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> | |
a27bb332 | 35 | #include <linux/aio.h> |
9ebefb18 | 36 | #include <linux/bit_spinlock.h> |
5103e947 | 37 | #include <linux/xattr.h> |
33268eaf | 38 | #include <linux/posix_acl.h> |
d899e052 | 39 | #include <linux/falloc.h> |
5a0e3ad6 | 40 | #include <linux/slab.h> |
7a36ddec | 41 | #include <linux/ratelimit.h> |
22c44fe6 | 42 | #include <linux/mount.h> |
55e301fd | 43 | #include <linux/btrfs.h> |
53b381b3 | 44 | #include <linux/blkdev.h> |
f23b5a59 | 45 | #include <linux/posix_acl_xattr.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" |
39279cc3 CM |
62 | |
63 | struct btrfs_iget_args { | |
90d3e592 | 64 | struct btrfs_key *location; |
39279cc3 CM |
65 | struct btrfs_root *root; |
66 | }; | |
67 | ||
6e1d5dcc AD |
68 | static const struct inode_operations btrfs_dir_inode_operations; |
69 | static const struct inode_operations btrfs_symlink_inode_operations; | |
70 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
71 | static const struct inode_operations btrfs_special_inode_operations; | |
72 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
73 | static const struct address_space_operations btrfs_aops; |
74 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 75 | static const struct file_operations btrfs_dir_file_operations; |
d1310b2e | 76 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
77 | |
78 | static struct kmem_cache *btrfs_inode_cachep; | |
8ccf6f19 | 79 | static struct kmem_cache *btrfs_delalloc_work_cachep; |
39279cc3 CM |
80 | struct kmem_cache *btrfs_trans_handle_cachep; |
81 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 82 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 83 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
84 | |
85 | #define S_SHIFT 12 | |
86 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
87 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
88 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
89 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
90 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
91 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
92 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
93 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
94 | }; | |
95 | ||
3972f260 | 96 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
a41ad394 | 97 | static int btrfs_truncate(struct inode *inode); |
5fd02043 | 98 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
99 | static noinline int cow_file_range(struct inode *inode, |
100 | struct page *locked_page, | |
101 | u64 start, u64 end, int *page_started, | |
102 | unsigned long *nr_written, int unlock); | |
70c8a91c JB |
103 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
104 | u64 len, u64 orig_start, | |
105 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
106 | u64 orig_block_len, u64 ram_bytes, |
107 | int type); | |
7b128766 | 108 | |
48a3b636 | 109 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 110 | |
f34f57a3 | 111 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
112 | struct inode *inode, struct inode *dir, |
113 | const struct qstr *qstr) | |
0279b4cd JO |
114 | { |
115 | int err; | |
116 | ||
f34f57a3 | 117 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 118 | if (!err) |
2a7dba39 | 119 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
120 | return err; |
121 | } | |
122 | ||
c8b97818 CM |
123 | /* |
124 | * this does all the hard work for inserting an inline extent into | |
125 | * the btree. The caller should have done a btrfs_drop_extents so that | |
126 | * no overlapping inline items exist in the btree | |
127 | */ | |
40f76580 | 128 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
1acae57b | 129 | struct btrfs_path *path, int extent_inserted, |
c8b97818 CM |
130 | struct btrfs_root *root, struct inode *inode, |
131 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 132 | int compress_type, |
c8b97818 CM |
133 | struct page **compressed_pages) |
134 | { | |
c8b97818 CM |
135 | struct extent_buffer *leaf; |
136 | struct page *page = NULL; | |
137 | char *kaddr; | |
138 | unsigned long ptr; | |
139 | struct btrfs_file_extent_item *ei; | |
140 | int err = 0; | |
141 | int ret; | |
142 | size_t cur_size = size; | |
c8b97818 | 143 | unsigned long offset; |
c8b97818 | 144 | |
fe3f566c | 145 | if (compressed_size && compressed_pages) |
c8b97818 | 146 | cur_size = compressed_size; |
c8b97818 | 147 | |
1acae57b | 148 | inode_add_bytes(inode, size); |
c8b97818 | 149 | |
1acae57b FDBM |
150 | if (!extent_inserted) { |
151 | struct btrfs_key key; | |
152 | size_t datasize; | |
c8b97818 | 153 | |
1acae57b FDBM |
154 | key.objectid = btrfs_ino(inode); |
155 | key.offset = start; | |
962a298f | 156 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 157 | |
1acae57b FDBM |
158 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
159 | path->leave_spinning = 1; | |
160 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
161 | datasize); | |
162 | if (ret) { | |
163 | err = ret; | |
164 | goto fail; | |
165 | } | |
c8b97818 CM |
166 | } |
167 | leaf = path->nodes[0]; | |
168 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
169 | struct btrfs_file_extent_item); | |
170 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
171 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
172 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
173 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
174 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
175 | ptr = btrfs_file_extent_inline_start(ei); | |
176 | ||
261507a0 | 177 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
178 | struct page *cpage; |
179 | int i = 0; | |
d397712b | 180 | while (compressed_size > 0) { |
c8b97818 | 181 | cpage = compressed_pages[i]; |
5b050f04 | 182 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
183 | PAGE_CACHE_SIZE); |
184 | ||
7ac687d9 | 185 | kaddr = kmap_atomic(cpage); |
c8b97818 | 186 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 187 | kunmap_atomic(kaddr); |
c8b97818 CM |
188 | |
189 | i++; | |
190 | ptr += cur_size; | |
191 | compressed_size -= cur_size; | |
192 | } | |
193 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 194 | compress_type); |
c8b97818 CM |
195 | } else { |
196 | page = find_get_page(inode->i_mapping, | |
197 | start >> PAGE_CACHE_SHIFT); | |
198 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7ac687d9 | 199 | kaddr = kmap_atomic(page); |
c8b97818 CM |
200 | offset = start & (PAGE_CACHE_SIZE - 1); |
201 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
7ac687d9 | 202 | kunmap_atomic(kaddr); |
c8b97818 CM |
203 | page_cache_release(page); |
204 | } | |
205 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 206 | btrfs_release_path(path); |
c8b97818 | 207 | |
c2167754 YZ |
208 | /* |
209 | * we're an inline extent, so nobody can | |
210 | * extend the file past i_size without locking | |
211 | * a page we already have locked. | |
212 | * | |
213 | * We must do any isize and inode updates | |
214 | * before we unlock the pages. Otherwise we | |
215 | * could end up racing with unlink. | |
216 | */ | |
c8b97818 | 217 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 218 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 219 | |
79787eaa | 220 | return ret; |
c8b97818 | 221 | fail: |
c8b97818 CM |
222 | return err; |
223 | } | |
224 | ||
225 | ||
226 | /* | |
227 | * conditionally insert an inline extent into the file. This | |
228 | * does the checks required to make sure the data is small enough | |
229 | * to fit as an inline extent. | |
230 | */ | |
00361589 JB |
231 | static noinline int cow_file_range_inline(struct btrfs_root *root, |
232 | struct inode *inode, u64 start, | |
233 | u64 end, size_t compressed_size, | |
234 | int compress_type, | |
235 | struct page **compressed_pages) | |
c8b97818 | 236 | { |
00361589 | 237 | struct btrfs_trans_handle *trans; |
c8b97818 CM |
238 | u64 isize = i_size_read(inode); |
239 | u64 actual_end = min(end + 1, isize); | |
240 | u64 inline_len = actual_end - start; | |
fda2832f | 241 | u64 aligned_end = ALIGN(end, root->sectorsize); |
c8b97818 CM |
242 | u64 data_len = inline_len; |
243 | int ret; | |
1acae57b FDBM |
244 | struct btrfs_path *path; |
245 | int extent_inserted = 0; | |
246 | u32 extent_item_size; | |
c8b97818 CM |
247 | |
248 | if (compressed_size) | |
249 | data_len = compressed_size; | |
250 | ||
251 | if (start > 0 || | |
354877be WS |
252 | actual_end > PAGE_CACHE_SIZE || |
253 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) || | |
c8b97818 CM |
254 | (!compressed_size && |
255 | (actual_end & (root->sectorsize - 1)) == 0) || | |
256 | end + 1 < isize || | |
257 | data_len > root->fs_info->max_inline) { | |
258 | return 1; | |
259 | } | |
260 | ||
1acae57b FDBM |
261 | path = btrfs_alloc_path(); |
262 | if (!path) | |
263 | return -ENOMEM; | |
264 | ||
00361589 | 265 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
266 | if (IS_ERR(trans)) { |
267 | btrfs_free_path(path); | |
00361589 | 268 | return PTR_ERR(trans); |
1acae57b | 269 | } |
00361589 JB |
270 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
271 | ||
1acae57b FDBM |
272 | if (compressed_size && compressed_pages) |
273 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
274 | compressed_size); | |
275 | else | |
276 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
277 | inline_len); | |
278 | ||
279 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
280 | start, aligned_end, NULL, | |
281 | 1, 1, extent_item_size, &extent_inserted); | |
00361589 JB |
282 | if (ret) { |
283 | btrfs_abort_transaction(trans, root, ret); | |
284 | goto out; | |
285 | } | |
c8b97818 CM |
286 | |
287 | if (isize > actual_end) | |
288 | inline_len = min_t(u64, isize, actual_end); | |
1acae57b FDBM |
289 | ret = insert_inline_extent(trans, path, extent_inserted, |
290 | root, inode, start, | |
c8b97818 | 291 | inline_len, compressed_size, |
fe3f566c | 292 | compress_type, compressed_pages); |
2adcac1a | 293 | if (ret && ret != -ENOSPC) { |
79787eaa | 294 | btrfs_abort_transaction(trans, root, ret); |
00361589 | 295 | goto out; |
2adcac1a | 296 | } else if (ret == -ENOSPC) { |
00361589 JB |
297 | ret = 1; |
298 | goto out; | |
79787eaa | 299 | } |
2adcac1a | 300 | |
bdc20e67 | 301 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
0ca1f7ce | 302 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 303 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
00361589 | 304 | out: |
1acae57b | 305 | btrfs_free_path(path); |
00361589 JB |
306 | btrfs_end_transaction(trans, root); |
307 | return ret; | |
c8b97818 CM |
308 | } |
309 | ||
771ed689 CM |
310 | struct async_extent { |
311 | u64 start; | |
312 | u64 ram_size; | |
313 | u64 compressed_size; | |
314 | struct page **pages; | |
315 | unsigned long nr_pages; | |
261507a0 | 316 | int compress_type; |
771ed689 CM |
317 | struct list_head list; |
318 | }; | |
319 | ||
320 | struct async_cow { | |
321 | struct inode *inode; | |
322 | struct btrfs_root *root; | |
323 | struct page *locked_page; | |
324 | u64 start; | |
325 | u64 end; | |
326 | struct list_head extents; | |
327 | struct btrfs_work work; | |
328 | }; | |
329 | ||
330 | static noinline int add_async_extent(struct async_cow *cow, | |
331 | u64 start, u64 ram_size, | |
332 | u64 compressed_size, | |
333 | struct page **pages, | |
261507a0 LZ |
334 | unsigned long nr_pages, |
335 | int compress_type) | |
771ed689 CM |
336 | { |
337 | struct async_extent *async_extent; | |
338 | ||
339 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 340 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
341 | async_extent->start = start; |
342 | async_extent->ram_size = ram_size; | |
343 | async_extent->compressed_size = compressed_size; | |
344 | async_extent->pages = pages; | |
345 | async_extent->nr_pages = nr_pages; | |
261507a0 | 346 | async_extent->compress_type = compress_type; |
771ed689 CM |
347 | list_add_tail(&async_extent->list, &cow->extents); |
348 | return 0; | |
349 | } | |
350 | ||
f79707b0 WS |
351 | static inline int inode_need_compress(struct inode *inode) |
352 | { | |
353 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
354 | ||
355 | /* force compress */ | |
356 | if (btrfs_test_opt(root, FORCE_COMPRESS)) | |
357 | return 1; | |
358 | /* bad compression ratios */ | |
359 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) | |
360 | return 0; | |
361 | if (btrfs_test_opt(root, COMPRESS) || | |
362 | BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS || | |
363 | BTRFS_I(inode)->force_compress) | |
364 | return 1; | |
365 | return 0; | |
366 | } | |
367 | ||
d352ac68 | 368 | /* |
771ed689 CM |
369 | * we create compressed extents in two phases. The first |
370 | * phase compresses a range of pages that have already been | |
371 | * locked (both pages and state bits are locked). | |
c8b97818 | 372 | * |
771ed689 CM |
373 | * This is done inside an ordered work queue, and the compression |
374 | * is spread across many cpus. The actual IO submission is step | |
375 | * two, and the ordered work queue takes care of making sure that | |
376 | * happens in the same order things were put onto the queue by | |
377 | * writepages and friends. | |
c8b97818 | 378 | * |
771ed689 CM |
379 | * If this code finds it can't get good compression, it puts an |
380 | * entry onto the work queue to write the uncompressed bytes. This | |
381 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
382 | * are written in the same order that the flusher thread sent them |
383 | * down. | |
d352ac68 | 384 | */ |
c44f649e | 385 | static noinline void compress_file_range(struct inode *inode, |
771ed689 CM |
386 | struct page *locked_page, |
387 | u64 start, u64 end, | |
388 | struct async_cow *async_cow, | |
389 | int *num_added) | |
b888db2b CM |
390 | { |
391 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
db94535d | 392 | u64 num_bytes; |
db94535d | 393 | u64 blocksize = root->sectorsize; |
c8b97818 | 394 | u64 actual_end; |
42dc7bab | 395 | u64 isize = i_size_read(inode); |
e6dcd2dc | 396 | int ret = 0; |
c8b97818 CM |
397 | struct page **pages = NULL; |
398 | unsigned long nr_pages; | |
399 | unsigned long nr_pages_ret = 0; | |
400 | unsigned long total_compressed = 0; | |
401 | unsigned long total_in = 0; | |
402 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 403 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
404 | int i; |
405 | int will_compress; | |
261507a0 | 406 | int compress_type = root->fs_info->compress_type; |
4adaa611 | 407 | int redirty = 0; |
b888db2b | 408 | |
4cb13e5d LB |
409 | /* if this is a small write inside eof, kick off a defrag */ |
410 | if ((end - start + 1) < 16 * 1024 && | |
411 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
412 | btrfs_add_inode_defrag(NULL, inode); |
413 | ||
42dc7bab | 414 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
415 | again: |
416 | will_compress = 0; | |
417 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
418 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 419 | |
f03d9301 CM |
420 | /* |
421 | * we don't want to send crud past the end of i_size through | |
422 | * compression, that's just a waste of CPU time. So, if the | |
423 | * end of the file is before the start of our current | |
424 | * requested range of bytes, we bail out to the uncompressed | |
425 | * cleanup code that can deal with all of this. | |
426 | * | |
427 | * It isn't really the fastest way to fix things, but this is a | |
428 | * very uncommon corner. | |
429 | */ | |
430 | if (actual_end <= start) | |
431 | goto cleanup_and_bail_uncompressed; | |
432 | ||
c8b97818 CM |
433 | total_compressed = actual_end - start; |
434 | ||
4bcbb332 SW |
435 | /* |
436 | * skip compression for a small file range(<=blocksize) that | |
437 | * isn't an inline extent, since it dosen't save disk space at all. | |
438 | */ | |
439 | if (total_compressed <= blocksize && | |
440 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
441 | goto cleanup_and_bail_uncompressed; | |
442 | ||
c8b97818 CM |
443 | /* we want to make sure that amount of ram required to uncompress |
444 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
445 | * of a compressed extent to 128k. This is a crucial number |
446 | * because it also controls how easily we can spread reads across | |
447 | * cpus for decompression. | |
448 | * | |
449 | * We also want to make sure the amount of IO required to do | |
450 | * a random read is reasonably small, so we limit the size of | |
451 | * a compressed extent to 128k. | |
c8b97818 CM |
452 | */ |
453 | total_compressed = min(total_compressed, max_uncompressed); | |
fda2832f | 454 | num_bytes = ALIGN(end - start + 1, blocksize); |
be20aa9d | 455 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
456 | total_in = 0; |
457 | ret = 0; | |
db94535d | 458 | |
771ed689 CM |
459 | /* |
460 | * we do compression for mount -o compress and when the | |
461 | * inode has not been flagged as nocompress. This flag can | |
462 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 463 | */ |
f79707b0 | 464 | if (inode_need_compress(inode)) { |
c8b97818 | 465 | WARN_ON(pages); |
cfbc246e | 466 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
560f7d75 LZ |
467 | if (!pages) { |
468 | /* just bail out to the uncompressed code */ | |
469 | goto cont; | |
470 | } | |
c8b97818 | 471 | |
261507a0 LZ |
472 | if (BTRFS_I(inode)->force_compress) |
473 | compress_type = BTRFS_I(inode)->force_compress; | |
474 | ||
4adaa611 CM |
475 | /* |
476 | * we need to call clear_page_dirty_for_io on each | |
477 | * page in the range. Otherwise applications with the file | |
478 | * mmap'd can wander in and change the page contents while | |
479 | * we are compressing them. | |
480 | * | |
481 | * If the compression fails for any reason, we set the pages | |
482 | * dirty again later on. | |
483 | */ | |
484 | extent_range_clear_dirty_for_io(inode, start, end); | |
485 | redirty = 1; | |
261507a0 LZ |
486 | ret = btrfs_compress_pages(compress_type, |
487 | inode->i_mapping, start, | |
488 | total_compressed, pages, | |
489 | nr_pages, &nr_pages_ret, | |
490 | &total_in, | |
491 | &total_compressed, | |
492 | max_compressed); | |
c8b97818 CM |
493 | |
494 | if (!ret) { | |
495 | unsigned long offset = total_compressed & | |
496 | (PAGE_CACHE_SIZE - 1); | |
497 | struct page *page = pages[nr_pages_ret - 1]; | |
498 | char *kaddr; | |
499 | ||
500 | /* zero the tail end of the last page, we might be | |
501 | * sending it down to disk | |
502 | */ | |
503 | if (offset) { | |
7ac687d9 | 504 | kaddr = kmap_atomic(page); |
c8b97818 CM |
505 | memset(kaddr + offset, 0, |
506 | PAGE_CACHE_SIZE - offset); | |
7ac687d9 | 507 | kunmap_atomic(kaddr); |
c8b97818 CM |
508 | } |
509 | will_compress = 1; | |
510 | } | |
511 | } | |
560f7d75 | 512 | cont: |
c8b97818 CM |
513 | if (start == 0) { |
514 | /* lets try to make an inline extent */ | |
771ed689 | 515 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 516 | /* we didn't compress the entire range, try |
771ed689 | 517 | * to make an uncompressed inline extent. |
c8b97818 | 518 | */ |
00361589 JB |
519 | ret = cow_file_range_inline(root, inode, start, end, |
520 | 0, 0, NULL); | |
c8b97818 | 521 | } else { |
771ed689 | 522 | /* try making a compressed inline extent */ |
00361589 | 523 | ret = cow_file_range_inline(root, inode, start, end, |
fe3f566c LZ |
524 | total_compressed, |
525 | compress_type, pages); | |
c8b97818 | 526 | } |
79787eaa | 527 | if (ret <= 0) { |
151a41bc JB |
528 | unsigned long clear_flags = EXTENT_DELALLOC | |
529 | EXTENT_DEFRAG; | |
530 | clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0; | |
531 | ||
771ed689 | 532 | /* |
79787eaa JM |
533 | * inline extent creation worked or returned error, |
534 | * we don't need to create any more async work items. | |
535 | * Unlock and free up our temp pages. | |
771ed689 | 536 | */ |
c2790a2e | 537 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
151a41bc | 538 | clear_flags, PAGE_UNLOCK | |
c2790a2e JB |
539 | PAGE_CLEAR_DIRTY | |
540 | PAGE_SET_WRITEBACK | | |
541 | PAGE_END_WRITEBACK); | |
c8b97818 CM |
542 | goto free_pages_out; |
543 | } | |
544 | } | |
545 | ||
546 | if (will_compress) { | |
547 | /* | |
548 | * we aren't doing an inline extent round the compressed size | |
549 | * up to a block size boundary so the allocator does sane | |
550 | * things | |
551 | */ | |
fda2832f | 552 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
553 | |
554 | /* | |
555 | * one last check to make sure the compression is really a | |
556 | * win, compare the page count read with the blocks on disk | |
557 | */ | |
fda2832f | 558 | total_in = ALIGN(total_in, PAGE_CACHE_SIZE); |
c8b97818 CM |
559 | if (total_compressed >= total_in) { |
560 | will_compress = 0; | |
561 | } else { | |
c8b97818 CM |
562 | num_bytes = total_in; |
563 | } | |
564 | } | |
565 | if (!will_compress && pages) { | |
566 | /* | |
567 | * the compression code ran but failed to make things smaller, | |
568 | * free any pages it allocated and our page pointer array | |
569 | */ | |
570 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 571 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
572 | page_cache_release(pages[i]); |
573 | } | |
574 | kfree(pages); | |
575 | pages = NULL; | |
576 | total_compressed = 0; | |
577 | nr_pages_ret = 0; | |
578 | ||
579 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
580 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
581 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 582 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 583 | } |
c8b97818 | 584 | } |
771ed689 CM |
585 | if (will_compress) { |
586 | *num_added += 1; | |
c8b97818 | 587 | |
771ed689 CM |
588 | /* the async work queues will take care of doing actual |
589 | * allocation on disk for these compressed pages, | |
590 | * and will submit them to the elevator. | |
591 | */ | |
592 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
593 | total_compressed, pages, nr_pages_ret, |
594 | compress_type); | |
179e29e4 | 595 | |
24ae6365 | 596 | if (start + num_bytes < end) { |
771ed689 CM |
597 | start += num_bytes; |
598 | pages = NULL; | |
599 | cond_resched(); | |
600 | goto again; | |
601 | } | |
602 | } else { | |
f03d9301 | 603 | cleanup_and_bail_uncompressed: |
771ed689 CM |
604 | /* |
605 | * No compression, but we still need to write the pages in | |
606 | * the file we've been given so far. redirty the locked | |
607 | * page if it corresponds to our extent and set things up | |
608 | * for the async work queue to run cow_file_range to do | |
609 | * the normal delalloc dance | |
610 | */ | |
611 | if (page_offset(locked_page) >= start && | |
612 | page_offset(locked_page) <= end) { | |
613 | __set_page_dirty_nobuffers(locked_page); | |
614 | /* unlocked later on in the async handlers */ | |
615 | } | |
4adaa611 CM |
616 | if (redirty) |
617 | extent_range_redirty_for_io(inode, start, end); | |
261507a0 LZ |
618 | add_async_extent(async_cow, start, end - start + 1, |
619 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
620 | *num_added += 1; |
621 | } | |
3b951516 | 622 | |
c44f649e | 623 | return; |
771ed689 CM |
624 | |
625 | free_pages_out: | |
626 | for (i = 0; i < nr_pages_ret; i++) { | |
627 | WARN_ON(pages[i]->mapping); | |
628 | page_cache_release(pages[i]); | |
629 | } | |
d397712b | 630 | kfree(pages); |
771ed689 CM |
631 | } |
632 | ||
40ae837b FM |
633 | static void free_async_extent_pages(struct async_extent *async_extent) |
634 | { | |
635 | int i; | |
636 | ||
637 | if (!async_extent->pages) | |
638 | return; | |
639 | ||
640 | for (i = 0; i < async_extent->nr_pages; i++) { | |
641 | WARN_ON(async_extent->pages[i]->mapping); | |
642 | page_cache_release(async_extent->pages[i]); | |
643 | } | |
644 | kfree(async_extent->pages); | |
645 | async_extent->nr_pages = 0; | |
646 | async_extent->pages = NULL; | |
647 | } | |
648 | ||
771ed689 CM |
649 | /* |
650 | * phase two of compressed writeback. This is the ordered portion | |
651 | * of the code, which only gets called in the order the work was | |
652 | * queued. We walk all the async extents created by compress_file_range | |
653 | * and send them down to the disk. | |
654 | */ | |
dec8f175 | 655 | static noinline void submit_compressed_extents(struct inode *inode, |
771ed689 CM |
656 | struct async_cow *async_cow) |
657 | { | |
658 | struct async_extent *async_extent; | |
659 | u64 alloc_hint = 0; | |
771ed689 CM |
660 | struct btrfs_key ins; |
661 | struct extent_map *em; | |
662 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
663 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
664 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 665 | int ret = 0; |
771ed689 | 666 | |
3e04e7f1 | 667 | again: |
d397712b | 668 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
669 | async_extent = list_entry(async_cow->extents.next, |
670 | struct async_extent, list); | |
671 | list_del(&async_extent->list); | |
c8b97818 | 672 | |
771ed689 CM |
673 | io_tree = &BTRFS_I(inode)->io_tree; |
674 | ||
f5a84ee3 | 675 | retry: |
771ed689 CM |
676 | /* did the compression code fall back to uncompressed IO? */ |
677 | if (!async_extent->pages) { | |
678 | int page_started = 0; | |
679 | unsigned long nr_written = 0; | |
680 | ||
681 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 682 | async_extent->start + |
d0082371 | 683 | async_extent->ram_size - 1); |
771ed689 CM |
684 | |
685 | /* allocate blocks */ | |
f5a84ee3 JB |
686 | ret = cow_file_range(inode, async_cow->locked_page, |
687 | async_extent->start, | |
688 | async_extent->start + | |
689 | async_extent->ram_size - 1, | |
690 | &page_started, &nr_written, 0); | |
771ed689 | 691 | |
79787eaa JM |
692 | /* JDM XXX */ |
693 | ||
771ed689 CM |
694 | /* |
695 | * if page_started, cow_file_range inserted an | |
696 | * inline extent and took care of all the unlocking | |
697 | * and IO for us. Otherwise, we need to submit | |
698 | * all those pages down to the drive. | |
699 | */ | |
f5a84ee3 | 700 | if (!page_started && !ret) |
771ed689 CM |
701 | extent_write_locked_range(io_tree, |
702 | inode, async_extent->start, | |
d397712b | 703 | async_extent->start + |
771ed689 CM |
704 | async_extent->ram_size - 1, |
705 | btrfs_get_extent, | |
706 | WB_SYNC_ALL); | |
3e04e7f1 JB |
707 | else if (ret) |
708 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
709 | kfree(async_extent); |
710 | cond_resched(); | |
711 | continue; | |
712 | } | |
713 | ||
714 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 715 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 716 | |
00361589 | 717 | ret = btrfs_reserve_extent(root, |
771ed689 CM |
718 | async_extent->compressed_size, |
719 | async_extent->compressed_size, | |
e570fd27 | 720 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 721 | if (ret) { |
40ae837b | 722 | free_async_extent_pages(async_extent); |
3e04e7f1 | 723 | |
fdf8e2ea JB |
724 | if (ret == -ENOSPC) { |
725 | unlock_extent(io_tree, async_extent->start, | |
726 | async_extent->start + | |
727 | async_extent->ram_size - 1); | |
ce62003f LB |
728 | |
729 | /* | |
730 | * we need to redirty the pages if we decide to | |
731 | * fallback to uncompressed IO, otherwise we | |
732 | * will not submit these pages down to lower | |
733 | * layers. | |
734 | */ | |
735 | extent_range_redirty_for_io(inode, | |
736 | async_extent->start, | |
737 | async_extent->start + | |
738 | async_extent->ram_size - 1); | |
739 | ||
79787eaa | 740 | goto retry; |
fdf8e2ea | 741 | } |
3e04e7f1 | 742 | goto out_free; |
f5a84ee3 JB |
743 | } |
744 | ||
c2167754 YZ |
745 | /* |
746 | * here we're doing allocation and writeback of the | |
747 | * compressed pages | |
748 | */ | |
749 | btrfs_drop_extent_cache(inode, async_extent->start, | |
750 | async_extent->start + | |
751 | async_extent->ram_size - 1, 0); | |
752 | ||
172ddd60 | 753 | em = alloc_extent_map(); |
b9aa55be LB |
754 | if (!em) { |
755 | ret = -ENOMEM; | |
3e04e7f1 | 756 | goto out_free_reserve; |
b9aa55be | 757 | } |
771ed689 CM |
758 | em->start = async_extent->start; |
759 | em->len = async_extent->ram_size; | |
445a6944 | 760 | em->orig_start = em->start; |
2ab28f32 JB |
761 | em->mod_start = em->start; |
762 | em->mod_len = em->len; | |
c8b97818 | 763 | |
771ed689 CM |
764 | em->block_start = ins.objectid; |
765 | em->block_len = ins.offset; | |
b4939680 | 766 | em->orig_block_len = ins.offset; |
cc95bef6 | 767 | em->ram_bytes = async_extent->ram_size; |
771ed689 | 768 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
261507a0 | 769 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
770 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
771 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
70c8a91c | 772 | em->generation = -1; |
771ed689 | 773 | |
d397712b | 774 | while (1) { |
890871be | 775 | write_lock(&em_tree->lock); |
09a2a8f9 | 776 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 777 | write_unlock(&em_tree->lock); |
771ed689 CM |
778 | if (ret != -EEXIST) { |
779 | free_extent_map(em); | |
780 | break; | |
781 | } | |
782 | btrfs_drop_extent_cache(inode, async_extent->start, | |
783 | async_extent->start + | |
784 | async_extent->ram_size - 1, 0); | |
785 | } | |
786 | ||
3e04e7f1 JB |
787 | if (ret) |
788 | goto out_free_reserve; | |
789 | ||
261507a0 LZ |
790 | ret = btrfs_add_ordered_extent_compress(inode, |
791 | async_extent->start, | |
792 | ins.objectid, | |
793 | async_extent->ram_size, | |
794 | ins.offset, | |
795 | BTRFS_ORDERED_COMPRESSED, | |
796 | async_extent->compress_type); | |
d9f85963 FM |
797 | if (ret) { |
798 | btrfs_drop_extent_cache(inode, async_extent->start, | |
799 | async_extent->start + | |
800 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 801 | goto out_free_reserve; |
d9f85963 | 802 | } |
771ed689 | 803 | |
771ed689 CM |
804 | /* |
805 | * clear dirty, set writeback and unlock the pages. | |
806 | */ | |
c2790a2e | 807 | extent_clear_unlock_delalloc(inode, async_extent->start, |
a791e35e CM |
808 | async_extent->start + |
809 | async_extent->ram_size - 1, | |
151a41bc JB |
810 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
811 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 812 | PAGE_SET_WRITEBACK); |
771ed689 | 813 | ret = btrfs_submit_compressed_write(inode, |
d397712b CM |
814 | async_extent->start, |
815 | async_extent->ram_size, | |
816 | ins.objectid, | |
817 | ins.offset, async_extent->pages, | |
818 | async_extent->nr_pages); | |
fce2a4e6 FM |
819 | if (ret) { |
820 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
821 | struct page *p = async_extent->pages[0]; | |
822 | const u64 start = async_extent->start; | |
823 | const u64 end = start + async_extent->ram_size - 1; | |
824 | ||
825 | p->mapping = inode->i_mapping; | |
826 | tree->ops->writepage_end_io_hook(p, start, end, | |
827 | NULL, 0); | |
828 | p->mapping = NULL; | |
829 | extent_clear_unlock_delalloc(inode, start, end, NULL, 0, | |
830 | PAGE_END_WRITEBACK | | |
831 | PAGE_SET_ERROR); | |
40ae837b | 832 | free_async_extent_pages(async_extent); |
fce2a4e6 | 833 | } |
771ed689 CM |
834 | alloc_hint = ins.objectid + ins.offset; |
835 | kfree(async_extent); | |
836 | cond_resched(); | |
837 | } | |
dec8f175 | 838 | return; |
3e04e7f1 | 839 | out_free_reserve: |
e570fd27 | 840 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
79787eaa | 841 | out_free: |
c2790a2e | 842 | extent_clear_unlock_delalloc(inode, async_extent->start, |
3e04e7f1 JB |
843 | async_extent->start + |
844 | async_extent->ram_size - 1, | |
c2790a2e | 845 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
846 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
847 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
848 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
849 | PAGE_SET_ERROR); | |
40ae837b | 850 | free_async_extent_pages(async_extent); |
79787eaa | 851 | kfree(async_extent); |
3e04e7f1 | 852 | goto again; |
771ed689 CM |
853 | } |
854 | ||
4b46fce2 JB |
855 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
856 | u64 num_bytes) | |
857 | { | |
858 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
859 | struct extent_map *em; | |
860 | u64 alloc_hint = 0; | |
861 | ||
862 | read_lock(&em_tree->lock); | |
863 | em = search_extent_mapping(em_tree, start, num_bytes); | |
864 | if (em) { | |
865 | /* | |
866 | * if block start isn't an actual block number then find the | |
867 | * first block in this inode and use that as a hint. If that | |
868 | * block is also bogus then just don't worry about it. | |
869 | */ | |
870 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
871 | free_extent_map(em); | |
872 | em = search_extent_mapping(em_tree, 0, 0); | |
873 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
874 | alloc_hint = em->block_start; | |
875 | if (em) | |
876 | free_extent_map(em); | |
877 | } else { | |
878 | alloc_hint = em->block_start; | |
879 | free_extent_map(em); | |
880 | } | |
881 | } | |
882 | read_unlock(&em_tree->lock); | |
883 | ||
884 | return alloc_hint; | |
885 | } | |
886 | ||
771ed689 CM |
887 | /* |
888 | * when extent_io.c finds a delayed allocation range in the file, | |
889 | * the call backs end up in this code. The basic idea is to | |
890 | * allocate extents on disk for the range, and create ordered data structs | |
891 | * in ram to track those extents. | |
892 | * | |
893 | * locked_page is the page that writepage had locked already. We use | |
894 | * it to make sure we don't do extra locks or unlocks. | |
895 | * | |
896 | * *page_started is set to one if we unlock locked_page and do everything | |
897 | * required to start IO on it. It may be clean and already done with | |
898 | * IO when we return. | |
899 | */ | |
00361589 JB |
900 | static noinline int cow_file_range(struct inode *inode, |
901 | struct page *locked_page, | |
902 | u64 start, u64 end, int *page_started, | |
903 | unsigned long *nr_written, | |
904 | int unlock) | |
771ed689 | 905 | { |
00361589 | 906 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
907 | u64 alloc_hint = 0; |
908 | u64 num_bytes; | |
909 | unsigned long ram_size; | |
910 | u64 disk_num_bytes; | |
911 | u64 cur_alloc_size; | |
912 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
913 | struct btrfs_key ins; |
914 | struct extent_map *em; | |
915 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
916 | int ret = 0; | |
917 | ||
02ecd2c2 JB |
918 | if (btrfs_is_free_space_inode(inode)) { |
919 | WARN_ON_ONCE(1); | |
29bce2f3 JB |
920 | ret = -EINVAL; |
921 | goto out_unlock; | |
02ecd2c2 | 922 | } |
771ed689 | 923 | |
fda2832f | 924 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 CM |
925 | num_bytes = max(blocksize, num_bytes); |
926 | disk_num_bytes = num_bytes; | |
771ed689 | 927 | |
4cb5300b | 928 | /* if this is a small write inside eof, kick off defrag */ |
4cb13e5d LB |
929 | if (num_bytes < 64 * 1024 && |
930 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
00361589 | 931 | btrfs_add_inode_defrag(NULL, inode); |
4cb5300b | 932 | |
771ed689 CM |
933 | if (start == 0) { |
934 | /* lets try to make an inline extent */ | |
00361589 JB |
935 | ret = cow_file_range_inline(root, inode, start, end, 0, 0, |
936 | NULL); | |
771ed689 | 937 | if (ret == 0) { |
c2790a2e JB |
938 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
939 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc | 940 | EXTENT_DEFRAG, PAGE_UNLOCK | |
c2790a2e JB |
941 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
942 | PAGE_END_WRITEBACK); | |
c2167754 | 943 | |
771ed689 CM |
944 | *nr_written = *nr_written + |
945 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
946 | *page_started = 1; | |
771ed689 | 947 | goto out; |
79787eaa | 948 | } else if (ret < 0) { |
79787eaa | 949 | goto out_unlock; |
771ed689 CM |
950 | } |
951 | } | |
952 | ||
953 | BUG_ON(disk_num_bytes > | |
6c41761f | 954 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 955 | |
4b46fce2 | 956 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
957 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
958 | ||
d397712b | 959 | while (disk_num_bytes > 0) { |
a791e35e CM |
960 | unsigned long op; |
961 | ||
287a0ab9 | 962 | cur_alloc_size = disk_num_bytes; |
00361589 | 963 | ret = btrfs_reserve_extent(root, cur_alloc_size, |
771ed689 | 964 | root->sectorsize, 0, alloc_hint, |
e570fd27 | 965 | &ins, 1, 1); |
00361589 | 966 | if (ret < 0) |
79787eaa | 967 | goto out_unlock; |
d397712b | 968 | |
172ddd60 | 969 | em = alloc_extent_map(); |
b9aa55be LB |
970 | if (!em) { |
971 | ret = -ENOMEM; | |
ace68bac | 972 | goto out_reserve; |
b9aa55be | 973 | } |
e6dcd2dc | 974 | em->start = start; |
445a6944 | 975 | em->orig_start = em->start; |
771ed689 CM |
976 | ram_size = ins.offset; |
977 | em->len = ins.offset; | |
2ab28f32 JB |
978 | em->mod_start = em->start; |
979 | em->mod_len = em->len; | |
c8b97818 | 980 | |
e6dcd2dc | 981 | em->block_start = ins.objectid; |
c8b97818 | 982 | em->block_len = ins.offset; |
b4939680 | 983 | em->orig_block_len = ins.offset; |
cc95bef6 | 984 | em->ram_bytes = ram_size; |
e6dcd2dc | 985 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 986 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
70c8a91c | 987 | em->generation = -1; |
c8b97818 | 988 | |
d397712b | 989 | while (1) { |
890871be | 990 | write_lock(&em_tree->lock); |
09a2a8f9 | 991 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 992 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
993 | if (ret != -EEXIST) { |
994 | free_extent_map(em); | |
995 | break; | |
996 | } | |
997 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 998 | start + ram_size - 1, 0); |
e6dcd2dc | 999 | } |
ace68bac LB |
1000 | if (ret) |
1001 | goto out_reserve; | |
e6dcd2dc | 1002 | |
98d20f67 | 1003 | cur_alloc_size = ins.offset; |
e6dcd2dc | 1004 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 1005 | ram_size, cur_alloc_size, 0); |
ace68bac | 1006 | if (ret) |
d9f85963 | 1007 | goto out_drop_extent_cache; |
c8b97818 | 1008 | |
17d217fe YZ |
1009 | if (root->root_key.objectid == |
1010 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1011 | ret = btrfs_reloc_clone_csums(inode, start, | |
1012 | cur_alloc_size); | |
00361589 | 1013 | if (ret) |
d9f85963 | 1014 | goto out_drop_extent_cache; |
17d217fe YZ |
1015 | } |
1016 | ||
d397712b | 1017 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 1018 | break; |
d397712b | 1019 | |
c8b97818 CM |
1020 | /* we're not doing compressed IO, don't unlock the first |
1021 | * page (which the caller expects to stay locked), don't | |
1022 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1023 | * |
1024 | * Do set the Private2 bit so we know this page was properly | |
1025 | * setup for writepage | |
c8b97818 | 1026 | */ |
c2790a2e JB |
1027 | op = unlock ? PAGE_UNLOCK : 0; |
1028 | op |= PAGE_SET_PRIVATE2; | |
a791e35e | 1029 | |
c2790a2e JB |
1030 | extent_clear_unlock_delalloc(inode, start, |
1031 | start + ram_size - 1, locked_page, | |
1032 | EXTENT_LOCKED | EXTENT_DELALLOC, | |
1033 | op); | |
c8b97818 | 1034 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
1035 | num_bytes -= cur_alloc_size; |
1036 | alloc_hint = ins.objectid + ins.offset; | |
1037 | start += cur_alloc_size; | |
b888db2b | 1038 | } |
79787eaa | 1039 | out: |
be20aa9d | 1040 | return ret; |
b7d5b0a8 | 1041 | |
d9f85963 FM |
1042 | out_drop_extent_cache: |
1043 | btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0); | |
ace68bac | 1044 | out_reserve: |
e570fd27 | 1045 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
79787eaa | 1046 | out_unlock: |
c2790a2e | 1047 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
151a41bc JB |
1048 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
1049 | EXTENT_DELALLOC | EXTENT_DEFRAG, | |
1050 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
1051 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 1052 | goto out; |
771ed689 | 1053 | } |
c8b97818 | 1054 | |
771ed689 CM |
1055 | /* |
1056 | * work queue call back to started compression on a file and pages | |
1057 | */ | |
1058 | static noinline void async_cow_start(struct btrfs_work *work) | |
1059 | { | |
1060 | struct async_cow *async_cow; | |
1061 | int num_added = 0; | |
1062 | async_cow = container_of(work, struct async_cow, work); | |
1063 | ||
1064 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1065 | async_cow->start, async_cow->end, async_cow, | |
1066 | &num_added); | |
8180ef88 | 1067 | if (num_added == 0) { |
cb77fcd8 | 1068 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1069 | async_cow->inode = NULL; |
8180ef88 | 1070 | } |
771ed689 CM |
1071 | } |
1072 | ||
1073 | /* | |
1074 | * work queue call back to submit previously compressed pages | |
1075 | */ | |
1076 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1077 | { | |
1078 | struct async_cow *async_cow; | |
1079 | struct btrfs_root *root; | |
1080 | unsigned long nr_pages; | |
1081 | ||
1082 | async_cow = container_of(work, struct async_cow, work); | |
1083 | ||
1084 | root = async_cow->root; | |
1085 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
1086 | PAGE_CACHE_SHIFT; | |
1087 | ||
66657b31 | 1088 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
287082b0 | 1089 | 5 * 1024 * 1024 && |
771ed689 CM |
1090 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1091 | wake_up(&root->fs_info->async_submit_wait); | |
1092 | ||
d397712b | 1093 | if (async_cow->inode) |
771ed689 | 1094 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1095 | } |
c8b97818 | 1096 | |
771ed689 CM |
1097 | static noinline void async_cow_free(struct btrfs_work *work) |
1098 | { | |
1099 | struct async_cow *async_cow; | |
1100 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1101 | if (async_cow->inode) |
cb77fcd8 | 1102 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1103 | kfree(async_cow); |
1104 | } | |
1105 | ||
1106 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1107 | u64 start, u64 end, int *page_started, | |
1108 | unsigned long *nr_written) | |
1109 | { | |
1110 | struct async_cow *async_cow; | |
1111 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1112 | unsigned long nr_pages; | |
1113 | u64 cur_end; | |
287082b0 | 1114 | int limit = 10 * 1024 * 1024; |
771ed689 | 1115 | |
a3429ab7 CM |
1116 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1117 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1118 | while (start < end) { |
771ed689 | 1119 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1120 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1121 | async_cow->inode = igrab(inode); |
771ed689 CM |
1122 | async_cow->root = root; |
1123 | async_cow->locked_page = locked_page; | |
1124 | async_cow->start = start; | |
1125 | ||
f79707b0 WS |
1126 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS && |
1127 | !btrfs_test_opt(root, FORCE_COMPRESS)) | |
771ed689 CM |
1128 | cur_end = end; |
1129 | else | |
1130 | cur_end = min(end, start + 512 * 1024 - 1); | |
1131 | ||
1132 | async_cow->end = cur_end; | |
1133 | INIT_LIST_HEAD(&async_cow->extents); | |
1134 | ||
9e0af237 LB |
1135 | btrfs_init_work(&async_cow->work, |
1136 | btrfs_delalloc_helper, | |
1137 | async_cow_start, async_cow_submit, | |
1138 | async_cow_free); | |
771ed689 | 1139 | |
771ed689 CM |
1140 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
1141 | PAGE_CACHE_SHIFT; | |
1142 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
1143 | ||
afe3d242 QW |
1144 | btrfs_queue_work(root->fs_info->delalloc_workers, |
1145 | &async_cow->work); | |
771ed689 CM |
1146 | |
1147 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1148 | wait_event(root->fs_info->async_submit_wait, | |
1149 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1150 | limit)); | |
1151 | } | |
1152 | ||
d397712b | 1153 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1154 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1155 | wait_event(root->fs_info->async_submit_wait, | |
1156 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1157 | 0)); | |
1158 | } | |
1159 | ||
1160 | *nr_written += nr_pages; | |
1161 | start = cur_end + 1; | |
1162 | } | |
1163 | *page_started = 1; | |
1164 | return 0; | |
be20aa9d CM |
1165 | } |
1166 | ||
d397712b | 1167 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1168 | u64 bytenr, u64 num_bytes) |
1169 | { | |
1170 | int ret; | |
1171 | struct btrfs_ordered_sum *sums; | |
1172 | LIST_HEAD(list); | |
1173 | ||
07d400a6 | 1174 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1175 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1176 | if (ret == 0 && list_empty(&list)) |
1177 | return 0; | |
1178 | ||
1179 | while (!list_empty(&list)) { | |
1180 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1181 | list_del(&sums->list); | |
1182 | kfree(sums); | |
1183 | } | |
1184 | return 1; | |
1185 | } | |
1186 | ||
d352ac68 CM |
1187 | /* |
1188 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1189 | * of the extents that exist in the file, and COWs the file as required. | |
1190 | * | |
1191 | * If no cow copies or snapshots exist, we write directly to the existing | |
1192 | * blocks on disk | |
1193 | */ | |
7f366cfe CM |
1194 | static noinline int run_delalloc_nocow(struct inode *inode, |
1195 | struct page *locked_page, | |
771ed689 CM |
1196 | u64 start, u64 end, int *page_started, int force, |
1197 | unsigned long *nr_written) | |
be20aa9d | 1198 | { |
be20aa9d | 1199 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1200 | struct btrfs_trans_handle *trans; |
be20aa9d | 1201 | struct extent_buffer *leaf; |
be20aa9d | 1202 | struct btrfs_path *path; |
80ff3856 | 1203 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1204 | struct btrfs_key found_key; |
80ff3856 YZ |
1205 | u64 cow_start; |
1206 | u64 cur_offset; | |
1207 | u64 extent_end; | |
5d4f98a2 | 1208 | u64 extent_offset; |
80ff3856 YZ |
1209 | u64 disk_bytenr; |
1210 | u64 num_bytes; | |
b4939680 | 1211 | u64 disk_num_bytes; |
cc95bef6 | 1212 | u64 ram_bytes; |
80ff3856 | 1213 | int extent_type; |
79787eaa | 1214 | int ret, err; |
d899e052 | 1215 | int type; |
80ff3856 YZ |
1216 | int nocow; |
1217 | int check_prev = 1; | |
82d5902d | 1218 | bool nolock; |
33345d01 | 1219 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1220 | |
1221 | path = btrfs_alloc_path(); | |
17ca04af | 1222 | if (!path) { |
c2790a2e JB |
1223 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1224 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1225 | EXTENT_DO_ACCOUNTING | |
1226 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1227 | PAGE_CLEAR_DIRTY | |
1228 | PAGE_SET_WRITEBACK | | |
1229 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1230 | return -ENOMEM; |
17ca04af | 1231 | } |
82d5902d | 1232 | |
83eea1f1 | 1233 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1234 | |
1235 | if (nolock) | |
7a7eaa40 | 1236 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1237 | else |
7a7eaa40 | 1238 | trans = btrfs_join_transaction(root); |
ff5714cc | 1239 | |
79787eaa | 1240 | if (IS_ERR(trans)) { |
c2790a2e JB |
1241 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1242 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1243 | EXTENT_DO_ACCOUNTING | |
1244 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1245 | PAGE_CLEAR_DIRTY | |
1246 | PAGE_SET_WRITEBACK | | |
1247 | PAGE_END_WRITEBACK); | |
79787eaa JM |
1248 | btrfs_free_path(path); |
1249 | return PTR_ERR(trans); | |
1250 | } | |
1251 | ||
74b21075 | 1252 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1253 | |
80ff3856 YZ |
1254 | cow_start = (u64)-1; |
1255 | cur_offset = start; | |
1256 | while (1) { | |
33345d01 | 1257 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1258 | cur_offset, 0); |
d788a349 | 1259 | if (ret < 0) |
79787eaa | 1260 | goto error; |
80ff3856 YZ |
1261 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1262 | leaf = path->nodes[0]; | |
1263 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1264 | path->slots[0] - 1); | |
33345d01 | 1265 | if (found_key.objectid == ino && |
80ff3856 YZ |
1266 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1267 | path->slots[0]--; | |
1268 | } | |
1269 | check_prev = 0; | |
1270 | next_slot: | |
1271 | leaf = path->nodes[0]; | |
1272 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1273 | ret = btrfs_next_leaf(root, path); | |
d788a349 | 1274 | if (ret < 0) |
79787eaa | 1275 | goto error; |
80ff3856 YZ |
1276 | if (ret > 0) |
1277 | break; | |
1278 | leaf = path->nodes[0]; | |
1279 | } | |
be20aa9d | 1280 | |
80ff3856 YZ |
1281 | nocow = 0; |
1282 | disk_bytenr = 0; | |
17d217fe | 1283 | num_bytes = 0; |
80ff3856 YZ |
1284 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1285 | ||
33345d01 | 1286 | if (found_key.objectid > ino || |
80ff3856 YZ |
1287 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1288 | found_key.offset > end) | |
1289 | break; | |
1290 | ||
1291 | if (found_key.offset > cur_offset) { | |
1292 | extent_end = found_key.offset; | |
e9061e21 | 1293 | extent_type = 0; |
80ff3856 YZ |
1294 | goto out_check; |
1295 | } | |
1296 | ||
1297 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1298 | struct btrfs_file_extent_item); | |
1299 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1300 | ||
cc95bef6 | 1301 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1302 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1303 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1304 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1305 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1306 | extent_end = found_key.offset + |
1307 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1308 | disk_num_bytes = |
1309 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1310 | if (extent_end <= start) { |
1311 | path->slots[0]++; | |
1312 | goto next_slot; | |
1313 | } | |
17d217fe YZ |
1314 | if (disk_bytenr == 0) |
1315 | goto out_check; | |
80ff3856 YZ |
1316 | if (btrfs_file_extent_compression(leaf, fi) || |
1317 | btrfs_file_extent_encryption(leaf, fi) || | |
1318 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1319 | goto out_check; | |
d899e052 YZ |
1320 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1321 | goto out_check; | |
d2fb3437 | 1322 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1323 | goto out_check; |
33345d01 | 1324 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1325 | found_key.offset - |
1326 | extent_offset, disk_bytenr)) | |
17d217fe | 1327 | goto out_check; |
5d4f98a2 | 1328 | disk_bytenr += extent_offset; |
17d217fe YZ |
1329 | disk_bytenr += cur_offset - found_key.offset; |
1330 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 WS |
1331 | /* |
1332 | * if there are pending snapshots for this root, | |
1333 | * we fall into common COW way. | |
1334 | */ | |
1335 | if (!nolock) { | |
1336 | err = btrfs_start_nocow_write(root); | |
1337 | if (!err) | |
1338 | goto out_check; | |
1339 | } | |
17d217fe YZ |
1340 | /* |
1341 | * force cow if csum exists in the range. | |
1342 | * this ensure that csum for a given extent are | |
1343 | * either valid or do not exist. | |
1344 | */ | |
1345 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1346 | goto out_check; | |
80ff3856 YZ |
1347 | nocow = 1; |
1348 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1349 | extent_end = found_key.offset + | |
514ac8ad CM |
1350 | btrfs_file_extent_inline_len(leaf, |
1351 | path->slots[0], fi); | |
80ff3856 YZ |
1352 | extent_end = ALIGN(extent_end, root->sectorsize); |
1353 | } else { | |
1354 | BUG_ON(1); | |
1355 | } | |
1356 | out_check: | |
1357 | if (extent_end <= start) { | |
1358 | path->slots[0]++; | |
e9894fd3 WS |
1359 | if (!nolock && nocow) |
1360 | btrfs_end_nocow_write(root); | |
80ff3856 YZ |
1361 | goto next_slot; |
1362 | } | |
1363 | if (!nocow) { | |
1364 | if (cow_start == (u64)-1) | |
1365 | cow_start = cur_offset; | |
1366 | cur_offset = extent_end; | |
1367 | if (cur_offset > end) | |
1368 | break; | |
1369 | path->slots[0]++; | |
1370 | goto next_slot; | |
7ea394f1 YZ |
1371 | } |
1372 | ||
b3b4aa74 | 1373 | btrfs_release_path(path); |
80ff3856 | 1374 | if (cow_start != (u64)-1) { |
00361589 JB |
1375 | ret = cow_file_range(inode, locked_page, |
1376 | cow_start, found_key.offset - 1, | |
1377 | page_started, nr_written, 1); | |
e9894fd3 WS |
1378 | if (ret) { |
1379 | if (!nolock && nocow) | |
1380 | btrfs_end_nocow_write(root); | |
79787eaa | 1381 | goto error; |
e9894fd3 | 1382 | } |
80ff3856 | 1383 | cow_start = (u64)-1; |
7ea394f1 | 1384 | } |
80ff3856 | 1385 | |
d899e052 YZ |
1386 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1387 | struct extent_map *em; | |
1388 | struct extent_map_tree *em_tree; | |
1389 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1390 | em = alloc_extent_map(); |
79787eaa | 1391 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1392 | em->start = cur_offset; |
70c8a91c | 1393 | em->orig_start = found_key.offset - extent_offset; |
d899e052 YZ |
1394 | em->len = num_bytes; |
1395 | em->block_len = num_bytes; | |
1396 | em->block_start = disk_bytenr; | |
b4939680 | 1397 | em->orig_block_len = disk_num_bytes; |
cc95bef6 | 1398 | em->ram_bytes = ram_bytes; |
d899e052 | 1399 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
2ab28f32 JB |
1400 | em->mod_start = em->start; |
1401 | em->mod_len = em->len; | |
d899e052 | 1402 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
b11e234d | 1403 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
70c8a91c | 1404 | em->generation = -1; |
d899e052 | 1405 | while (1) { |
890871be | 1406 | write_lock(&em_tree->lock); |
09a2a8f9 | 1407 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1408 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1409 | if (ret != -EEXIST) { |
1410 | free_extent_map(em); | |
1411 | break; | |
1412 | } | |
1413 | btrfs_drop_extent_cache(inode, em->start, | |
1414 | em->start + em->len - 1, 0); | |
1415 | } | |
1416 | type = BTRFS_ORDERED_PREALLOC; | |
1417 | } else { | |
1418 | type = BTRFS_ORDERED_NOCOW; | |
1419 | } | |
80ff3856 YZ |
1420 | |
1421 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1422 | num_bytes, num_bytes, type); |
79787eaa | 1423 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1424 | |
efa56464 YZ |
1425 | if (root->root_key.objectid == |
1426 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1427 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1428 | num_bytes); | |
e9894fd3 WS |
1429 | if (ret) { |
1430 | if (!nolock && nocow) | |
1431 | btrfs_end_nocow_write(root); | |
79787eaa | 1432 | goto error; |
e9894fd3 | 1433 | } |
efa56464 YZ |
1434 | } |
1435 | ||
c2790a2e JB |
1436 | extent_clear_unlock_delalloc(inode, cur_offset, |
1437 | cur_offset + num_bytes - 1, | |
1438 | locked_page, EXTENT_LOCKED | | |
1439 | EXTENT_DELALLOC, PAGE_UNLOCK | | |
1440 | PAGE_SET_PRIVATE2); | |
e9894fd3 WS |
1441 | if (!nolock && nocow) |
1442 | btrfs_end_nocow_write(root); | |
80ff3856 YZ |
1443 | cur_offset = extent_end; |
1444 | if (cur_offset > end) | |
1445 | break; | |
be20aa9d | 1446 | } |
b3b4aa74 | 1447 | btrfs_release_path(path); |
80ff3856 | 1448 | |
17ca04af | 1449 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1450 | cow_start = cur_offset; |
17ca04af JB |
1451 | cur_offset = end; |
1452 | } | |
1453 | ||
80ff3856 | 1454 | if (cow_start != (u64)-1) { |
00361589 JB |
1455 | ret = cow_file_range(inode, locked_page, cow_start, end, |
1456 | page_started, nr_written, 1); | |
d788a349 | 1457 | if (ret) |
79787eaa | 1458 | goto error; |
80ff3856 YZ |
1459 | } |
1460 | ||
79787eaa | 1461 | error: |
a698d075 | 1462 | err = btrfs_end_transaction(trans, root); |
79787eaa JM |
1463 | if (!ret) |
1464 | ret = err; | |
1465 | ||
17ca04af | 1466 | if (ret && cur_offset < end) |
c2790a2e JB |
1467 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
1468 | locked_page, EXTENT_LOCKED | | |
151a41bc JB |
1469 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1470 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1471 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1472 | PAGE_SET_WRITEBACK | |
1473 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1474 | btrfs_free_path(path); |
79787eaa | 1475 | return ret; |
be20aa9d CM |
1476 | } |
1477 | ||
47059d93 WS |
1478 | static inline int need_force_cow(struct inode *inode, u64 start, u64 end) |
1479 | { | |
1480 | ||
1481 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
1482 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) | |
1483 | return 0; | |
1484 | ||
1485 | /* | |
1486 | * @defrag_bytes is a hint value, no spinlock held here, | |
1487 | * if is not zero, it means the file is defragging. | |
1488 | * Force cow if given extent needs to be defragged. | |
1489 | */ | |
1490 | if (BTRFS_I(inode)->defrag_bytes && | |
1491 | test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1492 | EXTENT_DEFRAG, 0, NULL)) | |
1493 | return 1; | |
1494 | ||
1495 | return 0; | |
1496 | } | |
1497 | ||
d352ac68 CM |
1498 | /* |
1499 | * extent_io.c call back to do delayed allocation processing | |
1500 | */ | |
c8b97818 | 1501 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1502 | u64 start, u64 end, int *page_started, |
1503 | unsigned long *nr_written) | |
be20aa9d | 1504 | { |
be20aa9d | 1505 | int ret; |
47059d93 | 1506 | int force_cow = need_force_cow(inode, start, end); |
a2135011 | 1507 | |
47059d93 | 1508 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
c8b97818 | 1509 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1510 | page_started, 1, nr_written); |
47059d93 | 1511 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
d899e052 | 1512 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1513 | page_started, 0, nr_written); |
7816030e | 1514 | } else if (!inode_need_compress(inode)) { |
7f366cfe CM |
1515 | ret = cow_file_range(inode, locked_page, start, end, |
1516 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1517 | } else { |
1518 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1519 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1520 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1521 | page_started, nr_written); |
7ddf5a42 | 1522 | } |
b888db2b CM |
1523 | return ret; |
1524 | } | |
1525 | ||
1bf85046 JM |
1526 | static void btrfs_split_extent_hook(struct inode *inode, |
1527 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1528 | { |
0ca1f7ce | 1529 | /* not delalloc, ignore it */ |
9ed74f2d | 1530 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1531 | return; |
9ed74f2d | 1532 | |
9e0baf60 JB |
1533 | spin_lock(&BTRFS_I(inode)->lock); |
1534 | BTRFS_I(inode)->outstanding_extents++; | |
1535 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1536 | } |
1537 | ||
1538 | /* | |
1539 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1540 | * extents so we can keep track of new extents that are just merged onto old | |
1541 | * extents, such as when we are doing sequential writes, so we can properly | |
1542 | * account for the metadata space we'll need. | |
1543 | */ | |
1bf85046 JM |
1544 | static void btrfs_merge_extent_hook(struct inode *inode, |
1545 | struct extent_state *new, | |
1546 | struct extent_state *other) | |
9ed74f2d | 1547 | { |
9ed74f2d JB |
1548 | /* not delalloc, ignore it */ |
1549 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1550 | return; |
9ed74f2d | 1551 | |
9e0baf60 JB |
1552 | spin_lock(&BTRFS_I(inode)->lock); |
1553 | BTRFS_I(inode)->outstanding_extents--; | |
1554 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1555 | } |
1556 | ||
eb73c1b7 MX |
1557 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1558 | struct inode *inode) | |
1559 | { | |
1560 | spin_lock(&root->delalloc_lock); | |
1561 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1562 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1563 | &root->delalloc_inodes); | |
1564 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1565 | &BTRFS_I(inode)->runtime_flags); | |
1566 | root->nr_delalloc_inodes++; | |
1567 | if (root->nr_delalloc_inodes == 1) { | |
1568 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1569 | BUG_ON(!list_empty(&root->delalloc_root)); | |
1570 | list_add_tail(&root->delalloc_root, | |
1571 | &root->fs_info->delalloc_roots); | |
1572 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1573 | } | |
1574 | } | |
1575 | spin_unlock(&root->delalloc_lock); | |
1576 | } | |
1577 | ||
1578 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1579 | struct inode *inode) | |
1580 | { | |
1581 | spin_lock(&root->delalloc_lock); | |
1582 | if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1583 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1584 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1585 | &BTRFS_I(inode)->runtime_flags); | |
1586 | root->nr_delalloc_inodes--; | |
1587 | if (!root->nr_delalloc_inodes) { | |
1588 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1589 | BUG_ON(list_empty(&root->delalloc_root)); | |
1590 | list_del_init(&root->delalloc_root); | |
1591 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1592 | } | |
1593 | } | |
1594 | spin_unlock(&root->delalloc_lock); | |
1595 | } | |
1596 | ||
d352ac68 CM |
1597 | /* |
1598 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1599 | * bytes in this file, and to maintain the list of inodes that | |
1600 | * have pending delalloc work to be done. | |
1601 | */ | |
1bf85046 | 1602 | static void btrfs_set_bit_hook(struct inode *inode, |
41074888 | 1603 | struct extent_state *state, unsigned long *bits) |
291d673e | 1604 | { |
9ed74f2d | 1605 | |
47059d93 WS |
1606 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1607 | WARN_ON(1); | |
75eff68e CM |
1608 | /* |
1609 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1610 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1611 | * bit, which is only set or cleared with irqs on |
1612 | */ | |
0ca1f7ce | 1613 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1614 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1615 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1616 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1617 | |
9e0baf60 | 1618 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1619 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1620 | } else { |
1621 | spin_lock(&BTRFS_I(inode)->lock); | |
1622 | BTRFS_I(inode)->outstanding_extents++; | |
1623 | spin_unlock(&BTRFS_I(inode)->lock); | |
1624 | } | |
287a0ab9 | 1625 | |
963d678b MX |
1626 | __percpu_counter_add(&root->fs_info->delalloc_bytes, len, |
1627 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1628 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1629 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
1630 | if (*bits & EXTENT_DEFRAG) |
1631 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 1632 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1633 | &BTRFS_I(inode)->runtime_flags)) |
1634 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1635 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1636 | } |
291d673e CM |
1637 | } |
1638 | ||
d352ac68 CM |
1639 | /* |
1640 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1641 | */ | |
1bf85046 | 1642 | static void btrfs_clear_bit_hook(struct inode *inode, |
41074888 DS |
1643 | struct extent_state *state, |
1644 | unsigned long *bits) | |
291d673e | 1645 | { |
47059d93 WS |
1646 | u64 len = state->end + 1 - state->start; |
1647 | ||
1648 | spin_lock(&BTRFS_I(inode)->lock); | |
1649 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) | |
1650 | BTRFS_I(inode)->defrag_bytes -= len; | |
1651 | spin_unlock(&BTRFS_I(inode)->lock); | |
1652 | ||
75eff68e CM |
1653 | /* |
1654 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1655 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1656 | * bit, which is only set or cleared with irqs on |
1657 | */ | |
0ca1f7ce | 1658 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1659 | struct btrfs_root *root = BTRFS_I(inode)->root; |
83eea1f1 | 1660 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1661 | |
9e0baf60 | 1662 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1663 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1664 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1665 | spin_lock(&BTRFS_I(inode)->lock); | |
1666 | BTRFS_I(inode)->outstanding_extents--; | |
1667 | spin_unlock(&BTRFS_I(inode)->lock); | |
1668 | } | |
0ca1f7ce | 1669 | |
b6d08f06 JB |
1670 | /* |
1671 | * We don't reserve metadata space for space cache inodes so we | |
1672 | * don't need to call dellalloc_release_metadata if there is an | |
1673 | * error. | |
1674 | */ | |
1675 | if (*bits & EXTENT_DO_ACCOUNTING && | |
1676 | root != root->fs_info->tree_root) | |
0ca1f7ce YZ |
1677 | btrfs_delalloc_release_metadata(inode, len); |
1678 | ||
0cb59c99 | 1679 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
7ee9e440 | 1680 | && do_list && !(state->state & EXTENT_NORESERVE)) |
0ca1f7ce | 1681 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1682 | |
963d678b MX |
1683 | __percpu_counter_add(&root->fs_info->delalloc_bytes, -len, |
1684 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1685 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1686 | BTRFS_I(inode)->delalloc_bytes -= len; |
0cb59c99 | 1687 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
df0af1a5 | 1688 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1689 | &BTRFS_I(inode)->runtime_flags)) |
1690 | btrfs_del_delalloc_inode(root, inode); | |
df0af1a5 | 1691 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1692 | } |
291d673e CM |
1693 | } |
1694 | ||
d352ac68 CM |
1695 | /* |
1696 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1697 | * we don't create bios that span stripes or chunks | |
1698 | */ | |
64a16701 | 1699 | int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset, |
c8b97818 CM |
1700 | size_t size, struct bio *bio, |
1701 | unsigned long bio_flags) | |
239b14b3 CM |
1702 | { |
1703 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
4f024f37 | 1704 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
1705 | u64 length = 0; |
1706 | u64 map_length; | |
239b14b3 CM |
1707 | int ret; |
1708 | ||
771ed689 CM |
1709 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1710 | return 0; | |
1711 | ||
4f024f37 | 1712 | length = bio->bi_iter.bi_size; |
239b14b3 | 1713 | map_length = length; |
64a16701 | 1714 | ret = btrfs_map_block(root->fs_info, rw, logical, |
f188591e | 1715 | &map_length, NULL, 0); |
3ec706c8 | 1716 | /* Will always return 0 with map_multi == NULL */ |
3444a972 | 1717 | BUG_ON(ret < 0); |
d397712b | 1718 | if (map_length < length + size) |
239b14b3 | 1719 | return 1; |
3444a972 | 1720 | return 0; |
239b14b3 CM |
1721 | } |
1722 | ||
d352ac68 CM |
1723 | /* |
1724 | * in order to insert checksums into the metadata in large chunks, | |
1725 | * we wait until bio submission time. All the pages in the bio are | |
1726 | * checksummed and sums are attached onto the ordered extent record. | |
1727 | * | |
1728 | * At IO completion time the cums attached on the ordered extent record | |
1729 | * are inserted into the btree | |
1730 | */ | |
d397712b CM |
1731 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1732 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1733 | unsigned long bio_flags, |
1734 | u64 bio_offset) | |
065631f6 | 1735 | { |
065631f6 | 1736 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1737 | int ret = 0; |
e015640f | 1738 | |
d20f7043 | 1739 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1740 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1741 | return 0; |
1742 | } | |
e015640f | 1743 | |
4a69a410 CM |
1744 | /* |
1745 | * in order to insert checksums into the metadata in large chunks, | |
1746 | * we wait until bio submission time. All the pages in the bio are | |
1747 | * checksummed and sums are attached onto the ordered extent record. | |
1748 | * | |
1749 | * At IO completion time the cums attached on the ordered extent record | |
1750 | * are inserted into the btree | |
1751 | */ | |
b2950863 | 1752 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1753 | int mirror_num, unsigned long bio_flags, |
1754 | u64 bio_offset) | |
4a69a410 CM |
1755 | { |
1756 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
61891923 SB |
1757 | int ret; |
1758 | ||
1759 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 1); | |
1760 | if (ret) | |
1761 | bio_endio(bio, ret); | |
1762 | return ret; | |
44b8bd7e CM |
1763 | } |
1764 | ||
d352ac68 | 1765 | /* |
cad321ad CM |
1766 | * extent_io.c submission hook. This does the right thing for csum calculation |
1767 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1768 | */ |
b2950863 | 1769 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1770 | int mirror_num, unsigned long bio_flags, |
1771 | u64 bio_offset) | |
44b8bd7e CM |
1772 | { |
1773 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1774 | int ret = 0; | |
19b9bdb0 | 1775 | int skip_sum; |
0417341e | 1776 | int metadata = 0; |
b812ce28 | 1777 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1778 | |
6cbff00f | 1779 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1780 | |
83eea1f1 | 1781 | if (btrfs_is_free_space_inode(inode)) |
0417341e JM |
1782 | metadata = 2; |
1783 | ||
7b6d91da | 1784 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1785 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1786 | if (ret) | |
61891923 | 1787 | goto out; |
5fd02043 | 1788 | |
d20f7043 | 1789 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1790 | ret = btrfs_submit_compressed_read(inode, bio, |
1791 | mirror_num, | |
1792 | bio_flags); | |
1793 | goto out; | |
c2db1073 TI |
1794 | } else if (!skip_sum) { |
1795 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1796 | if (ret) | |
61891923 | 1797 | goto out; |
c2db1073 | 1798 | } |
4d1b5fb4 | 1799 | goto mapit; |
b812ce28 | 1800 | } else if (async && !skip_sum) { |
17d217fe YZ |
1801 | /* csum items have already been cloned */ |
1802 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1803 | goto mapit; | |
19b9bdb0 | 1804 | /* we're doing a write, do the async checksumming */ |
61891923 | 1805 | ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, |
44b8bd7e | 1806 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1807 | bio_flags, bio_offset, |
1808 | __btrfs_submit_bio_start, | |
4a69a410 | 1809 | __btrfs_submit_bio_done); |
61891923 | 1810 | goto out; |
b812ce28 JB |
1811 | } else if (!skip_sum) { |
1812 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); | |
1813 | if (ret) | |
1814 | goto out; | |
19b9bdb0 CM |
1815 | } |
1816 | ||
0b86a832 | 1817 | mapit: |
61891923 SB |
1818 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); |
1819 | ||
1820 | out: | |
1821 | if (ret < 0) | |
1822 | bio_endio(bio, ret); | |
1823 | return ret; | |
065631f6 | 1824 | } |
6885f308 | 1825 | |
d352ac68 CM |
1826 | /* |
1827 | * given a list of ordered sums record them in the inode. This happens | |
1828 | * at IO completion time based on sums calculated at bio submission time. | |
1829 | */ | |
ba1da2f4 | 1830 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1831 | struct inode *inode, u64 file_offset, |
1832 | struct list_head *list) | |
1833 | { | |
e6dcd2dc CM |
1834 | struct btrfs_ordered_sum *sum; |
1835 | ||
c6e30871 | 1836 | list_for_each_entry(sum, list, list) { |
39847c4d | 1837 | trans->adding_csums = 1; |
d20f7043 CM |
1838 | btrfs_csum_file_blocks(trans, |
1839 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 1840 | trans->adding_csums = 0; |
e6dcd2dc CM |
1841 | } |
1842 | return 0; | |
1843 | } | |
1844 | ||
2ac55d41 JB |
1845 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1846 | struct extent_state **cached_state) | |
ea8c2819 | 1847 | { |
6c1500f2 | 1848 | WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0); |
ea8c2819 | 1849 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1850 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1851 | } |
1852 | ||
d352ac68 | 1853 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1854 | struct btrfs_writepage_fixup { |
1855 | struct page *page; | |
1856 | struct btrfs_work work; | |
1857 | }; | |
1858 | ||
b2950863 | 1859 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1860 | { |
1861 | struct btrfs_writepage_fixup *fixup; | |
1862 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1863 | struct extent_state *cached_state = NULL; |
247e743c CM |
1864 | struct page *page; |
1865 | struct inode *inode; | |
1866 | u64 page_start; | |
1867 | u64 page_end; | |
87826df0 | 1868 | int ret; |
247e743c CM |
1869 | |
1870 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1871 | page = fixup->page; | |
4a096752 | 1872 | again: |
247e743c CM |
1873 | lock_page(page); |
1874 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1875 | ClearPageChecked(page); | |
1876 | goto out_page; | |
1877 | } | |
1878 | ||
1879 | inode = page->mapping->host; | |
1880 | page_start = page_offset(page); | |
1881 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1882 | ||
2ac55d41 | 1883 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
d0082371 | 1884 | &cached_state); |
4a096752 CM |
1885 | |
1886 | /* already ordered? We're done */ | |
8b62b72b | 1887 | if (PagePrivate2(page)) |
247e743c | 1888 | goto out; |
4a096752 CM |
1889 | |
1890 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1891 | if (ordered) { | |
2ac55d41 JB |
1892 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1893 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1894 | unlock_page(page); |
1895 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1896 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1897 | goto again; |
1898 | } | |
247e743c | 1899 | |
87826df0 JM |
1900 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1901 | if (ret) { | |
1902 | mapping_set_error(page->mapping, ret); | |
1903 | end_extent_writepage(page, ret, page_start, page_end); | |
1904 | ClearPageChecked(page); | |
1905 | goto out; | |
1906 | } | |
1907 | ||
2ac55d41 | 1908 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 1909 | ClearPageChecked(page); |
87826df0 | 1910 | set_page_dirty(page); |
247e743c | 1911 | out: |
2ac55d41 JB |
1912 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1913 | &cached_state, GFP_NOFS); | |
247e743c CM |
1914 | out_page: |
1915 | unlock_page(page); | |
1916 | page_cache_release(page); | |
b897abec | 1917 | kfree(fixup); |
247e743c CM |
1918 | } |
1919 | ||
1920 | /* | |
1921 | * There are a few paths in the higher layers of the kernel that directly | |
1922 | * set the page dirty bit without asking the filesystem if it is a | |
1923 | * good idea. This causes problems because we want to make sure COW | |
1924 | * properly happens and the data=ordered rules are followed. | |
1925 | * | |
c8b97818 | 1926 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1927 | * hasn't been properly setup for IO. We kick off an async process |
1928 | * to fix it up. The async helper will wait for ordered extents, set | |
1929 | * the delalloc bit and make it safe to write the page. | |
1930 | */ | |
b2950863 | 1931 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1932 | { |
1933 | struct inode *inode = page->mapping->host; | |
1934 | struct btrfs_writepage_fixup *fixup; | |
1935 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1936 | |
8b62b72b CM |
1937 | /* this page is properly in the ordered list */ |
1938 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1939 | return 0; |
1940 | ||
1941 | if (PageChecked(page)) | |
1942 | return -EAGAIN; | |
1943 | ||
1944 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1945 | if (!fixup) | |
1946 | return -EAGAIN; | |
f421950f | 1947 | |
247e743c CM |
1948 | SetPageChecked(page); |
1949 | page_cache_get(page); | |
9e0af237 LB |
1950 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
1951 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 1952 | fixup->page = page; |
dc6e3209 | 1953 | btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work); |
87826df0 | 1954 | return -EBUSY; |
247e743c CM |
1955 | } |
1956 | ||
d899e052 YZ |
1957 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1958 | struct inode *inode, u64 file_pos, | |
1959 | u64 disk_bytenr, u64 disk_num_bytes, | |
1960 | u64 num_bytes, u64 ram_bytes, | |
1961 | u8 compression, u8 encryption, | |
1962 | u16 other_encoding, int extent_type) | |
1963 | { | |
1964 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1965 | struct btrfs_file_extent_item *fi; | |
1966 | struct btrfs_path *path; | |
1967 | struct extent_buffer *leaf; | |
1968 | struct btrfs_key ins; | |
1acae57b | 1969 | int extent_inserted = 0; |
d899e052 YZ |
1970 | int ret; |
1971 | ||
1972 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1973 | if (!path) |
1974 | return -ENOMEM; | |
d899e052 | 1975 | |
a1ed835e CM |
1976 | /* |
1977 | * we may be replacing one extent in the tree with another. | |
1978 | * The new extent is pinned in the extent map, and we don't want | |
1979 | * to drop it from the cache until it is completely in the btree. | |
1980 | * | |
1981 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1982 | * the caller is expected to unpin it and allow it to be merged | |
1983 | * with the others. | |
1984 | */ | |
1acae57b FDBM |
1985 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
1986 | file_pos + num_bytes, NULL, 0, | |
1987 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
1988 | if (ret) |
1989 | goto out; | |
d899e052 | 1990 | |
1acae57b FDBM |
1991 | if (!extent_inserted) { |
1992 | ins.objectid = btrfs_ino(inode); | |
1993 | ins.offset = file_pos; | |
1994 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1995 | ||
1996 | path->leave_spinning = 1; | |
1997 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
1998 | sizeof(*fi)); | |
1999 | if (ret) | |
2000 | goto out; | |
2001 | } | |
d899e052 YZ |
2002 | leaf = path->nodes[0]; |
2003 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2004 | struct btrfs_file_extent_item); | |
2005 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
2006 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
2007 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
2008 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
2009 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2010 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2011 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2012 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2013 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2014 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2015 | |
d899e052 | 2016 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2017 | btrfs_release_path(path); |
d899e052 YZ |
2018 | |
2019 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2020 | |
2021 | ins.objectid = disk_bytenr; | |
2022 | ins.offset = disk_num_bytes; | |
2023 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
2024 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
2025 | root->root_key.objectid, | |
33345d01 | 2026 | btrfs_ino(inode), file_pos, &ins); |
79787eaa | 2027 | out: |
d899e052 | 2028 | btrfs_free_path(path); |
b9473439 | 2029 | |
79787eaa | 2030 | return ret; |
d899e052 YZ |
2031 | } |
2032 | ||
38c227d8 LB |
2033 | /* snapshot-aware defrag */ |
2034 | struct sa_defrag_extent_backref { | |
2035 | struct rb_node node; | |
2036 | struct old_sa_defrag_extent *old; | |
2037 | u64 root_id; | |
2038 | u64 inum; | |
2039 | u64 file_pos; | |
2040 | u64 extent_offset; | |
2041 | u64 num_bytes; | |
2042 | u64 generation; | |
2043 | }; | |
2044 | ||
2045 | struct old_sa_defrag_extent { | |
2046 | struct list_head list; | |
2047 | struct new_sa_defrag_extent *new; | |
2048 | ||
2049 | u64 extent_offset; | |
2050 | u64 bytenr; | |
2051 | u64 offset; | |
2052 | u64 len; | |
2053 | int count; | |
2054 | }; | |
2055 | ||
2056 | struct new_sa_defrag_extent { | |
2057 | struct rb_root root; | |
2058 | struct list_head head; | |
2059 | struct btrfs_path *path; | |
2060 | struct inode *inode; | |
2061 | u64 file_pos; | |
2062 | u64 len; | |
2063 | u64 bytenr; | |
2064 | u64 disk_len; | |
2065 | u8 compress_type; | |
2066 | }; | |
2067 | ||
2068 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2069 | struct sa_defrag_extent_backref *b2) | |
2070 | { | |
2071 | if (b1->root_id < b2->root_id) | |
2072 | return -1; | |
2073 | else if (b1->root_id > b2->root_id) | |
2074 | return 1; | |
2075 | ||
2076 | if (b1->inum < b2->inum) | |
2077 | return -1; | |
2078 | else if (b1->inum > b2->inum) | |
2079 | return 1; | |
2080 | ||
2081 | if (b1->file_pos < b2->file_pos) | |
2082 | return -1; | |
2083 | else if (b1->file_pos > b2->file_pos) | |
2084 | return 1; | |
2085 | ||
2086 | /* | |
2087 | * [------------------------------] ===> (a range of space) | |
2088 | * |<--->| |<---->| =============> (fs/file tree A) | |
2089 | * |<---------------------------->| ===> (fs/file tree B) | |
2090 | * | |
2091 | * A range of space can refer to two file extents in one tree while | |
2092 | * refer to only one file extent in another tree. | |
2093 | * | |
2094 | * So we may process a disk offset more than one time(two extents in A) | |
2095 | * and locate at the same extent(one extent in B), then insert two same | |
2096 | * backrefs(both refer to the extent in B). | |
2097 | */ | |
2098 | return 0; | |
2099 | } | |
2100 | ||
2101 | static void backref_insert(struct rb_root *root, | |
2102 | struct sa_defrag_extent_backref *backref) | |
2103 | { | |
2104 | struct rb_node **p = &root->rb_node; | |
2105 | struct rb_node *parent = NULL; | |
2106 | struct sa_defrag_extent_backref *entry; | |
2107 | int ret; | |
2108 | ||
2109 | while (*p) { | |
2110 | parent = *p; | |
2111 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2112 | ||
2113 | ret = backref_comp(backref, entry); | |
2114 | if (ret < 0) | |
2115 | p = &(*p)->rb_left; | |
2116 | else | |
2117 | p = &(*p)->rb_right; | |
2118 | } | |
2119 | ||
2120 | rb_link_node(&backref->node, parent, p); | |
2121 | rb_insert_color(&backref->node, root); | |
2122 | } | |
2123 | ||
2124 | /* | |
2125 | * Note the backref might has changed, and in this case we just return 0. | |
2126 | */ | |
2127 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2128 | void *ctx) | |
2129 | { | |
2130 | struct btrfs_file_extent_item *extent; | |
2131 | struct btrfs_fs_info *fs_info; | |
2132 | struct old_sa_defrag_extent *old = ctx; | |
2133 | struct new_sa_defrag_extent *new = old->new; | |
2134 | struct btrfs_path *path = new->path; | |
2135 | struct btrfs_key key; | |
2136 | struct btrfs_root *root; | |
2137 | struct sa_defrag_extent_backref *backref; | |
2138 | struct extent_buffer *leaf; | |
2139 | struct inode *inode = new->inode; | |
2140 | int slot; | |
2141 | int ret; | |
2142 | u64 extent_offset; | |
2143 | u64 num_bytes; | |
2144 | ||
2145 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
2146 | inum == btrfs_ino(inode)) | |
2147 | return 0; | |
2148 | ||
2149 | key.objectid = root_id; | |
2150 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2151 | key.offset = (u64)-1; | |
2152 | ||
2153 | fs_info = BTRFS_I(inode)->root->fs_info; | |
2154 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2155 | if (IS_ERR(root)) { | |
2156 | if (PTR_ERR(root) == -ENOENT) | |
2157 | return 0; | |
2158 | WARN_ON(1); | |
2159 | pr_debug("inum=%llu, offset=%llu, root_id=%llu\n", | |
2160 | inum, offset, root_id); | |
2161 | return PTR_ERR(root); | |
2162 | } | |
2163 | ||
2164 | key.objectid = inum; | |
2165 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2166 | if (offset > (u64)-1 << 32) | |
2167 | key.offset = 0; | |
2168 | else | |
2169 | key.offset = offset; | |
2170 | ||
2171 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2172 | if (WARN_ON(ret < 0)) |
38c227d8 | 2173 | return ret; |
50f1319c | 2174 | ret = 0; |
38c227d8 LB |
2175 | |
2176 | while (1) { | |
2177 | cond_resched(); | |
2178 | ||
2179 | leaf = path->nodes[0]; | |
2180 | slot = path->slots[0]; | |
2181 | ||
2182 | if (slot >= btrfs_header_nritems(leaf)) { | |
2183 | ret = btrfs_next_leaf(root, path); | |
2184 | if (ret < 0) { | |
2185 | goto out; | |
2186 | } else if (ret > 0) { | |
2187 | ret = 0; | |
2188 | goto out; | |
2189 | } | |
2190 | continue; | |
2191 | } | |
2192 | ||
2193 | path->slots[0]++; | |
2194 | ||
2195 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2196 | ||
2197 | if (key.objectid > inum) | |
2198 | goto out; | |
2199 | ||
2200 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2201 | continue; | |
2202 | ||
2203 | extent = btrfs_item_ptr(leaf, slot, | |
2204 | struct btrfs_file_extent_item); | |
2205 | ||
2206 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2207 | continue; | |
2208 | ||
e68afa49 LB |
2209 | /* |
2210 | * 'offset' refers to the exact key.offset, | |
2211 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2212 | * (key.offset - extent_offset). | |
2213 | */ | |
2214 | if (key.offset != offset) | |
38c227d8 LB |
2215 | continue; |
2216 | ||
e68afa49 | 2217 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2218 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2219 | |
38c227d8 LB |
2220 | if (extent_offset >= old->extent_offset + old->offset + |
2221 | old->len || extent_offset + num_bytes <= | |
2222 | old->extent_offset + old->offset) | |
2223 | continue; | |
38c227d8 LB |
2224 | break; |
2225 | } | |
2226 | ||
2227 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2228 | if (!backref) { | |
2229 | ret = -ENOENT; | |
2230 | goto out; | |
2231 | } | |
2232 | ||
2233 | backref->root_id = root_id; | |
2234 | backref->inum = inum; | |
e68afa49 | 2235 | backref->file_pos = offset; |
38c227d8 LB |
2236 | backref->num_bytes = num_bytes; |
2237 | backref->extent_offset = extent_offset; | |
2238 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2239 | backref->old = old; | |
2240 | backref_insert(&new->root, backref); | |
2241 | old->count++; | |
2242 | out: | |
2243 | btrfs_release_path(path); | |
2244 | WARN_ON(ret); | |
2245 | return ret; | |
2246 | } | |
2247 | ||
2248 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2249 | struct new_sa_defrag_extent *new) | |
2250 | { | |
2251 | struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info; | |
2252 | struct old_sa_defrag_extent *old, *tmp; | |
2253 | int ret; | |
2254 | ||
2255 | new->path = path; | |
2256 | ||
2257 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2258 | ret = iterate_inodes_from_logical(old->bytenr + |
2259 | old->extent_offset, fs_info, | |
38c227d8 LB |
2260 | path, record_one_backref, |
2261 | old); | |
4724b106 JB |
2262 | if (ret < 0 && ret != -ENOENT) |
2263 | return false; | |
38c227d8 LB |
2264 | |
2265 | /* no backref to be processed for this extent */ | |
2266 | if (!old->count) { | |
2267 | list_del(&old->list); | |
2268 | kfree(old); | |
2269 | } | |
2270 | } | |
2271 | ||
2272 | if (list_empty(&new->head)) | |
2273 | return false; | |
2274 | ||
2275 | return true; | |
2276 | } | |
2277 | ||
2278 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2279 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2280 | struct new_sa_defrag_extent *new) |
38c227d8 | 2281 | { |
116e0024 | 2282 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2283 | return 0; |
2284 | ||
2285 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2286 | return 0; | |
2287 | ||
116e0024 LB |
2288 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2289 | return 0; | |
2290 | ||
2291 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2292 | btrfs_file_extent_other_encoding(leaf, fi)) |
2293 | return 0; | |
2294 | ||
2295 | return 1; | |
2296 | } | |
2297 | ||
2298 | /* | |
2299 | * Note the backref might has changed, and in this case we just return 0. | |
2300 | */ | |
2301 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2302 | struct sa_defrag_extent_backref *prev, | |
2303 | struct sa_defrag_extent_backref *backref) | |
2304 | { | |
2305 | struct btrfs_file_extent_item *extent; | |
2306 | struct btrfs_file_extent_item *item; | |
2307 | struct btrfs_ordered_extent *ordered; | |
2308 | struct btrfs_trans_handle *trans; | |
2309 | struct btrfs_fs_info *fs_info; | |
2310 | struct btrfs_root *root; | |
2311 | struct btrfs_key key; | |
2312 | struct extent_buffer *leaf; | |
2313 | struct old_sa_defrag_extent *old = backref->old; | |
2314 | struct new_sa_defrag_extent *new = old->new; | |
2315 | struct inode *src_inode = new->inode; | |
2316 | struct inode *inode; | |
2317 | struct extent_state *cached = NULL; | |
2318 | int ret = 0; | |
2319 | u64 start; | |
2320 | u64 len; | |
2321 | u64 lock_start; | |
2322 | u64 lock_end; | |
2323 | bool merge = false; | |
2324 | int index; | |
2325 | ||
2326 | if (prev && prev->root_id == backref->root_id && | |
2327 | prev->inum == backref->inum && | |
2328 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2329 | merge = true; | |
2330 | ||
2331 | /* step 1: get root */ | |
2332 | key.objectid = backref->root_id; | |
2333 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2334 | key.offset = (u64)-1; | |
2335 | ||
2336 | fs_info = BTRFS_I(src_inode)->root->fs_info; | |
2337 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
2338 | ||
2339 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2340 | if (IS_ERR(root)) { | |
2341 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2342 | if (PTR_ERR(root) == -ENOENT) | |
2343 | return 0; | |
2344 | return PTR_ERR(root); | |
2345 | } | |
38c227d8 | 2346 | |
bcbba5e6 WS |
2347 | if (btrfs_root_readonly(root)) { |
2348 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2349 | return 0; | |
2350 | } | |
2351 | ||
38c227d8 LB |
2352 | /* step 2: get inode */ |
2353 | key.objectid = backref->inum; | |
2354 | key.type = BTRFS_INODE_ITEM_KEY; | |
2355 | key.offset = 0; | |
2356 | ||
2357 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2358 | if (IS_ERR(inode)) { | |
2359 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2360 | return 0; | |
2361 | } | |
2362 | ||
2363 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2364 | ||
2365 | /* step 3: relink backref */ | |
2366 | lock_start = backref->file_pos; | |
2367 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2368 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2369 | 0, &cached); | |
2370 | ||
2371 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2372 | if (ordered) { | |
2373 | btrfs_put_ordered_extent(ordered); | |
2374 | goto out_unlock; | |
2375 | } | |
2376 | ||
2377 | trans = btrfs_join_transaction(root); | |
2378 | if (IS_ERR(trans)) { | |
2379 | ret = PTR_ERR(trans); | |
2380 | goto out_unlock; | |
2381 | } | |
2382 | ||
2383 | key.objectid = backref->inum; | |
2384 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2385 | key.offset = backref->file_pos; | |
2386 | ||
2387 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2388 | if (ret < 0) { | |
2389 | goto out_free_path; | |
2390 | } else if (ret > 0) { | |
2391 | ret = 0; | |
2392 | goto out_free_path; | |
2393 | } | |
2394 | ||
2395 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2396 | struct btrfs_file_extent_item); | |
2397 | ||
2398 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2399 | backref->generation) | |
2400 | goto out_free_path; | |
2401 | ||
2402 | btrfs_release_path(path); | |
2403 | ||
2404 | start = backref->file_pos; | |
2405 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2406 | start += old->extent_offset + old->offset - | |
2407 | backref->extent_offset; | |
2408 | ||
2409 | len = min(backref->extent_offset + backref->num_bytes, | |
2410 | old->extent_offset + old->offset + old->len); | |
2411 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2412 | ||
2413 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2414 | start + len, 1); | |
2415 | if (ret) | |
2416 | goto out_free_path; | |
2417 | again: | |
2418 | key.objectid = btrfs_ino(inode); | |
2419 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2420 | key.offset = start; | |
2421 | ||
a09a0a70 | 2422 | path->leave_spinning = 1; |
38c227d8 LB |
2423 | if (merge) { |
2424 | struct btrfs_file_extent_item *fi; | |
2425 | u64 extent_len; | |
2426 | struct btrfs_key found_key; | |
2427 | ||
3c9665df | 2428 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2429 | if (ret < 0) |
2430 | goto out_free_path; | |
2431 | ||
2432 | path->slots[0]--; | |
2433 | leaf = path->nodes[0]; | |
2434 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2435 | ||
2436 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2437 | struct btrfs_file_extent_item); | |
2438 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2439 | ||
116e0024 LB |
2440 | if (extent_len + found_key.offset == start && |
2441 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2442 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2443 | extent_len + len); | |
2444 | btrfs_mark_buffer_dirty(leaf); | |
2445 | inode_add_bytes(inode, len); | |
2446 | ||
2447 | ret = 1; | |
2448 | goto out_free_path; | |
2449 | } else { | |
2450 | merge = false; | |
2451 | btrfs_release_path(path); | |
2452 | goto again; | |
2453 | } | |
2454 | } | |
2455 | ||
2456 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2457 | sizeof(*extent)); | |
2458 | if (ret) { | |
2459 | btrfs_abort_transaction(trans, root, ret); | |
2460 | goto out_free_path; | |
2461 | } | |
2462 | ||
2463 | leaf = path->nodes[0]; | |
2464 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2465 | struct btrfs_file_extent_item); | |
2466 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2467 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2468 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2469 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2470 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2471 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2472 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2473 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2474 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2475 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2476 | ||
2477 | btrfs_mark_buffer_dirty(leaf); | |
2478 | inode_add_bytes(inode, len); | |
a09a0a70 | 2479 | btrfs_release_path(path); |
38c227d8 LB |
2480 | |
2481 | ret = btrfs_inc_extent_ref(trans, root, new->bytenr, | |
2482 | new->disk_len, 0, | |
2483 | backref->root_id, backref->inum, | |
2484 | new->file_pos, 0); /* start - extent_offset */ | |
2485 | if (ret) { | |
2486 | btrfs_abort_transaction(trans, root, ret); | |
2487 | goto out_free_path; | |
2488 | } | |
2489 | ||
2490 | ret = 1; | |
2491 | out_free_path: | |
2492 | btrfs_release_path(path); | |
a09a0a70 | 2493 | path->leave_spinning = 0; |
38c227d8 LB |
2494 | btrfs_end_transaction(trans, root); |
2495 | out_unlock: | |
2496 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2497 | &cached, GFP_NOFS); | |
2498 | iput(inode); | |
2499 | return ret; | |
2500 | } | |
2501 | ||
6f519564 LB |
2502 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2503 | { | |
2504 | struct old_sa_defrag_extent *old, *tmp; | |
2505 | ||
2506 | if (!new) | |
2507 | return; | |
2508 | ||
2509 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
2510 | list_del(&old->list); | |
2511 | kfree(old); | |
2512 | } | |
2513 | kfree(new); | |
2514 | } | |
2515 | ||
38c227d8 LB |
2516 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2517 | { | |
2518 | struct btrfs_path *path; | |
38c227d8 LB |
2519 | struct sa_defrag_extent_backref *backref; |
2520 | struct sa_defrag_extent_backref *prev = NULL; | |
2521 | struct inode *inode; | |
2522 | struct btrfs_root *root; | |
2523 | struct rb_node *node; | |
2524 | int ret; | |
2525 | ||
2526 | inode = new->inode; | |
2527 | root = BTRFS_I(inode)->root; | |
2528 | ||
2529 | path = btrfs_alloc_path(); | |
2530 | if (!path) | |
2531 | return; | |
2532 | ||
2533 | if (!record_extent_backrefs(path, new)) { | |
2534 | btrfs_free_path(path); | |
2535 | goto out; | |
2536 | } | |
2537 | btrfs_release_path(path); | |
2538 | ||
2539 | while (1) { | |
2540 | node = rb_first(&new->root); | |
2541 | if (!node) | |
2542 | break; | |
2543 | rb_erase(node, &new->root); | |
2544 | ||
2545 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2546 | ||
2547 | ret = relink_extent_backref(path, prev, backref); | |
2548 | WARN_ON(ret < 0); | |
2549 | ||
2550 | kfree(prev); | |
2551 | ||
2552 | if (ret == 1) | |
2553 | prev = backref; | |
2554 | else | |
2555 | prev = NULL; | |
2556 | cond_resched(); | |
2557 | } | |
2558 | kfree(prev); | |
2559 | ||
2560 | btrfs_free_path(path); | |
38c227d8 | 2561 | out: |
6f519564 LB |
2562 | free_sa_defrag_extent(new); |
2563 | ||
38c227d8 LB |
2564 | atomic_dec(&root->fs_info->defrag_running); |
2565 | wake_up(&root->fs_info->transaction_wait); | |
38c227d8 LB |
2566 | } |
2567 | ||
2568 | static struct new_sa_defrag_extent * | |
2569 | record_old_file_extents(struct inode *inode, | |
2570 | struct btrfs_ordered_extent *ordered) | |
2571 | { | |
2572 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2573 | struct btrfs_path *path; | |
2574 | struct btrfs_key key; | |
6f519564 | 2575 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2576 | struct new_sa_defrag_extent *new; |
2577 | int ret; | |
2578 | ||
2579 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2580 | if (!new) | |
2581 | return NULL; | |
2582 | ||
2583 | new->inode = inode; | |
2584 | new->file_pos = ordered->file_offset; | |
2585 | new->len = ordered->len; | |
2586 | new->bytenr = ordered->start; | |
2587 | new->disk_len = ordered->disk_len; | |
2588 | new->compress_type = ordered->compress_type; | |
2589 | new->root = RB_ROOT; | |
2590 | INIT_LIST_HEAD(&new->head); | |
2591 | ||
2592 | path = btrfs_alloc_path(); | |
2593 | if (!path) | |
2594 | goto out_kfree; | |
2595 | ||
2596 | key.objectid = btrfs_ino(inode); | |
2597 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2598 | key.offset = new->file_pos; | |
2599 | ||
2600 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2601 | if (ret < 0) | |
2602 | goto out_free_path; | |
2603 | if (ret > 0 && path->slots[0] > 0) | |
2604 | path->slots[0]--; | |
2605 | ||
2606 | /* find out all the old extents for the file range */ | |
2607 | while (1) { | |
2608 | struct btrfs_file_extent_item *extent; | |
2609 | struct extent_buffer *l; | |
2610 | int slot; | |
2611 | u64 num_bytes; | |
2612 | u64 offset; | |
2613 | u64 end; | |
2614 | u64 disk_bytenr; | |
2615 | u64 extent_offset; | |
2616 | ||
2617 | l = path->nodes[0]; | |
2618 | slot = path->slots[0]; | |
2619 | ||
2620 | if (slot >= btrfs_header_nritems(l)) { | |
2621 | ret = btrfs_next_leaf(root, path); | |
2622 | if (ret < 0) | |
6f519564 | 2623 | goto out_free_path; |
38c227d8 LB |
2624 | else if (ret > 0) |
2625 | break; | |
2626 | continue; | |
2627 | } | |
2628 | ||
2629 | btrfs_item_key_to_cpu(l, &key, slot); | |
2630 | ||
2631 | if (key.objectid != btrfs_ino(inode)) | |
2632 | break; | |
2633 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2634 | break; | |
2635 | if (key.offset >= new->file_pos + new->len) | |
2636 | break; | |
2637 | ||
2638 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2639 | ||
2640 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2641 | if (key.offset + num_bytes < new->file_pos) | |
2642 | goto next; | |
2643 | ||
2644 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2645 | if (!disk_bytenr) | |
2646 | goto next; | |
2647 | ||
2648 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2649 | ||
2650 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2651 | if (!old) | |
6f519564 | 2652 | goto out_free_path; |
38c227d8 LB |
2653 | |
2654 | offset = max(new->file_pos, key.offset); | |
2655 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2656 | ||
2657 | old->bytenr = disk_bytenr; | |
2658 | old->extent_offset = extent_offset; | |
2659 | old->offset = offset - key.offset; | |
2660 | old->len = end - offset; | |
2661 | old->new = new; | |
2662 | old->count = 0; | |
2663 | list_add_tail(&old->list, &new->head); | |
2664 | next: | |
2665 | path->slots[0]++; | |
2666 | cond_resched(); | |
2667 | } | |
2668 | ||
2669 | btrfs_free_path(path); | |
2670 | atomic_inc(&root->fs_info->defrag_running); | |
2671 | ||
2672 | return new; | |
2673 | ||
38c227d8 LB |
2674 | out_free_path: |
2675 | btrfs_free_path(path); | |
2676 | out_kfree: | |
6f519564 | 2677 | free_sa_defrag_extent(new); |
38c227d8 LB |
2678 | return NULL; |
2679 | } | |
2680 | ||
e570fd27 MX |
2681 | static void btrfs_release_delalloc_bytes(struct btrfs_root *root, |
2682 | u64 start, u64 len) | |
2683 | { | |
2684 | struct btrfs_block_group_cache *cache; | |
2685 | ||
2686 | cache = btrfs_lookup_block_group(root->fs_info, start); | |
2687 | ASSERT(cache); | |
2688 | ||
2689 | spin_lock(&cache->lock); | |
2690 | cache->delalloc_bytes -= len; | |
2691 | spin_unlock(&cache->lock); | |
2692 | ||
2693 | btrfs_put_block_group(cache); | |
2694 | } | |
2695 | ||
d352ac68 CM |
2696 | /* as ordered data IO finishes, this gets called so we can finish |
2697 | * an ordered extent if the range of bytes in the file it covers are | |
2698 | * fully written. | |
2699 | */ | |
5fd02043 | 2700 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2701 | { |
5fd02043 | 2702 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 2703 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2704 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2705 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2706 | struct extent_state *cached_state = NULL; |
38c227d8 | 2707 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2708 | int compress_type = 0; |
77cef2ec JB |
2709 | int ret = 0; |
2710 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2711 | bool nolock; |
77cef2ec | 2712 | bool truncated = false; |
e6dcd2dc | 2713 | |
83eea1f1 | 2714 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 2715 | |
5fd02043 JB |
2716 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2717 | ret = -EIO; | |
2718 | goto out; | |
2719 | } | |
2720 | ||
f612496b MX |
2721 | btrfs_free_io_failure_record(inode, ordered_extent->file_offset, |
2722 | ordered_extent->file_offset + | |
2723 | ordered_extent->len - 1); | |
2724 | ||
77cef2ec JB |
2725 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2726 | truncated = true; | |
2727 | logical_len = ordered_extent->truncated_len; | |
2728 | /* Truncated the entire extent, don't bother adding */ | |
2729 | if (!logical_len) | |
2730 | goto out; | |
2731 | } | |
2732 | ||
c2167754 | 2733 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2734 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
6c760c07 JB |
2735 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2736 | if (nolock) | |
2737 | trans = btrfs_join_transaction_nolock(root); | |
2738 | else | |
2739 | trans = btrfs_join_transaction(root); | |
2740 | if (IS_ERR(trans)) { | |
2741 | ret = PTR_ERR(trans); | |
2742 | trans = NULL; | |
2743 | goto out; | |
c2167754 | 2744 | } |
6c760c07 JB |
2745 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2746 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2747 | if (ret) /* -ENOMEM or corruption */ | |
2748 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
2749 | goto out; |
2750 | } | |
e6dcd2dc | 2751 | |
2ac55d41 JB |
2752 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2753 | ordered_extent->file_offset + ordered_extent->len - 1, | |
d0082371 | 2754 | 0, &cached_state); |
e6dcd2dc | 2755 | |
38c227d8 LB |
2756 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2757 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2758 | EXTENT_DEFRAG, 1, cached_state); | |
2759 | if (ret) { | |
2760 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 2761 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
2762 | /* the inode is shared */ |
2763 | new = record_old_file_extents(inode, ordered_extent); | |
2764 | ||
2765 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
2766 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2767 | EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS); | |
2768 | } | |
2769 | ||
0cb59c99 | 2770 | if (nolock) |
7a7eaa40 | 2771 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 2772 | else |
7a7eaa40 | 2773 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2774 | if (IS_ERR(trans)) { |
2775 | ret = PTR_ERR(trans); | |
2776 | trans = NULL; | |
2777 | goto out_unlock; | |
2778 | } | |
a79b7d4b | 2779 | |
0ca1f7ce | 2780 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 2781 | |
c8b97818 | 2782 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 2783 | compress_type = ordered_extent->compress_type; |
d899e052 | 2784 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 2785 | BUG_ON(compress_type); |
920bbbfb | 2786 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
2787 | ordered_extent->file_offset, |
2788 | ordered_extent->file_offset + | |
77cef2ec | 2789 | logical_len); |
d899e052 | 2790 | } else { |
0af3d00b | 2791 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
2792 | ret = insert_reserved_file_extent(trans, inode, |
2793 | ordered_extent->file_offset, | |
2794 | ordered_extent->start, | |
2795 | ordered_extent->disk_len, | |
77cef2ec | 2796 | logical_len, logical_len, |
261507a0 | 2797 | compress_type, 0, 0, |
d899e052 | 2798 | BTRFS_FILE_EXTENT_REG); |
e570fd27 MX |
2799 | if (!ret) |
2800 | btrfs_release_delalloc_bytes(root, | |
2801 | ordered_extent->start, | |
2802 | ordered_extent->disk_len); | |
d899e052 | 2803 | } |
5dc562c5 JB |
2804 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
2805 | ordered_extent->file_offset, ordered_extent->len, | |
2806 | trans->transid); | |
79787eaa JM |
2807 | if (ret < 0) { |
2808 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 2809 | goto out_unlock; |
79787eaa | 2810 | } |
2ac55d41 | 2811 | |
e6dcd2dc CM |
2812 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
2813 | &ordered_extent->list); | |
2814 | ||
6c760c07 JB |
2815 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2816 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2817 | if (ret) { /* -ENOMEM or corruption */ | |
2818 | btrfs_abort_transaction(trans, root, ret); | |
2819 | goto out_unlock; | |
1ef30be1 JB |
2820 | } |
2821 | ret = 0; | |
5fd02043 JB |
2822 | out_unlock: |
2823 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
2824 | ordered_extent->file_offset + | |
2825 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 2826 | out: |
5b0e95bf | 2827 | if (root != root->fs_info->tree_root) |
0cb59c99 | 2828 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 MX |
2829 | if (trans) |
2830 | btrfs_end_transaction(trans, root); | |
0cb59c99 | 2831 | |
77cef2ec JB |
2832 | if (ret || truncated) { |
2833 | u64 start, end; | |
2834 | ||
2835 | if (truncated) | |
2836 | start = ordered_extent->file_offset + logical_len; | |
2837 | else | |
2838 | start = ordered_extent->file_offset; | |
2839 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
2840 | clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS); | |
2841 | ||
2842 | /* Drop the cache for the part of the extent we didn't write. */ | |
2843 | btrfs_drop_extent_cache(inode, start, end, 0); | |
5fd02043 | 2844 | |
0bec9ef5 JB |
2845 | /* |
2846 | * If the ordered extent had an IOERR or something else went | |
2847 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
2848 | * back to the allocator. We only free the extent in the |
2849 | * truncated case if we didn't write out the extent at all. | |
0bec9ef5 | 2850 | */ |
77cef2ec JB |
2851 | if ((ret || !logical_len) && |
2852 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
0bec9ef5 JB |
2853 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
2854 | btrfs_free_reserved_extent(root, ordered_extent->start, | |
e570fd27 | 2855 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
2856 | } |
2857 | ||
2858 | ||
5fd02043 | 2859 | /* |
8bad3c02 LB |
2860 | * This needs to be done to make sure anybody waiting knows we are done |
2861 | * updating everything for this ordered extent. | |
5fd02043 JB |
2862 | */ |
2863 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
2864 | ||
38c227d8 | 2865 | /* for snapshot-aware defrag */ |
6f519564 LB |
2866 | if (new) { |
2867 | if (ret) { | |
2868 | free_sa_defrag_extent(new); | |
2869 | atomic_dec(&root->fs_info->defrag_running); | |
2870 | } else { | |
2871 | relink_file_extents(new); | |
2872 | } | |
2873 | } | |
38c227d8 | 2874 | |
e6dcd2dc CM |
2875 | /* once for us */ |
2876 | btrfs_put_ordered_extent(ordered_extent); | |
2877 | /* once for the tree */ | |
2878 | btrfs_put_ordered_extent(ordered_extent); | |
2879 | ||
5fd02043 JB |
2880 | return ret; |
2881 | } | |
2882 | ||
2883 | static void finish_ordered_fn(struct btrfs_work *work) | |
2884 | { | |
2885 | struct btrfs_ordered_extent *ordered_extent; | |
2886 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
2887 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
2888 | } |
2889 | ||
b2950863 | 2890 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
2891 | struct extent_state *state, int uptodate) |
2892 | { | |
5fd02043 JB |
2893 | struct inode *inode = page->mapping->host; |
2894 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2895 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
9e0af237 LB |
2896 | struct btrfs_workqueue *wq; |
2897 | btrfs_work_func_t func; | |
5fd02043 | 2898 | |
1abe9b8a | 2899 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
2900 | ||
8b62b72b | 2901 | ClearPagePrivate2(page); |
5fd02043 JB |
2902 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
2903 | end - start + 1, uptodate)) | |
2904 | return 0; | |
2905 | ||
9e0af237 LB |
2906 | if (btrfs_is_free_space_inode(inode)) { |
2907 | wq = root->fs_info->endio_freespace_worker; | |
2908 | func = btrfs_freespace_write_helper; | |
2909 | } else { | |
2910 | wq = root->fs_info->endio_write_workers; | |
2911 | func = btrfs_endio_write_helper; | |
2912 | } | |
5fd02043 | 2913 | |
9e0af237 LB |
2914 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
2915 | NULL); | |
2916 | btrfs_queue_work(wq, &ordered_extent->work); | |
5fd02043 JB |
2917 | |
2918 | return 0; | |
211f90e6 CM |
2919 | } |
2920 | ||
dc380aea MX |
2921 | static int __readpage_endio_check(struct inode *inode, |
2922 | struct btrfs_io_bio *io_bio, | |
2923 | int icsum, struct page *page, | |
2924 | int pgoff, u64 start, size_t len) | |
2925 | { | |
2926 | char *kaddr; | |
2927 | u32 csum_expected; | |
2928 | u32 csum = ~(u32)0; | |
2929 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2930 | DEFAULT_RATELIMIT_BURST); | |
2931 | ||
2932 | csum_expected = *(((u32 *)io_bio->csum) + icsum); | |
2933 | ||
2934 | kaddr = kmap_atomic(page); | |
2935 | csum = btrfs_csum_data(kaddr + pgoff, csum, len); | |
2936 | btrfs_csum_final(csum, (char *)&csum); | |
2937 | if (csum != csum_expected) | |
2938 | goto zeroit; | |
2939 | ||
2940 | kunmap_atomic(kaddr); | |
2941 | return 0; | |
2942 | zeroit: | |
2943 | if (__ratelimit(&_rs)) | |
2944 | btrfs_info(BTRFS_I(inode)->root->fs_info, | |
2945 | "csum failed ino %llu off %llu csum %u expected csum %u", | |
2946 | btrfs_ino(inode), start, csum, csum_expected); | |
2947 | memset(kaddr + pgoff, 1, len); | |
2948 | flush_dcache_page(page); | |
2949 | kunmap_atomic(kaddr); | |
2950 | if (csum_expected == 0) | |
2951 | return 0; | |
2952 | return -EIO; | |
2953 | } | |
2954 | ||
d352ac68 CM |
2955 | /* |
2956 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
2957 | * if there's a match, we allow the bio to finish. If not, the code in |
2958 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 2959 | */ |
facc8a22 MX |
2960 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
2961 | u64 phy_offset, struct page *page, | |
2962 | u64 start, u64 end, int mirror) | |
07157aac | 2963 | { |
4eee4fa4 | 2964 | size_t offset = start - page_offset(page); |
07157aac | 2965 | struct inode *inode = page->mapping->host; |
d1310b2e | 2966 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 2967 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 2968 | |
d20f7043 CM |
2969 | if (PageChecked(page)) { |
2970 | ClearPageChecked(page); | |
dc380aea | 2971 | return 0; |
d20f7043 | 2972 | } |
6cbff00f CH |
2973 | |
2974 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 2975 | return 0; |
17d217fe YZ |
2976 | |
2977 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 2978 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
2979 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
2980 | GFP_NOFS); | |
b6cda9bc | 2981 | return 0; |
17d217fe | 2982 | } |
d20f7043 | 2983 | |
facc8a22 | 2984 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
dc380aea MX |
2985 | return __readpage_endio_check(inode, io_bio, phy_offset, page, offset, |
2986 | start, (size_t)(end - start + 1)); | |
07157aac | 2987 | } |
b888db2b | 2988 | |
24bbcf04 YZ |
2989 | struct delayed_iput { |
2990 | struct list_head list; | |
2991 | struct inode *inode; | |
2992 | }; | |
2993 | ||
79787eaa JM |
2994 | /* JDM: If this is fs-wide, why can't we add a pointer to |
2995 | * btrfs_inode instead and avoid the allocation? */ | |
24bbcf04 YZ |
2996 | void btrfs_add_delayed_iput(struct inode *inode) |
2997 | { | |
2998 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2999 | struct delayed_iput *delayed; | |
3000 | ||
3001 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3002 | return; | |
3003 | ||
3004 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
3005 | delayed->inode = inode; | |
3006 | ||
3007 | spin_lock(&fs_info->delayed_iput_lock); | |
3008 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
3009 | spin_unlock(&fs_info->delayed_iput_lock); | |
3010 | } | |
3011 | ||
3012 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
3013 | { | |
3014 | LIST_HEAD(list); | |
3015 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3016 | struct delayed_iput *delayed; | |
3017 | int empty; | |
3018 | ||
3019 | spin_lock(&fs_info->delayed_iput_lock); | |
3020 | empty = list_empty(&fs_info->delayed_iputs); | |
3021 | spin_unlock(&fs_info->delayed_iput_lock); | |
3022 | if (empty) | |
3023 | return; | |
3024 | ||
24bbcf04 YZ |
3025 | spin_lock(&fs_info->delayed_iput_lock); |
3026 | list_splice_init(&fs_info->delayed_iputs, &list); | |
3027 | spin_unlock(&fs_info->delayed_iput_lock); | |
3028 | ||
3029 | while (!list_empty(&list)) { | |
3030 | delayed = list_entry(list.next, struct delayed_iput, list); | |
3031 | list_del(&delayed->list); | |
3032 | iput(delayed->inode); | |
3033 | kfree(delayed); | |
3034 | } | |
24bbcf04 YZ |
3035 | } |
3036 | ||
d68fc57b | 3037 | /* |
42b2aa86 | 3038 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
3039 | * files in the subvolume, it removes orphan item and frees block_rsv |
3040 | * structure. | |
3041 | */ | |
3042 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
3043 | struct btrfs_root *root) | |
3044 | { | |
90290e19 | 3045 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
3046 | int ret; |
3047 | ||
8a35d95f | 3048 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
3049 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
3050 | return; | |
3051 | ||
90290e19 | 3052 | spin_lock(&root->orphan_lock); |
8a35d95f | 3053 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
3054 | spin_unlock(&root->orphan_lock); |
3055 | return; | |
3056 | } | |
3057 | ||
3058 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
3059 | spin_unlock(&root->orphan_lock); | |
3060 | return; | |
3061 | } | |
3062 | ||
3063 | block_rsv = root->orphan_block_rsv; | |
3064 | root->orphan_block_rsv = NULL; | |
3065 | spin_unlock(&root->orphan_lock); | |
3066 | ||
27cdeb70 | 3067 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) && |
d68fc57b YZ |
3068 | btrfs_root_refs(&root->root_item) > 0) { |
3069 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
3070 | root->root_key.objectid); | |
4ef31a45 JB |
3071 | if (ret) |
3072 | btrfs_abort_transaction(trans, root, ret); | |
3073 | else | |
27cdeb70 MX |
3074 | clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, |
3075 | &root->state); | |
d68fc57b YZ |
3076 | } |
3077 | ||
90290e19 JB |
3078 | if (block_rsv) { |
3079 | WARN_ON(block_rsv->size > 0); | |
3080 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
3081 | } |
3082 | } | |
3083 | ||
7b128766 JB |
3084 | /* |
3085 | * This creates an orphan entry for the given inode in case something goes | |
3086 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
3087 | * |
3088 | * NOTE: caller of this function should reserve 5 units of metadata for | |
3089 | * this function. | |
7b128766 JB |
3090 | */ |
3091 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
3092 | { | |
3093 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3094 | struct btrfs_block_rsv *block_rsv = NULL; |
3095 | int reserve = 0; | |
3096 | int insert = 0; | |
3097 | int ret; | |
7b128766 | 3098 | |
d68fc57b | 3099 | if (!root->orphan_block_rsv) { |
66d8f3dd | 3100 | block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
b532402e TI |
3101 | if (!block_rsv) |
3102 | return -ENOMEM; | |
d68fc57b | 3103 | } |
7b128766 | 3104 | |
d68fc57b YZ |
3105 | spin_lock(&root->orphan_lock); |
3106 | if (!root->orphan_block_rsv) { | |
3107 | root->orphan_block_rsv = block_rsv; | |
3108 | } else if (block_rsv) { | |
3109 | btrfs_free_block_rsv(root, block_rsv); | |
3110 | block_rsv = NULL; | |
7b128766 | 3111 | } |
7b128766 | 3112 | |
8a35d95f JB |
3113 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3114 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
3115 | #if 0 |
3116 | /* | |
3117 | * For proper ENOSPC handling, we should do orphan | |
3118 | * cleanup when mounting. But this introduces backward | |
3119 | * compatibility issue. | |
3120 | */ | |
3121 | if (!xchg(&root->orphan_item_inserted, 1)) | |
3122 | insert = 2; | |
3123 | else | |
3124 | insert = 1; | |
3125 | #endif | |
3126 | insert = 1; | |
321f0e70 | 3127 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
3128 | } |
3129 | ||
72ac3c0d JB |
3130 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3131 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3132 | reserve = 1; |
d68fc57b | 3133 | spin_unlock(&root->orphan_lock); |
7b128766 | 3134 | |
d68fc57b YZ |
3135 | /* grab metadata reservation from transaction handle */ |
3136 | if (reserve) { | |
3137 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 3138 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 3139 | } |
7b128766 | 3140 | |
d68fc57b YZ |
3141 | /* insert an orphan item to track this unlinked/truncated file */ |
3142 | if (insert >= 1) { | |
33345d01 | 3143 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
4ef31a45 | 3144 | if (ret) { |
703c88e0 | 3145 | atomic_dec(&root->orphan_inodes); |
4ef31a45 JB |
3146 | if (reserve) { |
3147 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
3148 | &BTRFS_I(inode)->runtime_flags); | |
3149 | btrfs_orphan_release_metadata(inode); | |
3150 | } | |
3151 | if (ret != -EEXIST) { | |
e8e7cff6 JB |
3152 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3153 | &BTRFS_I(inode)->runtime_flags); | |
4ef31a45 JB |
3154 | btrfs_abort_transaction(trans, root, ret); |
3155 | return ret; | |
3156 | } | |
79787eaa JM |
3157 | } |
3158 | ret = 0; | |
d68fc57b YZ |
3159 | } |
3160 | ||
3161 | /* insert an orphan item to track subvolume contains orphan files */ | |
3162 | if (insert >= 2) { | |
3163 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
3164 | root->root_key.objectid); | |
79787eaa JM |
3165 | if (ret && ret != -EEXIST) { |
3166 | btrfs_abort_transaction(trans, root, ret); | |
3167 | return ret; | |
3168 | } | |
d68fc57b YZ |
3169 | } |
3170 | return 0; | |
7b128766 JB |
3171 | } |
3172 | ||
3173 | /* | |
3174 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
3175 | * item for this particular inode. | |
3176 | */ | |
48a3b636 ES |
3177 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3178 | struct inode *inode) | |
7b128766 JB |
3179 | { |
3180 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3181 | int delete_item = 0; |
3182 | int release_rsv = 0; | |
7b128766 JB |
3183 | int ret = 0; |
3184 | ||
d68fc57b | 3185 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
3186 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3187 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3188 | delete_item = 1; |
7b128766 | 3189 | |
72ac3c0d JB |
3190 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3191 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3192 | release_rsv = 1; |
d68fc57b | 3193 | spin_unlock(&root->orphan_lock); |
7b128766 | 3194 | |
703c88e0 | 3195 | if (delete_item) { |
8a35d95f | 3196 | atomic_dec(&root->orphan_inodes); |
703c88e0 FDBM |
3197 | if (trans) |
3198 | ret = btrfs_del_orphan_item(trans, root, | |
3199 | btrfs_ino(inode)); | |
8a35d95f | 3200 | } |
7b128766 | 3201 | |
703c88e0 FDBM |
3202 | if (release_rsv) |
3203 | btrfs_orphan_release_metadata(inode); | |
3204 | ||
4ef31a45 | 3205 | return ret; |
7b128766 JB |
3206 | } |
3207 | ||
3208 | /* | |
3209 | * this cleans up any orphans that may be left on the list from the last use | |
3210 | * of this root. | |
3211 | */ | |
66b4ffd1 | 3212 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
3213 | { |
3214 | struct btrfs_path *path; | |
3215 | struct extent_buffer *leaf; | |
7b128766 JB |
3216 | struct btrfs_key key, found_key; |
3217 | struct btrfs_trans_handle *trans; | |
3218 | struct inode *inode; | |
8f6d7f4f | 3219 | u64 last_objectid = 0; |
7b128766 JB |
3220 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3221 | ||
d68fc57b | 3222 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3223 | return 0; |
c71bf099 YZ |
3224 | |
3225 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3226 | if (!path) { |
3227 | ret = -ENOMEM; | |
3228 | goto out; | |
3229 | } | |
7b128766 JB |
3230 | path->reada = -1; |
3231 | ||
3232 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3233 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3234 | key.offset = (u64)-1; |
3235 | ||
7b128766 JB |
3236 | while (1) { |
3237 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3238 | if (ret < 0) |
3239 | goto out; | |
7b128766 JB |
3240 | |
3241 | /* | |
3242 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3243 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3244 | * find the key and see if we have stuff that matches |
3245 | */ | |
3246 | if (ret > 0) { | |
66b4ffd1 | 3247 | ret = 0; |
7b128766 JB |
3248 | if (path->slots[0] == 0) |
3249 | break; | |
3250 | path->slots[0]--; | |
3251 | } | |
3252 | ||
3253 | /* pull out the item */ | |
3254 | leaf = path->nodes[0]; | |
7b128766 JB |
3255 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3256 | ||
3257 | /* make sure the item matches what we want */ | |
3258 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3259 | break; | |
962a298f | 3260 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3261 | break; |
3262 | ||
3263 | /* release the path since we're done with it */ | |
b3b4aa74 | 3264 | btrfs_release_path(path); |
7b128766 JB |
3265 | |
3266 | /* | |
3267 | * this is where we are basically btrfs_lookup, without the | |
3268 | * crossing root thing. we store the inode number in the | |
3269 | * offset of the orphan item. | |
3270 | */ | |
8f6d7f4f JB |
3271 | |
3272 | if (found_key.offset == last_objectid) { | |
c2cf52eb SK |
3273 | btrfs_err(root->fs_info, |
3274 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3275 | ret = -EINVAL; |
3276 | goto out; | |
3277 | } | |
3278 | ||
3279 | last_objectid = found_key.offset; | |
3280 | ||
5d4f98a2 YZ |
3281 | found_key.objectid = found_key.offset; |
3282 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3283 | found_key.offset = 0; | |
73f73415 | 3284 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3285 | ret = PTR_ERR_OR_ZERO(inode); |
a8c9e576 | 3286 | if (ret && ret != -ESTALE) |
66b4ffd1 | 3287 | goto out; |
7b128766 | 3288 | |
f8e9e0b0 AJ |
3289 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
3290 | struct btrfs_root *dead_root; | |
3291 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3292 | int is_dead_root = 0; | |
3293 | ||
3294 | /* | |
3295 | * this is an orphan in the tree root. Currently these | |
3296 | * could come from 2 sources: | |
3297 | * a) a snapshot deletion in progress | |
3298 | * b) a free space cache inode | |
3299 | * We need to distinguish those two, as the snapshot | |
3300 | * orphan must not get deleted. | |
3301 | * find_dead_roots already ran before us, so if this | |
3302 | * is a snapshot deletion, we should find the root | |
3303 | * in the dead_roots list | |
3304 | */ | |
3305 | spin_lock(&fs_info->trans_lock); | |
3306 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3307 | root_list) { | |
3308 | if (dead_root->root_key.objectid == | |
3309 | found_key.objectid) { | |
3310 | is_dead_root = 1; | |
3311 | break; | |
3312 | } | |
3313 | } | |
3314 | spin_unlock(&fs_info->trans_lock); | |
3315 | if (is_dead_root) { | |
3316 | /* prevent this orphan from being found again */ | |
3317 | key.offset = found_key.objectid - 1; | |
3318 | continue; | |
3319 | } | |
3320 | } | |
7b128766 | 3321 | /* |
a8c9e576 JB |
3322 | * Inode is already gone but the orphan item is still there, |
3323 | * kill the orphan item. | |
7b128766 | 3324 | */ |
a8c9e576 JB |
3325 | if (ret == -ESTALE) { |
3326 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
3327 | if (IS_ERR(trans)) { |
3328 | ret = PTR_ERR(trans); | |
3329 | goto out; | |
3330 | } | |
c2cf52eb SK |
3331 | btrfs_debug(root->fs_info, "auto deleting %Lu", |
3332 | found_key.objectid); | |
a8c9e576 JB |
3333 | ret = btrfs_del_orphan_item(trans, root, |
3334 | found_key.objectid); | |
5b21f2ed | 3335 | btrfs_end_transaction(trans, root); |
4ef31a45 JB |
3336 | if (ret) |
3337 | goto out; | |
7b128766 JB |
3338 | continue; |
3339 | } | |
3340 | ||
a8c9e576 JB |
3341 | /* |
3342 | * add this inode to the orphan list so btrfs_orphan_del does | |
3343 | * the proper thing when we hit it | |
3344 | */ | |
8a35d95f JB |
3345 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3346 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3347 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3348 | |
7b128766 JB |
3349 | /* if we have links, this was a truncate, lets do that */ |
3350 | if (inode->i_nlink) { | |
fae7f21c | 3351 | if (WARN_ON(!S_ISREG(inode->i_mode))) { |
a41ad394 JB |
3352 | iput(inode); |
3353 | continue; | |
3354 | } | |
7b128766 | 3355 | nr_truncate++; |
f3fe820c JB |
3356 | |
3357 | /* 1 for the orphan item deletion. */ | |
3358 | trans = btrfs_start_transaction(root, 1); | |
3359 | if (IS_ERR(trans)) { | |
c69b26b0 | 3360 | iput(inode); |
f3fe820c JB |
3361 | ret = PTR_ERR(trans); |
3362 | goto out; | |
3363 | } | |
3364 | ret = btrfs_orphan_add(trans, inode); | |
3365 | btrfs_end_transaction(trans, root); | |
c69b26b0 JB |
3366 | if (ret) { |
3367 | iput(inode); | |
f3fe820c | 3368 | goto out; |
c69b26b0 | 3369 | } |
f3fe820c | 3370 | |
66b4ffd1 | 3371 | ret = btrfs_truncate(inode); |
4a7d0f68 JB |
3372 | if (ret) |
3373 | btrfs_orphan_del(NULL, inode); | |
7b128766 JB |
3374 | } else { |
3375 | nr_unlink++; | |
3376 | } | |
3377 | ||
3378 | /* this will do delete_inode and everything for us */ | |
3379 | iput(inode); | |
66b4ffd1 JB |
3380 | if (ret) |
3381 | goto out; | |
7b128766 | 3382 | } |
3254c876 MX |
3383 | /* release the path since we're done with it */ |
3384 | btrfs_release_path(path); | |
3385 | ||
d68fc57b YZ |
3386 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3387 | ||
3388 | if (root->orphan_block_rsv) | |
3389 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
3390 | (u64)-1); | |
3391 | ||
27cdeb70 MX |
3392 | if (root->orphan_block_rsv || |
3393 | test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { | |
7a7eaa40 | 3394 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
3395 | if (!IS_ERR(trans)) |
3396 | btrfs_end_transaction(trans, root); | |
d68fc57b | 3397 | } |
7b128766 JB |
3398 | |
3399 | if (nr_unlink) | |
4884b476 | 3400 | btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3401 | if (nr_truncate) |
4884b476 | 3402 | btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3403 | |
3404 | out: | |
3405 | if (ret) | |
c2cf52eb SK |
3406 | btrfs_crit(root->fs_info, |
3407 | "could not do orphan cleanup %d", ret); | |
66b4ffd1 JB |
3408 | btrfs_free_path(path); |
3409 | return ret; | |
7b128766 JB |
3410 | } |
3411 | ||
46a53cca CM |
3412 | /* |
3413 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3414 | * don't find any xattrs, we know there can't be any acls. | |
3415 | * | |
3416 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3417 | */ | |
3418 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3419 | int slot, u64 objectid, |
3420 | int *first_xattr_slot) | |
46a53cca CM |
3421 | { |
3422 | u32 nritems = btrfs_header_nritems(leaf); | |
3423 | struct btrfs_key found_key; | |
f23b5a59 JB |
3424 | static u64 xattr_access = 0; |
3425 | static u64 xattr_default = 0; | |
46a53cca CM |
3426 | int scanned = 0; |
3427 | ||
f23b5a59 JB |
3428 | if (!xattr_access) { |
3429 | xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS, | |
3430 | strlen(POSIX_ACL_XATTR_ACCESS)); | |
3431 | xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT, | |
3432 | strlen(POSIX_ACL_XATTR_DEFAULT)); | |
3433 | } | |
3434 | ||
46a53cca | 3435 | slot++; |
63541927 | 3436 | *first_xattr_slot = -1; |
46a53cca CM |
3437 | while (slot < nritems) { |
3438 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3439 | ||
3440 | /* we found a different objectid, there must not be acls */ | |
3441 | if (found_key.objectid != objectid) | |
3442 | return 0; | |
3443 | ||
3444 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3445 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3446 | if (*first_xattr_slot == -1) |
3447 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3448 | if (found_key.offset == xattr_access || |
3449 | found_key.offset == xattr_default) | |
3450 | return 1; | |
3451 | } | |
46a53cca CM |
3452 | |
3453 | /* | |
3454 | * we found a key greater than an xattr key, there can't | |
3455 | * be any acls later on | |
3456 | */ | |
3457 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3458 | return 0; | |
3459 | ||
3460 | slot++; | |
3461 | scanned++; | |
3462 | ||
3463 | /* | |
3464 | * it goes inode, inode backrefs, xattrs, extents, | |
3465 | * so if there are a ton of hard links to an inode there can | |
3466 | * be a lot of backrefs. Don't waste time searching too hard, | |
3467 | * this is just an optimization | |
3468 | */ | |
3469 | if (scanned >= 8) | |
3470 | break; | |
3471 | } | |
3472 | /* we hit the end of the leaf before we found an xattr or | |
3473 | * something larger than an xattr. We have to assume the inode | |
3474 | * has acls | |
3475 | */ | |
63541927 FDBM |
3476 | if (*first_xattr_slot == -1) |
3477 | *first_xattr_slot = slot; | |
46a53cca CM |
3478 | return 1; |
3479 | } | |
3480 | ||
d352ac68 CM |
3481 | /* |
3482 | * read an inode from the btree into the in-memory inode | |
3483 | */ | |
5d4f98a2 | 3484 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
3485 | { |
3486 | struct btrfs_path *path; | |
5f39d397 | 3487 | struct extent_buffer *leaf; |
39279cc3 | 3488 | struct btrfs_inode_item *inode_item; |
0b86a832 | 3489 | struct btrfs_timespec *tspec; |
39279cc3 CM |
3490 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3491 | struct btrfs_key location; | |
67de1176 | 3492 | unsigned long ptr; |
46a53cca | 3493 | int maybe_acls; |
618e21d5 | 3494 | u32 rdev; |
39279cc3 | 3495 | int ret; |
2f7e33d4 | 3496 | bool filled = false; |
63541927 | 3497 | int first_xattr_slot; |
2f7e33d4 MX |
3498 | |
3499 | ret = btrfs_fill_inode(inode, &rdev); | |
3500 | if (!ret) | |
3501 | filled = true; | |
39279cc3 CM |
3502 | |
3503 | path = btrfs_alloc_path(); | |
1748f843 MF |
3504 | if (!path) |
3505 | goto make_bad; | |
3506 | ||
39279cc3 | 3507 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3508 | |
39279cc3 | 3509 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 3510 | if (ret) |
39279cc3 | 3511 | goto make_bad; |
39279cc3 | 3512 | |
5f39d397 | 3513 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3514 | |
3515 | if (filled) | |
67de1176 | 3516 | goto cache_index; |
2f7e33d4 | 3517 | |
5f39d397 CM |
3518 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3519 | struct btrfs_inode_item); | |
5f39d397 | 3520 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3521 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3522 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3523 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
dbe674a9 | 3524 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
3525 | |
3526 | tspec = btrfs_inode_atime(inode_item); | |
3527 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3528 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3529 | ||
3530 | tspec = btrfs_inode_mtime(inode_item); | |
3531 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3532 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3533 | ||
3534 | tspec = btrfs_inode_ctime(inode_item); | |
3535 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3536 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3537 | ||
a76a3cd4 | 3538 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3539 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3540 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3541 | ||
3542 | /* | |
3543 | * If we were modified in the current generation and evicted from memory | |
3544 | * and then re-read we need to do a full sync since we don't have any | |
3545 | * idea about which extents were modified before we were evicted from | |
3546 | * cache. | |
3547 | */ | |
3548 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
3549 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3550 | &BTRFS_I(inode)->runtime_flags); | |
3551 | ||
0c4d2d95 | 3552 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 3553 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 3554 | inode->i_rdev = 0; |
5f39d397 CM |
3555 | rdev = btrfs_inode_rdev(leaf, inode_item); |
3556 | ||
aec7477b | 3557 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 3558 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
67de1176 MX |
3559 | |
3560 | cache_index: | |
3561 | path->slots[0]++; | |
3562 | if (inode->i_nlink != 1 || | |
3563 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3564 | goto cache_acl; | |
3565 | ||
3566 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
3567 | if (location.objectid != btrfs_ino(inode)) | |
3568 | goto cache_acl; | |
3569 | ||
3570 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3571 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3572 | struct btrfs_inode_ref *ref; | |
3573 | ||
3574 | ref = (struct btrfs_inode_ref *)ptr; | |
3575 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3576 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3577 | struct btrfs_inode_extref *extref; | |
3578 | ||
3579 | extref = (struct btrfs_inode_extref *)ptr; | |
3580 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3581 | extref); | |
3582 | } | |
2f7e33d4 | 3583 | cache_acl: |
46a53cca CM |
3584 | /* |
3585 | * try to precache a NULL acl entry for files that don't have | |
3586 | * any xattrs or acls | |
3587 | */ | |
33345d01 | 3588 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
63541927 FDBM |
3589 | btrfs_ino(inode), &first_xattr_slot); |
3590 | if (first_xattr_slot != -1) { | |
3591 | path->slots[0] = first_xattr_slot; | |
3592 | ret = btrfs_load_inode_props(inode, path); | |
3593 | if (ret) | |
3594 | btrfs_err(root->fs_info, | |
351fd353 | 3595 | "error loading props for ino %llu (root %llu): %d", |
63541927 FDBM |
3596 | btrfs_ino(inode), |
3597 | root->root_key.objectid, ret); | |
3598 | } | |
3599 | btrfs_free_path(path); | |
3600 | ||
72c04902 AV |
3601 | if (!maybe_acls) |
3602 | cache_no_acl(inode); | |
46a53cca | 3603 | |
39279cc3 | 3604 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3605 | case S_IFREG: |
3606 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 3607 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 3608 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3609 | inode->i_fop = &btrfs_file_operations; |
3610 | inode->i_op = &btrfs_file_inode_operations; | |
3611 | break; | |
3612 | case S_IFDIR: | |
3613 | inode->i_fop = &btrfs_dir_file_operations; | |
3614 | if (root == root->fs_info->tree_root) | |
3615 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
3616 | else | |
3617 | inode->i_op = &btrfs_dir_inode_operations; | |
3618 | break; | |
3619 | case S_IFLNK: | |
3620 | inode->i_op = &btrfs_symlink_inode_operations; | |
3621 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 3622 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 3623 | break; |
618e21d5 | 3624 | default: |
0279b4cd | 3625 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3626 | init_special_inode(inode, inode->i_mode, rdev); |
3627 | break; | |
39279cc3 | 3628 | } |
6cbff00f CH |
3629 | |
3630 | btrfs_update_iflags(inode); | |
39279cc3 CM |
3631 | return; |
3632 | ||
3633 | make_bad: | |
39279cc3 | 3634 | btrfs_free_path(path); |
39279cc3 CM |
3635 | make_bad_inode(inode); |
3636 | } | |
3637 | ||
d352ac68 CM |
3638 | /* |
3639 | * given a leaf and an inode, copy the inode fields into the leaf | |
3640 | */ | |
e02119d5 CM |
3641 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3642 | struct extent_buffer *leaf, | |
5f39d397 | 3643 | struct btrfs_inode_item *item, |
39279cc3 CM |
3644 | struct inode *inode) |
3645 | { | |
51fab693 LB |
3646 | struct btrfs_map_token token; |
3647 | ||
3648 | btrfs_init_map_token(&token); | |
5f39d397 | 3649 | |
51fab693 LB |
3650 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3651 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3652 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3653 | &token); | |
3654 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3655 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3656 | |
51fab693 LB |
3657 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item), |
3658 | inode->i_atime.tv_sec, &token); | |
3659 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item), | |
3660 | inode->i_atime.tv_nsec, &token); | |
5f39d397 | 3661 | |
51fab693 LB |
3662 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item), |
3663 | inode->i_mtime.tv_sec, &token); | |
3664 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
3665 | inode->i_mtime.tv_nsec, &token); | |
5f39d397 | 3666 | |
51fab693 LB |
3667 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item), |
3668 | inode->i_ctime.tv_sec, &token); | |
3669 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
3670 | inode->i_ctime.tv_nsec, &token); | |
5f39d397 | 3671 | |
51fab693 LB |
3672 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3673 | &token); | |
3674 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3675 | &token); | |
3676 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3677 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3678 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3679 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3680 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3681 | } |
3682 | ||
d352ac68 CM |
3683 | /* |
3684 | * copy everything in the in-memory inode into the btree. | |
3685 | */ | |
2115133f | 3686 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3687 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3688 | { |
3689 | struct btrfs_inode_item *inode_item; | |
3690 | struct btrfs_path *path; | |
5f39d397 | 3691 | struct extent_buffer *leaf; |
39279cc3 CM |
3692 | int ret; |
3693 | ||
3694 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3695 | if (!path) |
3696 | return -ENOMEM; | |
3697 | ||
b9473439 | 3698 | path->leave_spinning = 1; |
16cdcec7 MX |
3699 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3700 | 1); | |
39279cc3 CM |
3701 | if (ret) { |
3702 | if (ret > 0) | |
3703 | ret = -ENOENT; | |
3704 | goto failed; | |
3705 | } | |
3706 | ||
5f39d397 CM |
3707 | leaf = path->nodes[0]; |
3708 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3709 | struct btrfs_inode_item); |
39279cc3 | 3710 | |
e02119d5 | 3711 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3712 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3713 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3714 | ret = 0; |
3715 | failed: | |
39279cc3 CM |
3716 | btrfs_free_path(path); |
3717 | return ret; | |
3718 | } | |
3719 | ||
2115133f CM |
3720 | /* |
3721 | * copy everything in the in-memory inode into the btree. | |
3722 | */ | |
3723 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3724 | struct btrfs_root *root, struct inode *inode) | |
3725 | { | |
3726 | int ret; | |
3727 | ||
3728 | /* | |
3729 | * If the inode is a free space inode, we can deadlock during commit | |
3730 | * if we put it into the delayed code. | |
3731 | * | |
3732 | * The data relocation inode should also be directly updated | |
3733 | * without delay | |
3734 | */ | |
83eea1f1 | 3735 | if (!btrfs_is_free_space_inode(inode) |
1d52c78a JB |
3736 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
3737 | && !root->fs_info->log_root_recovering) { | |
8ea05e3a AB |
3738 | btrfs_update_root_times(trans, root); |
3739 | ||
2115133f CM |
3740 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3741 | if (!ret) | |
3742 | btrfs_set_inode_last_trans(trans, inode); | |
3743 | return ret; | |
3744 | } | |
3745 | ||
3746 | return btrfs_update_inode_item(trans, root, inode); | |
3747 | } | |
3748 | ||
be6aef60 JB |
3749 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3750 | struct btrfs_root *root, | |
3751 | struct inode *inode) | |
2115133f CM |
3752 | { |
3753 | int ret; | |
3754 | ||
3755 | ret = btrfs_update_inode(trans, root, inode); | |
3756 | if (ret == -ENOSPC) | |
3757 | return btrfs_update_inode_item(trans, root, inode); | |
3758 | return ret; | |
3759 | } | |
3760 | ||
d352ac68 CM |
3761 | /* |
3762 | * unlink helper that gets used here in inode.c and in the tree logging | |
3763 | * recovery code. It remove a link in a directory with a given name, and | |
3764 | * also drops the back refs in the inode to the directory | |
3765 | */ | |
92986796 AV |
3766 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3767 | struct btrfs_root *root, | |
3768 | struct inode *dir, struct inode *inode, | |
3769 | const char *name, int name_len) | |
39279cc3 CM |
3770 | { |
3771 | struct btrfs_path *path; | |
39279cc3 | 3772 | int ret = 0; |
5f39d397 | 3773 | struct extent_buffer *leaf; |
39279cc3 | 3774 | struct btrfs_dir_item *di; |
5f39d397 | 3775 | struct btrfs_key key; |
aec7477b | 3776 | u64 index; |
33345d01 LZ |
3777 | u64 ino = btrfs_ino(inode); |
3778 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3779 | |
3780 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3781 | if (!path) { |
3782 | ret = -ENOMEM; | |
554233a6 | 3783 | goto out; |
54aa1f4d CM |
3784 | } |
3785 | ||
b9473439 | 3786 | path->leave_spinning = 1; |
33345d01 | 3787 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
3788 | name, name_len, -1); |
3789 | if (IS_ERR(di)) { | |
3790 | ret = PTR_ERR(di); | |
3791 | goto err; | |
3792 | } | |
3793 | if (!di) { | |
3794 | ret = -ENOENT; | |
3795 | goto err; | |
3796 | } | |
5f39d397 CM |
3797 | leaf = path->nodes[0]; |
3798 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 3799 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3800 | if (ret) |
3801 | goto err; | |
b3b4aa74 | 3802 | btrfs_release_path(path); |
39279cc3 | 3803 | |
67de1176 MX |
3804 | /* |
3805 | * If we don't have dir index, we have to get it by looking up | |
3806 | * the inode ref, since we get the inode ref, remove it directly, | |
3807 | * it is unnecessary to do delayed deletion. | |
3808 | * | |
3809 | * But if we have dir index, needn't search inode ref to get it. | |
3810 | * Since the inode ref is close to the inode item, it is better | |
3811 | * that we delay to delete it, and just do this deletion when | |
3812 | * we update the inode item. | |
3813 | */ | |
3814 | if (BTRFS_I(inode)->dir_index) { | |
3815 | ret = btrfs_delayed_delete_inode_ref(inode); | |
3816 | if (!ret) { | |
3817 | index = BTRFS_I(inode)->dir_index; | |
3818 | goto skip_backref; | |
3819 | } | |
3820 | } | |
3821 | ||
33345d01 LZ |
3822 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
3823 | dir_ino, &index); | |
aec7477b | 3824 | if (ret) { |
c2cf52eb SK |
3825 | btrfs_info(root->fs_info, |
3826 | "failed to delete reference to %.*s, inode %llu parent %llu", | |
c1c9ff7c | 3827 | name_len, name, ino, dir_ino); |
79787eaa | 3828 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
3829 | goto err; |
3830 | } | |
67de1176 | 3831 | skip_backref: |
16cdcec7 | 3832 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3833 | if (ret) { |
3834 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 3835 | goto err; |
79787eaa | 3836 | } |
39279cc3 | 3837 | |
e02119d5 | 3838 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 3839 | inode, dir_ino); |
79787eaa JM |
3840 | if (ret != 0 && ret != -ENOENT) { |
3841 | btrfs_abort_transaction(trans, root, ret); | |
3842 | goto err; | |
3843 | } | |
e02119d5 CM |
3844 | |
3845 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
3846 | dir, index); | |
6418c961 CM |
3847 | if (ret == -ENOENT) |
3848 | ret = 0; | |
d4e3991b ZB |
3849 | else if (ret) |
3850 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 CM |
3851 | err: |
3852 | btrfs_free_path(path); | |
e02119d5 CM |
3853 | if (ret) |
3854 | goto out; | |
3855 | ||
3856 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
3857 | inode_inc_iversion(inode); |
3858 | inode_inc_iversion(dir); | |
e02119d5 | 3859 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
b9959295 | 3860 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 3861 | out: |
39279cc3 CM |
3862 | return ret; |
3863 | } | |
3864 | ||
92986796 AV |
3865 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3866 | struct btrfs_root *root, | |
3867 | struct inode *dir, struct inode *inode, | |
3868 | const char *name, int name_len) | |
3869 | { | |
3870 | int ret; | |
3871 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
3872 | if (!ret) { | |
8b558c5f | 3873 | drop_nlink(inode); |
92986796 AV |
3874 | ret = btrfs_update_inode(trans, root, inode); |
3875 | } | |
3876 | return ret; | |
3877 | } | |
39279cc3 | 3878 | |
a22285a6 YZ |
3879 | /* |
3880 | * helper to start transaction for unlink and rmdir. | |
3881 | * | |
d52be818 JB |
3882 | * unlink and rmdir are special in btrfs, they do not always free space, so |
3883 | * if we cannot make our reservations the normal way try and see if there is | |
3884 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
3885 | * allow the unlink to occur. | |
a22285a6 | 3886 | */ |
d52be818 | 3887 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 3888 | { |
39279cc3 | 3889 | struct btrfs_trans_handle *trans; |
a22285a6 | 3890 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d YZ |
3891 | int ret; |
3892 | ||
e70bea5f JB |
3893 | /* |
3894 | * 1 for the possible orphan item | |
3895 | * 1 for the dir item | |
3896 | * 1 for the dir index | |
3897 | * 1 for the inode ref | |
e70bea5f JB |
3898 | * 1 for the inode |
3899 | */ | |
6e137ed3 | 3900 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
3901 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
3902 | return trans; | |
4df27c4d | 3903 | |
d52be818 JB |
3904 | if (PTR_ERR(trans) == -ENOSPC) { |
3905 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); | |
4df27c4d | 3906 | |
d52be818 JB |
3907 | trans = btrfs_start_transaction(root, 0); |
3908 | if (IS_ERR(trans)) | |
3909 | return trans; | |
3910 | ret = btrfs_cond_migrate_bytes(root->fs_info, | |
3911 | &root->fs_info->trans_block_rsv, | |
3912 | num_bytes, 5); | |
3913 | if (ret) { | |
3914 | btrfs_end_transaction(trans, root); | |
3915 | return ERR_PTR(ret); | |
a22285a6 | 3916 | } |
5a77d76c | 3917 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
d52be818 | 3918 | trans->bytes_reserved = num_bytes; |
a22285a6 | 3919 | } |
d52be818 | 3920 | return trans; |
a22285a6 YZ |
3921 | } |
3922 | ||
3923 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
3924 | { | |
3925 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3926 | struct btrfs_trans_handle *trans; | |
3927 | struct inode *inode = dentry->d_inode; | |
3928 | int ret; | |
a22285a6 | 3929 | |
d52be818 | 3930 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
3931 | if (IS_ERR(trans)) |
3932 | return PTR_ERR(trans); | |
5f39d397 | 3933 | |
12fcfd22 CM |
3934 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
3935 | ||
e02119d5 CM |
3936 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3937 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
3938 | if (ret) |
3939 | goto out; | |
7b128766 | 3940 | |
a22285a6 | 3941 | if (inode->i_nlink == 0) { |
7b128766 | 3942 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
3943 | if (ret) |
3944 | goto out; | |
a22285a6 | 3945 | } |
7b128766 | 3946 | |
b532402e | 3947 | out: |
d52be818 | 3948 | btrfs_end_transaction(trans, root); |
b53d3f5d | 3949 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
3950 | return ret; |
3951 | } | |
3952 | ||
4df27c4d YZ |
3953 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
3954 | struct btrfs_root *root, | |
3955 | struct inode *dir, u64 objectid, | |
3956 | const char *name, int name_len) | |
3957 | { | |
3958 | struct btrfs_path *path; | |
3959 | struct extent_buffer *leaf; | |
3960 | struct btrfs_dir_item *di; | |
3961 | struct btrfs_key key; | |
3962 | u64 index; | |
3963 | int ret; | |
33345d01 | 3964 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
3965 | |
3966 | path = btrfs_alloc_path(); | |
3967 | if (!path) | |
3968 | return -ENOMEM; | |
3969 | ||
33345d01 | 3970 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3971 | name, name_len, -1); |
79787eaa JM |
3972 | if (IS_ERR_OR_NULL(di)) { |
3973 | if (!di) | |
3974 | ret = -ENOENT; | |
3975 | else | |
3976 | ret = PTR_ERR(di); | |
3977 | goto out; | |
3978 | } | |
4df27c4d YZ |
3979 | |
3980 | leaf = path->nodes[0]; | |
3981 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3982 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3983 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
3984 | if (ret) { |
3985 | btrfs_abort_transaction(trans, root, ret); | |
3986 | goto out; | |
3987 | } | |
b3b4aa74 | 3988 | btrfs_release_path(path); |
4df27c4d YZ |
3989 | |
3990 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3991 | objectid, root->root_key.objectid, | |
33345d01 | 3992 | dir_ino, &index, name, name_len); |
4df27c4d | 3993 | if (ret < 0) { |
79787eaa JM |
3994 | if (ret != -ENOENT) { |
3995 | btrfs_abort_transaction(trans, root, ret); | |
3996 | goto out; | |
3997 | } | |
33345d01 | 3998 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3999 | name, name_len); |
79787eaa JM |
4000 | if (IS_ERR_OR_NULL(di)) { |
4001 | if (!di) | |
4002 | ret = -ENOENT; | |
4003 | else | |
4004 | ret = PTR_ERR(di); | |
4005 | btrfs_abort_transaction(trans, root, ret); | |
4006 | goto out; | |
4007 | } | |
4df27c4d YZ |
4008 | |
4009 | leaf = path->nodes[0]; | |
4010 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 4011 | btrfs_release_path(path); |
4df27c4d YZ |
4012 | index = key.offset; |
4013 | } | |
945d8962 | 4014 | btrfs_release_path(path); |
4df27c4d | 4015 | |
16cdcec7 | 4016 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
4017 | if (ret) { |
4018 | btrfs_abort_transaction(trans, root, ret); | |
4019 | goto out; | |
4020 | } | |
4df27c4d YZ |
4021 | |
4022 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 4023 | inode_inc_iversion(dir); |
4df27c4d | 4024 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
5a24e84c | 4025 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
4026 | if (ret) |
4027 | btrfs_abort_transaction(trans, root, ret); | |
4028 | out: | |
71d7aed0 | 4029 | btrfs_free_path(path); |
79787eaa | 4030 | return ret; |
4df27c4d YZ |
4031 | } |
4032 | ||
39279cc3 CM |
4033 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4034 | { | |
4035 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 4036 | int err = 0; |
39279cc3 | 4037 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4038 | struct btrfs_trans_handle *trans; |
39279cc3 | 4039 | |
b3ae244e | 4040 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4041 | return -ENOTEMPTY; |
b3ae244e DS |
4042 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
4043 | return -EPERM; | |
134d4512 | 4044 | |
d52be818 | 4045 | trans = __unlink_start_trans(dir); |
a22285a6 | 4046 | if (IS_ERR(trans)) |
5df6a9f6 | 4047 | return PTR_ERR(trans); |
5df6a9f6 | 4048 | |
33345d01 | 4049 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
4050 | err = btrfs_unlink_subvol(trans, root, dir, |
4051 | BTRFS_I(inode)->location.objectid, | |
4052 | dentry->d_name.name, | |
4053 | dentry->d_name.len); | |
4054 | goto out; | |
4055 | } | |
4056 | ||
7b128766 JB |
4057 | err = btrfs_orphan_add(trans, inode); |
4058 | if (err) | |
4df27c4d | 4059 | goto out; |
7b128766 | 4060 | |
39279cc3 | 4061 | /* now the directory is empty */ |
e02119d5 CM |
4062 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
4063 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 4064 | if (!err) |
dbe674a9 | 4065 | btrfs_i_size_write(inode, 0); |
4df27c4d | 4066 | out: |
d52be818 | 4067 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4068 | btrfs_btree_balance_dirty(root); |
3954401f | 4069 | |
39279cc3 CM |
4070 | return err; |
4071 | } | |
4072 | ||
39279cc3 CM |
4073 | /* |
4074 | * this can truncate away extent items, csum items and directory items. | |
4075 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4076 | * any higher than new_size |
39279cc3 CM |
4077 | * |
4078 | * csum items that cross the new i_size are truncated to the new size | |
4079 | * as well. | |
7b128766 JB |
4080 | * |
4081 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4082 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4083 | */ |
8082510e YZ |
4084 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4085 | struct btrfs_root *root, | |
4086 | struct inode *inode, | |
4087 | u64 new_size, u32 min_type) | |
39279cc3 | 4088 | { |
39279cc3 | 4089 | struct btrfs_path *path; |
5f39d397 | 4090 | struct extent_buffer *leaf; |
39279cc3 | 4091 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4092 | struct btrfs_key key; |
4093 | struct btrfs_key found_key; | |
39279cc3 | 4094 | u64 extent_start = 0; |
db94535d | 4095 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4096 | u64 extent_offset = 0; |
39279cc3 | 4097 | u64 item_end = 0; |
7f4f6e0a | 4098 | u64 last_size = (u64)-1; |
8082510e | 4099 | u32 found_type = (u8)-1; |
39279cc3 CM |
4100 | int found_extent; |
4101 | int del_item; | |
85e21bac CM |
4102 | int pending_del_nr = 0; |
4103 | int pending_del_slot = 0; | |
179e29e4 | 4104 | int extent_type = -1; |
8082510e YZ |
4105 | int ret; |
4106 | int err = 0; | |
33345d01 | 4107 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
4108 | |
4109 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4110 | |
0eb0e19c MF |
4111 | path = btrfs_alloc_path(); |
4112 | if (!path) | |
4113 | return -ENOMEM; | |
4114 | path->reada = -1; | |
4115 | ||
5dc562c5 JB |
4116 | /* |
4117 | * We want to drop from the next block forward in case this new size is | |
4118 | * not block aligned since we will be keeping the last block of the | |
4119 | * extent just the way it is. | |
4120 | */ | |
27cdeb70 MX |
4121 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
4122 | root == root->fs_info->tree_root) | |
fda2832f QW |
4123 | btrfs_drop_extent_cache(inode, ALIGN(new_size, |
4124 | root->sectorsize), (u64)-1, 0); | |
8082510e | 4125 | |
16cdcec7 MX |
4126 | /* |
4127 | * This function is also used to drop the items in the log tree before | |
4128 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
4129 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
4130 | * items. | |
4131 | */ | |
4132 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4133 | btrfs_kill_delayed_inode_items(inode); | |
4134 | ||
33345d01 | 4135 | key.objectid = ino; |
39279cc3 | 4136 | key.offset = (u64)-1; |
5f39d397 CM |
4137 | key.type = (u8)-1; |
4138 | ||
85e21bac | 4139 | search_again: |
b9473439 | 4140 | path->leave_spinning = 1; |
85e21bac | 4141 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
4142 | if (ret < 0) { |
4143 | err = ret; | |
4144 | goto out; | |
4145 | } | |
d397712b | 4146 | |
85e21bac | 4147 | if (ret > 0) { |
e02119d5 CM |
4148 | /* there are no items in the tree for us to truncate, we're |
4149 | * done | |
4150 | */ | |
8082510e YZ |
4151 | if (path->slots[0] == 0) |
4152 | goto out; | |
85e21bac CM |
4153 | path->slots[0]--; |
4154 | } | |
4155 | ||
d397712b | 4156 | while (1) { |
39279cc3 | 4157 | fi = NULL; |
5f39d397 CM |
4158 | leaf = path->nodes[0]; |
4159 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4160 | found_type = found_key.type; |
39279cc3 | 4161 | |
33345d01 | 4162 | if (found_key.objectid != ino) |
39279cc3 | 4163 | break; |
5f39d397 | 4164 | |
85e21bac | 4165 | if (found_type < min_type) |
39279cc3 CM |
4166 | break; |
4167 | ||
5f39d397 | 4168 | item_end = found_key.offset; |
39279cc3 | 4169 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4170 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4171 | struct btrfs_file_extent_item); |
179e29e4 CM |
4172 | extent_type = btrfs_file_extent_type(leaf, fi); |
4173 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4174 | item_end += |
db94535d | 4175 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 4176 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 4177 | item_end += btrfs_file_extent_inline_len(leaf, |
514ac8ad | 4178 | path->slots[0], fi); |
39279cc3 | 4179 | } |
008630c1 | 4180 | item_end--; |
39279cc3 | 4181 | } |
8082510e YZ |
4182 | if (found_type > min_type) { |
4183 | del_item = 1; | |
4184 | } else { | |
4185 | if (item_end < new_size) | |
b888db2b | 4186 | break; |
8082510e YZ |
4187 | if (found_key.offset >= new_size) |
4188 | del_item = 1; | |
4189 | else | |
4190 | del_item = 0; | |
39279cc3 | 4191 | } |
39279cc3 | 4192 | found_extent = 0; |
39279cc3 | 4193 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4194 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4195 | goto delete; | |
4196 | ||
7f4f6e0a JB |
4197 | if (del_item) |
4198 | last_size = found_key.offset; | |
4199 | else | |
4200 | last_size = new_size; | |
4201 | ||
179e29e4 | 4202 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
39279cc3 | 4203 | u64 num_dec; |
db94535d | 4204 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4205 | if (!del_item) { |
db94535d CM |
4206 | u64 orig_num_bytes = |
4207 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4208 | extent_num_bytes = ALIGN(new_size - |
4209 | found_key.offset, | |
4210 | root->sectorsize); | |
db94535d CM |
4211 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4212 | extent_num_bytes); | |
4213 | num_dec = (orig_num_bytes - | |
9069218d | 4214 | extent_num_bytes); |
27cdeb70 MX |
4215 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4216 | &root->state) && | |
4217 | extent_start != 0) | |
a76a3cd4 | 4218 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4219 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4220 | } else { |
db94535d CM |
4221 | extent_num_bytes = |
4222 | btrfs_file_extent_disk_num_bytes(leaf, | |
4223 | fi); | |
5d4f98a2 YZ |
4224 | extent_offset = found_key.offset - |
4225 | btrfs_file_extent_offset(leaf, fi); | |
4226 | ||
39279cc3 | 4227 | /* FIXME blocksize != 4096 */ |
9069218d | 4228 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4229 | if (extent_start != 0) { |
4230 | found_extent = 1; | |
27cdeb70 MX |
4231 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4232 | &root->state)) | |
a76a3cd4 | 4233 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4234 | } |
39279cc3 | 4235 | } |
9069218d | 4236 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4237 | /* |
4238 | * we can't truncate inline items that have had | |
4239 | * special encodings | |
4240 | */ | |
4241 | if (!del_item && | |
4242 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
4243 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
4244 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
4245 | u32 size = new_size - found_key.offset; |
4246 | ||
27cdeb70 | 4247 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
a76a3cd4 YZ |
4248 | inode_sub_bytes(inode, item_end + 1 - |
4249 | new_size); | |
514ac8ad CM |
4250 | |
4251 | /* | |
4252 | * update the ram bytes to properly reflect | |
4253 | * the new size of our item | |
4254 | */ | |
4255 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
e02119d5 CM |
4256 | size = |
4257 | btrfs_file_extent_calc_inline_size(size); | |
afe5fea7 | 4258 | btrfs_truncate_item(root, path, size, 1); |
27cdeb70 MX |
4259 | } else if (test_bit(BTRFS_ROOT_REF_COWS, |
4260 | &root->state)) { | |
a76a3cd4 YZ |
4261 | inode_sub_bytes(inode, item_end + 1 - |
4262 | found_key.offset); | |
9069218d | 4263 | } |
39279cc3 | 4264 | } |
179e29e4 | 4265 | delete: |
39279cc3 | 4266 | if (del_item) { |
85e21bac CM |
4267 | if (!pending_del_nr) { |
4268 | /* no pending yet, add ourselves */ | |
4269 | pending_del_slot = path->slots[0]; | |
4270 | pending_del_nr = 1; | |
4271 | } else if (pending_del_nr && | |
4272 | path->slots[0] + 1 == pending_del_slot) { | |
4273 | /* hop on the pending chunk */ | |
4274 | pending_del_nr++; | |
4275 | pending_del_slot = path->slots[0]; | |
4276 | } else { | |
d397712b | 4277 | BUG(); |
85e21bac | 4278 | } |
39279cc3 CM |
4279 | } else { |
4280 | break; | |
4281 | } | |
27cdeb70 MX |
4282 | if (found_extent && |
4283 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
4284 | root == root->fs_info->tree_root)) { | |
b9473439 | 4285 | btrfs_set_path_blocking(path); |
39279cc3 | 4286 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
4287 | extent_num_bytes, 0, |
4288 | btrfs_header_owner(leaf), | |
66d7e7f0 | 4289 | ino, extent_offset, 0); |
39279cc3 CM |
4290 | BUG_ON(ret); |
4291 | } | |
85e21bac | 4292 | |
8082510e YZ |
4293 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4294 | break; | |
4295 | ||
4296 | if (path->slots[0] == 0 || | |
4297 | path->slots[0] != pending_del_slot) { | |
8082510e YZ |
4298 | if (pending_del_nr) { |
4299 | ret = btrfs_del_items(trans, root, path, | |
4300 | pending_del_slot, | |
4301 | pending_del_nr); | |
79787eaa JM |
4302 | if (ret) { |
4303 | btrfs_abort_transaction(trans, | |
4304 | root, ret); | |
4305 | goto error; | |
4306 | } | |
8082510e YZ |
4307 | pending_del_nr = 0; |
4308 | } | |
b3b4aa74 | 4309 | btrfs_release_path(path); |
85e21bac | 4310 | goto search_again; |
8082510e YZ |
4311 | } else { |
4312 | path->slots[0]--; | |
85e21bac | 4313 | } |
39279cc3 | 4314 | } |
8082510e | 4315 | out: |
85e21bac CM |
4316 | if (pending_del_nr) { |
4317 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4318 | pending_del_nr); | |
79787eaa JM |
4319 | if (ret) |
4320 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 4321 | } |
79787eaa | 4322 | error: |
dac5705c FM |
4323 | if (last_size != (u64)-1 && |
4324 | root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
7f4f6e0a | 4325 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
39279cc3 | 4326 | btrfs_free_path(path); |
8082510e | 4327 | return err; |
39279cc3 CM |
4328 | } |
4329 | ||
4330 | /* | |
2aaa6655 JB |
4331 | * btrfs_truncate_page - read, zero a chunk and write a page |
4332 | * @inode - inode that we're zeroing | |
4333 | * @from - the offset to start zeroing | |
4334 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4335 | * offset | |
4336 | * @front - zero up to the offset instead of from the offset on | |
4337 | * | |
4338 | * This will find the page for the "from" offset and cow the page and zero the | |
4339 | * part we want to zero. This is used with truncate and hole punching. | |
39279cc3 | 4340 | */ |
2aaa6655 JB |
4341 | int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len, |
4342 | int front) | |
39279cc3 | 4343 | { |
2aaa6655 | 4344 | struct address_space *mapping = inode->i_mapping; |
db94535d | 4345 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4346 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4347 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4348 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 4349 | char *kaddr; |
db94535d | 4350 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
4351 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
4352 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
4353 | struct page *page; | |
3b16a4e3 | 4354 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4355 | int ret = 0; |
a52d9a80 | 4356 | u64 page_start; |
e6dcd2dc | 4357 | u64 page_end; |
39279cc3 | 4358 | |
2aaa6655 JB |
4359 | if ((offset & (blocksize - 1)) == 0 && |
4360 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4361 | goto out; |
0ca1f7ce | 4362 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
4363 | if (ret) |
4364 | goto out; | |
39279cc3 | 4365 | |
211c17f5 | 4366 | again: |
3b16a4e3 | 4367 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4368 | if (!page) { |
0ca1f7ce | 4369 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
ac6a2b36 | 4370 | ret = -ENOMEM; |
39279cc3 | 4371 | goto out; |
5d5e103a | 4372 | } |
e6dcd2dc CM |
4373 | |
4374 | page_start = page_offset(page); | |
4375 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
4376 | ||
39279cc3 | 4377 | if (!PageUptodate(page)) { |
9ebefb18 | 4378 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4379 | lock_page(page); |
211c17f5 CM |
4380 | if (page->mapping != mapping) { |
4381 | unlock_page(page); | |
4382 | page_cache_release(page); | |
4383 | goto again; | |
4384 | } | |
39279cc3 CM |
4385 | if (!PageUptodate(page)) { |
4386 | ret = -EIO; | |
89642229 | 4387 | goto out_unlock; |
39279cc3 CM |
4388 | } |
4389 | } | |
211c17f5 | 4390 | wait_on_page_writeback(page); |
e6dcd2dc | 4391 | |
d0082371 | 4392 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
4393 | set_page_extent_mapped(page); |
4394 | ||
4395 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
4396 | if (ordered) { | |
2ac55d41 JB |
4397 | unlock_extent_cached(io_tree, page_start, page_end, |
4398 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
4399 | unlock_page(page); |
4400 | page_cache_release(page); | |
eb84ae03 | 4401 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4402 | btrfs_put_ordered_extent(ordered); |
4403 | goto again; | |
4404 | } | |
4405 | ||
2ac55d41 | 4406 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
4407 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4408 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 4409 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 4410 | |
2ac55d41 JB |
4411 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
4412 | &cached_state); | |
9ed74f2d | 4413 | if (ret) { |
2ac55d41 JB |
4414 | unlock_extent_cached(io_tree, page_start, page_end, |
4415 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
4416 | goto out_unlock; |
4417 | } | |
4418 | ||
e6dcd2dc | 4419 | if (offset != PAGE_CACHE_SIZE) { |
2aaa6655 JB |
4420 | if (!len) |
4421 | len = PAGE_CACHE_SIZE - offset; | |
e6dcd2dc | 4422 | kaddr = kmap(page); |
2aaa6655 JB |
4423 | if (front) |
4424 | memset(kaddr, 0, offset); | |
4425 | else | |
4426 | memset(kaddr + offset, 0, len); | |
e6dcd2dc CM |
4427 | flush_dcache_page(page); |
4428 | kunmap(page); | |
4429 | } | |
247e743c | 4430 | ClearPageChecked(page); |
e6dcd2dc | 4431 | set_page_dirty(page); |
2ac55d41 JB |
4432 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
4433 | GFP_NOFS); | |
39279cc3 | 4434 | |
89642229 | 4435 | out_unlock: |
5d5e103a | 4436 | if (ret) |
0ca1f7ce | 4437 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
4438 | unlock_page(page); |
4439 | page_cache_release(page); | |
4440 | out: | |
4441 | return ret; | |
4442 | } | |
4443 | ||
16e7549f JB |
4444 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4445 | u64 offset, u64 len) | |
4446 | { | |
4447 | struct btrfs_trans_handle *trans; | |
4448 | int ret; | |
4449 | ||
4450 | /* | |
4451 | * Still need to make sure the inode looks like it's been updated so | |
4452 | * that any holes get logged if we fsync. | |
4453 | */ | |
4454 | if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) { | |
4455 | BTRFS_I(inode)->last_trans = root->fs_info->generation; | |
4456 | BTRFS_I(inode)->last_sub_trans = root->log_transid; | |
4457 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4458 | return 0; | |
4459 | } | |
4460 | ||
4461 | /* | |
4462 | * 1 - for the one we're dropping | |
4463 | * 1 - for the one we're adding | |
4464 | * 1 - for updating the inode. | |
4465 | */ | |
4466 | trans = btrfs_start_transaction(root, 3); | |
4467 | if (IS_ERR(trans)) | |
4468 | return PTR_ERR(trans); | |
4469 | ||
4470 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4471 | if (ret) { | |
4472 | btrfs_abort_transaction(trans, root, ret); | |
4473 | btrfs_end_transaction(trans, root); | |
4474 | return ret; | |
4475 | } | |
4476 | ||
4477 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset, | |
4478 | 0, 0, len, 0, len, 0, 0, 0); | |
4479 | if (ret) | |
4480 | btrfs_abort_transaction(trans, root, ret); | |
4481 | else | |
4482 | btrfs_update_inode(trans, root, inode); | |
4483 | btrfs_end_transaction(trans, root); | |
4484 | return ret; | |
4485 | } | |
4486 | ||
695a0d0d JB |
4487 | /* |
4488 | * This function puts in dummy file extents for the area we're creating a hole | |
4489 | * for. So if we are truncating this file to a larger size we need to insert | |
4490 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4491 | * the range between oldsize and size | |
4492 | */ | |
a41ad394 | 4493 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4494 | { |
9036c102 YZ |
4495 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4496 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4497 | struct extent_map *em = NULL; |
2ac55d41 | 4498 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4499 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
fda2832f QW |
4500 | u64 hole_start = ALIGN(oldsize, root->sectorsize); |
4501 | u64 block_end = ALIGN(size, root->sectorsize); | |
9036c102 YZ |
4502 | u64 last_byte; |
4503 | u64 cur_offset; | |
4504 | u64 hole_size; | |
9ed74f2d | 4505 | int err = 0; |
39279cc3 | 4506 | |
a71754fc JB |
4507 | /* |
4508 | * If our size started in the middle of a page we need to zero out the | |
4509 | * rest of the page before we expand the i_size, otherwise we could | |
4510 | * expose stale data. | |
4511 | */ | |
4512 | err = btrfs_truncate_page(inode, oldsize, 0, 0); | |
4513 | if (err) | |
4514 | return err; | |
4515 | ||
9036c102 YZ |
4516 | if (size <= hole_start) |
4517 | return 0; | |
4518 | ||
9036c102 YZ |
4519 | while (1) { |
4520 | struct btrfs_ordered_extent *ordered; | |
fa7c1494 | 4521 | |
2ac55d41 | 4522 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
d0082371 | 4523 | &cached_state); |
fa7c1494 MX |
4524 | ordered = btrfs_lookup_ordered_range(inode, hole_start, |
4525 | block_end - hole_start); | |
9036c102 YZ |
4526 | if (!ordered) |
4527 | break; | |
2ac55d41 JB |
4528 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
4529 | &cached_state, GFP_NOFS); | |
fa7c1494 | 4530 | btrfs_start_ordered_extent(inode, ordered, 1); |
9036c102 YZ |
4531 | btrfs_put_ordered_extent(ordered); |
4532 | } | |
39279cc3 | 4533 | |
9036c102 YZ |
4534 | cur_offset = hole_start; |
4535 | while (1) { | |
4536 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
4537 | block_end - cur_offset, 0); | |
79787eaa JM |
4538 | if (IS_ERR(em)) { |
4539 | err = PTR_ERR(em); | |
f2767956 | 4540 | em = NULL; |
79787eaa JM |
4541 | break; |
4542 | } | |
9036c102 | 4543 | last_byte = min(extent_map_end(em), block_end); |
fda2832f | 4544 | last_byte = ALIGN(last_byte , root->sectorsize); |
8082510e | 4545 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4546 | struct extent_map *hole_em; |
9036c102 | 4547 | hole_size = last_byte - cur_offset; |
9ed74f2d | 4548 | |
16e7549f JB |
4549 | err = maybe_insert_hole(root, inode, cur_offset, |
4550 | hole_size); | |
4551 | if (err) | |
3893e33b | 4552 | break; |
5dc562c5 JB |
4553 | btrfs_drop_extent_cache(inode, cur_offset, |
4554 | cur_offset + hole_size - 1, 0); | |
4555 | hole_em = alloc_extent_map(); | |
4556 | if (!hole_em) { | |
4557 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4558 | &BTRFS_I(inode)->runtime_flags); | |
4559 | goto next; | |
4560 | } | |
4561 | hole_em->start = cur_offset; | |
4562 | hole_em->len = hole_size; | |
4563 | hole_em->orig_start = cur_offset; | |
8082510e | 4564 | |
5dc562c5 JB |
4565 | hole_em->block_start = EXTENT_MAP_HOLE; |
4566 | hole_em->block_len = 0; | |
b4939680 | 4567 | hole_em->orig_block_len = 0; |
cc95bef6 | 4568 | hole_em->ram_bytes = hole_size; |
5dc562c5 JB |
4569 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
4570 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
16e7549f | 4571 | hole_em->generation = root->fs_info->generation; |
8082510e | 4572 | |
5dc562c5 JB |
4573 | while (1) { |
4574 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4575 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4576 | write_unlock(&em_tree->lock); |
4577 | if (err != -EEXIST) | |
4578 | break; | |
4579 | btrfs_drop_extent_cache(inode, cur_offset, | |
4580 | cur_offset + | |
4581 | hole_size - 1, 0); | |
4582 | } | |
4583 | free_extent_map(hole_em); | |
9036c102 | 4584 | } |
16e7549f | 4585 | next: |
9036c102 | 4586 | free_extent_map(em); |
a22285a6 | 4587 | em = NULL; |
9036c102 | 4588 | cur_offset = last_byte; |
8082510e | 4589 | if (cur_offset >= block_end) |
9036c102 YZ |
4590 | break; |
4591 | } | |
a22285a6 | 4592 | free_extent_map(em); |
2ac55d41 JB |
4593 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
4594 | GFP_NOFS); | |
9036c102 YZ |
4595 | return err; |
4596 | } | |
39279cc3 | 4597 | |
3972f260 | 4598 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4599 | { |
f4a2f4c5 MX |
4600 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4601 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4602 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4603 | loff_t newsize = attr->ia_size; |
4604 | int mask = attr->ia_valid; | |
8082510e YZ |
4605 | int ret; |
4606 | ||
3972f260 ES |
4607 | /* |
4608 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4609 | * special case where we need to update the times despite not having | |
4610 | * these flags set. For all other operations the VFS set these flags | |
4611 | * explicitly if it wants a timestamp update. | |
4612 | */ | |
dff6efc3 CH |
4613 | if (newsize != oldsize) { |
4614 | inode_inc_iversion(inode); | |
4615 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
4616 | inode->i_ctime = inode->i_mtime = | |
4617 | current_fs_time(inode->i_sb); | |
4618 | } | |
3972f260 | 4619 | |
a41ad394 | 4620 | if (newsize > oldsize) { |
7caef267 | 4621 | truncate_pagecache(inode, newsize); |
a41ad394 | 4622 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
f4a2f4c5 | 4623 | if (ret) |
8082510e | 4624 | return ret; |
8082510e | 4625 | |
f4a2f4c5 MX |
4626 | trans = btrfs_start_transaction(root, 1); |
4627 | if (IS_ERR(trans)) | |
4628 | return PTR_ERR(trans); | |
4629 | ||
4630 | i_size_write(inode, newsize); | |
4631 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
4632 | ret = btrfs_update_inode(trans, root, inode); | |
7ad85bb7 | 4633 | btrfs_end_transaction(trans, root); |
a41ad394 | 4634 | } else { |
8082510e | 4635 | |
a41ad394 JB |
4636 | /* |
4637 | * We're truncating a file that used to have good data down to | |
4638 | * zero. Make sure it gets into the ordered flush list so that | |
4639 | * any new writes get down to disk quickly. | |
4640 | */ | |
4641 | if (newsize == 0) | |
72ac3c0d JB |
4642 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
4643 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 4644 | |
f3fe820c JB |
4645 | /* |
4646 | * 1 for the orphan item we're going to add | |
4647 | * 1 for the orphan item deletion. | |
4648 | */ | |
4649 | trans = btrfs_start_transaction(root, 2); | |
4650 | if (IS_ERR(trans)) | |
4651 | return PTR_ERR(trans); | |
4652 | ||
4653 | /* | |
4654 | * We need to do this in case we fail at _any_ point during the | |
4655 | * actual truncate. Once we do the truncate_setsize we could | |
4656 | * invalidate pages which forces any outstanding ordered io to | |
4657 | * be instantly completed which will give us extents that need | |
4658 | * to be truncated. If we fail to get an orphan inode down we | |
4659 | * could have left over extents that were never meant to live, | |
4660 | * so we need to garuntee from this point on that everything | |
4661 | * will be consistent. | |
4662 | */ | |
4663 | ret = btrfs_orphan_add(trans, inode); | |
4664 | btrfs_end_transaction(trans, root); | |
4665 | if (ret) | |
4666 | return ret; | |
4667 | ||
a41ad394 JB |
4668 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
4669 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
4670 | |
4671 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
4672 | btrfs_inode_block_unlocked_dio(inode); | |
4673 | inode_dio_wait(inode); | |
4674 | btrfs_inode_resume_unlocked_dio(inode); | |
4675 | ||
a41ad394 | 4676 | ret = btrfs_truncate(inode); |
7f4f6e0a JB |
4677 | if (ret && inode->i_nlink) { |
4678 | int err; | |
4679 | ||
4680 | /* | |
4681 | * failed to truncate, disk_i_size is only adjusted down | |
4682 | * as we remove extents, so it should represent the true | |
4683 | * size of the inode, so reset the in memory size and | |
4684 | * delete our orphan entry. | |
4685 | */ | |
4686 | trans = btrfs_join_transaction(root); | |
4687 | if (IS_ERR(trans)) { | |
4688 | btrfs_orphan_del(NULL, inode); | |
4689 | return ret; | |
4690 | } | |
4691 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
4692 | err = btrfs_orphan_del(trans, inode); | |
4693 | if (err) | |
4694 | btrfs_abort_transaction(trans, root, err); | |
4695 | btrfs_end_transaction(trans, root); | |
4696 | } | |
8082510e YZ |
4697 | } |
4698 | ||
a41ad394 | 4699 | return ret; |
8082510e YZ |
4700 | } |
4701 | ||
9036c102 YZ |
4702 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
4703 | { | |
4704 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 4705 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 4706 | int err; |
39279cc3 | 4707 | |
b83cc969 LZ |
4708 | if (btrfs_root_readonly(root)) |
4709 | return -EROFS; | |
4710 | ||
9036c102 YZ |
4711 | err = inode_change_ok(inode, attr); |
4712 | if (err) | |
4713 | return err; | |
2bf5a725 | 4714 | |
5a3f23d5 | 4715 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 4716 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
4717 | if (err) |
4718 | return err; | |
39279cc3 | 4719 | } |
9036c102 | 4720 | |
1025774c CH |
4721 | if (attr->ia_valid) { |
4722 | setattr_copy(inode, attr); | |
0c4d2d95 | 4723 | inode_inc_iversion(inode); |
22c44fe6 | 4724 | err = btrfs_dirty_inode(inode); |
1025774c | 4725 | |
22c44fe6 | 4726 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 4727 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 4728 | } |
33268eaf | 4729 | |
39279cc3 CM |
4730 | return err; |
4731 | } | |
61295eb8 | 4732 | |
131e404a FDBM |
4733 | /* |
4734 | * While truncating the inode pages during eviction, we get the VFS calling | |
4735 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
4736 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
4737 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
4738 | * extent_state structures over and over, wasting lots of time. | |
4739 | * | |
4740 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
4741 | * those expensive operations on a per page basis and do only the ordered io | |
4742 | * finishing, while we release here the extent_map and extent_state structures, | |
4743 | * without the excessive merging and splitting. | |
4744 | */ | |
4745 | static void evict_inode_truncate_pages(struct inode *inode) | |
4746 | { | |
4747 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
4748 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
4749 | struct rb_node *node; | |
4750 | ||
4751 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 4752 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
4753 | |
4754 | write_lock(&map_tree->lock); | |
4755 | while (!RB_EMPTY_ROOT(&map_tree->map)) { | |
4756 | struct extent_map *em; | |
4757 | ||
4758 | node = rb_first(&map_tree->map); | |
4759 | em = rb_entry(node, struct extent_map, rb_node); | |
180589ef WS |
4760 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
4761 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
4762 | remove_extent_mapping(map_tree, em); |
4763 | free_extent_map(em); | |
7064dd5c FM |
4764 | if (need_resched()) { |
4765 | write_unlock(&map_tree->lock); | |
4766 | cond_resched(); | |
4767 | write_lock(&map_tree->lock); | |
4768 | } | |
131e404a FDBM |
4769 | } |
4770 | write_unlock(&map_tree->lock); | |
4771 | ||
4772 | spin_lock(&io_tree->lock); | |
4773 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
4774 | struct extent_state *state; | |
4775 | struct extent_state *cached_state = NULL; | |
4776 | ||
4777 | node = rb_first(&io_tree->state); | |
4778 | state = rb_entry(node, struct extent_state, rb_node); | |
4779 | atomic_inc(&state->refs); | |
4780 | spin_unlock(&io_tree->lock); | |
4781 | ||
4782 | lock_extent_bits(io_tree, state->start, state->end, | |
4783 | 0, &cached_state); | |
4784 | clear_extent_bit(io_tree, state->start, state->end, | |
4785 | EXTENT_LOCKED | EXTENT_DIRTY | | |
4786 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
4787 | EXTENT_DEFRAG, 1, 1, | |
4788 | &cached_state, GFP_NOFS); | |
4789 | free_extent_state(state); | |
4790 | ||
7064dd5c | 4791 | cond_resched(); |
131e404a FDBM |
4792 | spin_lock(&io_tree->lock); |
4793 | } | |
4794 | spin_unlock(&io_tree->lock); | |
4795 | } | |
4796 | ||
bd555975 | 4797 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
4798 | { |
4799 | struct btrfs_trans_handle *trans; | |
4800 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 4801 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 4802 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 CM |
4803 | int ret; |
4804 | ||
1abe9b8a | 4805 | trace_btrfs_inode_evict(inode); |
4806 | ||
131e404a FDBM |
4807 | evict_inode_truncate_pages(inode); |
4808 | ||
69e9c6c6 SB |
4809 | if (inode->i_nlink && |
4810 | ((btrfs_root_refs(&root->root_item) != 0 && | |
4811 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
4812 | btrfs_is_free_space_inode(inode))) | |
bd555975 AV |
4813 | goto no_delete; |
4814 | ||
39279cc3 | 4815 | if (is_bad_inode(inode)) { |
7b128766 | 4816 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
4817 | goto no_delete; |
4818 | } | |
bd555975 | 4819 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 4820 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 4821 | |
f612496b MX |
4822 | btrfs_free_io_failure_record(inode, 0, (u64)-1); |
4823 | ||
c71bf099 | 4824 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 4825 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 4826 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
4827 | goto no_delete; |
4828 | } | |
4829 | ||
76dda93c | 4830 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
4831 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
4832 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
4833 | goto no_delete; |
4834 | } | |
4835 | ||
0e8c36a9 MX |
4836 | ret = btrfs_commit_inode_delayed_inode(inode); |
4837 | if (ret) { | |
4838 | btrfs_orphan_del(NULL, inode); | |
4839 | goto no_delete; | |
4840 | } | |
4841 | ||
66d8f3dd | 4842 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
4843 | if (!rsv) { |
4844 | btrfs_orphan_del(NULL, inode); | |
4845 | goto no_delete; | |
4846 | } | |
4a338542 | 4847 | rsv->size = min_size; |
ca7e70f5 | 4848 | rsv->failfast = 1; |
726c35fa | 4849 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 4850 | |
dbe674a9 | 4851 | btrfs_i_size_write(inode, 0); |
5f39d397 | 4852 | |
4289a667 | 4853 | /* |
8407aa46 MX |
4854 | * This is a bit simpler than btrfs_truncate since we've already |
4855 | * reserved our space for our orphan item in the unlink, so we just | |
4856 | * need to reserve some slack space in case we add bytes and update | |
4857 | * inode item when doing the truncate. | |
4289a667 | 4858 | */ |
8082510e | 4859 | while (1) { |
08e007d2 MX |
4860 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
4861 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
4862 | |
4863 | /* | |
4864 | * Try and steal from the global reserve since we will | |
4865 | * likely not use this space anyway, we want to try as | |
4866 | * hard as possible to get this to work. | |
4867 | */ | |
4868 | if (ret) | |
4869 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
d68fc57b | 4870 | |
d68fc57b | 4871 | if (ret) { |
c2cf52eb SK |
4872 | btrfs_warn(root->fs_info, |
4873 | "Could not get space for a delete, will truncate on mount %d", | |
4874 | ret); | |
4289a667 JB |
4875 | btrfs_orphan_del(NULL, inode); |
4876 | btrfs_free_block_rsv(root, rsv); | |
4877 | goto no_delete; | |
d68fc57b | 4878 | } |
7b128766 | 4879 | |
0e8c36a9 | 4880 | trans = btrfs_join_transaction(root); |
4289a667 JB |
4881 | if (IS_ERR(trans)) { |
4882 | btrfs_orphan_del(NULL, inode); | |
4883 | btrfs_free_block_rsv(root, rsv); | |
4884 | goto no_delete; | |
d68fc57b | 4885 | } |
7b128766 | 4886 | |
4289a667 JB |
4887 | trans->block_rsv = rsv; |
4888 | ||
d68fc57b | 4889 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
ca7e70f5 | 4890 | if (ret != -ENOSPC) |
8082510e | 4891 | break; |
85e21bac | 4892 | |
8407aa46 | 4893 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e YZ |
4894 | btrfs_end_transaction(trans, root); |
4895 | trans = NULL; | |
b53d3f5d | 4896 | btrfs_btree_balance_dirty(root); |
8082510e | 4897 | } |
5f39d397 | 4898 | |
4289a667 JB |
4899 | btrfs_free_block_rsv(root, rsv); |
4900 | ||
4ef31a45 JB |
4901 | /* |
4902 | * Errors here aren't a big deal, it just means we leave orphan items | |
4903 | * in the tree. They will be cleaned up on the next mount. | |
4904 | */ | |
8082510e | 4905 | if (ret == 0) { |
4289a667 | 4906 | trans->block_rsv = root->orphan_block_rsv; |
4ef31a45 JB |
4907 | btrfs_orphan_del(trans, inode); |
4908 | } else { | |
4909 | btrfs_orphan_del(NULL, inode); | |
8082510e | 4910 | } |
54aa1f4d | 4911 | |
4289a667 | 4912 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
4913 | if (!(root == root->fs_info->tree_root || |
4914 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 4915 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 4916 | |
54aa1f4d | 4917 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4918 | btrfs_btree_balance_dirty(root); |
39279cc3 | 4919 | no_delete: |
89042e5a | 4920 | btrfs_remove_delayed_node(inode); |
dbd5768f | 4921 | clear_inode(inode); |
8082510e | 4922 | return; |
39279cc3 CM |
4923 | } |
4924 | ||
4925 | /* | |
4926 | * this returns the key found in the dir entry in the location pointer. | |
4927 | * If no dir entries were found, location->objectid is 0. | |
4928 | */ | |
4929 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
4930 | struct btrfs_key *location) | |
4931 | { | |
4932 | const char *name = dentry->d_name.name; | |
4933 | int namelen = dentry->d_name.len; | |
4934 | struct btrfs_dir_item *di; | |
4935 | struct btrfs_path *path; | |
4936 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 4937 | int ret = 0; |
39279cc3 CM |
4938 | |
4939 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
4940 | if (!path) |
4941 | return -ENOMEM; | |
3954401f | 4942 | |
33345d01 | 4943 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 4944 | namelen, 0); |
0d9f7f3e Y |
4945 | if (IS_ERR(di)) |
4946 | ret = PTR_ERR(di); | |
d397712b | 4947 | |
c704005d | 4948 | if (IS_ERR_OR_NULL(di)) |
3954401f | 4949 | goto out_err; |
d397712b | 4950 | |
5f39d397 | 4951 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 4952 | out: |
39279cc3 CM |
4953 | btrfs_free_path(path); |
4954 | return ret; | |
3954401f CM |
4955 | out_err: |
4956 | location->objectid = 0; | |
4957 | goto out; | |
39279cc3 CM |
4958 | } |
4959 | ||
4960 | /* | |
4961 | * when we hit a tree root in a directory, the btrfs part of the inode | |
4962 | * needs to be changed to reflect the root directory of the tree root. This | |
4963 | * is kind of like crossing a mount point. | |
4964 | */ | |
4965 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
4966 | struct inode *dir, |
4967 | struct dentry *dentry, | |
4968 | struct btrfs_key *location, | |
4969 | struct btrfs_root **sub_root) | |
39279cc3 | 4970 | { |
4df27c4d YZ |
4971 | struct btrfs_path *path; |
4972 | struct btrfs_root *new_root; | |
4973 | struct btrfs_root_ref *ref; | |
4974 | struct extent_buffer *leaf; | |
4975 | int ret; | |
4976 | int err = 0; | |
39279cc3 | 4977 | |
4df27c4d YZ |
4978 | path = btrfs_alloc_path(); |
4979 | if (!path) { | |
4980 | err = -ENOMEM; | |
4981 | goto out; | |
4982 | } | |
39279cc3 | 4983 | |
4df27c4d | 4984 | err = -ENOENT; |
75ac2dd9 KN |
4985 | ret = btrfs_find_item(root->fs_info->tree_root, path, |
4986 | BTRFS_I(dir)->root->root_key.objectid, | |
4987 | location->objectid, BTRFS_ROOT_REF_KEY, NULL); | |
4df27c4d YZ |
4988 | if (ret) { |
4989 | if (ret < 0) | |
4990 | err = ret; | |
4991 | goto out; | |
4992 | } | |
39279cc3 | 4993 | |
4df27c4d YZ |
4994 | leaf = path->nodes[0]; |
4995 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 4996 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
4997 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
4998 | goto out; | |
39279cc3 | 4999 | |
4df27c4d YZ |
5000 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5001 | (unsigned long)(ref + 1), | |
5002 | dentry->d_name.len); | |
5003 | if (ret) | |
5004 | goto out; | |
5005 | ||
b3b4aa74 | 5006 | btrfs_release_path(path); |
4df27c4d YZ |
5007 | |
5008 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
5009 | if (IS_ERR(new_root)) { | |
5010 | err = PTR_ERR(new_root); | |
5011 | goto out; | |
5012 | } | |
5013 | ||
4df27c4d YZ |
5014 | *sub_root = new_root; |
5015 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5016 | location->type = BTRFS_INODE_ITEM_KEY; | |
5017 | location->offset = 0; | |
5018 | err = 0; | |
5019 | out: | |
5020 | btrfs_free_path(path); | |
5021 | return err; | |
39279cc3 CM |
5022 | } |
5023 | ||
5d4f98a2 YZ |
5024 | static void inode_tree_add(struct inode *inode) |
5025 | { | |
5026 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5027 | struct btrfs_inode *entry; | |
03e860bd FNP |
5028 | struct rb_node **p; |
5029 | struct rb_node *parent; | |
cef21937 | 5030 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
33345d01 | 5031 | u64 ino = btrfs_ino(inode); |
5d4f98a2 | 5032 | |
1d3382cb | 5033 | if (inode_unhashed(inode)) |
76dda93c | 5034 | return; |
e1409cef | 5035 | parent = NULL; |
5d4f98a2 | 5036 | spin_lock(&root->inode_lock); |
e1409cef | 5037 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5038 | while (*p) { |
5039 | parent = *p; | |
5040 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5041 | ||
33345d01 | 5042 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 5043 | p = &parent->rb_left; |
33345d01 | 5044 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 5045 | p = &parent->rb_right; |
5d4f98a2 YZ |
5046 | else { |
5047 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5048 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5049 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5050 | RB_CLEAR_NODE(parent); |
5051 | spin_unlock(&root->inode_lock); | |
cef21937 | 5052 | return; |
5d4f98a2 YZ |
5053 | } |
5054 | } | |
cef21937 FDBM |
5055 | rb_link_node(new, parent, p); |
5056 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5057 | spin_unlock(&root->inode_lock); |
5058 | } | |
5059 | ||
5060 | static void inode_tree_del(struct inode *inode) | |
5061 | { | |
5062 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 5063 | int empty = 0; |
5d4f98a2 | 5064 | |
03e860bd | 5065 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5066 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5067 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5068 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5069 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5070 | } |
03e860bd | 5071 | spin_unlock(&root->inode_lock); |
76dda93c | 5072 | |
69e9c6c6 | 5073 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5074 | synchronize_srcu(&root->fs_info->subvol_srcu); |
5075 | spin_lock(&root->inode_lock); | |
5076 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5077 | spin_unlock(&root->inode_lock); | |
5078 | if (empty) | |
5079 | btrfs_add_dead_root(root); | |
5080 | } | |
5081 | } | |
5082 | ||
143bede5 | 5083 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
5084 | { |
5085 | struct rb_node *node; | |
5086 | struct rb_node *prev; | |
5087 | struct btrfs_inode *entry; | |
5088 | struct inode *inode; | |
5089 | u64 objectid = 0; | |
5090 | ||
7813b3db LB |
5091 | if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
5092 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
76dda93c YZ |
5093 | |
5094 | spin_lock(&root->inode_lock); | |
5095 | again: | |
5096 | node = root->inode_tree.rb_node; | |
5097 | prev = NULL; | |
5098 | while (node) { | |
5099 | prev = node; | |
5100 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
5101 | ||
33345d01 | 5102 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 5103 | node = node->rb_left; |
33345d01 | 5104 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
5105 | node = node->rb_right; |
5106 | else | |
5107 | break; | |
5108 | } | |
5109 | if (!node) { | |
5110 | while (prev) { | |
5111 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 5112 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
5113 | node = prev; |
5114 | break; | |
5115 | } | |
5116 | prev = rb_next(prev); | |
5117 | } | |
5118 | } | |
5119 | while (node) { | |
5120 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 5121 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
5122 | inode = igrab(&entry->vfs_inode); |
5123 | if (inode) { | |
5124 | spin_unlock(&root->inode_lock); | |
5125 | if (atomic_read(&inode->i_count) > 1) | |
5126 | d_prune_aliases(inode); | |
5127 | /* | |
45321ac5 | 5128 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
5129 | * the inode cache when its usage count |
5130 | * hits zero. | |
5131 | */ | |
5132 | iput(inode); | |
5133 | cond_resched(); | |
5134 | spin_lock(&root->inode_lock); | |
5135 | goto again; | |
5136 | } | |
5137 | ||
5138 | if (cond_resched_lock(&root->inode_lock)) | |
5139 | goto again; | |
5140 | ||
5141 | node = rb_next(node); | |
5142 | } | |
5143 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
5144 | } |
5145 | ||
e02119d5 CM |
5146 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5147 | { | |
5148 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5149 | inode->i_ino = args->location->objectid; |
5150 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5151 | sizeof(*args->location)); | |
e02119d5 | 5152 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5153 | return 0; |
5154 | } | |
5155 | ||
5156 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5157 | { | |
5158 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5159 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5160 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5161 | } |
5162 | ||
5d4f98a2 | 5163 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5164 | struct btrfs_key *location, |
5d4f98a2 | 5165 | struct btrfs_root *root) |
39279cc3 CM |
5166 | { |
5167 | struct inode *inode; | |
5168 | struct btrfs_iget_args args; | |
90d3e592 | 5169 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5170 | |
90d3e592 | 5171 | args.location = location; |
39279cc3 CM |
5172 | args.root = root; |
5173 | ||
778ba82b | 5174 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5175 | btrfs_init_locked_inode, |
5176 | (void *)&args); | |
5177 | return inode; | |
5178 | } | |
5179 | ||
1a54ef8c BR |
5180 | /* Get an inode object given its location and corresponding root. |
5181 | * Returns in *is_new if the inode was read from disk | |
5182 | */ | |
5183 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 5184 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
5185 | { |
5186 | struct inode *inode; | |
5187 | ||
90d3e592 | 5188 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5189 | if (!inode) |
5d4f98a2 | 5190 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5191 | |
5192 | if (inode->i_state & I_NEW) { | |
1a54ef8c | 5193 | btrfs_read_locked_inode(inode); |
1748f843 MF |
5194 | if (!is_bad_inode(inode)) { |
5195 | inode_tree_add(inode); | |
5196 | unlock_new_inode(inode); | |
5197 | if (new) | |
5198 | *new = 1; | |
5199 | } else { | |
e0b6d65b ST |
5200 | unlock_new_inode(inode); |
5201 | iput(inode); | |
5202 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
5203 | } |
5204 | } | |
5205 | ||
1a54ef8c BR |
5206 | return inode; |
5207 | } | |
5208 | ||
4df27c4d YZ |
5209 | static struct inode *new_simple_dir(struct super_block *s, |
5210 | struct btrfs_key *key, | |
5211 | struct btrfs_root *root) | |
5212 | { | |
5213 | struct inode *inode = new_inode(s); | |
5214 | ||
5215 | if (!inode) | |
5216 | return ERR_PTR(-ENOMEM); | |
5217 | ||
4df27c4d YZ |
5218 | BTRFS_I(inode)->root = root; |
5219 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5220 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5221 | |
5222 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5223 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
5224 | inode->i_fop = &simple_dir_operations; |
5225 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
5226 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
5227 | ||
5228 | return inode; | |
5229 | } | |
5230 | ||
3de4586c | 5231 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5232 | { |
d397712b | 5233 | struct inode *inode; |
4df27c4d | 5234 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5235 | struct btrfs_root *sub_root = root; |
5236 | struct btrfs_key location; | |
76dda93c | 5237 | int index; |
b4aff1f8 | 5238 | int ret = 0; |
39279cc3 CM |
5239 | |
5240 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5241 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5242 | |
39e3c955 | 5243 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
5244 | if (ret < 0) |
5245 | return ERR_PTR(ret); | |
5f39d397 | 5246 | |
4df27c4d | 5247 | if (location.objectid == 0) |
5662344b | 5248 | return ERR_PTR(-ENOENT); |
4df27c4d YZ |
5249 | |
5250 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 5251 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
5252 | return inode; |
5253 | } | |
5254 | ||
5255 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
5256 | ||
76dda93c | 5257 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
5258 | ret = fixup_tree_root_location(root, dir, dentry, |
5259 | &location, &sub_root); | |
5260 | if (ret < 0) { | |
5261 | if (ret != -ENOENT) | |
5262 | inode = ERR_PTR(ret); | |
5263 | else | |
5264 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5265 | } else { | |
73f73415 | 5266 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5267 | } |
76dda93c YZ |
5268 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
5269 | ||
34d19bad | 5270 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
5271 | down_read(&root->fs_info->cleanup_work_sem); |
5272 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 5273 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 5274 | up_read(&root->fs_info->cleanup_work_sem); |
01cd3367 JB |
5275 | if (ret) { |
5276 | iput(inode); | |
66b4ffd1 | 5277 | inode = ERR_PTR(ret); |
01cd3367 | 5278 | } |
c71bf099 YZ |
5279 | } |
5280 | ||
3de4586c CM |
5281 | return inode; |
5282 | } | |
5283 | ||
fe15ce44 | 5284 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5285 | { |
5286 | struct btrfs_root *root; | |
848cce0d | 5287 | struct inode *inode = dentry->d_inode; |
76dda93c | 5288 | |
848cce0d LZ |
5289 | if (!inode && !IS_ROOT(dentry)) |
5290 | inode = dentry->d_parent->d_inode; | |
76dda93c | 5291 | |
848cce0d LZ |
5292 | if (inode) { |
5293 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5294 | if (btrfs_root_refs(&root->root_item) == 0) |
5295 | return 1; | |
848cce0d LZ |
5296 | |
5297 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
5298 | return 1; | |
efefb143 | 5299 | } |
76dda93c YZ |
5300 | return 0; |
5301 | } | |
5302 | ||
b4aff1f8 JB |
5303 | static void btrfs_dentry_release(struct dentry *dentry) |
5304 | { | |
944a4515 | 5305 | kfree(dentry->d_fsdata); |
b4aff1f8 JB |
5306 | } |
5307 | ||
3de4586c | 5308 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5309 | unsigned int flags) |
3de4586c | 5310 | { |
5662344b | 5311 | struct inode *inode; |
a66e7cc6 | 5312 | |
5662344b TI |
5313 | inode = btrfs_lookup_dentry(dir, dentry); |
5314 | if (IS_ERR(inode)) { | |
5315 | if (PTR_ERR(inode) == -ENOENT) | |
5316 | inode = NULL; | |
5317 | else | |
5318 | return ERR_CAST(inode); | |
5319 | } | |
5320 | ||
3a0dfa6a | 5321 | return d_materialise_unique(dentry, inode); |
39279cc3 CM |
5322 | } |
5323 | ||
16cdcec7 | 5324 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
5325 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
5326 | }; | |
5327 | ||
9cdda8d3 | 5328 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5329 | { |
9cdda8d3 | 5330 | struct inode *inode = file_inode(file); |
39279cc3 CM |
5331 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5332 | struct btrfs_item *item; | |
5333 | struct btrfs_dir_item *di; | |
5334 | struct btrfs_key key; | |
5f39d397 | 5335 | struct btrfs_key found_key; |
39279cc3 | 5336 | struct btrfs_path *path; |
16cdcec7 MX |
5337 | struct list_head ins_list; |
5338 | struct list_head del_list; | |
39279cc3 | 5339 | int ret; |
5f39d397 | 5340 | struct extent_buffer *leaf; |
39279cc3 | 5341 | int slot; |
39279cc3 CM |
5342 | unsigned char d_type; |
5343 | int over = 0; | |
5344 | u32 di_cur; | |
5345 | u32 di_total; | |
5346 | u32 di_len; | |
5347 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
5348 | char tmp_name[32]; |
5349 | char *name_ptr; | |
5350 | int name_len; | |
9cdda8d3 | 5351 | int is_curr = 0; /* ctx->pos points to the current index? */ |
39279cc3 CM |
5352 | |
5353 | /* FIXME, use a real flag for deciding about the key type */ | |
5354 | if (root->fs_info->tree_root == root) | |
5355 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 5356 | |
9cdda8d3 AV |
5357 | if (!dir_emit_dots(file, ctx)) |
5358 | return 0; | |
5359 | ||
49593bfa | 5360 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5361 | if (!path) |
5362 | return -ENOMEM; | |
ff5714cc | 5363 | |
026fd317 | 5364 | path->reada = 1; |
49593bfa | 5365 | |
16cdcec7 MX |
5366 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5367 | INIT_LIST_HEAD(&ins_list); | |
5368 | INIT_LIST_HEAD(&del_list); | |
5369 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
5370 | } | |
5371 | ||
962a298f | 5372 | key.type = key_type; |
9cdda8d3 | 5373 | key.offset = ctx->pos; |
33345d01 | 5374 | key.objectid = btrfs_ino(inode); |
5f39d397 | 5375 | |
39279cc3 CM |
5376 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5377 | if (ret < 0) | |
5378 | goto err; | |
49593bfa DW |
5379 | |
5380 | while (1) { | |
5f39d397 | 5381 | leaf = path->nodes[0]; |
39279cc3 | 5382 | slot = path->slots[0]; |
b9e03af0 LZ |
5383 | if (slot >= btrfs_header_nritems(leaf)) { |
5384 | ret = btrfs_next_leaf(root, path); | |
5385 | if (ret < 0) | |
5386 | goto err; | |
5387 | else if (ret > 0) | |
5388 | break; | |
5389 | continue; | |
39279cc3 | 5390 | } |
3de4586c | 5391 | |
dd3cc16b | 5392 | item = btrfs_item_nr(slot); |
5f39d397 CM |
5393 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5394 | ||
5395 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5396 | break; |
962a298f | 5397 | if (found_key.type != key_type) |
39279cc3 | 5398 | break; |
9cdda8d3 | 5399 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5400 | goto next; |
16cdcec7 MX |
5401 | if (key_type == BTRFS_DIR_INDEX_KEY && |
5402 | btrfs_should_delete_dir_index(&del_list, | |
5403 | found_key.offset)) | |
5404 | goto next; | |
5f39d397 | 5405 | |
9cdda8d3 | 5406 | ctx->pos = found_key.offset; |
16cdcec7 | 5407 | is_curr = 1; |
49593bfa | 5408 | |
39279cc3 CM |
5409 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
5410 | di_cur = 0; | |
5f39d397 | 5411 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
5412 | |
5413 | while (di_cur < di_total) { | |
5f39d397 CM |
5414 | struct btrfs_key location; |
5415 | ||
22a94d44 JB |
5416 | if (verify_dir_item(root, leaf, di)) |
5417 | break; | |
5418 | ||
5f39d397 | 5419 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 5420 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
5421 | name_ptr = tmp_name; |
5422 | } else { | |
5423 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
5424 | if (!name_ptr) { |
5425 | ret = -ENOMEM; | |
5426 | goto err; | |
5427 | } | |
5f39d397 CM |
5428 | } |
5429 | read_extent_buffer(leaf, name_ptr, | |
5430 | (unsigned long)(di + 1), name_len); | |
5431 | ||
5432 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
5433 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 5434 | |
fede766f | 5435 | |
3de4586c | 5436 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
5437 | * skip it. |
5438 | * | |
5439 | * In contrast to old kernels, we insert the snapshot's | |
5440 | * dir item and dir index after it has been created, so | |
5441 | * we won't find a reference to our own snapshot. We | |
5442 | * still keep the following code for backward | |
5443 | * compatibility. | |
3de4586c CM |
5444 | */ |
5445 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
5446 | location.objectid == root->root_key.objectid) { | |
5447 | over = 0; | |
5448 | goto skip; | |
5449 | } | |
9cdda8d3 AV |
5450 | over = !dir_emit(ctx, name_ptr, name_len, |
5451 | location.objectid, d_type); | |
5f39d397 | 5452 | |
3de4586c | 5453 | skip: |
5f39d397 CM |
5454 | if (name_ptr != tmp_name) |
5455 | kfree(name_ptr); | |
5456 | ||
39279cc3 CM |
5457 | if (over) |
5458 | goto nopos; | |
5103e947 | 5459 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 5460 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
5461 | di_cur += di_len; |
5462 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
5463 | } | |
b9e03af0 LZ |
5464 | next: |
5465 | path->slots[0]++; | |
39279cc3 | 5466 | } |
49593bfa | 5467 | |
16cdcec7 MX |
5468 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5469 | if (is_curr) | |
9cdda8d3 AV |
5470 | ctx->pos++; |
5471 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); | |
16cdcec7 MX |
5472 | if (ret) |
5473 | goto nopos; | |
5474 | } | |
5475 | ||
49593bfa | 5476 | /* Reached end of directory/root. Bump pos past the last item. */ |
db62efbb ZB |
5477 | ctx->pos++; |
5478 | ||
5479 | /* | |
5480 | * Stop new entries from being returned after we return the last | |
5481 | * entry. | |
5482 | * | |
5483 | * New directory entries are assigned a strictly increasing | |
5484 | * offset. This means that new entries created during readdir | |
5485 | * are *guaranteed* to be seen in the future by that readdir. | |
5486 | * This has broken buggy programs which operate on names as | |
5487 | * they're returned by readdir. Until we re-use freed offsets | |
5488 | * we have this hack to stop new entries from being returned | |
5489 | * under the assumption that they'll never reach this huge | |
5490 | * offset. | |
5491 | * | |
5492 | * This is being careful not to overflow 32bit loff_t unless the | |
5493 | * last entry requires it because doing so has broken 32bit apps | |
5494 | * in the past. | |
5495 | */ | |
5496 | if (key_type == BTRFS_DIR_INDEX_KEY) { | |
5497 | if (ctx->pos >= INT_MAX) | |
5498 | ctx->pos = LLONG_MAX; | |
5499 | else | |
5500 | ctx->pos = INT_MAX; | |
5501 | } | |
39279cc3 CM |
5502 | nopos: |
5503 | ret = 0; | |
5504 | err: | |
16cdcec7 MX |
5505 | if (key_type == BTRFS_DIR_INDEX_KEY) |
5506 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 5507 | btrfs_free_path(path); |
39279cc3 CM |
5508 | return ret; |
5509 | } | |
5510 | ||
a9185b41 | 5511 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
5512 | { |
5513 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5514 | struct btrfs_trans_handle *trans; | |
5515 | int ret = 0; | |
0af3d00b | 5516 | bool nolock = false; |
39279cc3 | 5517 | |
72ac3c0d | 5518 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
5519 | return 0; |
5520 | ||
83eea1f1 | 5521 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 5522 | nolock = true; |
0af3d00b | 5523 | |
a9185b41 | 5524 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 5525 | if (nolock) |
7a7eaa40 | 5526 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 5527 | else |
7a7eaa40 | 5528 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5529 | if (IS_ERR(trans)) |
5530 | return PTR_ERR(trans); | |
a698d075 | 5531 | ret = btrfs_commit_transaction(trans, root); |
39279cc3 CM |
5532 | } |
5533 | return ret; | |
5534 | } | |
5535 | ||
5536 | /* | |
54aa1f4d | 5537 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5538 | * inode changes. But, it is most likely to find the inode in cache. |
5539 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5540 | * to keep or drop this code. | |
5541 | */ | |
48a3b636 | 5542 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
5543 | { |
5544 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5545 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5546 | int ret; |
5547 | ||
72ac3c0d | 5548 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5549 | return 0; |
39279cc3 | 5550 | |
7a7eaa40 | 5551 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5552 | if (IS_ERR(trans)) |
5553 | return PTR_ERR(trans); | |
8929ecfa YZ |
5554 | |
5555 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
5556 | if (ret && ret == -ENOSPC) { |
5557 | /* whoops, lets try again with the full transaction */ | |
5558 | btrfs_end_transaction(trans, root); | |
5559 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
5560 | if (IS_ERR(trans)) |
5561 | return PTR_ERR(trans); | |
8929ecfa | 5562 | |
94b60442 | 5563 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 5564 | } |
39279cc3 | 5565 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
5566 | if (BTRFS_I(inode)->delayed_node) |
5567 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
5568 | |
5569 | return ret; | |
5570 | } | |
5571 | ||
5572 | /* | |
5573 | * This is a copy of file_update_time. We need this so we can return error on | |
5574 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
5575 | */ | |
e41f941a JB |
5576 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
5577 | int flags) | |
22c44fe6 | 5578 | { |
2bc55652 AB |
5579 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5580 | ||
5581 | if (btrfs_root_readonly(root)) | |
5582 | return -EROFS; | |
5583 | ||
e41f941a | 5584 | if (flags & S_VERSION) |
22c44fe6 | 5585 | inode_inc_iversion(inode); |
e41f941a JB |
5586 | if (flags & S_CTIME) |
5587 | inode->i_ctime = *now; | |
5588 | if (flags & S_MTIME) | |
5589 | inode->i_mtime = *now; | |
5590 | if (flags & S_ATIME) | |
5591 | inode->i_atime = *now; | |
5592 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
5593 | } |
5594 | ||
d352ac68 CM |
5595 | /* |
5596 | * find the highest existing sequence number in a directory | |
5597 | * and then set the in-memory index_cnt variable to reflect | |
5598 | * free sequence numbers | |
5599 | */ | |
aec7477b JB |
5600 | static int btrfs_set_inode_index_count(struct inode *inode) |
5601 | { | |
5602 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5603 | struct btrfs_key key, found_key; | |
5604 | struct btrfs_path *path; | |
5605 | struct extent_buffer *leaf; | |
5606 | int ret; | |
5607 | ||
33345d01 | 5608 | key.objectid = btrfs_ino(inode); |
962a298f | 5609 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
5610 | key.offset = (u64)-1; |
5611 | ||
5612 | path = btrfs_alloc_path(); | |
5613 | if (!path) | |
5614 | return -ENOMEM; | |
5615 | ||
5616 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5617 | if (ret < 0) | |
5618 | goto out; | |
5619 | /* FIXME: we should be able to handle this */ | |
5620 | if (ret == 0) | |
5621 | goto out; | |
5622 | ret = 0; | |
5623 | ||
5624 | /* | |
5625 | * MAGIC NUMBER EXPLANATION: | |
5626 | * since we search a directory based on f_pos we have to start at 2 | |
5627 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
5628 | * else has to start at 2 | |
5629 | */ | |
5630 | if (path->slots[0] == 0) { | |
5631 | BTRFS_I(inode)->index_cnt = 2; | |
5632 | goto out; | |
5633 | } | |
5634 | ||
5635 | path->slots[0]--; | |
5636 | ||
5637 | leaf = path->nodes[0]; | |
5638 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5639 | ||
33345d01 | 5640 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 5641 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
aec7477b JB |
5642 | BTRFS_I(inode)->index_cnt = 2; |
5643 | goto out; | |
5644 | } | |
5645 | ||
5646 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
5647 | out: | |
5648 | btrfs_free_path(path); | |
5649 | return ret; | |
5650 | } | |
5651 | ||
d352ac68 CM |
5652 | /* |
5653 | * helper to find a free sequence number in a given directory. This current | |
5654 | * code is very simple, later versions will do smarter things in the btree | |
5655 | */ | |
3de4586c | 5656 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
5657 | { |
5658 | int ret = 0; | |
5659 | ||
5660 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
5661 | ret = btrfs_inode_delayed_dir_index_count(dir); |
5662 | if (ret) { | |
5663 | ret = btrfs_set_inode_index_count(dir); | |
5664 | if (ret) | |
5665 | return ret; | |
5666 | } | |
aec7477b JB |
5667 | } |
5668 | ||
00e4e6b3 | 5669 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
5670 | BTRFS_I(dir)->index_cnt++; |
5671 | ||
5672 | return ret; | |
5673 | } | |
5674 | ||
b0d5d10f CM |
5675 | static int btrfs_insert_inode_locked(struct inode *inode) |
5676 | { | |
5677 | struct btrfs_iget_args args; | |
5678 | args.location = &BTRFS_I(inode)->location; | |
5679 | args.root = BTRFS_I(inode)->root; | |
5680 | ||
5681 | return insert_inode_locked4(inode, | |
5682 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
5683 | btrfs_find_actor, &args); | |
5684 | } | |
5685 | ||
39279cc3 CM |
5686 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
5687 | struct btrfs_root *root, | |
aec7477b | 5688 | struct inode *dir, |
9c58309d | 5689 | const char *name, int name_len, |
175a4eb7 AV |
5690 | u64 ref_objectid, u64 objectid, |
5691 | umode_t mode, u64 *index) | |
39279cc3 CM |
5692 | { |
5693 | struct inode *inode; | |
5f39d397 | 5694 | struct btrfs_inode_item *inode_item; |
39279cc3 | 5695 | struct btrfs_key *location; |
5f39d397 | 5696 | struct btrfs_path *path; |
9c58309d CM |
5697 | struct btrfs_inode_ref *ref; |
5698 | struct btrfs_key key[2]; | |
5699 | u32 sizes[2]; | |
ef3b9af5 | 5700 | int nitems = name ? 2 : 1; |
9c58309d | 5701 | unsigned long ptr; |
39279cc3 | 5702 | int ret; |
39279cc3 | 5703 | |
5f39d397 | 5704 | path = btrfs_alloc_path(); |
d8926bb3 MF |
5705 | if (!path) |
5706 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 5707 | |
39279cc3 | 5708 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
5709 | if (!inode) { |
5710 | btrfs_free_path(path); | |
39279cc3 | 5711 | return ERR_PTR(-ENOMEM); |
8fb27640 | 5712 | } |
39279cc3 | 5713 | |
5762b5c9 FM |
5714 | /* |
5715 | * O_TMPFILE, set link count to 0, so that after this point, | |
5716 | * we fill in an inode item with the correct link count. | |
5717 | */ | |
5718 | if (!name) | |
5719 | set_nlink(inode, 0); | |
5720 | ||
581bb050 LZ |
5721 | /* |
5722 | * we have to initialize this early, so we can reclaim the inode | |
5723 | * number if we fail afterwards in this function. | |
5724 | */ | |
5725 | inode->i_ino = objectid; | |
5726 | ||
ef3b9af5 | 5727 | if (dir && name) { |
1abe9b8a | 5728 | trace_btrfs_inode_request(dir); |
5729 | ||
3de4586c | 5730 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 5731 | if (ret) { |
8fb27640 | 5732 | btrfs_free_path(path); |
09771430 | 5733 | iput(inode); |
aec7477b | 5734 | return ERR_PTR(ret); |
09771430 | 5735 | } |
ef3b9af5 FM |
5736 | } else if (dir) { |
5737 | *index = 0; | |
aec7477b JB |
5738 | } |
5739 | /* | |
5740 | * index_cnt is ignored for everything but a dir, | |
5741 | * btrfs_get_inode_index_count has an explanation for the magic | |
5742 | * number | |
5743 | */ | |
5744 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 5745 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 5746 | BTRFS_I(inode)->root = root; |
e02119d5 | 5747 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 5748 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 5749 | |
5dc562c5 JB |
5750 | /* |
5751 | * We could have gotten an inode number from somebody who was fsynced | |
5752 | * and then removed in this same transaction, so let's just set full | |
5753 | * sync since it will be a full sync anyway and this will blow away the | |
5754 | * old info in the log. | |
5755 | */ | |
5756 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
5757 | ||
9c58309d | 5758 | key[0].objectid = objectid; |
962a298f | 5759 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
5760 | key[0].offset = 0; |
5761 | ||
9c58309d | 5762 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
5763 | |
5764 | if (name) { | |
5765 | /* | |
5766 | * Start new inodes with an inode_ref. This is slightly more | |
5767 | * efficient for small numbers of hard links since they will | |
5768 | * be packed into one item. Extended refs will kick in if we | |
5769 | * add more hard links than can fit in the ref item. | |
5770 | */ | |
5771 | key[1].objectid = objectid; | |
962a298f | 5772 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
5773 | key[1].offset = ref_objectid; |
5774 | ||
5775 | sizes[1] = name_len + sizeof(*ref); | |
5776 | } | |
9c58309d | 5777 | |
b0d5d10f CM |
5778 | location = &BTRFS_I(inode)->location; |
5779 | location->objectid = objectid; | |
5780 | location->offset = 0; | |
962a298f | 5781 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
5782 | |
5783 | ret = btrfs_insert_inode_locked(inode); | |
5784 | if (ret < 0) | |
5785 | goto fail; | |
5786 | ||
b9473439 | 5787 | path->leave_spinning = 1; |
ef3b9af5 | 5788 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 5789 | if (ret != 0) |
b0d5d10f | 5790 | goto fail_unlock; |
5f39d397 | 5791 | |
ecc11fab | 5792 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 5793 | inode_set_bytes(inode, 0); |
39279cc3 | 5794 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
5795 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
5796 | struct btrfs_inode_item); | |
293f7e07 LZ |
5797 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
5798 | sizeof(*inode_item)); | |
e02119d5 | 5799 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 5800 | |
ef3b9af5 FM |
5801 | if (name) { |
5802 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
5803 | struct btrfs_inode_ref); | |
5804 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
5805 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
5806 | ptr = (unsigned long)(ref + 1); | |
5807 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
5808 | } | |
9c58309d | 5809 | |
5f39d397 CM |
5810 | btrfs_mark_buffer_dirty(path->nodes[0]); |
5811 | btrfs_free_path(path); | |
5812 | ||
6cbff00f CH |
5813 | btrfs_inherit_iflags(inode, dir); |
5814 | ||
569254b0 | 5815 | if (S_ISREG(mode)) { |
94272164 CM |
5816 | if (btrfs_test_opt(root, NODATASUM)) |
5817 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
213490b3 | 5818 | if (btrfs_test_opt(root, NODATACOW)) |
f2bdf9a8 JB |
5819 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
5820 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
5821 | } |
5822 | ||
5d4f98a2 | 5823 | inode_tree_add(inode); |
1abe9b8a | 5824 | |
5825 | trace_btrfs_inode_new(inode); | |
1973f0fa | 5826 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 5827 | |
8ea05e3a AB |
5828 | btrfs_update_root_times(trans, root); |
5829 | ||
63541927 FDBM |
5830 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
5831 | if (ret) | |
5832 | btrfs_err(root->fs_info, | |
5833 | "error inheriting props for ino %llu (root %llu): %d", | |
5834 | btrfs_ino(inode), root->root_key.objectid, ret); | |
5835 | ||
39279cc3 | 5836 | return inode; |
b0d5d10f CM |
5837 | |
5838 | fail_unlock: | |
5839 | unlock_new_inode(inode); | |
5f39d397 | 5840 | fail: |
ef3b9af5 | 5841 | if (dir && name) |
aec7477b | 5842 | BTRFS_I(dir)->index_cnt--; |
5f39d397 | 5843 | btrfs_free_path(path); |
09771430 | 5844 | iput(inode); |
5f39d397 | 5845 | return ERR_PTR(ret); |
39279cc3 CM |
5846 | } |
5847 | ||
5848 | static inline u8 btrfs_inode_type(struct inode *inode) | |
5849 | { | |
5850 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
5851 | } | |
5852 | ||
d352ac68 CM |
5853 | /* |
5854 | * utility function to add 'inode' into 'parent_inode' with | |
5855 | * a give name and a given sequence number. | |
5856 | * if 'add_backref' is true, also insert a backref from the | |
5857 | * inode to the parent directory. | |
5858 | */ | |
e02119d5 CM |
5859 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
5860 | struct inode *parent_inode, struct inode *inode, | |
5861 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 5862 | { |
4df27c4d | 5863 | int ret = 0; |
39279cc3 | 5864 | struct btrfs_key key; |
e02119d5 | 5865 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
5866 | u64 ino = btrfs_ino(inode); |
5867 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 5868 | |
33345d01 | 5869 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5870 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
5871 | } else { | |
33345d01 | 5872 | key.objectid = ino; |
962a298f | 5873 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
5874 | key.offset = 0; |
5875 | } | |
5876 | ||
33345d01 | 5877 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5878 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
5879 | key.objectid, root->root_key.objectid, | |
33345d01 | 5880 | parent_ino, index, name, name_len); |
4df27c4d | 5881 | } else if (add_backref) { |
33345d01 LZ |
5882 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
5883 | parent_ino, index); | |
4df27c4d | 5884 | } |
39279cc3 | 5885 | |
79787eaa JM |
5886 | /* Nothing to clean up yet */ |
5887 | if (ret) | |
5888 | return ret; | |
4df27c4d | 5889 | |
79787eaa JM |
5890 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
5891 | parent_inode, &key, | |
5892 | btrfs_inode_type(inode), index); | |
9c52057c | 5893 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
5894 | goto fail_dir_item; |
5895 | else if (ret) { | |
5896 | btrfs_abort_transaction(trans, root, ret); | |
5897 | return ret; | |
39279cc3 | 5898 | } |
79787eaa JM |
5899 | |
5900 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
5901 | name_len * 2); | |
0c4d2d95 | 5902 | inode_inc_iversion(parent_inode); |
79787eaa JM |
5903 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
5904 | ret = btrfs_update_inode(trans, root, parent_inode); | |
5905 | if (ret) | |
5906 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 5907 | return ret; |
fe66a05a CM |
5908 | |
5909 | fail_dir_item: | |
5910 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
5911 | u64 local_index; | |
5912 | int err; | |
5913 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
5914 | key.objectid, root->root_key.objectid, | |
5915 | parent_ino, &local_index, name, name_len); | |
5916 | ||
5917 | } else if (add_backref) { | |
5918 | u64 local_index; | |
5919 | int err; | |
5920 | ||
5921 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
5922 | ino, parent_ino, &local_index); | |
5923 | } | |
5924 | return ret; | |
39279cc3 CM |
5925 | } |
5926 | ||
5927 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
5928 | struct inode *dir, struct dentry *dentry, |
5929 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 5930 | { |
a1b075d2 JB |
5931 | int err = btrfs_add_link(trans, dir, inode, |
5932 | dentry->d_name.name, dentry->d_name.len, | |
5933 | backref, index); | |
39279cc3 CM |
5934 | if (err > 0) |
5935 | err = -EEXIST; | |
5936 | return err; | |
5937 | } | |
5938 | ||
618e21d5 | 5939 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 5940 | umode_t mode, dev_t rdev) |
618e21d5 JB |
5941 | { |
5942 | struct btrfs_trans_handle *trans; | |
5943 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 5944 | struct inode *inode = NULL; |
618e21d5 JB |
5945 | int err; |
5946 | int drop_inode = 0; | |
5947 | u64 objectid; | |
00e4e6b3 | 5948 | u64 index = 0; |
618e21d5 JB |
5949 | |
5950 | if (!new_valid_dev(rdev)) | |
5951 | return -EINVAL; | |
5952 | ||
9ed74f2d JB |
5953 | /* |
5954 | * 2 for inode item and ref | |
5955 | * 2 for dir items | |
5956 | * 1 for xattr if selinux is on | |
5957 | */ | |
a22285a6 YZ |
5958 | trans = btrfs_start_transaction(root, 5); |
5959 | if (IS_ERR(trans)) | |
5960 | return PTR_ERR(trans); | |
1832a6d5 | 5961 | |
581bb050 LZ |
5962 | err = btrfs_find_free_ino(root, &objectid); |
5963 | if (err) | |
5964 | goto out_unlock; | |
5965 | ||
aec7477b | 5966 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5967 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5968 | mode, &index); |
7cf96da3 TI |
5969 | if (IS_ERR(inode)) { |
5970 | err = PTR_ERR(inode); | |
618e21d5 | 5971 | goto out_unlock; |
7cf96da3 | 5972 | } |
618e21d5 | 5973 | |
ad19db71 CS |
5974 | /* |
5975 | * If the active LSM wants to access the inode during | |
5976 | * d_instantiate it needs these. Smack checks to see | |
5977 | * if the filesystem supports xattrs by looking at the | |
5978 | * ops vector. | |
5979 | */ | |
ad19db71 | 5980 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
5981 | init_special_inode(inode, inode->i_mode, rdev); |
5982 | ||
5983 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 5984 | if (err) |
b0d5d10f CM |
5985 | goto out_unlock_inode; |
5986 | ||
5987 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); | |
5988 | if (err) { | |
5989 | goto out_unlock_inode; | |
5990 | } else { | |
1b4ab1bb | 5991 | btrfs_update_inode(trans, root, inode); |
b0d5d10f | 5992 | unlock_new_inode(inode); |
08c422c2 | 5993 | d_instantiate(dentry, inode); |
618e21d5 | 5994 | } |
b0d5d10f | 5995 | |
618e21d5 | 5996 | out_unlock: |
7ad85bb7 | 5997 | btrfs_end_transaction(trans, root); |
c581afc8 | 5998 | btrfs_balance_delayed_items(root); |
b53d3f5d | 5999 | btrfs_btree_balance_dirty(root); |
618e21d5 JB |
6000 | if (drop_inode) { |
6001 | inode_dec_link_count(inode); | |
6002 | iput(inode); | |
6003 | } | |
618e21d5 | 6004 | return err; |
b0d5d10f CM |
6005 | |
6006 | out_unlock_inode: | |
6007 | drop_inode = 1; | |
6008 | unlock_new_inode(inode); | |
6009 | goto out_unlock; | |
6010 | ||
618e21d5 JB |
6011 | } |
6012 | ||
39279cc3 | 6013 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6014 | umode_t mode, bool excl) |
39279cc3 CM |
6015 | { |
6016 | struct btrfs_trans_handle *trans; | |
6017 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6018 | struct inode *inode = NULL; |
43baa579 | 6019 | int drop_inode_on_err = 0; |
a22285a6 | 6020 | int err; |
39279cc3 | 6021 | u64 objectid; |
00e4e6b3 | 6022 | u64 index = 0; |
39279cc3 | 6023 | |
9ed74f2d JB |
6024 | /* |
6025 | * 2 for inode item and ref | |
6026 | * 2 for dir items | |
6027 | * 1 for xattr if selinux is on | |
6028 | */ | |
a22285a6 YZ |
6029 | trans = btrfs_start_transaction(root, 5); |
6030 | if (IS_ERR(trans)) | |
6031 | return PTR_ERR(trans); | |
9ed74f2d | 6032 | |
581bb050 LZ |
6033 | err = btrfs_find_free_ino(root, &objectid); |
6034 | if (err) | |
6035 | goto out_unlock; | |
6036 | ||
aec7477b | 6037 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6038 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6039 | mode, &index); |
7cf96da3 TI |
6040 | if (IS_ERR(inode)) { |
6041 | err = PTR_ERR(inode); | |
39279cc3 | 6042 | goto out_unlock; |
7cf96da3 | 6043 | } |
43baa579 | 6044 | drop_inode_on_err = 1; |
ad19db71 CS |
6045 | /* |
6046 | * If the active LSM wants to access the inode during | |
6047 | * d_instantiate it needs these. Smack checks to see | |
6048 | * if the filesystem supports xattrs by looking at the | |
6049 | * ops vector. | |
6050 | */ | |
6051 | inode->i_fop = &btrfs_file_operations; | |
6052 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f CM |
6053 | inode->i_mapping->a_ops = &btrfs_aops; |
6054 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
6055 | ||
6056 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6057 | if (err) | |
6058 | goto out_unlock_inode; | |
6059 | ||
6060 | err = btrfs_update_inode(trans, root, inode); | |
6061 | if (err) | |
6062 | goto out_unlock_inode; | |
ad19db71 | 6063 | |
a1b075d2 | 6064 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 6065 | if (err) |
b0d5d10f | 6066 | goto out_unlock_inode; |
43baa579 | 6067 | |
43baa579 | 6068 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
b0d5d10f | 6069 | unlock_new_inode(inode); |
43baa579 FB |
6070 | d_instantiate(dentry, inode); |
6071 | ||
39279cc3 | 6072 | out_unlock: |
7ad85bb7 | 6073 | btrfs_end_transaction(trans, root); |
43baa579 | 6074 | if (err && drop_inode_on_err) { |
39279cc3 CM |
6075 | inode_dec_link_count(inode); |
6076 | iput(inode); | |
6077 | } | |
c581afc8 | 6078 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6079 | btrfs_btree_balance_dirty(root); |
39279cc3 | 6080 | return err; |
b0d5d10f CM |
6081 | |
6082 | out_unlock_inode: | |
6083 | unlock_new_inode(inode); | |
6084 | goto out_unlock; | |
6085 | ||
39279cc3 CM |
6086 | } |
6087 | ||
6088 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6089 | struct dentry *dentry) | |
6090 | { | |
6091 | struct btrfs_trans_handle *trans; | |
6092 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6093 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 6094 | u64 index; |
39279cc3 CM |
6095 | int err; |
6096 | int drop_inode = 0; | |
6097 | ||
4a8be425 TH |
6098 | /* do not allow sys_link's with other subvols of the same device */ |
6099 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 6100 | return -EXDEV; |
4a8be425 | 6101 | |
f186373f | 6102 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6103 | return -EMLINK; |
4a8be425 | 6104 | |
3de4586c | 6105 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
6106 | if (err) |
6107 | goto fail; | |
6108 | ||
a22285a6 | 6109 | /* |
7e6b6465 | 6110 | * 2 items for inode and inode ref |
a22285a6 | 6111 | * 2 items for dir items |
7e6b6465 | 6112 | * 1 item for parent inode |
a22285a6 | 6113 | */ |
7e6b6465 | 6114 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
6115 | if (IS_ERR(trans)) { |
6116 | err = PTR_ERR(trans); | |
6117 | goto fail; | |
6118 | } | |
5f39d397 | 6119 | |
67de1176 MX |
6120 | /* There are several dir indexes for this inode, clear the cache. */ |
6121 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6122 | inc_nlink(inode); |
0c4d2d95 | 6123 | inode_inc_iversion(inode); |
3153495d | 6124 | inode->i_ctime = CURRENT_TIME; |
7de9c6ee | 6125 | ihold(inode); |
e9976151 | 6126 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6127 | |
a1b075d2 | 6128 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 6129 | |
a5719521 | 6130 | if (err) { |
54aa1f4d | 6131 | drop_inode = 1; |
a5719521 | 6132 | } else { |
10d9f309 | 6133 | struct dentry *parent = dentry->d_parent; |
a5719521 | 6134 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6135 | if (err) |
6136 | goto fail; | |
ef3b9af5 FM |
6137 | if (inode->i_nlink == 1) { |
6138 | /* | |
6139 | * If new hard link count is 1, it's a file created | |
6140 | * with open(2) O_TMPFILE flag. | |
6141 | */ | |
6142 | err = btrfs_orphan_del(trans, inode); | |
6143 | if (err) | |
6144 | goto fail; | |
6145 | } | |
08c422c2 | 6146 | d_instantiate(dentry, inode); |
6a912213 | 6147 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 6148 | } |
39279cc3 | 6149 | |
7ad85bb7 | 6150 | btrfs_end_transaction(trans, root); |
c581afc8 | 6151 | btrfs_balance_delayed_items(root); |
1832a6d5 | 6152 | fail: |
39279cc3 CM |
6153 | if (drop_inode) { |
6154 | inode_dec_link_count(inode); | |
6155 | iput(inode); | |
6156 | } | |
b53d3f5d | 6157 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
6158 | return err; |
6159 | } | |
6160 | ||
18bb1db3 | 6161 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6162 | { |
b9d86667 | 6163 | struct inode *inode = NULL; |
39279cc3 CM |
6164 | struct btrfs_trans_handle *trans; |
6165 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6166 | int err = 0; | |
6167 | int drop_on_err = 0; | |
b9d86667 | 6168 | u64 objectid = 0; |
00e4e6b3 | 6169 | u64 index = 0; |
39279cc3 | 6170 | |
9ed74f2d JB |
6171 | /* |
6172 | * 2 items for inode and ref | |
6173 | * 2 items for dir items | |
6174 | * 1 for xattr if selinux is on | |
6175 | */ | |
a22285a6 YZ |
6176 | trans = btrfs_start_transaction(root, 5); |
6177 | if (IS_ERR(trans)) | |
6178 | return PTR_ERR(trans); | |
39279cc3 | 6179 | |
581bb050 LZ |
6180 | err = btrfs_find_free_ino(root, &objectid); |
6181 | if (err) | |
6182 | goto out_fail; | |
6183 | ||
aec7477b | 6184 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6185 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6186 | S_IFDIR | mode, &index); |
39279cc3 CM |
6187 | if (IS_ERR(inode)) { |
6188 | err = PTR_ERR(inode); | |
6189 | goto out_fail; | |
6190 | } | |
5f39d397 | 6191 | |
39279cc3 | 6192 | drop_on_err = 1; |
b0d5d10f CM |
6193 | /* these must be set before we unlock the inode */ |
6194 | inode->i_op = &btrfs_dir_inode_operations; | |
6195 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6196 | |
2a7dba39 | 6197 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6198 | if (err) |
b0d5d10f | 6199 | goto out_fail_inode; |
39279cc3 | 6200 | |
dbe674a9 | 6201 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
6202 | err = btrfs_update_inode(trans, root, inode); |
6203 | if (err) | |
b0d5d10f | 6204 | goto out_fail_inode; |
5f39d397 | 6205 | |
a1b075d2 JB |
6206 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
6207 | dentry->d_name.len, 0, index); | |
39279cc3 | 6208 | if (err) |
b0d5d10f | 6209 | goto out_fail_inode; |
5f39d397 | 6210 | |
39279cc3 | 6211 | d_instantiate(dentry, inode); |
b0d5d10f CM |
6212 | /* |
6213 | * mkdir is special. We're unlocking after we call d_instantiate | |
6214 | * to avoid a race with nfsd calling d_instantiate. | |
6215 | */ | |
6216 | unlock_new_inode(inode); | |
39279cc3 | 6217 | drop_on_err = 0; |
39279cc3 CM |
6218 | |
6219 | out_fail: | |
7ad85bb7 | 6220 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
6221 | if (drop_on_err) |
6222 | iput(inode); | |
c581afc8 | 6223 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6224 | btrfs_btree_balance_dirty(root); |
39279cc3 | 6225 | return err; |
b0d5d10f CM |
6226 | |
6227 | out_fail_inode: | |
6228 | unlock_new_inode(inode); | |
6229 | goto out_fail; | |
39279cc3 CM |
6230 | } |
6231 | ||
e6c4efd8 QW |
6232 | /* Find next extent map of a given extent map, caller needs to ensure locks */ |
6233 | static struct extent_map *next_extent_map(struct extent_map *em) | |
6234 | { | |
6235 | struct rb_node *next; | |
6236 | ||
6237 | next = rb_next(&em->rb_node); | |
6238 | if (!next) | |
6239 | return NULL; | |
6240 | return container_of(next, struct extent_map, rb_node); | |
6241 | } | |
6242 | ||
6243 | static struct extent_map *prev_extent_map(struct extent_map *em) | |
6244 | { | |
6245 | struct rb_node *prev; | |
6246 | ||
6247 | prev = rb_prev(&em->rb_node); | |
6248 | if (!prev) | |
6249 | return NULL; | |
6250 | return container_of(prev, struct extent_map, rb_node); | |
6251 | } | |
6252 | ||
d352ac68 | 6253 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
e6c4efd8 | 6254 | * the existing extent is the nearest extent to map_start, |
d352ac68 | 6255 | * and an extent that you want to insert, deal with overlap and insert |
e6c4efd8 | 6256 | * the best fitted new extent into the tree. |
d352ac68 | 6257 | */ |
3b951516 CM |
6258 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
6259 | struct extent_map *existing, | |
e6dcd2dc | 6260 | struct extent_map *em, |
51f395ad | 6261 | u64 map_start) |
3b951516 | 6262 | { |
e6c4efd8 QW |
6263 | struct extent_map *prev; |
6264 | struct extent_map *next; | |
6265 | u64 start; | |
6266 | u64 end; | |
3b951516 | 6267 | u64 start_diff; |
3b951516 | 6268 | |
e6dcd2dc | 6269 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
e6c4efd8 QW |
6270 | |
6271 | if (existing->start > map_start) { | |
6272 | next = existing; | |
6273 | prev = prev_extent_map(next); | |
6274 | } else { | |
6275 | prev = existing; | |
6276 | next = next_extent_map(prev); | |
6277 | } | |
6278 | ||
6279 | start = prev ? extent_map_end(prev) : em->start; | |
6280 | start = max_t(u64, start, em->start); | |
6281 | end = next ? next->start : extent_map_end(em); | |
6282 | end = min_t(u64, end, extent_map_end(em)); | |
6283 | start_diff = start - em->start; | |
6284 | em->start = start; | |
6285 | em->len = end - start; | |
c8b97818 CM |
6286 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
6287 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 6288 | em->block_start += start_diff; |
c8b97818 CM |
6289 | em->block_len -= start_diff; |
6290 | } | |
09a2a8f9 | 6291 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6292 | } |
6293 | ||
c8b97818 CM |
6294 | static noinline int uncompress_inline(struct btrfs_path *path, |
6295 | struct inode *inode, struct page *page, | |
6296 | size_t pg_offset, u64 extent_offset, | |
6297 | struct btrfs_file_extent_item *item) | |
6298 | { | |
6299 | int ret; | |
6300 | struct extent_buffer *leaf = path->nodes[0]; | |
6301 | char *tmp; | |
6302 | size_t max_size; | |
6303 | unsigned long inline_size; | |
6304 | unsigned long ptr; | |
261507a0 | 6305 | int compress_type; |
c8b97818 CM |
6306 | |
6307 | WARN_ON(pg_offset != 0); | |
261507a0 | 6308 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6309 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6310 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6311 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6312 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6313 | if (!tmp) |
6314 | return -ENOMEM; | |
c8b97818 CM |
6315 | ptr = btrfs_file_extent_inline_start(item); |
6316 | ||
6317 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6318 | ||
5b050f04 | 6319 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
6320 | ret = btrfs_decompress(compress_type, tmp, page, |
6321 | extent_offset, inline_size, max_size); | |
c8b97818 | 6322 | kfree(tmp); |
166ae5a4 | 6323 | return ret; |
c8b97818 CM |
6324 | } |
6325 | ||
d352ac68 CM |
6326 | /* |
6327 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
6328 | * the ugly parts come from merging extents from the disk with the in-ram |
6329 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
6330 | * where the in-ram extents might be locked pending data=ordered completion. |
6331 | * | |
6332 | * This also copies inline extents directly into the page. | |
6333 | */ | |
d397712b | 6334 | |
a52d9a80 | 6335 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 6336 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
6337 | int create) |
6338 | { | |
6339 | int ret; | |
6340 | int err = 0; | |
a52d9a80 CM |
6341 | u64 extent_start = 0; |
6342 | u64 extent_end = 0; | |
33345d01 | 6343 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 6344 | u32 found_type; |
f421950f | 6345 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
6346 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6347 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
6348 | struct extent_buffer *leaf; |
6349 | struct btrfs_key found_key; | |
a52d9a80 CM |
6350 | struct extent_map *em = NULL; |
6351 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 6352 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 6353 | struct btrfs_trans_handle *trans = NULL; |
7ffbb598 | 6354 | const bool new_inline = !page || create; |
a52d9a80 | 6355 | |
a52d9a80 | 6356 | again: |
890871be | 6357 | read_lock(&em_tree->lock); |
d1310b2e | 6358 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
6359 | if (em) |
6360 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 6361 | read_unlock(&em_tree->lock); |
d1310b2e | 6362 | |
a52d9a80 | 6363 | if (em) { |
e1c4b745 CM |
6364 | if (em->start > start || em->start + em->len <= start) |
6365 | free_extent_map(em); | |
6366 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6367 | free_extent_map(em); |
6368 | else | |
6369 | goto out; | |
a52d9a80 | 6370 | } |
172ddd60 | 6371 | em = alloc_extent_map(); |
a52d9a80 | 6372 | if (!em) { |
d1310b2e CM |
6373 | err = -ENOMEM; |
6374 | goto out; | |
a52d9a80 | 6375 | } |
e6dcd2dc | 6376 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 6377 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6378 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6379 | em->len = (u64)-1; |
c8b97818 | 6380 | em->block_len = (u64)-1; |
f421950f CM |
6381 | |
6382 | if (!path) { | |
6383 | path = btrfs_alloc_path(); | |
026fd317 JB |
6384 | if (!path) { |
6385 | err = -ENOMEM; | |
6386 | goto out; | |
6387 | } | |
6388 | /* | |
6389 | * Chances are we'll be called again, so go ahead and do | |
6390 | * readahead | |
6391 | */ | |
6392 | path->reada = 1; | |
f421950f CM |
6393 | } |
6394 | ||
179e29e4 CM |
6395 | ret = btrfs_lookup_file_extent(trans, root, path, |
6396 | objectid, start, trans != NULL); | |
a52d9a80 CM |
6397 | if (ret < 0) { |
6398 | err = ret; | |
6399 | goto out; | |
6400 | } | |
6401 | ||
6402 | if (ret != 0) { | |
6403 | if (path->slots[0] == 0) | |
6404 | goto not_found; | |
6405 | path->slots[0]--; | |
6406 | } | |
6407 | ||
5f39d397 CM |
6408 | leaf = path->nodes[0]; |
6409 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6410 | struct btrfs_file_extent_item); |
a52d9a80 | 6411 | /* are we inside the extent that was found? */ |
5f39d397 | 6412 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
962a298f | 6413 | found_type = found_key.type; |
5f39d397 | 6414 | if (found_key.objectid != objectid || |
a52d9a80 | 6415 | found_type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6416 | /* |
6417 | * If we backup past the first extent we want to move forward | |
6418 | * and see if there is an extent in front of us, otherwise we'll | |
6419 | * say there is a hole for our whole search range which can | |
6420 | * cause problems. | |
6421 | */ | |
6422 | extent_end = start; | |
6423 | goto next; | |
a52d9a80 CM |
6424 | } |
6425 | ||
5f39d397 CM |
6426 | found_type = btrfs_file_extent_type(leaf, item); |
6427 | extent_start = found_key.offset; | |
d899e052 YZ |
6428 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6429 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6430 | extent_end = extent_start + |
db94535d | 6431 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
6432 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
6433 | size_t size; | |
514ac8ad | 6434 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
fda2832f | 6435 | extent_end = ALIGN(extent_start + size, root->sectorsize); |
9036c102 | 6436 | } |
25a50341 | 6437 | next: |
9036c102 YZ |
6438 | if (start >= extent_end) { |
6439 | path->slots[0]++; | |
6440 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6441 | ret = btrfs_next_leaf(root, path); | |
6442 | if (ret < 0) { | |
6443 | err = ret; | |
6444 | goto out; | |
a52d9a80 | 6445 | } |
9036c102 YZ |
6446 | if (ret > 0) |
6447 | goto not_found; | |
6448 | leaf = path->nodes[0]; | |
a52d9a80 | 6449 | } |
9036c102 YZ |
6450 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6451 | if (found_key.objectid != objectid || | |
6452 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6453 | goto not_found; | |
6454 | if (start + len <= found_key.offset) | |
6455 | goto not_found; | |
e2eca69d WS |
6456 | if (start > found_key.offset) |
6457 | goto next; | |
9036c102 | 6458 | em->start = start; |
70c8a91c | 6459 | em->orig_start = start; |
9036c102 YZ |
6460 | em->len = found_key.offset - start; |
6461 | goto not_found_em; | |
6462 | } | |
6463 | ||
7ffbb598 FM |
6464 | btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em); |
6465 | ||
d899e052 YZ |
6466 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6467 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 CM |
6468 | goto insert; |
6469 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 6470 | unsigned long ptr; |
a52d9a80 | 6471 | char *map; |
3326d1b0 CM |
6472 | size_t size; |
6473 | size_t extent_offset; | |
6474 | size_t copy_size; | |
a52d9a80 | 6475 | |
7ffbb598 | 6476 | if (new_inline) |
689f9346 | 6477 | goto out; |
5f39d397 | 6478 | |
514ac8ad | 6479 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
9036c102 | 6480 | extent_offset = page_offset(page) + pg_offset - extent_start; |
70dec807 | 6481 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 6482 | size - extent_offset); |
3326d1b0 | 6483 | em->start = extent_start + extent_offset; |
fda2832f | 6484 | em->len = ALIGN(copy_size, root->sectorsize); |
b4939680 | 6485 | em->orig_block_len = em->len; |
70c8a91c | 6486 | em->orig_start = em->start; |
689f9346 | 6487 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 6488 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
6489 | if (btrfs_file_extent_compression(leaf, item) != |
6490 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
6491 | ret = uncompress_inline(path, inode, page, |
6492 | pg_offset, | |
6493 | extent_offset, item); | |
166ae5a4 ZB |
6494 | if (ret) { |
6495 | err = ret; | |
6496 | goto out; | |
6497 | } | |
c8b97818 CM |
6498 | } else { |
6499 | map = kmap(page); | |
6500 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6501 | copy_size); | |
93c82d57 CM |
6502 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
6503 | memset(map + pg_offset + copy_size, 0, | |
6504 | PAGE_CACHE_SIZE - pg_offset - | |
6505 | copy_size); | |
6506 | } | |
c8b97818 CM |
6507 | kunmap(page); |
6508 | } | |
179e29e4 CM |
6509 | flush_dcache_page(page); |
6510 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 6511 | BUG(); |
179e29e4 CM |
6512 | if (!trans) { |
6513 | kunmap(page); | |
6514 | free_extent_map(em); | |
6515 | em = NULL; | |
ff5714cc | 6516 | |
b3b4aa74 | 6517 | btrfs_release_path(path); |
7a7eaa40 | 6518 | trans = btrfs_join_transaction(root); |
ff5714cc | 6519 | |
3612b495 TI |
6520 | if (IS_ERR(trans)) |
6521 | return ERR_CAST(trans); | |
179e29e4 CM |
6522 | goto again; |
6523 | } | |
c8b97818 | 6524 | map = kmap(page); |
70dec807 | 6525 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 6526 | copy_size); |
c8b97818 | 6527 | kunmap(page); |
179e29e4 | 6528 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 6529 | } |
d1310b2e | 6530 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6531 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 6532 | goto insert; |
a52d9a80 CM |
6533 | } |
6534 | not_found: | |
6535 | em->start = start; | |
70c8a91c | 6536 | em->orig_start = start; |
d1310b2e | 6537 | em->len = len; |
a52d9a80 | 6538 | not_found_em: |
5f39d397 | 6539 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 6540 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 6541 | insert: |
b3b4aa74 | 6542 | btrfs_release_path(path); |
d1310b2e | 6543 | if (em->start > start || extent_map_end(em) <= start) { |
c2cf52eb | 6544 | btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]", |
c1c9ff7c | 6545 | em->start, em->len, start, len); |
a52d9a80 CM |
6546 | err = -EIO; |
6547 | goto out; | |
6548 | } | |
d1310b2e CM |
6549 | |
6550 | err = 0; | |
890871be | 6551 | write_lock(&em_tree->lock); |
09a2a8f9 | 6552 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6553 | /* it is possible that someone inserted the extent into the tree |
6554 | * while we had the lock dropped. It is also possible that | |
6555 | * an overlapping map exists in the tree | |
6556 | */ | |
a52d9a80 | 6557 | if (ret == -EEXIST) { |
3b951516 | 6558 | struct extent_map *existing; |
e6dcd2dc CM |
6559 | |
6560 | ret = 0; | |
6561 | ||
e6c4efd8 QW |
6562 | existing = search_extent_mapping(em_tree, start, len); |
6563 | /* | |
6564 | * existing will always be non-NULL, since there must be | |
6565 | * extent causing the -EEXIST. | |
6566 | */ | |
6567 | if (start >= extent_map_end(existing) || | |
32be3a1a | 6568 | start <= existing->start) { |
e6c4efd8 QW |
6569 | /* |
6570 | * The existing extent map is the one nearest to | |
6571 | * the [start, start + len) range which overlaps | |
6572 | */ | |
6573 | err = merge_extent_mapping(em_tree, existing, | |
6574 | em, start); | |
e1c4b745 | 6575 | free_extent_map(existing); |
e6c4efd8 | 6576 | if (err) { |
3b951516 CM |
6577 | free_extent_map(em); |
6578 | em = NULL; | |
6579 | } | |
6580 | } else { | |
6581 | free_extent_map(em); | |
6582 | em = existing; | |
e6dcd2dc | 6583 | err = 0; |
a52d9a80 | 6584 | } |
a52d9a80 | 6585 | } |
890871be | 6586 | write_unlock(&em_tree->lock); |
a52d9a80 | 6587 | out: |
1abe9b8a | 6588 | |
4cd8587c | 6589 | trace_btrfs_get_extent(root, em); |
1abe9b8a | 6590 | |
f421950f CM |
6591 | if (path) |
6592 | btrfs_free_path(path); | |
a52d9a80 CM |
6593 | if (trans) { |
6594 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 6595 | if (!err) |
a52d9a80 CM |
6596 | err = ret; |
6597 | } | |
a52d9a80 CM |
6598 | if (err) { |
6599 | free_extent_map(em); | |
a52d9a80 CM |
6600 | return ERR_PTR(err); |
6601 | } | |
79787eaa | 6602 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
6603 | return em; |
6604 | } | |
6605 | ||
ec29ed5b CM |
6606 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
6607 | size_t pg_offset, u64 start, u64 len, | |
6608 | int create) | |
6609 | { | |
6610 | struct extent_map *em; | |
6611 | struct extent_map *hole_em = NULL; | |
6612 | u64 range_start = start; | |
6613 | u64 end; | |
6614 | u64 found; | |
6615 | u64 found_end; | |
6616 | int err = 0; | |
6617 | ||
6618 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
6619 | if (IS_ERR(em)) | |
6620 | return em; | |
6621 | if (em) { | |
6622 | /* | |
f9e4fb53 LB |
6623 | * if our em maps to |
6624 | * - a hole or | |
6625 | * - a pre-alloc extent, | |
6626 | * there might actually be delalloc bytes behind it. | |
ec29ed5b | 6627 | */ |
f9e4fb53 LB |
6628 | if (em->block_start != EXTENT_MAP_HOLE && |
6629 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
ec29ed5b CM |
6630 | return em; |
6631 | else | |
6632 | hole_em = em; | |
6633 | } | |
6634 | ||
6635 | /* check to see if we've wrapped (len == -1 or similar) */ | |
6636 | end = start + len; | |
6637 | if (end < start) | |
6638 | end = (u64)-1; | |
6639 | else | |
6640 | end -= 1; | |
6641 | ||
6642 | em = NULL; | |
6643 | ||
6644 | /* ok, we didn't find anything, lets look for delalloc */ | |
6645 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
6646 | end, len, EXTENT_DELALLOC, 1); | |
6647 | found_end = range_start + found; | |
6648 | if (found_end < range_start) | |
6649 | found_end = (u64)-1; | |
6650 | ||
6651 | /* | |
6652 | * we didn't find anything useful, return | |
6653 | * the original results from get_extent() | |
6654 | */ | |
6655 | if (range_start > end || found_end <= start) { | |
6656 | em = hole_em; | |
6657 | hole_em = NULL; | |
6658 | goto out; | |
6659 | } | |
6660 | ||
6661 | /* adjust the range_start to make sure it doesn't | |
6662 | * go backwards from the start they passed in | |
6663 | */ | |
67871254 | 6664 | range_start = max(start, range_start); |
ec29ed5b CM |
6665 | found = found_end - range_start; |
6666 | ||
6667 | if (found > 0) { | |
6668 | u64 hole_start = start; | |
6669 | u64 hole_len = len; | |
6670 | ||
172ddd60 | 6671 | em = alloc_extent_map(); |
ec29ed5b CM |
6672 | if (!em) { |
6673 | err = -ENOMEM; | |
6674 | goto out; | |
6675 | } | |
6676 | /* | |
6677 | * when btrfs_get_extent can't find anything it | |
6678 | * returns one huge hole | |
6679 | * | |
6680 | * make sure what it found really fits our range, and | |
6681 | * adjust to make sure it is based on the start from | |
6682 | * the caller | |
6683 | */ | |
6684 | if (hole_em) { | |
6685 | u64 calc_end = extent_map_end(hole_em); | |
6686 | ||
6687 | if (calc_end <= start || (hole_em->start > end)) { | |
6688 | free_extent_map(hole_em); | |
6689 | hole_em = NULL; | |
6690 | } else { | |
6691 | hole_start = max(hole_em->start, start); | |
6692 | hole_len = calc_end - hole_start; | |
6693 | } | |
6694 | } | |
6695 | em->bdev = NULL; | |
6696 | if (hole_em && range_start > hole_start) { | |
6697 | /* our hole starts before our delalloc, so we | |
6698 | * have to return just the parts of the hole | |
6699 | * that go until the delalloc starts | |
6700 | */ | |
6701 | em->len = min(hole_len, | |
6702 | range_start - hole_start); | |
6703 | em->start = hole_start; | |
6704 | em->orig_start = hole_start; | |
6705 | /* | |
6706 | * don't adjust block start at all, | |
6707 | * it is fixed at EXTENT_MAP_HOLE | |
6708 | */ | |
6709 | em->block_start = hole_em->block_start; | |
6710 | em->block_len = hole_len; | |
f9e4fb53 LB |
6711 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
6712 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
6713 | } else { |
6714 | em->start = range_start; | |
6715 | em->len = found; | |
6716 | em->orig_start = range_start; | |
6717 | em->block_start = EXTENT_MAP_DELALLOC; | |
6718 | em->block_len = found; | |
6719 | } | |
6720 | } else if (hole_em) { | |
6721 | return hole_em; | |
6722 | } | |
6723 | out: | |
6724 | ||
6725 | free_extent_map(hole_em); | |
6726 | if (err) { | |
6727 | free_extent_map(em); | |
6728 | return ERR_PTR(err); | |
6729 | } | |
6730 | return em; | |
6731 | } | |
6732 | ||
4b46fce2 JB |
6733 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
6734 | u64 start, u64 len) | |
6735 | { | |
6736 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
70c8a91c | 6737 | struct extent_map *em; |
4b46fce2 JB |
6738 | struct btrfs_key ins; |
6739 | u64 alloc_hint; | |
6740 | int ret; | |
4b46fce2 | 6741 | |
4b46fce2 | 6742 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
00361589 | 6743 | ret = btrfs_reserve_extent(root, len, root->sectorsize, 0, |
e570fd27 | 6744 | alloc_hint, &ins, 1, 1); |
00361589 JB |
6745 | if (ret) |
6746 | return ERR_PTR(ret); | |
4b46fce2 | 6747 | |
70c8a91c | 6748 | em = create_pinned_em(inode, start, ins.offset, start, ins.objectid, |
cc95bef6 | 6749 | ins.offset, ins.offset, ins.offset, 0); |
00361589 | 6750 | if (IS_ERR(em)) { |
e570fd27 | 6751 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
00361589 JB |
6752 | return em; |
6753 | } | |
4b46fce2 JB |
6754 | |
6755 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
6756 | ins.offset, ins.offset, 0); | |
6757 | if (ret) { | |
e570fd27 | 6758 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
00361589 JB |
6759 | free_extent_map(em); |
6760 | return ERR_PTR(ret); | |
4b46fce2 | 6761 | } |
00361589 | 6762 | |
4b46fce2 JB |
6763 | return em; |
6764 | } | |
6765 | ||
46bfbb5c CM |
6766 | /* |
6767 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
6768 | * block must be cow'd | |
6769 | */ | |
00361589 | 6770 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
6771 | u64 *orig_start, u64 *orig_block_len, |
6772 | u64 *ram_bytes) | |
46bfbb5c | 6773 | { |
00361589 | 6774 | struct btrfs_trans_handle *trans; |
46bfbb5c CM |
6775 | struct btrfs_path *path; |
6776 | int ret; | |
6777 | struct extent_buffer *leaf; | |
6778 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 6779 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
6780 | struct btrfs_file_extent_item *fi; |
6781 | struct btrfs_key key; | |
6782 | u64 disk_bytenr; | |
6783 | u64 backref_offset; | |
6784 | u64 extent_end; | |
6785 | u64 num_bytes; | |
6786 | int slot; | |
6787 | int found_type; | |
7ee9e440 | 6788 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 6789 | |
46bfbb5c CM |
6790 | path = btrfs_alloc_path(); |
6791 | if (!path) | |
6792 | return -ENOMEM; | |
6793 | ||
00361589 | 6794 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), |
46bfbb5c CM |
6795 | offset, 0); |
6796 | if (ret < 0) | |
6797 | goto out; | |
6798 | ||
6799 | slot = path->slots[0]; | |
6800 | if (ret == 1) { | |
6801 | if (slot == 0) { | |
6802 | /* can't find the item, must cow */ | |
6803 | ret = 0; | |
6804 | goto out; | |
6805 | } | |
6806 | slot--; | |
6807 | } | |
6808 | ret = 0; | |
6809 | leaf = path->nodes[0]; | |
6810 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 6811 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
6812 | key.type != BTRFS_EXTENT_DATA_KEY) { |
6813 | /* not our file or wrong item type, must cow */ | |
6814 | goto out; | |
6815 | } | |
6816 | ||
6817 | if (key.offset > offset) { | |
6818 | /* Wrong offset, must cow */ | |
6819 | goto out; | |
6820 | } | |
6821 | ||
6822 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
6823 | found_type = btrfs_file_extent_type(leaf, fi); | |
6824 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
6825 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
6826 | /* not a regular extent, must cow */ | |
6827 | goto out; | |
6828 | } | |
7ee9e440 JB |
6829 | |
6830 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
6831 | goto out; | |
6832 | ||
e77751aa MX |
6833 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
6834 | if (extent_end <= offset) | |
6835 | goto out; | |
6836 | ||
46bfbb5c | 6837 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
6838 | if (disk_bytenr == 0) |
6839 | goto out; | |
6840 | ||
6841 | if (btrfs_file_extent_compression(leaf, fi) || | |
6842 | btrfs_file_extent_encryption(leaf, fi) || | |
6843 | btrfs_file_extent_other_encoding(leaf, fi)) | |
6844 | goto out; | |
6845 | ||
46bfbb5c CM |
6846 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
6847 | ||
7ee9e440 JB |
6848 | if (orig_start) { |
6849 | *orig_start = key.offset - backref_offset; | |
6850 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
6851 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
6852 | } | |
eb384b55 | 6853 | |
46bfbb5c CM |
6854 | if (btrfs_extent_readonly(root, disk_bytenr)) |
6855 | goto out; | |
7b2b7085 MX |
6856 | |
6857 | num_bytes = min(offset + *len, extent_end) - offset; | |
6858 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
6859 | u64 range_end; | |
6860 | ||
6861 | range_end = round_up(offset + num_bytes, root->sectorsize) - 1; | |
6862 | ret = test_range_bit(io_tree, offset, range_end, | |
6863 | EXTENT_DELALLOC, 0, NULL); | |
6864 | if (ret) { | |
6865 | ret = -EAGAIN; | |
6866 | goto out; | |
6867 | } | |
6868 | } | |
6869 | ||
1bda19eb | 6870 | btrfs_release_path(path); |
46bfbb5c CM |
6871 | |
6872 | /* | |
6873 | * look for other files referencing this extent, if we | |
6874 | * find any we must cow | |
6875 | */ | |
00361589 JB |
6876 | trans = btrfs_join_transaction(root); |
6877 | if (IS_ERR(trans)) { | |
6878 | ret = 0; | |
46bfbb5c | 6879 | goto out; |
00361589 JB |
6880 | } |
6881 | ||
6882 | ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), | |
6883 | key.offset - backref_offset, disk_bytenr); | |
6884 | btrfs_end_transaction(trans, root); | |
6885 | if (ret) { | |
6886 | ret = 0; | |
6887 | goto out; | |
6888 | } | |
46bfbb5c CM |
6889 | |
6890 | /* | |
6891 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
6892 | * in this extent we are about to write. If there | |
6893 | * are any csums in that range we have to cow in order | |
6894 | * to keep the csums correct | |
6895 | */ | |
6896 | disk_bytenr += backref_offset; | |
6897 | disk_bytenr += offset - key.offset; | |
46bfbb5c CM |
6898 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) |
6899 | goto out; | |
6900 | /* | |
6901 | * all of the above have passed, it is safe to overwrite this extent | |
6902 | * without cow | |
6903 | */ | |
eb384b55 | 6904 | *len = num_bytes; |
46bfbb5c CM |
6905 | ret = 1; |
6906 | out: | |
6907 | btrfs_free_path(path); | |
6908 | return ret; | |
6909 | } | |
6910 | ||
fc4adbff AG |
6911 | bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end) |
6912 | { | |
6913 | struct radix_tree_root *root = &inode->i_mapping->page_tree; | |
6914 | int found = false; | |
6915 | void **pagep = NULL; | |
6916 | struct page *page = NULL; | |
6917 | int start_idx; | |
6918 | int end_idx; | |
6919 | ||
6920 | start_idx = start >> PAGE_CACHE_SHIFT; | |
6921 | ||
6922 | /* | |
6923 | * end is the last byte in the last page. end == start is legal | |
6924 | */ | |
6925 | end_idx = end >> PAGE_CACHE_SHIFT; | |
6926 | ||
6927 | rcu_read_lock(); | |
6928 | ||
6929 | /* Most of the code in this while loop is lifted from | |
6930 | * find_get_page. It's been modified to begin searching from a | |
6931 | * page and return just the first page found in that range. If the | |
6932 | * found idx is less than or equal to the end idx then we know that | |
6933 | * a page exists. If no pages are found or if those pages are | |
6934 | * outside of the range then we're fine (yay!) */ | |
6935 | while (page == NULL && | |
6936 | radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) { | |
6937 | page = radix_tree_deref_slot(pagep); | |
6938 | if (unlikely(!page)) | |
6939 | break; | |
6940 | ||
6941 | if (radix_tree_exception(page)) { | |
809f9016 FM |
6942 | if (radix_tree_deref_retry(page)) { |
6943 | page = NULL; | |
fc4adbff | 6944 | continue; |
809f9016 | 6945 | } |
fc4adbff AG |
6946 | /* |
6947 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
6948 | * here as an exceptional entry: so return it without | |
6949 | * attempting to raise page count. | |
6950 | */ | |
6fdef6d4 | 6951 | page = NULL; |
fc4adbff AG |
6952 | break; /* TODO: Is this relevant for this use case? */ |
6953 | } | |
6954 | ||
91405151 FM |
6955 | if (!page_cache_get_speculative(page)) { |
6956 | page = NULL; | |
fc4adbff | 6957 | continue; |
91405151 | 6958 | } |
fc4adbff AG |
6959 | |
6960 | /* | |
6961 | * Has the page moved? | |
6962 | * This is part of the lockless pagecache protocol. See | |
6963 | * include/linux/pagemap.h for details. | |
6964 | */ | |
6965 | if (unlikely(page != *pagep)) { | |
6966 | page_cache_release(page); | |
6967 | page = NULL; | |
6968 | } | |
6969 | } | |
6970 | ||
6971 | if (page) { | |
6972 | if (page->index <= end_idx) | |
6973 | found = true; | |
6974 | page_cache_release(page); | |
6975 | } | |
6976 | ||
6977 | rcu_read_unlock(); | |
6978 | return found; | |
6979 | } | |
6980 | ||
eb838e73 JB |
6981 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
6982 | struct extent_state **cached_state, int writing) | |
6983 | { | |
6984 | struct btrfs_ordered_extent *ordered; | |
6985 | int ret = 0; | |
6986 | ||
6987 | while (1) { | |
6988 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6989 | 0, cached_state); | |
6990 | /* | |
6991 | * We're concerned with the entire range that we're going to be | |
6992 | * doing DIO to, so we need to make sure theres no ordered | |
6993 | * extents in this range. | |
6994 | */ | |
6995 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
6996 | lockend - lockstart + 1); | |
6997 | ||
6998 | /* | |
6999 | * We need to make sure there are no buffered pages in this | |
7000 | * range either, we could have raced between the invalidate in | |
7001 | * generic_file_direct_write and locking the extent. The | |
7002 | * invalidate needs to happen so that reads after a write do not | |
7003 | * get stale data. | |
7004 | */ | |
fc4adbff AG |
7005 | if (!ordered && |
7006 | (!writing || | |
7007 | !btrfs_page_exists_in_range(inode, lockstart, lockend))) | |
eb838e73 JB |
7008 | break; |
7009 | ||
7010 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7011 | cached_state, GFP_NOFS); | |
7012 | ||
7013 | if (ordered) { | |
7014 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7015 | btrfs_put_ordered_extent(ordered); | |
7016 | } else { | |
7017 | /* Screw you mmap */ | |
728404da | 7018 | ret = btrfs_fdatawrite_range(inode, lockstart, lockend); |
075bdbdb FM |
7019 | if (ret) |
7020 | break; | |
7021 | ret = filemap_fdatawait_range(inode->i_mapping, | |
7022 | lockstart, | |
7023 | lockend); | |
eb838e73 JB |
7024 | if (ret) |
7025 | break; | |
7026 | ||
7027 | /* | |
7028 | * If we found a page that couldn't be invalidated just | |
7029 | * fall back to buffered. | |
7030 | */ | |
7031 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
7032 | lockstart >> PAGE_CACHE_SHIFT, | |
7033 | lockend >> PAGE_CACHE_SHIFT); | |
7034 | if (ret) | |
7035 | break; | |
7036 | } | |
7037 | ||
7038 | cond_resched(); | |
7039 | } | |
7040 | ||
7041 | return ret; | |
7042 | } | |
7043 | ||
69ffb543 JB |
7044 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
7045 | u64 len, u64 orig_start, | |
7046 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
7047 | u64 orig_block_len, u64 ram_bytes, |
7048 | int type) | |
69ffb543 JB |
7049 | { |
7050 | struct extent_map_tree *em_tree; | |
7051 | struct extent_map *em; | |
7052 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7053 | int ret; | |
7054 | ||
7055 | em_tree = &BTRFS_I(inode)->extent_tree; | |
7056 | em = alloc_extent_map(); | |
7057 | if (!em) | |
7058 | return ERR_PTR(-ENOMEM); | |
7059 | ||
7060 | em->start = start; | |
7061 | em->orig_start = orig_start; | |
2ab28f32 JB |
7062 | em->mod_start = start; |
7063 | em->mod_len = len; | |
69ffb543 JB |
7064 | em->len = len; |
7065 | em->block_len = block_len; | |
7066 | em->block_start = block_start; | |
7067 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7068 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7069 | em->ram_bytes = ram_bytes; |
70c8a91c | 7070 | em->generation = -1; |
69ffb543 JB |
7071 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
7072 | if (type == BTRFS_ORDERED_PREALLOC) | |
b11e234d | 7073 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
69ffb543 JB |
7074 | |
7075 | do { | |
7076 | btrfs_drop_extent_cache(inode, em->start, | |
7077 | em->start + em->len - 1, 0); | |
7078 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7079 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 JB |
7080 | write_unlock(&em_tree->lock); |
7081 | } while (ret == -EEXIST); | |
7082 | ||
7083 | if (ret) { | |
7084 | free_extent_map(em); | |
7085 | return ERR_PTR(ret); | |
7086 | } | |
7087 | ||
7088 | return em; | |
7089 | } | |
7090 | ||
7091 | ||
4b46fce2 JB |
7092 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7093 | struct buffer_head *bh_result, int create) | |
7094 | { | |
7095 | struct extent_map *em; | |
7096 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 7097 | struct extent_state *cached_state = NULL; |
4b46fce2 | 7098 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7099 | u64 lockstart, lockend; |
4b46fce2 | 7100 | u64 len = bh_result->b_size; |
eb838e73 | 7101 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7102 | int ret = 0; |
eb838e73 | 7103 | |
172a5049 | 7104 | if (create) |
eb838e73 | 7105 | unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY; |
172a5049 | 7106 | else |
c329861d | 7107 | len = min_t(u64, len, root->sectorsize); |
eb838e73 | 7108 | |
c329861d JB |
7109 | lockstart = start; |
7110 | lockend = start + len - 1; | |
7111 | ||
eb838e73 JB |
7112 | /* |
7113 | * If this errors out it's because we couldn't invalidate pagecache for | |
7114 | * this range and we need to fallback to buffered. | |
7115 | */ | |
7116 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create)) | |
7117 | return -ENOTBLK; | |
7118 | ||
4b46fce2 | 7119 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); |
eb838e73 JB |
7120 | if (IS_ERR(em)) { |
7121 | ret = PTR_ERR(em); | |
7122 | goto unlock_err; | |
7123 | } | |
4b46fce2 JB |
7124 | |
7125 | /* | |
7126 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7127 | * io. INLINE is special, and we could probably kludge it in here, but | |
7128 | * it's still buffered so for safety lets just fall back to the generic | |
7129 | * buffered path. | |
7130 | * | |
7131 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7132 | * decompress it, so there will be buffering required no matter what we | |
7133 | * do, so go ahead and fallback to buffered. | |
7134 | * | |
7135 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
7136 | * to buffered IO. Don't blame me, this is the price we pay for using | |
7137 | * the generic code. | |
7138 | */ | |
7139 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7140 | em->block_start == EXTENT_MAP_INLINE) { | |
7141 | free_extent_map(em); | |
eb838e73 JB |
7142 | ret = -ENOTBLK; |
7143 | goto unlock_err; | |
4b46fce2 JB |
7144 | } |
7145 | ||
7146 | /* Just a good old fashioned hole, return */ | |
7147 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
7148 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
7149 | free_extent_map(em); | |
eb838e73 | 7150 | goto unlock_err; |
4b46fce2 JB |
7151 | } |
7152 | ||
7153 | /* | |
7154 | * We don't allocate a new extent in the following cases | |
7155 | * | |
7156 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7157 | * existing extent. | |
7158 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7159 | * just use the extent. | |
7160 | * | |
7161 | */ | |
46bfbb5c | 7162 | if (!create) { |
eb838e73 JB |
7163 | len = min(len, em->len - (start - em->start)); |
7164 | lockstart = start + len; | |
7165 | goto unlock; | |
46bfbb5c | 7166 | } |
4b46fce2 JB |
7167 | |
7168 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7169 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7170 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
7171 | int type; |
7172 | int ret; | |
eb384b55 | 7173 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
7174 | |
7175 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7176 | type = BTRFS_ORDERED_PREALLOC; | |
7177 | else | |
7178 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 7179 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 7180 | block_start = em->block_start + (start - em->start); |
46bfbb5c | 7181 | |
00361589 | 7182 | if (can_nocow_extent(inode, start, &len, &orig_start, |
7ee9e440 | 7183 | &orig_block_len, &ram_bytes) == 1) { |
69ffb543 JB |
7184 | if (type == BTRFS_ORDERED_PREALLOC) { |
7185 | free_extent_map(em); | |
7186 | em = create_pinned_em(inode, start, len, | |
7187 | orig_start, | |
b4939680 | 7188 | block_start, len, |
cc95bef6 JB |
7189 | orig_block_len, |
7190 | ram_bytes, type); | |
555e1286 FM |
7191 | if (IS_ERR(em)) { |
7192 | ret = PTR_ERR(em); | |
69ffb543 | 7193 | goto unlock_err; |
555e1286 | 7194 | } |
69ffb543 JB |
7195 | } |
7196 | ||
46bfbb5c CM |
7197 | ret = btrfs_add_ordered_extent_dio(inode, start, |
7198 | block_start, len, len, type); | |
46bfbb5c CM |
7199 | if (ret) { |
7200 | free_extent_map(em); | |
eb838e73 | 7201 | goto unlock_err; |
46bfbb5c CM |
7202 | } |
7203 | goto unlock; | |
4b46fce2 | 7204 | } |
4b46fce2 | 7205 | } |
00361589 | 7206 | |
46bfbb5c CM |
7207 | /* |
7208 | * this will cow the extent, reset the len in case we changed | |
7209 | * it above | |
7210 | */ | |
7211 | len = bh_result->b_size; | |
70c8a91c JB |
7212 | free_extent_map(em); |
7213 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
7214 | if (IS_ERR(em)) { |
7215 | ret = PTR_ERR(em); | |
7216 | goto unlock_err; | |
7217 | } | |
46bfbb5c CM |
7218 | len = min(len, em->len - (start - em->start)); |
7219 | unlock: | |
4b46fce2 JB |
7220 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
7221 | inode->i_blkbits; | |
46bfbb5c | 7222 | bh_result->b_size = len; |
4b46fce2 JB |
7223 | bh_result->b_bdev = em->bdev; |
7224 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
7225 | if (create) { |
7226 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7227 | set_buffer_new(bh_result); | |
7228 | ||
7229 | /* | |
7230 | * Need to update the i_size under the extent lock so buffered | |
7231 | * readers will get the updated i_size when we unlock. | |
7232 | */ | |
7233 | if (start + len > i_size_read(inode)) | |
7234 | i_size_write(inode, start + len); | |
0934856d | 7235 | |
172a5049 MX |
7236 | spin_lock(&BTRFS_I(inode)->lock); |
7237 | BTRFS_I(inode)->outstanding_extents++; | |
7238 | spin_unlock(&BTRFS_I(inode)->lock); | |
7239 | ||
0934856d MX |
7240 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
7241 | lockstart + len - 1, EXTENT_DELALLOC, NULL, | |
7242 | &cached_state, GFP_NOFS); | |
7243 | BUG_ON(ret); | |
c3473e83 | 7244 | } |
4b46fce2 | 7245 | |
eb838e73 JB |
7246 | /* |
7247 | * In the case of write we need to clear and unlock the entire range, | |
7248 | * in the case of read we need to unlock only the end area that we | |
7249 | * aren't using if there is any left over space. | |
7250 | */ | |
24c03fa5 | 7251 | if (lockstart < lockend) { |
0934856d MX |
7252 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
7253 | lockend, unlock_bits, 1, 0, | |
7254 | &cached_state, GFP_NOFS); | |
24c03fa5 | 7255 | } else { |
eb838e73 | 7256 | free_extent_state(cached_state); |
24c03fa5 | 7257 | } |
eb838e73 | 7258 | |
4b46fce2 JB |
7259 | free_extent_map(em); |
7260 | ||
7261 | return 0; | |
eb838e73 JB |
7262 | |
7263 | unlock_err: | |
eb838e73 JB |
7264 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7265 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
7266 | return ret; | |
4b46fce2 JB |
7267 | } |
7268 | ||
8b110e39 MX |
7269 | static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio, |
7270 | int rw, int mirror_num) | |
7271 | { | |
7272 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7273 | int ret; | |
7274 | ||
7275 | BUG_ON(rw & REQ_WRITE); | |
7276 | ||
7277 | bio_get(bio); | |
7278 | ||
7279 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, | |
7280 | BTRFS_WQ_ENDIO_DIO_REPAIR); | |
7281 | if (ret) | |
7282 | goto err; | |
7283 | ||
7284 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); | |
7285 | err: | |
7286 | bio_put(bio); | |
7287 | return ret; | |
7288 | } | |
7289 | ||
7290 | static int btrfs_check_dio_repairable(struct inode *inode, | |
7291 | struct bio *failed_bio, | |
7292 | struct io_failure_record *failrec, | |
7293 | int failed_mirror) | |
7294 | { | |
7295 | int num_copies; | |
7296 | ||
7297 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, | |
7298 | failrec->logical, failrec->len); | |
7299 | if (num_copies == 1) { | |
7300 | /* | |
7301 | * we only have a single copy of the data, so don't bother with | |
7302 | * all the retry and error correction code that follows. no | |
7303 | * matter what the error is, it is very likely to persist. | |
7304 | */ | |
7305 | pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
7306 | num_copies, failrec->this_mirror, failed_mirror); | |
7307 | return 0; | |
7308 | } | |
7309 | ||
7310 | failrec->failed_mirror = failed_mirror; | |
7311 | failrec->this_mirror++; | |
7312 | if (failrec->this_mirror == failed_mirror) | |
7313 | failrec->this_mirror++; | |
7314 | ||
7315 | if (failrec->this_mirror > num_copies) { | |
7316 | pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
7317 | num_copies, failrec->this_mirror, failed_mirror); | |
7318 | return 0; | |
7319 | } | |
7320 | ||
7321 | return 1; | |
7322 | } | |
7323 | ||
7324 | static int dio_read_error(struct inode *inode, struct bio *failed_bio, | |
7325 | struct page *page, u64 start, u64 end, | |
7326 | int failed_mirror, bio_end_io_t *repair_endio, | |
7327 | void *repair_arg) | |
7328 | { | |
7329 | struct io_failure_record *failrec; | |
7330 | struct bio *bio; | |
7331 | int isector; | |
7332 | int read_mode; | |
7333 | int ret; | |
7334 | ||
7335 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
7336 | ||
7337 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
7338 | if (ret) | |
7339 | return ret; | |
7340 | ||
7341 | ret = btrfs_check_dio_repairable(inode, failed_bio, failrec, | |
7342 | failed_mirror); | |
7343 | if (!ret) { | |
7344 | free_io_failure(inode, failrec); | |
7345 | return -EIO; | |
7346 | } | |
7347 | ||
7348 | if (failed_bio->bi_vcnt > 1) | |
7349 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
7350 | else | |
7351 | read_mode = READ_SYNC; | |
7352 | ||
7353 | isector = start - btrfs_io_bio(failed_bio)->logical; | |
7354 | isector >>= inode->i_sb->s_blocksize_bits; | |
7355 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
7356 | 0, isector, repair_endio, repair_arg); | |
7357 | if (!bio) { | |
7358 | free_io_failure(inode, failrec); | |
7359 | return -EIO; | |
7360 | } | |
7361 | ||
7362 | btrfs_debug(BTRFS_I(inode)->root->fs_info, | |
7363 | "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n", | |
7364 | read_mode, failrec->this_mirror, failrec->in_validation); | |
7365 | ||
7366 | ret = submit_dio_repair_bio(inode, bio, read_mode, | |
7367 | failrec->this_mirror); | |
7368 | if (ret) { | |
7369 | free_io_failure(inode, failrec); | |
7370 | bio_put(bio); | |
7371 | } | |
7372 | ||
7373 | return ret; | |
7374 | } | |
7375 | ||
7376 | struct btrfs_retry_complete { | |
7377 | struct completion done; | |
7378 | struct inode *inode; | |
7379 | u64 start; | |
7380 | int uptodate; | |
7381 | }; | |
7382 | ||
7383 | static void btrfs_retry_endio_nocsum(struct bio *bio, int err) | |
7384 | { | |
7385 | struct btrfs_retry_complete *done = bio->bi_private; | |
7386 | struct bio_vec *bvec; | |
7387 | int i; | |
7388 | ||
7389 | if (err) | |
7390 | goto end; | |
7391 | ||
7392 | done->uptodate = 1; | |
7393 | bio_for_each_segment_all(bvec, bio, i) | |
7394 | clean_io_failure(done->inode, done->start, bvec->bv_page, 0); | |
7395 | end: | |
7396 | complete(&done->done); | |
7397 | bio_put(bio); | |
7398 | } | |
7399 | ||
7400 | static int __btrfs_correct_data_nocsum(struct inode *inode, | |
7401 | struct btrfs_io_bio *io_bio) | |
4b46fce2 | 7402 | { |
2c30c71b | 7403 | struct bio_vec *bvec; |
8b110e39 | 7404 | struct btrfs_retry_complete done; |
4b46fce2 | 7405 | u64 start; |
2c30c71b | 7406 | int i; |
c1dc0896 | 7407 | int ret; |
4b46fce2 | 7408 | |
8b110e39 MX |
7409 | start = io_bio->logical; |
7410 | done.inode = inode; | |
7411 | ||
7412 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { | |
7413 | try_again: | |
7414 | done.uptodate = 0; | |
7415 | done.start = start; | |
7416 | init_completion(&done.done); | |
7417 | ||
7418 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start, | |
7419 | start + bvec->bv_len - 1, | |
7420 | io_bio->mirror_num, | |
7421 | btrfs_retry_endio_nocsum, &done); | |
7422 | if (ret) | |
7423 | return ret; | |
7424 | ||
7425 | wait_for_completion(&done.done); | |
7426 | ||
7427 | if (!done.uptodate) { | |
7428 | /* We might have another mirror, so try again */ | |
7429 | goto try_again; | |
7430 | } | |
7431 | ||
7432 | start += bvec->bv_len; | |
7433 | } | |
7434 | ||
7435 | return 0; | |
7436 | } | |
7437 | ||
7438 | static void btrfs_retry_endio(struct bio *bio, int err) | |
7439 | { | |
7440 | struct btrfs_retry_complete *done = bio->bi_private; | |
7441 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
7442 | struct bio_vec *bvec; | |
7443 | int uptodate; | |
7444 | int ret; | |
7445 | int i; | |
7446 | ||
7447 | if (err) | |
7448 | goto end; | |
7449 | ||
7450 | uptodate = 1; | |
7451 | bio_for_each_segment_all(bvec, bio, i) { | |
7452 | ret = __readpage_endio_check(done->inode, io_bio, i, | |
7453 | bvec->bv_page, 0, | |
7454 | done->start, bvec->bv_len); | |
7455 | if (!ret) | |
7456 | clean_io_failure(done->inode, done->start, | |
7457 | bvec->bv_page, 0); | |
7458 | else | |
7459 | uptodate = 0; | |
7460 | } | |
7461 | ||
7462 | done->uptodate = uptodate; | |
7463 | end: | |
7464 | complete(&done->done); | |
7465 | bio_put(bio); | |
7466 | } | |
7467 | ||
7468 | static int __btrfs_subio_endio_read(struct inode *inode, | |
7469 | struct btrfs_io_bio *io_bio, int err) | |
7470 | { | |
7471 | struct bio_vec *bvec; | |
7472 | struct btrfs_retry_complete done; | |
7473 | u64 start; | |
7474 | u64 offset = 0; | |
7475 | int i; | |
7476 | int ret; | |
dc380aea | 7477 | |
8b110e39 | 7478 | err = 0; |
c1dc0896 | 7479 | start = io_bio->logical; |
8b110e39 MX |
7480 | done.inode = inode; |
7481 | ||
c1dc0896 | 7482 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { |
dc380aea MX |
7483 | ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page, |
7484 | 0, start, bvec->bv_len); | |
8b110e39 MX |
7485 | if (likely(!ret)) |
7486 | goto next; | |
7487 | try_again: | |
7488 | done.uptodate = 0; | |
7489 | done.start = start; | |
7490 | init_completion(&done.done); | |
7491 | ||
7492 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start, | |
7493 | start + bvec->bv_len - 1, | |
7494 | io_bio->mirror_num, | |
7495 | btrfs_retry_endio, &done); | |
7496 | if (ret) { | |
7497 | err = ret; | |
7498 | goto next; | |
7499 | } | |
7500 | ||
7501 | wait_for_completion(&done.done); | |
7502 | ||
7503 | if (!done.uptodate) { | |
7504 | /* We might have another mirror, so try again */ | |
7505 | goto try_again; | |
7506 | } | |
7507 | next: | |
7508 | offset += bvec->bv_len; | |
4b46fce2 | 7509 | start += bvec->bv_len; |
2c30c71b | 7510 | } |
c1dc0896 MX |
7511 | |
7512 | return err; | |
7513 | } | |
7514 | ||
8b110e39 MX |
7515 | static int btrfs_subio_endio_read(struct inode *inode, |
7516 | struct btrfs_io_bio *io_bio, int err) | |
7517 | { | |
7518 | bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
7519 | ||
7520 | if (skip_csum) { | |
7521 | if (unlikely(err)) | |
7522 | return __btrfs_correct_data_nocsum(inode, io_bio); | |
7523 | else | |
7524 | return 0; | |
7525 | } else { | |
7526 | return __btrfs_subio_endio_read(inode, io_bio, err); | |
7527 | } | |
7528 | } | |
7529 | ||
c1dc0896 MX |
7530 | static void btrfs_endio_direct_read(struct bio *bio, int err) |
7531 | { | |
7532 | struct btrfs_dio_private *dip = bio->bi_private; | |
7533 | struct inode *inode = dip->inode; | |
7534 | struct bio *dio_bio; | |
7535 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
7536 | ||
8b110e39 MX |
7537 | if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) |
7538 | err = btrfs_subio_endio_read(inode, io_bio, err); | |
c1dc0896 | 7539 | |
4b46fce2 | 7540 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, |
d0082371 | 7541 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 7542 | dio_bio = dip->dio_bio; |
4b46fce2 | 7543 | |
4b46fce2 | 7544 | kfree(dip); |
c0da7aa1 JB |
7545 | |
7546 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
7547 | if (err) | |
9be3395b CM |
7548 | clear_bit(BIO_UPTODATE, &dio_bio->bi_flags); |
7549 | dio_end_io(dio_bio, err); | |
23ea8e5a MX |
7550 | |
7551 | if (io_bio->end_io) | |
7552 | io_bio->end_io(io_bio, err); | |
9be3395b | 7553 | bio_put(bio); |
4b46fce2 JB |
7554 | } |
7555 | ||
7556 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
7557 | { | |
7558 | struct btrfs_dio_private *dip = bio->bi_private; | |
7559 | struct inode *inode = dip->inode; | |
7560 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b46fce2 | 7561 | struct btrfs_ordered_extent *ordered = NULL; |
163cf09c CM |
7562 | u64 ordered_offset = dip->logical_offset; |
7563 | u64 ordered_bytes = dip->bytes; | |
9be3395b | 7564 | struct bio *dio_bio; |
4b46fce2 JB |
7565 | int ret; |
7566 | ||
7567 | if (err) | |
7568 | goto out_done; | |
163cf09c CM |
7569 | again: |
7570 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
7571 | &ordered_offset, | |
5fd02043 | 7572 | ordered_bytes, !err); |
4b46fce2 | 7573 | if (!ret) |
163cf09c | 7574 | goto out_test; |
4b46fce2 | 7575 | |
9e0af237 LB |
7576 | btrfs_init_work(&ordered->work, btrfs_endio_write_helper, |
7577 | finish_ordered_fn, NULL, NULL); | |
fccb5d86 QW |
7578 | btrfs_queue_work(root->fs_info->endio_write_workers, |
7579 | &ordered->work); | |
163cf09c CM |
7580 | out_test: |
7581 | /* | |
7582 | * our bio might span multiple ordered extents. If we haven't | |
7583 | * completed the accounting for the whole dio, go back and try again | |
7584 | */ | |
7585 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
7586 | ordered_bytes = dip->logical_offset + dip->bytes - | |
7587 | ordered_offset; | |
5fd02043 | 7588 | ordered = NULL; |
163cf09c CM |
7589 | goto again; |
7590 | } | |
4b46fce2 | 7591 | out_done: |
9be3395b | 7592 | dio_bio = dip->dio_bio; |
4b46fce2 | 7593 | |
4b46fce2 | 7594 | kfree(dip); |
c0da7aa1 JB |
7595 | |
7596 | /* If we had an error make sure to clear the uptodate flag */ | |
7597 | if (err) | |
9be3395b CM |
7598 | clear_bit(BIO_UPTODATE, &dio_bio->bi_flags); |
7599 | dio_end_io(dio_bio, err); | |
7600 | bio_put(bio); | |
4b46fce2 JB |
7601 | } |
7602 | ||
eaf25d93 CM |
7603 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
7604 | struct bio *bio, int mirror_num, | |
7605 | unsigned long bio_flags, u64 offset) | |
7606 | { | |
7607 | int ret; | |
7608 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7609 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 7610 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
7611 | return 0; |
7612 | } | |
7613 | ||
e65e1535 MX |
7614 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
7615 | { | |
7616 | struct btrfs_dio_private *dip = bio->bi_private; | |
7617 | ||
8b110e39 MX |
7618 | if (err) |
7619 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
7620 | "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d", | |
7621 | btrfs_ino(dip->inode), bio->bi_rw, | |
7622 | (unsigned long long)bio->bi_iter.bi_sector, | |
7623 | bio->bi_iter.bi_size, err); | |
7624 | ||
7625 | if (dip->subio_endio) | |
7626 | err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err); | |
c1dc0896 MX |
7627 | |
7628 | if (err) { | |
e65e1535 MX |
7629 | dip->errors = 1; |
7630 | ||
7631 | /* | |
7632 | * before atomic variable goto zero, we must make sure | |
7633 | * dip->errors is perceived to be set. | |
7634 | */ | |
4e857c58 | 7635 | smp_mb__before_atomic(); |
e65e1535 MX |
7636 | } |
7637 | ||
7638 | /* if there are more bios still pending for this dio, just exit */ | |
7639 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
7640 | goto out; | |
7641 | ||
9be3395b | 7642 | if (dip->errors) { |
e65e1535 | 7643 | bio_io_error(dip->orig_bio); |
9be3395b CM |
7644 | } else { |
7645 | set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags); | |
e65e1535 MX |
7646 | bio_endio(dip->orig_bio, 0); |
7647 | } | |
7648 | out: | |
7649 | bio_put(bio); | |
7650 | } | |
7651 | ||
7652 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
7653 | u64 first_sector, gfp_t gfp_flags) | |
7654 | { | |
7655 | int nr_vecs = bio_get_nr_vecs(bdev); | |
7656 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
7657 | } | |
7658 | ||
c1dc0896 MX |
7659 | static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root, |
7660 | struct inode *inode, | |
7661 | struct btrfs_dio_private *dip, | |
7662 | struct bio *bio, | |
7663 | u64 file_offset) | |
7664 | { | |
7665 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
7666 | struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio); | |
7667 | int ret; | |
7668 | ||
7669 | /* | |
7670 | * We load all the csum data we need when we submit | |
7671 | * the first bio to reduce the csum tree search and | |
7672 | * contention. | |
7673 | */ | |
7674 | if (dip->logical_offset == file_offset) { | |
7675 | ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio, | |
7676 | file_offset); | |
7677 | if (ret) | |
7678 | return ret; | |
7679 | } | |
7680 | ||
7681 | if (bio == dip->orig_bio) | |
7682 | return 0; | |
7683 | ||
7684 | file_offset -= dip->logical_offset; | |
7685 | file_offset >>= inode->i_sb->s_blocksize_bits; | |
7686 | io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset); | |
7687 | ||
7688 | return 0; | |
7689 | } | |
7690 | ||
e65e1535 MX |
7691 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, |
7692 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 7693 | int async_submit) |
e65e1535 | 7694 | { |
facc8a22 | 7695 | struct btrfs_dio_private *dip = bio->bi_private; |
e65e1535 MX |
7696 | int write = rw & REQ_WRITE; |
7697 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7698 | int ret; | |
7699 | ||
b812ce28 JB |
7700 | if (async_submit) |
7701 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
7702 | ||
e65e1535 | 7703 | bio_get(bio); |
5fd02043 JB |
7704 | |
7705 | if (!write) { | |
bfebd8b5 DS |
7706 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, |
7707 | BTRFS_WQ_ENDIO_DATA); | |
5fd02043 JB |
7708 | if (ret) |
7709 | goto err; | |
7710 | } | |
e65e1535 | 7711 | |
1ae39938 JB |
7712 | if (skip_sum) |
7713 | goto map; | |
7714 | ||
7715 | if (write && async_submit) { | |
e65e1535 MX |
7716 | ret = btrfs_wq_submit_bio(root->fs_info, |
7717 | inode, rw, bio, 0, 0, | |
7718 | file_offset, | |
7719 | __btrfs_submit_bio_start_direct_io, | |
7720 | __btrfs_submit_bio_done); | |
7721 | goto err; | |
1ae39938 JB |
7722 | } else if (write) { |
7723 | /* | |
7724 | * If we aren't doing async submit, calculate the csum of the | |
7725 | * bio now. | |
7726 | */ | |
7727 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
7728 | if (ret) | |
7729 | goto err; | |
23ea8e5a | 7730 | } else { |
c1dc0896 MX |
7731 | ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio, |
7732 | file_offset); | |
c2db1073 TI |
7733 | if (ret) |
7734 | goto err; | |
7735 | } | |
1ae39938 JB |
7736 | map: |
7737 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
7738 | err: |
7739 | bio_put(bio); | |
7740 | return ret; | |
7741 | } | |
7742 | ||
7743 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
7744 | int skip_sum) | |
7745 | { | |
7746 | struct inode *inode = dip->inode; | |
7747 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
e65e1535 MX |
7748 | struct bio *bio; |
7749 | struct bio *orig_bio = dip->orig_bio; | |
7750 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
4f024f37 | 7751 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 MX |
7752 | u64 file_offset = dip->logical_offset; |
7753 | u64 submit_len = 0; | |
7754 | u64 map_length; | |
7755 | int nr_pages = 0; | |
23ea8e5a | 7756 | int ret; |
1ae39938 | 7757 | int async_submit = 0; |
e65e1535 | 7758 | |
4f024f37 | 7759 | map_length = orig_bio->bi_iter.bi_size; |
53b381b3 | 7760 | ret = btrfs_map_block(root->fs_info, rw, start_sector << 9, |
e65e1535 | 7761 | &map_length, NULL, 0); |
7a5c3c9b | 7762 | if (ret) |
e65e1535 | 7763 | return -EIO; |
facc8a22 | 7764 | |
4f024f37 | 7765 | if (map_length >= orig_bio->bi_iter.bi_size) { |
02f57c7a | 7766 | bio = orig_bio; |
c1dc0896 | 7767 | dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED; |
02f57c7a JB |
7768 | goto submit; |
7769 | } | |
7770 | ||
53b381b3 DW |
7771 | /* async crcs make it difficult to collect full stripe writes. */ |
7772 | if (btrfs_get_alloc_profile(root, 1) & | |
7773 | (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) | |
7774 | async_submit = 0; | |
7775 | else | |
7776 | async_submit = 1; | |
7777 | ||
02f57c7a JB |
7778 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
7779 | if (!bio) | |
7780 | return -ENOMEM; | |
7a5c3c9b | 7781 | |
02f57c7a JB |
7782 | bio->bi_private = dip; |
7783 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 7784 | btrfs_io_bio(bio)->logical = file_offset; |
02f57c7a JB |
7785 | atomic_inc(&dip->pending_bios); |
7786 | ||
e65e1535 | 7787 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
ee39b432 | 7788 | if (map_length < submit_len + bvec->bv_len || |
e65e1535 | 7789 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, |
ee39b432 | 7790 | bvec->bv_offset) < bvec->bv_len) { |
e65e1535 MX |
7791 | /* |
7792 | * inc the count before we submit the bio so | |
7793 | * we know the end IO handler won't happen before | |
7794 | * we inc the count. Otherwise, the dip might get freed | |
7795 | * before we're done setting it up | |
7796 | */ | |
7797 | atomic_inc(&dip->pending_bios); | |
7798 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
7799 | file_offset, skip_sum, | |
c329861d | 7800 | async_submit); |
e65e1535 MX |
7801 | if (ret) { |
7802 | bio_put(bio); | |
7803 | atomic_dec(&dip->pending_bios); | |
7804 | goto out_err; | |
7805 | } | |
7806 | ||
e65e1535 MX |
7807 | start_sector += submit_len >> 9; |
7808 | file_offset += submit_len; | |
7809 | ||
7810 | submit_len = 0; | |
7811 | nr_pages = 0; | |
7812 | ||
7813 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
7814 | start_sector, GFP_NOFS); | |
7815 | if (!bio) | |
7816 | goto out_err; | |
7817 | bio->bi_private = dip; | |
7818 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 7819 | btrfs_io_bio(bio)->logical = file_offset; |
e65e1535 | 7820 | |
4f024f37 | 7821 | map_length = orig_bio->bi_iter.bi_size; |
53b381b3 | 7822 | ret = btrfs_map_block(root->fs_info, rw, |
3ec706c8 | 7823 | start_sector << 9, |
e65e1535 MX |
7824 | &map_length, NULL, 0); |
7825 | if (ret) { | |
7826 | bio_put(bio); | |
7827 | goto out_err; | |
7828 | } | |
7829 | } else { | |
7830 | submit_len += bvec->bv_len; | |
67871254 | 7831 | nr_pages++; |
e65e1535 MX |
7832 | bvec++; |
7833 | } | |
7834 | } | |
7835 | ||
02f57c7a | 7836 | submit: |
e65e1535 | 7837 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 7838 | async_submit); |
e65e1535 MX |
7839 | if (!ret) |
7840 | return 0; | |
7841 | ||
7842 | bio_put(bio); | |
7843 | out_err: | |
7844 | dip->errors = 1; | |
7845 | /* | |
7846 | * before atomic variable goto zero, we must | |
7847 | * make sure dip->errors is perceived to be set. | |
7848 | */ | |
4e857c58 | 7849 | smp_mb__before_atomic(); |
e65e1535 MX |
7850 | if (atomic_dec_and_test(&dip->pending_bios)) |
7851 | bio_io_error(dip->orig_bio); | |
7852 | ||
7853 | /* bio_end_io() will handle error, so we needn't return it */ | |
7854 | return 0; | |
7855 | } | |
7856 | ||
9be3395b CM |
7857 | static void btrfs_submit_direct(int rw, struct bio *dio_bio, |
7858 | struct inode *inode, loff_t file_offset) | |
4b46fce2 JB |
7859 | { |
7860 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7861 | struct btrfs_dio_private *dip; | |
9be3395b | 7862 | struct bio *io_bio; |
23ea8e5a | 7863 | struct btrfs_io_bio *btrfs_bio; |
4b46fce2 | 7864 | int skip_sum; |
7b6d91da | 7865 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
7866 | int ret = 0; |
7867 | ||
7868 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
7869 | ||
9be3395b | 7870 | io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS); |
9be3395b CM |
7871 | if (!io_bio) { |
7872 | ret = -ENOMEM; | |
7873 | goto free_ordered; | |
7874 | } | |
7875 | ||
c1dc0896 | 7876 | dip = kzalloc(sizeof(*dip), GFP_NOFS); |
4b46fce2 JB |
7877 | if (!dip) { |
7878 | ret = -ENOMEM; | |
9be3395b | 7879 | goto free_io_bio; |
4b46fce2 | 7880 | } |
4b46fce2 | 7881 | |
9be3395b | 7882 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
7883 | dip->inode = inode; |
7884 | dip->logical_offset = file_offset; | |
4f024f37 KO |
7885 | dip->bytes = dio_bio->bi_iter.bi_size; |
7886 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
9be3395b | 7887 | io_bio->bi_private = dip; |
9be3395b CM |
7888 | dip->orig_bio = io_bio; |
7889 | dip->dio_bio = dio_bio; | |
e65e1535 | 7890 | atomic_set(&dip->pending_bios, 0); |
c1dc0896 MX |
7891 | btrfs_bio = btrfs_io_bio(io_bio); |
7892 | btrfs_bio->logical = file_offset; | |
4b46fce2 | 7893 | |
c1dc0896 | 7894 | if (write) { |
9be3395b | 7895 | io_bio->bi_end_io = btrfs_endio_direct_write; |
c1dc0896 | 7896 | } else { |
9be3395b | 7897 | io_bio->bi_end_io = btrfs_endio_direct_read; |
c1dc0896 MX |
7898 | dip->subio_endio = btrfs_subio_endio_read; |
7899 | } | |
4b46fce2 | 7900 | |
e65e1535 MX |
7901 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
7902 | if (!ret) | |
eaf25d93 | 7903 | return; |
9be3395b | 7904 | |
23ea8e5a MX |
7905 | if (btrfs_bio->end_io) |
7906 | btrfs_bio->end_io(btrfs_bio, ret); | |
9be3395b CM |
7907 | free_io_bio: |
7908 | bio_put(io_bio); | |
7909 | ||
4b46fce2 JB |
7910 | free_ordered: |
7911 | /* | |
7912 | * If this is a write, we need to clean up the reserved space and kill | |
7913 | * the ordered extent. | |
7914 | */ | |
7915 | if (write) { | |
7916 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 7917 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
7918 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
7919 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
7920 | btrfs_free_reserved_extent(root, ordered->start, | |
e570fd27 | 7921 | ordered->disk_len, 1); |
4b46fce2 JB |
7922 | btrfs_put_ordered_extent(ordered); |
7923 | btrfs_put_ordered_extent(ordered); | |
7924 | } | |
9be3395b | 7925 | bio_endio(dio_bio, ret); |
4b46fce2 JB |
7926 | } |
7927 | ||
5a5f79b5 | 7928 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
28060d5d | 7929 | const struct iov_iter *iter, loff_t offset) |
5a5f79b5 CM |
7930 | { |
7931 | int seg; | |
a1b75f7d | 7932 | int i; |
5a5f79b5 CM |
7933 | unsigned blocksize_mask = root->sectorsize - 1; |
7934 | ssize_t retval = -EINVAL; | |
5a5f79b5 CM |
7935 | |
7936 | if (offset & blocksize_mask) | |
7937 | goto out; | |
7938 | ||
28060d5d AV |
7939 | if (iov_iter_alignment(iter) & blocksize_mask) |
7940 | goto out; | |
a1b75f7d | 7941 | |
28060d5d AV |
7942 | /* If this is a write we don't need to check anymore */ |
7943 | if (rw & WRITE) | |
7944 | return 0; | |
7945 | /* | |
7946 | * Check to make sure we don't have duplicate iov_base's in this | |
7947 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
7948 | * when reading back. | |
7949 | */ | |
7950 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
7951 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
7952 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
7953 | goto out; |
7954 | } | |
5a5f79b5 CM |
7955 | } |
7956 | retval = 0; | |
7957 | out: | |
7958 | return retval; | |
7959 | } | |
eb838e73 | 7960 | |
16432985 | 7961 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
d8d3d94b | 7962 | struct iov_iter *iter, loff_t offset) |
16432985 | 7963 | { |
4b46fce2 JB |
7964 | struct file *file = iocb->ki_filp; |
7965 | struct inode *inode = file->f_mapping->host; | |
0934856d | 7966 | size_t count = 0; |
2e60a51e | 7967 | int flags = 0; |
38851cc1 MX |
7968 | bool wakeup = true; |
7969 | bool relock = false; | |
0934856d | 7970 | ssize_t ret; |
4b46fce2 | 7971 | |
28060d5d | 7972 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset)) |
5a5f79b5 | 7973 | return 0; |
3f7c579c | 7974 | |
38851cc1 | 7975 | atomic_inc(&inode->i_dio_count); |
4e857c58 | 7976 | smp_mb__after_atomic(); |
38851cc1 | 7977 | |
0e267c44 | 7978 | /* |
41bd9ca4 MX |
7979 | * The generic stuff only does filemap_write_and_wait_range, which |
7980 | * isn't enough if we've written compressed pages to this area, so | |
7981 | * we need to flush the dirty pages again to make absolutely sure | |
7982 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 7983 | */ |
a6cbcd4a | 7984 | count = iov_iter_count(iter); |
41bd9ca4 MX |
7985 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
7986 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
7987 | filemap_fdatawrite_range(inode->i_mapping, offset, |
7988 | offset + count - 1); | |
0e267c44 | 7989 | |
0934856d | 7990 | if (rw & WRITE) { |
38851cc1 MX |
7991 | /* |
7992 | * If the write DIO is beyond the EOF, we need update | |
7993 | * the isize, but it is protected by i_mutex. So we can | |
7994 | * not unlock the i_mutex at this case. | |
7995 | */ | |
7996 | if (offset + count <= inode->i_size) { | |
7997 | mutex_unlock(&inode->i_mutex); | |
7998 | relock = true; | |
7999 | } | |
0934856d MX |
8000 | ret = btrfs_delalloc_reserve_space(inode, count); |
8001 | if (ret) | |
38851cc1 | 8002 | goto out; |
ee39b432 DS |
8003 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
8004 | &BTRFS_I(inode)->runtime_flags)) { | |
38851cc1 MX |
8005 | inode_dio_done(inode); |
8006 | flags = DIO_LOCKING | DIO_SKIP_HOLES; | |
8007 | wakeup = false; | |
0934856d MX |
8008 | } |
8009 | ||
8010 | ret = __blockdev_direct_IO(rw, iocb, inode, | |
8011 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
31b14039 | 8012 | iter, offset, btrfs_get_blocks_direct, NULL, |
2e60a51e | 8013 | btrfs_submit_direct, flags); |
0934856d MX |
8014 | if (rw & WRITE) { |
8015 | if (ret < 0 && ret != -EIOCBQUEUED) | |
8016 | btrfs_delalloc_release_space(inode, count); | |
172a5049 | 8017 | else if (ret >= 0 && (size_t)ret < count) |
0934856d MX |
8018 | btrfs_delalloc_release_space(inode, |
8019 | count - (size_t)ret); | |
172a5049 MX |
8020 | else |
8021 | btrfs_delalloc_release_metadata(inode, 0); | |
0934856d | 8022 | } |
38851cc1 | 8023 | out: |
2e60a51e MX |
8024 | if (wakeup) |
8025 | inode_dio_done(inode); | |
38851cc1 MX |
8026 | if (relock) |
8027 | mutex_lock(&inode->i_mutex); | |
0934856d MX |
8028 | |
8029 | return ret; | |
16432985 CM |
8030 | } |
8031 | ||
05dadc09 TI |
8032 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
8033 | ||
1506fcc8 YS |
8034 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
8035 | __u64 start, __u64 len) | |
8036 | { | |
05dadc09 TI |
8037 | int ret; |
8038 | ||
8039 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
8040 | if (ret) | |
8041 | return ret; | |
8042 | ||
ec29ed5b | 8043 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
8044 | } |
8045 | ||
a52d9a80 | 8046 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8047 | { |
d1310b2e CM |
8048 | struct extent_io_tree *tree; |
8049 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 8050 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 8051 | } |
1832a6d5 | 8052 | |
a52d9a80 | 8053 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8054 | { |
d1310b2e | 8055 | struct extent_io_tree *tree; |
b888db2b CM |
8056 | |
8057 | ||
8058 | if (current->flags & PF_MEMALLOC) { | |
8059 | redirty_page_for_writepage(wbc, page); | |
8060 | unlock_page(page); | |
8061 | return 0; | |
8062 | } | |
d1310b2e | 8063 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 8064 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
8065 | } |
8066 | ||
48a3b636 ES |
8067 | static int btrfs_writepages(struct address_space *mapping, |
8068 | struct writeback_control *wbc) | |
b293f02e | 8069 | { |
d1310b2e | 8070 | struct extent_io_tree *tree; |
771ed689 | 8071 | |
d1310b2e | 8072 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
8073 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
8074 | } | |
8075 | ||
3ab2fb5a CM |
8076 | static int |
8077 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
8078 | struct list_head *pages, unsigned nr_pages) | |
8079 | { | |
d1310b2e CM |
8080 | struct extent_io_tree *tree; |
8081 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
8082 | return extent_readpages(tree, mapping, pages, nr_pages, |
8083 | btrfs_get_extent); | |
8084 | } | |
e6dcd2dc | 8085 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8086 | { |
d1310b2e CM |
8087 | struct extent_io_tree *tree; |
8088 | struct extent_map_tree *map; | |
a52d9a80 | 8089 | int ret; |
8c2383c3 | 8090 | |
d1310b2e CM |
8091 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
8092 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 8093 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
8094 | if (ret == 1) { |
8095 | ClearPagePrivate(page); | |
8096 | set_page_private(page, 0); | |
8097 | page_cache_release(page); | |
39279cc3 | 8098 | } |
a52d9a80 | 8099 | return ret; |
39279cc3 CM |
8100 | } |
8101 | ||
e6dcd2dc CM |
8102 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8103 | { | |
98509cfc CM |
8104 | if (PageWriteback(page) || PageDirty(page)) |
8105 | return 0; | |
b335b003 | 8106 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
8107 | } |
8108 | ||
d47992f8 LC |
8109 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8110 | unsigned int length) | |
39279cc3 | 8111 | { |
5fd02043 | 8112 | struct inode *inode = page->mapping->host; |
d1310b2e | 8113 | struct extent_io_tree *tree; |
e6dcd2dc | 8114 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8115 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
8116 | u64 page_start = page_offset(page); |
8117 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
131e404a | 8118 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 8119 | |
8b62b72b CM |
8120 | /* |
8121 | * we have the page locked, so new writeback can't start, | |
8122 | * and the dirty bit won't be cleared while we are here. | |
8123 | * | |
8124 | * Wait for IO on this page so that we can safely clear | |
8125 | * the PagePrivate2 bit and do ordered accounting | |
8126 | */ | |
e6dcd2dc | 8127 | wait_on_page_writeback(page); |
8b62b72b | 8128 | |
5fd02043 | 8129 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
8130 | if (offset) { |
8131 | btrfs_releasepage(page, GFP_NOFS); | |
8132 | return; | |
8133 | } | |
131e404a FDBM |
8134 | |
8135 | if (!inode_evicting) | |
8136 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); | |
8137 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
e6dcd2dc | 8138 | if (ordered) { |
eb84ae03 CM |
8139 | /* |
8140 | * IO on this page will never be started, so we need | |
8141 | * to account for any ordered extents now | |
8142 | */ | |
131e404a FDBM |
8143 | if (!inode_evicting) |
8144 | clear_extent_bit(tree, page_start, page_end, | |
8145 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
8146 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | | |
8147 | EXTENT_DEFRAG, 1, 0, &cached_state, | |
8148 | GFP_NOFS); | |
8b62b72b CM |
8149 | /* |
8150 | * whoever cleared the private bit is responsible | |
8151 | * for the finish_ordered_io | |
8152 | */ | |
77cef2ec JB |
8153 | if (TestClearPagePrivate2(page)) { |
8154 | struct btrfs_ordered_inode_tree *tree; | |
8155 | u64 new_len; | |
8156 | ||
8157 | tree = &BTRFS_I(inode)->ordered_tree; | |
8158 | ||
8159 | spin_lock_irq(&tree->lock); | |
8160 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
8161 | new_len = page_start - ordered->file_offset; | |
8162 | if (new_len < ordered->truncated_len) | |
8163 | ordered->truncated_len = new_len; | |
8164 | spin_unlock_irq(&tree->lock); | |
8165 | ||
8166 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
8167 | page_start, | |
8168 | PAGE_CACHE_SIZE, 1)) | |
8169 | btrfs_finish_ordered_io(ordered); | |
8b62b72b | 8170 | } |
e6dcd2dc | 8171 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
8172 | if (!inode_evicting) { |
8173 | cached_state = NULL; | |
8174 | lock_extent_bits(tree, page_start, page_end, 0, | |
8175 | &cached_state); | |
8176 | } | |
8177 | } | |
8178 | ||
8179 | if (!inode_evicting) { | |
8180 | clear_extent_bit(tree, page_start, page_end, | |
8181 | EXTENT_LOCKED | EXTENT_DIRTY | | |
8182 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
8183 | EXTENT_DEFRAG, 1, 1, | |
8184 | &cached_state, GFP_NOFS); | |
8185 | ||
8186 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 8187 | } |
e6dcd2dc | 8188 | |
4a096752 | 8189 | ClearPageChecked(page); |
9ad6b7bc | 8190 | if (PagePrivate(page)) { |
9ad6b7bc CM |
8191 | ClearPagePrivate(page); |
8192 | set_page_private(page, 0); | |
8193 | page_cache_release(page); | |
8194 | } | |
39279cc3 CM |
8195 | } |
8196 | ||
9ebefb18 CM |
8197 | /* |
8198 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8199 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8200 | * be careful to check for EOF conditions here. We set the page up correctly | |
8201 | * for a written page which means we get ENOSPC checking when writing into | |
8202 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8203 | * support these features. | |
8204 | * | |
8205 | * We are not allowed to take the i_mutex here so we have to play games to | |
8206 | * protect against truncate races as the page could now be beyond EOF. Because | |
8207 | * vmtruncate() writes the inode size before removing pages, once we have the | |
8208 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
8209 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
8210 | * unlock the page. | |
8211 | */ | |
c2ec175c | 8212 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 8213 | { |
c2ec175c | 8214 | struct page *page = vmf->page; |
496ad9aa | 8215 | struct inode *inode = file_inode(vma->vm_file); |
1832a6d5 | 8216 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
8217 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8218 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8219 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
8220 | char *kaddr; |
8221 | unsigned long zero_start; | |
9ebefb18 | 8222 | loff_t size; |
1832a6d5 | 8223 | int ret; |
9998eb70 | 8224 | int reserved = 0; |
a52d9a80 | 8225 | u64 page_start; |
e6dcd2dc | 8226 | u64 page_end; |
9ebefb18 | 8227 | |
b2b5ef5c | 8228 | sb_start_pagefault(inode->i_sb); |
0ca1f7ce | 8229 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 8230 | if (!ret) { |
e41f941a | 8231 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
8232 | reserved = 1; |
8233 | } | |
56a76f82 NP |
8234 | if (ret) { |
8235 | if (ret == -ENOMEM) | |
8236 | ret = VM_FAULT_OOM; | |
8237 | else /* -ENOSPC, -EIO, etc */ | |
8238 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
8239 | if (reserved) |
8240 | goto out; | |
8241 | goto out_noreserve; | |
56a76f82 | 8242 | } |
1832a6d5 | 8243 | |
56a76f82 | 8244 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8245 | again: |
9ebefb18 | 8246 | lock_page(page); |
9ebefb18 | 8247 | size = i_size_read(inode); |
e6dcd2dc CM |
8248 | page_start = page_offset(page); |
8249 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 8250 | |
9ebefb18 | 8251 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8252 | (page_start >= size)) { |
9ebefb18 CM |
8253 | /* page got truncated out from underneath us */ |
8254 | goto out_unlock; | |
8255 | } | |
e6dcd2dc CM |
8256 | wait_on_page_writeback(page); |
8257 | ||
d0082371 | 8258 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
8259 | set_page_extent_mapped(page); |
8260 | ||
eb84ae03 CM |
8261 | /* |
8262 | * we can't set the delalloc bits if there are pending ordered | |
8263 | * extents. Drop our locks and wait for them to finish | |
8264 | */ | |
e6dcd2dc CM |
8265 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
8266 | if (ordered) { | |
2ac55d41 JB |
8267 | unlock_extent_cached(io_tree, page_start, page_end, |
8268 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 8269 | unlock_page(page); |
eb84ae03 | 8270 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
8271 | btrfs_put_ordered_extent(ordered); |
8272 | goto again; | |
8273 | } | |
8274 | ||
fbf19087 JB |
8275 | /* |
8276 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
8277 | * if it was already dirty, so for space accounting reasons we need to | |
8278 | * clear any delalloc bits for the range we are fixing to save. There | |
8279 | * is probably a better way to do this, but for now keep consistent with | |
8280 | * prepare_pages in the normal write path. | |
8281 | */ | |
2ac55d41 | 8282 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
8283 | EXTENT_DIRTY | EXTENT_DELALLOC | |
8284 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 8285 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 8286 | |
2ac55d41 JB |
8287 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
8288 | &cached_state); | |
9ed74f2d | 8289 | if (ret) { |
2ac55d41 JB |
8290 | unlock_extent_cached(io_tree, page_start, page_end, |
8291 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
8292 | ret = VM_FAULT_SIGBUS; |
8293 | goto out_unlock; | |
8294 | } | |
e6dcd2dc | 8295 | ret = 0; |
9ebefb18 CM |
8296 | |
8297 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 8298 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 8299 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 8300 | else |
e6dcd2dc | 8301 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 8302 | |
e6dcd2dc CM |
8303 | if (zero_start != PAGE_CACHE_SIZE) { |
8304 | kaddr = kmap(page); | |
8305 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
8306 | flush_dcache_page(page); | |
8307 | kunmap(page); | |
8308 | } | |
247e743c | 8309 | ClearPageChecked(page); |
e6dcd2dc | 8310 | set_page_dirty(page); |
50a9b214 | 8311 | SetPageUptodate(page); |
5a3f23d5 | 8312 | |
257c62e1 CM |
8313 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
8314 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 8315 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 8316 | |
2ac55d41 | 8317 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
8318 | |
8319 | out_unlock: | |
b2b5ef5c JK |
8320 | if (!ret) { |
8321 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 8322 | return VM_FAULT_LOCKED; |
b2b5ef5c | 8323 | } |
9ebefb18 | 8324 | unlock_page(page); |
1832a6d5 | 8325 | out: |
ec39e180 | 8326 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 8327 | out_noreserve: |
b2b5ef5c | 8328 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
8329 | return ret; |
8330 | } | |
8331 | ||
a41ad394 | 8332 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
8333 | { |
8334 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 8335 | struct btrfs_block_rsv *rsv; |
a71754fc | 8336 | int ret = 0; |
3893e33b | 8337 | int err = 0; |
39279cc3 | 8338 | struct btrfs_trans_handle *trans; |
dbe674a9 | 8339 | u64 mask = root->sectorsize - 1; |
07127184 | 8340 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 8341 | |
0ef8b726 JB |
8342 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), |
8343 | (u64)-1); | |
8344 | if (ret) | |
8345 | return ret; | |
39279cc3 | 8346 | |
fcb80c2a JB |
8347 | /* |
8348 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
8349 | * 3 things going on here | |
8350 | * | |
8351 | * 1) We need to reserve space for our orphan item and the space to | |
8352 | * delete our orphan item. Lord knows we don't want to have a dangling | |
8353 | * orphan item because we didn't reserve space to remove it. | |
8354 | * | |
8355 | * 2) We need to reserve space to update our inode. | |
8356 | * | |
8357 | * 3) We need to have something to cache all the space that is going to | |
8358 | * be free'd up by the truncate operation, but also have some slack | |
8359 | * space reserved in case it uses space during the truncate (thank you | |
8360 | * very much snapshotting). | |
8361 | * | |
8362 | * And we need these to all be seperate. The fact is we can use alot of | |
8363 | * space doing the truncate, and we have no earthly idea how much space | |
8364 | * we will use, so we need the truncate reservation to be seperate so it | |
8365 | * doesn't end up using space reserved for updating the inode or | |
8366 | * removing the orphan item. We also need to be able to stop the | |
8367 | * transaction and start a new one, which means we need to be able to | |
8368 | * update the inode several times, and we have no idea of knowing how | |
8369 | * many times that will be, so we can't just reserve 1 item for the | |
8370 | * entirety of the opration, so that has to be done seperately as well. | |
8371 | * Then there is the orphan item, which does indeed need to be held on | |
8372 | * to for the whole operation, and we need nobody to touch this reserved | |
8373 | * space except the orphan code. | |
8374 | * | |
8375 | * So that leaves us with | |
8376 | * | |
8377 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
8378 | * 2) rsv - for the truncate reservation, which we will steal from the | |
8379 | * transaction reservation. | |
8380 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
8381 | * updating the inode. | |
8382 | */ | |
66d8f3dd | 8383 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
8384 | if (!rsv) |
8385 | return -ENOMEM; | |
4a338542 | 8386 | rsv->size = min_size; |
ca7e70f5 | 8387 | rsv->failfast = 1; |
f0cd846e | 8388 | |
907cbceb | 8389 | /* |
07127184 | 8390 | * 1 for the truncate slack space |
907cbceb JB |
8391 | * 1 for updating the inode. |
8392 | */ | |
f3fe820c | 8393 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
8394 | if (IS_ERR(trans)) { |
8395 | err = PTR_ERR(trans); | |
8396 | goto out; | |
8397 | } | |
f0cd846e | 8398 | |
907cbceb JB |
8399 | /* Migrate the slack space for the truncate to our reserve */ |
8400 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
8401 | min_size); | |
fcb80c2a | 8402 | BUG_ON(ret); |
f0cd846e | 8403 | |
5dc562c5 JB |
8404 | /* |
8405 | * So if we truncate and then write and fsync we normally would just | |
8406 | * write the extents that changed, which is a problem if we need to | |
8407 | * first truncate that entire inode. So set this flag so we write out | |
8408 | * all of the extents in the inode to the sync log so we're completely | |
8409 | * safe. | |
8410 | */ | |
8411 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 8412 | trans->block_rsv = rsv; |
907cbceb | 8413 | |
8082510e YZ |
8414 | while (1) { |
8415 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
8416 | inode->i_size, | |
8417 | BTRFS_EXTENT_DATA_KEY); | |
ca7e70f5 | 8418 | if (ret != -ENOSPC) { |
3893e33b | 8419 | err = ret; |
8082510e | 8420 | break; |
3893e33b | 8421 | } |
39279cc3 | 8422 | |
fcb80c2a | 8423 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 8424 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
8425 | if (ret) { |
8426 | err = ret; | |
8427 | break; | |
8428 | } | |
ca7e70f5 | 8429 | |
8082510e | 8430 | btrfs_end_transaction(trans, root); |
b53d3f5d | 8431 | btrfs_btree_balance_dirty(root); |
ca7e70f5 JB |
8432 | |
8433 | trans = btrfs_start_transaction(root, 2); | |
8434 | if (IS_ERR(trans)) { | |
8435 | ret = err = PTR_ERR(trans); | |
8436 | trans = NULL; | |
8437 | break; | |
8438 | } | |
8439 | ||
8440 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
8441 | rsv, min_size); | |
8442 | BUG_ON(ret); /* shouldn't happen */ | |
8443 | trans->block_rsv = rsv; | |
8082510e YZ |
8444 | } |
8445 | ||
8446 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 8447 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 8448 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
8449 | if (ret) |
8450 | err = ret; | |
8082510e YZ |
8451 | } |
8452 | ||
917c16b2 CM |
8453 | if (trans) { |
8454 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
8455 | ret = btrfs_update_inode(trans, root, inode); | |
8456 | if (ret && !err) | |
8457 | err = ret; | |
7b128766 | 8458 | |
7ad85bb7 | 8459 | ret = btrfs_end_transaction(trans, root); |
b53d3f5d | 8460 | btrfs_btree_balance_dirty(root); |
917c16b2 | 8461 | } |
fcb80c2a JB |
8462 | |
8463 | out: | |
8464 | btrfs_free_block_rsv(root, rsv); | |
8465 | ||
3893e33b JB |
8466 | if (ret && !err) |
8467 | err = ret; | |
a41ad394 | 8468 | |
3893e33b | 8469 | return err; |
39279cc3 CM |
8470 | } |
8471 | ||
d352ac68 CM |
8472 | /* |
8473 | * create a new subvolume directory/inode (helper for the ioctl). | |
8474 | */ | |
d2fb3437 | 8475 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
8476 | struct btrfs_root *new_root, |
8477 | struct btrfs_root *parent_root, | |
8478 | u64 new_dirid) | |
39279cc3 | 8479 | { |
39279cc3 | 8480 | struct inode *inode; |
76dda93c | 8481 | int err; |
00e4e6b3 | 8482 | u64 index = 0; |
39279cc3 | 8483 | |
12fc9d09 FA |
8484 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
8485 | new_dirid, new_dirid, | |
8486 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
8487 | &index); | |
54aa1f4d | 8488 | if (IS_ERR(inode)) |
f46b5a66 | 8489 | return PTR_ERR(inode); |
39279cc3 CM |
8490 | inode->i_op = &btrfs_dir_inode_operations; |
8491 | inode->i_fop = &btrfs_dir_file_operations; | |
8492 | ||
bfe86848 | 8493 | set_nlink(inode, 1); |
dbe674a9 | 8494 | btrfs_i_size_write(inode, 0); |
b0d5d10f | 8495 | unlock_new_inode(inode); |
3b96362c | 8496 | |
63541927 FDBM |
8497 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
8498 | if (err) | |
8499 | btrfs_err(new_root->fs_info, | |
351fd353 | 8500 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
8501 | new_root->root_key.objectid, err); |
8502 | ||
76dda93c | 8503 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 8504 | |
76dda93c | 8505 | iput(inode); |
ce598979 | 8506 | return err; |
39279cc3 CM |
8507 | } |
8508 | ||
39279cc3 CM |
8509 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
8510 | { | |
8511 | struct btrfs_inode *ei; | |
2ead6ae7 | 8512 | struct inode *inode; |
39279cc3 CM |
8513 | |
8514 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
8515 | if (!ei) | |
8516 | return NULL; | |
2ead6ae7 YZ |
8517 | |
8518 | ei->root = NULL; | |
2ead6ae7 | 8519 | ei->generation = 0; |
15ee9bc7 | 8520 | ei->last_trans = 0; |
257c62e1 | 8521 | ei->last_sub_trans = 0; |
e02119d5 | 8522 | ei->logged_trans = 0; |
2ead6ae7 | 8523 | ei->delalloc_bytes = 0; |
47059d93 | 8524 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
8525 | ei->disk_i_size = 0; |
8526 | ei->flags = 0; | |
7709cde3 | 8527 | ei->csum_bytes = 0; |
2ead6ae7 | 8528 | ei->index_cnt = (u64)-1; |
67de1176 | 8529 | ei->dir_index = 0; |
2ead6ae7 | 8530 | ei->last_unlink_trans = 0; |
46d8bc34 | 8531 | ei->last_log_commit = 0; |
2ead6ae7 | 8532 | |
9e0baf60 JB |
8533 | spin_lock_init(&ei->lock); |
8534 | ei->outstanding_extents = 0; | |
8535 | ei->reserved_extents = 0; | |
2ead6ae7 | 8536 | |
72ac3c0d | 8537 | ei->runtime_flags = 0; |
261507a0 | 8538 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 8539 | |
16cdcec7 MX |
8540 | ei->delayed_node = NULL; |
8541 | ||
2ead6ae7 | 8542 | inode = &ei->vfs_inode; |
a8067e02 | 8543 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
8544 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
8545 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
8546 | ei->io_tree.track_uptodate = 1; |
8547 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 8548 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 8549 | mutex_init(&ei->log_mutex); |
f248679e | 8550 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 8551 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 8552 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
2ead6ae7 YZ |
8553 | RB_CLEAR_NODE(&ei->rb_node); |
8554 | ||
8555 | return inode; | |
39279cc3 CM |
8556 | } |
8557 | ||
aaedb55b JB |
8558 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
8559 | void btrfs_test_destroy_inode(struct inode *inode) | |
8560 | { | |
8561 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); | |
8562 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
8563 | } | |
8564 | #endif | |
8565 | ||
fa0d7e3d NP |
8566 | static void btrfs_i_callback(struct rcu_head *head) |
8567 | { | |
8568 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
8569 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8570 | } | |
8571 | ||
39279cc3 CM |
8572 | void btrfs_destroy_inode(struct inode *inode) |
8573 | { | |
e6dcd2dc | 8574 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
8575 | struct btrfs_root *root = BTRFS_I(inode)->root; |
8576 | ||
b3d9b7a3 | 8577 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 8578 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
8579 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
8580 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
8581 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
8582 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
47059d93 | 8583 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 8584 | |
a6dbd429 JB |
8585 | /* |
8586 | * This can happen where we create an inode, but somebody else also | |
8587 | * created the same inode and we need to destroy the one we already | |
8588 | * created. | |
8589 | */ | |
8590 | if (!root) | |
8591 | goto free; | |
8592 | ||
8a35d95f JB |
8593 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8594 | &BTRFS_I(inode)->runtime_flags)) { | |
c2cf52eb | 8595 | btrfs_info(root->fs_info, "inode %llu still on the orphan list", |
c1c9ff7c | 8596 | btrfs_ino(inode)); |
8a35d95f | 8597 | atomic_dec(&root->orphan_inodes); |
7b128766 | 8598 | } |
7b128766 | 8599 | |
d397712b | 8600 | while (1) { |
e6dcd2dc CM |
8601 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
8602 | if (!ordered) | |
8603 | break; | |
8604 | else { | |
c2cf52eb | 8605 | btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup", |
c1c9ff7c | 8606 | ordered->file_offset, ordered->len); |
e6dcd2dc CM |
8607 | btrfs_remove_ordered_extent(inode, ordered); |
8608 | btrfs_put_ordered_extent(ordered); | |
8609 | btrfs_put_ordered_extent(ordered); | |
8610 | } | |
8611 | } | |
5d4f98a2 | 8612 | inode_tree_del(inode); |
5b21f2ed | 8613 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 8614 | free: |
fa0d7e3d | 8615 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
8616 | } |
8617 | ||
45321ac5 | 8618 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
8619 | { |
8620 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 8621 | |
6379ef9f NA |
8622 | if (root == NULL) |
8623 | return 1; | |
8624 | ||
fa6ac876 | 8625 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 8626 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 8627 | return 1; |
76dda93c | 8628 | else |
45321ac5 | 8629 | return generic_drop_inode(inode); |
76dda93c YZ |
8630 | } |
8631 | ||
0ee0fda0 | 8632 | static void init_once(void *foo) |
39279cc3 CM |
8633 | { |
8634 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
8635 | ||
8636 | inode_init_once(&ei->vfs_inode); | |
8637 | } | |
8638 | ||
8639 | void btrfs_destroy_cachep(void) | |
8640 | { | |
8c0a8537 KS |
8641 | /* |
8642 | * Make sure all delayed rcu free inodes are flushed before we | |
8643 | * destroy cache. | |
8644 | */ | |
8645 | rcu_barrier(); | |
39279cc3 CM |
8646 | if (btrfs_inode_cachep) |
8647 | kmem_cache_destroy(btrfs_inode_cachep); | |
8648 | if (btrfs_trans_handle_cachep) | |
8649 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
8650 | if (btrfs_transaction_cachep) | |
8651 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
8652 | if (btrfs_path_cachep) |
8653 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
8654 | if (btrfs_free_space_cachep) |
8655 | kmem_cache_destroy(btrfs_free_space_cachep); | |
8ccf6f19 MX |
8656 | if (btrfs_delalloc_work_cachep) |
8657 | kmem_cache_destroy(btrfs_delalloc_work_cachep); | |
39279cc3 CM |
8658 | } |
8659 | ||
8660 | int btrfs_init_cachep(void) | |
8661 | { | |
837e1972 | 8662 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 CH |
8663 | sizeof(struct btrfs_inode), 0, |
8664 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
8665 | if (!btrfs_inode_cachep) |
8666 | goto fail; | |
9601e3f6 | 8667 | |
837e1972 | 8668 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 CH |
8669 | sizeof(struct btrfs_trans_handle), 0, |
8670 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
8671 | if (!btrfs_trans_handle_cachep) |
8672 | goto fail; | |
9601e3f6 | 8673 | |
837e1972 | 8674 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 CH |
8675 | sizeof(struct btrfs_transaction), 0, |
8676 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
8677 | if (!btrfs_transaction_cachep) |
8678 | goto fail; | |
9601e3f6 | 8679 | |
837e1972 | 8680 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 CH |
8681 | sizeof(struct btrfs_path), 0, |
8682 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
8683 | if (!btrfs_path_cachep) |
8684 | goto fail; | |
9601e3f6 | 8685 | |
837e1972 | 8686 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 JB |
8687 | sizeof(struct btrfs_free_space), 0, |
8688 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
8689 | if (!btrfs_free_space_cachep) | |
8690 | goto fail; | |
8691 | ||
8ccf6f19 MX |
8692 | btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work", |
8693 | sizeof(struct btrfs_delalloc_work), 0, | |
8694 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
8695 | NULL); | |
8696 | if (!btrfs_delalloc_work_cachep) | |
8697 | goto fail; | |
8698 | ||
39279cc3 CM |
8699 | return 0; |
8700 | fail: | |
8701 | btrfs_destroy_cachep(); | |
8702 | return -ENOMEM; | |
8703 | } | |
8704 | ||
8705 | static int btrfs_getattr(struct vfsmount *mnt, | |
8706 | struct dentry *dentry, struct kstat *stat) | |
8707 | { | |
df0af1a5 | 8708 | u64 delalloc_bytes; |
39279cc3 | 8709 | struct inode *inode = dentry->d_inode; |
fadc0d8b DS |
8710 | u32 blocksize = inode->i_sb->s_blocksize; |
8711 | ||
39279cc3 | 8712 | generic_fillattr(inode, stat); |
0ee5dc67 | 8713 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 8714 | stat->blksize = PAGE_CACHE_SIZE; |
df0af1a5 MX |
8715 | |
8716 | spin_lock(&BTRFS_I(inode)->lock); | |
8717 | delalloc_bytes = BTRFS_I(inode)->delalloc_bytes; | |
8718 | spin_unlock(&BTRFS_I(inode)->lock); | |
fadc0d8b | 8719 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 8720 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
8721 | return 0; |
8722 | } | |
8723 | ||
d397712b CM |
8724 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
8725 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
8726 | { |
8727 | struct btrfs_trans_handle *trans; | |
8728 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 8729 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
8730 | struct inode *new_inode = new_dentry->d_inode; |
8731 | struct inode *old_inode = old_dentry->d_inode; | |
8732 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 8733 | u64 index = 0; |
4df27c4d | 8734 | u64 root_objectid; |
39279cc3 | 8735 | int ret; |
33345d01 | 8736 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 8737 | |
33345d01 | 8738 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
8739 | return -EPERM; |
8740 | ||
4df27c4d | 8741 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 8742 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
8743 | return -EXDEV; |
8744 | ||
33345d01 LZ |
8745 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
8746 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 8747 | return -ENOTEMPTY; |
5f39d397 | 8748 | |
4df27c4d YZ |
8749 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
8750 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
8751 | return -ENOTEMPTY; | |
9c52057c CM |
8752 | |
8753 | ||
8754 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 8755 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
8756 | new_dentry->d_name.name, |
8757 | new_dentry->d_name.len); | |
8758 | ||
8759 | if (ret) { | |
8760 | if (ret == -EEXIST) { | |
8761 | /* we shouldn't get | |
8762 | * eexist without a new_inode */ | |
fae7f21c | 8763 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
8764 | return ret; |
8765 | } | |
8766 | } else { | |
8767 | /* maybe -EOVERFLOW */ | |
8768 | return ret; | |
8769 | } | |
8770 | } | |
8771 | ret = 0; | |
8772 | ||
5a3f23d5 | 8773 | /* |
8d875f95 CM |
8774 | * we're using rename to replace one file with another. Start IO on it |
8775 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 8776 | */ |
8d875f95 | 8777 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
8778 | filemap_flush(old_inode->i_mapping); |
8779 | ||
76dda93c | 8780 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 8781 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 8782 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
8783 | /* |
8784 | * We want to reserve the absolute worst case amount of items. So if | |
8785 | * both inodes are subvols and we need to unlink them then that would | |
8786 | * require 4 item modifications, but if they are both normal inodes it | |
8787 | * would require 5 item modifications, so we'll assume their normal | |
8788 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
8789 | * should cover the worst case number of items we'll modify. | |
8790 | */ | |
6e137ed3 | 8791 | trans = btrfs_start_transaction(root, 11); |
b44c59a8 JL |
8792 | if (IS_ERR(trans)) { |
8793 | ret = PTR_ERR(trans); | |
8794 | goto out_notrans; | |
8795 | } | |
76dda93c | 8796 | |
4df27c4d YZ |
8797 | if (dest != root) |
8798 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 8799 | |
a5719521 YZ |
8800 | ret = btrfs_set_inode_index(new_dir, &index); |
8801 | if (ret) | |
8802 | goto out_fail; | |
5a3f23d5 | 8803 | |
67de1176 | 8804 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 8805 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 8806 | /* force full log commit if subvolume involved. */ |
995946dd | 8807 | btrfs_set_log_full_commit(root->fs_info, trans); |
4df27c4d | 8808 | } else { |
a5719521 YZ |
8809 | ret = btrfs_insert_inode_ref(trans, dest, |
8810 | new_dentry->d_name.name, | |
8811 | new_dentry->d_name.len, | |
33345d01 LZ |
8812 | old_ino, |
8813 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
8814 | if (ret) |
8815 | goto out_fail; | |
4df27c4d YZ |
8816 | /* |
8817 | * this is an ugly little race, but the rename is required | |
8818 | * to make sure that if we crash, the inode is either at the | |
8819 | * old name or the new one. pinning the log transaction lets | |
8820 | * us make sure we don't allow a log commit to come in after | |
8821 | * we unlink the name but before we add the new name back in. | |
8822 | */ | |
8823 | btrfs_pin_log_trans(root); | |
8824 | } | |
5a3f23d5 | 8825 | |
0c4d2d95 JB |
8826 | inode_inc_iversion(old_dir); |
8827 | inode_inc_iversion(new_dir); | |
8828 | inode_inc_iversion(old_inode); | |
39279cc3 CM |
8829 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
8830 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
8831 | old_inode->i_ctime = ctime; | |
5f39d397 | 8832 | |
12fcfd22 CM |
8833 | if (old_dentry->d_parent != new_dentry->d_parent) |
8834 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
8835 | ||
33345d01 | 8836 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
8837 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
8838 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
8839 | old_dentry->d_name.name, | |
8840 | old_dentry->d_name.len); | |
8841 | } else { | |
92986796 AV |
8842 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
8843 | old_dentry->d_inode, | |
8844 | old_dentry->d_name.name, | |
8845 | old_dentry->d_name.len); | |
8846 | if (!ret) | |
8847 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 8848 | } |
79787eaa JM |
8849 | if (ret) { |
8850 | btrfs_abort_transaction(trans, root, ret); | |
8851 | goto out_fail; | |
8852 | } | |
39279cc3 CM |
8853 | |
8854 | if (new_inode) { | |
0c4d2d95 | 8855 | inode_inc_iversion(new_inode); |
39279cc3 | 8856 | new_inode->i_ctime = CURRENT_TIME; |
33345d01 | 8857 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
8858 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
8859 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
8860 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
8861 | root_objectid, | |
8862 | new_dentry->d_name.name, | |
8863 | new_dentry->d_name.len); | |
8864 | BUG_ON(new_inode->i_nlink == 0); | |
8865 | } else { | |
8866 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
8867 | new_dentry->d_inode, | |
8868 | new_dentry->d_name.name, | |
8869 | new_dentry->d_name.len); | |
8870 | } | |
4ef31a45 | 8871 | if (!ret && new_inode->i_nlink == 0) |
e02119d5 | 8872 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
79787eaa JM |
8873 | if (ret) { |
8874 | btrfs_abort_transaction(trans, root, ret); | |
8875 | goto out_fail; | |
8876 | } | |
39279cc3 | 8877 | } |
aec7477b | 8878 | |
4df27c4d YZ |
8879 | ret = btrfs_add_link(trans, new_dir, old_inode, |
8880 | new_dentry->d_name.name, | |
a5719521 | 8881 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
8882 | if (ret) { |
8883 | btrfs_abort_transaction(trans, root, ret); | |
8884 | goto out_fail; | |
8885 | } | |
39279cc3 | 8886 | |
67de1176 MX |
8887 | if (old_inode->i_nlink == 1) |
8888 | BTRFS_I(old_inode)->dir_index = index; | |
8889 | ||
33345d01 | 8890 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 8891 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 8892 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
8893 | btrfs_end_log_trans(root); |
8894 | } | |
39279cc3 | 8895 | out_fail: |
7ad85bb7 | 8896 | btrfs_end_transaction(trans, root); |
b44c59a8 | 8897 | out_notrans: |
33345d01 | 8898 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 8899 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 8900 | |
39279cc3 CM |
8901 | return ret; |
8902 | } | |
8903 | ||
80ace85c MS |
8904 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
8905 | struct inode *new_dir, struct dentry *new_dentry, | |
8906 | unsigned int flags) | |
8907 | { | |
8908 | if (flags & ~RENAME_NOREPLACE) | |
8909 | return -EINVAL; | |
8910 | ||
8911 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry); | |
8912 | } | |
8913 | ||
8ccf6f19 MX |
8914 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
8915 | { | |
8916 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 8917 | struct inode *inode; |
8ccf6f19 MX |
8918 | |
8919 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
8920 | work); | |
9f23e289 JB |
8921 | inode = delalloc_work->inode; |
8922 | if (delalloc_work->wait) { | |
8923 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
8924 | } else { | |
8925 | filemap_flush(inode->i_mapping); | |
8926 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
8927 | &BTRFS_I(inode)->runtime_flags)) | |
8928 | filemap_flush(inode->i_mapping); | |
8929 | } | |
8ccf6f19 MX |
8930 | |
8931 | if (delalloc_work->delay_iput) | |
9f23e289 | 8932 | btrfs_add_delayed_iput(inode); |
8ccf6f19 | 8933 | else |
9f23e289 | 8934 | iput(inode); |
8ccf6f19 MX |
8935 | complete(&delalloc_work->completion); |
8936 | } | |
8937 | ||
8938 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
8939 | int wait, int delay_iput) | |
8940 | { | |
8941 | struct btrfs_delalloc_work *work; | |
8942 | ||
8943 | work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS); | |
8944 | if (!work) | |
8945 | return NULL; | |
8946 | ||
8947 | init_completion(&work->completion); | |
8948 | INIT_LIST_HEAD(&work->list); | |
8949 | work->inode = inode; | |
8950 | work->wait = wait; | |
8951 | work->delay_iput = delay_iput; | |
9e0af237 LB |
8952 | WARN_ON_ONCE(!inode); |
8953 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, | |
8954 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
8955 | |
8956 | return work; | |
8957 | } | |
8958 | ||
8959 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
8960 | { | |
8961 | wait_for_completion(&work->completion); | |
8962 | kmem_cache_free(btrfs_delalloc_work_cachep, work); | |
8963 | } | |
8964 | ||
d352ac68 CM |
8965 | /* |
8966 | * some fairly slow code that needs optimization. This walks the list | |
8967 | * of all the inodes with pending delalloc and forces them to disk. | |
8968 | */ | |
6c255e67 MX |
8969 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput, |
8970 | int nr) | |
ea8c2819 | 8971 | { |
ea8c2819 | 8972 | struct btrfs_inode *binode; |
5b21f2ed | 8973 | struct inode *inode; |
8ccf6f19 MX |
8974 | struct btrfs_delalloc_work *work, *next; |
8975 | struct list_head works; | |
1eafa6c7 | 8976 | struct list_head splice; |
8ccf6f19 | 8977 | int ret = 0; |
ea8c2819 | 8978 | |
8ccf6f19 | 8979 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 8980 | INIT_LIST_HEAD(&splice); |
63607cc8 | 8981 | |
573bfb72 | 8982 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
8983 | spin_lock(&root->delalloc_lock); |
8984 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
8985 | while (!list_empty(&splice)) { |
8986 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 8987 | delalloc_inodes); |
1eafa6c7 | 8988 | |
eb73c1b7 MX |
8989 | list_move_tail(&binode->delalloc_inodes, |
8990 | &root->delalloc_inodes); | |
5b21f2ed | 8991 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 8992 | if (!inode) { |
eb73c1b7 | 8993 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 8994 | continue; |
df0af1a5 | 8995 | } |
eb73c1b7 | 8996 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 MX |
8997 | |
8998 | work = btrfs_alloc_delalloc_work(inode, 0, delay_iput); | |
5d99a998 | 8999 | if (!work) { |
f4ab9ea7 JB |
9000 | if (delay_iput) |
9001 | btrfs_add_delayed_iput(inode); | |
9002 | else | |
9003 | iput(inode); | |
1eafa6c7 | 9004 | ret = -ENOMEM; |
a1ecaabb | 9005 | goto out; |
5b21f2ed | 9006 | } |
1eafa6c7 | 9007 | list_add_tail(&work->list, &works); |
a44903ab QW |
9008 | btrfs_queue_work(root->fs_info->flush_workers, |
9009 | &work->work); | |
6c255e67 MX |
9010 | ret++; |
9011 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 9012 | goto out; |
5b21f2ed | 9013 | cond_resched(); |
eb73c1b7 | 9014 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 9015 | } |
eb73c1b7 | 9016 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 9017 | |
a1ecaabb | 9018 | out: |
eb73c1b7 MX |
9019 | list_for_each_entry_safe(work, next, &works, list) { |
9020 | list_del_init(&work->list); | |
9021 | btrfs_wait_and_free_delalloc_work(work); | |
9022 | } | |
9023 | ||
9024 | if (!list_empty_careful(&splice)) { | |
9025 | spin_lock(&root->delalloc_lock); | |
9026 | list_splice_tail(&splice, &root->delalloc_inodes); | |
9027 | spin_unlock(&root->delalloc_lock); | |
9028 | } | |
573bfb72 | 9029 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
9030 | return ret; |
9031 | } | |
1eafa6c7 | 9032 | |
eb73c1b7 MX |
9033 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
9034 | { | |
9035 | int ret; | |
1eafa6c7 | 9036 | |
2c21b4d7 | 9037 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
eb73c1b7 MX |
9038 | return -EROFS; |
9039 | ||
6c255e67 MX |
9040 | ret = __start_delalloc_inodes(root, delay_iput, -1); |
9041 | if (ret > 0) | |
9042 | ret = 0; | |
eb73c1b7 MX |
9043 | /* |
9044 | * the filemap_flush will queue IO into the worker threads, but | |
8c8bee1d CM |
9045 | * we have to make sure the IO is actually started and that |
9046 | * ordered extents get created before we return | |
9047 | */ | |
9048 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 9049 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 9050 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 9051 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
9052 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
9053 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
9054 | } |
9055 | atomic_dec(&root->fs_info->async_submit_draining); | |
eb73c1b7 MX |
9056 | return ret; |
9057 | } | |
9058 | ||
6c255e67 MX |
9059 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput, |
9060 | int nr) | |
eb73c1b7 MX |
9061 | { |
9062 | struct btrfs_root *root; | |
9063 | struct list_head splice; | |
9064 | int ret; | |
9065 | ||
2c21b4d7 | 9066 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
9067 | return -EROFS; |
9068 | ||
9069 | INIT_LIST_HEAD(&splice); | |
9070 | ||
573bfb72 | 9071 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
9072 | spin_lock(&fs_info->delalloc_root_lock); |
9073 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 9074 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
9075 | root = list_first_entry(&splice, struct btrfs_root, |
9076 | delalloc_root); | |
9077 | root = btrfs_grab_fs_root(root); | |
9078 | BUG_ON(!root); | |
9079 | list_move_tail(&root->delalloc_root, | |
9080 | &fs_info->delalloc_roots); | |
9081 | spin_unlock(&fs_info->delalloc_root_lock); | |
9082 | ||
6c255e67 | 9083 | ret = __start_delalloc_inodes(root, delay_iput, nr); |
eb73c1b7 | 9084 | btrfs_put_fs_root(root); |
6c255e67 | 9085 | if (ret < 0) |
eb73c1b7 MX |
9086 | goto out; |
9087 | ||
6c255e67 MX |
9088 | if (nr != -1) { |
9089 | nr -= ret; | |
9090 | WARN_ON(nr < 0); | |
9091 | } | |
eb73c1b7 | 9092 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 9093 | } |
eb73c1b7 | 9094 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 9095 | |
6c255e67 | 9096 | ret = 0; |
eb73c1b7 MX |
9097 | atomic_inc(&fs_info->async_submit_draining); |
9098 | while (atomic_read(&fs_info->nr_async_submits) || | |
9099 | atomic_read(&fs_info->async_delalloc_pages)) { | |
9100 | wait_event(fs_info->async_submit_wait, | |
9101 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
9102 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
9103 | } | |
9104 | atomic_dec(&fs_info->async_submit_draining); | |
eb73c1b7 | 9105 | out: |
1eafa6c7 | 9106 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
9107 | spin_lock(&fs_info->delalloc_root_lock); |
9108 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
9109 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 9110 | } |
573bfb72 | 9111 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 9112 | return ret; |
ea8c2819 CM |
9113 | } |
9114 | ||
39279cc3 CM |
9115 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
9116 | const char *symname) | |
9117 | { | |
9118 | struct btrfs_trans_handle *trans; | |
9119 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9120 | struct btrfs_path *path; | |
9121 | struct btrfs_key key; | |
1832a6d5 | 9122 | struct inode *inode = NULL; |
39279cc3 CM |
9123 | int err; |
9124 | int drop_inode = 0; | |
9125 | u64 objectid; | |
67871254 | 9126 | u64 index = 0; |
39279cc3 CM |
9127 | int name_len; |
9128 | int datasize; | |
5f39d397 | 9129 | unsigned long ptr; |
39279cc3 | 9130 | struct btrfs_file_extent_item *ei; |
5f39d397 | 9131 | struct extent_buffer *leaf; |
39279cc3 | 9132 | |
f06becc4 | 9133 | name_len = strlen(symname); |
39279cc3 CM |
9134 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) |
9135 | return -ENAMETOOLONG; | |
1832a6d5 | 9136 | |
9ed74f2d JB |
9137 | /* |
9138 | * 2 items for inode item and ref | |
9139 | * 2 items for dir items | |
9140 | * 1 item for xattr if selinux is on | |
9141 | */ | |
a22285a6 YZ |
9142 | trans = btrfs_start_transaction(root, 5); |
9143 | if (IS_ERR(trans)) | |
9144 | return PTR_ERR(trans); | |
1832a6d5 | 9145 | |
581bb050 LZ |
9146 | err = btrfs_find_free_ino(root, &objectid); |
9147 | if (err) | |
9148 | goto out_unlock; | |
9149 | ||
aec7477b | 9150 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 9151 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 9152 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
9153 | if (IS_ERR(inode)) { |
9154 | err = PTR_ERR(inode); | |
39279cc3 | 9155 | goto out_unlock; |
7cf96da3 | 9156 | } |
39279cc3 | 9157 | |
ad19db71 CS |
9158 | /* |
9159 | * If the active LSM wants to access the inode during | |
9160 | * d_instantiate it needs these. Smack checks to see | |
9161 | * if the filesystem supports xattrs by looking at the | |
9162 | * ops vector. | |
9163 | */ | |
9164 | inode->i_fop = &btrfs_file_operations; | |
9165 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f CM |
9166 | inode->i_mapping->a_ops = &btrfs_aops; |
9167 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
9168 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
9169 | ||
9170 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
9171 | if (err) | |
9172 | goto out_unlock_inode; | |
ad19db71 | 9173 | |
a1b075d2 | 9174 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 9175 | if (err) |
b0d5d10f | 9176 | goto out_unlock_inode; |
39279cc3 CM |
9177 | |
9178 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
9179 | if (!path) { |
9180 | err = -ENOMEM; | |
b0d5d10f | 9181 | goto out_unlock_inode; |
d8926bb3 | 9182 | } |
33345d01 | 9183 | key.objectid = btrfs_ino(inode); |
39279cc3 | 9184 | key.offset = 0; |
962a298f | 9185 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
9186 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
9187 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
9188 | datasize); | |
54aa1f4d | 9189 | if (err) { |
b0839166 | 9190 | btrfs_free_path(path); |
b0d5d10f | 9191 | goto out_unlock_inode; |
54aa1f4d | 9192 | } |
5f39d397 CM |
9193 | leaf = path->nodes[0]; |
9194 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
9195 | struct btrfs_file_extent_item); | |
9196 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
9197 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 9198 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
9199 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
9200 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
9201 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
9202 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
9203 | ||
39279cc3 | 9204 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
9205 | write_extent_buffer(leaf, symname, ptr, name_len); |
9206 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 9207 | btrfs_free_path(path); |
5f39d397 | 9208 | |
39279cc3 CM |
9209 | inode->i_op = &btrfs_symlink_inode_operations; |
9210 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 9211 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 9212 | inode_set_bytes(inode, name_len); |
f06becc4 | 9213 | btrfs_i_size_write(inode, name_len); |
54aa1f4d | 9214 | err = btrfs_update_inode(trans, root, inode); |
b0d5d10f | 9215 | if (err) { |
54aa1f4d | 9216 | drop_inode = 1; |
b0d5d10f CM |
9217 | goto out_unlock_inode; |
9218 | } | |
9219 | ||
9220 | unlock_new_inode(inode); | |
9221 | d_instantiate(dentry, inode); | |
39279cc3 CM |
9222 | |
9223 | out_unlock: | |
7ad85bb7 | 9224 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
9225 | if (drop_inode) { |
9226 | inode_dec_link_count(inode); | |
9227 | iput(inode); | |
9228 | } | |
b53d3f5d | 9229 | btrfs_btree_balance_dirty(root); |
39279cc3 | 9230 | return err; |
b0d5d10f CM |
9231 | |
9232 | out_unlock_inode: | |
9233 | drop_inode = 1; | |
9234 | unlock_new_inode(inode); | |
9235 | goto out_unlock; | |
39279cc3 | 9236 | } |
16432985 | 9237 | |
0af3d00b JB |
9238 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
9239 | u64 start, u64 num_bytes, u64 min_size, | |
9240 | loff_t actual_len, u64 *alloc_hint, | |
9241 | struct btrfs_trans_handle *trans) | |
d899e052 | 9242 | { |
5dc562c5 JB |
9243 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
9244 | struct extent_map *em; | |
d899e052 YZ |
9245 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9246 | struct btrfs_key ins; | |
d899e052 | 9247 | u64 cur_offset = start; |
55a61d1d | 9248 | u64 i_size; |
154ea289 | 9249 | u64 cur_bytes; |
d899e052 | 9250 | int ret = 0; |
0af3d00b | 9251 | bool own_trans = true; |
d899e052 | 9252 | |
0af3d00b JB |
9253 | if (trans) |
9254 | own_trans = false; | |
d899e052 | 9255 | while (num_bytes > 0) { |
0af3d00b JB |
9256 | if (own_trans) { |
9257 | trans = btrfs_start_transaction(root, 3); | |
9258 | if (IS_ERR(trans)) { | |
9259 | ret = PTR_ERR(trans); | |
9260 | break; | |
9261 | } | |
5a303d5d YZ |
9262 | } |
9263 | ||
154ea289 CM |
9264 | cur_bytes = min(num_bytes, 256ULL * 1024 * 1024); |
9265 | cur_bytes = max(cur_bytes, min_size); | |
00361589 | 9266 | ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0, |
e570fd27 | 9267 | *alloc_hint, &ins, 1, 0); |
5a303d5d | 9268 | if (ret) { |
0af3d00b JB |
9269 | if (own_trans) |
9270 | btrfs_end_transaction(trans, root); | |
a22285a6 | 9271 | break; |
d899e052 | 9272 | } |
5a303d5d | 9273 | |
d899e052 YZ |
9274 | ret = insert_reserved_file_extent(trans, inode, |
9275 | cur_offset, ins.objectid, | |
9276 | ins.offset, ins.offset, | |
920bbbfb | 9277 | ins.offset, 0, 0, 0, |
d899e052 | 9278 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 9279 | if (ret) { |
857cc2fc | 9280 | btrfs_free_reserved_extent(root, ins.objectid, |
e570fd27 | 9281 | ins.offset, 0); |
79787eaa JM |
9282 | btrfs_abort_transaction(trans, root, ret); |
9283 | if (own_trans) | |
9284 | btrfs_end_transaction(trans, root); | |
9285 | break; | |
9286 | } | |
a1ed835e CM |
9287 | btrfs_drop_extent_cache(inode, cur_offset, |
9288 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 9289 | |
5dc562c5 JB |
9290 | em = alloc_extent_map(); |
9291 | if (!em) { | |
9292 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
9293 | &BTRFS_I(inode)->runtime_flags); | |
9294 | goto next; | |
9295 | } | |
9296 | ||
9297 | em->start = cur_offset; | |
9298 | em->orig_start = cur_offset; | |
9299 | em->len = ins.offset; | |
9300 | em->block_start = ins.objectid; | |
9301 | em->block_len = ins.offset; | |
b4939680 | 9302 | em->orig_block_len = ins.offset; |
cc95bef6 | 9303 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
9304 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
9305 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
9306 | em->generation = trans->transid; | |
9307 | ||
9308 | while (1) { | |
9309 | write_lock(&em_tree->lock); | |
09a2a8f9 | 9310 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
9311 | write_unlock(&em_tree->lock); |
9312 | if (ret != -EEXIST) | |
9313 | break; | |
9314 | btrfs_drop_extent_cache(inode, cur_offset, | |
9315 | cur_offset + ins.offset - 1, | |
9316 | 0); | |
9317 | } | |
9318 | free_extent_map(em); | |
9319 | next: | |
d899e052 YZ |
9320 | num_bytes -= ins.offset; |
9321 | cur_offset += ins.offset; | |
efa56464 | 9322 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 9323 | |
0c4d2d95 | 9324 | inode_inc_iversion(inode); |
d899e052 | 9325 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 9326 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 9327 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
9328 | (actual_len > inode->i_size) && |
9329 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 9330 | if (cur_offset > actual_len) |
55a61d1d | 9331 | i_size = actual_len; |
d1ea6a61 | 9332 | else |
55a61d1d JB |
9333 | i_size = cur_offset; |
9334 | i_size_write(inode, i_size); | |
9335 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
9336 | } |
9337 | ||
d899e052 | 9338 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
9339 | |
9340 | if (ret) { | |
9341 | btrfs_abort_transaction(trans, root, ret); | |
9342 | if (own_trans) | |
9343 | btrfs_end_transaction(trans, root); | |
9344 | break; | |
9345 | } | |
d899e052 | 9346 | |
0af3d00b JB |
9347 | if (own_trans) |
9348 | btrfs_end_transaction(trans, root); | |
5a303d5d | 9349 | } |
d899e052 YZ |
9350 | return ret; |
9351 | } | |
9352 | ||
0af3d00b JB |
9353 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
9354 | u64 start, u64 num_bytes, u64 min_size, | |
9355 | loff_t actual_len, u64 *alloc_hint) | |
9356 | { | |
9357 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9358 | min_size, actual_len, alloc_hint, | |
9359 | NULL); | |
9360 | } | |
9361 | ||
9362 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
9363 | struct btrfs_trans_handle *trans, int mode, | |
9364 | u64 start, u64 num_bytes, u64 min_size, | |
9365 | loff_t actual_len, u64 *alloc_hint) | |
9366 | { | |
9367 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9368 | min_size, actual_len, alloc_hint, trans); | |
9369 | } | |
9370 | ||
e6dcd2dc CM |
9371 | static int btrfs_set_page_dirty(struct page *page) |
9372 | { | |
e6dcd2dc CM |
9373 | return __set_page_dirty_nobuffers(page); |
9374 | } | |
9375 | ||
10556cb2 | 9376 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 9377 | { |
b83cc969 | 9378 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 9379 | umode_t mode = inode->i_mode; |
b83cc969 | 9380 | |
cb6db4e5 JM |
9381 | if (mask & MAY_WRITE && |
9382 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
9383 | if (btrfs_root_readonly(root)) | |
9384 | return -EROFS; | |
9385 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
9386 | return -EACCES; | |
9387 | } | |
2830ba7f | 9388 | return generic_permission(inode, mask); |
fdebe2bd | 9389 | } |
39279cc3 | 9390 | |
ef3b9af5 FM |
9391 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
9392 | { | |
9393 | struct btrfs_trans_handle *trans; | |
9394 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9395 | struct inode *inode = NULL; | |
9396 | u64 objectid; | |
9397 | u64 index; | |
9398 | int ret = 0; | |
9399 | ||
9400 | /* | |
9401 | * 5 units required for adding orphan entry | |
9402 | */ | |
9403 | trans = btrfs_start_transaction(root, 5); | |
9404 | if (IS_ERR(trans)) | |
9405 | return PTR_ERR(trans); | |
9406 | ||
9407 | ret = btrfs_find_free_ino(root, &objectid); | |
9408 | if (ret) | |
9409 | goto out; | |
9410 | ||
9411 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
9412 | btrfs_ino(dir), objectid, mode, &index); | |
9413 | if (IS_ERR(inode)) { | |
9414 | ret = PTR_ERR(inode); | |
9415 | inode = NULL; | |
9416 | goto out; | |
9417 | } | |
9418 | ||
ef3b9af5 FM |
9419 | inode->i_fop = &btrfs_file_operations; |
9420 | inode->i_op = &btrfs_file_inode_operations; | |
9421 | ||
9422 | inode->i_mapping->a_ops = &btrfs_aops; | |
9423 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
9424 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
9425 | ||
b0d5d10f CM |
9426 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
9427 | if (ret) | |
9428 | goto out_inode; | |
9429 | ||
9430 | ret = btrfs_update_inode(trans, root, inode); | |
9431 | if (ret) | |
9432 | goto out_inode; | |
ef3b9af5 FM |
9433 | ret = btrfs_orphan_add(trans, inode); |
9434 | if (ret) | |
b0d5d10f | 9435 | goto out_inode; |
ef3b9af5 | 9436 | |
5762b5c9 FM |
9437 | /* |
9438 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
9439 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
9440 | * through: | |
9441 | * | |
9442 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
9443 | */ | |
9444 | set_nlink(inode, 1); | |
b0d5d10f | 9445 | unlock_new_inode(inode); |
ef3b9af5 FM |
9446 | d_tmpfile(dentry, inode); |
9447 | mark_inode_dirty(inode); | |
9448 | ||
9449 | out: | |
9450 | btrfs_end_transaction(trans, root); | |
9451 | if (ret) | |
9452 | iput(inode); | |
9453 | btrfs_balance_delayed_items(root); | |
9454 | btrfs_btree_balance_dirty(root); | |
ef3b9af5 | 9455 | return ret; |
b0d5d10f CM |
9456 | |
9457 | out_inode: | |
9458 | unlock_new_inode(inode); | |
9459 | goto out; | |
9460 | ||
ef3b9af5 FM |
9461 | } |
9462 | ||
6e1d5dcc | 9463 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 9464 | .getattr = btrfs_getattr, |
39279cc3 CM |
9465 | .lookup = btrfs_lookup, |
9466 | .create = btrfs_create, | |
9467 | .unlink = btrfs_unlink, | |
9468 | .link = btrfs_link, | |
9469 | .mkdir = btrfs_mkdir, | |
9470 | .rmdir = btrfs_rmdir, | |
80ace85c | 9471 | .rename2 = btrfs_rename2, |
39279cc3 CM |
9472 | .symlink = btrfs_symlink, |
9473 | .setattr = btrfs_setattr, | |
618e21d5 | 9474 | .mknod = btrfs_mknod, |
95819c05 CH |
9475 | .setxattr = btrfs_setxattr, |
9476 | .getxattr = btrfs_getxattr, | |
5103e947 | 9477 | .listxattr = btrfs_listxattr, |
95819c05 | 9478 | .removexattr = btrfs_removexattr, |
fdebe2bd | 9479 | .permission = btrfs_permission, |
4e34e719 | 9480 | .get_acl = btrfs_get_acl, |
996a710d | 9481 | .set_acl = btrfs_set_acl, |
93fd63c2 | 9482 | .update_time = btrfs_update_time, |
ef3b9af5 | 9483 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 9484 | }; |
6e1d5dcc | 9485 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 9486 | .lookup = btrfs_lookup, |
fdebe2bd | 9487 | .permission = btrfs_permission, |
4e34e719 | 9488 | .get_acl = btrfs_get_acl, |
996a710d | 9489 | .set_acl = btrfs_set_acl, |
93fd63c2 | 9490 | .update_time = btrfs_update_time, |
39279cc3 | 9491 | }; |
76dda93c | 9492 | |
828c0950 | 9493 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
9494 | .llseek = generic_file_llseek, |
9495 | .read = generic_read_dir, | |
9cdda8d3 | 9496 | .iterate = btrfs_real_readdir, |
34287aa3 | 9497 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 9498 | #ifdef CONFIG_COMPAT |
34287aa3 | 9499 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 9500 | #endif |
6bf13c0c | 9501 | .release = btrfs_release_file, |
e02119d5 | 9502 | .fsync = btrfs_sync_file, |
39279cc3 CM |
9503 | }; |
9504 | ||
d1310b2e | 9505 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 9506 | .fill_delalloc = run_delalloc_range, |
065631f6 | 9507 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 9508 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 9509 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 9510 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 9511 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
9512 | .set_bit_hook = btrfs_set_bit_hook, |
9513 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
9514 | .merge_extent_hook = btrfs_merge_extent_hook, |
9515 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
9516 | }; |
9517 | ||
35054394 CM |
9518 | /* |
9519 | * btrfs doesn't support the bmap operation because swapfiles | |
9520 | * use bmap to make a mapping of extents in the file. They assume | |
9521 | * these extents won't change over the life of the file and they | |
9522 | * use the bmap result to do IO directly to the drive. | |
9523 | * | |
9524 | * the btrfs bmap call would return logical addresses that aren't | |
9525 | * suitable for IO and they also will change frequently as COW | |
9526 | * operations happen. So, swapfile + btrfs == corruption. | |
9527 | * | |
9528 | * For now we're avoiding this by dropping bmap. | |
9529 | */ | |
7f09410b | 9530 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
9531 | .readpage = btrfs_readpage, |
9532 | .writepage = btrfs_writepage, | |
b293f02e | 9533 | .writepages = btrfs_writepages, |
3ab2fb5a | 9534 | .readpages = btrfs_readpages, |
16432985 | 9535 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
9536 | .invalidatepage = btrfs_invalidatepage, |
9537 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 9538 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 9539 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
9540 | }; |
9541 | ||
7f09410b | 9542 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
9543 | .readpage = btrfs_readpage, |
9544 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
9545 | .invalidatepage = btrfs_invalidatepage, |
9546 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
9547 | }; |
9548 | ||
6e1d5dcc | 9549 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
9550 | .getattr = btrfs_getattr, |
9551 | .setattr = btrfs_setattr, | |
95819c05 CH |
9552 | .setxattr = btrfs_setxattr, |
9553 | .getxattr = btrfs_getxattr, | |
5103e947 | 9554 | .listxattr = btrfs_listxattr, |
95819c05 | 9555 | .removexattr = btrfs_removexattr, |
fdebe2bd | 9556 | .permission = btrfs_permission, |
1506fcc8 | 9557 | .fiemap = btrfs_fiemap, |
4e34e719 | 9558 | .get_acl = btrfs_get_acl, |
996a710d | 9559 | .set_acl = btrfs_set_acl, |
e41f941a | 9560 | .update_time = btrfs_update_time, |
39279cc3 | 9561 | }; |
6e1d5dcc | 9562 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
9563 | .getattr = btrfs_getattr, |
9564 | .setattr = btrfs_setattr, | |
fdebe2bd | 9565 | .permission = btrfs_permission, |
95819c05 CH |
9566 | .setxattr = btrfs_setxattr, |
9567 | .getxattr = btrfs_getxattr, | |
33268eaf | 9568 | .listxattr = btrfs_listxattr, |
95819c05 | 9569 | .removexattr = btrfs_removexattr, |
4e34e719 | 9570 | .get_acl = btrfs_get_acl, |
996a710d | 9571 | .set_acl = btrfs_set_acl, |
e41f941a | 9572 | .update_time = btrfs_update_time, |
618e21d5 | 9573 | }; |
6e1d5dcc | 9574 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
9575 | .readlink = generic_readlink, |
9576 | .follow_link = page_follow_link_light, | |
9577 | .put_link = page_put_link, | |
f209561a | 9578 | .getattr = btrfs_getattr, |
22c44fe6 | 9579 | .setattr = btrfs_setattr, |
fdebe2bd | 9580 | .permission = btrfs_permission, |
0279b4cd JO |
9581 | .setxattr = btrfs_setxattr, |
9582 | .getxattr = btrfs_getxattr, | |
9583 | .listxattr = btrfs_listxattr, | |
9584 | .removexattr = btrfs_removexattr, | |
e41f941a | 9585 | .update_time = btrfs_update_time, |
39279cc3 | 9586 | }; |
76dda93c | 9587 | |
82d339d9 | 9588 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 9589 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 9590 | .d_release = btrfs_dentry_release, |
76dda93c | 9591 | }; |