path(path); |
3875 | return ret; | |
3876 | } | |
3877 | ||
2115133f CM |
3878 | /* |
3879 | * copy everything in the in-memory inode into the btree. | |
3880 | */ | |
3881 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3882 | struct btrfs_root *root, struct inode *inode) | |
3883 | { | |
3884 | int ret; | |
3885 | ||
3886 | /* | |
3887 | * If the inode is a free space inode, we can deadlock during commit | |
3888 | * if we put it into the delayed code. | |
3889 | * | |
3890 | * The data relocation inode should also be directly updated | |
3891 | * without delay | |
3892 | */ | |
83eea1f1 | 3893 | if (!btrfs_is_free_space_inode(inode) |
1d52c78a JB |
3894 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
3895 | && !root->fs_info->log_root_recovering) { | |
8ea05e3a AB |
3896 | btrfs_update_root_times(trans, root); |
3897 | ||
2115133f CM |
3898 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3899 | if (!ret) | |
3900 | btrfs_set_inode_last_trans(trans, inode); | |
3901 | return ret; | |
3902 | } | |
3903 | ||
3904 | return btrfs_update_inode_item(trans, root, inode); | |
3905 | } | |
3906 | ||
be6aef60 JB |
3907 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3908 | struct btrfs_root *root, | |
3909 | struct inode *inode) | |
2115133f CM |
3910 | { |
3911 | int ret; | |
3912 | ||
3913 | ret = btrfs_update_inode(trans, root, inode); | |
3914 | if (ret == -ENOSPC) | |
3915 | return btrfs_update_inode_item(trans, root, inode); | |
3916 | return ret; | |
3917 | } | |
3918 | ||
d352ac68 CM |
3919 | /* |
3920 | * unlink helper that gets used here in inode.c and in the tree logging | |
3921 | * recovery code. It remove a link in a directory with a given name, and | |
3922 | * also drops the back refs in the inode to the directory | |
3923 | */ | |
92986796 AV |
3924 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3925 | struct btrfs_root *root, | |
3926 | struct inode *dir, struct inode *inode, | |
3927 | const char *name, int name_len) | |
39279cc3 CM |
3928 | { |
3929 | struct btrfs_path *path; | |
39279cc3 | 3930 | int ret = 0; |
5f39d397 | 3931 | struct extent_buffer *leaf; |
39279cc3 | 3932 | struct btrfs_dir_item *di; |
5f39d397 | 3933 | struct btrfs_key key; |
aec7477b | 3934 | u64 index; |
33345d01 LZ |
3935 | u64 ino = btrfs_ino(inode); |
3936 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3937 | |
3938 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3939 | if (!path) { |
3940 | ret = -ENOMEM; | |
554233a6 | 3941 | goto out; |
54aa1f4d CM |
3942 | } |
3943 | ||
b9473439 | 3944 | path->leave_spinning = 1; |
33345d01 | 3945 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
3946 | name, name_len, -1); |
3947 | if (IS_ERR(di)) { | |
3948 | ret = PTR_ERR(di); | |
3949 | goto err; | |
3950 | } | |
3951 | if (!di) { | |
3952 | ret = -ENOENT; | |
3953 | goto err; | |
3954 | } | |
5f39d397 CM |
3955 | leaf = path->nodes[0]; |
3956 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 3957 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3958 | if (ret) |
3959 | goto err; | |
b3b4aa74 | 3960 | btrfs_release_path(path); |
39279cc3 | 3961 | |
67de1176 MX |
3962 | /* |
3963 | * If we don't have dir index, we have to get it by looking up | |
3964 | * the inode ref, since we get the inode ref, remove it directly, | |
3965 | * it is unnecessary to do delayed deletion. | |
3966 | * | |
3967 | * But if we have dir index, needn't search inode ref to get it. | |
3968 | * Since the inode ref is close to the inode item, it is better | |
3969 | * that we delay to delete it, and just do this deletion when | |
3970 | * we update the inode item. | |
3971 | */ | |
3972 | if (BTRFS_I(inode)->dir_index) { | |
3973 | ret = btrfs_delayed_delete_inode_ref(inode); | |
3974 | if (!ret) { | |
3975 | index = BTRFS_I(inode)->dir_index; | |
3976 | goto skip_backref; | |
3977 | } | |
3978 | } | |
3979 | ||
33345d01 LZ |
3980 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
3981 | dir_ino, &index); | |
aec7477b | 3982 | if (ret) { |
c2cf52eb SK |
3983 | btrfs_info(root->fs_info, |
3984 | "failed to delete reference to %.*s, inode %llu parent %llu", | |
c1c9ff7c | 3985 | name_len, name, ino, dir_ino); |
79787eaa | 3986 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
3987 | goto err; |
3988 | } | |
67de1176 | 3989 | skip_backref: |
16cdcec7 | 3990 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3991 | if (ret) { |
3992 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 3993 | goto err; |
79787eaa | 3994 | } |
39279cc3 | 3995 | |
e02119d5 | 3996 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 3997 | inode, dir_ino); |
79787eaa JM |
3998 | if (ret != 0 && ret != -ENOENT) { |
3999 | btrfs_abort_transaction(trans, root, ret); | |
4000 | goto err; | |
4001 | } | |
e02119d5 CM |
4002 | |
4003 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
4004 | dir, index); | |
6418c961 CM |
4005 | if (ret == -ENOENT) |
4006 | ret = 0; | |
d4e3991b ZB |
4007 | else if (ret) |
4008 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 CM |
4009 | err: |
4010 | btrfs_free_path(path); | |
e02119d5 CM |
4011 | if (ret) |
4012 | goto out; | |
4013 | ||
4014 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
4015 | inode_inc_iversion(inode); |
4016 | inode_inc_iversion(dir); | |
04b285f3 DD |
4017 | inode->i_ctime = dir->i_mtime = |
4018 | dir->i_ctime = current_fs_time(inode->i_sb); | |
b9959295 | 4019 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 4020 | out: |
39279cc3 CM |
4021 | return ret; |
4022 | } | |
4023 | ||
92986796 AV |
4024 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4025 | struct btrfs_root *root, | |
4026 | struct inode *dir, struct inode *inode, | |
4027 | const char *name, int name_len) | |
4028 | { | |
4029 | int ret; | |
4030 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
4031 | if (!ret) { | |
8b558c5f | 4032 | drop_nlink(inode); |
92986796 AV |
4033 | ret = btrfs_update_inode(trans, root, inode); |
4034 | } | |
4035 | return ret; | |
4036 | } | |
39279cc3 | 4037 | |
a22285a6 YZ |
4038 | /* |
4039 | * helper to start transaction for unlink and rmdir. | |
4040 | * | |
d52be818 JB |
4041 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4042 | * if we cannot make our reservations the normal way try and see if there is | |
4043 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4044 | * allow the unlink to occur. | |
a22285a6 | 4045 | */ |
d52be818 | 4046 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4047 | { |
a22285a6 | 4048 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 4049 | |
e70bea5f JB |
4050 | /* |
4051 | * 1 for the possible orphan item | |
4052 | * 1 for the dir item | |
4053 | * 1 for the dir index | |
4054 | * 1 for the inode ref | |
e70bea5f JB |
4055 | * 1 for the inode |
4056 | */ | |
8eab77ff | 4057 | return btrfs_start_transaction_fallback_global_rsv(root, 5, 5); |
a22285a6 YZ |
4058 | } |
4059 | ||
4060 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4061 | { | |
4062 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4063 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 4064 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4065 | int ret; |
a22285a6 | 4066 | |
d52be818 | 4067 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4068 | if (IS_ERR(trans)) |
4069 | return PTR_ERR(trans); | |
5f39d397 | 4070 | |
2b0143b5 | 4071 | btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0); |
12fcfd22 | 4072 | |
2b0143b5 | 4073 | ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry), |
e02119d5 | 4074 | dentry->d_name.name, dentry->d_name.len); |
b532402e TI |
4075 | if (ret) |
4076 | goto out; | |
7b128766 | 4077 | |
a22285a6 | 4078 | if (inode->i_nlink == 0) { |
7b128766 | 4079 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
4080 | if (ret) |
4081 | goto out; | |
a22285a6 | 4082 | } |
7b128766 | 4083 | |
b532402e | 4084 | out: |
d52be818 | 4085 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4086 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
4087 | return ret; |
4088 | } | |
4089 | ||
4df27c4d YZ |
4090 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
4091 | struct btrfs_root *root, | |
4092 | struct inode *dir, u64 objectid, | |
4093 | const char *name, int name_len) | |
4094 | { | |
4095 | struct btrfs_path *path; | |
4096 | struct extent_buffer *leaf; | |
4097 | struct btrfs_dir_item *di; | |
4098 | struct btrfs_key key; | |
4099 | u64 index; | |
4100 | int ret; | |
33345d01 | 4101 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
4102 | |
4103 | path = btrfs_alloc_path(); | |
4104 | if (!path) | |
4105 | return -ENOMEM; | |
4106 | ||
33345d01 | 4107 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4108 | name, name_len, -1); |
79787eaa JM |
4109 | if (IS_ERR_OR_NULL(di)) { |
4110 | if (!di) | |
4111 | ret = -ENOENT; | |
4112 | else | |
4113 | ret = PTR_ERR(di); | |
4114 | goto out; | |
4115 | } | |
4df27c4d YZ |
4116 | |
4117 | leaf = path->nodes[0]; | |
4118 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4119 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4120 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
4121 | if (ret) { |
4122 | btrfs_abort_transaction(trans, root, ret); | |
4123 | goto out; | |
4124 | } | |
b3b4aa74 | 4125 | btrfs_release_path(path); |
4df27c4d YZ |
4126 | |
4127 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
4128 | objectid, root->root_key.objectid, | |
33345d01 | 4129 | dir_ino, &index, name, name_len); |
4df27c4d | 4130 | if (ret < 0) { |
79787eaa JM |
4131 | if (ret != -ENOENT) { |
4132 | btrfs_abort_transaction(trans, root, ret); | |
4133 | goto out; | |
4134 | } | |
33345d01 | 4135 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4136 | name, name_len); |
79787eaa JM |
4137 | if (IS_ERR_OR_NULL(di)) { |
4138 | if (!di) | |
4139 | ret = -ENOENT; | |
4140 | else | |
4141 | ret = PTR_ERR(di); | |
4142 | btrfs_abort_transaction(trans, root, ret); | |
4143 | goto out; | |
4144 | } | |
4df27c4d YZ |
4145 | |
4146 | leaf = path->nodes[0]; | |
4147 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 4148 | btrfs_release_path(path); |
4df27c4d YZ |
4149 | index = key.offset; |
4150 | } | |
945d8962 | 4151 | btrfs_release_path(path); |
4df27c4d | 4152 | |
16cdcec7 | 4153 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
4154 | if (ret) { |
4155 | btrfs_abort_transaction(trans, root, ret); | |
4156 | goto out; | |
4157 | } | |
4df27c4d YZ |
4158 | |
4159 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 4160 | inode_inc_iversion(dir); |
04b285f3 | 4161 | dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb); |
5a24e84c | 4162 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
4163 | if (ret) |
4164 | btrfs_abort_transaction(trans, root, ret); | |
4165 | out: | |
71d7aed0 | 4166 | btrfs_free_path(path); |
79787eaa | 4167 | return ret; |
4df27c4d YZ |
4168 | } |
4169 | ||
39279cc3 CM |
4170 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4171 | { | |
2b0143b5 | 4172 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4173 | int err = 0; |
39279cc3 | 4174 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4175 | struct btrfs_trans_handle *trans; |
39279cc3 | 4176 | |
b3ae244e | 4177 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4178 | return -ENOTEMPTY; |
b3ae244e DS |
4179 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
4180 | return -EPERM; | |
134d4512 | 4181 | |
d52be818 | 4182 | trans = __unlink_start_trans(dir); |
a22285a6 | 4183 | if (IS_ERR(trans)) |
5df6a9f6 | 4184 | return PTR_ERR(trans); |
5df6a9f6 | 4185 | |
33345d01 | 4186 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
4187 | err = btrfs_unlink_subvol(trans, root, dir, |
4188 | BTRFS_I(inode)->location.objectid, | |
4189 | dentry->d_name.name, | |
4190 | dentry->d_name.len); | |
4191 | goto out; | |
4192 | } | |
4193 | ||
7b128766 JB |
4194 | err = btrfs_orphan_add(trans, inode); |
4195 | if (err) | |
4df27c4d | 4196 | goto out; |
7b128766 | 4197 | |
39279cc3 | 4198 | /* now the directory is empty */ |
2b0143b5 | 4199 | err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry), |
e02119d5 | 4200 | dentry->d_name.name, dentry->d_name.len); |
d397712b | 4201 | if (!err) |
dbe674a9 | 4202 | btrfs_i_size_write(inode, 0); |
4df27c4d | 4203 | out: |
d52be818 | 4204 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4205 | btrfs_btree_balance_dirty(root); |
3954401f | 4206 | |
39279cc3 CM |
4207 | return err; |
4208 | } | |
4209 | ||
28f75a0e CM |
4210 | static int truncate_space_check(struct btrfs_trans_handle *trans, |
4211 | struct btrfs_root *root, | |
4212 | u64 bytes_deleted) | |
4213 | { | |
4214 | int ret; | |
4215 | ||
dc95f7bf JB |
4216 | /* |
4217 | * This is only used to apply pressure to the enospc system, we don't | |
4218 | * intend to use this reservation at all. | |
4219 | */ | |
28f75a0e | 4220 | bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted); |
dc95f7bf | 4221 | bytes_deleted *= root->nodesize; |
28f75a0e CM |
4222 | ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv, |
4223 | bytes_deleted, BTRFS_RESERVE_NO_FLUSH); | |
dc95f7bf JB |
4224 | if (!ret) { |
4225 | trace_btrfs_space_reservation(root->fs_info, "transaction", | |
4226 | trans->transid, | |
4227 | bytes_deleted, 1); | |
28f75a0e | 4228 | trans->bytes_reserved += bytes_deleted; |
dc95f7bf | 4229 | } |
28f75a0e CM |
4230 | return ret; |
4231 | ||
4232 | } | |
4233 | ||
0305cd5f FM |
4234 | static int truncate_inline_extent(struct inode *inode, |
4235 | struct btrfs_path *path, | |
4236 | struct btrfs_key *found_key, | |
4237 | const u64 item_end, | |
4238 | const u64 new_size) | |
4239 | { | |
4240 | struct extent_buffer *leaf = path->nodes[0]; | |
4241 | int slot = path->slots[0]; | |
4242 | struct btrfs_file_extent_item *fi; | |
4243 | u32 size = (u32)(new_size - found_key->offset); | |
4244 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4245 | ||
4246 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
4247 | ||
4248 | if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) { | |
4249 | loff_t offset = new_size; | |
09cbfeaf | 4250 | loff_t page_end = ALIGN(offset, PAGE_SIZE); |
0305cd5f FM |
4251 | |
4252 | /* | |
4253 | * Zero out the remaining of the last page of our inline extent, | |
4254 | * instead of directly truncating our inline extent here - that | |
4255 | * would be much more complex (decompressing all the data, then | |
4256 | * compressing the truncated data, which might be bigger than | |
4257 | * the size of the inline extent, resize the extent, etc). | |
4258 | * We release the path because to get the page we might need to | |
4259 | * read the extent item from disk (data not in the page cache). | |
4260 | */ | |
4261 | btrfs_release_path(path); | |
9703fefe CR |
4262 | return btrfs_truncate_block(inode, offset, page_end - offset, |
4263 | 0); | |
0305cd5f FM |
4264 | } |
4265 | ||
4266 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4267 | size = btrfs_file_extent_calc_inline_size(size); | |
4268 | btrfs_truncate_item(root, path, size, 1); | |
4269 | ||
4270 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) | |
4271 | inode_sub_bytes(inode, item_end + 1 - new_size); | |
4272 | ||
4273 | return 0; | |
4274 | } | |
4275 | ||
39279cc3 CM |
4276 | /* |
4277 | * this can truncate away extent items, csum items and directory items. | |
4278 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4279 | * any higher than new_size |
39279cc3 CM |
4280 | * |
4281 | * csum items that cross the new i_size are truncated to the new size | |
4282 | * as well. | |
7b128766 JB |
4283 | * |
4284 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4285 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4286 | */ |
8082510e YZ |
4287 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4288 | struct btrfs_root *root, | |
4289 | struct inode *inode, | |
4290 | u64 new_size, u32 min_type) | |
39279cc3 | 4291 | { |
39279cc3 | 4292 | struct btrfs_path *path; |
5f39d397 | 4293 | struct extent_buffer *leaf; |
39279cc3 | 4294 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4295 | struct btrfs_key key; |
4296 | struct btrfs_key found_key; | |
39279cc3 | 4297 | u64 extent_start = 0; |
db94535d | 4298 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4299 | u64 extent_offset = 0; |
39279cc3 | 4300 | u64 item_end = 0; |
c1aa4575 | 4301 | u64 last_size = new_size; |
8082510e | 4302 | u32 found_type = (u8)-1; |
39279cc3 CM |
4303 | int found_extent; |
4304 | int del_item; | |
85e21bac CM |
4305 | int pending_del_nr = 0; |
4306 | int pending_del_slot = 0; | |
179e29e4 | 4307 | int extent_type = -1; |
8082510e YZ |
4308 | int ret; |
4309 | int err = 0; | |
33345d01 | 4310 | u64 ino = btrfs_ino(inode); |
28ed1345 | 4311 | u64 bytes_deleted = 0; |
1262133b JB |
4312 | bool be_nice = 0; |
4313 | bool should_throttle = 0; | |
28f75a0e | 4314 | bool should_end = 0; |
8082510e YZ |
4315 | |
4316 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4317 | |
28ed1345 CM |
4318 | /* |
4319 | * for non-free space inodes and ref cows, we want to back off from | |
4320 | * time to time | |
4321 | */ | |
4322 | if (!btrfs_is_free_space_inode(inode) && | |
4323 | test_bit(BTRFS_ROOT_REF_COWS, &root->state)) | |
4324 | be_nice = 1; | |
4325 | ||
0eb0e19c MF |
4326 | path = btrfs_alloc_path(); |
4327 | if (!path) | |
4328 | return -ENOMEM; | |
e4058b54 | 4329 | path->reada = READA_BACK; |
0eb0e19c | 4330 | |
5dc562c5 JB |
4331 | /* |
4332 | * We want to drop from the next block forward in case this new size is | |
4333 | * not block aligned since we will be keeping the last block of the | |
4334 | * extent just the way it is. | |
4335 | */ | |
27cdeb70 MX |
4336 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
4337 | root == root->fs_info->tree_root) | |
fda2832f QW |
4338 | btrfs_drop_extent_cache(inode, ALIGN(new_size, |
4339 | root->sectorsize), (u64)-1, 0); | |
8082510e | 4340 | |
16cdcec7 MX |
4341 | /* |
4342 | * This function is also used to drop the items in the log tree before | |
4343 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
4344 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
4345 | * items. | |
4346 | */ | |
4347 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4348 | btrfs_kill_delayed_inode_items(inode); | |
4349 | ||
33345d01 | 4350 | key.objectid = ino; |
39279cc3 | 4351 | key.offset = (u64)-1; |
5f39d397 CM |
4352 | key.type = (u8)-1; |
4353 | ||
85e21bac | 4354 | search_again: |
28ed1345 CM |
4355 | /* |
4356 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4357 | * up a huge file in a single leaf. Most of the time that | |
4358 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4359 | */ | |
ee22184b | 4360 | if (be_nice && bytes_deleted > SZ_32M) { |
28ed1345 CM |
4361 | if (btrfs_should_end_transaction(trans, root)) { |
4362 | err = -EAGAIN; | |
4363 | goto error; | |
4364 | } | |
4365 | } | |
4366 | ||
4367 | ||
b9473439 | 4368 | path->leave_spinning = 1; |
85e21bac | 4369 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
4370 | if (ret < 0) { |
4371 | err = ret; | |
4372 | goto out; | |
4373 | } | |
d397712b | 4374 | |
85e21bac | 4375 | if (ret > 0) { |
e02119d5 CM |
4376 | /* there are no items in the tree for us to truncate, we're |
4377 | * done | |
4378 | */ | |
8082510e YZ |
4379 | if (path->slots[0] == 0) |
4380 | goto out; | |
85e21bac CM |
4381 | path->slots[0]--; |
4382 | } | |
4383 | ||
d397712b | 4384 | while (1) { |
39279cc3 | 4385 | fi = NULL; |
5f39d397 CM |
4386 | leaf = path->nodes[0]; |
4387 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4388 | found_type = found_key.type; |
39279cc3 | 4389 | |
33345d01 | 4390 | if (found_key.objectid != ino) |
39279cc3 | 4391 | break; |
5f39d397 | 4392 | |
85e21bac | 4393 | if (found_type < min_type) |
39279cc3 CM |
4394 | break; |
4395 | ||
5f39d397 | 4396 | item_end = found_key.offset; |
39279cc3 | 4397 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4398 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4399 | struct btrfs_file_extent_item); |
179e29e4 CM |
4400 | extent_type = btrfs_file_extent_type(leaf, fi); |
4401 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4402 | item_end += |
db94535d | 4403 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 4404 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 4405 | item_end += btrfs_file_extent_inline_len(leaf, |
514ac8ad | 4406 | path->slots[0], fi); |
39279cc3 | 4407 | } |
008630c1 | 4408 | item_end--; |
39279cc3 | 4409 | } |
8082510e YZ |
4410 | if (found_type > min_type) { |
4411 | del_item = 1; | |
4412 | } else { | |
4413 | if (item_end < new_size) | |
b888db2b | 4414 | break; |
8082510e YZ |
4415 | if (found_key.offset >= new_size) |
4416 | del_item = 1; | |
4417 | else | |
4418 | del_item = 0; | |
39279cc3 | 4419 | } |
39279cc3 | 4420 | found_extent = 0; |
39279cc3 | 4421 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4422 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4423 | goto delete; | |
4424 | ||
7f4f6e0a JB |
4425 | if (del_item) |
4426 | last_size = found_key.offset; | |
4427 | else | |
4428 | last_size = new_size; | |
4429 | ||
179e29e4 | 4430 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
39279cc3 | 4431 | u64 num_dec; |
db94535d | 4432 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4433 | if (!del_item) { |
db94535d CM |
4434 | u64 orig_num_bytes = |
4435 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4436 | extent_num_bytes = ALIGN(new_size - |
4437 | found_key.offset, | |
4438 | root->sectorsize); | |
db94535d CM |
4439 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4440 | extent_num_bytes); | |
4441 | num_dec = (orig_num_bytes - | |
9069218d | 4442 | extent_num_bytes); |
27cdeb70 MX |
4443 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4444 | &root->state) && | |
4445 | extent_start != 0) | |
a76a3cd4 | 4446 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4447 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4448 | } else { |
db94535d CM |
4449 | extent_num_bytes = |
4450 | btrfs_file_extent_disk_num_bytes(leaf, | |
4451 | fi); | |
5d4f98a2 YZ |
4452 | extent_offset = found_key.offset - |
4453 | btrfs_file_extent_offset(leaf, fi); | |
4454 | ||
39279cc3 | 4455 | /* FIXME blocksize != 4096 */ |
9069218d | 4456 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4457 | if (extent_start != 0) { |
4458 | found_extent = 1; | |
27cdeb70 MX |
4459 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4460 | &root->state)) | |
a76a3cd4 | 4461 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4462 | } |
39279cc3 | 4463 | } |
9069218d | 4464 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4465 | /* |
4466 | * we can't truncate inline items that have had | |
4467 | * special encodings | |
4468 | */ | |
4469 | if (!del_item && | |
c8b97818 CM |
4470 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
4471 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
514ac8ad CM |
4472 | |
4473 | /* | |
0305cd5f FM |
4474 | * Need to release path in order to truncate a |
4475 | * compressed extent. So delete any accumulated | |
4476 | * extent items so far. | |
514ac8ad | 4477 | */ |
0305cd5f FM |
4478 | if (btrfs_file_extent_compression(leaf, fi) != |
4479 | BTRFS_COMPRESS_NONE && pending_del_nr) { | |
4480 | err = btrfs_del_items(trans, root, path, | |
4481 | pending_del_slot, | |
4482 | pending_del_nr); | |
4483 | if (err) { | |
4484 | btrfs_abort_transaction(trans, | |
4485 | root, | |
4486 | err); | |
4487 | goto error; | |
4488 | } | |
4489 | pending_del_nr = 0; | |
4490 | } | |
4491 | ||
4492 | err = truncate_inline_extent(inode, path, | |
4493 | &found_key, | |
4494 | item_end, | |
4495 | new_size); | |
4496 | if (err) { | |
4497 | btrfs_abort_transaction(trans, | |
4498 | root, err); | |
4499 | goto error; | |
4500 | } | |
27cdeb70 MX |
4501 | } else if (test_bit(BTRFS_ROOT_REF_COWS, |
4502 | &root->state)) { | |
0305cd5f | 4503 | inode_sub_bytes(inode, item_end + 1 - new_size); |
9069218d | 4504 | } |
39279cc3 | 4505 | } |
179e29e4 | 4506 | delete: |
39279cc3 | 4507 | if (del_item) { |
85e21bac CM |
4508 | if (!pending_del_nr) { |
4509 | /* no pending yet, add ourselves */ | |
4510 | pending_del_slot = path->slots[0]; | |
4511 | pending_del_nr = 1; | |
4512 | } else if (pending_del_nr && | |
4513 | path->slots[0] + 1 == pending_del_slot) { | |
4514 | /* hop on the pending chunk */ | |
4515 | pending_del_nr++; | |
4516 | pending_del_slot = path->slots[0]; | |
4517 | } else { | |
d397712b | 4518 | BUG(); |
85e21bac | 4519 | } |
39279cc3 CM |
4520 | } else { |
4521 | break; | |
4522 | } | |
28f75a0e CM |
4523 | should_throttle = 0; |
4524 | ||
27cdeb70 MX |
4525 | if (found_extent && |
4526 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
4527 | root == root->fs_info->tree_root)) { | |
b9473439 | 4528 | btrfs_set_path_blocking(path); |
28ed1345 | 4529 | bytes_deleted += extent_num_bytes; |
39279cc3 | 4530 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
4531 | extent_num_bytes, 0, |
4532 | btrfs_header_owner(leaf), | |
b06c4bf5 | 4533 | ino, extent_offset); |
39279cc3 | 4534 | BUG_ON(ret); |
1262133b | 4535 | if (btrfs_should_throttle_delayed_refs(trans, root)) |
28ed1345 CM |
4536 | btrfs_async_run_delayed_refs(root, |
4537 | trans->delayed_ref_updates * 2, 0); | |
28f75a0e CM |
4538 | if (be_nice) { |
4539 | if (truncate_space_check(trans, root, | |
4540 | extent_num_bytes)) { | |
4541 | should_end = 1; | |
4542 | } | |
4543 | if (btrfs_should_throttle_delayed_refs(trans, | |
4544 | root)) { | |
4545 | should_throttle = 1; | |
4546 | } | |
4547 | } | |
39279cc3 | 4548 | } |
85e21bac | 4549 | |
8082510e YZ |
4550 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4551 | break; | |
4552 | ||
4553 | if (path->slots[0] == 0 || | |
1262133b | 4554 | path->slots[0] != pending_del_slot || |
28f75a0e | 4555 | should_throttle || should_end) { |
8082510e YZ |
4556 | if (pending_del_nr) { |
4557 | ret = btrfs_del_items(trans, root, path, | |
4558 | pending_del_slot, | |
4559 | pending_del_nr); | |
79787eaa JM |
4560 | if (ret) { |
4561 | btrfs_abort_transaction(trans, | |
4562 | root, ret); | |
4563 | goto error; | |
4564 | } | |
8082510e YZ |
4565 | pending_del_nr = 0; |
4566 | } | |
b3b4aa74 | 4567 | btrfs_release_path(path); |
28f75a0e | 4568 | if (should_throttle) { |
1262133b JB |
4569 | unsigned long updates = trans->delayed_ref_updates; |
4570 | if (updates) { | |
4571 | trans->delayed_ref_updates = 0; | |
4572 | ret = btrfs_run_delayed_refs(trans, root, updates * 2); | |
4573 | if (ret && !err) | |
4574 | err = ret; | |
4575 | } | |
4576 | } | |
28f75a0e CM |
4577 | /* |
4578 | * if we failed to refill our space rsv, bail out | |
4579 | * and let the transaction restart | |
4580 | */ | |
4581 | if (should_end) { | |
4582 | err = -EAGAIN; | |
4583 | goto error; | |
4584 | } | |
85e21bac | 4585 | goto search_again; |
8082510e YZ |
4586 | } else { |
4587 | path->slots[0]--; | |
85e21bac | 4588 | } |
39279cc3 | 4589 | } |
8082510e | 4590 | out: |
85e21bac CM |
4591 | if (pending_del_nr) { |
4592 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4593 | pending_del_nr); | |
79787eaa JM |
4594 | if (ret) |
4595 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 4596 | } |
79787eaa | 4597 | error: |
c1aa4575 | 4598 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
7f4f6e0a | 4599 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
28ed1345 | 4600 | |
39279cc3 | 4601 | btrfs_free_path(path); |
28ed1345 | 4602 | |
ee22184b | 4603 | if (be_nice && bytes_deleted > SZ_32M) { |
28ed1345 CM |
4604 | unsigned long updates = trans->delayed_ref_updates; |
4605 | if (updates) { | |
4606 | trans->delayed_ref_updates = 0; | |
4607 | ret = btrfs_run_delayed_refs(trans, root, updates * 2); | |
4608 | if (ret && !err) | |
4609 | err = ret; | |
4610 | } | |
4611 | } | |
8082510e | 4612 | return err; |
39279cc3 CM |
4613 | } |
4614 | ||
4615 | /* | |
9703fefe | 4616 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4617 | * @inode - inode that we're zeroing |
4618 | * @from - the offset to start zeroing | |
4619 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4620 | * offset | |
4621 | * @front - zero up to the offset instead of from the offset on | |
4622 | * | |
9703fefe | 4623 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4624 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4625 | */ |
9703fefe | 4626 | int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len, |
2aaa6655 | 4627 | int front) |
39279cc3 | 4628 | { |
2aaa6655 | 4629 | struct address_space *mapping = inode->i_mapping; |
db94535d | 4630 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4631 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4632 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4633 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 4634 | char *kaddr; |
db94535d | 4635 | u32 blocksize = root->sectorsize; |
09cbfeaf | 4636 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4637 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4638 | struct page *page; |
3b16a4e3 | 4639 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4640 | int ret = 0; |
9703fefe CR |
4641 | u64 block_start; |
4642 | u64 block_end; | |
39279cc3 | 4643 | |
2aaa6655 JB |
4644 | if ((offset & (blocksize - 1)) == 0 && |
4645 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4646 | goto out; |
9703fefe | 4647 | |
7cf5b976 | 4648 | ret = btrfs_delalloc_reserve_space(inode, |
9703fefe | 4649 | round_down(from, blocksize), blocksize); |
5d5e103a JB |
4650 | if (ret) |
4651 | goto out; | |
39279cc3 | 4652 | |
211c17f5 | 4653 | again: |
3b16a4e3 | 4654 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4655 | if (!page) { |
7cf5b976 | 4656 | btrfs_delalloc_release_space(inode, |
9703fefe CR |
4657 | round_down(from, blocksize), |
4658 | blocksize); | |
ac6a2b36 | 4659 | ret = -ENOMEM; |
39279cc3 | 4660 | goto out; |
5d5e103a | 4661 | } |
e6dcd2dc | 4662 | |
9703fefe CR |
4663 | block_start = round_down(from, blocksize); |
4664 | block_end = block_start + blocksize - 1; | |
e6dcd2dc | 4665 | |
39279cc3 | 4666 | if (!PageUptodate(page)) { |
9ebefb18 | 4667 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4668 | lock_page(page); |
211c17f5 CM |
4669 | if (page->mapping != mapping) { |
4670 | unlock_page(page); | |
09cbfeaf | 4671 | put_page(page); |
211c17f5 CM |
4672 | goto again; |
4673 | } | |
39279cc3 CM |
4674 | if (!PageUptodate(page)) { |
4675 | ret = -EIO; | |
89642229 | 4676 | goto out_unlock; |
39279cc3 CM |
4677 | } |
4678 | } | |
211c17f5 | 4679 | wait_on_page_writeback(page); |
e6dcd2dc | 4680 | |
9703fefe | 4681 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc CM |
4682 | set_page_extent_mapped(page); |
4683 | ||
9703fefe | 4684 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4685 | if (ordered) { |
9703fefe | 4686 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4687 | &cached_state, GFP_NOFS); |
e6dcd2dc | 4688 | unlock_page(page); |
09cbfeaf | 4689 | put_page(page); |
eb84ae03 | 4690 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4691 | btrfs_put_ordered_extent(ordered); |
4692 | goto again; | |
4693 | } | |
4694 | ||
9703fefe | 4695 | clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end, |
9e8a4a8b LB |
4696 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4697 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 4698 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 4699 | |
9703fefe | 4700 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, |
2ac55d41 | 4701 | &cached_state); |
9ed74f2d | 4702 | if (ret) { |
9703fefe | 4703 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4704 | &cached_state, GFP_NOFS); |
9ed74f2d JB |
4705 | goto out_unlock; |
4706 | } | |
4707 | ||
9703fefe | 4708 | if (offset != blocksize) { |
2aaa6655 | 4709 | if (!len) |
9703fefe | 4710 | len = blocksize - offset; |
e6dcd2dc | 4711 | kaddr = kmap(page); |
2aaa6655 | 4712 | if (front) |
9703fefe CR |
4713 | memset(kaddr + (block_start - page_offset(page)), |
4714 | 0, offset); | |
2aaa6655 | 4715 | else |
9703fefe CR |
4716 | memset(kaddr + (block_start - page_offset(page)) + offset, |
4717 | 0, len); | |
e6dcd2dc CM |
4718 | flush_dcache_page(page); |
4719 | kunmap(page); | |
4720 | } | |
247e743c | 4721 | ClearPageChecked(page); |
e6dcd2dc | 4722 | set_page_dirty(page); |
9703fefe | 4723 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state, |
2ac55d41 | 4724 | GFP_NOFS); |
39279cc3 | 4725 | |
89642229 | 4726 | out_unlock: |
5d5e103a | 4727 | if (ret) |
9703fefe CR |
4728 | btrfs_delalloc_release_space(inode, block_start, |
4729 | blocksize); | |
39279cc3 | 4730 | unlock_page(page); |
09cbfeaf | 4731 | put_page(page); |
39279cc3 CM |
4732 | out: |
4733 | return ret; | |
4734 | } | |
4735 | ||
16e7549f JB |
4736 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4737 | u64 offset, u64 len) | |
4738 | { | |
4739 | struct btrfs_trans_handle *trans; | |
4740 | int ret; | |
4741 | ||
4742 | /* | |
4743 | * Still need to make sure the inode looks like it's been updated so | |
4744 | * that any holes get logged if we fsync. | |
4745 | */ | |
4746 | if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) { | |
4747 | BTRFS_I(inode)->last_trans = root->fs_info->generation; | |
4748 | BTRFS_I(inode)->last_sub_trans = root->log_transid; | |
4749 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4750 | return 0; | |
4751 | } | |
4752 | ||
4753 | /* | |
4754 | * 1 - for the one we're dropping | |
4755 | * 1 - for the one we're adding | |
4756 | * 1 - for updating the inode. | |
4757 | */ | |
4758 | trans = btrfs_start_transaction(root, 3); | |
4759 | if (IS_ERR(trans)) | |
4760 | return PTR_ERR(trans); | |
4761 | ||
4762 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4763 | if (ret) { | |
4764 | btrfs_abort_transaction(trans, root, ret); | |
4765 | btrfs_end_transaction(trans, root); | |
4766 | return ret; | |
4767 | } | |
4768 | ||
4769 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset, | |
4770 | 0, 0, len, 0, len, 0, 0, 0); | |
4771 | if (ret) | |
4772 | btrfs_abort_transaction(trans, root, ret); | |
4773 | else | |
4774 | btrfs_update_inode(trans, root, inode); | |
4775 | btrfs_end_transaction(trans, root); | |
4776 | return ret; | |
4777 | } | |
4778 | ||
695a0d0d JB |
4779 | /* |
4780 | * This function puts in dummy file extents for the area we're creating a hole | |
4781 | * for. So if we are truncating this file to a larger size we need to insert | |
4782 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4783 | * the range between oldsize and size | |
4784 | */ | |
a41ad394 | 4785 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4786 | { |
9036c102 YZ |
4787 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4788 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4789 | struct extent_map *em = NULL; |
2ac55d41 | 4790 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4791 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
fda2832f QW |
4792 | u64 hole_start = ALIGN(oldsize, root->sectorsize); |
4793 | u64 block_end = ALIGN(size, root->sectorsize); | |
9036c102 YZ |
4794 | u64 last_byte; |
4795 | u64 cur_offset; | |
4796 | u64 hole_size; | |
9ed74f2d | 4797 | int err = 0; |
39279cc3 | 4798 | |
a71754fc | 4799 | /* |
9703fefe CR |
4800 | * If our size started in the middle of a block we need to zero out the |
4801 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
4802 | * expose stale data. |
4803 | */ | |
9703fefe | 4804 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
4805 | if (err) |
4806 | return err; | |
4807 | ||
9036c102 YZ |
4808 | if (size <= hole_start) |
4809 | return 0; | |
4810 | ||
9036c102 YZ |
4811 | while (1) { |
4812 | struct btrfs_ordered_extent *ordered; | |
fa7c1494 | 4813 | |
ff13db41 | 4814 | lock_extent_bits(io_tree, hole_start, block_end - 1, |
d0082371 | 4815 | &cached_state); |
fa7c1494 MX |
4816 | ordered = btrfs_lookup_ordered_range(inode, hole_start, |
4817 | block_end - hole_start); | |
9036c102 YZ |
4818 | if (!ordered) |
4819 | break; | |
2ac55d41 JB |
4820 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
4821 | &cached_state, GFP_NOFS); | |
fa7c1494 | 4822 | btrfs_start_ordered_extent(inode, ordered, 1); |
9036c102 YZ |
4823 | btrfs_put_ordered_extent(ordered); |
4824 | } | |
39279cc3 | 4825 | |
9036c102 YZ |
4826 | cur_offset = hole_start; |
4827 | while (1) { | |
4828 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
4829 | block_end - cur_offset, 0); | |
79787eaa JM |
4830 | if (IS_ERR(em)) { |
4831 | err = PTR_ERR(em); | |
f2767956 | 4832 | em = NULL; |
79787eaa JM |
4833 | break; |
4834 | } | |
9036c102 | 4835 | last_byte = min(extent_map_end(em), block_end); |
fda2832f | 4836 | last_byte = ALIGN(last_byte , root->sectorsize); |
8082510e | 4837 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4838 | struct extent_map *hole_em; |
9036c102 | 4839 | hole_size = last_byte - cur_offset; |
9ed74f2d | 4840 | |
16e7549f JB |
4841 | err = maybe_insert_hole(root, inode, cur_offset, |
4842 | hole_size); | |
4843 | if (err) | |
3893e33b | 4844 | break; |
5dc562c5 JB |
4845 | btrfs_drop_extent_cache(inode, cur_offset, |
4846 | cur_offset + hole_size - 1, 0); | |
4847 | hole_em = alloc_extent_map(); | |
4848 | if (!hole_em) { | |
4849 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4850 | &BTRFS_I(inode)->runtime_flags); | |
4851 | goto next; | |
4852 | } | |
4853 | hole_em->start = cur_offset; | |
4854 | hole_em->len = hole_size; | |
4855 | hole_em->orig_start = cur_offset; | |
8082510e | 4856 | |
5dc562c5 JB |
4857 | hole_em->block_start = EXTENT_MAP_HOLE; |
4858 | hole_em->block_len = 0; | |
b4939680 | 4859 | hole_em->orig_block_len = 0; |
cc95bef6 | 4860 | hole_em->ram_bytes = hole_size; |
5dc562c5 JB |
4861 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
4862 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
16e7549f | 4863 | hole_em->generation = root->fs_info->generation; |
8082510e | 4864 | |
5dc562c5 JB |
4865 | while (1) { |
4866 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4867 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4868 | write_unlock(&em_tree->lock); |
4869 | if (err != -EEXIST) | |
4870 | break; | |
4871 | btrfs_drop_extent_cache(inode, cur_offset, | |
4872 | cur_offset + | |
4873 | hole_size - 1, 0); | |
4874 | } | |
4875 | free_extent_map(hole_em); | |
9036c102 | 4876 | } |
16e7549f | 4877 | next: |
9036c102 | 4878 | free_extent_map(em); |
a22285a6 | 4879 | em = NULL; |
9036c102 | 4880 | cur_offset = last_byte; |
8082510e | 4881 | if (cur_offset >= block_end) |
9036c102 YZ |
4882 | break; |
4883 | } | |
a22285a6 | 4884 | free_extent_map(em); |
2ac55d41 JB |
4885 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
4886 | GFP_NOFS); | |
9036c102 YZ |
4887 | return err; |
4888 | } | |
39279cc3 | 4889 | |
3972f260 | 4890 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4891 | { |
f4a2f4c5 MX |
4892 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4893 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4894 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4895 | loff_t newsize = attr->ia_size; |
4896 | int mask = attr->ia_valid; | |
8082510e YZ |
4897 | int ret; |
4898 | ||
3972f260 ES |
4899 | /* |
4900 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4901 | * special case where we need to update the times despite not having | |
4902 | * these flags set. For all other operations the VFS set these flags | |
4903 | * explicitly if it wants a timestamp update. | |
4904 | */ | |
dff6efc3 CH |
4905 | if (newsize != oldsize) { |
4906 | inode_inc_iversion(inode); | |
4907 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
4908 | inode->i_ctime = inode->i_mtime = | |
4909 | current_fs_time(inode->i_sb); | |
4910 | } | |
3972f260 | 4911 | |
a41ad394 | 4912 | if (newsize > oldsize) { |
9ea24bbe FM |
4913 | /* |
4914 | * Don't do an expanding truncate while snapshoting is ongoing. | |
4915 | * This is to ensure the snapshot captures a fully consistent | |
4916 | * state of this file - if the snapshot captures this expanding | |
4917 | * truncation, it must capture all writes that happened before | |
4918 | * this truncation. | |
4919 | */ | |
0bc19f90 | 4920 | btrfs_wait_for_snapshot_creation(root); |
a41ad394 | 4921 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
9ea24bbe FM |
4922 | if (ret) { |
4923 | btrfs_end_write_no_snapshoting(root); | |
8082510e | 4924 | return ret; |
9ea24bbe | 4925 | } |
8082510e | 4926 | |
f4a2f4c5 | 4927 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe FM |
4928 | if (IS_ERR(trans)) { |
4929 | btrfs_end_write_no_snapshoting(root); | |
f4a2f4c5 | 4930 | return PTR_ERR(trans); |
9ea24bbe | 4931 | } |
f4a2f4c5 MX |
4932 | |
4933 | i_size_write(inode, newsize); | |
4934 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
27772b68 | 4935 | pagecache_isize_extended(inode, oldsize, newsize); |
f4a2f4c5 | 4936 | ret = btrfs_update_inode(trans, root, inode); |
9ea24bbe | 4937 | btrfs_end_write_no_snapshoting(root); |
7ad85bb7 | 4938 | btrfs_end_transaction(trans, root); |
a41ad394 | 4939 | } else { |
8082510e | 4940 | |
a41ad394 JB |
4941 | /* |
4942 | * We're truncating a file that used to have good data down to | |
4943 | * zero. Make sure it gets into the ordered flush list so that | |
4944 | * any new writes get down to disk quickly. | |
4945 | */ | |
4946 | if (newsize == 0) | |
72ac3c0d JB |
4947 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
4948 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 4949 | |
f3fe820c JB |
4950 | /* |
4951 | * 1 for the orphan item we're going to add | |
4952 | * 1 for the orphan item deletion. | |
4953 | */ | |
4954 | trans = btrfs_start_transaction(root, 2); | |
4955 | if (IS_ERR(trans)) | |
4956 | return PTR_ERR(trans); | |
4957 | ||
4958 | /* | |
4959 | * We need to do this in case we fail at _any_ point during the | |
4960 | * actual truncate. Once we do the truncate_setsize we could | |
4961 | * invalidate pages which forces any outstanding ordered io to | |
4962 | * be instantly completed which will give us extents that need | |
4963 | * to be truncated. If we fail to get an orphan inode down we | |
4964 | * could have left over extents that were never meant to live, | |
4965 | * so we need to garuntee from this point on that everything | |
4966 | * will be consistent. | |
4967 | */ | |
4968 | ret = btrfs_orphan_add(trans, inode); | |
4969 | btrfs_end_transaction(trans, root); | |
4970 | if (ret) | |
4971 | return ret; | |
4972 | ||
a41ad394 JB |
4973 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
4974 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
4975 | |
4976 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
4977 | btrfs_inode_block_unlocked_dio(inode); | |
4978 | inode_dio_wait(inode); | |
4979 | btrfs_inode_resume_unlocked_dio(inode); | |
4980 | ||
a41ad394 | 4981 | ret = btrfs_truncate(inode); |
7f4f6e0a JB |
4982 | if (ret && inode->i_nlink) { |
4983 | int err; | |
4984 | ||
4985 | /* | |
4986 | * failed to truncate, disk_i_size is only adjusted down | |
4987 | * as we remove extents, so it should represent the true | |
4988 | * size of the inode, so reset the in memory size and | |
4989 | * delete our orphan entry. | |
4990 | */ | |
4991 | trans = btrfs_join_transaction(root); | |
4992 | if (IS_ERR(trans)) { | |
4993 | btrfs_orphan_del(NULL, inode); | |
4994 | return ret; | |
4995 | } | |
4996 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
4997 | err = btrfs_orphan_del(trans, inode); | |
4998 | if (err) | |
4999 | btrfs_abort_transaction(trans, root, err); | |
5000 | btrfs_end_transaction(trans, root); | |
5001 | } | |
8082510e YZ |
5002 | } |
5003 | ||
a41ad394 | 5004 | return ret; |
8082510e YZ |
5005 | } |
5006 | ||
9036c102 YZ |
5007 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
5008 | { | |
2b0143b5 | 5009 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5010 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5011 | int err; |
39279cc3 | 5012 | |
b83cc969 LZ |
5013 | if (btrfs_root_readonly(root)) |
5014 | return -EROFS; | |
5015 | ||
9036c102 YZ |
5016 | err = inode_change_ok(inode, attr); |
5017 | if (err) | |
5018 | return err; | |
2bf5a725 | 5019 | |
5a3f23d5 | 5020 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5021 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5022 | if (err) |
5023 | return err; | |
39279cc3 | 5024 | } |
9036c102 | 5025 | |
1025774c CH |
5026 | if (attr->ia_valid) { |
5027 | setattr_copy(inode, attr); | |
0c4d2d95 | 5028 | inode_inc_iversion(inode); |
22c44fe6 | 5029 | err = btrfs_dirty_inode(inode); |
1025774c | 5030 | |
22c44fe6 | 5031 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 5032 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 5033 | } |
33268eaf | 5034 | |
39279cc3 CM |
5035 | return err; |
5036 | } | |
61295eb8 | 5037 | |
131e404a FDBM |
5038 | /* |
5039 | * While truncating the inode pages during eviction, we get the VFS calling | |
5040 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
5041 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
5042 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
5043 | * extent_state structures over and over, wasting lots of time. | |
5044 | * | |
5045 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
5046 | * those expensive operations on a per page basis and do only the ordered io | |
5047 | * finishing, while we release here the extent_map and extent_state structures, | |
5048 | * without the excessive merging and splitting. | |
5049 | */ | |
5050 | static void evict_inode_truncate_pages(struct inode *inode) | |
5051 | { | |
5052 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5053 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5054 | struct rb_node *node; | |
5055 | ||
5056 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5057 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5058 | |
5059 | write_lock(&map_tree->lock); | |
5060 | while (!RB_EMPTY_ROOT(&map_tree->map)) { | |
5061 | struct extent_map *em; | |
5062 | ||
5063 | node = rb_first(&map_tree->map); | |
5064 | em = rb_entry(node, struct extent_map, rb_node); | |
180589ef WS |
5065 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5066 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5067 | remove_extent_mapping(map_tree, em); |
5068 | free_extent_map(em); | |
7064dd5c FM |
5069 | if (need_resched()) { |
5070 | write_unlock(&map_tree->lock); | |
5071 | cond_resched(); | |
5072 | write_lock(&map_tree->lock); | |
5073 | } | |
131e404a FDBM |
5074 | } |
5075 | write_unlock(&map_tree->lock); | |
5076 | ||
6ca07097 FM |
5077 | /* |
5078 | * Keep looping until we have no more ranges in the io tree. | |
5079 | * We can have ongoing bios started by readpages (called from readahead) | |
9c6429d9 FM |
5080 | * that have their endio callback (extent_io.c:end_bio_extent_readpage) |
5081 | * still in progress (unlocked the pages in the bio but did not yet | |
5082 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5083 | * ranges can still be locked and eviction started because before |
5084 | * submitting those bios, which are executed by a separate task (work | |
5085 | * queue kthread), inode references (inode->i_count) were not taken | |
5086 | * (which would be dropped in the end io callback of each bio). | |
5087 | * Therefore here we effectively end up waiting for those bios and | |
5088 | * anyone else holding locked ranges without having bumped the inode's | |
5089 | * reference count - if we don't do it, when they access the inode's | |
5090 | * io_tree to unlock a range it may be too late, leading to an | |
5091 | * use-after-free issue. | |
5092 | */ | |
131e404a FDBM |
5093 | spin_lock(&io_tree->lock); |
5094 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5095 | struct extent_state *state; | |
5096 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5097 | u64 start; |
5098 | u64 end; | |
131e404a FDBM |
5099 | |
5100 | node = rb_first(&io_tree->state); | |
5101 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5102 | start = state->start; |
5103 | end = state->end; | |
131e404a FDBM |
5104 | spin_unlock(&io_tree->lock); |
5105 | ||
ff13db41 | 5106 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5107 | |
5108 | /* | |
5109 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5110 | * and its reserved space won't be freed by delayed_ref. | |
5111 | * So we need to free its reserved space here. | |
5112 | * (Refer to comment in btrfs_invalidatepage, case 2) | |
5113 | * | |
5114 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5115 | */ | |
5116 | if (state->state & EXTENT_DELALLOC) | |
5117 | btrfs_qgroup_free_data(inode, start, end - start + 1); | |
5118 | ||
6ca07097 | 5119 | clear_extent_bit(io_tree, start, end, |
131e404a FDBM |
5120 | EXTENT_LOCKED | EXTENT_DIRTY | |
5121 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
5122 | EXTENT_DEFRAG, 1, 1, | |
5123 | &cached_state, GFP_NOFS); | |
131e404a | 5124 | |
7064dd5c | 5125 | cond_resched(); |
131e404a FDBM |
5126 | spin_lock(&io_tree->lock); |
5127 | } | |
5128 | spin_unlock(&io_tree->lock); | |
5129 | } | |
5130 | ||
bd555975 | 5131 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
5132 | { |
5133 | struct btrfs_trans_handle *trans; | |
5134 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 5135 | struct btrfs_block_rsv *rsv, *global_rsv; |
3bce876f | 5136 | int steal_from_global = 0; |
07127184 | 5137 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 CM |
5138 | int ret; |
5139 | ||
1abe9b8a | 5140 | trace_btrfs_inode_evict(inode); |
5141 | ||
131e404a FDBM |
5142 | evict_inode_truncate_pages(inode); |
5143 | ||
69e9c6c6 SB |
5144 | if (inode->i_nlink && |
5145 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5146 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
5147 | btrfs_is_free_space_inode(inode))) | |
bd555975 AV |
5148 | goto no_delete; |
5149 | ||
39279cc3 | 5150 | if (is_bad_inode(inode)) { |
7b128766 | 5151 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
5152 | goto no_delete; |
5153 | } | |
bd555975 | 5154 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
a30e577c JM |
5155 | if (!special_file(inode->i_mode)) |
5156 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
5f39d397 | 5157 | |
f612496b MX |
5158 | btrfs_free_io_failure_record(inode, 0, (u64)-1); |
5159 | ||
c71bf099 | 5160 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 5161 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 5162 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
5163 | goto no_delete; |
5164 | } | |
5165 | ||
76dda93c | 5166 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5167 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5168 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5169 | goto no_delete; |
5170 | } | |
5171 | ||
0e8c36a9 MX |
5172 | ret = btrfs_commit_inode_delayed_inode(inode); |
5173 | if (ret) { | |
5174 | btrfs_orphan_del(NULL, inode); | |
5175 | goto no_delete; | |
5176 | } | |
5177 | ||
66d8f3dd | 5178 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
5179 | if (!rsv) { |
5180 | btrfs_orphan_del(NULL, inode); | |
5181 | goto no_delete; | |
5182 | } | |
4a338542 | 5183 | rsv->size = min_size; |
ca7e70f5 | 5184 | rsv->failfast = 1; |
726c35fa | 5185 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 5186 | |
dbe674a9 | 5187 | btrfs_i_size_write(inode, 0); |
5f39d397 | 5188 | |
4289a667 | 5189 | /* |
8407aa46 MX |
5190 | * This is a bit simpler than btrfs_truncate since we've already |
5191 | * reserved our space for our orphan item in the unlink, so we just | |
5192 | * need to reserve some slack space in case we add bytes and update | |
5193 | * inode item when doing the truncate. | |
4289a667 | 5194 | */ |
8082510e | 5195 | while (1) { |
08e007d2 MX |
5196 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
5197 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
5198 | |
5199 | /* | |
5200 | * Try and steal from the global reserve since we will | |
5201 | * likely not use this space anyway, we want to try as | |
5202 | * hard as possible to get this to work. | |
5203 | */ | |
5204 | if (ret) | |
3bce876f JB |
5205 | steal_from_global++; |
5206 | else | |
5207 | steal_from_global = 0; | |
5208 | ret = 0; | |
d68fc57b | 5209 | |
3bce876f JB |
5210 | /* |
5211 | * steal_from_global == 0: we reserved stuff, hooray! | |
5212 | * steal_from_global == 1: we didn't reserve stuff, boo! | |
5213 | * steal_from_global == 2: we've committed, still not a lot of | |
5214 | * room but maybe we'll have room in the global reserve this | |
5215 | * time. | |
5216 | * steal_from_global == 3: abandon all hope! | |
5217 | */ | |
5218 | if (steal_from_global > 2) { | |
c2cf52eb SK |
5219 | btrfs_warn(root->fs_info, |
5220 | "Could not get space for a delete, will truncate on mount %d", | |
5221 | ret); | |
4289a667 JB |
5222 | btrfs_orphan_del(NULL, inode); |
5223 | btrfs_free_block_rsv(root, rsv); | |
5224 | goto no_delete; | |
d68fc57b | 5225 | } |
7b128766 | 5226 | |
0e8c36a9 | 5227 | trans = btrfs_join_transaction(root); |
4289a667 JB |
5228 | if (IS_ERR(trans)) { |
5229 | btrfs_orphan_del(NULL, inode); | |
5230 | btrfs_free_block_rsv(root, rsv); | |
5231 | goto no_delete; | |
d68fc57b | 5232 | } |
7b128766 | 5233 | |
3bce876f JB |
5234 | /* |
5235 | * We can't just steal from the global reserve, we need tomake | |
5236 | * sure there is room to do it, if not we need to commit and try | |
5237 | * again. | |
5238 | */ | |
5239 | if (steal_from_global) { | |
5240 | if (!btrfs_check_space_for_delayed_refs(trans, root)) | |
5241 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, | |
5242 | min_size); | |
5243 | else | |
5244 | ret = -ENOSPC; | |
5245 | } | |
5246 | ||
5247 | /* | |
5248 | * Couldn't steal from the global reserve, we have too much | |
5249 | * pending stuff built up, commit the transaction and try it | |
5250 | * again. | |
5251 | */ | |
5252 | if (ret) { | |
5253 | ret = btrfs_commit_transaction(trans, root); | |
5254 | if (ret) { | |
5255 | btrfs_orphan_del(NULL, inode); | |
5256 | btrfs_free_block_rsv(root, rsv); | |
5257 | goto no_delete; | |
5258 | } | |
5259 | continue; | |
5260 | } else { | |
5261 | steal_from_global = 0; | |
5262 | } | |
5263 | ||
4289a667 JB |
5264 | trans->block_rsv = rsv; |
5265 | ||
d68fc57b | 5266 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
28ed1345 | 5267 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 5268 | break; |
85e21bac | 5269 | |
8407aa46 | 5270 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e YZ |
5271 | btrfs_end_transaction(trans, root); |
5272 | trans = NULL; | |
b53d3f5d | 5273 | btrfs_btree_balance_dirty(root); |
8082510e | 5274 | } |
5f39d397 | 5275 | |
4289a667 JB |
5276 | btrfs_free_block_rsv(root, rsv); |
5277 | ||
4ef31a45 JB |
5278 | /* |
5279 | * Errors here aren't a big deal, it just means we leave orphan items | |
5280 | * in the tree. They will be cleaned up on the next mount. | |
5281 | */ | |
8082510e | 5282 | if (ret == 0) { |
4289a667 | 5283 | trans->block_rsv = root->orphan_block_rsv; |
4ef31a45 JB |
5284 | btrfs_orphan_del(trans, inode); |
5285 | } else { | |
5286 | btrfs_orphan_del(NULL, inode); | |
8082510e | 5287 | } |
54aa1f4d | 5288 | |
4289a667 | 5289 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
5290 | if (!(root == root->fs_info->tree_root || |
5291 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 5292 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 5293 | |
54aa1f4d | 5294 | btrfs_end_transaction(trans, root); |
b53d3f5d | 5295 | btrfs_btree_balance_dirty(root); |
39279cc3 | 5296 | no_delete: |
89042e5a | 5297 | btrfs_remove_delayed_node(inode); |
dbd5768f | 5298 | clear_inode(inode); |
39279cc3 CM |
5299 | } |
5300 | ||
5301 | /* | |
5302 | * this returns the key found in the dir entry in the location pointer. | |
5303 | * If no dir entries were found, location->objectid is 0. | |
5304 | */ | |
5305 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
5306 | struct btrfs_key *location) | |
5307 | { | |
5308 | const char *name = dentry->d_name.name; | |
5309 | int namelen = dentry->d_name.len; | |
5310 | struct btrfs_dir_item *di; | |
5311 | struct btrfs_path *path; | |
5312 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5313 | int ret = 0; |
39279cc3 CM |
5314 | |
5315 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5316 | if (!path) |
5317 | return -ENOMEM; | |
3954401f | 5318 | |
33345d01 | 5319 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 5320 | namelen, 0); |
0d9f7f3e Y |
5321 | if (IS_ERR(di)) |
5322 | ret = PTR_ERR(di); | |
d397712b | 5323 | |
c704005d | 5324 | if (IS_ERR_OR_NULL(di)) |
3954401f | 5325 | goto out_err; |
d397712b | 5326 | |
5f39d397 | 5327 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 5328 | out: |
39279cc3 CM |
5329 | btrfs_free_path(path); |
5330 | return ret; | |
3954401f CM |
5331 | out_err: |
5332 | location->objectid = 0; | |
5333 | goto out; | |
39279cc3 CM |
5334 | } |
5335 | ||
5336 | /* | |
5337 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5338 | * needs to be changed to reflect the root directory of the tree root. This | |
5339 | * is kind of like crossing a mount point. | |
5340 | */ | |
5341 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
5342 | struct inode *dir, |
5343 | struct dentry *dentry, | |
5344 | struct btrfs_key *location, | |
5345 | struct btrfs_root **sub_root) | |
39279cc3 | 5346 | { |
4df27c4d YZ |
5347 | struct btrfs_path *path; |
5348 | struct btrfs_root *new_root; | |
5349 | struct btrfs_root_ref *ref; | |
5350 | struct extent_buffer *leaf; | |
1d4c08e0 | 5351 | struct btrfs_key key; |
4df27c4d YZ |
5352 | int ret; |
5353 | int err = 0; | |
39279cc3 | 5354 | |
4df27c4d YZ |
5355 | path = btrfs_alloc_path(); |
5356 | if (!path) { | |
5357 | err = -ENOMEM; | |
5358 | goto out; | |
5359 | } | |
39279cc3 | 5360 | |
4df27c4d | 5361 | err = -ENOENT; |
1d4c08e0 DS |
5362 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5363 | key.type = BTRFS_ROOT_REF_KEY; | |
5364 | key.offset = location->objectid; | |
5365 | ||
5366 | ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path, | |
5367 | 0, 0); | |
4df27c4d YZ |
5368 | if (ret) { |
5369 | if (ret < 0) | |
5370 | err = ret; | |
5371 | goto out; | |
5372 | } | |
39279cc3 | 5373 | |
4df27c4d YZ |
5374 | leaf = path->nodes[0]; |
5375 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 5376 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
5377 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5378 | goto out; | |
39279cc3 | 5379 | |
4df27c4d YZ |
5380 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5381 | (unsigned long)(ref + 1), | |
5382 | dentry->d_name.len); | |
5383 | if (ret) | |
5384 | goto out; | |
5385 | ||
b3b4aa74 | 5386 | btrfs_release_path(path); |
4df27c4d YZ |
5387 | |
5388 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
5389 | if (IS_ERR(new_root)) { | |
5390 | err = PTR_ERR(new_root); | |
5391 | goto out; | |
5392 | } | |
5393 | ||
4df27c4d YZ |
5394 | *sub_root = new_root; |
5395 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5396 | location->type = BTRFS_INODE_ITEM_KEY; | |
5397 | location->offset = 0; | |
5398 | err = 0; | |
5399 | out: | |
5400 | btrfs_free_path(path); | |
5401 | return err; | |
39279cc3 CM |
5402 | } |
5403 | ||
5d4f98a2 YZ |
5404 | static void inode_tree_add(struct inode *inode) |
5405 | { | |
5406 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5407 | struct btrfs_inode *entry; | |
03e860bd FNP |
5408 | struct rb_node **p; |
5409 | struct rb_node *parent; | |
cef21937 | 5410 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
33345d01 | 5411 | u64 ino = btrfs_ino(inode); |
5d4f98a2 | 5412 | |
1d3382cb | 5413 | if (inode_unhashed(inode)) |
76dda93c | 5414 | return; |
e1409cef | 5415 | parent = NULL; |
5d4f98a2 | 5416 | spin_lock(&root->inode_lock); |
e1409cef | 5417 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5418 | while (*p) { |
5419 | parent = *p; | |
5420 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5421 | ||
33345d01 | 5422 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 5423 | p = &parent->rb_left; |
33345d01 | 5424 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 5425 | p = &parent->rb_right; |
5d4f98a2 YZ |
5426 | else { |
5427 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5428 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5429 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5430 | RB_CLEAR_NODE(parent); |
5431 | spin_unlock(&root->inode_lock); | |
cef21937 | 5432 | return; |
5d4f98a2 YZ |
5433 | } |
5434 | } | |
cef21937 FDBM |
5435 | rb_link_node(new, parent, p); |
5436 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5437 | spin_unlock(&root->inode_lock); |
5438 | } | |
5439 | ||
5440 | static void inode_tree_del(struct inode *inode) | |
5441 | { | |
5442 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 5443 | int empty = 0; |
5d4f98a2 | 5444 | |
03e860bd | 5445 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5446 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5447 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5448 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5449 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5450 | } |
03e860bd | 5451 | spin_unlock(&root->inode_lock); |
76dda93c | 5452 | |
69e9c6c6 | 5453 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5454 | synchronize_srcu(&root->fs_info->subvol_srcu); |
5455 | spin_lock(&root->inode_lock); | |
5456 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5457 | spin_unlock(&root->inode_lock); | |
5458 | if (empty) | |
5459 | btrfs_add_dead_root(root); | |
5460 | } | |
5461 | } | |
5462 | ||
143bede5 | 5463 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
5464 | { |
5465 | struct rb_node *node; | |
5466 | struct rb_node *prev; | |
5467 | struct btrfs_inode *entry; | |
5468 | struct inode *inode; | |
5469 | u64 objectid = 0; | |
5470 | ||
7813b3db LB |
5471 | if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
5472 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
76dda93c YZ |
5473 | |
5474 | spin_lock(&root->inode_lock); | |
5475 | again: | |
5476 | node = root->inode_tree.rb_node; | |
5477 | prev = NULL; | |
5478 | while (node) { | |
5479 | prev = node; | |
5480 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
5481 | ||
33345d01 | 5482 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 5483 | node = node->rb_left; |
33345d01 | 5484 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
5485 | node = node->rb_right; |
5486 | else | |
5487 | break; | |
5488 | } | |
5489 | if (!node) { | |
5490 | while (prev) { | |
5491 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 5492 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
5493 | node = prev; |
5494 | break; | |
5495 | } | |
5496 | prev = rb_next(prev); | |
5497 | } | |
5498 | } | |
5499 | while (node) { | |
5500 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 5501 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
5502 | inode = igrab(&entry->vfs_inode); |
5503 | if (inode) { | |
5504 | spin_unlock(&root->inode_lock); | |
5505 | if (atomic_read(&inode->i_count) > 1) | |
5506 | d_prune_aliases(inode); | |
5507 | /* | |
45321ac5 | 5508 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
5509 | * the inode cache when its usage count |
5510 | * hits zero. | |
5511 | */ | |
5512 | iput(inode); | |
5513 | cond_resched(); | |
5514 | spin_lock(&root->inode_lock); | |
5515 | goto again; | |
5516 | } | |
5517 | ||
5518 | if (cond_resched_lock(&root->inode_lock)) | |
5519 | goto again; | |
5520 | ||
5521 | node = rb_next(node); | |
5522 | } | |
5523 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
5524 | } |
5525 | ||
e02119d5 CM |
5526 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5527 | { | |
5528 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5529 | inode->i_ino = args->location->objectid; |
5530 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5531 | sizeof(*args->location)); | |
e02119d5 | 5532 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5533 | return 0; |
5534 | } | |
5535 | ||
5536 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5537 | { | |
5538 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5539 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5540 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5541 | } |
5542 | ||
5d4f98a2 | 5543 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5544 | struct btrfs_key *location, |
5d4f98a2 | 5545 | struct btrfs_root *root) |
39279cc3 CM |
5546 | { |
5547 | struct inode *inode; | |
5548 | struct btrfs_iget_args args; | |
90d3e592 | 5549 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5550 | |
90d3e592 | 5551 | args.location = location; |
39279cc3 CM |
5552 | args.root = root; |
5553 | ||
778ba82b | 5554 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5555 | btrfs_init_locked_inode, |
5556 | (void *)&args); | |
5557 | return inode; | |
5558 | } | |
5559 | ||
1a54ef8c BR |
5560 | /* Get an inode object given its location and corresponding root. |
5561 | * Returns in *is_new if the inode was read from disk | |
5562 | */ | |
5563 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 5564 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
5565 | { |
5566 | struct inode *inode; | |
5567 | ||
90d3e592 | 5568 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5569 | if (!inode) |
5d4f98a2 | 5570 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5571 | |
5572 | if (inode->i_state & I_NEW) { | |
1a54ef8c | 5573 | btrfs_read_locked_inode(inode); |
1748f843 MF |
5574 | if (!is_bad_inode(inode)) { |
5575 | inode_tree_add(inode); | |
5576 | unlock_new_inode(inode); | |
5577 | if (new) | |
5578 | *new = 1; | |
5579 | } else { | |
e0b6d65b ST |
5580 | unlock_new_inode(inode); |
5581 | iput(inode); | |
5582 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
5583 | } |
5584 | } | |
5585 | ||
1a54ef8c BR |
5586 | return inode; |
5587 | } | |
5588 | ||
4df27c4d YZ |
5589 | static struct inode *new_simple_dir(struct super_block *s, |
5590 | struct btrfs_key *key, | |
5591 | struct btrfs_root *root) | |
5592 | { | |
5593 | struct inode *inode = new_inode(s); | |
5594 | ||
5595 | if (!inode) | |
5596 | return ERR_PTR(-ENOMEM); | |
5597 | ||
4df27c4d YZ |
5598 | BTRFS_I(inode)->root = root; |
5599 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5600 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5601 | |
5602 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5603 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
5604 | inode->i_fop = &simple_dir_operations; |
5605 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
04b285f3 | 5606 | inode->i_mtime = current_fs_time(inode->i_sb); |
9cc97d64 | 5607 | inode->i_atime = inode->i_mtime; |
5608 | inode->i_ctime = inode->i_mtime; | |
5609 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
4df27c4d YZ |
5610 | |
5611 | return inode; | |
5612 | } | |
5613 | ||
3de4586c | 5614 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5615 | { |
d397712b | 5616 | struct inode *inode; |
4df27c4d | 5617 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5618 | struct btrfs_root *sub_root = root; |
5619 | struct btrfs_key location; | |
76dda93c | 5620 | int index; |
b4aff1f8 | 5621 | int ret = 0; |
39279cc3 CM |
5622 | |
5623 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5624 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5625 | |
39e3c955 | 5626 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
5627 | if (ret < 0) |
5628 | return ERR_PTR(ret); | |
5f39d397 | 5629 | |
4df27c4d | 5630 | if (location.objectid == 0) |
5662344b | 5631 | return ERR_PTR(-ENOENT); |
4df27c4d YZ |
5632 | |
5633 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 5634 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
5635 | return inode; |
5636 | } | |
5637 | ||
5638 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
5639 | ||
76dda93c | 5640 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
5641 | ret = fixup_tree_root_location(root, dir, dentry, |
5642 | &location, &sub_root); | |
5643 | if (ret < 0) { | |
5644 | if (ret != -ENOENT) | |
5645 | inode = ERR_PTR(ret); | |
5646 | else | |
5647 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5648 | } else { | |
73f73415 | 5649 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5650 | } |
76dda93c YZ |
5651 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
5652 | ||
34d19bad | 5653 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
5654 | down_read(&root->fs_info->cleanup_work_sem); |
5655 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 5656 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 5657 | up_read(&root->fs_info->cleanup_work_sem); |
01cd3367 JB |
5658 | if (ret) { |
5659 | iput(inode); | |
66b4ffd1 | 5660 | inode = ERR_PTR(ret); |
01cd3367 | 5661 | } |
c71bf099 YZ |
5662 | } |
5663 | ||
3de4586c CM |
5664 | return inode; |
5665 | } | |
5666 | ||
fe15ce44 | 5667 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5668 | { |
5669 | struct btrfs_root *root; | |
2b0143b5 | 5670 | struct inode *inode = d_inode(dentry); |
76dda93c | 5671 | |
848cce0d | 5672 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5673 | inode = d_inode(dentry->d_parent); |
76dda93c | 5674 | |
848cce0d LZ |
5675 | if (inode) { |
5676 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5677 | if (btrfs_root_refs(&root->root_item) == 0) |
5678 | return 1; | |
848cce0d LZ |
5679 | |
5680 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
5681 | return 1; | |
efefb143 | 5682 | } |
76dda93c YZ |
5683 | return 0; |
5684 | } | |
5685 | ||
b4aff1f8 JB |
5686 | static void btrfs_dentry_release(struct dentry *dentry) |
5687 | { | |
944a4515 | 5688 | kfree(dentry->d_fsdata); |
b4aff1f8 JB |
5689 | } |
5690 | ||
3de4586c | 5691 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5692 | unsigned int flags) |
3de4586c | 5693 | { |
5662344b | 5694 | struct inode *inode; |
a66e7cc6 | 5695 | |
5662344b TI |
5696 | inode = btrfs_lookup_dentry(dir, dentry); |
5697 | if (IS_ERR(inode)) { | |
5698 | if (PTR_ERR(inode) == -ENOENT) | |
5699 | inode = NULL; | |
5700 | else | |
5701 | return ERR_CAST(inode); | |
5702 | } | |
5703 | ||
41d28bca | 5704 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5705 | } |
5706 | ||
16cdcec7 | 5707 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
5708 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
5709 | }; | |
5710 | ||
9cdda8d3 | 5711 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5712 | { |
9cdda8d3 | 5713 | struct inode *inode = file_inode(file); |
39279cc3 CM |
5714 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5715 | struct btrfs_item *item; | |
5716 | struct btrfs_dir_item *di; | |
5717 | struct btrfs_key key; | |
5f39d397 | 5718 | struct btrfs_key found_key; |
39279cc3 | 5719 | struct btrfs_path *path; |
16cdcec7 MX |
5720 | struct list_head ins_list; |
5721 | struct list_head del_list; | |
39279cc3 | 5722 | int ret; |
5f39d397 | 5723 | struct extent_buffer *leaf; |
39279cc3 | 5724 | int slot; |
39279cc3 CM |
5725 | unsigned char d_type; |
5726 | int over = 0; | |
5727 | u32 di_cur; | |
5728 | u32 di_total; | |
5729 | u32 di_len; | |
5730 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
5731 | char tmp_name[32]; |
5732 | char *name_ptr; | |
5733 | int name_len; | |
9cdda8d3 | 5734 | int is_curr = 0; /* ctx->pos points to the current index? */ |
bc4ef759 | 5735 | bool emitted; |
39279cc3 CM |
5736 | |
5737 | /* FIXME, use a real flag for deciding about the key type */ | |
5738 | if (root->fs_info->tree_root == root) | |
5739 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 5740 | |
9cdda8d3 AV |
5741 | if (!dir_emit_dots(file, ctx)) |
5742 | return 0; | |
5743 | ||
49593bfa | 5744 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5745 | if (!path) |
5746 | return -ENOMEM; | |
ff5714cc | 5747 | |
e4058b54 | 5748 | path->reada = READA_FORWARD; |
49593bfa | 5749 | |
16cdcec7 MX |
5750 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5751 | INIT_LIST_HEAD(&ins_list); | |
5752 | INIT_LIST_HEAD(&del_list); | |
5753 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
5754 | } | |
5755 | ||
962a298f | 5756 | key.type = key_type; |
9cdda8d3 | 5757 | key.offset = ctx->pos; |
33345d01 | 5758 | key.objectid = btrfs_ino(inode); |
5f39d397 | 5759 | |
39279cc3 CM |
5760 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5761 | if (ret < 0) | |
5762 | goto err; | |
49593bfa | 5763 | |
bc4ef759 | 5764 | emitted = false; |
49593bfa | 5765 | while (1) { |
5f39d397 | 5766 | leaf = path->nodes[0]; |
39279cc3 | 5767 | slot = path->slots[0]; |
b9e03af0 LZ |
5768 | if (slot >= btrfs_header_nritems(leaf)) { |
5769 | ret = btrfs_next_leaf(root, path); | |
5770 | if (ret < 0) | |
5771 | goto err; | |
5772 | else if (ret > 0) | |
5773 | break; | |
5774 | continue; | |
39279cc3 | 5775 | } |
3de4586c | 5776 | |
dd3cc16b | 5777 | item = btrfs_item_nr(slot); |
5f39d397 CM |
5778 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5779 | ||
5780 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5781 | break; |
962a298f | 5782 | if (found_key.type != key_type) |
39279cc3 | 5783 | break; |
9cdda8d3 | 5784 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5785 | goto next; |
16cdcec7 MX |
5786 | if (key_type == BTRFS_DIR_INDEX_KEY && |
5787 | btrfs_should_delete_dir_index(&del_list, | |
5788 | found_key.offset)) | |
5789 | goto next; | |
5f39d397 | 5790 | |
9cdda8d3 | 5791 | ctx->pos = found_key.offset; |
16cdcec7 | 5792 | is_curr = 1; |
49593bfa | 5793 | |
39279cc3 CM |
5794 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
5795 | di_cur = 0; | |
5f39d397 | 5796 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
5797 | |
5798 | while (di_cur < di_total) { | |
5f39d397 CM |
5799 | struct btrfs_key location; |
5800 | ||
22a94d44 JB |
5801 | if (verify_dir_item(root, leaf, di)) |
5802 | break; | |
5803 | ||
5f39d397 | 5804 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 5805 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
5806 | name_ptr = tmp_name; |
5807 | } else { | |
49e350a4 | 5808 | name_ptr = kmalloc(name_len, GFP_KERNEL); |
49593bfa DW |
5809 | if (!name_ptr) { |
5810 | ret = -ENOMEM; | |
5811 | goto err; | |
5812 | } | |
5f39d397 CM |
5813 | } |
5814 | read_extent_buffer(leaf, name_ptr, | |
5815 | (unsigned long)(di + 1), name_len); | |
5816 | ||
5817 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
5818 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 5819 | |
fede766f | 5820 | |
3de4586c | 5821 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
5822 | * skip it. |
5823 | * | |
5824 | * In contrast to old kernels, we insert the snapshot's | |
5825 | * dir item and dir index after it has been created, so | |
5826 | * we won't find a reference to our own snapshot. We | |
5827 | * still keep the following code for backward | |
5828 | * compatibility. | |
3de4586c CM |
5829 | */ |
5830 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
5831 | location.objectid == root->root_key.objectid) { | |
5832 | over = 0; | |
5833 | goto skip; | |
5834 | } | |
9cdda8d3 AV |
5835 | over = !dir_emit(ctx, name_ptr, name_len, |
5836 | location.objectid, d_type); | |
5f39d397 | 5837 | |
3de4586c | 5838 | skip: |
5f39d397 CM |
5839 | if (name_ptr != tmp_name) |
5840 | kfree(name_ptr); | |
5841 | ||
39279cc3 CM |
5842 | if (over) |
5843 | goto nopos; | |
bc4ef759 | 5844 | emitted = true; |
5103e947 | 5845 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 5846 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
5847 | di_cur += di_len; |
5848 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
5849 | } | |
b9e03af0 LZ |
5850 | next: |
5851 | path->slots[0]++; | |
39279cc3 | 5852 | } |
49593bfa | 5853 | |
16cdcec7 MX |
5854 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5855 | if (is_curr) | |
9cdda8d3 | 5856 | ctx->pos++; |
bc4ef759 | 5857 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted); |
16cdcec7 MX |
5858 | if (ret) |
5859 | goto nopos; | |
5860 | } | |
5861 | ||
bc4ef759 DS |
5862 | /* |
5863 | * If we haven't emitted any dir entry, we must not touch ctx->pos as | |
5864 | * it was was set to the termination value in previous call. We assume | |
5865 | * that "." and ".." were emitted if we reach this point and set the | |
5866 | * termination value as well for an empty directory. | |
5867 | */ | |
5868 | if (ctx->pos > 2 && !emitted) | |
5869 | goto nopos; | |
5870 | ||
49593bfa | 5871 | /* Reached end of directory/root. Bump pos past the last item. */ |
db62efbb ZB |
5872 | ctx->pos++; |
5873 | ||
5874 | /* | |
5875 | * Stop new entries from being returned after we return the last | |
5876 | * entry. | |
5877 | * | |
5878 | * New directory entries are assigned a strictly increasing | |
5879 | * offset. This means that new entries created during readdir | |
5880 | * are *guaranteed* to be seen in the future by that readdir. | |
5881 | * This has broken buggy programs which operate on names as | |
5882 | * they're returned by readdir. Until we re-use freed offsets | |
5883 | * we have this hack to stop new entries from being returned | |
5884 | * under the assumption that they'll never reach this huge | |
5885 | * offset. | |
5886 | * | |
5887 | * This is being careful not to overflow 32bit loff_t unless the | |
5888 | * last entry requires it because doing so has broken 32bit apps | |
5889 | * in the past. | |
5890 | */ | |
5891 | if (key_type == BTRFS_DIR_INDEX_KEY) { | |
5892 | if (ctx->pos >= INT_MAX) | |
5893 | ctx->pos = LLONG_MAX; | |
5894 | else | |
5895 | ctx->pos = INT_MAX; | |
5896 | } | |
39279cc3 CM |
5897 | nopos: |
5898 | ret = 0; | |
5899 | err: | |
16cdcec7 MX |
5900 | if (key_type == BTRFS_DIR_INDEX_KEY) |
5901 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 5902 | btrfs_free_path(path); |
39279cc3 CM |
5903 | return ret; |
5904 | } | |
5905 | ||
a9185b41 | 5906 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
5907 | { |
5908 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5909 | struct btrfs_trans_handle *trans; | |
5910 | int ret = 0; | |
0af3d00b | 5911 | bool nolock = false; |
39279cc3 | 5912 | |
72ac3c0d | 5913 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
5914 | return 0; |
5915 | ||
83eea1f1 | 5916 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 5917 | nolock = true; |
0af3d00b | 5918 | |
a9185b41 | 5919 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 5920 | if (nolock) |
7a7eaa40 | 5921 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 5922 | else |
7a7eaa40 | 5923 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5924 | if (IS_ERR(trans)) |
5925 | return PTR_ERR(trans); | |
a698d075 | 5926 | ret = btrfs_commit_transaction(trans, root); |
39279cc3 CM |
5927 | } |
5928 | return ret; | |
5929 | } | |
5930 | ||
5931 | /* | |
54aa1f4d | 5932 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5933 | * inode changes. But, it is most likely to find the inode in cache. |
5934 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5935 | * to keep or drop this code. | |
5936 | */ | |
48a3b636 | 5937 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
5938 | { |
5939 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5940 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5941 | int ret; |
5942 | ||
72ac3c0d | 5943 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5944 | return 0; |
39279cc3 | 5945 | |
7a7eaa40 | 5946 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5947 | if (IS_ERR(trans)) |
5948 | return PTR_ERR(trans); | |
8929ecfa YZ |
5949 | |
5950 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
5951 | if (ret && ret == -ENOSPC) { |
5952 | /* whoops, lets try again with the full transaction */ | |
5953 | btrfs_end_transaction(trans, root); | |
5954 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
5955 | if (IS_ERR(trans)) |
5956 | return PTR_ERR(trans); | |
8929ecfa | 5957 | |
94b60442 | 5958 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 5959 | } |
39279cc3 | 5960 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
5961 | if (BTRFS_I(inode)->delayed_node) |
5962 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
5963 | |
5964 | return ret; | |
5965 | } | |
5966 | ||
5967 | /* | |
5968 | * This is a copy of file_update_time. We need this so we can return error on | |
5969 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
5970 | */ | |
e41f941a JB |
5971 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
5972 | int flags) | |
22c44fe6 | 5973 | { |
2bc55652 AB |
5974 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5975 | ||
5976 | if (btrfs_root_readonly(root)) | |
5977 | return -EROFS; | |
5978 | ||
e41f941a | 5979 | if (flags & S_VERSION) |
22c44fe6 | 5980 | inode_inc_iversion(inode); |
e41f941a JB |
5981 | if (flags & S_CTIME) |
5982 | inode->i_ctime = *now; | |
5983 | if (flags & S_MTIME) | |
5984 | inode->i_mtime = *now; | |
5985 | if (flags & S_ATIME) | |
5986 | inode->i_atime = *now; | |
5987 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
5988 | } |
5989 | ||
d352ac68 CM |
5990 | /* |
5991 | * find the highest existing sequence number in a directory | |
5992 | * and then set the in-memory index_cnt variable to reflect | |
5993 | * free sequence numbers | |
5994 | */ | |
aec7477b JB |
5995 | static int btrfs_set_inode_index_count(struct inode *inode) |
5996 | { | |
5997 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5998 | struct btrfs_key key, found_key; | |
5999 | struct btrfs_path *path; | |
6000 | struct extent_buffer *leaf; | |
6001 | int ret; | |
6002 | ||
33345d01 | 6003 | key.objectid = btrfs_ino(inode); |
962a298f | 6004 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
6005 | key.offset = (u64)-1; |
6006 | ||
6007 | path = btrfs_alloc_path(); | |
6008 | if (!path) | |
6009 | return -ENOMEM; | |
6010 | ||
6011 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6012 | if (ret < 0) | |
6013 | goto out; | |
6014 | /* FIXME: we should be able to handle this */ | |
6015 | if (ret == 0) | |
6016 | goto out; | |
6017 | ret = 0; | |
6018 | ||
6019 | /* | |
6020 | * MAGIC NUMBER EXPLANATION: | |
6021 | * since we search a directory based on f_pos we have to start at 2 | |
6022 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
6023 | * else has to start at 2 | |
6024 | */ | |
6025 | if (path->slots[0] == 0) { | |
6026 | BTRFS_I(inode)->index_cnt = 2; | |
6027 | goto out; | |
6028 | } | |
6029 | ||
6030 | path->slots[0]--; | |
6031 | ||
6032 | leaf = path->nodes[0]; | |
6033 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6034 | ||
33345d01 | 6035 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 6036 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
aec7477b JB |
6037 | BTRFS_I(inode)->index_cnt = 2; |
6038 | goto out; | |
6039 | } | |
6040 | ||
6041 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
6042 | out: | |
6043 | btrfs_free_path(path); | |
6044 | return ret; | |
6045 | } | |
6046 | ||
d352ac68 CM |
6047 | /* |
6048 | * helper to find a free sequence number in a given directory. This current | |
6049 | * code is very simple, later versions will do smarter things in the btree | |
6050 | */ | |
3de4586c | 6051 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
6052 | { |
6053 | int ret = 0; | |
6054 | ||
6055 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
6056 | ret = btrfs_inode_delayed_dir_index_count(dir); |
6057 | if (ret) { | |
6058 | ret = btrfs_set_inode_index_count(dir); | |
6059 | if (ret) | |
6060 | return ret; | |
6061 | } | |
aec7477b JB |
6062 | } |
6063 | ||
00e4e6b3 | 6064 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
6065 | BTRFS_I(dir)->index_cnt++; |
6066 | ||
6067 | return ret; | |
6068 | } | |
6069 | ||
b0d5d10f CM |
6070 | static int btrfs_insert_inode_locked(struct inode *inode) |
6071 | { | |
6072 | struct btrfs_iget_args args; | |
6073 | args.location = &BTRFS_I(inode)->location; | |
6074 | args.root = BTRFS_I(inode)->root; | |
6075 | ||
6076 | return insert_inode_locked4(inode, | |
6077 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6078 | btrfs_find_actor, &args); | |
6079 | } | |
6080 | ||
39279cc3 CM |
6081 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6082 | struct btrfs_root *root, | |
aec7477b | 6083 | struct inode *dir, |
9c58309d | 6084 | const char *name, int name_len, |
175a4eb7 AV |
6085 | u64 ref_objectid, u64 objectid, |
6086 | umode_t mode, u64 *index) | |
39279cc3 CM |
6087 | { |
6088 | struct inode *inode; | |
5f39d397 | 6089 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6090 | struct btrfs_key *location; |
5f39d397 | 6091 | struct btrfs_path *path; |
9c58309d CM |
6092 | struct btrfs_inode_ref *ref; |
6093 | struct btrfs_key key[2]; | |
6094 | u32 sizes[2]; | |
ef3b9af5 | 6095 | int nitems = name ? 2 : 1; |
9c58309d | 6096 | unsigned long ptr; |
39279cc3 | 6097 | int ret; |
39279cc3 | 6098 | |
5f39d397 | 6099 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6100 | if (!path) |
6101 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6102 | |
39279cc3 | 6103 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
6104 | if (!inode) { |
6105 | btrfs_free_path(path); | |
39279cc3 | 6106 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6107 | } |
39279cc3 | 6108 | |
5762b5c9 FM |
6109 | /* |
6110 | * O_TMPFILE, set link count to 0, so that after this point, | |
6111 | * we fill in an inode item with the correct link count. | |
6112 | */ | |
6113 | if (!name) | |
6114 | set_nlink(inode, 0); | |
6115 | ||
581bb050 LZ |
6116 | /* |
6117 | * we have to initialize this early, so we can reclaim the inode | |
6118 | * number if we fail afterwards in this function. | |
6119 | */ | |
6120 | inode->i_ino = objectid; | |
6121 | ||
ef3b9af5 | 6122 | if (dir && name) { |
1abe9b8a | 6123 | trace_btrfs_inode_request(dir); |
6124 | ||
3de4586c | 6125 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 6126 | if (ret) { |
8fb27640 | 6127 | btrfs_free_path(path); |
09771430 | 6128 | iput(inode); |
aec7477b | 6129 | return ERR_PTR(ret); |
09771430 | 6130 | } |
ef3b9af5 FM |
6131 | } else if (dir) { |
6132 | *index = 0; | |
aec7477b JB |
6133 | } |
6134 | /* | |
6135 | * index_cnt is ignored for everything but a dir, | |
6136 | * btrfs_get_inode_index_count has an explanation for the magic | |
6137 | * number | |
6138 | */ | |
6139 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6140 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 6141 | BTRFS_I(inode)->root = root; |
e02119d5 | 6142 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6143 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6144 | |
5dc562c5 JB |
6145 | /* |
6146 | * We could have gotten an inode number from somebody who was fsynced | |
6147 | * and then removed in this same transaction, so let's just set full | |
6148 | * sync since it will be a full sync anyway and this will blow away the | |
6149 | * old info in the log. | |
6150 | */ | |
6151 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
6152 | ||
9c58309d | 6153 | key[0].objectid = objectid; |
962a298f | 6154 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6155 | key[0].offset = 0; |
6156 | ||
9c58309d | 6157 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6158 | |
6159 | if (name) { | |
6160 | /* | |
6161 | * Start new inodes with an inode_ref. This is slightly more | |
6162 | * efficient for small numbers of hard links since they will | |
6163 | * be packed into one item. Extended refs will kick in if we | |
6164 | * add more hard links than can fit in the ref item. | |
6165 | */ | |
6166 | key[1].objectid = objectid; | |
962a298f | 6167 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
6168 | key[1].offset = ref_objectid; |
6169 | ||
6170 | sizes[1] = name_len + sizeof(*ref); | |
6171 | } | |
9c58309d | 6172 | |
b0d5d10f CM |
6173 | location = &BTRFS_I(inode)->location; |
6174 | location->objectid = objectid; | |
6175 | location->offset = 0; | |
962a298f | 6176 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6177 | |
6178 | ret = btrfs_insert_inode_locked(inode); | |
6179 | if (ret < 0) | |
6180 | goto fail; | |
6181 | ||
b9473439 | 6182 | path->leave_spinning = 1; |
ef3b9af5 | 6183 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6184 | if (ret != 0) |
b0d5d10f | 6185 | goto fail_unlock; |
5f39d397 | 6186 | |
ecc11fab | 6187 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 6188 | inode_set_bytes(inode, 0); |
9cc97d64 | 6189 | |
04b285f3 | 6190 | inode->i_mtime = current_fs_time(inode->i_sb); |
9cc97d64 | 6191 | inode->i_atime = inode->i_mtime; |
6192 | inode->i_ctime = inode->i_mtime; | |
6193 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
6194 | ||
5f39d397 CM |
6195 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6196 | struct btrfs_inode_item); | |
293f7e07 LZ |
6197 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
6198 | sizeof(*inode_item)); | |
e02119d5 | 6199 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6200 | |
ef3b9af5 FM |
6201 | if (name) { |
6202 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6203 | struct btrfs_inode_ref); | |
6204 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6205 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6206 | ptr = (unsigned long)(ref + 1); | |
6207 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6208 | } | |
9c58309d | 6209 | |
5f39d397 CM |
6210 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6211 | btrfs_free_path(path); | |
6212 | ||
6cbff00f CH |
6213 | btrfs_inherit_iflags(inode, dir); |
6214 | ||
569254b0 | 6215 | if (S_ISREG(mode)) { |
94272164 CM |
6216 | if (btrfs_test_opt(root, NODATASUM)) |
6217 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
213490b3 | 6218 | if (btrfs_test_opt(root, NODATACOW)) |
f2bdf9a8 JB |
6219 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6220 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6221 | } |
6222 | ||
5d4f98a2 | 6223 | inode_tree_add(inode); |
1abe9b8a | 6224 | |
6225 | trace_btrfs_inode_new(inode); | |
1973f0fa | 6226 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 6227 | |
8ea05e3a AB |
6228 | btrfs_update_root_times(trans, root); |
6229 | ||
63541927 FDBM |
6230 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6231 | if (ret) | |
6232 | btrfs_err(root->fs_info, | |
6233 | "error inheriting props for ino %llu (root %llu): %d", | |
6234 | btrfs_ino(inode), root->root_key.objectid, ret); | |
6235 | ||
39279cc3 | 6236 | return inode; |
b0d5d10f CM |
6237 | |
6238 | fail_unlock: | |
6239 | unlock_new_inode(inode); | |
5f39d397 | 6240 | fail: |
ef3b9af5 | 6241 | if (dir && name) |
aec7477b | 6242 | BTRFS_I(dir)->index_cnt--; |
5f39d397 | 6243 | btrfs_free_path(path); |
09771430 | 6244 | iput(inode); |
5f39d397 | 6245 | return ERR_PTR(ret); |
39279cc3 CM |
6246 | } |
6247 | ||
6248 | static inline u8 btrfs_inode_type(struct inode *inode) | |
6249 | { | |
6250 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
6251 | } | |
6252 | ||
d352ac68 CM |
6253 | /* |
6254 | * utility function to add 'inode' into 'parent_inode' with | |
6255 | * a give name and a given sequence number. | |
6256 | * if 'add_backref' is true, also insert a backref from the | |
6257 | * inode to the parent directory. | |
6258 | */ | |
e02119d5 CM |
6259 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
6260 | struct inode *parent_inode, struct inode *inode, | |
6261 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 6262 | { |
4df27c4d | 6263 | int ret = 0; |
39279cc3 | 6264 | struct btrfs_key key; |
e02119d5 | 6265 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
6266 | u64 ino = btrfs_ino(inode); |
6267 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6268 | |
33345d01 | 6269 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
6270 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
6271 | } else { | |
33345d01 | 6272 | key.objectid = ino; |
962a298f | 6273 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6274 | key.offset = 0; |
6275 | } | |
6276 | ||
33345d01 | 6277 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
6278 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
6279 | key.objectid, root->root_key.objectid, | |
33345d01 | 6280 | parent_ino, index, name, name_len); |
4df27c4d | 6281 | } else if (add_backref) { |
33345d01 LZ |
6282 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6283 | parent_ino, index); | |
4df27c4d | 6284 | } |
39279cc3 | 6285 | |
79787eaa JM |
6286 | /* Nothing to clean up yet */ |
6287 | if (ret) | |
6288 | return ret; | |
4df27c4d | 6289 | |
79787eaa JM |
6290 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
6291 | parent_inode, &key, | |
6292 | btrfs_inode_type(inode), index); | |
9c52057c | 6293 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6294 | goto fail_dir_item; |
6295 | else if (ret) { | |
6296 | btrfs_abort_transaction(trans, root, ret); | |
6297 | return ret; | |
39279cc3 | 6298 | } |
79787eaa JM |
6299 | |
6300 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
6301 | name_len * 2); | |
0c4d2d95 | 6302 | inode_inc_iversion(parent_inode); |
04b285f3 DD |
6303 | parent_inode->i_mtime = parent_inode->i_ctime = |
6304 | current_fs_time(parent_inode->i_sb); | |
79787eaa JM |
6305 | ret = btrfs_update_inode(trans, root, parent_inode); |
6306 | if (ret) | |
6307 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 6308 | return ret; |
fe66a05a CM |
6309 | |
6310 | fail_dir_item: | |
6311 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6312 | u64 local_index; | |
6313 | int err; | |
6314 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
6315 | key.objectid, root->root_key.objectid, | |
6316 | parent_ino, &local_index, name, name_len); | |
6317 | ||
6318 | } else if (add_backref) { | |
6319 | u64 local_index; | |
6320 | int err; | |
6321 | ||
6322 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6323 | ino, parent_ino, &local_index); | |
6324 | } | |
6325 | return ret; | |
39279cc3 CM |
6326 | } |
6327 | ||
6328 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
6329 | struct inode *dir, struct dentry *dentry, |
6330 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 6331 | { |
a1b075d2 JB |
6332 | int err = btrfs_add_link(trans, dir, inode, |
6333 | dentry->d_name.name, dentry->d_name.len, | |
6334 | backref, index); | |
39279cc3 CM |
6335 | if (err > 0) |
6336 | err = -EEXIST; | |
6337 | return err; | |
6338 | } | |
6339 | ||
618e21d5 | 6340 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6341 | umode_t mode, dev_t rdev) |
618e21d5 JB |
6342 | { |
6343 | struct btrfs_trans_handle *trans; | |
6344 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6345 | struct inode *inode = NULL; |
618e21d5 JB |
6346 | int err; |
6347 | int drop_inode = 0; | |
6348 | u64 objectid; | |
00e4e6b3 | 6349 | u64 index = 0; |
618e21d5 | 6350 | |
9ed74f2d JB |
6351 | /* |
6352 | * 2 for inode item and ref | |
6353 | * 2 for dir items | |
6354 | * 1 for xattr if selinux is on | |
6355 | */ | |
a22285a6 YZ |
6356 | trans = btrfs_start_transaction(root, 5); |
6357 | if (IS_ERR(trans)) | |
6358 | return PTR_ERR(trans); | |
1832a6d5 | 6359 | |
581bb050 LZ |
6360 | err = btrfs_find_free_ino(root, &objectid); |
6361 | if (err) | |
6362 | goto out_unlock; | |
6363 | ||
aec7477b | 6364 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6365 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6366 | mode, &index); |
7cf96da3 TI |
6367 | if (IS_ERR(inode)) { |
6368 | err = PTR_ERR(inode); | |
618e21d5 | 6369 | goto out_unlock; |
7cf96da3 | 6370 | } |
618e21d5 | 6371 | |
ad19db71 CS |
6372 | /* |
6373 | * If the active LSM wants to access the inode during | |
6374 | * d_instantiate it needs these. Smack checks to see | |
6375 | * if the filesystem supports xattrs by looking at the | |
6376 | * ops vector. | |
6377 | */ | |
ad19db71 | 6378 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6379 | init_special_inode(inode, inode->i_mode, rdev); |
6380 | ||
6381 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6382 | if (err) |
b0d5d10f CM |
6383 | goto out_unlock_inode; |
6384 | ||
6385 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); | |
6386 | if (err) { | |
6387 | goto out_unlock_inode; | |
6388 | } else { | |
1b4ab1bb | 6389 | btrfs_update_inode(trans, root, inode); |
b0d5d10f | 6390 | unlock_new_inode(inode); |
08c422c2 | 6391 | d_instantiate(dentry, inode); |
618e21d5 | 6392 | } |
b0d5d10f | 6393 | |
618e21d5 | 6394 | out_unlock: |
7ad85bb7 | 6395 | btrfs_end_transaction(trans, root); |
c581afc8 | 6396 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6397 | btrfs_btree_balance_dirty(root); |
618e21d5 JB |
6398 | if (drop_inode) { |
6399 | inode_dec_link_count(inode); | |
6400 | iput(inode); | |
6401 | } | |
618e21d5 | 6402 | return err; |
b0d5d10f CM |
6403 | |
6404 | out_unlock_inode: | |
6405 | drop_inode = 1; | |
6406 | unlock_new_inode(inode); | |
6407 | goto out_unlock; | |
6408 | ||
618e21d5 JB |
6409 | } |
6410 | ||
39279cc3 | 6411 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6412 | umode_t mode, bool excl) |
39279cc3 CM |
6413 | { |
6414 | struct btrfs_trans_handle *trans; | |
6415 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6416 | struct inode *inode = NULL; |
43baa579 | 6417 | int drop_inode_on_err = 0; |
a22285a6 | 6418 | int err; |
39279cc3 | 6419 | u64 objectid; |
00e4e6b3 | 6420 | u64 index = 0; |
39279cc3 | 6421 | |
9ed74f2d JB |
6422 | /* |
6423 | * 2 for inode item and ref | |
6424 | * 2 for dir items | |
6425 | * 1 for xattr if selinux is on | |
6426 | */ | |
a22285a6 YZ |
6427 | trans = btrfs_start_transaction(root, 5); |
6428 | if (IS_ERR(trans)) | |
6429 | return PTR_ERR(trans); | |
9ed74f2d | 6430 | |
581bb050 LZ |
6431 | err = btrfs_find_free_ino(root, &objectid); |
6432 | if (err) | |
6433 | goto out_unlock; | |
6434 | ||
aec7477b | 6435 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6436 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6437 | mode, &index); |
7cf96da3 TI |
6438 | if (IS_ERR(inode)) { |
6439 | err = PTR_ERR(inode); | |
39279cc3 | 6440 | goto out_unlock; |
7cf96da3 | 6441 | } |
43baa579 | 6442 | drop_inode_on_err = 1; |
ad19db71 CS |
6443 | /* |
6444 | * If the active LSM wants to access the inode during | |
6445 | * d_instantiate it needs these. Smack checks to see | |
6446 | * if the filesystem supports xattrs by looking at the | |
6447 | * ops vector. | |
6448 | */ | |
6449 | inode->i_fop = &btrfs_file_operations; | |
6450 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6451 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6452 | |
6453 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6454 | if (err) | |
6455 | goto out_unlock_inode; | |
6456 | ||
6457 | err = btrfs_update_inode(trans, root, inode); | |
6458 | if (err) | |
6459 | goto out_unlock_inode; | |
ad19db71 | 6460 | |
a1b075d2 | 6461 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 6462 | if (err) |
b0d5d10f | 6463 | goto out_unlock_inode; |
43baa579 | 6464 | |
43baa579 | 6465 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
b0d5d10f | 6466 | unlock_new_inode(inode); |
43baa579 FB |
6467 | d_instantiate(dentry, inode); |
6468 | ||
39279cc3 | 6469 | out_unlock: |
7ad85bb7 | 6470 | btrfs_end_transaction(trans, root); |
43baa579 | 6471 | if (err && drop_inode_on_err) { |
39279cc3 CM |
6472 | inode_dec_link_count(inode); |
6473 | iput(inode); | |
6474 | } | |
c581afc8 | 6475 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6476 | btrfs_btree_balance_dirty(root); |
39279cc3 | 6477 | return err; |
b0d5d10f CM |
6478 | |
6479 | out_unlock_inode: | |
6480 | unlock_new_inode(inode); | |
6481 | goto out_unlock; | |
6482 | ||
39279cc3 CM |
6483 | } |
6484 | ||
6485 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6486 | struct dentry *dentry) | |
6487 | { | |
271dba45 | 6488 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6489 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6490 | struct inode *inode = d_inode(old_dentry); |
00e4e6b3 | 6491 | u64 index; |
39279cc3 CM |
6492 | int err; |
6493 | int drop_inode = 0; | |
6494 | ||
4a8be425 TH |
6495 | /* do not allow sys_link's with other subvols of the same device */ |
6496 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 6497 | return -EXDEV; |
4a8be425 | 6498 | |
f186373f | 6499 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6500 | return -EMLINK; |
4a8be425 | 6501 | |
3de4586c | 6502 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
6503 | if (err) |
6504 | goto fail; | |
6505 | ||
a22285a6 | 6506 | /* |
7e6b6465 | 6507 | * 2 items for inode and inode ref |
a22285a6 | 6508 | * 2 items for dir items |
7e6b6465 | 6509 | * 1 item for parent inode |
a22285a6 | 6510 | */ |
7e6b6465 | 6511 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
6512 | if (IS_ERR(trans)) { |
6513 | err = PTR_ERR(trans); | |
271dba45 | 6514 | trans = NULL; |
a22285a6 YZ |
6515 | goto fail; |
6516 | } | |
5f39d397 | 6517 | |
67de1176 MX |
6518 | /* There are several dir indexes for this inode, clear the cache. */ |
6519 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6520 | inc_nlink(inode); |
0c4d2d95 | 6521 | inode_inc_iversion(inode); |
04b285f3 | 6522 | inode->i_ctime = current_fs_time(inode->i_sb); |
7de9c6ee | 6523 | ihold(inode); |
e9976151 | 6524 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6525 | |
a1b075d2 | 6526 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 6527 | |
a5719521 | 6528 | if (err) { |
54aa1f4d | 6529 | drop_inode = 1; |
a5719521 | 6530 | } else { |
10d9f309 | 6531 | struct dentry *parent = dentry->d_parent; |
a5719521 | 6532 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6533 | if (err) |
6534 | goto fail; | |
ef3b9af5 FM |
6535 | if (inode->i_nlink == 1) { |
6536 | /* | |
6537 | * If new hard link count is 1, it's a file created | |
6538 | * with open(2) O_TMPFILE flag. | |
6539 | */ | |
6540 | err = btrfs_orphan_del(trans, inode); | |
6541 | if (err) | |
6542 | goto fail; | |
6543 | } | |
08c422c2 | 6544 | d_instantiate(dentry, inode); |
6a912213 | 6545 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 6546 | } |
39279cc3 | 6547 | |
c581afc8 | 6548 | btrfs_balance_delayed_items(root); |
1832a6d5 | 6549 | fail: |
271dba45 FM |
6550 | if (trans) |
6551 | btrfs_end_transaction(trans, root); | |
39279cc3 CM |
6552 | if (drop_inode) { |
6553 | inode_dec_link_count(inode); | |
6554 | iput(inode); | |
6555 | } | |
b53d3f5d | 6556 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
6557 | return err; |
6558 | } | |
6559 | ||
18bb1db3 | 6560 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6561 | { |
b9d86667 | 6562 | struct inode *inode = NULL; |
39279cc3 CM |
6563 | struct btrfs_trans_handle *trans; |
6564 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6565 | int err = 0; | |
6566 | int drop_on_err = 0; | |
b9d86667 | 6567 | u64 objectid = 0; |
00e4e6b3 | 6568 | u64 index = 0; |
39279cc3 | 6569 | |
9ed74f2d JB |
6570 | /* |
6571 | * 2 items for inode and ref | |
6572 | * 2 items for dir items | |
6573 | * 1 for xattr if selinux is on | |
6574 | */ | |
a22285a6 YZ |
6575 | trans = btrfs_start_transaction(root, 5); |
6576 | if (IS_ERR(trans)) | |
6577 | return PTR_ERR(trans); | |
39279cc3 | 6578 | |
581bb050 LZ |
6579 | err = btrfs_find_free_ino(root, &objectid); |
6580 | if (err) | |
6581 | goto out_fail; | |
6582 | ||
aec7477b | 6583 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6584 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6585 | S_IFDIR | mode, &index); |
39279cc3 CM |
6586 | if (IS_ERR(inode)) { |
6587 | err = PTR_ERR(inode); | |
6588 | goto out_fail; | |
6589 | } | |
5f39d397 | 6590 | |
39279cc3 | 6591 | drop_on_err = 1; |
b0d5d10f CM |
6592 | /* these must be set before we unlock the inode */ |
6593 | inode->i_op = &btrfs_dir_inode_operations; | |
6594 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6595 | |
2a7dba39 | 6596 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6597 | if (err) |
b0d5d10f | 6598 | goto out_fail_inode; |
39279cc3 | 6599 | |
dbe674a9 | 6600 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
6601 | err = btrfs_update_inode(trans, root, inode); |
6602 | if (err) | |
b0d5d10f | 6603 | goto out_fail_inode; |
5f39d397 | 6604 | |
a1b075d2 JB |
6605 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
6606 | dentry->d_name.len, 0, index); | |
39279cc3 | 6607 | if (err) |
b0d5d10f | 6608 | goto out_fail_inode; |
5f39d397 | 6609 | |
39279cc3 | 6610 | d_instantiate(dentry, inode); |
b0d5d10f CM |
6611 | /* |
6612 | * mkdir is special. We're unlocking after we call d_instantiate | |
6613 | * to avoid a race with nfsd calling d_instantiate. | |
6614 | */ | |
6615 | unlock_new_inode(inode); | |
39279cc3 | 6616 | drop_on_err = 0; |
39279cc3 CM |
6617 | |
6618 | out_fail: | |
7ad85bb7 | 6619 | btrfs_end_transaction(trans, root); |
c7cfb8a5 WS |
6620 | if (drop_on_err) { |
6621 | inode_dec_link_count(inode); | |
39279cc3 | 6622 | iput(inode); |
c7cfb8a5 | 6623 | } |
c581afc8 | 6624 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6625 | btrfs_btree_balance_dirty(root); |
39279cc3 | 6626 | return err; |
b0d5d10f CM |
6627 | |
6628 | out_fail_inode: | |
6629 | unlock_new_inode(inode); | |
6630 | goto out_fail; | |
39279cc3 CM |
6631 | } |
6632 | ||
e6c4efd8 QW |
6633 | /* Find next extent map of a given extent map, caller needs to ensure locks */ |
6634 | static struct extent_map *next_extent_map(struct extent_map *em) | |
6635 | { | |
6636 | struct rb_node *next; | |
6637 | ||
6638 | next = rb_next(&em->rb_node); | |
6639 | if (!next) | |
6640 | return NULL; | |
6641 | return container_of(next, struct extent_map, rb_node); | |
6642 | } | |
6643 | ||
6644 | static struct extent_map *prev_extent_map(struct extent_map *em) | |
6645 | { | |
6646 | struct rb_node *prev; | |
6647 | ||
6648 | prev = rb_prev(&em->rb_node); | |
6649 | if (!prev) | |
6650 | return NULL; | |
6651 | return container_of(prev, struct extent_map, rb_node); | |
6652 | } | |
6653 | ||
d352ac68 | 6654 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
e6c4efd8 | 6655 | * the existing extent is the nearest extent to map_start, |
d352ac68 | 6656 | * and an extent that you want to insert, deal with overlap and insert |
e6c4efd8 | 6657 | * the best fitted new extent into the tree. |
d352ac68 | 6658 | */ |
3b951516 CM |
6659 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
6660 | struct extent_map *existing, | |
e6dcd2dc | 6661 | struct extent_map *em, |
51f395ad | 6662 | u64 map_start) |
3b951516 | 6663 | { |
e6c4efd8 QW |
6664 | struct extent_map *prev; |
6665 | struct extent_map *next; | |
6666 | u64 start; | |
6667 | u64 end; | |
3b951516 | 6668 | u64 start_diff; |
3b951516 | 6669 | |
e6dcd2dc | 6670 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
e6c4efd8 QW |
6671 | |
6672 | if (existing->start > map_start) { | |
6673 | next = existing; | |
6674 | prev = prev_extent_map(next); | |
6675 | } else { | |
6676 | prev = existing; | |
6677 | next = next_extent_map(prev); | |
6678 | } | |
6679 | ||
6680 | start = prev ? extent_map_end(prev) : em->start; | |
6681 | start = max_t(u64, start, em->start); | |
6682 | end = next ? next->start : extent_map_end(em); | |
6683 | end = min_t(u64, end, extent_map_end(em)); | |
6684 | start_diff = start - em->start; | |
6685 | em->start = start; | |
6686 | em->len = end - start; | |
c8b97818 CM |
6687 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
6688 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 6689 | em->block_start += start_diff; |
c8b97818 CM |
6690 | em->block_len -= start_diff; |
6691 | } | |
09a2a8f9 | 6692 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6693 | } |
6694 | ||
c8b97818 | 6695 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6696 | struct page *page, |
c8b97818 CM |
6697 | size_t pg_offset, u64 extent_offset, |
6698 | struct btrfs_file_extent_item *item) | |
6699 | { | |
6700 | int ret; | |
6701 | struct extent_buffer *leaf = path->nodes[0]; | |
6702 | char *tmp; | |
6703 | size_t max_size; | |
6704 | unsigned long inline_size; | |
6705 | unsigned long ptr; | |
261507a0 | 6706 | int compress_type; |
c8b97818 CM |
6707 | |
6708 | WARN_ON(pg_offset != 0); | |
261507a0 | 6709 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6710 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6711 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6712 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6713 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6714 | if (!tmp) |
6715 | return -ENOMEM; | |
c8b97818 CM |
6716 | ptr = btrfs_file_extent_inline_start(item); |
6717 | ||
6718 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6719 | ||
09cbfeaf | 6720 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6721 | ret = btrfs_decompress(compress_type, tmp, page, |
6722 | extent_offset, inline_size, max_size); | |
c8b97818 | 6723 | kfree(tmp); |
166ae5a4 | 6724 | return ret; |
c8b97818 CM |
6725 | } |
6726 | ||
d352ac68 CM |
6727 | /* |
6728 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
6729 | * the ugly parts come from merging extents from the disk with the in-ram |
6730 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
6731 | * where the in-ram extents might be locked pending data=ordered completion. |
6732 | * | |
6733 | * This also copies inline extents directly into the page. | |
6734 | */ | |
d397712b | 6735 | |
a52d9a80 | 6736 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 6737 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
6738 | int create) |
6739 | { | |
6740 | int ret; | |
6741 | int err = 0; | |
a52d9a80 CM |
6742 | u64 extent_start = 0; |
6743 | u64 extent_end = 0; | |
33345d01 | 6744 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 6745 | u32 found_type; |
f421950f | 6746 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
6747 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6748 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
6749 | struct extent_buffer *leaf; |
6750 | struct btrfs_key found_key; | |
a52d9a80 CM |
6751 | struct extent_map *em = NULL; |
6752 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 6753 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 6754 | struct btrfs_trans_handle *trans = NULL; |
7ffbb598 | 6755 | const bool new_inline = !page || create; |
a52d9a80 | 6756 | |
a52d9a80 | 6757 | again: |
890871be | 6758 | read_lock(&em_tree->lock); |
d1310b2e | 6759 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
6760 | if (em) |
6761 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 6762 | read_unlock(&em_tree->lock); |
d1310b2e | 6763 | |
a52d9a80 | 6764 | if (em) { |
e1c4b745 CM |
6765 | if (em->start > start || em->start + em->len <= start) |
6766 | free_extent_map(em); | |
6767 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6768 | free_extent_map(em); |
6769 | else | |
6770 | goto out; | |
a52d9a80 | 6771 | } |
172ddd60 | 6772 | em = alloc_extent_map(); |
a52d9a80 | 6773 | if (!em) { |
d1310b2e CM |
6774 | err = -ENOMEM; |
6775 | goto out; | |
a52d9a80 | 6776 | } |
e6dcd2dc | 6777 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 6778 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6779 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6780 | em->len = (u64)-1; |
c8b97818 | 6781 | em->block_len = (u64)-1; |
f421950f CM |
6782 | |
6783 | if (!path) { | |
6784 | path = btrfs_alloc_path(); | |
026fd317 JB |
6785 | if (!path) { |
6786 | err = -ENOMEM; | |
6787 | goto out; | |
6788 | } | |
6789 | /* | |
6790 | * Chances are we'll be called again, so go ahead and do | |
6791 | * readahead | |
6792 | */ | |
e4058b54 | 6793 | path->reada = READA_FORWARD; |
f421950f CM |
6794 | } |
6795 | ||
179e29e4 CM |
6796 | ret = btrfs_lookup_file_extent(trans, root, path, |
6797 | objectid, start, trans != NULL); | |
a52d9a80 CM |
6798 | if (ret < 0) { |
6799 | err = ret; | |
6800 | goto out; | |
6801 | } | |
6802 | ||
6803 | if (ret != 0) { | |
6804 | if (path->slots[0] == 0) | |
6805 | goto not_found; | |
6806 | path->slots[0]--; | |
6807 | } | |
6808 | ||
5f39d397 CM |
6809 | leaf = path->nodes[0]; |
6810 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6811 | struct btrfs_file_extent_item); |
a52d9a80 | 6812 | /* are we inside the extent that was found? */ |
5f39d397 | 6813 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
962a298f | 6814 | found_type = found_key.type; |
5f39d397 | 6815 | if (found_key.objectid != objectid || |
a52d9a80 | 6816 | found_type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6817 | /* |
6818 | * If we backup past the first extent we want to move forward | |
6819 | * and see if there is an extent in front of us, otherwise we'll | |
6820 | * say there is a hole for our whole search range which can | |
6821 | * cause problems. | |
6822 | */ | |
6823 | extent_end = start; | |
6824 | goto next; | |
a52d9a80 CM |
6825 | } |
6826 | ||
5f39d397 CM |
6827 | found_type = btrfs_file_extent_type(leaf, item); |
6828 | extent_start = found_key.offset; | |
d899e052 YZ |
6829 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6830 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6831 | extent_end = extent_start + |
db94535d | 6832 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
6833 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
6834 | size_t size; | |
514ac8ad | 6835 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
fda2832f | 6836 | extent_end = ALIGN(extent_start + size, root->sectorsize); |
9036c102 | 6837 | } |
25a50341 | 6838 | next: |
9036c102 YZ |
6839 | if (start >= extent_end) { |
6840 | path->slots[0]++; | |
6841 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6842 | ret = btrfs_next_leaf(root, path); | |
6843 | if (ret < 0) { | |
6844 | err = ret; | |
6845 | goto out; | |
a52d9a80 | 6846 | } |
9036c102 YZ |
6847 | if (ret > 0) |
6848 | goto not_found; | |
6849 | leaf = path->nodes[0]; | |
a52d9a80 | 6850 | } |
9036c102 YZ |
6851 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6852 | if (found_key.objectid != objectid || | |
6853 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6854 | goto not_found; | |
6855 | if (start + len <= found_key.offset) | |
6856 | goto not_found; | |
e2eca69d WS |
6857 | if (start > found_key.offset) |
6858 | goto next; | |
9036c102 | 6859 | em->start = start; |
70c8a91c | 6860 | em->orig_start = start; |
9036c102 YZ |
6861 | em->len = found_key.offset - start; |
6862 | goto not_found_em; | |
6863 | } | |
6864 | ||
7ffbb598 FM |
6865 | btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em); |
6866 | ||
d899e052 YZ |
6867 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6868 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 CM |
6869 | goto insert; |
6870 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 6871 | unsigned long ptr; |
a52d9a80 | 6872 | char *map; |
3326d1b0 CM |
6873 | size_t size; |
6874 | size_t extent_offset; | |
6875 | size_t copy_size; | |
a52d9a80 | 6876 | |
7ffbb598 | 6877 | if (new_inline) |
689f9346 | 6878 | goto out; |
5f39d397 | 6879 | |
514ac8ad | 6880 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
9036c102 | 6881 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
6882 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
6883 | size - extent_offset); | |
3326d1b0 | 6884 | em->start = extent_start + extent_offset; |
fda2832f | 6885 | em->len = ALIGN(copy_size, root->sectorsize); |
b4939680 | 6886 | em->orig_block_len = em->len; |
70c8a91c | 6887 | em->orig_start = em->start; |
689f9346 | 6888 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 6889 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
6890 | if (btrfs_file_extent_compression(leaf, item) != |
6891 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 6892 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 6893 | extent_offset, item); |
166ae5a4 ZB |
6894 | if (ret) { |
6895 | err = ret; | |
6896 | goto out; | |
6897 | } | |
c8b97818 CM |
6898 | } else { |
6899 | map = kmap(page); | |
6900 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6901 | copy_size); | |
09cbfeaf | 6902 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 6903 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 6904 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
6905 | copy_size); |
6906 | } | |
c8b97818 CM |
6907 | kunmap(page); |
6908 | } | |
179e29e4 CM |
6909 | flush_dcache_page(page); |
6910 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 6911 | BUG(); |
179e29e4 CM |
6912 | if (!trans) { |
6913 | kunmap(page); | |
6914 | free_extent_map(em); | |
6915 | em = NULL; | |
ff5714cc | 6916 | |
b3b4aa74 | 6917 | btrfs_release_path(path); |
7a7eaa40 | 6918 | trans = btrfs_join_transaction(root); |
ff5714cc | 6919 | |
3612b495 TI |
6920 | if (IS_ERR(trans)) |
6921 | return ERR_CAST(trans); | |
179e29e4 CM |
6922 | goto again; |
6923 | } | |
c8b97818 | 6924 | map = kmap(page); |
70dec807 | 6925 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 6926 | copy_size); |
c8b97818 | 6927 | kunmap(page); |
179e29e4 | 6928 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 6929 | } |
d1310b2e | 6930 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6931 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 6932 | goto insert; |
a52d9a80 CM |
6933 | } |
6934 | not_found: | |
6935 | em->start = start; | |
70c8a91c | 6936 | em->orig_start = start; |
d1310b2e | 6937 | em->len = len; |
a52d9a80 | 6938 | not_found_em: |
5f39d397 | 6939 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 6940 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 6941 | insert: |
b3b4aa74 | 6942 | btrfs_release_path(path); |
d1310b2e | 6943 | if (em->start > start || extent_map_end(em) <= start) { |
c2cf52eb | 6944 | btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]", |
c1c9ff7c | 6945 | em->start, em->len, start, len); |
a52d9a80 CM |
6946 | err = -EIO; |
6947 | goto out; | |
6948 | } | |
d1310b2e CM |
6949 | |
6950 | err = 0; | |
890871be | 6951 | write_lock(&em_tree->lock); |
09a2a8f9 | 6952 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6953 | /* it is possible that someone inserted the extent into the tree |
6954 | * while we had the lock dropped. It is also possible that | |
6955 | * an overlapping map exists in the tree | |
6956 | */ | |
a52d9a80 | 6957 | if (ret == -EEXIST) { |
3b951516 | 6958 | struct extent_map *existing; |
e6dcd2dc CM |
6959 | |
6960 | ret = 0; | |
6961 | ||
e6c4efd8 QW |
6962 | existing = search_extent_mapping(em_tree, start, len); |
6963 | /* | |
6964 | * existing will always be non-NULL, since there must be | |
6965 | * extent causing the -EEXIST. | |
6966 | */ | |
6967 | if (start >= extent_map_end(existing) || | |
32be3a1a | 6968 | start <= existing->start) { |
e6c4efd8 QW |
6969 | /* |
6970 | * The existing extent map is the one nearest to | |
6971 | * the [start, start + len) range which overlaps | |
6972 | */ | |
6973 | err = merge_extent_mapping(em_tree, existing, | |
6974 | em, start); | |
e1c4b745 | 6975 | free_extent_map(existing); |
e6c4efd8 | 6976 | if (err) { |
3b951516 CM |
6977 | free_extent_map(em); |
6978 | em = NULL; | |
6979 | } | |
6980 | } else { | |
6981 | free_extent_map(em); | |
6982 | em = existing; | |
e6dcd2dc | 6983 | err = 0; |
a52d9a80 | 6984 | } |
a52d9a80 | 6985 | } |
890871be | 6986 | write_unlock(&em_tree->lock); |
a52d9a80 | 6987 | out: |
1abe9b8a | 6988 | |
4cd8587c | 6989 | trace_btrfs_get_extent(root, em); |
1abe9b8a | 6990 | |
527afb44 | 6991 | btrfs_free_path(path); |
a52d9a80 CM |
6992 | if (trans) { |
6993 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 6994 | if (!err) |
a52d9a80 CM |
6995 | err = ret; |
6996 | } | |
a52d9a80 CM |
6997 | if (err) { |
6998 | free_extent_map(em); | |
a52d9a80 CM |
6999 | return ERR_PTR(err); |
7000 | } | |
79787eaa | 7001 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
7002 | return em; |
7003 | } | |
7004 | ||
ec29ed5b CM |
7005 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
7006 | size_t pg_offset, u64 start, u64 len, | |
7007 | int create) | |
7008 | { | |
7009 | struct extent_map *em; | |
7010 | struct extent_map *hole_em = NULL; | |
7011 | u64 range_start = start; | |
7012 | u64 end; | |
7013 | u64 found; | |
7014 | u64 found_end; | |
7015 | int err = 0; | |
7016 | ||
7017 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
7018 | if (IS_ERR(em)) | |
7019 | return em; | |
7020 | if (em) { | |
7021 | /* | |
f9e4fb53 LB |
7022 | * if our em maps to |
7023 | * - a hole or | |
7024 | * - a pre-alloc extent, | |
7025 | * there might actually be delalloc bytes behind it. | |
ec29ed5b | 7026 | */ |
f9e4fb53 LB |
7027 | if (em->block_start != EXTENT_MAP_HOLE && |
7028 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
ec29ed5b CM |
7029 | return em; |
7030 | else | |
7031 | hole_em = em; | |
7032 | } | |
7033 | ||
7034 | /* check to see if we've wrapped (len == -1 or similar) */ | |
7035 | end = start + len; | |
7036 | if (end < start) | |
7037 | end = (u64)-1; | |
7038 | else | |
7039 | end -= 1; | |
7040 | ||
7041 | em = NULL; | |
7042 | ||
7043 | /* ok, we didn't find anything, lets look for delalloc */ | |
7044 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
7045 | end, len, EXTENT_DELALLOC, 1); | |
7046 | found_end = range_start + found; | |
7047 | if (found_end < range_start) | |
7048 | found_end = (u64)-1; | |
7049 | ||
7050 | /* | |
7051 | * we didn't find anything useful, return | |
7052 | * the original results from get_extent() | |
7053 | */ | |
7054 | if (range_start > end || found_end <= start) { | |
7055 | em = hole_em; | |
7056 | hole_em = NULL; | |
7057 | goto out; | |
7058 | } | |
7059 | ||
7060 | /* adjust the range_start to make sure it doesn't | |
7061 | * go backwards from the start they passed in | |
7062 | */ | |
67871254 | 7063 | range_start = max(start, range_start); |
ec29ed5b CM |
7064 | found = found_end - range_start; |
7065 | ||
7066 | if (found > 0) { | |
7067 | u64 hole_start = start; | |
7068 | u64 hole_len = len; | |
7069 | ||
172ddd60 | 7070 | em = alloc_extent_map(); |
ec29ed5b CM |
7071 | if (!em) { |
7072 | err = -ENOMEM; | |
7073 | goto out; | |
7074 | } | |
7075 | /* | |
7076 | * when btrfs_get_extent can't find anything it | |
7077 | * returns one huge hole | |
7078 | * | |
7079 | * make sure what it found really fits our range, and | |
7080 | * adjust to make sure it is based on the start from | |
7081 | * the caller | |
7082 | */ | |
7083 | if (hole_em) { | |
7084 | u64 calc_end = extent_map_end(hole_em); | |
7085 | ||
7086 | if (calc_end <= start || (hole_em->start > end)) { | |
7087 | free_extent_map(hole_em); | |
7088 | hole_em = NULL; | |
7089 | } else { | |
7090 | hole_start = max(hole_em->start, start); | |
7091 | hole_len = calc_end - hole_start; | |
7092 | } | |
7093 | } | |
7094 | em->bdev = NULL; | |
7095 | if (hole_em && range_start > hole_start) { | |
7096 | /* our hole starts before our delalloc, so we | |
7097 | * have to return just the parts of the hole | |
7098 | * that go until the delalloc starts | |
7099 | */ | |
7100 | em->len = min(hole_len, | |
7101 | range_start - hole_start); | |
7102 | em->start = hole_start; | |
7103 | em->orig_start = hole_start; | |
7104 | /* | |
7105 | * don't adjust block start at all, | |
7106 | * it is fixed at EXTENT_MAP_HOLE | |
7107 | */ | |
7108 | em->block_start = hole_em->block_start; | |
7109 | em->block_len = hole_len; | |
f9e4fb53 LB |
7110 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
7111 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
7112 | } else { |
7113 | em->start = range_start; | |
7114 | em->len = found; | |
7115 | em->orig_start = range_start; | |
7116 | em->block_start = EXTENT_MAP_DELALLOC; | |
7117 | em->block_len = found; | |
7118 | } | |
7119 | } else if (hole_em) { | |
7120 | return hole_em; | |
7121 | } | |
7122 | out: | |
7123 | ||
7124 | free_extent_map(hole_em); | |
7125 | if (err) { | |
7126 | free_extent_map(em); | |
7127 | return ERR_PTR(err); | |
7128 | } | |
7129 | return em; | |
7130 | } | |
7131 | ||
4b46fce2 JB |
7132 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
7133 | u64 start, u64 len) | |
7134 | { | |
7135 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
70c8a91c | 7136 | struct extent_map *em; |
4b46fce2 JB |
7137 | struct btrfs_key ins; |
7138 | u64 alloc_hint; | |
7139 | int ret; | |
4b46fce2 | 7140 | |
4b46fce2 | 7141 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
00361589 | 7142 | ret = btrfs_reserve_extent(root, len, root->sectorsize, 0, |
e570fd27 | 7143 | alloc_hint, &ins, 1, 1); |
00361589 JB |
7144 | if (ret) |
7145 | return ERR_PTR(ret); | |
4b46fce2 | 7146 | |
de0ee0ed FM |
7147 | /* |
7148 | * Create the ordered extent before the extent map. This is to avoid | |
7149 | * races with the fast fsync path that would lead to it logging file | |
7150 | * extent items that point to disk extents that were not yet written to. | |
7151 | * The fast fsync path collects ordered extents into a local list and | |
7152 | * then collects all the new extent maps, so we must create the ordered | |
7153 | * extent first and make sure the fast fsync path collects any new | |
7154 | * ordered extents after collecting new extent maps as well. | |
7155 | * The fsync path simply can not rely on inode_dio_wait() because it | |
7156 | * causes deadlock with AIO. | |
7157 | */ | |
4b46fce2 JB |
7158 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, |
7159 | ins.offset, ins.offset, 0); | |
7160 | if (ret) { | |
e570fd27 | 7161 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
00361589 | 7162 | return ERR_PTR(ret); |
4b46fce2 | 7163 | } |
00361589 | 7164 | |
de0ee0ed FM |
7165 | em = create_pinned_em(inode, start, ins.offset, start, ins.objectid, |
7166 | ins.offset, ins.offset, ins.offset, 0); | |
7167 | if (IS_ERR(em)) { | |
7168 | struct btrfs_ordered_extent *oe; | |
7169 | ||
7170 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); | |
7171 | oe = btrfs_lookup_ordered_extent(inode, start); | |
7172 | ASSERT(oe); | |
7173 | if (WARN_ON(!oe)) | |
7174 | return em; | |
7175 | set_bit(BTRFS_ORDERED_IOERR, &oe->flags); | |
7176 | set_bit(BTRFS_ORDERED_IO_DONE, &oe->flags); | |
7177 | btrfs_remove_ordered_extent(inode, oe); | |
7178 | /* Once for our lookup and once for the ordered extents tree. */ | |
7179 | btrfs_put_ordered_extent(oe); | |
7180 | btrfs_put_ordered_extent(oe); | |
7181 | } | |
4b46fce2 JB |
7182 | return em; |
7183 | } | |
7184 | ||
46bfbb5c CM |
7185 | /* |
7186 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
7187 | * block must be cow'd | |
7188 | */ | |
00361589 | 7189 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
7190 | u64 *orig_start, u64 *orig_block_len, |
7191 | u64 *ram_bytes) | |
46bfbb5c | 7192 | { |
00361589 | 7193 | struct btrfs_trans_handle *trans; |
46bfbb5c CM |
7194 | struct btrfs_path *path; |
7195 | int ret; | |
7196 | struct extent_buffer *leaf; | |
7197 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7198 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7199 | struct btrfs_file_extent_item *fi; |
7200 | struct btrfs_key key; | |
7201 | u64 disk_bytenr; | |
7202 | u64 backref_offset; | |
7203 | u64 extent_end; | |
7204 | u64 num_bytes; | |
7205 | int slot; | |
7206 | int found_type; | |
7ee9e440 | 7207 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7208 | |
46bfbb5c CM |
7209 | path = btrfs_alloc_path(); |
7210 | if (!path) | |
7211 | return -ENOMEM; | |
7212 | ||
00361589 | 7213 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), |
46bfbb5c CM |
7214 | offset, 0); |
7215 | if (ret < 0) | |
7216 | goto out; | |
7217 | ||
7218 | slot = path->slots[0]; | |
7219 | if (ret == 1) { | |
7220 | if (slot == 0) { | |
7221 | /* can't find the item, must cow */ | |
7222 | ret = 0; | |
7223 | goto out; | |
7224 | } | |
7225 | slot--; | |
7226 | } | |
7227 | ret = 0; | |
7228 | leaf = path->nodes[0]; | |
7229 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 7230 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
7231 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7232 | /* not our file or wrong item type, must cow */ | |
7233 | goto out; | |
7234 | } | |
7235 | ||
7236 | if (key.offset > offset) { | |
7237 | /* Wrong offset, must cow */ | |
7238 | goto out; | |
7239 | } | |
7240 | ||
7241 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7242 | found_type = btrfs_file_extent_type(leaf, fi); | |
7243 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7244 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7245 | /* not a regular extent, must cow */ | |
7246 | goto out; | |
7247 | } | |
7ee9e440 JB |
7248 | |
7249 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7250 | goto out; | |
7251 | ||
e77751aa MX |
7252 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7253 | if (extent_end <= offset) | |
7254 | goto out; | |
7255 | ||
46bfbb5c | 7256 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7257 | if (disk_bytenr == 0) |
7258 | goto out; | |
7259 | ||
7260 | if (btrfs_file_extent_compression(leaf, fi) || | |
7261 | btrfs_file_extent_encryption(leaf, fi) || | |
7262 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7263 | goto out; | |
7264 | ||
46bfbb5c CM |
7265 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7266 | ||
7ee9e440 JB |
7267 | if (orig_start) { |
7268 | *orig_start = key.offset - backref_offset; | |
7269 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7270 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7271 | } | |
eb384b55 | 7272 | |
46bfbb5c CM |
7273 | if (btrfs_extent_readonly(root, disk_bytenr)) |
7274 | goto out; | |
7b2b7085 MX |
7275 | |
7276 | num_bytes = min(offset + *len, extent_end) - offset; | |
7277 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7278 | u64 range_end; | |
7279 | ||
7280 | range_end = round_up(offset + num_bytes, root->sectorsize) - 1; | |
7281 | ret = test_range_bit(io_tree, offset, range_end, | |
7282 | EXTENT_DELALLOC, 0, NULL); | |
7283 | if (ret) { | |
7284 | ret = -EAGAIN; | |
7285 | goto out; | |
7286 | } | |
7287 | } | |
7288 | ||
1bda19eb | 7289 | btrfs_release_path(path); |
46bfbb5c CM |
7290 | |
7291 | /* | |
7292 | * look for other files referencing this extent, if we | |
7293 | * find any we must cow | |
7294 | */ | |
00361589 JB |
7295 | trans = btrfs_join_transaction(root); |
7296 | if (IS_ERR(trans)) { | |
7297 | ret = 0; | |
46bfbb5c | 7298 | goto out; |
00361589 JB |
7299 | } |
7300 | ||
7301 | ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), | |
7302 | key.offset - backref_offset, disk_bytenr); | |
7303 | btrfs_end_transaction(trans, root); | |
7304 | if (ret) { | |
7305 | ret = 0; | |
7306 | goto out; | |
7307 | } | |
46bfbb5c CM |
7308 | |
7309 | /* | |
7310 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7311 | * in this extent we are about to write. If there | |
7312 | * are any csums in that range we have to cow in order | |
7313 | * to keep the csums correct | |
7314 | */ | |
7315 | disk_bytenr += backref_offset; | |
7316 | disk_bytenr += offset - key.offset; | |
46bfbb5c CM |
7317 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) |
7318 | goto out; | |
7319 | /* | |
7320 | * all of the above have passed, it is safe to overwrite this extent | |
7321 | * without cow | |
7322 | */ | |
eb384b55 | 7323 | *len = num_bytes; |
46bfbb5c CM |
7324 | ret = 1; |
7325 | out: | |
7326 | btrfs_free_path(path); | |
7327 | return ret; | |
7328 | } | |
7329 | ||
fc4adbff AG |
7330 | bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end) |
7331 | { | |
7332 | struct radix_tree_root *root = &inode->i_mapping->page_tree; | |
7333 | int found = false; | |
7334 | void **pagep = NULL; | |
7335 | struct page *page = NULL; | |
7336 | int start_idx; | |
7337 | int end_idx; | |
7338 | ||
09cbfeaf | 7339 | start_idx = start >> PAGE_SHIFT; |
fc4adbff AG |
7340 | |
7341 | /* | |
7342 | * end is the last byte in the last page. end == start is legal | |
7343 | */ | |
09cbfeaf | 7344 | end_idx = end >> PAGE_SHIFT; |
fc4adbff AG |
7345 | |
7346 | rcu_read_lock(); | |
7347 | ||
7348 | /* Most of the code in this while loop is lifted from | |
7349 | * find_get_page. It's been modified to begin searching from a | |
7350 | * page and return just the first page found in that range. If the | |
7351 | * found idx is less than or equal to the end idx then we know that | |
7352 | * a page exists. If no pages are found or if those pages are | |
7353 | * outside of the range then we're fine (yay!) */ | |
7354 | while (page == NULL && | |
7355 | radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) { | |
7356 | page = radix_tree_deref_slot(pagep); | |
7357 | if (unlikely(!page)) | |
7358 | break; | |
7359 | ||
7360 | if (radix_tree_exception(page)) { | |
809f9016 FM |
7361 | if (radix_tree_deref_retry(page)) { |
7362 | page = NULL; | |
fc4adbff | 7363 | continue; |
809f9016 | 7364 | } |
fc4adbff AG |
7365 | /* |
7366 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
7367 | * here as an exceptional entry: so return it without | |
7368 | * attempting to raise page count. | |
7369 | */ | |
6fdef6d4 | 7370 | page = NULL; |
fc4adbff AG |
7371 | break; /* TODO: Is this relevant for this use case? */ |
7372 | } | |
7373 | ||
91405151 FM |
7374 | if (!page_cache_get_speculative(page)) { |
7375 | page = NULL; | |
fc4adbff | 7376 | continue; |
91405151 | 7377 | } |
fc4adbff AG |
7378 | |
7379 | /* | |
7380 | * Has the page moved? | |
7381 | * This is part of the lockless pagecache protocol. See | |
7382 | * include/linux/pagemap.h for details. | |
7383 | */ | |
7384 | if (unlikely(page != *pagep)) { | |
09cbfeaf | 7385 | put_page(page); |
fc4adbff AG |
7386 | page = NULL; |
7387 | } | |
7388 | } | |
7389 | ||
7390 | if (page) { | |
7391 | if (page->index <= end_idx) | |
7392 | found = true; | |
09cbfeaf | 7393 | put_page(page); |
fc4adbff AG |
7394 | } |
7395 | ||
7396 | rcu_read_unlock(); | |
7397 | return found; | |
7398 | } | |
7399 | ||
eb838e73 JB |
7400 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
7401 | struct extent_state **cached_state, int writing) | |
7402 | { | |
7403 | struct btrfs_ordered_extent *ordered; | |
7404 | int ret = 0; | |
7405 | ||
7406 | while (1) { | |
7407 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7408 | cached_state); |
eb838e73 JB |
7409 | /* |
7410 | * We're concerned with the entire range that we're going to be | |
7411 | * doing DIO to, so we need to make sure theres no ordered | |
7412 | * extents in this range. | |
7413 | */ | |
7414 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
7415 | lockend - lockstart + 1); | |
7416 | ||
7417 | /* | |
7418 | * We need to make sure there are no buffered pages in this | |
7419 | * range either, we could have raced between the invalidate in | |
7420 | * generic_file_direct_write and locking the extent. The | |
7421 | * invalidate needs to happen so that reads after a write do not | |
7422 | * get stale data. | |
7423 | */ | |
fc4adbff AG |
7424 | if (!ordered && |
7425 | (!writing || | |
7426 | !btrfs_page_exists_in_range(inode, lockstart, lockend))) | |
eb838e73 JB |
7427 | break; |
7428 | ||
7429 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7430 | cached_state, GFP_NOFS); | |
7431 | ||
7432 | if (ordered) { | |
ade77029 FM |
7433 | /* |
7434 | * If we are doing a DIO read and the ordered extent we | |
7435 | * found is for a buffered write, we can not wait for it | |
7436 | * to complete and retry, because if we do so we can | |
7437 | * deadlock with concurrent buffered writes on page | |
7438 | * locks. This happens only if our DIO read covers more | |
7439 | * than one extent map, if at this point has already | |
7440 | * created an ordered extent for a previous extent map | |
7441 | * and locked its range in the inode's io tree, and a | |
7442 | * concurrent write against that previous extent map's | |
7443 | * range and this range started (we unlock the ranges | |
7444 | * in the io tree only when the bios complete and | |
7445 | * buffered writes always lock pages before attempting | |
7446 | * to lock range in the io tree). | |
7447 | */ | |
7448 | if (writing || | |
7449 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
7450 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7451 | else | |
7452 | ret = -ENOTBLK; | |
eb838e73 JB |
7453 | btrfs_put_ordered_extent(ordered); |
7454 | } else { | |
eb838e73 | 7455 | /* |
b850ae14 FM |
7456 | * We could trigger writeback for this range (and wait |
7457 | * for it to complete) and then invalidate the pages for | |
7458 | * this range (through invalidate_inode_pages2_range()), | |
7459 | * but that can lead us to a deadlock with a concurrent | |
7460 | * call to readpages() (a buffered read or a defrag call | |
7461 | * triggered a readahead) on a page lock due to an | |
7462 | * ordered dio extent we created before but did not have | |
7463 | * yet a corresponding bio submitted (whence it can not | |
7464 | * complete), which makes readpages() wait for that | |
7465 | * ordered extent to complete while holding a lock on | |
7466 | * that page. | |
eb838e73 | 7467 | */ |
b850ae14 | 7468 | ret = -ENOTBLK; |
eb838e73 JB |
7469 | } |
7470 | ||
ade77029 FM |
7471 | if (ret) |
7472 | break; | |
7473 | ||
eb838e73 JB |
7474 | cond_resched(); |
7475 | } | |
7476 | ||
7477 | return ret; | |
7478 | } | |
7479 | ||
69ffb543 JB |
7480 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
7481 | u64 len, u64 orig_start, | |
7482 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
7483 | u64 orig_block_len, u64 ram_bytes, |
7484 | int type) | |
69ffb543 JB |
7485 | { |
7486 | struct extent_map_tree *em_tree; | |
7487 | struct extent_map *em; | |
7488 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7489 | int ret; | |
7490 | ||
7491 | em_tree = &BTRFS_I(inode)->extent_tree; | |
7492 | em = alloc_extent_map(); | |
7493 | if (!em) | |
7494 | return ERR_PTR(-ENOMEM); | |
7495 | ||
7496 | em->start = start; | |
7497 | em->orig_start = orig_start; | |
2ab28f32 JB |
7498 | em->mod_start = start; |
7499 | em->mod_len = len; | |
69ffb543 JB |
7500 | em->len = len; |
7501 | em->block_len = block_len; | |
7502 | em->block_start = block_start; | |
7503 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7504 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7505 | em->ram_bytes = ram_bytes; |
70c8a91c | 7506 | em->generation = -1; |
69ffb543 JB |
7507 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
7508 | if (type == BTRFS_ORDERED_PREALLOC) | |
b11e234d | 7509 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
69ffb543 JB |
7510 | |
7511 | do { | |
7512 | btrfs_drop_extent_cache(inode, em->start, | |
7513 | em->start + em->len - 1, 0); | |
7514 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7515 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 JB |
7516 | write_unlock(&em_tree->lock); |
7517 | } while (ret == -EEXIST); | |
7518 | ||
7519 | if (ret) { | |
7520 | free_extent_map(em); | |
7521 | return ERR_PTR(ret); | |
7522 | } | |
7523 | ||
7524 | return em; | |
7525 | } | |
7526 | ||
9c9464cc FM |
7527 | static void adjust_dio_outstanding_extents(struct inode *inode, |
7528 | struct btrfs_dio_data *dio_data, | |
7529 | const u64 len) | |
7530 | { | |
7531 | unsigned num_extents; | |
7532 | ||
7533 | num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1, | |
7534 | BTRFS_MAX_EXTENT_SIZE); | |
7535 | /* | |
7536 | * If we have an outstanding_extents count still set then we're | |
7537 | * within our reservation, otherwise we need to adjust our inode | |
7538 | * counter appropriately. | |
7539 | */ | |
7540 | if (dio_data->outstanding_extents) { | |
7541 | dio_data->outstanding_extents -= num_extents; | |
7542 | } else { | |
7543 | spin_lock(&BTRFS_I(inode)->lock); | |
7544 | BTRFS_I(inode)->outstanding_extents += num_extents; | |
7545 | spin_unlock(&BTRFS_I(inode)->lock); | |
7546 | } | |
7547 | } | |
7548 | ||
4b46fce2 JB |
7549 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7550 | struct buffer_head *bh_result, int create) | |
7551 | { | |
7552 | struct extent_map *em; | |
7553 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 7554 | struct extent_state *cached_state = NULL; |
50745b0a | 7555 | struct btrfs_dio_data *dio_data = NULL; |
4b46fce2 | 7556 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7557 | u64 lockstart, lockend; |
4b46fce2 | 7558 | u64 len = bh_result->b_size; |
eb838e73 | 7559 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7560 | int ret = 0; |
eb838e73 | 7561 | |
172a5049 | 7562 | if (create) |
3266789f | 7563 | unlock_bits |= EXTENT_DIRTY; |
172a5049 | 7564 | else |
c329861d | 7565 | len = min_t(u64, len, root->sectorsize); |
eb838e73 | 7566 | |
c329861d JB |
7567 | lockstart = start; |
7568 | lockend = start + len - 1; | |
7569 | ||
e1cbbfa5 JB |
7570 | if (current->journal_info) { |
7571 | /* | |
7572 | * Need to pull our outstanding extents and set journal_info to NULL so | |
7573 | * that anything that needs to check if there's a transction doesn't get | |
7574 | * confused. | |
7575 | */ | |
50745b0a | 7576 | dio_data = current->journal_info; |
e1cbbfa5 JB |
7577 | current->journal_info = NULL; |
7578 | } | |
7579 | ||
eb838e73 JB |
7580 | /* |
7581 | * If this errors out it's because we couldn't invalidate pagecache for | |
7582 | * this range and we need to fallback to buffered. | |
7583 | */ | |
9c9464cc FM |
7584 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, |
7585 | create)) { | |
7586 | ret = -ENOTBLK; | |
7587 | goto err; | |
7588 | } | |
eb838e73 | 7589 | |
4b46fce2 | 7590 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); |
eb838e73 JB |
7591 | if (IS_ERR(em)) { |
7592 | ret = PTR_ERR(em); | |
7593 | goto unlock_err; | |
7594 | } | |
4b46fce2 JB |
7595 | |
7596 | /* | |
7597 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7598 | * io. INLINE is special, and we could probably kludge it in here, but | |
7599 | * it's still buffered so for safety lets just fall back to the generic | |
7600 | * buffered path. | |
7601 | * | |
7602 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7603 | * decompress it, so there will be buffering required no matter what we | |
7604 | * do, so go ahead and fallback to buffered. | |
7605 | * | |
7606 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
7607 | * to buffered IO. Don't blame me, this is the price we pay for using | |
7608 | * the generic code. | |
7609 | */ | |
7610 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7611 | em->block_start == EXTENT_MAP_INLINE) { | |
7612 | free_extent_map(em); | |
eb838e73 JB |
7613 | ret = -ENOTBLK; |
7614 | goto unlock_err; | |
4b46fce2 JB |
7615 | } |
7616 | ||
7617 | /* Just a good old fashioned hole, return */ | |
7618 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
7619 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
7620 | free_extent_map(em); | |
eb838e73 | 7621 | goto unlock_err; |
4b46fce2 JB |
7622 | } |
7623 | ||
7624 | /* | |
7625 | * We don't allocate a new extent in the following cases | |
7626 | * | |
7627 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7628 | * existing extent. | |
7629 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7630 | * just use the extent. | |
7631 | * | |
7632 | */ | |
46bfbb5c | 7633 | if (!create) { |
eb838e73 JB |
7634 | len = min(len, em->len - (start - em->start)); |
7635 | lockstart = start + len; | |
7636 | goto unlock; | |
46bfbb5c | 7637 | } |
4b46fce2 JB |
7638 | |
7639 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7640 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7641 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 | 7642 | int type; |
eb384b55 | 7643 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
7644 | |
7645 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7646 | type = BTRFS_ORDERED_PREALLOC; | |
7647 | else | |
7648 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 7649 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 7650 | block_start = em->block_start + (start - em->start); |
46bfbb5c | 7651 | |
00361589 | 7652 | if (can_nocow_extent(inode, start, &len, &orig_start, |
7ee9e440 | 7653 | &orig_block_len, &ram_bytes) == 1) { |
69ffb543 JB |
7654 | if (type == BTRFS_ORDERED_PREALLOC) { |
7655 | free_extent_map(em); | |
7656 | em = create_pinned_em(inode, start, len, | |
7657 | orig_start, | |
b4939680 | 7658 | block_start, len, |
cc95bef6 JB |
7659 | orig_block_len, |
7660 | ram_bytes, type); | |
555e1286 FM |
7661 | if (IS_ERR(em)) { |
7662 | ret = PTR_ERR(em); | |
69ffb543 | 7663 | goto unlock_err; |
555e1286 | 7664 | } |
69ffb543 JB |
7665 | } |
7666 | ||
46bfbb5c CM |
7667 | ret = btrfs_add_ordered_extent_dio(inode, start, |
7668 | block_start, len, len, type); | |
46bfbb5c CM |
7669 | if (ret) { |
7670 | free_extent_map(em); | |
eb838e73 | 7671 | goto unlock_err; |
46bfbb5c CM |
7672 | } |
7673 | goto unlock; | |
4b46fce2 | 7674 | } |
4b46fce2 | 7675 | } |
00361589 | 7676 | |
46bfbb5c CM |
7677 | /* |
7678 | * this will cow the extent, reset the len in case we changed | |
7679 | * it above | |
7680 | */ | |
7681 | len = bh_result->b_size; | |
70c8a91c JB |
7682 | free_extent_map(em); |
7683 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
7684 | if (IS_ERR(em)) { |
7685 | ret = PTR_ERR(em); | |
7686 | goto unlock_err; | |
7687 | } | |
46bfbb5c CM |
7688 | len = min(len, em->len - (start - em->start)); |
7689 | unlock: | |
4b46fce2 JB |
7690 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
7691 | inode->i_blkbits; | |
46bfbb5c | 7692 | bh_result->b_size = len; |
4b46fce2 JB |
7693 | bh_result->b_bdev = em->bdev; |
7694 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
7695 | if (create) { |
7696 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7697 | set_buffer_new(bh_result); | |
7698 | ||
7699 | /* | |
7700 | * Need to update the i_size under the extent lock so buffered | |
7701 | * readers will get the updated i_size when we unlock. | |
7702 | */ | |
7703 | if (start + len > i_size_read(inode)) | |
7704 | i_size_write(inode, start + len); | |
0934856d | 7705 | |
9c9464cc | 7706 | adjust_dio_outstanding_extents(inode, dio_data, len); |
7cf5b976 | 7707 | btrfs_free_reserved_data_space(inode, start, len); |
50745b0a | 7708 | WARN_ON(dio_data->reserve < len); |
7709 | dio_data->reserve -= len; | |
f28a4928 | 7710 | dio_data->unsubmitted_oe_range_end = start + len; |
50745b0a | 7711 | current->journal_info = dio_data; |
c3473e83 | 7712 | } |
4b46fce2 | 7713 | |
eb838e73 JB |
7714 | /* |
7715 | * In the case of write we need to clear and unlock the entire range, | |
7716 | * in the case of read we need to unlock only the end area that we | |
7717 | * aren't using if there is any left over space. | |
7718 | */ | |
24c03fa5 | 7719 | if (lockstart < lockend) { |
0934856d MX |
7720 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
7721 | lockend, unlock_bits, 1, 0, | |
7722 | &cached_state, GFP_NOFS); | |
24c03fa5 | 7723 | } else { |
eb838e73 | 7724 | free_extent_state(cached_state); |
24c03fa5 | 7725 | } |
eb838e73 | 7726 | |
4b46fce2 JB |
7727 | free_extent_map(em); |
7728 | ||
7729 | return 0; | |
eb838e73 JB |
7730 | |
7731 | unlock_err: | |
eb838e73 JB |
7732 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7733 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
9c9464cc | 7734 | err: |
50745b0a | 7735 | if (dio_data) |
7736 | current->journal_info = dio_data; | |
9c9464cc FM |
7737 | /* |
7738 | * Compensate the delalloc release we do in btrfs_direct_IO() when we | |
7739 | * write less data then expected, so that we don't underflow our inode's | |
7740 | * outstanding extents counter. | |
7741 | */ | |
7742 | if (create && dio_data) | |
7743 | adjust_dio_outstanding_extents(inode, dio_data, len); | |
7744 | ||
eb838e73 | 7745 | return ret; |
4b46fce2 JB |
7746 | } |
7747 | ||
8b110e39 MX |
7748 | static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio, |
7749 | int rw, int mirror_num) | |
7750 | { | |
7751 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7752 | int ret; | |
7753 | ||
7754 | BUG_ON(rw & REQ_WRITE); | |
7755 | ||
7756 | bio_get(bio); | |
7757 | ||
7758 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, | |
7759 | BTRFS_WQ_ENDIO_DIO_REPAIR); | |
7760 | if (ret) | |
7761 | goto err; | |
7762 | ||
7763 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); | |
7764 | err: | |
7765 | bio_put(bio); | |
7766 | return ret; | |
7767 | } | |
7768 | ||
7769 | static int btrfs_check_dio_repairable(struct inode *inode, | |
7770 | struct bio *failed_bio, | |
7771 | struct io_failure_record *failrec, | |
7772 | int failed_mirror) | |
7773 | { | |
7774 | int num_copies; | |
7775 | ||
7776 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, | |
7777 | failrec->logical, failrec->len); | |
7778 | if (num_copies == 1) { | |
7779 | /* | |
7780 | * we only have a single copy of the data, so don't bother with | |
7781 | * all the retry and error correction code that follows. no | |
7782 | * matter what the error is, it is very likely to persist. | |
7783 | */ | |
7784 | pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
7785 | num_copies, failrec->this_mirror, failed_mirror); | |
7786 | return 0; | |
7787 | } | |
7788 | ||
7789 | failrec->failed_mirror = failed_mirror; | |
7790 | failrec->this_mirror++; | |
7791 | if (failrec->this_mirror == failed_mirror) | |
7792 | failrec->this_mirror++; | |
7793 | ||
7794 | if (failrec->this_mirror > num_copies) { | |
7795 | pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
7796 | num_copies, failrec->this_mirror, failed_mirror); | |
7797 | return 0; | |
7798 | } | |
7799 | ||
7800 | return 1; | |
7801 | } | |
7802 | ||
7803 | static int dio_read_error(struct inode *inode, struct bio *failed_bio, | |
2dabb324 CR |
7804 | struct page *page, unsigned int pgoff, |
7805 | u64 start, u64 end, int failed_mirror, | |
7806 | bio_end_io_t *repair_endio, void *repair_arg) | |
8b110e39 MX |
7807 | { |
7808 | struct io_failure_record *failrec; | |
7809 | struct bio *bio; | |
7810 | int isector; | |
7811 | int read_mode; | |
7812 | int ret; | |
7813 | ||
7814 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
7815 | ||
7816 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
7817 | if (ret) | |
7818 | return ret; | |
7819 | ||
7820 | ret = btrfs_check_dio_repairable(inode, failed_bio, failrec, | |
7821 | failed_mirror); | |
7822 | if (!ret) { | |
7823 | free_io_failure(inode, failrec); | |
7824 | return -EIO; | |
7825 | } | |
7826 | ||
2dabb324 CR |
7827 | if ((failed_bio->bi_vcnt > 1) |
7828 | || (failed_bio->bi_io_vec->bv_len | |
7829 | > BTRFS_I(inode)->root->sectorsize)) | |
8b110e39 MX |
7830 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; |
7831 | else | |
7832 | read_mode = READ_SYNC; | |
7833 | ||
7834 | isector = start - btrfs_io_bio(failed_bio)->logical; | |
7835 | isector >>= inode->i_sb->s_blocksize_bits; | |
7836 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2dabb324 | 7837 | pgoff, isector, repair_endio, repair_arg); |
8b110e39 MX |
7838 | if (!bio) { |
7839 | free_io_failure(inode, failrec); | |
7840 | return -EIO; | |
7841 | } | |
7842 | ||
7843 | btrfs_debug(BTRFS_I(inode)->root->fs_info, | |
7844 | "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n", | |
7845 | read_mode, failrec->this_mirror, failrec->in_validation); | |
7846 | ||
7847 | ret = submit_dio_repair_bio(inode, bio, read_mode, | |
7848 | failrec->this_mirror); | |
7849 | if (ret) { | |
7850 | free_io_failure(inode, failrec); | |
7851 | bio_put(bio); | |
7852 | } | |
7853 | ||
7854 | return ret; | |
7855 | } | |
7856 | ||
7857 | struct btrfs_retry_complete { | |
7858 | struct completion done; | |
7859 | struct inode *inode; | |
7860 | u64 start; | |
7861 | int uptodate; | |
7862 | }; | |
7863 | ||
4246a0b6 | 7864 | static void btrfs_retry_endio_nocsum(struct bio *bio) |
8b110e39 MX |
7865 | { |
7866 | struct btrfs_retry_complete *done = bio->bi_private; | |
2dabb324 | 7867 | struct inode *inode; |
8b110e39 MX |
7868 | struct bio_vec *bvec; |
7869 | int i; | |
7870 | ||
4246a0b6 | 7871 | if (bio->bi_error) |
8b110e39 MX |
7872 | goto end; |
7873 | ||
2dabb324 CR |
7874 | ASSERT(bio->bi_vcnt == 1); |
7875 | inode = bio->bi_io_vec->bv_page->mapping->host; | |
7876 | ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize); | |
7877 | ||
8b110e39 MX |
7878 | done->uptodate = 1; |
7879 | bio_for_each_segment_all(bvec, bio, i) | |
7880 | clean_io_failure(done->inode, done->start, bvec->bv_page, 0); | |
7881 | end: | |
7882 | complete(&done->done); | |
7883 | bio_put(bio); | |
7884 | } | |
7885 | ||
7886 | static int __btrfs_correct_data_nocsum(struct inode *inode, | |
7887 | struct btrfs_io_bio *io_bio) | |
4b46fce2 | 7888 | { |
2dabb324 | 7889 | struct btrfs_fs_info *fs_info; |
2c30c71b | 7890 | struct bio_vec *bvec; |
8b110e39 | 7891 | struct btrfs_retry_complete done; |
4b46fce2 | 7892 | u64 start; |
2dabb324 CR |
7893 | unsigned int pgoff; |
7894 | u32 sectorsize; | |
7895 | int nr_sectors; | |
2c30c71b | 7896 | int i; |
c1dc0896 | 7897 | int ret; |
4b46fce2 | 7898 | |
2dabb324 CR |
7899 | fs_info = BTRFS_I(inode)->root->fs_info; |
7900 | sectorsize = BTRFS_I(inode)->root->sectorsize; | |
7901 | ||
8b110e39 MX |
7902 | start = io_bio->logical; |
7903 | done.inode = inode; | |
7904 | ||
7905 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { | |
2dabb324 CR |
7906 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len); |
7907 | pgoff = bvec->bv_offset; | |
7908 | ||
7909 | next_block_or_try_again: | |
8b110e39 MX |
7910 | done.uptodate = 0; |
7911 | done.start = start; | |
7912 | init_completion(&done.done); | |
7913 | ||
2dabb324 CR |
7914 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, |
7915 | pgoff, start, start + sectorsize - 1, | |
7916 | io_bio->mirror_num, | |
7917 | btrfs_retry_endio_nocsum, &done); | |
8b110e39 MX |
7918 | if (ret) |
7919 | return ret; | |
7920 | ||
7921 | wait_for_completion(&done.done); | |
7922 | ||
7923 | if (!done.uptodate) { | |
7924 | /* We might have another mirror, so try again */ | |
2dabb324 | 7925 | goto next_block_or_try_again; |
8b110e39 MX |
7926 | } |
7927 | ||
2dabb324 CR |
7928 | start += sectorsize; |
7929 | ||
7930 | if (nr_sectors--) { | |
7931 | pgoff += sectorsize; | |
7932 | goto next_block_or_try_again; | |
7933 | } | |
8b110e39 MX |
7934 | } |
7935 | ||
7936 | return 0; | |
7937 | } | |
7938 | ||
4246a0b6 | 7939 | static void btrfs_retry_endio(struct bio *bio) |
8b110e39 MX |
7940 | { |
7941 | struct btrfs_retry_complete *done = bio->bi_private; | |
7942 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
2dabb324 | 7943 | struct inode *inode; |
8b110e39 | 7944 | struct bio_vec *bvec; |
2dabb324 | 7945 | u64 start; |
8b110e39 MX |
7946 | int uptodate; |
7947 | int ret; | |
7948 | int i; | |
7949 | ||
4246a0b6 | 7950 | if (bio->bi_error) |
8b110e39 MX |
7951 | goto end; |
7952 | ||
7953 | uptodate = 1; | |
2dabb324 CR |
7954 | |
7955 | start = done->start; | |
7956 | ||
7957 | ASSERT(bio->bi_vcnt == 1); | |
7958 | inode = bio->bi_io_vec->bv_page->mapping->host; | |
7959 | ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize); | |
7960 | ||
8b110e39 MX |
7961 | bio_for_each_segment_all(bvec, bio, i) { |
7962 | ret = __readpage_endio_check(done->inode, io_bio, i, | |
2dabb324 CR |
7963 | bvec->bv_page, bvec->bv_offset, |
7964 | done->start, bvec->bv_len); | |
8b110e39 MX |
7965 | if (!ret) |
7966 | clean_io_failure(done->inode, done->start, | |
2dabb324 | 7967 | bvec->bv_page, bvec->bv_offset); |
8b110e39 MX |
7968 | else |
7969 | uptodate = 0; | |
7970 | } | |
7971 | ||
7972 | done->uptodate = uptodate; | |
7973 | end: | |
7974 | complete(&done->done); | |
7975 | bio_put(bio); | |
7976 | } | |
7977 | ||
7978 | static int __btrfs_subio_endio_read(struct inode *inode, | |
7979 | struct btrfs_io_bio *io_bio, int err) | |
7980 | { | |
2dabb324 | 7981 | struct btrfs_fs_info *fs_info; |
8b110e39 MX |
7982 | struct bio_vec *bvec; |
7983 | struct btrfs_retry_complete done; | |
7984 | u64 start; | |
7985 | u64 offset = 0; | |
2dabb324 CR |
7986 | u32 sectorsize; |
7987 | int nr_sectors; | |
7988 | unsigned int pgoff; | |
7989 | int csum_pos; | |
8b110e39 MX |
7990 | int i; |
7991 | int ret; | |
dc380aea | 7992 | |
2dabb324 CR |
7993 | fs_info = BTRFS_I(inode)->root->fs_info; |
7994 | sectorsize = BTRFS_I(inode)->root->sectorsize; | |
7995 | ||
8b110e39 | 7996 | err = 0; |
c1dc0896 | 7997 | start = io_bio->logical; |
8b110e39 MX |
7998 | done.inode = inode; |
7999 | ||
c1dc0896 | 8000 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { |
2dabb324 CR |
8001 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len); |
8002 | ||
8003 | pgoff = bvec->bv_offset; | |
8004 | next_block: | |
8005 | csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset); | |
8006 | ret = __readpage_endio_check(inode, io_bio, csum_pos, | |
8007 | bvec->bv_page, pgoff, start, | |
8008 | sectorsize); | |
8b110e39 MX |
8009 | if (likely(!ret)) |
8010 | goto next; | |
8011 | try_again: | |
8012 | done.uptodate = 0; | |
8013 | done.start = start; | |
8014 | init_completion(&done.done); | |
8015 | ||
2dabb324 CR |
8016 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, |
8017 | pgoff, start, start + sectorsize - 1, | |
8018 | io_bio->mirror_num, | |
8019 | btrfs_retry_endio, &done); | |
8b110e39 MX |
8020 | if (ret) { |
8021 | err = ret; | |
8022 | goto next; | |
8023 | } | |
8024 | ||
8025 | wait_for_completion(&done.done); | |
8026 | ||
8027 | if (!done.uptodate) { | |
8028 | /* We might have another mirror, so try again */ | |
8029 | goto try_again; | |
8030 | } | |
8031 | next: | |
2dabb324 CR |
8032 | offset += sectorsize; |
8033 | start += sectorsize; | |
8034 | ||
8035 | ASSERT(nr_sectors); | |
8036 | ||
8037 | if (--nr_sectors) { | |
8038 | pgoff += sectorsize; | |
8039 | goto next_block; | |
8040 | } | |
2c30c71b | 8041 | } |
c1dc0896 MX |
8042 | |
8043 | return err; | |
8044 | } | |
8045 | ||
8b110e39 MX |
8046 | static int btrfs_subio_endio_read(struct inode *inode, |
8047 | struct btrfs_io_bio *io_bio, int err) | |
8048 | { | |
8049 | bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8050 | ||
8051 | if (skip_csum) { | |
8052 | if (unlikely(err)) | |
8053 | return __btrfs_correct_data_nocsum(inode, io_bio); | |
8054 | else | |
8055 | return 0; | |
8056 | } else { | |
8057 | return __btrfs_subio_endio_read(inode, io_bio, err); | |
8058 | } | |
8059 | } | |
8060 | ||
4246a0b6 | 8061 | static void btrfs_endio_direct_read(struct bio *bio) |
c1dc0896 MX |
8062 | { |
8063 | struct btrfs_dio_private *dip = bio->bi_private; | |
8064 | struct inode *inode = dip->inode; | |
8065 | struct bio *dio_bio; | |
8066 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
4246a0b6 | 8067 | int err = bio->bi_error; |
c1dc0896 | 8068 | |
8b110e39 MX |
8069 | if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) |
8070 | err = btrfs_subio_endio_read(inode, io_bio, err); | |
c1dc0896 | 8071 | |
4b46fce2 | 8072 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, |
d0082371 | 8073 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 8074 | dio_bio = dip->dio_bio; |
4b46fce2 | 8075 | |
4b46fce2 | 8076 | kfree(dip); |
c0da7aa1 | 8077 | |
1636d1d7 | 8078 | dio_bio->bi_error = bio->bi_error; |
4246a0b6 | 8079 | dio_end_io(dio_bio, bio->bi_error); |
23ea8e5a MX |
8080 | |
8081 | if (io_bio->end_io) | |
8082 | io_bio->end_io(io_bio, err); | |
9be3395b | 8083 | bio_put(bio); |
4b46fce2 JB |
8084 | } |
8085 | ||
14543774 FM |
8086 | static void btrfs_endio_direct_write_update_ordered(struct inode *inode, |
8087 | const u64 offset, | |
8088 | const u64 bytes, | |
8089 | const int uptodate) | |
4b46fce2 | 8090 | { |
4b46fce2 | 8091 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4b46fce2 | 8092 | struct btrfs_ordered_extent *ordered = NULL; |
14543774 FM |
8093 | u64 ordered_offset = offset; |
8094 | u64 ordered_bytes = bytes; | |
4b46fce2 JB |
8095 | int ret; |
8096 | ||
163cf09c CM |
8097 | again: |
8098 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
8099 | &ordered_offset, | |
4246a0b6 | 8100 | ordered_bytes, |
14543774 | 8101 | uptodate); |
4b46fce2 | 8102 | if (!ret) |
163cf09c | 8103 | goto out_test; |
4b46fce2 | 8104 | |
9e0af237 LB |
8105 | btrfs_init_work(&ordered->work, btrfs_endio_write_helper, |
8106 | finish_ordered_fn, NULL, NULL); | |
fccb5d86 QW |
8107 | btrfs_queue_work(root->fs_info->endio_write_workers, |
8108 | &ordered->work); | |
163cf09c CM |
8109 | out_test: |
8110 | /* | |
8111 | * our bio might span multiple ordered extents. If we haven't | |
8112 | * completed the accounting for the whole dio, go back and try again | |
8113 | */ | |
14543774 FM |
8114 | if (ordered_offset < offset + bytes) { |
8115 | ordered_bytes = offset + bytes - ordered_offset; | |
5fd02043 | 8116 | ordered = NULL; |
163cf09c CM |
8117 | goto again; |
8118 | } | |
14543774 FM |
8119 | } |
8120 | ||
8121 | static void btrfs_endio_direct_write(struct bio *bio) | |
8122 | { | |
8123 | struct btrfs_dio_private *dip = bio->bi_private; | |
8124 | struct bio *dio_bio = dip->dio_bio; | |
8125 | ||
8126 | btrfs_endio_direct_write_update_ordered(dip->inode, | |
8127 | dip->logical_offset, | |
8128 | dip->bytes, | |
8129 | !bio->bi_error); | |
4b46fce2 | 8130 | |
4b46fce2 | 8131 | kfree(dip); |
c0da7aa1 | 8132 | |
1636d1d7 | 8133 | dio_bio->bi_error = bio->bi_error; |
4246a0b6 | 8134 | dio_end_io(dio_bio, bio->bi_error); |
9be3395b | 8135 | bio_put(bio); |
4b46fce2 JB |
8136 | } |
8137 | ||
eaf25d93 CM |
8138 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
8139 | struct bio *bio, int mirror_num, | |
8140 | unsigned long bio_flags, u64 offset) | |
8141 | { | |
8142 | int ret; | |
8143 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8144 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 8145 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
8146 | return 0; |
8147 | } | |
8148 | ||
4246a0b6 | 8149 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
8150 | { |
8151 | struct btrfs_dio_private *dip = bio->bi_private; | |
4246a0b6 | 8152 | int err = bio->bi_error; |
e65e1535 | 8153 | |
8b110e39 MX |
8154 | if (err) |
8155 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
8156 | "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d", | |
8157 | btrfs_ino(dip->inode), bio->bi_rw, | |
8158 | (unsigned long long)bio->bi_iter.bi_sector, | |
8159 | bio->bi_iter.bi_size, err); | |
8160 | ||
8161 | if (dip->subio_endio) | |
8162 | err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err); | |
c1dc0896 MX |
8163 | |
8164 | if (err) { | |
e65e1535 MX |
8165 | dip->errors = 1; |
8166 | ||
8167 | /* | |
8168 | * before atomic variable goto zero, we must make sure | |
8169 | * dip->errors is perceived to be set. | |
8170 | */ | |
4e857c58 | 8171 | smp_mb__before_atomic(); |
e65e1535 MX |
8172 | } |
8173 | ||
8174 | /* if there are more bios still pending for this dio, just exit */ | |
8175 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
8176 | goto out; | |
8177 | ||
9be3395b | 8178 | if (dip->errors) { |
e65e1535 | 8179 | bio_io_error(dip->orig_bio); |
9be3395b | 8180 | } else { |
4246a0b6 CH |
8181 | dip->dio_bio->bi_error = 0; |
8182 | bio_endio(dip->orig_bio); | |
e65e1535 MX |
8183 | } |
8184 | out: | |
8185 | bio_put(bio); | |
8186 | } | |
8187 | ||
8188 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
8189 | u64 first_sector, gfp_t gfp_flags) | |
8190 | { | |
da2f0f74 | 8191 | struct bio *bio; |
22365979 | 8192 | bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags); |
da2f0f74 CM |
8193 | if (bio) |
8194 | bio_associate_current(bio); | |
8195 | return bio; | |
e65e1535 MX |
8196 | } |
8197 | ||
c1dc0896 MX |
8198 | static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root, |
8199 | struct inode *inode, | |
8200 | struct btrfs_dio_private *dip, | |
8201 | struct bio *bio, | |
8202 | u64 file_offset) | |
8203 | { | |
8204 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
8205 | struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio); | |
8206 | int ret; | |
8207 | ||
8208 | /* | |
8209 | * We load all the csum data we need when we submit | |
8210 | * the first bio to reduce the csum tree search and | |
8211 | * contention. | |
8212 | */ | |
8213 | if (dip->logical_offset == file_offset) { | |
8214 | ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio, | |
8215 | file_offset); | |
8216 | if (ret) | |
8217 | return ret; | |
8218 | } | |
8219 | ||
8220 | if (bio == dip->orig_bio) | |
8221 | return 0; | |
8222 | ||
8223 | file_offset -= dip->logical_offset; | |
8224 | file_offset >>= inode->i_sb->s_blocksize_bits; | |
8225 | io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset); | |
8226 | ||
8227 | return 0; | |
8228 | } | |
8229 | ||
e65e1535 MX |
8230 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, |
8231 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 8232 | int async_submit) |
e65e1535 | 8233 | { |
facc8a22 | 8234 | struct btrfs_dio_private *dip = bio->bi_private; |
e65e1535 MX |
8235 | int write = rw & REQ_WRITE; |
8236 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8237 | int ret; | |
8238 | ||
b812ce28 JB |
8239 | if (async_submit) |
8240 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
8241 | ||
e65e1535 | 8242 | bio_get(bio); |
5fd02043 JB |
8243 | |
8244 | if (!write) { | |
bfebd8b5 DS |
8245 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, |
8246 | BTRFS_WQ_ENDIO_DATA); | |
5fd02043 JB |
8247 | if (ret) |
8248 | goto err; | |
8249 | } | |
e65e1535 | 8250 | |
1ae39938 JB |
8251 | if (skip_sum) |
8252 | goto map; | |
8253 | ||
8254 | if (write && async_submit) { | |
e65e1535 MX |
8255 | ret = btrfs_wq_submit_bio(root->fs_info, |
8256 | inode, rw, bio, 0, 0, | |
8257 | file_offset, | |
8258 | __btrfs_submit_bio_start_direct_io, | |
8259 | __btrfs_submit_bio_done); | |
8260 | goto err; | |
1ae39938 JB |
8261 | } else if (write) { |
8262 | /* | |
8263 | * If we aren't doing async submit, calculate the csum of the | |
8264 | * bio now. | |
8265 | */ | |
8266 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
8267 | if (ret) | |
8268 | goto err; | |
23ea8e5a | 8269 | } else { |
c1dc0896 MX |
8270 | ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio, |
8271 | file_offset); | |
c2db1073 TI |
8272 | if (ret) |
8273 | goto err; | |
8274 | } | |
1ae39938 JB |
8275 | map: |
8276 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
8277 | err: |
8278 | bio_put(bio); | |
8279 | return ret; | |
8280 | } | |
8281 | ||
8282 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
8283 | int skip_sum) | |
8284 | { | |
8285 | struct inode *inode = dip->inode; | |
8286 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
e65e1535 MX |
8287 | struct bio *bio; |
8288 | struct bio *orig_bio = dip->orig_bio; | |
8289 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
4f024f37 | 8290 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 MX |
8291 | u64 file_offset = dip->logical_offset; |
8292 | u64 submit_len = 0; | |
8293 | u64 map_length; | |
5f4dc8fc | 8294 | u32 blocksize = root->sectorsize; |
1ae39938 | 8295 | int async_submit = 0; |
5f4dc8fc CR |
8296 | int nr_sectors; |
8297 | int ret; | |
8298 | int i; | |
e65e1535 | 8299 | |
4f024f37 | 8300 | map_length = orig_bio->bi_iter.bi_size; |
53b381b3 | 8301 | ret = btrfs_map_block(root->fs_info, rw, start_sector << 9, |
e65e1535 | 8302 | &map_length, NULL, 0); |
7a5c3c9b | 8303 | if (ret) |
e65e1535 | 8304 | return -EIO; |
facc8a22 | 8305 | |
4f024f37 | 8306 | if (map_length >= orig_bio->bi_iter.bi_size) { |
02f57c7a | 8307 | bio = orig_bio; |
c1dc0896 | 8308 | dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED; |
02f57c7a JB |
8309 | goto submit; |
8310 | } | |
8311 | ||
53b381b3 | 8312 | /* async crcs make it difficult to collect full stripe writes. */ |
ffe2d203 | 8313 | if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
8314 | async_submit = 0; |
8315 | else | |
8316 | async_submit = 1; | |
8317 | ||
02f57c7a JB |
8318 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
8319 | if (!bio) | |
8320 | return -ENOMEM; | |
7a5c3c9b | 8321 | |
02f57c7a JB |
8322 | bio->bi_private = dip; |
8323 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 8324 | btrfs_io_bio(bio)->logical = file_offset; |
02f57c7a JB |
8325 | atomic_inc(&dip->pending_bios); |
8326 | ||
e65e1535 | 8327 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
5f4dc8fc CR |
8328 | nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len); |
8329 | i = 0; | |
8330 | next_block: | |
8331 | if (unlikely(map_length < submit_len + blocksize || | |
8332 | bio_add_page(bio, bvec->bv_page, blocksize, | |
8333 | bvec->bv_offset + (i * blocksize)) < blocksize)) { | |
e65e1535 MX |
8334 | /* |
8335 | * inc the count before we submit the bio so | |
8336 | * we know the end IO handler won't happen before | |
8337 | * we inc the count. Otherwise, the dip might get freed | |
8338 | * before we're done setting it up | |
8339 | */ | |
8340 | atomic_inc(&dip->pending_bios); | |
8341 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
8342 | file_offset, skip_sum, | |
c329861d | 8343 | async_submit); |
e65e1535 MX |
8344 | if (ret) { |
8345 | bio_put(bio); | |
8346 | atomic_dec(&dip->pending_bios); | |
8347 | goto out_err; | |
8348 | } | |
8349 | ||
e65e1535 MX |
8350 | start_sector += submit_len >> 9; |
8351 | file_offset += submit_len; | |
8352 | ||
8353 | submit_len = 0; | |
e65e1535 MX |
8354 | |
8355 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
8356 | start_sector, GFP_NOFS); | |
8357 | if (!bio) | |
8358 | goto out_err; | |
8359 | bio->bi_private = dip; | |
8360 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 8361 | btrfs_io_bio(bio)->logical = file_offset; |
e65e1535 | 8362 | |
4f024f37 | 8363 | map_length = orig_bio->bi_iter.bi_size; |
53b381b3 | 8364 | ret = btrfs_map_block(root->fs_info, rw, |
3ec706c8 | 8365 | start_sector << 9, |
e65e1535 MX |
8366 | &map_length, NULL, 0); |
8367 | if (ret) { | |
8368 | bio_put(bio); | |
8369 | goto out_err; | |
8370 | } | |
5f4dc8fc CR |
8371 | |
8372 | goto next_block; | |
e65e1535 | 8373 | } else { |
5f4dc8fc CR |
8374 | submit_len += blocksize; |
8375 | if (--nr_sectors) { | |
8376 | i++; | |
8377 | goto next_block; | |
8378 | } | |
e65e1535 MX |
8379 | bvec++; |
8380 | } | |
8381 | } | |
8382 | ||
02f57c7a | 8383 | submit: |
e65e1535 | 8384 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 8385 | async_submit); |
e65e1535 MX |
8386 | if (!ret) |
8387 | return 0; | |
8388 | ||
8389 | bio_put(bio); | |
8390 | out_err: | |
8391 | dip->errors = 1; | |
8392 | /* | |
8393 | * before atomic variable goto zero, we must | |
8394 | * make sure dip->errors is perceived to be set. | |
8395 | */ | |
4e857c58 | 8396 | smp_mb__before_atomic(); |
e65e1535 MX |
8397 | if (atomic_dec_and_test(&dip->pending_bios)) |
8398 | bio_io_error(dip->orig_bio); | |
8399 | ||
8400 | /* bio_end_io() will handle error, so we needn't return it */ | |
8401 | return 0; | |
8402 | } | |
8403 | ||
9be3395b CM |
8404 | static void btrfs_submit_direct(int rw, struct bio *dio_bio, |
8405 | struct inode *inode, loff_t file_offset) | |
4b46fce2 | 8406 | { |
61de718f FM |
8407 | struct btrfs_dio_private *dip = NULL; |
8408 | struct bio *io_bio = NULL; | |
23ea8e5a | 8409 | struct btrfs_io_bio *btrfs_bio; |
4b46fce2 | 8410 | int skip_sum; |
7b6d91da | 8411 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
8412 | int ret = 0; |
8413 | ||
8414 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8415 | ||
9be3395b | 8416 | io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS); |
9be3395b CM |
8417 | if (!io_bio) { |
8418 | ret = -ENOMEM; | |
8419 | goto free_ordered; | |
8420 | } | |
8421 | ||
c1dc0896 | 8422 | dip = kzalloc(sizeof(*dip), GFP_NOFS); |
4b46fce2 JB |
8423 | if (!dip) { |
8424 | ret = -ENOMEM; | |
61de718f | 8425 | goto free_ordered; |
4b46fce2 | 8426 | } |
4b46fce2 | 8427 | |
9be3395b | 8428 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
8429 | dip->inode = inode; |
8430 | dip->logical_offset = file_offset; | |
4f024f37 KO |
8431 | dip->bytes = dio_bio->bi_iter.bi_size; |
8432 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
9be3395b | 8433 | io_bio->bi_private = dip; |
9be3395b CM |
8434 | dip->orig_bio = io_bio; |
8435 | dip->dio_bio = dio_bio; | |
e65e1535 | 8436 | atomic_set(&dip->pending_bios, 0); |
c1dc0896 MX |
8437 | btrfs_bio = btrfs_io_bio(io_bio); |
8438 | btrfs_bio->logical = file_offset; | |
4b46fce2 | 8439 | |
c1dc0896 | 8440 | if (write) { |
9be3395b | 8441 | io_bio->bi_end_io = btrfs_endio_direct_write; |
c1dc0896 | 8442 | } else { |
9be3395b | 8443 | io_bio->bi_end_io = btrfs_endio_direct_read; |
c1dc0896 MX |
8444 | dip->subio_endio = btrfs_subio_endio_read; |
8445 | } | |
4b46fce2 | 8446 | |
f28a4928 FM |
8447 | /* |
8448 | * Reset the range for unsubmitted ordered extents (to a 0 length range) | |
8449 | * even if we fail to submit a bio, because in such case we do the | |
8450 | * corresponding error handling below and it must not be done a second | |
8451 | * time by btrfs_direct_IO(). | |
8452 | */ | |
8453 | if (write) { | |
8454 | struct btrfs_dio_data *dio_data = current->journal_info; | |
8455 | ||
8456 | dio_data->unsubmitted_oe_range_end = dip->logical_offset + | |
8457 | dip->bytes; | |
8458 | dio_data->unsubmitted_oe_range_start = | |
8459 | dio_data->unsubmitted_oe_range_end; | |
8460 | } | |
8461 | ||
e65e1535 MX |
8462 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
8463 | if (!ret) | |
eaf25d93 | 8464 | return; |
9be3395b | 8465 | |
23ea8e5a MX |
8466 | if (btrfs_bio->end_io) |
8467 | btrfs_bio->end_io(btrfs_bio, ret); | |
9be3395b | 8468 | |
4b46fce2 JB |
8469 | free_ordered: |
8470 | /* | |
61de718f FM |
8471 | * If we arrived here it means either we failed to submit the dip |
8472 | * or we either failed to clone the dio_bio or failed to allocate the | |
8473 | * dip. If we cloned the dio_bio and allocated the dip, we can just | |
8474 | * call bio_endio against our io_bio so that we get proper resource | |
8475 | * cleanup if we fail to submit the dip, otherwise, we must do the | |
8476 | * same as btrfs_endio_direct_[write|read] because we can't call these | |
8477 | * callbacks - they require an allocated dip and a clone of dio_bio. | |
4b46fce2 | 8478 | */ |
61de718f | 8479 | if (io_bio && dip) { |
4246a0b6 CH |
8480 | io_bio->bi_error = -EIO; |
8481 | bio_endio(io_bio); | |
61de718f FM |
8482 | /* |
8483 | * The end io callbacks free our dip, do the final put on io_bio | |
8484 | * and all the cleanup and final put for dio_bio (through | |
8485 | * dio_end_io()). | |
8486 | */ | |
8487 | dip = NULL; | |
8488 | io_bio = NULL; | |
8489 | } else { | |
14543774 FM |
8490 | if (write) |
8491 | btrfs_endio_direct_write_update_ordered(inode, | |
8492 | file_offset, | |
8493 | dio_bio->bi_iter.bi_size, | |
8494 | 0); | |
8495 | else | |
61de718f FM |
8496 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, |
8497 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
14543774 | 8498 | |
4246a0b6 | 8499 | dio_bio->bi_error = -EIO; |
61de718f FM |
8500 | /* |
8501 | * Releases and cleans up our dio_bio, no need to bio_put() | |
8502 | * nor bio_endio()/bio_io_error() against dio_bio. | |
8503 | */ | |
8504 | dio_end_io(dio_bio, ret); | |
4b46fce2 | 8505 | } |
61de718f FM |
8506 | if (io_bio) |
8507 | bio_put(io_bio); | |
8508 | kfree(dip); | |
4b46fce2 JB |
8509 | } |
8510 | ||
6f673763 | 8511 | static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb, |
28060d5d | 8512 | const struct iov_iter *iter, loff_t offset) |
5a5f79b5 CM |
8513 | { |
8514 | int seg; | |
a1b75f7d | 8515 | int i; |
5a5f79b5 CM |
8516 | unsigned blocksize_mask = root->sectorsize - 1; |
8517 | ssize_t retval = -EINVAL; | |
5a5f79b5 CM |
8518 | |
8519 | if (offset & blocksize_mask) | |
8520 | goto out; | |
8521 | ||
28060d5d AV |
8522 | if (iov_iter_alignment(iter) & blocksize_mask) |
8523 | goto out; | |
a1b75f7d | 8524 | |
28060d5d | 8525 | /* If this is a write we don't need to check anymore */ |
6f673763 | 8526 | if (iov_iter_rw(iter) == WRITE) |
28060d5d AV |
8527 | return 0; |
8528 | /* | |
8529 | * Check to make sure we don't have duplicate iov_base's in this | |
8530 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
8531 | * when reading back. | |
8532 | */ | |
8533 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
8534 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
8535 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
8536 | goto out; |
8537 | } | |
5a5f79b5 CM |
8538 | } |
8539 | retval = 0; | |
8540 | out: | |
8541 | return retval; | |
8542 | } | |
eb838e73 | 8543 | |
c8b8e32d | 8544 | static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
16432985 | 8545 | { |
4b46fce2 JB |
8546 | struct file *file = iocb->ki_filp; |
8547 | struct inode *inode = file->f_mapping->host; | |
50745b0a | 8548 | struct btrfs_root *root = BTRFS_I(inode)->root; |
8549 | struct btrfs_dio_data dio_data = { 0 }; | |
c8b8e32d | 8550 | loff_t offset = iocb->ki_pos; |
0934856d | 8551 | size_t count = 0; |
2e60a51e | 8552 | int flags = 0; |
38851cc1 MX |
8553 | bool wakeup = true; |
8554 | bool relock = false; | |
0934856d | 8555 | ssize_t ret; |
4b46fce2 | 8556 | |
6f673763 | 8557 | if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset)) |
5a5f79b5 | 8558 | return 0; |
3f7c579c | 8559 | |
fe0f07d0 | 8560 | inode_dio_begin(inode); |
4e857c58 | 8561 | smp_mb__after_atomic(); |
38851cc1 | 8562 | |
0e267c44 | 8563 | /* |
41bd9ca4 MX |
8564 | * The generic stuff only does filemap_write_and_wait_range, which |
8565 | * isn't enough if we've written compressed pages to this area, so | |
8566 | * we need to flush the dirty pages again to make absolutely sure | |
8567 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 8568 | */ |
a6cbcd4a | 8569 | count = iov_iter_count(iter); |
41bd9ca4 MX |
8570 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
8571 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
8572 | filemap_fdatawrite_range(inode->i_mapping, offset, |
8573 | offset + count - 1); | |
0e267c44 | 8574 | |
6f673763 | 8575 | if (iov_iter_rw(iter) == WRITE) { |
38851cc1 MX |
8576 | /* |
8577 | * If the write DIO is beyond the EOF, we need update | |
8578 | * the isize, but it is protected by i_mutex. So we can | |
8579 | * not unlock the i_mutex at this case. | |
8580 | */ | |
8581 | if (offset + count <= inode->i_size) { | |
5955102c | 8582 | inode_unlock(inode); |
38851cc1 MX |
8583 | relock = true; |
8584 | } | |
7cf5b976 | 8585 | ret = btrfs_delalloc_reserve_space(inode, offset, count); |
0934856d | 8586 | if (ret) |
38851cc1 | 8587 | goto out; |
50745b0a | 8588 | dio_data.outstanding_extents = div64_u64(count + |
e1cbbfa5 JB |
8589 | BTRFS_MAX_EXTENT_SIZE - 1, |
8590 | BTRFS_MAX_EXTENT_SIZE); | |
8591 | ||
8592 | /* | |
8593 | * We need to know how many extents we reserved so that we can | |
8594 | * do the accounting properly if we go over the number we | |
8595 | * originally calculated. Abuse current->journal_info for this. | |
8596 | */ | |
50745b0a | 8597 | dio_data.reserve = round_up(count, root->sectorsize); |
f28a4928 FM |
8598 | dio_data.unsubmitted_oe_range_start = (u64)offset; |
8599 | dio_data.unsubmitted_oe_range_end = (u64)offset; | |
50745b0a | 8600 | current->journal_info = &dio_data; |
ee39b432 DS |
8601 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
8602 | &BTRFS_I(inode)->runtime_flags)) { | |
fe0f07d0 | 8603 | inode_dio_end(inode); |
38851cc1 MX |
8604 | flags = DIO_LOCKING | DIO_SKIP_HOLES; |
8605 | wakeup = false; | |
0934856d MX |
8606 | } |
8607 | ||
17f8c842 OS |
8608 | ret = __blockdev_direct_IO(iocb, inode, |
8609 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
c8b8e32d | 8610 | iter, btrfs_get_blocks_direct, NULL, |
17f8c842 | 8611 | btrfs_submit_direct, flags); |
6f673763 | 8612 | if (iov_iter_rw(iter) == WRITE) { |
e1cbbfa5 | 8613 | current->journal_info = NULL; |
ddba1bfc | 8614 | if (ret < 0 && ret != -EIOCBQUEUED) { |
50745b0a | 8615 | if (dio_data.reserve) |
7cf5b976 QW |
8616 | btrfs_delalloc_release_space(inode, offset, |
8617 | dio_data.reserve); | |
f28a4928 FM |
8618 | /* |
8619 | * On error we might have left some ordered extents | |
8620 | * without submitting corresponding bios for them, so | |
8621 | * cleanup them up to avoid other tasks getting them | |
8622 | * and waiting for them to complete forever. | |
8623 | */ | |
8624 | if (dio_data.unsubmitted_oe_range_start < | |
8625 | dio_data.unsubmitted_oe_range_end) | |
8626 | btrfs_endio_direct_write_update_ordered(inode, | |
8627 | dio_data.unsubmitted_oe_range_start, | |
8628 | dio_data.unsubmitted_oe_range_end - | |
8629 | dio_data.unsubmitted_oe_range_start, | |
8630 | 0); | |
ddba1bfc | 8631 | } else if (ret >= 0 && (size_t)ret < count) |
7cf5b976 QW |
8632 | btrfs_delalloc_release_space(inode, offset, |
8633 | count - (size_t)ret); | |
0934856d | 8634 | } |
38851cc1 | 8635 | out: |
2e60a51e | 8636 | if (wakeup) |
fe0f07d0 | 8637 | inode_dio_end(inode); |
38851cc1 | 8638 | if (relock) |
5955102c | 8639 | inode_lock(inode); |
0934856d MX |
8640 | |
8641 | return ret; | |
16432985 CM |
8642 | } |
8643 | ||
05dadc09 TI |
8644 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
8645 | ||
1506fcc8 YS |
8646 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
8647 | __u64 start, __u64 len) | |
8648 | { | |
05dadc09 TI |
8649 | int ret; |
8650 | ||
8651 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
8652 | if (ret) | |
8653 | return ret; | |
8654 | ||
ec29ed5b | 8655 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
8656 | } |
8657 | ||
a52d9a80 | 8658 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8659 | { |
d1310b2e CM |
8660 | struct extent_io_tree *tree; |
8661 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 8662 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 8663 | } |
1832a6d5 | 8664 | |
a52d9a80 | 8665 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8666 | { |
d1310b2e | 8667 | struct extent_io_tree *tree; |
be7bd730 JB |
8668 | struct inode *inode = page->mapping->host; |
8669 | int ret; | |
b888db2b CM |
8670 | |
8671 | if (current->flags & PF_MEMALLOC) { | |
8672 | redirty_page_for_writepage(wbc, page); | |
8673 | unlock_page(page); | |
8674 | return 0; | |
8675 | } | |
be7bd730 JB |
8676 | |
8677 | /* | |
8678 | * If we are under memory pressure we will call this directly from the | |
8679 | * VM, we need to make sure we have the inode referenced for the ordered | |
8680 | * extent. If not just return like we didn't do anything. | |
8681 | */ | |
8682 | if (!igrab(inode)) { | |
8683 | redirty_page_for_writepage(wbc, page); | |
8684 | return AOP_WRITEPAGE_ACTIVATE; | |
8685 | } | |
d1310b2e | 8686 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
be7bd730 JB |
8687 | ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
8688 | btrfs_add_delayed_iput(inode); | |
8689 | return ret; | |
9ebefb18 CM |
8690 | } |
8691 | ||
48a3b636 ES |
8692 | static int btrfs_writepages(struct address_space *mapping, |
8693 | struct writeback_control *wbc) | |
b293f02e | 8694 | { |
d1310b2e | 8695 | struct extent_io_tree *tree; |
771ed689 | 8696 | |
d1310b2e | 8697 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
8698 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
8699 | } | |
8700 | ||
3ab2fb5a CM |
8701 | static int |
8702 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
8703 | struct list_head *pages, unsigned nr_pages) | |
8704 | { | |
d1310b2e CM |
8705 | struct extent_io_tree *tree; |
8706 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
8707 | return extent_readpages(tree, mapping, pages, nr_pages, |
8708 | btrfs_get_extent); | |
8709 | } | |
e6dcd2dc | 8710 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8711 | { |
d1310b2e CM |
8712 | struct extent_io_tree *tree; |
8713 | struct extent_map_tree *map; | |
a52d9a80 | 8714 | int ret; |
8c2383c3 | 8715 | |
d1310b2e CM |
8716 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
8717 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 8718 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
8719 | if (ret == 1) { |
8720 | ClearPagePrivate(page); | |
8721 | set_page_private(page, 0); | |
09cbfeaf | 8722 | put_page(page); |
39279cc3 | 8723 | } |
a52d9a80 | 8724 | return ret; |
39279cc3 CM |
8725 | } |
8726 | ||
e6dcd2dc CM |
8727 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8728 | { | |
98509cfc CM |
8729 | if (PageWriteback(page) || PageDirty(page)) |
8730 | return 0; | |
b335b003 | 8731 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
8732 | } |
8733 | ||
d47992f8 LC |
8734 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8735 | unsigned int length) | |
39279cc3 | 8736 | { |
5fd02043 | 8737 | struct inode *inode = page->mapping->host; |
d1310b2e | 8738 | struct extent_io_tree *tree; |
e6dcd2dc | 8739 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8740 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 8741 | u64 page_start = page_offset(page); |
09cbfeaf | 8742 | u64 page_end = page_start + PAGE_SIZE - 1; |
dbfdb6d1 CR |
8743 | u64 start; |
8744 | u64 end; | |
131e404a | 8745 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 8746 | |
8b62b72b CM |
8747 | /* |
8748 | * we have the page locked, so new writeback can't start, | |
8749 | * and the dirty bit won't be cleared while we are here. | |
8750 | * | |
8751 | * Wait for IO on this page so that we can safely clear | |
8752 | * the PagePrivate2 bit and do ordered accounting | |
8753 | */ | |
e6dcd2dc | 8754 | wait_on_page_writeback(page); |
8b62b72b | 8755 | |
5fd02043 | 8756 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
8757 | if (offset) { |
8758 | btrfs_releasepage(page, GFP_NOFS); | |
8759 | return; | |
8760 | } | |
131e404a FDBM |
8761 | |
8762 | if (!inode_evicting) | |
ff13db41 | 8763 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
dbfdb6d1 CR |
8764 | again: |
8765 | start = page_start; | |
8766 | ordered = btrfs_lookup_ordered_range(inode, start, | |
8767 | page_end - start + 1); | |
e6dcd2dc | 8768 | if (ordered) { |
dbfdb6d1 | 8769 | end = min(page_end, ordered->file_offset + ordered->len - 1); |
eb84ae03 CM |
8770 | /* |
8771 | * IO on this page will never be started, so we need | |
8772 | * to account for any ordered extents now | |
8773 | */ | |
131e404a | 8774 | if (!inode_evicting) |
dbfdb6d1 | 8775 | clear_extent_bit(tree, start, end, |
131e404a FDBM |
8776 | EXTENT_DIRTY | EXTENT_DELALLOC | |
8777 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | | |
8778 | EXTENT_DEFRAG, 1, 0, &cached_state, | |
8779 | GFP_NOFS); | |
8b62b72b CM |
8780 | /* |
8781 | * whoever cleared the private bit is responsible | |
8782 | * for the finish_ordered_io | |
8783 | */ | |
77cef2ec JB |
8784 | if (TestClearPagePrivate2(page)) { |
8785 | struct btrfs_ordered_inode_tree *tree; | |
8786 | u64 new_len; | |
8787 | ||
8788 | tree = &BTRFS_I(inode)->ordered_tree; | |
8789 | ||
8790 | spin_lock_irq(&tree->lock); | |
8791 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
dbfdb6d1 | 8792 | new_len = start - ordered->file_offset; |
77cef2ec JB |
8793 | if (new_len < ordered->truncated_len) |
8794 | ordered->truncated_len = new_len; | |
8795 | spin_unlock_irq(&tree->lock); | |
8796 | ||
8797 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
dbfdb6d1 CR |
8798 | start, |
8799 | end - start + 1, 1)) | |
77cef2ec | 8800 | btrfs_finish_ordered_io(ordered); |
8b62b72b | 8801 | } |
e6dcd2dc | 8802 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
8803 | if (!inode_evicting) { |
8804 | cached_state = NULL; | |
dbfdb6d1 | 8805 | lock_extent_bits(tree, start, end, |
131e404a FDBM |
8806 | &cached_state); |
8807 | } | |
dbfdb6d1 CR |
8808 | |
8809 | start = end + 1; | |
8810 | if (start < page_end) | |
8811 | goto again; | |
131e404a FDBM |
8812 | } |
8813 | ||
b9d0b389 QW |
8814 | /* |
8815 | * Qgroup reserved space handler | |
8816 | * Page here will be either | |
8817 | * 1) Already written to disk | |
8818 | * In this case, its reserved space is released from data rsv map | |
8819 | * and will be freed by delayed_ref handler finally. | |
8820 | * So even we call qgroup_free_data(), it won't decrease reserved | |
8821 | * space. | |
8822 | * 2) Not written to disk | |
8823 | * This means the reserved space should be freed here. | |
8824 | */ | |
09cbfeaf | 8825 | btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE); |
131e404a FDBM |
8826 | if (!inode_evicting) { |
8827 | clear_extent_bit(tree, page_start, page_end, | |
8828 | EXTENT_LOCKED | EXTENT_DIRTY | | |
8829 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
8830 | EXTENT_DEFRAG, 1, 1, | |
8831 | &cached_state, GFP_NOFS); | |
8832 | ||
8833 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 8834 | } |
e6dcd2dc | 8835 | |
4a096752 | 8836 | ClearPageChecked(page); |
9ad6b7bc | 8837 | if (PagePrivate(page)) { |
9ad6b7bc CM |
8838 | ClearPagePrivate(page); |
8839 | set_page_private(page, 0); | |
09cbfeaf | 8840 | put_page(page); |
9ad6b7bc | 8841 | } |
39279cc3 CM |
8842 | } |
8843 | ||
9ebefb18 CM |
8844 | /* |
8845 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8846 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8847 | * be careful to check for EOF conditions here. We set the page up correctly | |
8848 | * for a written page which means we get ENOSPC checking when writing into | |
8849 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8850 | * support these features. | |
8851 | * | |
8852 | * We are not allowed to take the i_mutex here so we have to play games to | |
8853 | * protect against truncate races as the page could now be beyond EOF. Because | |
8854 | * vmtruncate() writes the inode size before removing pages, once we have the | |
8855 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
8856 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
8857 | * unlock the page. | |
8858 | */ | |
c2ec175c | 8859 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 8860 | { |
c2ec175c | 8861 | struct page *page = vmf->page; |
496ad9aa | 8862 | struct inode *inode = file_inode(vma->vm_file); |
1832a6d5 | 8863 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
8864 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8865 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8866 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
8867 | char *kaddr; |
8868 | unsigned long zero_start; | |
9ebefb18 | 8869 | loff_t size; |
1832a6d5 | 8870 | int ret; |
9998eb70 | 8871 | int reserved = 0; |
d0b7da88 | 8872 | u64 reserved_space; |
a52d9a80 | 8873 | u64 page_start; |
e6dcd2dc | 8874 | u64 page_end; |
d0b7da88 CR |
8875 | u64 end; |
8876 | ||
09cbfeaf | 8877 | reserved_space = PAGE_SIZE; |
9ebefb18 | 8878 | |
b2b5ef5c | 8879 | sb_start_pagefault(inode->i_sb); |
df480633 | 8880 | page_start = page_offset(page); |
09cbfeaf | 8881 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 8882 | end = page_end; |
df480633 | 8883 | |
d0b7da88 CR |
8884 | /* |
8885 | * Reserving delalloc space after obtaining the page lock can lead to | |
8886 | * deadlock. For example, if a dirty page is locked by this function | |
8887 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
8888 | * dirty page write out, then the btrfs_writepage() function could | |
8889 | * end up waiting indefinitely to get a lock on the page currently | |
8890 | * being processed by btrfs_page_mkwrite() function. | |
8891 | */ | |
7cf5b976 | 8892 | ret = btrfs_delalloc_reserve_space(inode, page_start, |
d0b7da88 | 8893 | reserved_space); |
9998eb70 | 8894 | if (!ret) { |
e41f941a | 8895 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
8896 | reserved = 1; |
8897 | } | |
56a76f82 NP |
8898 | if (ret) { |
8899 | if (ret == -ENOMEM) | |
8900 | ret = VM_FAULT_OOM; | |
8901 | else /* -ENOSPC, -EIO, etc */ | |
8902 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
8903 | if (reserved) |
8904 | goto out; | |
8905 | goto out_noreserve; | |
56a76f82 | 8906 | } |
1832a6d5 | 8907 | |
56a76f82 | 8908 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8909 | again: |
9ebefb18 | 8910 | lock_page(page); |
9ebefb18 | 8911 | size = i_size_read(inode); |
a52d9a80 | 8912 | |
9ebefb18 | 8913 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8914 | (page_start >= size)) { |
9ebefb18 CM |
8915 | /* page got truncated out from underneath us */ |
8916 | goto out_unlock; | |
8917 | } | |
e6dcd2dc CM |
8918 | wait_on_page_writeback(page); |
8919 | ||
ff13db41 | 8920 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc CM |
8921 | set_page_extent_mapped(page); |
8922 | ||
eb84ae03 CM |
8923 | /* |
8924 | * we can't set the delalloc bits if there are pending ordered | |
8925 | * extents. Drop our locks and wait for them to finish | |
8926 | */ | |
d0b7da88 | 8927 | ordered = btrfs_lookup_ordered_range(inode, page_start, page_end); |
e6dcd2dc | 8928 | if (ordered) { |
2ac55d41 JB |
8929 | unlock_extent_cached(io_tree, page_start, page_end, |
8930 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 8931 | unlock_page(page); |
eb84ae03 | 8932 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
8933 | btrfs_put_ordered_extent(ordered); |
8934 | goto again; | |
8935 | } | |
8936 | ||
09cbfeaf | 8937 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
d0b7da88 | 8938 | reserved_space = round_up(size - page_start, root->sectorsize); |
09cbfeaf | 8939 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 CR |
8940 | end = page_start + reserved_space - 1; |
8941 | spin_lock(&BTRFS_I(inode)->lock); | |
8942 | BTRFS_I(inode)->outstanding_extents++; | |
8943 | spin_unlock(&BTRFS_I(inode)->lock); | |
8944 | btrfs_delalloc_release_space(inode, page_start, | |
09cbfeaf | 8945 | PAGE_SIZE - reserved_space); |
d0b7da88 CR |
8946 | } |
8947 | } | |
8948 | ||
fbf19087 JB |
8949 | /* |
8950 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
8951 | * if it was already dirty, so for space accounting reasons we need to | |
8952 | * clear any delalloc bits for the range we are fixing to save. There | |
8953 | * is probably a better way to do this, but for now keep consistent with | |
8954 | * prepare_pages in the normal write path. | |
8955 | */ | |
d0b7da88 | 8956 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
9e8a4a8b LB |
8957 | EXTENT_DIRTY | EXTENT_DELALLOC | |
8958 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 8959 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 8960 | |
d0b7da88 | 8961 | ret = btrfs_set_extent_delalloc(inode, page_start, end, |
2ac55d41 | 8962 | &cached_state); |
9ed74f2d | 8963 | if (ret) { |
2ac55d41 JB |
8964 | unlock_extent_cached(io_tree, page_start, page_end, |
8965 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
8966 | ret = VM_FAULT_SIGBUS; |
8967 | goto out_unlock; | |
8968 | } | |
e6dcd2dc | 8969 | ret = 0; |
9ebefb18 CM |
8970 | |
8971 | /* page is wholly or partially inside EOF */ | |
09cbfeaf KS |
8972 | if (page_start + PAGE_SIZE > size) |
8973 | zero_start = size & ~PAGE_MASK; | |
9ebefb18 | 8974 | else |
09cbfeaf | 8975 | zero_start = PAGE_SIZE; |
9ebefb18 | 8976 | |
09cbfeaf | 8977 | if (zero_start != PAGE_SIZE) { |
e6dcd2dc | 8978 | kaddr = kmap(page); |
09cbfeaf | 8979 | memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start); |
e6dcd2dc CM |
8980 | flush_dcache_page(page); |
8981 | kunmap(page); | |
8982 | } | |
247e743c | 8983 | ClearPageChecked(page); |
e6dcd2dc | 8984 | set_page_dirty(page); |
50a9b214 | 8985 | SetPageUptodate(page); |
5a3f23d5 | 8986 | |
257c62e1 CM |
8987 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
8988 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 8989 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 8990 | |
2ac55d41 | 8991 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
8992 | |
8993 | out_unlock: | |
b2b5ef5c JK |
8994 | if (!ret) { |
8995 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 8996 | return VM_FAULT_LOCKED; |
b2b5ef5c | 8997 | } |
9ebefb18 | 8998 | unlock_page(page); |
1832a6d5 | 8999 | out: |
d0b7da88 | 9000 | btrfs_delalloc_release_space(inode, page_start, reserved_space); |
9998eb70 | 9001 | out_noreserve: |
b2b5ef5c | 9002 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
9003 | return ret; |
9004 | } | |
9005 | ||
a41ad394 | 9006 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
9007 | { |
9008 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 9009 | struct btrfs_block_rsv *rsv; |
a71754fc | 9010 | int ret = 0; |
3893e33b | 9011 | int err = 0; |
39279cc3 | 9012 | struct btrfs_trans_handle *trans; |
dbe674a9 | 9013 | u64 mask = root->sectorsize - 1; |
07127184 | 9014 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 9015 | |
0ef8b726 JB |
9016 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), |
9017 | (u64)-1); | |
9018 | if (ret) | |
9019 | return ret; | |
39279cc3 | 9020 | |
fcb80c2a JB |
9021 | /* |
9022 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
9023 | * 3 things going on here | |
9024 | * | |
9025 | * 1) We need to reserve space for our orphan item and the space to | |
9026 | * delete our orphan item. Lord knows we don't want to have a dangling | |
9027 | * orphan item because we didn't reserve space to remove it. | |
9028 | * | |
9029 | * 2) We need to reserve space to update our inode. | |
9030 | * | |
9031 | * 3) We need to have something to cache all the space that is going to | |
9032 | * be free'd up by the truncate operation, but also have some slack | |
9033 | * space reserved in case it uses space during the truncate (thank you | |
9034 | * very much snapshotting). | |
9035 | * | |
9036 | * And we need these to all be seperate. The fact is we can use alot of | |
9037 | * space doing the truncate, and we have no earthly idea how much space | |
9038 | * we will use, so we need the truncate reservation to be seperate so it | |
9039 | * doesn't end up using space reserved for updating the inode or | |
9040 | * removing the orphan item. We also need to be able to stop the | |
9041 | * transaction and start a new one, which means we need to be able to | |
9042 | * update the inode several times, and we have no idea of knowing how | |
9043 | * many times that will be, so we can't just reserve 1 item for the | |
9044 | * entirety of the opration, so that has to be done seperately as well. | |
9045 | * Then there is the orphan item, which does indeed need to be held on | |
9046 | * to for the whole operation, and we need nobody to touch this reserved | |
9047 | * space except the orphan code. | |
9048 | * | |
9049 | * So that leaves us with | |
9050 | * | |
9051 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
9052 | * 2) rsv - for the truncate reservation, which we will steal from the | |
9053 | * transaction reservation. | |
9054 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
9055 | * updating the inode. | |
9056 | */ | |
66d8f3dd | 9057 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
9058 | if (!rsv) |
9059 | return -ENOMEM; | |
4a338542 | 9060 | rsv->size = min_size; |
ca7e70f5 | 9061 | rsv->failfast = 1; |
f0cd846e | 9062 | |
907cbceb | 9063 | /* |
07127184 | 9064 | * 1 for the truncate slack space |
907cbceb JB |
9065 | * 1 for updating the inode. |
9066 | */ | |
f3fe820c | 9067 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
9068 | if (IS_ERR(trans)) { |
9069 | err = PTR_ERR(trans); | |
9070 | goto out; | |
9071 | } | |
f0cd846e | 9072 | |
907cbceb JB |
9073 | /* Migrate the slack space for the truncate to our reserve */ |
9074 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
9075 | min_size); | |
fcb80c2a | 9076 | BUG_ON(ret); |
f0cd846e | 9077 | |
5dc562c5 JB |
9078 | /* |
9079 | * So if we truncate and then write and fsync we normally would just | |
9080 | * write the extents that changed, which is a problem if we need to | |
9081 | * first truncate that entire inode. So set this flag so we write out | |
9082 | * all of the extents in the inode to the sync log so we're completely | |
9083 | * safe. | |
9084 | */ | |
9085 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 9086 | trans->block_rsv = rsv; |
907cbceb | 9087 | |
8082510e YZ |
9088 | while (1) { |
9089 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
9090 | inode->i_size, | |
9091 | BTRFS_EXTENT_DATA_KEY); | |
28ed1345 | 9092 | if (ret != -ENOSPC && ret != -EAGAIN) { |
3893e33b | 9093 | err = ret; |
8082510e | 9094 | break; |
3893e33b | 9095 | } |
39279cc3 | 9096 | |
fcb80c2a | 9097 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 9098 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
9099 | if (ret) { |
9100 | err = ret; | |
9101 | break; | |
9102 | } | |
ca7e70f5 | 9103 | |
8082510e | 9104 | btrfs_end_transaction(trans, root); |
b53d3f5d | 9105 | btrfs_btree_balance_dirty(root); |
ca7e70f5 JB |
9106 | |
9107 | trans = btrfs_start_transaction(root, 2); | |
9108 | if (IS_ERR(trans)) { | |
9109 | ret = err = PTR_ERR(trans); | |
9110 | trans = NULL; | |
9111 | break; | |
9112 | } | |
9113 | ||
9114 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
9115 | rsv, min_size); | |
9116 | BUG_ON(ret); /* shouldn't happen */ | |
9117 | trans->block_rsv = rsv; | |
8082510e YZ |
9118 | } |
9119 | ||
9120 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 9121 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 9122 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
9123 | if (ret) |
9124 | err = ret; | |
8082510e YZ |
9125 | } |
9126 | ||
917c16b2 CM |
9127 | if (trans) { |
9128 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
9129 | ret = btrfs_update_inode(trans, root, inode); | |
9130 | if (ret && !err) | |
9131 | err = ret; | |
7b128766 | 9132 | |
7ad85bb7 | 9133 | ret = btrfs_end_transaction(trans, root); |
b53d3f5d | 9134 | btrfs_btree_balance_dirty(root); |
917c16b2 | 9135 | } |
fcb80c2a JB |
9136 | |
9137 | out: | |
9138 | btrfs_free_block_rsv(root, rsv); | |
9139 | ||
3893e33b JB |
9140 | if (ret && !err) |
9141 | err = ret; | |
a41ad394 | 9142 | |
3893e33b | 9143 | return err; |
39279cc3 CM |
9144 | } |
9145 | ||
d352ac68 CM |
9146 | /* |
9147 | * create a new subvolume directory/inode (helper for the ioctl). | |
9148 | */ | |
d2fb3437 | 9149 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
9150 | struct btrfs_root *new_root, |
9151 | struct btrfs_root *parent_root, | |
9152 | u64 new_dirid) | |
39279cc3 | 9153 | { |
39279cc3 | 9154 | struct inode *inode; |
76dda93c | 9155 | int err; |
00e4e6b3 | 9156 | u64 index = 0; |
39279cc3 | 9157 | |
12fc9d09 FA |
9158 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
9159 | new_dirid, new_dirid, | |
9160 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
9161 | &index); | |
54aa1f4d | 9162 | if (IS_ERR(inode)) |
f46b5a66 | 9163 | return PTR_ERR(inode); |
39279cc3 CM |
9164 | inode->i_op = &btrfs_dir_inode_operations; |
9165 | inode->i_fop = &btrfs_dir_file_operations; | |
9166 | ||
bfe86848 | 9167 | set_nlink(inode, 1); |
dbe674a9 | 9168 | btrfs_i_size_write(inode, 0); |
b0d5d10f | 9169 | unlock_new_inode(inode); |
3b96362c | 9170 | |
63541927 FDBM |
9171 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
9172 | if (err) | |
9173 | btrfs_err(new_root->fs_info, | |
351fd353 | 9174 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
9175 | new_root->root_key.objectid, err); |
9176 | ||
76dda93c | 9177 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 9178 | |
76dda93c | 9179 | iput(inode); |
ce598979 | 9180 | return err; |
39279cc3 CM |
9181 | } |
9182 | ||
39279cc3 CM |
9183 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
9184 | { | |
9185 | struct btrfs_inode *ei; | |
2ead6ae7 | 9186 | struct inode *inode; |
39279cc3 CM |
9187 | |
9188 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
9189 | if (!ei) | |
9190 | return NULL; | |
2ead6ae7 YZ |
9191 | |
9192 | ei->root = NULL; | |
2ead6ae7 | 9193 | ei->generation = 0; |
15ee9bc7 | 9194 | ei->last_trans = 0; |
257c62e1 | 9195 | ei->last_sub_trans = 0; |
e02119d5 | 9196 | ei->logged_trans = 0; |
2ead6ae7 | 9197 | ei->delalloc_bytes = 0; |
47059d93 | 9198 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
9199 | ei->disk_i_size = 0; |
9200 | ei->flags = 0; | |
7709cde3 | 9201 | ei->csum_bytes = 0; |
2ead6ae7 | 9202 | ei->index_cnt = (u64)-1; |
67de1176 | 9203 | ei->dir_index = 0; |
2ead6ae7 | 9204 | ei->last_unlink_trans = 0; |
46d8bc34 | 9205 | ei->last_log_commit = 0; |
8089fe62 | 9206 | ei->delayed_iput_count = 0; |
2ead6ae7 | 9207 | |
9e0baf60 JB |
9208 | spin_lock_init(&ei->lock); |
9209 | ei->outstanding_extents = 0; | |
9210 | ei->reserved_extents = 0; | |
2ead6ae7 | 9211 | |
72ac3c0d | 9212 | ei->runtime_flags = 0; |
261507a0 | 9213 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 9214 | |
16cdcec7 MX |
9215 | ei->delayed_node = NULL; |
9216 | ||
9cc97d64 | 9217 | ei->i_otime.tv_sec = 0; |
9218 | ei->i_otime.tv_nsec = 0; | |
9219 | ||
2ead6ae7 | 9220 | inode = &ei->vfs_inode; |
a8067e02 | 9221 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
9222 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
9223 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
9224 | ei->io_tree.track_uptodate = 1; |
9225 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 9226 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 9227 | mutex_init(&ei->log_mutex); |
f248679e | 9228 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 9229 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 9230 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 9231 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 YZ |
9232 | RB_CLEAR_NODE(&ei->rb_node); |
9233 | ||
9234 | return inode; | |
39279cc3 CM |
9235 | } |
9236 | ||
aaedb55b JB |
9237 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
9238 | void btrfs_test_destroy_inode(struct inode *inode) | |
9239 | { | |
9240 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); | |
9241 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
9242 | } | |
9243 | #endif | |
9244 | ||
fa0d7e3d NP |
9245 | static void btrfs_i_callback(struct rcu_head *head) |
9246 | { | |
9247 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
9248 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9249 | } | |
9250 | ||
39279cc3 CM |
9251 | void btrfs_destroy_inode(struct inode *inode) |
9252 | { | |
e6dcd2dc | 9253 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
9254 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9255 | ||
b3d9b7a3 | 9256 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 9257 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
9258 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
9259 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
9260 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
9261 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
47059d93 | 9262 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 9263 | |
a6dbd429 JB |
9264 | /* |
9265 | * This can happen where we create an inode, but somebody else also | |
9266 | * created the same inode and we need to destroy the one we already | |
9267 | * created. | |
9268 | */ | |
9269 | if (!root) | |
9270 | goto free; | |
9271 | ||
8a35d95f JB |
9272 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
9273 | &BTRFS_I(inode)->runtime_flags)) { | |
c2cf52eb | 9274 | btrfs_info(root->fs_info, "inode %llu still on the orphan list", |
c1c9ff7c | 9275 | btrfs_ino(inode)); |
8a35d95f | 9276 | atomic_dec(&root->orphan_inodes); |
7b128766 | 9277 | } |
7b128766 | 9278 | |
d397712b | 9279 | while (1) { |
e6dcd2dc CM |
9280 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
9281 | if (!ordered) | |
9282 | break; | |
9283 | else { | |
c2cf52eb | 9284 | btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup", |
c1c9ff7c | 9285 | ordered->file_offset, ordered->len); |
e6dcd2dc CM |
9286 | btrfs_remove_ordered_extent(inode, ordered); |
9287 | btrfs_put_ordered_extent(ordered); | |
9288 | btrfs_put_ordered_extent(ordered); | |
9289 | } | |
9290 | } | |
56fa9d07 | 9291 | btrfs_qgroup_check_reserved_leak(inode); |
5d4f98a2 | 9292 | inode_tree_del(inode); |
5b21f2ed | 9293 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 9294 | free: |
fa0d7e3d | 9295 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
9296 | } |
9297 | ||
45321ac5 | 9298 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
9299 | { |
9300 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 9301 | |
6379ef9f NA |
9302 | if (root == NULL) |
9303 | return 1; | |
9304 | ||
fa6ac876 | 9305 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 9306 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 9307 | return 1; |
76dda93c | 9308 | else |
45321ac5 | 9309 | return generic_drop_inode(inode); |
76dda93c YZ |
9310 | } |
9311 | ||
0ee0fda0 | 9312 | static void init_once(void *foo) |
39279cc3 CM |
9313 | { |
9314 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
9315 | ||
9316 | inode_init_once(&ei->vfs_inode); | |
9317 | } | |
9318 | ||
9319 | void btrfs_destroy_cachep(void) | |
9320 | { | |
8c0a8537 KS |
9321 | /* |
9322 | * Make sure all delayed rcu free inodes are flushed before we | |
9323 | * destroy cache. | |
9324 | */ | |
9325 | rcu_barrier(); | |
5598e900 KM |
9326 | kmem_cache_destroy(btrfs_inode_cachep); |
9327 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
9328 | kmem_cache_destroy(btrfs_transaction_cachep); | |
9329 | kmem_cache_destroy(btrfs_path_cachep); | |
9330 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
9331 | } |
9332 | ||
9333 | int btrfs_init_cachep(void) | |
9334 | { | |
837e1972 | 9335 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 9336 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
9337 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
9338 | init_once); | |
39279cc3 CM |
9339 | if (!btrfs_inode_cachep) |
9340 | goto fail; | |
9601e3f6 | 9341 | |
837e1972 | 9342 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 CH |
9343 | sizeof(struct btrfs_trans_handle), 0, |
9344 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
9345 | if (!btrfs_trans_handle_cachep) |
9346 | goto fail; | |
9601e3f6 | 9347 | |
837e1972 | 9348 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 CH |
9349 | sizeof(struct btrfs_transaction), 0, |
9350 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
9351 | if (!btrfs_transaction_cachep) |
9352 | goto fail; | |
9601e3f6 | 9353 | |
837e1972 | 9354 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 CH |
9355 | sizeof(struct btrfs_path), 0, |
9356 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
9357 | if (!btrfs_path_cachep) |
9358 | goto fail; | |
9601e3f6 | 9359 | |
837e1972 | 9360 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 JB |
9361 | sizeof(struct btrfs_free_space), 0, |
9362 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
9363 | if (!btrfs_free_space_cachep) | |
9364 | goto fail; | |
9365 | ||
39279cc3 CM |
9366 | return 0; |
9367 | fail: | |
9368 | btrfs_destroy_cachep(); | |
9369 | return -ENOMEM; | |
9370 | } | |
9371 | ||
9372 | static int btrfs_getattr(struct vfsmount *mnt, | |
9373 | struct dentry *dentry, struct kstat *stat) | |
9374 | { | |
df0af1a5 | 9375 | u64 delalloc_bytes; |
2b0143b5 | 9376 | struct inode *inode = d_inode(dentry); |
fadc0d8b DS |
9377 | u32 blocksize = inode->i_sb->s_blocksize; |
9378 | ||
39279cc3 | 9379 | generic_fillattr(inode, stat); |
0ee5dc67 | 9380 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
9381 | |
9382 | spin_lock(&BTRFS_I(inode)->lock); | |
9383 | delalloc_bytes = BTRFS_I(inode)->delalloc_bytes; | |
9384 | spin_unlock(&BTRFS_I(inode)->lock); | |
fadc0d8b | 9385 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 9386 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
9387 | return 0; |
9388 | } | |
9389 | ||
d397712b CM |
9390 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
9391 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
9392 | { |
9393 | struct btrfs_trans_handle *trans; | |
9394 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 9395 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9396 | struct inode *new_inode = d_inode(new_dentry); |
9397 | struct inode *old_inode = d_inode(old_dentry); | |
00e4e6b3 | 9398 | u64 index = 0; |
4df27c4d | 9399 | u64 root_objectid; |
39279cc3 | 9400 | int ret; |
33345d01 | 9401 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 9402 | |
33345d01 | 9403 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9404 | return -EPERM; |
9405 | ||
4df27c4d | 9406 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9407 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9408 | return -EXDEV; |
9409 | ||
33345d01 LZ |
9410 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
9411 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 9412 | return -ENOTEMPTY; |
5f39d397 | 9413 | |
4df27c4d YZ |
9414 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9415 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9416 | return -ENOTEMPTY; | |
9c52057c CM |
9417 | |
9418 | ||
9419 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9420 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9421 | new_dentry->d_name.name, |
9422 | new_dentry->d_name.len); | |
9423 | ||
9424 | if (ret) { | |
9425 | if (ret == -EEXIST) { | |
9426 | /* we shouldn't get | |
9427 | * eexist without a new_inode */ | |
fae7f21c | 9428 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9429 | return ret; |
9430 | } | |
9431 | } else { | |
9432 | /* maybe -EOVERFLOW */ | |
9433 | return ret; | |
9434 | } | |
9435 | } | |
9436 | ret = 0; | |
9437 | ||
5a3f23d5 | 9438 | /* |
8d875f95 CM |
9439 | * we're using rename to replace one file with another. Start IO on it |
9440 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9441 | */ |
8d875f95 | 9442 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9443 | filemap_flush(old_inode->i_mapping); |
9444 | ||
76dda93c | 9445 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 9446 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 9447 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
9448 | /* |
9449 | * We want to reserve the absolute worst case amount of items. So if | |
9450 | * both inodes are subvols and we need to unlink them then that would | |
9451 | * require 4 item modifications, but if they are both normal inodes it | |
9452 | * would require 5 item modifications, so we'll assume their normal | |
9453 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
9454 | * should cover the worst case number of items we'll modify. | |
9455 | */ | |
6e137ed3 | 9456 | trans = btrfs_start_transaction(root, 11); |
b44c59a8 JL |
9457 | if (IS_ERR(trans)) { |
9458 | ret = PTR_ERR(trans); | |
9459 | goto out_notrans; | |
9460 | } | |
76dda93c | 9461 | |
4df27c4d YZ |
9462 | if (dest != root) |
9463 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 9464 | |
a5719521 YZ |
9465 | ret = btrfs_set_inode_index(new_dir, &index); |
9466 | if (ret) | |
9467 | goto out_fail; | |
5a3f23d5 | 9468 | |
67de1176 | 9469 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9470 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9471 | /* force full log commit if subvolume involved. */ |
995946dd | 9472 | btrfs_set_log_full_commit(root->fs_info, trans); |
4df27c4d | 9473 | } else { |
a5719521 YZ |
9474 | ret = btrfs_insert_inode_ref(trans, dest, |
9475 | new_dentry->d_name.name, | |
9476 | new_dentry->d_name.len, | |
33345d01 LZ |
9477 | old_ino, |
9478 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
9479 | if (ret) |
9480 | goto out_fail; | |
4df27c4d YZ |
9481 | /* |
9482 | * this is an ugly little race, but the rename is required | |
9483 | * to make sure that if we crash, the inode is either at the | |
9484 | * old name or the new one. pinning the log transaction lets | |
9485 | * us make sure we don't allow a log commit to come in after | |
9486 | * we unlink the name but before we add the new name back in. | |
9487 | */ | |
9488 | btrfs_pin_log_trans(root); | |
9489 | } | |
5a3f23d5 | 9490 | |
0c4d2d95 JB |
9491 | inode_inc_iversion(old_dir); |
9492 | inode_inc_iversion(new_dir); | |
9493 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
9494 | old_dir->i_ctime = old_dir->i_mtime = |
9495 | new_dir->i_ctime = new_dir->i_mtime = | |
9496 | old_inode->i_ctime = current_fs_time(old_dir->i_sb); | |
5f39d397 | 9497 | |
12fcfd22 CM |
9498 | if (old_dentry->d_parent != new_dentry->d_parent) |
9499 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
9500 | ||
33345d01 | 9501 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
9502 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
9503 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
9504 | old_dentry->d_name.name, | |
9505 | old_dentry->d_name.len); | |
9506 | } else { | |
92986796 | 9507 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
2b0143b5 | 9508 | d_inode(old_dentry), |
92986796 AV |
9509 | old_dentry->d_name.name, |
9510 | old_dentry->d_name.len); | |
9511 | if (!ret) | |
9512 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 9513 | } |
79787eaa JM |
9514 | if (ret) { |
9515 | btrfs_abort_transaction(trans, root, ret); | |
9516 | goto out_fail; | |
9517 | } | |
39279cc3 CM |
9518 | |
9519 | if (new_inode) { | |
0c4d2d95 | 9520 | inode_inc_iversion(new_inode); |
04b285f3 | 9521 | new_inode->i_ctime = current_fs_time(new_inode->i_sb); |
33345d01 | 9522 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
9523 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
9524 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
9525 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
9526 | root_objectid, | |
9527 | new_dentry->d_name.name, | |
9528 | new_dentry->d_name.len); | |
9529 | BUG_ON(new_inode->i_nlink == 0); | |
9530 | } else { | |
9531 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
2b0143b5 | 9532 | d_inode(new_dentry), |
4df27c4d YZ |
9533 | new_dentry->d_name.name, |
9534 | new_dentry->d_name.len); | |
9535 | } | |
4ef31a45 | 9536 | if (!ret && new_inode->i_nlink == 0) |
2b0143b5 | 9537 | ret = btrfs_orphan_add(trans, d_inode(new_dentry)); |
79787eaa JM |
9538 | if (ret) { |
9539 | btrfs_abort_transaction(trans, root, ret); | |
9540 | goto out_fail; | |
9541 | } | |
39279cc3 | 9542 | } |
aec7477b | 9543 | |
4df27c4d YZ |
9544 | ret = btrfs_add_link(trans, new_dir, old_inode, |
9545 | new_dentry->d_name.name, | |
a5719521 | 9546 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
9547 | if (ret) { |
9548 | btrfs_abort_transaction(trans, root, ret); | |
9549 | goto out_fail; | |
9550 | } | |
39279cc3 | 9551 | |
67de1176 MX |
9552 | if (old_inode->i_nlink == 1) |
9553 | BTRFS_I(old_inode)->dir_index = index; | |
9554 | ||
33345d01 | 9555 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 9556 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 9557 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
9558 | btrfs_end_log_trans(root); |
9559 | } | |
39279cc3 | 9560 | out_fail: |
7ad85bb7 | 9561 | btrfs_end_transaction(trans, root); |
b44c59a8 | 9562 | out_notrans: |
33345d01 | 9563 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 9564 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 9565 | |
39279cc3 CM |
9566 | return ret; |
9567 | } | |
9568 | ||
80ace85c MS |
9569 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
9570 | struct inode *new_dir, struct dentry *new_dentry, | |
9571 | unsigned int flags) | |
9572 | { | |
9573 | if (flags & ~RENAME_NOREPLACE) | |
9574 | return -EINVAL; | |
9575 | ||
9576 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry); | |
9577 | } | |
9578 | ||
8ccf6f19 MX |
9579 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
9580 | { | |
9581 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 9582 | struct inode *inode; |
8ccf6f19 MX |
9583 | |
9584 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
9585 | work); | |
9f23e289 | 9586 | inode = delalloc_work->inode; |
30424601 DS |
9587 | filemap_flush(inode->i_mapping); |
9588 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9589 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 9590 | filemap_flush(inode->i_mapping); |
8ccf6f19 MX |
9591 | |
9592 | if (delalloc_work->delay_iput) | |
9f23e289 | 9593 | btrfs_add_delayed_iput(inode); |
8ccf6f19 | 9594 | else |
9f23e289 | 9595 | iput(inode); |
8ccf6f19 MX |
9596 | complete(&delalloc_work->completion); |
9597 | } | |
9598 | ||
9599 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
651d494a | 9600 | int delay_iput) |
8ccf6f19 MX |
9601 | { |
9602 | struct btrfs_delalloc_work *work; | |
9603 | ||
100d5702 | 9604 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
9605 | if (!work) |
9606 | return NULL; | |
9607 | ||
9608 | init_completion(&work->completion); | |
9609 | INIT_LIST_HEAD(&work->list); | |
9610 | work->inode = inode; | |
8ccf6f19 | 9611 | work->delay_iput = delay_iput; |
9e0af237 LB |
9612 | WARN_ON_ONCE(!inode); |
9613 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, | |
9614 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
9615 | |
9616 | return work; | |
9617 | } | |
9618 | ||
9619 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
9620 | { | |
9621 | wait_for_completion(&work->completion); | |
100d5702 | 9622 | kfree(work); |
8ccf6f19 MX |
9623 | } |
9624 | ||
d352ac68 CM |
9625 | /* |
9626 | * some fairly slow code that needs optimization. This walks the list | |
9627 | * of all the inodes with pending delalloc and forces them to disk. | |
9628 | */ | |
6c255e67 MX |
9629 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput, |
9630 | int nr) | |
ea8c2819 | 9631 | { |
ea8c2819 | 9632 | struct btrfs_inode *binode; |
5b21f2ed | 9633 | struct inode *inode; |
8ccf6f19 MX |
9634 | struct btrfs_delalloc_work *work, *next; |
9635 | struct list_head works; | |
1eafa6c7 | 9636 | struct list_head splice; |
8ccf6f19 | 9637 | int ret = 0; |
ea8c2819 | 9638 | |
8ccf6f19 | 9639 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 9640 | INIT_LIST_HEAD(&splice); |
63607cc8 | 9641 | |
573bfb72 | 9642 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
9643 | spin_lock(&root->delalloc_lock); |
9644 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
9645 | while (!list_empty(&splice)) { |
9646 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 9647 | delalloc_inodes); |
1eafa6c7 | 9648 | |
eb73c1b7 MX |
9649 | list_move_tail(&binode->delalloc_inodes, |
9650 | &root->delalloc_inodes); | |
5b21f2ed | 9651 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 9652 | if (!inode) { |
eb73c1b7 | 9653 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 9654 | continue; |
df0af1a5 | 9655 | } |
eb73c1b7 | 9656 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 9657 | |
651d494a | 9658 | work = btrfs_alloc_delalloc_work(inode, delay_iput); |
5d99a998 | 9659 | if (!work) { |
f4ab9ea7 JB |
9660 | if (delay_iput) |
9661 | btrfs_add_delayed_iput(inode); | |
9662 | else | |
9663 | iput(inode); | |
1eafa6c7 | 9664 | ret = -ENOMEM; |
a1ecaabb | 9665 | goto out; |
5b21f2ed | 9666 | } |
1eafa6c7 | 9667 | list_add_tail(&work->list, &works); |
a44903ab QW |
9668 | btrfs_queue_work(root->fs_info->flush_workers, |
9669 | &work->work); | |
6c255e67 MX |
9670 | ret++; |
9671 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 9672 | goto out; |
5b21f2ed | 9673 | cond_resched(); |
eb73c1b7 | 9674 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 9675 | } |
eb73c1b7 | 9676 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 9677 | |
a1ecaabb | 9678 | out: |
eb73c1b7 MX |
9679 | list_for_each_entry_safe(work, next, &works, list) { |
9680 | list_del_init(&work->list); | |
9681 | btrfs_wait_and_free_delalloc_work(work); | |
9682 | } | |
9683 | ||
9684 | if (!list_empty_careful(&splice)) { | |
9685 | spin_lock(&root->delalloc_lock); | |
9686 | list_splice_tail(&splice, &root->delalloc_inodes); | |
9687 | spin_unlock(&root->delalloc_lock); | |
9688 | } | |
573bfb72 | 9689 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
9690 | return ret; |
9691 | } | |
1eafa6c7 | 9692 | |
eb73c1b7 MX |
9693 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
9694 | { | |
9695 | int ret; | |
1eafa6c7 | 9696 | |
2c21b4d7 | 9697 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
eb73c1b7 MX |
9698 | return -EROFS; |
9699 | ||
6c255e67 MX |
9700 | ret = __start_delalloc_inodes(root, delay_iput, -1); |
9701 | if (ret > 0) | |
9702 | ret = 0; | |
eb73c1b7 MX |
9703 | /* |
9704 | * the filemap_flush will queue IO into the worker threads, but | |
8c8bee1d CM |
9705 | * we have to make sure the IO is actually started and that |
9706 | * ordered extents get created before we return | |
9707 | */ | |
9708 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 9709 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 9710 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 9711 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
9712 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
9713 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
9714 | } |
9715 | atomic_dec(&root->fs_info->async_submit_draining); | |
eb73c1b7 MX |
9716 | return ret; |
9717 | } | |
9718 | ||
6c255e67 MX |
9719 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput, |
9720 | int nr) | |
eb73c1b7 MX |
9721 | { |
9722 | struct btrfs_root *root; | |
9723 | struct list_head splice; | |
9724 | int ret; | |
9725 | ||
2c21b4d7 | 9726 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
9727 | return -EROFS; |
9728 | ||
9729 | INIT_LIST_HEAD(&splice); | |
9730 | ||
573bfb72 | 9731 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
9732 | spin_lock(&fs_info->delalloc_root_lock); |
9733 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 9734 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
9735 | root = list_first_entry(&splice, struct btrfs_root, |
9736 | delalloc_root); | |
9737 | root = btrfs_grab_fs_root(root); | |
9738 | BUG_ON(!root); | |
9739 | list_move_tail(&root->delalloc_root, | |
9740 | &fs_info->delalloc_roots); | |
9741 | spin_unlock(&fs_info->delalloc_root_lock); | |
9742 | ||
6c255e67 | 9743 | ret = __start_delalloc_inodes(root, delay_iput, nr); |
eb73c1b7 | 9744 | btrfs_put_fs_root(root); |
6c255e67 | 9745 | if (ret < 0) |
eb73c1b7 MX |
9746 | goto out; |
9747 | ||
6c255e67 MX |
9748 | if (nr != -1) { |
9749 | nr -= ret; | |
9750 | WARN_ON(nr < 0); | |
9751 | } | |
eb73c1b7 | 9752 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 9753 | } |
eb73c1b7 | 9754 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 9755 | |
6c255e67 | 9756 | ret = 0; |
eb73c1b7 MX |
9757 | atomic_inc(&fs_info->async_submit_draining); |
9758 | while (atomic_read(&fs_info->nr_async_submits) || | |
9759 | atomic_read(&fs_info->async_delalloc_pages)) { | |
9760 | wait_event(fs_info->async_submit_wait, | |
9761 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
9762 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
9763 | } | |
9764 | atomic_dec(&fs_info->async_submit_draining); | |
eb73c1b7 | 9765 | out: |
1eafa6c7 | 9766 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
9767 | spin_lock(&fs_info->delalloc_root_lock); |
9768 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
9769 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 9770 | } |
573bfb72 | 9771 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 9772 | return ret; |
ea8c2819 CM |
9773 | } |
9774 | ||
39279cc3 CM |
9775 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
9776 | const char *symname) | |
9777 | { | |
9778 | struct btrfs_trans_handle *trans; | |
9779 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9780 | struct btrfs_path *path; | |
9781 | struct btrfs_key key; | |
1832a6d5 | 9782 | struct inode *inode = NULL; |
39279cc3 CM |
9783 | int err; |
9784 | int drop_inode = 0; | |
9785 | u64 objectid; | |
67871254 | 9786 | u64 index = 0; |
39279cc3 CM |
9787 | int name_len; |
9788 | int datasize; | |
5f39d397 | 9789 | unsigned long ptr; |
39279cc3 | 9790 | struct btrfs_file_extent_item *ei; |
5f39d397 | 9791 | struct extent_buffer *leaf; |
39279cc3 | 9792 | |
f06becc4 | 9793 | name_len = strlen(symname); |
39279cc3 CM |
9794 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) |
9795 | return -ENAMETOOLONG; | |
1832a6d5 | 9796 | |
9ed74f2d JB |
9797 | /* |
9798 | * 2 items for inode item and ref | |
9799 | * 2 items for dir items | |
9269d12b FM |
9800 | * 1 item for updating parent inode item |
9801 | * 1 item for the inline extent item | |
9ed74f2d JB |
9802 | * 1 item for xattr if selinux is on |
9803 | */ | |
9269d12b | 9804 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
9805 | if (IS_ERR(trans)) |
9806 | return PTR_ERR(trans); | |
1832a6d5 | 9807 | |
581bb050 LZ |
9808 | err = btrfs_find_free_ino(root, &objectid); |
9809 | if (err) | |
9810 | goto out_unlock; | |
9811 | ||
aec7477b | 9812 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 9813 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 9814 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
9815 | if (IS_ERR(inode)) { |
9816 | err = PTR_ERR(inode); | |
39279cc3 | 9817 | goto out_unlock; |
7cf96da3 | 9818 | } |
39279cc3 | 9819 | |
ad19db71 CS |
9820 | /* |
9821 | * If the active LSM wants to access the inode during | |
9822 | * d_instantiate it needs these. Smack checks to see | |
9823 | * if the filesystem supports xattrs by looking at the | |
9824 | * ops vector. | |
9825 | */ | |
9826 | inode->i_fop = &btrfs_file_operations; | |
9827 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 9828 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
9829 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
9830 | ||
9831 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
9832 | if (err) | |
9833 | goto out_unlock_inode; | |
ad19db71 | 9834 | |
39279cc3 | 9835 | path = btrfs_alloc_path(); |
d8926bb3 MF |
9836 | if (!path) { |
9837 | err = -ENOMEM; | |
b0d5d10f | 9838 | goto out_unlock_inode; |
d8926bb3 | 9839 | } |
33345d01 | 9840 | key.objectid = btrfs_ino(inode); |
39279cc3 | 9841 | key.offset = 0; |
962a298f | 9842 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
9843 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
9844 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
9845 | datasize); | |
54aa1f4d | 9846 | if (err) { |
b0839166 | 9847 | btrfs_free_path(path); |
b0d5d10f | 9848 | goto out_unlock_inode; |
54aa1f4d | 9849 | } |
5f39d397 CM |
9850 | leaf = path->nodes[0]; |
9851 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
9852 | struct btrfs_file_extent_item); | |
9853 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
9854 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 9855 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
9856 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
9857 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
9858 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
9859 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
9860 | ||
39279cc3 | 9861 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
9862 | write_extent_buffer(leaf, symname, ptr, name_len); |
9863 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 9864 | btrfs_free_path(path); |
5f39d397 | 9865 | |
39279cc3 | 9866 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 9867 | inode_nohighmem(inode); |
39279cc3 | 9868 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
d899e052 | 9869 | inode_set_bytes(inode, name_len); |
f06becc4 | 9870 | btrfs_i_size_write(inode, name_len); |
54aa1f4d | 9871 | err = btrfs_update_inode(trans, root, inode); |
d50866d0 FM |
9872 | /* |
9873 | * Last step, add directory indexes for our symlink inode. This is the | |
9874 | * last step to avoid extra cleanup of these indexes if an error happens | |
9875 | * elsewhere above. | |
9876 | */ | |
9877 | if (!err) | |
9878 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); | |
b0d5d10f | 9879 | if (err) { |
54aa1f4d | 9880 | drop_inode = 1; |
b0d5d10f CM |
9881 | goto out_unlock_inode; |
9882 | } | |
9883 | ||
9884 | unlock_new_inode(inode); | |
9885 | d_instantiate(dentry, inode); | |
39279cc3 CM |
9886 | |
9887 | out_unlock: | |
7ad85bb7 | 9888 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
9889 | if (drop_inode) { |
9890 | inode_dec_link_count(inode); | |
9891 | iput(inode); | |
9892 | } | |
b53d3f5d | 9893 | btrfs_btree_balance_dirty(root); |
39279cc3 | 9894 | return err; |
b0d5d10f CM |
9895 | |
9896 | out_unlock_inode: | |
9897 | drop_inode = 1; | |
9898 | unlock_new_inode(inode); | |
9899 | goto out_unlock; | |
39279cc3 | 9900 | } |
16432985 | 9901 | |
0af3d00b JB |
9902 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
9903 | u64 start, u64 num_bytes, u64 min_size, | |
9904 | loff_t actual_len, u64 *alloc_hint, | |
9905 | struct btrfs_trans_handle *trans) | |
d899e052 | 9906 | { |
5dc562c5 JB |
9907 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
9908 | struct extent_map *em; | |
d899e052 YZ |
9909 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9910 | struct btrfs_key ins; | |
d899e052 | 9911 | u64 cur_offset = start; |
55a61d1d | 9912 | u64 i_size; |
154ea289 | 9913 | u64 cur_bytes; |
0b670dc4 | 9914 | u64 last_alloc = (u64)-1; |
d899e052 | 9915 | int ret = 0; |
0af3d00b | 9916 | bool own_trans = true; |
d899e052 | 9917 | |
0af3d00b JB |
9918 | if (trans) |
9919 | own_trans = false; | |
d899e052 | 9920 | while (num_bytes > 0) { |
0af3d00b JB |
9921 | if (own_trans) { |
9922 | trans = btrfs_start_transaction(root, 3); | |
9923 | if (IS_ERR(trans)) { | |
9924 | ret = PTR_ERR(trans); | |
9925 | break; | |
9926 | } | |
5a303d5d YZ |
9927 | } |
9928 | ||
ee22184b | 9929 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 9930 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
9931 | /* |
9932 | * If we are severely fragmented we could end up with really | |
9933 | * small allocations, so if the allocator is returning small | |
9934 | * chunks lets make its job easier by only searching for those | |
9935 | * sized chunks. | |
9936 | */ | |
9937 | cur_bytes = min(cur_bytes, last_alloc); | |
00361589 | 9938 | ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0, |
e570fd27 | 9939 | *alloc_hint, &ins, 1, 0); |
5a303d5d | 9940 | if (ret) { |
0af3d00b JB |
9941 | if (own_trans) |
9942 | btrfs_end_transaction(trans, root); | |
a22285a6 | 9943 | break; |
d899e052 | 9944 | } |
5a303d5d | 9945 | |
0b670dc4 | 9946 | last_alloc = ins.offset; |
d899e052 YZ |
9947 | ret = insert_reserved_file_extent(trans, inode, |
9948 | cur_offset, ins.objectid, | |
9949 | ins.offset, ins.offset, | |
920bbbfb | 9950 | ins.offset, 0, 0, 0, |
d899e052 | 9951 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 9952 | if (ret) { |
857cc2fc | 9953 | btrfs_free_reserved_extent(root, ins.objectid, |
e570fd27 | 9954 | ins.offset, 0); |
79787eaa JM |
9955 | btrfs_abort_transaction(trans, root, ret); |
9956 | if (own_trans) | |
9957 | btrfs_end_transaction(trans, root); | |
9958 | break; | |
9959 | } | |
31193213 | 9960 | |
a1ed835e CM |
9961 | btrfs_drop_extent_cache(inode, cur_offset, |
9962 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 9963 | |
5dc562c5 JB |
9964 | em = alloc_extent_map(); |
9965 | if (!em) { | |
9966 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
9967 | &BTRFS_I(inode)->runtime_flags); | |
9968 | goto next; | |
9969 | } | |
9970 | ||
9971 | em->start = cur_offset; | |
9972 | em->orig_start = cur_offset; | |
9973 | em->len = ins.offset; | |
9974 | em->block_start = ins.objectid; | |
9975 | em->block_len = ins.offset; | |
b4939680 | 9976 | em->orig_block_len = ins.offset; |
cc95bef6 | 9977 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
9978 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
9979 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
9980 | em->generation = trans->transid; | |
9981 | ||
9982 | while (1) { | |
9983 | write_lock(&em_tree->lock); | |
09a2a8f9 | 9984 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
9985 | write_unlock(&em_tree->lock); |
9986 | if (ret != -EEXIST) | |
9987 | break; | |
9988 | btrfs_drop_extent_cache(inode, cur_offset, | |
9989 | cur_offset + ins.offset - 1, | |
9990 | 0); | |
9991 | } | |
9992 | free_extent_map(em); | |
9993 | next: | |
d899e052 YZ |
9994 | num_bytes -= ins.offset; |
9995 | cur_offset += ins.offset; | |
efa56464 | 9996 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 9997 | |
0c4d2d95 | 9998 | inode_inc_iversion(inode); |
04b285f3 | 9999 | inode->i_ctime = current_fs_time(inode->i_sb); |
6cbff00f | 10000 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 10001 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
10002 | (actual_len > inode->i_size) && |
10003 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 10004 | if (cur_offset > actual_len) |
55a61d1d | 10005 | i_size = actual_len; |
d1ea6a61 | 10006 | else |
55a61d1d JB |
10007 | i_size = cur_offset; |
10008 | i_size_write(inode, i_size); | |
10009 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
10010 | } |
10011 | ||
d899e052 | 10012 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
10013 | |
10014 | if (ret) { | |
10015 | btrfs_abort_transaction(trans, root, ret); | |
10016 | if (own_trans) | |
10017 | btrfs_end_transaction(trans, root); | |
10018 | break; | |
10019 | } | |
d899e052 | 10020 | |
0af3d00b JB |
10021 | if (own_trans) |
10022 | btrfs_end_transaction(trans, root); | |
5a303d5d | 10023 | } |
d899e052 YZ |
10024 | return ret; |
10025 | } | |
10026 | ||
0af3d00b JB |
10027 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
10028 | u64 start, u64 num_bytes, u64 min_size, | |
10029 | loff_t actual_len, u64 *alloc_hint) | |
10030 | { | |
10031 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10032 | min_size, actual_len, alloc_hint, | |
10033 | NULL); | |
10034 | } | |
10035 | ||
10036 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
10037 | struct btrfs_trans_handle *trans, int mode, | |
10038 | u64 start, u64 num_bytes, u64 min_size, | |
10039 | loff_t actual_len, u64 *alloc_hint) | |
10040 | { | |
10041 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10042 | min_size, actual_len, alloc_hint, trans); | |
10043 | } | |
10044 | ||
e6dcd2dc CM |
10045 | static int btrfs_set_page_dirty(struct page *page) |
10046 | { | |
e6dcd2dc CM |
10047 | return __set_page_dirty_nobuffers(page); |
10048 | } | |
10049 | ||
10556cb2 | 10050 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 10051 | { |
b83cc969 | 10052 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 10053 | umode_t mode = inode->i_mode; |
b83cc969 | 10054 | |
cb6db4e5 JM |
10055 | if (mask & MAY_WRITE && |
10056 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
10057 | if (btrfs_root_readonly(root)) | |
10058 | return -EROFS; | |
10059 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
10060 | return -EACCES; | |
10061 | } | |
2830ba7f | 10062 | return generic_permission(inode, mask); |
fdebe2bd | 10063 | } |
39279cc3 | 10064 | |
ef3b9af5 FM |
10065 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
10066 | { | |
10067 | struct btrfs_trans_handle *trans; | |
10068 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10069 | struct inode *inode = NULL; | |
10070 | u64 objectid; | |
10071 | u64 index; | |
10072 | int ret = 0; | |
10073 | ||
10074 | /* | |
10075 | * 5 units required for adding orphan entry | |
10076 | */ | |
10077 | trans = btrfs_start_transaction(root, 5); | |
10078 | if (IS_ERR(trans)) | |
10079 | return PTR_ERR(trans); | |
10080 | ||
10081 | ret = btrfs_find_free_ino(root, &objectid); | |
10082 | if (ret) | |
10083 | goto out; | |
10084 | ||
10085 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
10086 | btrfs_ino(dir), objectid, mode, &index); | |
10087 | if (IS_ERR(inode)) { | |
10088 | ret = PTR_ERR(inode); | |
10089 | inode = NULL; | |
10090 | goto out; | |
10091 | } | |
10092 | ||
ef3b9af5 FM |
10093 | inode->i_fop = &btrfs_file_operations; |
10094 | inode->i_op = &btrfs_file_inode_operations; | |
10095 | ||
10096 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 FM |
10097 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10098 | ||
b0d5d10f CM |
10099 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
10100 | if (ret) | |
10101 | goto out_inode; | |
10102 | ||
10103 | ret = btrfs_update_inode(trans, root, inode); | |
10104 | if (ret) | |
10105 | goto out_inode; | |
ef3b9af5 FM |
10106 | ret = btrfs_orphan_add(trans, inode); |
10107 | if (ret) | |
b0d5d10f | 10108 | goto out_inode; |
ef3b9af5 | 10109 | |
5762b5c9 FM |
10110 | /* |
10111 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
10112 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
10113 | * through: | |
10114 | * | |
10115 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
10116 | */ | |
10117 | set_nlink(inode, 1); | |
b0d5d10f | 10118 | unlock_new_inode(inode); |
ef3b9af5 FM |
10119 | d_tmpfile(dentry, inode); |
10120 | mark_inode_dirty(inode); | |
10121 | ||
10122 | out: | |
10123 | btrfs_end_transaction(trans, root); | |
10124 | if (ret) | |
10125 | iput(inode); | |
10126 | btrfs_balance_delayed_items(root); | |
10127 | btrfs_btree_balance_dirty(root); | |
ef3b9af5 | 10128 | return ret; |
b0d5d10f CM |
10129 | |
10130 | out_inode: | |
10131 | unlock_new_inode(inode); | |
10132 | goto out; | |
10133 | ||
ef3b9af5 FM |
10134 | } |
10135 | ||
b38ef71c FM |
10136 | /* Inspired by filemap_check_errors() */ |
10137 | int btrfs_inode_check_errors(struct inode *inode) | |
10138 | { | |
10139 | int ret = 0; | |
10140 | ||
10141 | if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) && | |
10142 | test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags)) | |
10143 | ret = -ENOSPC; | |
10144 | if (test_bit(AS_EIO, &inode->i_mapping->flags) && | |
10145 | test_and_clear_bit(AS_EIO, &inode->i_mapping->flags)) | |
10146 | ret = -EIO; | |
10147 | ||
10148 | return ret; | |
10149 | } | |
10150 | ||
6e1d5dcc | 10151 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10152 | .getattr = btrfs_getattr, |
39279cc3 CM |
10153 | .lookup = btrfs_lookup, |
10154 | .create = btrfs_create, | |
10155 | .unlink = btrfs_unlink, | |
10156 | .link = btrfs_link, | |
10157 | .mkdir = btrfs_mkdir, | |
10158 | .rmdir = btrfs_rmdir, | |
80ace85c | 10159 | .rename2 = btrfs_rename2, |
39279cc3 CM |
10160 | .symlink = btrfs_symlink, |
10161 | .setattr = btrfs_setattr, | |
618e21d5 | 10162 | .mknod = btrfs_mknod, |
e0d46f5c | 10163 | .setxattr = generic_setxattr, |
9172abbc | 10164 | .getxattr = generic_getxattr, |
5103e947 | 10165 | .listxattr = btrfs_listxattr, |
e0d46f5c | 10166 | .removexattr = generic_removexattr, |
fdebe2bd | 10167 | .permission = btrfs_permission, |
4e34e719 | 10168 | .get_acl = btrfs_get_acl, |
996a710d | 10169 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10170 | .update_time = btrfs_update_time, |
ef3b9af5 | 10171 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 10172 | }; |
6e1d5dcc | 10173 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 10174 | .lookup = btrfs_lookup, |
fdebe2bd | 10175 | .permission = btrfs_permission, |
4e34e719 | 10176 | .get_acl = btrfs_get_acl, |
996a710d | 10177 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10178 | .update_time = btrfs_update_time, |
39279cc3 | 10179 | }; |
76dda93c | 10180 | |
828c0950 | 10181 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10182 | .llseek = generic_file_llseek, |
10183 | .read = generic_read_dir, | |
fe742fd4 | 10184 | .iterate = btrfs_real_readdir, |
34287aa3 | 10185 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10186 | #ifdef CONFIG_COMPAT |
34287aa3 | 10187 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 10188 | #endif |
6bf13c0c | 10189 | .release = btrfs_release_file, |
e02119d5 | 10190 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10191 | }; |
10192 | ||
20e5506b | 10193 | static const struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 10194 | .fill_delalloc = run_delalloc_range, |
065631f6 | 10195 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 10196 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 10197 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 10198 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 10199 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
10200 | .set_bit_hook = btrfs_set_bit_hook, |
10201 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
10202 | .merge_extent_hook = btrfs_merge_extent_hook, |
10203 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
10204 | }; |
10205 | ||
35054394 CM |
10206 | /* |
10207 | * btrfs doesn't support the bmap operation because swapfiles | |
10208 | * use bmap to make a mapping of extents in the file. They assume | |
10209 | * these extents won't change over the life of the file and they | |
10210 | * use the bmap result to do IO directly to the drive. | |
10211 | * | |
10212 | * the btrfs bmap call would return logical addresses that aren't | |
10213 | * suitable for IO and they also will change frequently as COW | |
10214 | * operations happen. So, swapfile + btrfs == corruption. | |
10215 | * | |
10216 | * For now we're avoiding this by dropping bmap. | |
10217 | */ | |
7f09410b | 10218 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
10219 | .readpage = btrfs_readpage, |
10220 | .writepage = btrfs_writepage, | |
b293f02e | 10221 | .writepages = btrfs_writepages, |
3ab2fb5a | 10222 | .readpages = btrfs_readpages, |
16432985 | 10223 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
10224 | .invalidatepage = btrfs_invalidatepage, |
10225 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 10226 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 10227 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
10228 | }; |
10229 | ||
7f09410b | 10230 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
10231 | .readpage = btrfs_readpage, |
10232 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
10233 | .invalidatepage = btrfs_invalidatepage, |
10234 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
10235 | }; |
10236 | ||
6e1d5dcc | 10237 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10238 | .getattr = btrfs_getattr, |
10239 | .setattr = btrfs_setattr, | |
e0d46f5c | 10240 | .setxattr = generic_setxattr, |
9172abbc | 10241 | .getxattr = generic_getxattr, |
5103e947 | 10242 | .listxattr = btrfs_listxattr, |
e0d46f5c | 10243 | .removexattr = generic_removexattr, |
fdebe2bd | 10244 | .permission = btrfs_permission, |
1506fcc8 | 10245 | .fiemap = btrfs_fiemap, |
4e34e719 | 10246 | .get_acl = btrfs_get_acl, |
996a710d | 10247 | .set_acl = btrfs_set_acl, |
e41f941a | 10248 | .update_time = btrfs_update_time, |
39279cc3 | 10249 | }; |
6e1d5dcc | 10250 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10251 | .getattr = btrfs_getattr, |
10252 | .setattr = btrfs_setattr, | |
fdebe2bd | 10253 | .permission = btrfs_permission, |
e0d46f5c | 10254 | .setxattr = generic_setxattr, |
9172abbc | 10255 | .getxattr = generic_getxattr, |
33268eaf | 10256 | .listxattr = btrfs_listxattr, |
e0d46f5c | 10257 | .removexattr = generic_removexattr, |
4e34e719 | 10258 | .get_acl = btrfs_get_acl, |
996a710d | 10259 | .set_acl = btrfs_set_acl, |
e41f941a | 10260 | .update_time = btrfs_update_time, |
618e21d5 | 10261 | }; |
6e1d5dcc | 10262 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 | 10263 | .readlink = generic_readlink, |
6b255391 | 10264 | .get_link = page_get_link, |
f209561a | 10265 | .getattr = btrfs_getattr, |
22c44fe6 | 10266 | .setattr = btrfs_setattr, |
fdebe2bd | 10267 | .permission = btrfs_permission, |
e0d46f5c | 10268 | .setxattr = generic_setxattr, |
9172abbc | 10269 | .getxattr = generic_getxattr, |
0279b4cd | 10270 | .listxattr = btrfs_listxattr, |
e0d46f5c | 10271 | .removexattr = generic_removexattr, |
e41f941a | 10272 | .update_time = btrfs_update_time, |
39279cc3 | 10273 | }; |
76dda93c | 10274 | |
82d339d9 | 10275 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 10276 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 10277 | .d_release = btrfs_dentry_release, |
76dda93c | 10278 | }; |