Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 CM |
4 | */ |
5 | ||
7999096f | 6 | #include <crypto/hash.h> |
8f18cf13 | 7 | #include <linux/kernel.h> |
065631f6 | 8 | #include <linux/bio.h> |
f2eb0a24 | 9 | #include <linux/file.h> |
39279cc3 CM |
10 | #include <linux/fs.h> |
11 | #include <linux/pagemap.h> | |
12 | #include <linux/highmem.h> | |
13 | #include <linux/time.h> | |
14 | #include <linux/init.h> | |
15 | #include <linux/string.h> | |
39279cc3 | 16 | #include <linux/backing-dev.h> |
39279cc3 | 17 | #include <linux/writeback.h> |
39279cc3 | 18 | #include <linux/compat.h> |
5103e947 | 19 | #include <linux/xattr.h> |
33268eaf | 20 | #include <linux/posix_acl.h> |
d899e052 | 21 | #include <linux/falloc.h> |
5a0e3ad6 | 22 | #include <linux/slab.h> |
7a36ddec | 23 | #include <linux/ratelimit.h> |
55e301fd | 24 | #include <linux/btrfs.h> |
53b381b3 | 25 | #include <linux/blkdev.h> |
f23b5a59 | 26 | #include <linux/posix_acl_xattr.h> |
e2e40f2c | 27 | #include <linux/uio.h> |
69fe2d75 | 28 | #include <linux/magic.h> |
ae5e165d | 29 | #include <linux/iversion.h> |
ed46ff3d | 30 | #include <linux/swap.h> |
f8e66081 | 31 | #include <linux/migrate.h> |
b1c16ac9 | 32 | #include <linux/sched/mm.h> |
f85781fb | 33 | #include <linux/iomap.h> |
92d32170 | 34 | #include <asm/unaligned.h> |
602cbe91 | 35 | #include "misc.h" |
39279cc3 CM |
36 | #include "ctree.h" |
37 | #include "disk-io.h" | |
38 | #include "transaction.h" | |
39 | #include "btrfs_inode.h" | |
39279cc3 | 40 | #include "print-tree.h" |
e6dcd2dc | 41 | #include "ordered-data.h" |
95819c05 | 42 | #include "xattr.h" |
e02119d5 | 43 | #include "tree-log.h" |
4a54c8c1 | 44 | #include "volumes.h" |
c8b97818 | 45 | #include "compression.h" |
b4ce94de | 46 | #include "locking.h" |
dc89e982 | 47 | #include "free-space-cache.h" |
63541927 | 48 | #include "props.h" |
31193213 | 49 | #include "qgroup.h" |
86736342 | 50 | #include "delalloc-space.h" |
aac0023c | 51 | #include "block-group.h" |
467dc47e | 52 | #include "space-info.h" |
d8e3fb10 | 53 | #include "zoned.h" |
b945a463 | 54 | #include "subpage.h" |
39279cc3 CM |
55 | |
56 | struct btrfs_iget_args { | |
0202e83f | 57 | u64 ino; |
39279cc3 CM |
58 | struct btrfs_root *root; |
59 | }; | |
60 | ||
f28a4928 | 61 | struct btrfs_dio_data { |
f28a4928 | 62 | u64 reserve; |
f85781fb GR |
63 | loff_t length; |
64 | ssize_t submitted; | |
65 | struct extent_changeset *data_reserved; | |
f28a4928 FM |
66 | }; |
67 | ||
6e1d5dcc AD |
68 | static const struct inode_operations btrfs_dir_inode_operations; |
69 | static const struct inode_operations btrfs_symlink_inode_operations; | |
6e1d5dcc AD |
70 | static const struct inode_operations btrfs_special_inode_operations; |
71 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b | 72 | static const struct address_space_operations btrfs_aops; |
828c0950 | 73 | static const struct file_operations btrfs_dir_file_operations; |
39279cc3 CM |
74 | |
75 | static struct kmem_cache *btrfs_inode_cachep; | |
76 | struct kmem_cache *btrfs_trans_handle_cachep; | |
39279cc3 | 77 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 78 | struct kmem_cache *btrfs_free_space_cachep; |
3acd4850 | 79 | struct kmem_cache *btrfs_free_space_bitmap_cachep; |
39279cc3 | 80 | |
3972f260 | 81 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
213e8c55 | 82 | static int btrfs_truncate(struct inode *inode, bool skip_writeback); |
5fd02043 | 83 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
6e26c442 | 84 | static noinline int cow_file_range(struct btrfs_inode *inode, |
771ed689 | 85 | struct page *locked_page, |
74e9194a | 86 | u64 start, u64 end, int *page_started, |
330a5827 | 87 | unsigned long *nr_written, int unlock); |
4b67c11d NB |
88 | static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start, |
89 | u64 len, u64 orig_start, u64 block_start, | |
6f9994db LB |
90 | u64 block_len, u64 orig_block_len, |
91 | u64 ram_bytes, int compress_type, | |
92 | int type); | |
7b128766 | 93 | |
b672b5c1 | 94 | static void __endio_write_update_ordered(struct btrfs_inode *inode, |
52427260 QW |
95 | const u64 offset, const u64 bytes, |
96 | const bool uptodate); | |
97 | ||
a14b78ad GR |
98 | /* |
99 | * btrfs_inode_lock - lock inode i_rwsem based on arguments passed | |
100 | * | |
101 | * ilock_flags can have the following bit set: | |
102 | * | |
103 | * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode | |
104 | * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt | |
105 | * return -EAGAIN | |
8318ba79 | 106 | * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock |
a14b78ad GR |
107 | */ |
108 | int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags) | |
109 | { | |
110 | if (ilock_flags & BTRFS_ILOCK_SHARED) { | |
111 | if (ilock_flags & BTRFS_ILOCK_TRY) { | |
112 | if (!inode_trylock_shared(inode)) | |
113 | return -EAGAIN; | |
114 | else | |
115 | return 0; | |
116 | } | |
117 | inode_lock_shared(inode); | |
118 | } else { | |
119 | if (ilock_flags & BTRFS_ILOCK_TRY) { | |
120 | if (!inode_trylock(inode)) | |
121 | return -EAGAIN; | |
122 | else | |
123 | return 0; | |
124 | } | |
125 | inode_lock(inode); | |
126 | } | |
8318ba79 JB |
127 | if (ilock_flags & BTRFS_ILOCK_MMAP) |
128 | down_write(&BTRFS_I(inode)->i_mmap_lock); | |
a14b78ad GR |
129 | return 0; |
130 | } | |
131 | ||
132 | /* | |
133 | * btrfs_inode_unlock - unock inode i_rwsem | |
134 | * | |
135 | * ilock_flags should contain the same bits set as passed to btrfs_inode_lock() | |
136 | * to decide whether the lock acquired is shared or exclusive. | |
137 | */ | |
138 | void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags) | |
139 | { | |
8318ba79 JB |
140 | if (ilock_flags & BTRFS_ILOCK_MMAP) |
141 | up_write(&BTRFS_I(inode)->i_mmap_lock); | |
a14b78ad GR |
142 | if (ilock_flags & BTRFS_ILOCK_SHARED) |
143 | inode_unlock_shared(inode); | |
144 | else | |
145 | inode_unlock(inode); | |
146 | } | |
147 | ||
52427260 QW |
148 | /* |
149 | * Cleanup all submitted ordered extents in specified range to handle errors | |
52042d8e | 150 | * from the btrfs_run_delalloc_range() callback. |
52427260 QW |
151 | * |
152 | * NOTE: caller must ensure that when an error happens, it can not call | |
153 | * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING | |
154 | * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata | |
155 | * to be released, which we want to happen only when finishing the ordered | |
d1051d6e | 156 | * extent (btrfs_finish_ordered_io()). |
52427260 | 157 | */ |
64e1db56 | 158 | static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode, |
d1051d6e NB |
159 | struct page *locked_page, |
160 | u64 offset, u64 bytes) | |
52427260 | 161 | { |
63d71450 NA |
162 | unsigned long index = offset >> PAGE_SHIFT; |
163 | unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT; | |
d1051d6e NB |
164 | u64 page_start = page_offset(locked_page); |
165 | u64 page_end = page_start + PAGE_SIZE - 1; | |
166 | ||
63d71450 NA |
167 | struct page *page; |
168 | ||
169 | while (index <= end_index) { | |
968f2566 QW |
170 | /* |
171 | * For locked page, we will call end_extent_writepage() on it | |
172 | * in run_delalloc_range() for the error handling. That | |
173 | * end_extent_writepage() function will call | |
174 | * btrfs_mark_ordered_io_finished() to clear page Ordered and | |
175 | * run the ordered extent accounting. | |
176 | * | |
177 | * Here we can't just clear the Ordered bit, or | |
178 | * btrfs_mark_ordered_io_finished() would skip the accounting | |
179 | * for the page range, and the ordered extent will never finish. | |
180 | */ | |
181 | if (index == (page_offset(locked_page) >> PAGE_SHIFT)) { | |
182 | index++; | |
183 | continue; | |
184 | } | |
64e1db56 | 185 | page = find_get_page(inode->vfs_inode.i_mapping, index); |
63d71450 NA |
186 | index++; |
187 | if (!page) | |
188 | continue; | |
968f2566 QW |
189 | |
190 | /* | |
191 | * Here we just clear all Ordered bits for every page in the | |
192 | * range, then __endio_write_update_ordered() will handle | |
193 | * the ordered extent accounting for the range. | |
194 | */ | |
b945a463 QW |
195 | btrfs_page_clamp_clear_ordered(inode->root->fs_info, page, |
196 | offset, bytes); | |
63d71450 NA |
197 | put_page(page); |
198 | } | |
d1051d6e | 199 | |
b945a463 QW |
200 | /* The locked page covers the full range, nothing needs to be done */ |
201 | if (bytes + offset <= page_offset(locked_page) + PAGE_SIZE) | |
202 | return; | |
d1051d6e NB |
203 | /* |
204 | * In case this page belongs to the delalloc range being instantiated | |
205 | * then skip it, since the first page of a range is going to be | |
206 | * properly cleaned up by the caller of run_delalloc_range | |
207 | */ | |
208 | if (page_start >= offset && page_end <= (offset + bytes - 1)) { | |
b945a463 QW |
209 | bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE; |
210 | offset = page_offset(locked_page) + PAGE_SIZE; | |
d1051d6e NB |
211 | } |
212 | ||
64e1db56 | 213 | return __endio_write_update_ordered(inode, offset, bytes, false); |
52427260 QW |
214 | } |
215 | ||
48a3b636 | 216 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 217 | |
f34f57a3 | 218 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
219 | struct inode *inode, struct inode *dir, |
220 | const struct qstr *qstr) | |
0279b4cd JO |
221 | { |
222 | int err; | |
223 | ||
f34f57a3 | 224 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 225 | if (!err) |
2a7dba39 | 226 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
227 | return err; |
228 | } | |
229 | ||
c8b97818 CM |
230 | /* |
231 | * this does all the hard work for inserting an inline extent into | |
232 | * the btree. The caller should have done a btrfs_drop_extents so that | |
233 | * no overlapping inline items exist in the btree | |
234 | */ | |
40f76580 | 235 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
5893dfb9 | 236 | struct btrfs_path *path, bool extent_inserted, |
c8b97818 CM |
237 | struct btrfs_root *root, struct inode *inode, |
238 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 239 | int compress_type, |
c8b97818 CM |
240 | struct page **compressed_pages) |
241 | { | |
c8b97818 CM |
242 | struct extent_buffer *leaf; |
243 | struct page *page = NULL; | |
244 | char *kaddr; | |
245 | unsigned long ptr; | |
246 | struct btrfs_file_extent_item *ei; | |
c8b97818 CM |
247 | int ret; |
248 | size_t cur_size = size; | |
c8b97818 | 249 | unsigned long offset; |
c8b97818 | 250 | |
982f1f5d JJB |
251 | ASSERT((compressed_size > 0 && compressed_pages) || |
252 | (compressed_size == 0 && !compressed_pages)); | |
253 | ||
fe3f566c | 254 | if (compressed_size && compressed_pages) |
c8b97818 | 255 | cur_size = compressed_size; |
c8b97818 | 256 | |
1acae57b FDBM |
257 | if (!extent_inserted) { |
258 | struct btrfs_key key; | |
259 | size_t datasize; | |
c8b97818 | 260 | |
4a0cc7ca | 261 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b | 262 | key.offset = start; |
962a298f | 263 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 264 | |
1acae57b | 265 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
1acae57b FDBM |
266 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
267 | datasize); | |
79b4f4c6 | 268 | if (ret) |
1acae57b | 269 | goto fail; |
c8b97818 CM |
270 | } |
271 | leaf = path->nodes[0]; | |
272 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
273 | struct btrfs_file_extent_item); | |
274 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
275 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
276 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
277 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
278 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
279 | ptr = btrfs_file_extent_inline_start(ei); | |
280 | ||
261507a0 | 281 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
282 | struct page *cpage; |
283 | int i = 0; | |
d397712b | 284 | while (compressed_size > 0) { |
c8b97818 | 285 | cpage = compressed_pages[i]; |
5b050f04 | 286 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 287 | PAGE_SIZE); |
c8b97818 | 288 | |
4c2bf276 | 289 | kaddr = page_address(cpage); |
c8b97818 | 290 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
c8b97818 CM |
291 | |
292 | i++; | |
293 | ptr += cur_size; | |
294 | compressed_size -= cur_size; | |
295 | } | |
296 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 297 | compress_type); |
c8b97818 CM |
298 | } else { |
299 | page = find_get_page(inode->i_mapping, | |
09cbfeaf | 300 | start >> PAGE_SHIFT); |
c8b97818 | 301 | btrfs_set_file_extent_compression(leaf, ei, 0); |
7ac687d9 | 302 | kaddr = kmap_atomic(page); |
7073017a | 303 | offset = offset_in_page(start); |
c8b97818 | 304 | write_extent_buffer(leaf, kaddr + offset, ptr, size); |
7ac687d9 | 305 | kunmap_atomic(kaddr); |
09cbfeaf | 306 | put_page(page); |
c8b97818 CM |
307 | } |
308 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 309 | btrfs_release_path(path); |
c8b97818 | 310 | |
9ddc959e JB |
311 | /* |
312 | * We align size to sectorsize for inline extents just for simplicity | |
313 | * sake. | |
314 | */ | |
315 | size = ALIGN(size, root->fs_info->sectorsize); | |
316 | ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size); | |
317 | if (ret) | |
318 | goto fail; | |
319 | ||
c2167754 YZ |
320 | /* |
321 | * we're an inline extent, so nobody can | |
322 | * extend the file past i_size without locking | |
323 | * a page we already have locked. | |
324 | * | |
325 | * We must do any isize and inode updates | |
326 | * before we unlock the pages. Otherwise we | |
327 | * could end up racing with unlink. | |
328 | */ | |
c8b97818 | 329 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
c8b97818 | 330 | fail: |
79b4f4c6 | 331 | return ret; |
c8b97818 CM |
332 | } |
333 | ||
334 | ||
335 | /* | |
336 | * conditionally insert an inline extent into the file. This | |
337 | * does the checks required to make sure the data is small enough | |
338 | * to fit as an inline extent. | |
339 | */ | |
a0349401 | 340 | static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start, |
00361589 JB |
341 | u64 end, size_t compressed_size, |
342 | int compress_type, | |
343 | struct page **compressed_pages) | |
c8b97818 | 344 | { |
5893dfb9 | 345 | struct btrfs_drop_extents_args drop_args = { 0 }; |
a0349401 | 346 | struct btrfs_root *root = inode->root; |
0b246afa | 347 | struct btrfs_fs_info *fs_info = root->fs_info; |
00361589 | 348 | struct btrfs_trans_handle *trans; |
a0349401 | 349 | u64 isize = i_size_read(&inode->vfs_inode); |
c8b97818 CM |
350 | u64 actual_end = min(end + 1, isize); |
351 | u64 inline_len = actual_end - start; | |
0b246afa | 352 | u64 aligned_end = ALIGN(end, fs_info->sectorsize); |
c8b97818 CM |
353 | u64 data_len = inline_len; |
354 | int ret; | |
1acae57b | 355 | struct btrfs_path *path; |
c8b97818 CM |
356 | |
357 | if (compressed_size) | |
358 | data_len = compressed_size; | |
359 | ||
360 | if (start > 0 || | |
0b246afa JM |
361 | actual_end > fs_info->sectorsize || |
362 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) || | |
c8b97818 | 363 | (!compressed_size && |
0b246afa | 364 | (actual_end & (fs_info->sectorsize - 1)) == 0) || |
c8b97818 | 365 | end + 1 < isize || |
0b246afa | 366 | data_len > fs_info->max_inline) { |
c8b97818 CM |
367 | return 1; |
368 | } | |
369 | ||
1acae57b FDBM |
370 | path = btrfs_alloc_path(); |
371 | if (!path) | |
372 | return -ENOMEM; | |
373 | ||
00361589 | 374 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
375 | if (IS_ERR(trans)) { |
376 | btrfs_free_path(path); | |
00361589 | 377 | return PTR_ERR(trans); |
1acae57b | 378 | } |
a0349401 | 379 | trans->block_rsv = &inode->block_rsv; |
00361589 | 380 | |
5893dfb9 FM |
381 | drop_args.path = path; |
382 | drop_args.start = start; | |
383 | drop_args.end = aligned_end; | |
384 | drop_args.drop_cache = true; | |
385 | drop_args.replace_extent = true; | |
386 | ||
1acae57b | 387 | if (compressed_size && compressed_pages) |
5893dfb9 | 388 | drop_args.extent_item_size = btrfs_file_extent_calc_inline_size( |
1acae57b FDBM |
389 | compressed_size); |
390 | else | |
5893dfb9 | 391 | drop_args.extent_item_size = btrfs_file_extent_calc_inline_size( |
1acae57b FDBM |
392 | inline_len); |
393 | ||
5893dfb9 | 394 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
00361589 | 395 | if (ret) { |
66642832 | 396 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
397 | goto out; |
398 | } | |
c8b97818 CM |
399 | |
400 | if (isize > actual_end) | |
401 | inline_len = min_t(u64, isize, actual_end); | |
5893dfb9 | 402 | ret = insert_inline_extent(trans, path, drop_args.extent_inserted, |
a0349401 | 403 | root, &inode->vfs_inode, start, |
c8b97818 | 404 | inline_len, compressed_size, |
fe3f566c | 405 | compress_type, compressed_pages); |
2adcac1a | 406 | if (ret && ret != -ENOSPC) { |
66642832 | 407 | btrfs_abort_transaction(trans, ret); |
00361589 | 408 | goto out; |
2adcac1a | 409 | } else if (ret == -ENOSPC) { |
00361589 JB |
410 | ret = 1; |
411 | goto out; | |
79787eaa | 412 | } |
2adcac1a | 413 | |
2766ff61 | 414 | btrfs_update_inode_bytes(inode, inline_len, drop_args.bytes_found); |
9a56fcd1 | 415 | ret = btrfs_update_inode(trans, root, inode); |
2766ff61 FM |
416 | if (ret && ret != -ENOSPC) { |
417 | btrfs_abort_transaction(trans, ret); | |
418 | goto out; | |
419 | } else if (ret == -ENOSPC) { | |
420 | ret = 1; | |
421 | goto out; | |
422 | } | |
423 | ||
a0349401 | 424 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags); |
00361589 | 425 | out: |
94ed938a QW |
426 | /* |
427 | * Don't forget to free the reserved space, as for inlined extent | |
428 | * it won't count as data extent, free them directly here. | |
429 | * And at reserve time, it's always aligned to page size, so | |
430 | * just free one page here. | |
431 | */ | |
a0349401 | 432 | btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE); |
1acae57b | 433 | btrfs_free_path(path); |
3a45bb20 | 434 | btrfs_end_transaction(trans); |
00361589 | 435 | return ret; |
c8b97818 CM |
436 | } |
437 | ||
771ed689 CM |
438 | struct async_extent { |
439 | u64 start; | |
440 | u64 ram_size; | |
441 | u64 compressed_size; | |
442 | struct page **pages; | |
443 | unsigned long nr_pages; | |
261507a0 | 444 | int compress_type; |
771ed689 CM |
445 | struct list_head list; |
446 | }; | |
447 | ||
97db1204 | 448 | struct async_chunk { |
771ed689 | 449 | struct inode *inode; |
771ed689 CM |
450 | struct page *locked_page; |
451 | u64 start; | |
452 | u64 end; | |
f82b7359 | 453 | unsigned int write_flags; |
771ed689 | 454 | struct list_head extents; |
ec39f769 | 455 | struct cgroup_subsys_state *blkcg_css; |
771ed689 | 456 | struct btrfs_work work; |
97db1204 | 457 | atomic_t *pending; |
771ed689 CM |
458 | }; |
459 | ||
97db1204 NB |
460 | struct async_cow { |
461 | /* Number of chunks in flight; must be first in the structure */ | |
462 | atomic_t num_chunks; | |
463 | struct async_chunk chunks[]; | |
771ed689 CM |
464 | }; |
465 | ||
97db1204 | 466 | static noinline int add_async_extent(struct async_chunk *cow, |
771ed689 CM |
467 | u64 start, u64 ram_size, |
468 | u64 compressed_size, | |
469 | struct page **pages, | |
261507a0 LZ |
470 | unsigned long nr_pages, |
471 | int compress_type) | |
771ed689 CM |
472 | { |
473 | struct async_extent *async_extent; | |
474 | ||
475 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 476 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
477 | async_extent->start = start; |
478 | async_extent->ram_size = ram_size; | |
479 | async_extent->compressed_size = compressed_size; | |
480 | async_extent->pages = pages; | |
481 | async_extent->nr_pages = nr_pages; | |
261507a0 | 482 | async_extent->compress_type = compress_type; |
771ed689 CM |
483 | list_add_tail(&async_extent->list, &cow->extents); |
484 | return 0; | |
485 | } | |
486 | ||
42c16da6 QW |
487 | /* |
488 | * Check if the inode has flags compatible with compression | |
489 | */ | |
99c88dc7 | 490 | static inline bool inode_can_compress(struct btrfs_inode *inode) |
42c16da6 | 491 | { |
99c88dc7 NB |
492 | if (inode->flags & BTRFS_INODE_NODATACOW || |
493 | inode->flags & BTRFS_INODE_NODATASUM) | |
42c16da6 QW |
494 | return false; |
495 | return true; | |
496 | } | |
497 | ||
498 | /* | |
499 | * Check if the inode needs to be submitted to compression, based on mount | |
500 | * options, defragmentation, properties or heuristics. | |
501 | */ | |
808a1292 NB |
502 | static inline int inode_need_compress(struct btrfs_inode *inode, u64 start, |
503 | u64 end) | |
f79707b0 | 504 | { |
808a1292 | 505 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
f79707b0 | 506 | |
808a1292 | 507 | if (!inode_can_compress(inode)) { |
42c16da6 QW |
508 | WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), |
509 | KERN_ERR "BTRFS: unexpected compression for ino %llu\n", | |
808a1292 | 510 | btrfs_ino(inode)); |
42c16da6 QW |
511 | return 0; |
512 | } | |
f79707b0 | 513 | /* force compress */ |
0b246afa | 514 | if (btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
f79707b0 | 515 | return 1; |
eec63c65 | 516 | /* defrag ioctl */ |
808a1292 | 517 | if (inode->defrag_compress) |
eec63c65 | 518 | return 1; |
f79707b0 | 519 | /* bad compression ratios */ |
808a1292 | 520 | if (inode->flags & BTRFS_INODE_NOCOMPRESS) |
f79707b0 | 521 | return 0; |
0b246afa | 522 | if (btrfs_test_opt(fs_info, COMPRESS) || |
808a1292 NB |
523 | inode->flags & BTRFS_INODE_COMPRESS || |
524 | inode->prop_compress) | |
525 | return btrfs_compress_heuristic(&inode->vfs_inode, start, end); | |
f79707b0 WS |
526 | return 0; |
527 | } | |
528 | ||
6158e1ce | 529 | static inline void inode_should_defrag(struct btrfs_inode *inode, |
26d30f85 AJ |
530 | u64 start, u64 end, u64 num_bytes, u64 small_write) |
531 | { | |
532 | /* If this is a small write inside eof, kick off a defrag */ | |
533 | if (num_bytes < small_write && | |
6158e1ce | 534 | (start > 0 || end + 1 < inode->disk_i_size)) |
26d30f85 AJ |
535 | btrfs_add_inode_defrag(NULL, inode); |
536 | } | |
537 | ||
d352ac68 | 538 | /* |
771ed689 CM |
539 | * we create compressed extents in two phases. The first |
540 | * phase compresses a range of pages that have already been | |
541 | * locked (both pages and state bits are locked). | |
c8b97818 | 542 | * |
771ed689 CM |
543 | * This is done inside an ordered work queue, and the compression |
544 | * is spread across many cpus. The actual IO submission is step | |
545 | * two, and the ordered work queue takes care of making sure that | |
546 | * happens in the same order things were put onto the queue by | |
547 | * writepages and friends. | |
c8b97818 | 548 | * |
771ed689 CM |
549 | * If this code finds it can't get good compression, it puts an |
550 | * entry onto the work queue to write the uncompressed bytes. This | |
551 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
552 | * are written in the same order that the flusher thread sent them |
553 | * down. | |
d352ac68 | 554 | */ |
ac3e9933 | 555 | static noinline int compress_file_range(struct async_chunk *async_chunk) |
b888db2b | 556 | { |
1368c6da | 557 | struct inode *inode = async_chunk->inode; |
0b246afa | 558 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
0b246afa | 559 | u64 blocksize = fs_info->sectorsize; |
1368c6da NB |
560 | u64 start = async_chunk->start; |
561 | u64 end = async_chunk->end; | |
c8b97818 | 562 | u64 actual_end; |
d98da499 | 563 | u64 i_size; |
e6dcd2dc | 564 | int ret = 0; |
c8b97818 CM |
565 | struct page **pages = NULL; |
566 | unsigned long nr_pages; | |
c8b97818 CM |
567 | unsigned long total_compressed = 0; |
568 | unsigned long total_in = 0; | |
c8b97818 CM |
569 | int i; |
570 | int will_compress; | |
0b246afa | 571 | int compress_type = fs_info->compress_type; |
ac3e9933 | 572 | int compressed_extents = 0; |
4adaa611 | 573 | int redirty = 0; |
b888db2b | 574 | |
6158e1ce NB |
575 | inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1, |
576 | SZ_16K); | |
4cb5300b | 577 | |
d98da499 JB |
578 | /* |
579 | * We need to save i_size before now because it could change in between | |
580 | * us evaluating the size and assigning it. This is because we lock and | |
581 | * unlock the page in truncate and fallocate, and then modify the i_size | |
582 | * later on. | |
583 | * | |
584 | * The barriers are to emulate READ_ONCE, remove that once i_size_read | |
585 | * does that for us. | |
586 | */ | |
587 | barrier(); | |
588 | i_size = i_size_read(inode); | |
589 | barrier(); | |
590 | actual_end = min_t(u64, i_size, end + 1); | |
c8b97818 CM |
591 | again: |
592 | will_compress = 0; | |
09cbfeaf | 593 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
069eac78 DS |
594 | BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0); |
595 | nr_pages = min_t(unsigned long, nr_pages, | |
596 | BTRFS_MAX_COMPRESSED / PAGE_SIZE); | |
be20aa9d | 597 | |
f03d9301 CM |
598 | /* |
599 | * we don't want to send crud past the end of i_size through | |
600 | * compression, that's just a waste of CPU time. So, if the | |
601 | * end of the file is before the start of our current | |
602 | * requested range of bytes, we bail out to the uncompressed | |
603 | * cleanup code that can deal with all of this. | |
604 | * | |
605 | * It isn't really the fastest way to fix things, but this is a | |
606 | * very uncommon corner. | |
607 | */ | |
608 | if (actual_end <= start) | |
609 | goto cleanup_and_bail_uncompressed; | |
610 | ||
c8b97818 CM |
611 | total_compressed = actual_end - start; |
612 | ||
4bcbb332 SW |
613 | /* |
614 | * skip compression for a small file range(<=blocksize) that | |
01327610 | 615 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
616 | */ |
617 | if (total_compressed <= blocksize && | |
618 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
619 | goto cleanup_and_bail_uncompressed; | |
620 | ||
069eac78 DS |
621 | total_compressed = min_t(unsigned long, total_compressed, |
622 | BTRFS_MAX_UNCOMPRESSED); | |
c8b97818 CM |
623 | total_in = 0; |
624 | ret = 0; | |
db94535d | 625 | |
771ed689 CM |
626 | /* |
627 | * we do compression for mount -o compress and when the | |
628 | * inode has not been flagged as nocompress. This flag can | |
629 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 630 | */ |
f2165627 | 631 | if (nr_pages > 1 && inode_need_compress(BTRFS_I(inode), start, end)) { |
c8b97818 | 632 | WARN_ON(pages); |
31e818fe | 633 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
634 | if (!pages) { |
635 | /* just bail out to the uncompressed code */ | |
3527a018 | 636 | nr_pages = 0; |
560f7d75 LZ |
637 | goto cont; |
638 | } | |
c8b97818 | 639 | |
eec63c65 DS |
640 | if (BTRFS_I(inode)->defrag_compress) |
641 | compress_type = BTRFS_I(inode)->defrag_compress; | |
642 | else if (BTRFS_I(inode)->prop_compress) | |
b52aa8c9 | 643 | compress_type = BTRFS_I(inode)->prop_compress; |
261507a0 | 644 | |
4adaa611 CM |
645 | /* |
646 | * we need to call clear_page_dirty_for_io on each | |
647 | * page in the range. Otherwise applications with the file | |
648 | * mmap'd can wander in and change the page contents while | |
649 | * we are compressing them. | |
650 | * | |
651 | * If the compression fails for any reason, we set the pages | |
652 | * dirty again later on. | |
e9679de3 TT |
653 | * |
654 | * Note that the remaining part is redirtied, the start pointer | |
655 | * has moved, the end is the original one. | |
4adaa611 | 656 | */ |
e9679de3 TT |
657 | if (!redirty) { |
658 | extent_range_clear_dirty_for_io(inode, start, end); | |
659 | redirty = 1; | |
660 | } | |
f51d2b59 DS |
661 | |
662 | /* Compression level is applied here and only here */ | |
663 | ret = btrfs_compress_pages( | |
664 | compress_type | (fs_info->compress_level << 4), | |
261507a0 | 665 | inode->i_mapping, start, |
38c31464 | 666 | pages, |
4d3a800e | 667 | &nr_pages, |
261507a0 | 668 | &total_in, |
e5d74902 | 669 | &total_compressed); |
c8b97818 CM |
670 | |
671 | if (!ret) { | |
7073017a | 672 | unsigned long offset = offset_in_page(total_compressed); |
4d3a800e | 673 | struct page *page = pages[nr_pages - 1]; |
c8b97818 CM |
674 | |
675 | /* zero the tail end of the last page, we might be | |
676 | * sending it down to disk | |
677 | */ | |
d048b9c2 IW |
678 | if (offset) |
679 | memzero_page(page, offset, PAGE_SIZE - offset); | |
c8b97818 CM |
680 | will_compress = 1; |
681 | } | |
682 | } | |
560f7d75 | 683 | cont: |
7367253a QW |
684 | /* |
685 | * Check cow_file_range() for why we don't even try to create inline | |
686 | * extent for subpage case. | |
687 | */ | |
688 | if (start == 0 && fs_info->sectorsize == PAGE_SIZE) { | |
c8b97818 | 689 | /* lets try to make an inline extent */ |
6018ba0a | 690 | if (ret || total_in < actual_end) { |
c8b97818 | 691 | /* we didn't compress the entire range, try |
771ed689 | 692 | * to make an uncompressed inline extent. |
c8b97818 | 693 | */ |
a0349401 NB |
694 | ret = cow_file_range_inline(BTRFS_I(inode), start, end, |
695 | 0, BTRFS_COMPRESS_NONE, | |
696 | NULL); | |
c8b97818 | 697 | } else { |
771ed689 | 698 | /* try making a compressed inline extent */ |
a0349401 | 699 | ret = cow_file_range_inline(BTRFS_I(inode), start, end, |
fe3f566c LZ |
700 | total_compressed, |
701 | compress_type, pages); | |
c8b97818 | 702 | } |
79787eaa | 703 | if (ret <= 0) { |
151a41bc | 704 | unsigned long clear_flags = EXTENT_DELALLOC | |
8b62f87b JB |
705 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
706 | EXTENT_DO_ACCOUNTING; | |
e6eb4314 FM |
707 | unsigned long page_error_op; |
708 | ||
e6eb4314 | 709 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 710 | |
771ed689 | 711 | /* |
79787eaa JM |
712 | * inline extent creation worked or returned error, |
713 | * we don't need to create any more async work items. | |
714 | * Unlock and free up our temp pages. | |
8b62f87b JB |
715 | * |
716 | * We use DO_ACCOUNTING here because we need the | |
717 | * delalloc_release_metadata to be done _after_ we drop | |
718 | * our outstanding extent for clearing delalloc for this | |
719 | * range. | |
771ed689 | 720 | */ |
ad7ff17b NB |
721 | extent_clear_unlock_delalloc(BTRFS_I(inode), start, end, |
722 | NULL, | |
74e9194a | 723 | clear_flags, |
ba8b04c1 | 724 | PAGE_UNLOCK | |
6869b0a8 | 725 | PAGE_START_WRITEBACK | |
e6eb4314 | 726 | page_error_op | |
c2790a2e | 727 | PAGE_END_WRITEBACK); |
cecc8d90 | 728 | |
1e6e238c QW |
729 | /* |
730 | * Ensure we only free the compressed pages if we have | |
731 | * them allocated, as we can still reach here with | |
732 | * inode_need_compress() == false. | |
733 | */ | |
734 | if (pages) { | |
735 | for (i = 0; i < nr_pages; i++) { | |
736 | WARN_ON(pages[i]->mapping); | |
737 | put_page(pages[i]); | |
738 | } | |
739 | kfree(pages); | |
cecc8d90 | 740 | } |
cecc8d90 | 741 | return 0; |
c8b97818 CM |
742 | } |
743 | } | |
744 | ||
745 | if (will_compress) { | |
746 | /* | |
747 | * we aren't doing an inline extent round the compressed size | |
748 | * up to a block size boundary so the allocator does sane | |
749 | * things | |
750 | */ | |
fda2832f | 751 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
752 | |
753 | /* | |
754 | * one last check to make sure the compression is really a | |
170607eb TT |
755 | * win, compare the page count read with the blocks on disk, |
756 | * compression must free at least one sector size | |
c8b97818 | 757 | */ |
09cbfeaf | 758 | total_in = ALIGN(total_in, PAGE_SIZE); |
170607eb | 759 | if (total_compressed + blocksize <= total_in) { |
ac3e9933 | 760 | compressed_extents++; |
c8bb0c8b AS |
761 | |
762 | /* | |
763 | * The async work queues will take care of doing actual | |
764 | * allocation on disk for these compressed pages, and | |
765 | * will submit them to the elevator. | |
766 | */ | |
b5326271 | 767 | add_async_extent(async_chunk, start, total_in, |
4d3a800e | 768 | total_compressed, pages, nr_pages, |
c8bb0c8b AS |
769 | compress_type); |
770 | ||
1170862d TT |
771 | if (start + total_in < end) { |
772 | start += total_in; | |
c8bb0c8b AS |
773 | pages = NULL; |
774 | cond_resched(); | |
775 | goto again; | |
776 | } | |
ac3e9933 | 777 | return compressed_extents; |
c8b97818 CM |
778 | } |
779 | } | |
c8bb0c8b | 780 | if (pages) { |
c8b97818 CM |
781 | /* |
782 | * the compression code ran but failed to make things smaller, | |
783 | * free any pages it allocated and our page pointer array | |
784 | */ | |
4d3a800e | 785 | for (i = 0; i < nr_pages; i++) { |
70b99e69 | 786 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 787 | put_page(pages[i]); |
c8b97818 CM |
788 | } |
789 | kfree(pages); | |
790 | pages = NULL; | |
791 | total_compressed = 0; | |
4d3a800e | 792 | nr_pages = 0; |
c8b97818 CM |
793 | |
794 | /* flag the file so we don't compress in the future */ | |
0b246afa | 795 | if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && |
b52aa8c9 | 796 | !(BTRFS_I(inode)->prop_compress)) { |
a555f810 | 797 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 798 | } |
c8b97818 | 799 | } |
f03d9301 | 800 | cleanup_and_bail_uncompressed: |
c8bb0c8b AS |
801 | /* |
802 | * No compression, but we still need to write the pages in the file | |
803 | * we've been given so far. redirty the locked page if it corresponds | |
804 | * to our extent and set things up for the async work queue to run | |
805 | * cow_file_range to do the normal delalloc dance. | |
806 | */ | |
1d53c9e6 CM |
807 | if (async_chunk->locked_page && |
808 | (page_offset(async_chunk->locked_page) >= start && | |
809 | page_offset(async_chunk->locked_page)) <= end) { | |
1368c6da | 810 | __set_page_dirty_nobuffers(async_chunk->locked_page); |
c8bb0c8b | 811 | /* unlocked later on in the async handlers */ |
1d53c9e6 | 812 | } |
c8bb0c8b AS |
813 | |
814 | if (redirty) | |
815 | extent_range_redirty_for_io(inode, start, end); | |
b5326271 | 816 | add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0, |
c8bb0c8b | 817 | BTRFS_COMPRESS_NONE); |
ac3e9933 | 818 | compressed_extents++; |
3b951516 | 819 | |
ac3e9933 | 820 | return compressed_extents; |
771ed689 | 821 | } |
771ed689 | 822 | |
40ae837b FM |
823 | static void free_async_extent_pages(struct async_extent *async_extent) |
824 | { | |
825 | int i; | |
826 | ||
827 | if (!async_extent->pages) | |
828 | return; | |
829 | ||
830 | for (i = 0; i < async_extent->nr_pages; i++) { | |
831 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 832 | put_page(async_extent->pages[i]); |
40ae837b FM |
833 | } |
834 | kfree(async_extent->pages); | |
835 | async_extent->nr_pages = 0; | |
836 | async_extent->pages = NULL; | |
771ed689 CM |
837 | } |
838 | ||
839 | /* | |
840 | * phase two of compressed writeback. This is the ordered portion | |
841 | * of the code, which only gets called in the order the work was | |
842 | * queued. We walk all the async extents created by compress_file_range | |
843 | * and send them down to the disk. | |
844 | */ | |
b5326271 | 845 | static noinline void submit_compressed_extents(struct async_chunk *async_chunk) |
771ed689 | 846 | { |
a0ff10dc NB |
847 | struct btrfs_inode *inode = BTRFS_I(async_chunk->inode); |
848 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
771ed689 CM |
849 | struct async_extent *async_extent; |
850 | u64 alloc_hint = 0; | |
771ed689 CM |
851 | struct btrfs_key ins; |
852 | struct extent_map *em; | |
a0ff10dc NB |
853 | struct btrfs_root *root = inode->root; |
854 | struct extent_io_tree *io_tree = &inode->io_tree; | |
f5a84ee3 | 855 | int ret = 0; |
771ed689 | 856 | |
3e04e7f1 | 857 | again: |
b5326271 NB |
858 | while (!list_empty(&async_chunk->extents)) { |
859 | async_extent = list_entry(async_chunk->extents.next, | |
771ed689 CM |
860 | struct async_extent, list); |
861 | list_del(&async_extent->list); | |
c8b97818 | 862 | |
f5a84ee3 | 863 | retry: |
7447555f NB |
864 | lock_extent(io_tree, async_extent->start, |
865 | async_extent->start + async_extent->ram_size - 1); | |
771ed689 CM |
866 | /* did the compression code fall back to uncompressed IO? */ |
867 | if (!async_extent->pages) { | |
868 | int page_started = 0; | |
869 | unsigned long nr_written = 0; | |
870 | ||
771ed689 | 871 | /* allocate blocks */ |
a0ff10dc | 872 | ret = cow_file_range(inode, async_chunk->locked_page, |
f5a84ee3 JB |
873 | async_extent->start, |
874 | async_extent->start + | |
875 | async_extent->ram_size - 1, | |
330a5827 | 876 | &page_started, &nr_written, 0); |
771ed689 | 877 | |
79787eaa JM |
878 | /* JDM XXX */ |
879 | ||
771ed689 CM |
880 | /* |
881 | * if page_started, cow_file_range inserted an | |
882 | * inline extent and took care of all the unlocking | |
883 | * and IO for us. Otherwise, we need to submit | |
884 | * all those pages down to the drive. | |
885 | */ | |
f5a84ee3 | 886 | if (!page_started && !ret) |
a0ff10dc | 887 | extent_write_locked_range(&inode->vfs_inode, |
5e3ee236 | 888 | async_extent->start, |
d397712b | 889 | async_extent->start + |
771ed689 | 890 | async_extent->ram_size - 1, |
771ed689 | 891 | WB_SYNC_ALL); |
1d53c9e6 | 892 | else if (ret && async_chunk->locked_page) |
b5326271 | 893 | unlock_page(async_chunk->locked_page); |
771ed689 CM |
894 | kfree(async_extent); |
895 | cond_resched(); | |
896 | continue; | |
897 | } | |
898 | ||
18513091 | 899 | ret = btrfs_reserve_extent(root, async_extent->ram_size, |
771ed689 CM |
900 | async_extent->compressed_size, |
901 | async_extent->compressed_size, | |
e570fd27 | 902 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 903 | if (ret) { |
40ae837b | 904 | free_async_extent_pages(async_extent); |
3e04e7f1 | 905 | |
fdf8e2ea JB |
906 | if (ret == -ENOSPC) { |
907 | unlock_extent(io_tree, async_extent->start, | |
908 | async_extent->start + | |
909 | async_extent->ram_size - 1); | |
ce62003f LB |
910 | |
911 | /* | |
912 | * we need to redirty the pages if we decide to | |
913 | * fallback to uncompressed IO, otherwise we | |
914 | * will not submit these pages down to lower | |
915 | * layers. | |
916 | */ | |
a0ff10dc | 917 | extent_range_redirty_for_io(&inode->vfs_inode, |
ce62003f LB |
918 | async_extent->start, |
919 | async_extent->start + | |
920 | async_extent->ram_size - 1); | |
921 | ||
79787eaa | 922 | goto retry; |
fdf8e2ea | 923 | } |
3e04e7f1 | 924 | goto out_free; |
f5a84ee3 | 925 | } |
c2167754 YZ |
926 | /* |
927 | * here we're doing allocation and writeback of the | |
928 | * compressed pages | |
929 | */ | |
a0ff10dc | 930 | em = create_io_em(inode, async_extent->start, |
6f9994db LB |
931 | async_extent->ram_size, /* len */ |
932 | async_extent->start, /* orig_start */ | |
933 | ins.objectid, /* block_start */ | |
934 | ins.offset, /* block_len */ | |
935 | ins.offset, /* orig_block_len */ | |
936 | async_extent->ram_size, /* ram_bytes */ | |
937 | async_extent->compress_type, | |
938 | BTRFS_ORDERED_COMPRESSED); | |
939 | if (IS_ERR(em)) | |
940 | /* ret value is not necessary due to void function */ | |
3e04e7f1 | 941 | goto out_free_reserve; |
6f9994db | 942 | free_extent_map(em); |
3e04e7f1 | 943 | |
a0ff10dc | 944 | ret = btrfs_add_ordered_extent_compress(inode, |
261507a0 LZ |
945 | async_extent->start, |
946 | ins.objectid, | |
947 | async_extent->ram_size, | |
948 | ins.offset, | |
261507a0 | 949 | async_extent->compress_type); |
d9f85963 | 950 | if (ret) { |
a0ff10dc | 951 | btrfs_drop_extent_cache(inode, async_extent->start, |
d9f85963 FM |
952 | async_extent->start + |
953 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 954 | goto out_free_reserve; |
d9f85963 | 955 | } |
0b246afa | 956 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
771ed689 | 957 | |
771ed689 CM |
958 | /* |
959 | * clear dirty, set writeback and unlock the pages. | |
960 | */ | |
a0ff10dc | 961 | extent_clear_unlock_delalloc(inode, async_extent->start, |
a791e35e CM |
962 | async_extent->start + |
963 | async_extent->ram_size - 1, | |
151a41bc | 964 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
6869b0a8 | 965 | PAGE_UNLOCK | PAGE_START_WRITEBACK); |
a0ff10dc | 966 | if (btrfs_submit_compressed_write(inode, async_extent->start, |
d397712b CM |
967 | async_extent->ram_size, |
968 | ins.objectid, | |
969 | ins.offset, async_extent->pages, | |
f82b7359 | 970 | async_extent->nr_pages, |
ec39f769 CM |
971 | async_chunk->write_flags, |
972 | async_chunk->blkcg_css)) { | |
fce2a4e6 FM |
973 | struct page *p = async_extent->pages[0]; |
974 | const u64 start = async_extent->start; | |
975 | const u64 end = start + async_extent->ram_size - 1; | |
976 | ||
a0ff10dc | 977 | p->mapping = inode->vfs_inode.i_mapping; |
38a39ac7 | 978 | btrfs_writepage_endio_finish_ordered(inode, p, start, |
25c1252a | 979 | end, false); |
7087a9d8 | 980 | |
fce2a4e6 | 981 | p->mapping = NULL; |
a0ff10dc | 982 | extent_clear_unlock_delalloc(inode, start, end, NULL, 0, |
fce2a4e6 FM |
983 | PAGE_END_WRITEBACK | |
984 | PAGE_SET_ERROR); | |
40ae837b | 985 | free_async_extent_pages(async_extent); |
fce2a4e6 | 986 | } |
771ed689 CM |
987 | alloc_hint = ins.objectid + ins.offset; |
988 | kfree(async_extent); | |
989 | cond_resched(); | |
990 | } | |
dec8f175 | 991 | return; |
3e04e7f1 | 992 | out_free_reserve: |
0b246afa | 993 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 994 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 995 | out_free: |
a0ff10dc | 996 | extent_clear_unlock_delalloc(inode, async_extent->start, |
3e04e7f1 JB |
997 | async_extent->start + |
998 | async_extent->ram_size - 1, | |
c2790a2e | 999 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
a7e3b975 | 1000 | EXTENT_DELALLOC_NEW | |
151a41bc | 1001 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
6869b0a8 QW |
1002 | PAGE_UNLOCK | PAGE_START_WRITEBACK | |
1003 | PAGE_END_WRITEBACK | PAGE_SET_ERROR); | |
40ae837b | 1004 | free_async_extent_pages(async_extent); |
79787eaa | 1005 | kfree(async_extent); |
3e04e7f1 | 1006 | goto again; |
771ed689 CM |
1007 | } |
1008 | ||
43c69849 | 1009 | static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start, |
4b46fce2 JB |
1010 | u64 num_bytes) |
1011 | { | |
43c69849 | 1012 | struct extent_map_tree *em_tree = &inode->extent_tree; |
4b46fce2 JB |
1013 | struct extent_map *em; |
1014 | u64 alloc_hint = 0; | |
1015 | ||
1016 | read_lock(&em_tree->lock); | |
1017 | em = search_extent_mapping(em_tree, start, num_bytes); | |
1018 | if (em) { | |
1019 | /* | |
1020 | * if block start isn't an actual block number then find the | |
1021 | * first block in this inode and use that as a hint. If that | |
1022 | * block is also bogus then just don't worry about it. | |
1023 | */ | |
1024 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
1025 | free_extent_map(em); | |
1026 | em = search_extent_mapping(em_tree, 0, 0); | |
1027 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
1028 | alloc_hint = em->block_start; | |
1029 | if (em) | |
1030 | free_extent_map(em); | |
1031 | } else { | |
1032 | alloc_hint = em->block_start; | |
1033 | free_extent_map(em); | |
1034 | } | |
1035 | } | |
1036 | read_unlock(&em_tree->lock); | |
1037 | ||
1038 | return alloc_hint; | |
1039 | } | |
1040 | ||
771ed689 CM |
1041 | /* |
1042 | * when extent_io.c finds a delayed allocation range in the file, | |
1043 | * the call backs end up in this code. The basic idea is to | |
1044 | * allocate extents on disk for the range, and create ordered data structs | |
1045 | * in ram to track those extents. | |
1046 | * | |
1047 | * locked_page is the page that writepage had locked already. We use | |
1048 | * it to make sure we don't do extra locks or unlocks. | |
1049 | * | |
1050 | * *page_started is set to one if we unlock locked_page and do everything | |
1051 | * required to start IO on it. It may be clean and already done with | |
1052 | * IO when we return. | |
1053 | */ | |
6e26c442 | 1054 | static noinline int cow_file_range(struct btrfs_inode *inode, |
00361589 | 1055 | struct page *locked_page, |
74e9194a | 1056 | u64 start, u64 end, int *page_started, |
330a5827 | 1057 | unsigned long *nr_written, int unlock) |
771ed689 | 1058 | { |
6e26c442 NB |
1059 | struct btrfs_root *root = inode->root; |
1060 | struct btrfs_fs_info *fs_info = root->fs_info; | |
771ed689 CM |
1061 | u64 alloc_hint = 0; |
1062 | u64 num_bytes; | |
1063 | unsigned long ram_size; | |
a315e68f | 1064 | u64 cur_alloc_size = 0; |
432cd2a1 | 1065 | u64 min_alloc_size; |
0b246afa | 1066 | u64 blocksize = fs_info->sectorsize; |
771ed689 CM |
1067 | struct btrfs_key ins; |
1068 | struct extent_map *em; | |
a315e68f FM |
1069 | unsigned clear_bits; |
1070 | unsigned long page_ops; | |
1071 | bool extent_reserved = false; | |
771ed689 CM |
1072 | int ret = 0; |
1073 | ||
6e26c442 | 1074 | if (btrfs_is_free_space_inode(inode)) { |
02ecd2c2 | 1075 | WARN_ON_ONCE(1); |
29bce2f3 JB |
1076 | ret = -EINVAL; |
1077 | goto out_unlock; | |
02ecd2c2 | 1078 | } |
771ed689 | 1079 | |
fda2832f | 1080 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 | 1081 | num_bytes = max(blocksize, num_bytes); |
566b1760 | 1082 | ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy)); |
771ed689 | 1083 | |
6e26c442 | 1084 | inode_should_defrag(inode, start, end, num_bytes, SZ_64K); |
4cb5300b | 1085 | |
7367253a QW |
1086 | /* |
1087 | * Due to the page size limit, for subpage we can only trigger the | |
1088 | * writeback for the dirty sectors of page, that means data writeback | |
1089 | * is doing more writeback than what we want. | |
1090 | * | |
1091 | * This is especially unexpected for some call sites like fallocate, | |
1092 | * where we only increase i_size after everything is done. | |
1093 | * This means we can trigger inline extent even if we didn't want to. | |
1094 | * So here we skip inline extent creation completely. | |
1095 | */ | |
1096 | if (start == 0 && fs_info->sectorsize == PAGE_SIZE) { | |
771ed689 | 1097 | /* lets try to make an inline extent */ |
6e26c442 | 1098 | ret = cow_file_range_inline(inode, start, end, 0, |
d02c0e20 | 1099 | BTRFS_COMPRESS_NONE, NULL); |
771ed689 | 1100 | if (ret == 0) { |
8b62f87b JB |
1101 | /* |
1102 | * We use DO_ACCOUNTING here because we need the | |
1103 | * delalloc_release_metadata to be run _after_ we drop | |
1104 | * our outstanding extent for clearing delalloc for this | |
1105 | * range. | |
1106 | */ | |
4750af3b QW |
1107 | extent_clear_unlock_delalloc(inode, start, end, |
1108 | locked_page, | |
c2790a2e | 1109 | EXTENT_LOCKED | EXTENT_DELALLOC | |
8b62f87b JB |
1110 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
1111 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
6869b0a8 | 1112 | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK); |
771ed689 | 1113 | *nr_written = *nr_written + |
09cbfeaf | 1114 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 1115 | *page_started = 1; |
4750af3b QW |
1116 | /* |
1117 | * locked_page is locked by the caller of | |
1118 | * writepage_delalloc(), not locked by | |
1119 | * __process_pages_contig(). | |
1120 | * | |
1121 | * We can't let __process_pages_contig() to unlock it, | |
1122 | * as it doesn't have any subpage::writers recorded. | |
1123 | * | |
1124 | * Here we manually unlock the page, since the caller | |
1125 | * can't use page_started to determine if it's an | |
1126 | * inline extent or a compressed extent. | |
1127 | */ | |
1128 | unlock_page(locked_page); | |
771ed689 | 1129 | goto out; |
79787eaa | 1130 | } else if (ret < 0) { |
79787eaa | 1131 | goto out_unlock; |
771ed689 CM |
1132 | } |
1133 | } | |
1134 | ||
6e26c442 NB |
1135 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
1136 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); | |
771ed689 | 1137 | |
432cd2a1 FM |
1138 | /* |
1139 | * Relocation relies on the relocated extents to have exactly the same | |
1140 | * size as the original extents. Normally writeback for relocation data | |
1141 | * extents follows a NOCOW path because relocation preallocates the | |
1142 | * extents. However, due to an operation such as scrub turning a block | |
1143 | * group to RO mode, it may fallback to COW mode, so we must make sure | |
1144 | * an extent allocated during COW has exactly the requested size and can | |
1145 | * not be split into smaller extents, otherwise relocation breaks and | |
1146 | * fails during the stage where it updates the bytenr of file extent | |
1147 | * items. | |
1148 | */ | |
1149 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1150 | min_alloc_size = num_bytes; | |
1151 | else | |
1152 | min_alloc_size = fs_info->sectorsize; | |
1153 | ||
3752d22f AJ |
1154 | while (num_bytes > 0) { |
1155 | cur_alloc_size = num_bytes; | |
18513091 | 1156 | ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size, |
432cd2a1 | 1157 | min_alloc_size, 0, alloc_hint, |
e570fd27 | 1158 | &ins, 1, 1); |
00361589 | 1159 | if (ret < 0) |
79787eaa | 1160 | goto out_unlock; |
a315e68f FM |
1161 | cur_alloc_size = ins.offset; |
1162 | extent_reserved = true; | |
d397712b | 1163 | |
771ed689 | 1164 | ram_size = ins.offset; |
6e26c442 | 1165 | em = create_io_em(inode, start, ins.offset, /* len */ |
6f9994db LB |
1166 | start, /* orig_start */ |
1167 | ins.objectid, /* block_start */ | |
1168 | ins.offset, /* block_len */ | |
1169 | ins.offset, /* orig_block_len */ | |
1170 | ram_size, /* ram_bytes */ | |
1171 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
1af4a0aa | 1172 | BTRFS_ORDERED_REGULAR /* type */); |
090a127a SY |
1173 | if (IS_ERR(em)) { |
1174 | ret = PTR_ERR(em); | |
ace68bac | 1175 | goto out_reserve; |
090a127a | 1176 | } |
6f9994db | 1177 | free_extent_map(em); |
e6dcd2dc | 1178 | |
6e26c442 | 1179 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
3c198fe0 QW |
1180 | ram_size, cur_alloc_size, |
1181 | BTRFS_ORDERED_REGULAR); | |
ace68bac | 1182 | if (ret) |
d9f85963 | 1183 | goto out_drop_extent_cache; |
c8b97818 | 1184 | |
17d217fe YZ |
1185 | if (root->root_key.objectid == |
1186 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
6e26c442 | 1187 | ret = btrfs_reloc_clone_csums(inode, start, |
17d217fe | 1188 | cur_alloc_size); |
4dbd80fb QW |
1189 | /* |
1190 | * Only drop cache here, and process as normal. | |
1191 | * | |
1192 | * We must not allow extent_clear_unlock_delalloc() | |
1193 | * at out_unlock label to free meta of this ordered | |
1194 | * extent, as its meta should be freed by | |
1195 | * btrfs_finish_ordered_io(). | |
1196 | * | |
1197 | * So we must continue until @start is increased to | |
1198 | * skip current ordered extent. | |
1199 | */ | |
00361589 | 1200 | if (ret) |
6e26c442 | 1201 | btrfs_drop_extent_cache(inode, start, |
4dbd80fb | 1202 | start + ram_size - 1, 0); |
17d217fe YZ |
1203 | } |
1204 | ||
0b246afa | 1205 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
9cfa3e34 | 1206 | |
f57ad937 QW |
1207 | /* |
1208 | * We're not doing compressed IO, don't unlock the first page | |
1209 | * (which the caller expects to stay locked), don't clear any | |
1210 | * dirty bits and don't set any writeback bits | |
8b62b72b | 1211 | * |
f57ad937 QW |
1212 | * Do set the Ordered (Private2) bit so we know this page was |
1213 | * properly setup for writepage. | |
c8b97818 | 1214 | */ |
a315e68f | 1215 | page_ops = unlock ? PAGE_UNLOCK : 0; |
f57ad937 | 1216 | page_ops |= PAGE_SET_ORDERED; |
a791e35e | 1217 | |
6e26c442 | 1218 | extent_clear_unlock_delalloc(inode, start, start + ram_size - 1, |
74e9194a | 1219 | locked_page, |
c2790a2e | 1220 | EXTENT_LOCKED | EXTENT_DELALLOC, |
a315e68f | 1221 | page_ops); |
3752d22f AJ |
1222 | if (num_bytes < cur_alloc_size) |
1223 | num_bytes = 0; | |
4dbd80fb | 1224 | else |
3752d22f | 1225 | num_bytes -= cur_alloc_size; |
c59f8951 CM |
1226 | alloc_hint = ins.objectid + ins.offset; |
1227 | start += cur_alloc_size; | |
a315e68f | 1228 | extent_reserved = false; |
4dbd80fb QW |
1229 | |
1230 | /* | |
1231 | * btrfs_reloc_clone_csums() error, since start is increased | |
1232 | * extent_clear_unlock_delalloc() at out_unlock label won't | |
1233 | * free metadata of current ordered extent, we're OK to exit. | |
1234 | */ | |
1235 | if (ret) | |
1236 | goto out_unlock; | |
b888db2b | 1237 | } |
79787eaa | 1238 | out: |
be20aa9d | 1239 | return ret; |
b7d5b0a8 | 1240 | |
d9f85963 | 1241 | out_drop_extent_cache: |
6e26c442 | 1242 | btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0); |
ace68bac | 1243 | out_reserve: |
0b246afa | 1244 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1245 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1246 | out_unlock: |
a7e3b975 FM |
1247 | clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
1248 | EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV; | |
6869b0a8 | 1249 | page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK; |
a315e68f FM |
1250 | /* |
1251 | * If we reserved an extent for our delalloc range (or a subrange) and | |
1252 | * failed to create the respective ordered extent, then it means that | |
1253 | * when we reserved the extent we decremented the extent's size from | |
1254 | * the data space_info's bytes_may_use counter and incremented the | |
1255 | * space_info's bytes_reserved counter by the same amount. We must make | |
1256 | * sure extent_clear_unlock_delalloc() does not try to decrement again | |
1257 | * the data space_info's bytes_may_use counter, therefore we do not pass | |
1258 | * it the flag EXTENT_CLEAR_DATA_RESV. | |
1259 | */ | |
1260 | if (extent_reserved) { | |
6e26c442 | 1261 | extent_clear_unlock_delalloc(inode, start, |
e2c8e92d | 1262 | start + cur_alloc_size - 1, |
a315e68f FM |
1263 | locked_page, |
1264 | clear_bits, | |
1265 | page_ops); | |
1266 | start += cur_alloc_size; | |
1267 | if (start >= end) | |
1268 | goto out; | |
1269 | } | |
6e26c442 | 1270 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
a315e68f FM |
1271 | clear_bits | EXTENT_CLEAR_DATA_RESV, |
1272 | page_ops); | |
79787eaa | 1273 | goto out; |
771ed689 | 1274 | } |
c8b97818 | 1275 | |
771ed689 CM |
1276 | /* |
1277 | * work queue call back to started compression on a file and pages | |
1278 | */ | |
1279 | static noinline void async_cow_start(struct btrfs_work *work) | |
1280 | { | |
b5326271 | 1281 | struct async_chunk *async_chunk; |
ac3e9933 | 1282 | int compressed_extents; |
771ed689 | 1283 | |
b5326271 | 1284 | async_chunk = container_of(work, struct async_chunk, work); |
771ed689 | 1285 | |
ac3e9933 NB |
1286 | compressed_extents = compress_file_range(async_chunk); |
1287 | if (compressed_extents == 0) { | |
b5326271 NB |
1288 | btrfs_add_delayed_iput(async_chunk->inode); |
1289 | async_chunk->inode = NULL; | |
8180ef88 | 1290 | } |
771ed689 CM |
1291 | } |
1292 | ||
1293 | /* | |
1294 | * work queue call back to submit previously compressed pages | |
1295 | */ | |
1296 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1297 | { | |
c5a68aec NB |
1298 | struct async_chunk *async_chunk = container_of(work, struct async_chunk, |
1299 | work); | |
1300 | struct btrfs_fs_info *fs_info = btrfs_work_owner(work); | |
771ed689 CM |
1301 | unsigned long nr_pages; |
1302 | ||
b5326271 | 1303 | nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >> |
09cbfeaf | 1304 | PAGE_SHIFT; |
771ed689 | 1305 | |
093258e6 | 1306 | /* atomic_sub_return implies a barrier */ |
0b246afa | 1307 | if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) < |
093258e6 DS |
1308 | 5 * SZ_1M) |
1309 | cond_wake_up_nomb(&fs_info->async_submit_wait); | |
771ed689 | 1310 | |
4546d178 | 1311 | /* |
b5326271 | 1312 | * ->inode could be NULL if async_chunk_start has failed to compress, |
4546d178 NB |
1313 | * in which case we don't have anything to submit, yet we need to |
1314 | * always adjust ->async_delalloc_pages as its paired with the init | |
1315 | * happening in cow_file_range_async | |
1316 | */ | |
b5326271 NB |
1317 | if (async_chunk->inode) |
1318 | submit_compressed_extents(async_chunk); | |
771ed689 | 1319 | } |
c8b97818 | 1320 | |
771ed689 CM |
1321 | static noinline void async_cow_free(struct btrfs_work *work) |
1322 | { | |
b5326271 | 1323 | struct async_chunk *async_chunk; |
97db1204 | 1324 | |
b5326271 NB |
1325 | async_chunk = container_of(work, struct async_chunk, work); |
1326 | if (async_chunk->inode) | |
1327 | btrfs_add_delayed_iput(async_chunk->inode); | |
ec39f769 CM |
1328 | if (async_chunk->blkcg_css) |
1329 | css_put(async_chunk->blkcg_css); | |
97db1204 NB |
1330 | /* |
1331 | * Since the pointer to 'pending' is at the beginning of the array of | |
b5326271 | 1332 | * async_chunk's, freeing it ensures the whole array has been freed. |
97db1204 | 1333 | */ |
b5326271 | 1334 | if (atomic_dec_and_test(async_chunk->pending)) |
b1c16ac9 | 1335 | kvfree(async_chunk->pending); |
771ed689 CM |
1336 | } |
1337 | ||
751b6431 | 1338 | static int cow_file_range_async(struct btrfs_inode *inode, |
ec39f769 CM |
1339 | struct writeback_control *wbc, |
1340 | struct page *locked_page, | |
771ed689 | 1341 | u64 start, u64 end, int *page_started, |
fac07d2b | 1342 | unsigned long *nr_written) |
771ed689 | 1343 | { |
751b6431 | 1344 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
ec39f769 | 1345 | struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc); |
97db1204 NB |
1346 | struct async_cow *ctx; |
1347 | struct async_chunk *async_chunk; | |
771ed689 CM |
1348 | unsigned long nr_pages; |
1349 | u64 cur_end; | |
97db1204 NB |
1350 | u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K); |
1351 | int i; | |
1352 | bool should_compress; | |
b1c16ac9 | 1353 | unsigned nofs_flag; |
fac07d2b | 1354 | const unsigned int write_flags = wbc_to_write_flags(wbc); |
771ed689 | 1355 | |
751b6431 | 1356 | unlock_extent(&inode->io_tree, start, end); |
97db1204 | 1357 | |
751b6431 | 1358 | if (inode->flags & BTRFS_INODE_NOCOMPRESS && |
97db1204 NB |
1359 | !btrfs_test_opt(fs_info, FORCE_COMPRESS)) { |
1360 | num_chunks = 1; | |
1361 | should_compress = false; | |
1362 | } else { | |
1363 | should_compress = true; | |
1364 | } | |
1365 | ||
b1c16ac9 NB |
1366 | nofs_flag = memalloc_nofs_save(); |
1367 | ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL); | |
1368 | memalloc_nofs_restore(nofs_flag); | |
1369 | ||
97db1204 NB |
1370 | if (!ctx) { |
1371 | unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | | |
1372 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | | |
1373 | EXTENT_DO_ACCOUNTING; | |
6869b0a8 QW |
1374 | unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | |
1375 | PAGE_END_WRITEBACK | PAGE_SET_ERROR; | |
97db1204 | 1376 | |
751b6431 NB |
1377 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1378 | clear_bits, page_ops); | |
97db1204 NB |
1379 | return -ENOMEM; |
1380 | } | |
1381 | ||
1382 | async_chunk = ctx->chunks; | |
1383 | atomic_set(&ctx->num_chunks, num_chunks); | |
1384 | ||
1385 | for (i = 0; i < num_chunks; i++) { | |
1386 | if (should_compress) | |
1387 | cur_end = min(end, start + SZ_512K - 1); | |
1388 | else | |
1389 | cur_end = end; | |
771ed689 | 1390 | |
bd4691a0 NB |
1391 | /* |
1392 | * igrab is called higher up in the call chain, take only the | |
1393 | * lightweight reference for the callback lifetime | |
1394 | */ | |
751b6431 | 1395 | ihold(&inode->vfs_inode); |
97db1204 | 1396 | async_chunk[i].pending = &ctx->num_chunks; |
751b6431 | 1397 | async_chunk[i].inode = &inode->vfs_inode; |
97db1204 NB |
1398 | async_chunk[i].start = start; |
1399 | async_chunk[i].end = cur_end; | |
97db1204 NB |
1400 | async_chunk[i].write_flags = write_flags; |
1401 | INIT_LIST_HEAD(&async_chunk[i].extents); | |
1402 | ||
1d53c9e6 CM |
1403 | /* |
1404 | * The locked_page comes all the way from writepage and its | |
1405 | * the original page we were actually given. As we spread | |
1406 | * this large delalloc region across multiple async_chunk | |
1407 | * structs, only the first struct needs a pointer to locked_page | |
1408 | * | |
1409 | * This way we don't need racey decisions about who is supposed | |
1410 | * to unlock it. | |
1411 | */ | |
1412 | if (locked_page) { | |
ec39f769 CM |
1413 | /* |
1414 | * Depending on the compressibility, the pages might or | |
1415 | * might not go through async. We want all of them to | |
1416 | * be accounted against wbc once. Let's do it here | |
1417 | * before the paths diverge. wbc accounting is used | |
1418 | * only for foreign writeback detection and doesn't | |
1419 | * need full accuracy. Just account the whole thing | |
1420 | * against the first page. | |
1421 | */ | |
1422 | wbc_account_cgroup_owner(wbc, locked_page, | |
1423 | cur_end - start); | |
1d53c9e6 CM |
1424 | async_chunk[i].locked_page = locked_page; |
1425 | locked_page = NULL; | |
1426 | } else { | |
1427 | async_chunk[i].locked_page = NULL; | |
1428 | } | |
1429 | ||
ec39f769 CM |
1430 | if (blkcg_css != blkcg_root_css) { |
1431 | css_get(blkcg_css); | |
1432 | async_chunk[i].blkcg_css = blkcg_css; | |
1433 | } else { | |
1434 | async_chunk[i].blkcg_css = NULL; | |
1435 | } | |
1436 | ||
a0cac0ec OS |
1437 | btrfs_init_work(&async_chunk[i].work, async_cow_start, |
1438 | async_cow_submit, async_cow_free); | |
771ed689 | 1439 | |
97db1204 | 1440 | nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE); |
0b246afa | 1441 | atomic_add(nr_pages, &fs_info->async_delalloc_pages); |
771ed689 | 1442 | |
97db1204 | 1443 | btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work); |
771ed689 | 1444 | |
771ed689 CM |
1445 | *nr_written += nr_pages; |
1446 | start = cur_end + 1; | |
1447 | } | |
1448 | *page_started = 1; | |
1449 | return 0; | |
be20aa9d CM |
1450 | } |
1451 | ||
42c01100 NA |
1452 | static noinline int run_delalloc_zoned(struct btrfs_inode *inode, |
1453 | struct page *locked_page, u64 start, | |
1454 | u64 end, int *page_started, | |
1455 | unsigned long *nr_written) | |
1456 | { | |
1457 | int ret; | |
1458 | ||
1459 | ret = cow_file_range(inode, locked_page, start, end, page_started, | |
1460 | nr_written, 0); | |
1461 | if (ret) | |
1462 | return ret; | |
1463 | ||
1464 | if (*page_started) | |
1465 | return 0; | |
1466 | ||
1467 | __set_page_dirty_nobuffers(locked_page); | |
1468 | account_page_redirty(locked_page); | |
1469 | extent_write_locked_range(&inode->vfs_inode, start, end, WB_SYNC_ALL); | |
1470 | *page_started = 1; | |
1471 | ||
1472 | return 0; | |
1473 | } | |
1474 | ||
2ff7e61e | 1475 | static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, |
17d217fe YZ |
1476 | u64 bytenr, u64 num_bytes) |
1477 | { | |
1478 | int ret; | |
1479 | struct btrfs_ordered_sum *sums; | |
1480 | LIST_HEAD(list); | |
1481 | ||
0b246afa | 1482 | ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr, |
a2de733c | 1483 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1484 | if (ret == 0 && list_empty(&list)) |
1485 | return 0; | |
1486 | ||
1487 | while (!list_empty(&list)) { | |
1488 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1489 | list_del(&sums->list); | |
1490 | kfree(sums); | |
1491 | } | |
58113753 LB |
1492 | if (ret < 0) |
1493 | return ret; | |
17d217fe YZ |
1494 | return 1; |
1495 | } | |
1496 | ||
8ba96f3d | 1497 | static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page, |
467dc47e FM |
1498 | const u64 start, const u64 end, |
1499 | int *page_started, unsigned long *nr_written) | |
1500 | { | |
8ba96f3d NB |
1501 | const bool is_space_ino = btrfs_is_free_space_inode(inode); |
1502 | const bool is_reloc_ino = (inode->root->root_key.objectid == | |
6bd335b4 | 1503 | BTRFS_DATA_RELOC_TREE_OBJECTID); |
2166e5ed | 1504 | const u64 range_bytes = end + 1 - start; |
8ba96f3d | 1505 | struct extent_io_tree *io_tree = &inode->io_tree; |
467dc47e FM |
1506 | u64 range_start = start; |
1507 | u64 count; | |
1508 | ||
1509 | /* | |
1510 | * If EXTENT_NORESERVE is set it means that when the buffered write was | |
1511 | * made we had not enough available data space and therefore we did not | |
1512 | * reserve data space for it, since we though we could do NOCOW for the | |
1513 | * respective file range (either there is prealloc extent or the inode | |
1514 | * has the NOCOW bit set). | |
1515 | * | |
1516 | * However when we need to fallback to COW mode (because for example the | |
1517 | * block group for the corresponding extent was turned to RO mode by a | |
1518 | * scrub or relocation) we need to do the following: | |
1519 | * | |
1520 | * 1) We increment the bytes_may_use counter of the data space info. | |
1521 | * If COW succeeds, it allocates a new data extent and after doing | |
1522 | * that it decrements the space info's bytes_may_use counter and | |
1523 | * increments its bytes_reserved counter by the same amount (we do | |
1524 | * this at btrfs_add_reserved_bytes()). So we need to increment the | |
1525 | * bytes_may_use counter to compensate (when space is reserved at | |
1526 | * buffered write time, the bytes_may_use counter is incremented); | |
1527 | * | |
1528 | * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so | |
1529 | * that if the COW path fails for any reason, it decrements (through | |
1530 | * extent_clear_unlock_delalloc()) the bytes_may_use counter of the | |
1531 | * data space info, which we incremented in the step above. | |
2166e5ed FM |
1532 | * |
1533 | * If we need to fallback to cow and the inode corresponds to a free | |
6bd335b4 FM |
1534 | * space cache inode or an inode of the data relocation tree, we must |
1535 | * also increment bytes_may_use of the data space_info for the same | |
1536 | * reason. Space caches and relocated data extents always get a prealloc | |
2166e5ed | 1537 | * extent for them, however scrub or balance may have set the block |
6bd335b4 FM |
1538 | * group that contains that extent to RO mode and therefore force COW |
1539 | * when starting writeback. | |
467dc47e | 1540 | */ |
2166e5ed | 1541 | count = count_range_bits(io_tree, &range_start, end, range_bytes, |
467dc47e | 1542 | EXTENT_NORESERVE, 0); |
6bd335b4 FM |
1543 | if (count > 0 || is_space_ino || is_reloc_ino) { |
1544 | u64 bytes = count; | |
8ba96f3d | 1545 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
467dc47e FM |
1546 | struct btrfs_space_info *sinfo = fs_info->data_sinfo; |
1547 | ||
6bd335b4 FM |
1548 | if (is_space_ino || is_reloc_ino) |
1549 | bytes = range_bytes; | |
1550 | ||
467dc47e | 1551 | spin_lock(&sinfo->lock); |
2166e5ed | 1552 | btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes); |
467dc47e FM |
1553 | spin_unlock(&sinfo->lock); |
1554 | ||
2166e5ed FM |
1555 | if (count > 0) |
1556 | clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE, | |
1557 | 0, 0, NULL); | |
467dc47e FM |
1558 | } |
1559 | ||
8ba96f3d NB |
1560 | return cow_file_range(inode, locked_page, start, end, page_started, |
1561 | nr_written, 1); | |
467dc47e FM |
1562 | } |
1563 | ||
d352ac68 CM |
1564 | /* |
1565 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1566 | * of the extents that exist in the file, and COWs the file as required. | |
1567 | * | |
1568 | * If no cow copies or snapshots exist, we write directly to the existing | |
1569 | * blocks on disk | |
1570 | */ | |
968322c8 | 1571 | static noinline int run_delalloc_nocow(struct btrfs_inode *inode, |
7f366cfe | 1572 | struct page *locked_page, |
3e024846 | 1573 | const u64 start, const u64 end, |
6e65ae76 | 1574 | int *page_started, |
3e024846 | 1575 | unsigned long *nr_written) |
be20aa9d | 1576 | { |
968322c8 NB |
1577 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
1578 | struct btrfs_root *root = inode->root; | |
be20aa9d | 1579 | struct btrfs_path *path; |
3e024846 NB |
1580 | u64 cow_start = (u64)-1; |
1581 | u64 cur_offset = start; | |
8ecebf4d | 1582 | int ret; |
3e024846 | 1583 | bool check_prev = true; |
968322c8 NB |
1584 | const bool freespace_inode = btrfs_is_free_space_inode(inode); |
1585 | u64 ino = btrfs_ino(inode); | |
762bf098 NB |
1586 | bool nocow = false; |
1587 | u64 disk_bytenr = 0; | |
6e65ae76 | 1588 | const bool force = inode->flags & BTRFS_INODE_NODATACOW; |
be20aa9d CM |
1589 | |
1590 | path = btrfs_alloc_path(); | |
17ca04af | 1591 | if (!path) { |
968322c8 | 1592 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
c2790a2e | 1593 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
1594 | EXTENT_DO_ACCOUNTING | |
1595 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
6869b0a8 | 1596 | PAGE_START_WRITEBACK | |
c2790a2e | 1597 | PAGE_END_WRITEBACK); |
d8926bb3 | 1598 | return -ENOMEM; |
17ca04af | 1599 | } |
82d5902d | 1600 | |
80ff3856 | 1601 | while (1) { |
3e024846 NB |
1602 | struct btrfs_key found_key; |
1603 | struct btrfs_file_extent_item *fi; | |
1604 | struct extent_buffer *leaf; | |
1605 | u64 extent_end; | |
1606 | u64 extent_offset; | |
3e024846 NB |
1607 | u64 num_bytes = 0; |
1608 | u64 disk_num_bytes; | |
3e024846 NB |
1609 | u64 ram_bytes; |
1610 | int extent_type; | |
762bf098 NB |
1611 | |
1612 | nocow = false; | |
3e024846 | 1613 | |
e4c3b2dc | 1614 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, |
80ff3856 | 1615 | cur_offset, 0); |
d788a349 | 1616 | if (ret < 0) |
79787eaa | 1617 | goto error; |
a6bd9cd1 NB |
1618 | |
1619 | /* | |
1620 | * If there is no extent for our range when doing the initial | |
1621 | * search, then go back to the previous slot as it will be the | |
1622 | * one containing the search offset | |
1623 | */ | |
80ff3856 YZ |
1624 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1625 | leaf = path->nodes[0]; | |
1626 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1627 | path->slots[0] - 1); | |
33345d01 | 1628 | if (found_key.objectid == ino && |
80ff3856 YZ |
1629 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1630 | path->slots[0]--; | |
1631 | } | |
3e024846 | 1632 | check_prev = false; |
80ff3856 | 1633 | next_slot: |
a6bd9cd1 | 1634 | /* Go to next leaf if we have exhausted the current one */ |
80ff3856 YZ |
1635 | leaf = path->nodes[0]; |
1636 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1637 | ret = btrfs_next_leaf(root, path); | |
e8916699 LB |
1638 | if (ret < 0) { |
1639 | if (cow_start != (u64)-1) | |
1640 | cur_offset = cow_start; | |
79787eaa | 1641 | goto error; |
e8916699 | 1642 | } |
80ff3856 YZ |
1643 | if (ret > 0) |
1644 | break; | |
1645 | leaf = path->nodes[0]; | |
1646 | } | |
be20aa9d | 1647 | |
80ff3856 YZ |
1648 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1649 | ||
a6bd9cd1 | 1650 | /* Didn't find anything for our INO */ |
1d512cb7 FM |
1651 | if (found_key.objectid > ino) |
1652 | break; | |
a6bd9cd1 NB |
1653 | /* |
1654 | * Keep searching until we find an EXTENT_ITEM or there are no | |
1655 | * more extents for this inode | |
1656 | */ | |
1d512cb7 FM |
1657 | if (WARN_ON_ONCE(found_key.objectid < ino) || |
1658 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1659 | path->slots[0]++; | |
1660 | goto next_slot; | |
1661 | } | |
a6bd9cd1 NB |
1662 | |
1663 | /* Found key is not EXTENT_DATA_KEY or starts after req range */ | |
1d512cb7 | 1664 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || |
80ff3856 YZ |
1665 | found_key.offset > end) |
1666 | break; | |
1667 | ||
a6bd9cd1 NB |
1668 | /* |
1669 | * If the found extent starts after requested offset, then | |
1670 | * adjust extent_end to be right before this extent begins | |
1671 | */ | |
80ff3856 YZ |
1672 | if (found_key.offset > cur_offset) { |
1673 | extent_end = found_key.offset; | |
e9061e21 | 1674 | extent_type = 0; |
80ff3856 YZ |
1675 | goto out_check; |
1676 | } | |
1677 | ||
a6bd9cd1 NB |
1678 | /* |
1679 | * Found extent which begins before our range and potentially | |
1680 | * intersect it | |
1681 | */ | |
80ff3856 YZ |
1682 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1683 | struct btrfs_file_extent_item); | |
1684 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1685 | ||
cc95bef6 | 1686 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1687 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1688 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1689 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1690 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1691 | extent_end = found_key.offset + |
1692 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1693 | disk_num_bytes = |
1694 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
a6bd9cd1 | 1695 | /* |
de7999af FM |
1696 | * If the extent we got ends before our current offset, |
1697 | * skip to the next extent. | |
a6bd9cd1 | 1698 | */ |
de7999af | 1699 | if (extent_end <= cur_offset) { |
80ff3856 YZ |
1700 | path->slots[0]++; |
1701 | goto next_slot; | |
1702 | } | |
a6bd9cd1 | 1703 | /* Skip holes */ |
17d217fe YZ |
1704 | if (disk_bytenr == 0) |
1705 | goto out_check; | |
a6bd9cd1 | 1706 | /* Skip compressed/encrypted/encoded extents */ |
80ff3856 YZ |
1707 | if (btrfs_file_extent_compression(leaf, fi) || |
1708 | btrfs_file_extent_encryption(leaf, fi) || | |
1709 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1710 | goto out_check; | |
78d4295b | 1711 | /* |
a6bd9cd1 NB |
1712 | * If extent is created before the last volume's snapshot |
1713 | * this implies the extent is shared, hence we can't do | |
1714 | * nocow. This is the same check as in | |
1715 | * btrfs_cross_ref_exist but without calling | |
1716 | * btrfs_search_slot. | |
78d4295b | 1717 | */ |
3e024846 | 1718 | if (!freespace_inode && |
27a7ff55 | 1719 | btrfs_file_extent_generation(leaf, fi) <= |
78d4295b EL |
1720 | btrfs_root_last_snapshot(&root->root_item)) |
1721 | goto out_check; | |
d899e052 YZ |
1722 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1723 | goto out_check; | |
c65ca98f FM |
1724 | |
1725 | /* | |
1726 | * The following checks can be expensive, as they need to | |
1727 | * take other locks and do btree or rbtree searches, so | |
1728 | * release the path to avoid blocking other tasks for too | |
1729 | * long. | |
1730 | */ | |
1731 | btrfs_release_path(path); | |
1732 | ||
58113753 LB |
1733 | ret = btrfs_cross_ref_exist(root, ino, |
1734 | found_key.offset - | |
a84d5d42 | 1735 | extent_offset, disk_bytenr, false); |
58113753 LB |
1736 | if (ret) { |
1737 | /* | |
1738 | * ret could be -EIO if the above fails to read | |
1739 | * metadata. | |
1740 | */ | |
1741 | if (ret < 0) { | |
1742 | if (cow_start != (u64)-1) | |
1743 | cur_offset = cow_start; | |
1744 | goto error; | |
1745 | } | |
1746 | ||
3e024846 | 1747 | WARN_ON_ONCE(freespace_inode); |
17d217fe | 1748 | goto out_check; |
58113753 | 1749 | } |
5d4f98a2 | 1750 | disk_bytenr += extent_offset; |
17d217fe YZ |
1751 | disk_bytenr += cur_offset - found_key.offset; |
1752 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 | 1753 | /* |
a6bd9cd1 NB |
1754 | * If there are pending snapshots for this root, we |
1755 | * fall into common COW way | |
e9894fd3 | 1756 | */ |
3e024846 | 1757 | if (!freespace_inode && atomic_read(&root->snapshot_force_cow)) |
8ecebf4d | 1758 | goto out_check; |
17d217fe YZ |
1759 | /* |
1760 | * force cow if csum exists in the range. | |
1761 | * this ensure that csum for a given extent are | |
1762 | * either valid or do not exist. | |
1763 | */ | |
58113753 LB |
1764 | ret = csum_exist_in_range(fs_info, disk_bytenr, |
1765 | num_bytes); | |
1766 | if (ret) { | |
58113753 LB |
1767 | /* |
1768 | * ret could be -EIO if the above fails to read | |
1769 | * metadata. | |
1770 | */ | |
1771 | if (ret < 0) { | |
1772 | if (cow_start != (u64)-1) | |
1773 | cur_offset = cow_start; | |
1774 | goto error; | |
1775 | } | |
3e024846 | 1776 | WARN_ON_ONCE(freespace_inode); |
17d217fe | 1777 | goto out_check; |
91e1f56a | 1778 | } |
20903032 | 1779 | /* If the extent's block group is RO, we must COW */ |
8ecebf4d | 1780 | if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) |
f78c436c | 1781 | goto out_check; |
3e024846 | 1782 | nocow = true; |
80ff3856 | 1783 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
e8e21007 NB |
1784 | extent_end = found_key.offset + ram_bytes; |
1785 | extent_end = ALIGN(extent_end, fs_info->sectorsize); | |
922f0518 NB |
1786 | /* Skip extents outside of our requested range */ |
1787 | if (extent_end <= start) { | |
1788 | path->slots[0]++; | |
1789 | goto next_slot; | |
1790 | } | |
80ff3856 | 1791 | } else { |
e8e21007 | 1792 | /* If this triggers then we have a memory corruption */ |
290342f6 | 1793 | BUG(); |
80ff3856 YZ |
1794 | } |
1795 | out_check: | |
a6bd9cd1 NB |
1796 | /* |
1797 | * If nocow is false then record the beginning of the range | |
1798 | * that needs to be COWed | |
1799 | */ | |
80ff3856 YZ |
1800 | if (!nocow) { |
1801 | if (cow_start == (u64)-1) | |
1802 | cow_start = cur_offset; | |
1803 | cur_offset = extent_end; | |
1804 | if (cur_offset > end) | |
1805 | break; | |
c65ca98f FM |
1806 | if (!path->nodes[0]) |
1807 | continue; | |
80ff3856 YZ |
1808 | path->slots[0]++; |
1809 | goto next_slot; | |
7ea394f1 YZ |
1810 | } |
1811 | ||
a6bd9cd1 NB |
1812 | /* |
1813 | * COW range from cow_start to found_key.offset - 1. As the key | |
1814 | * will contain the beginning of the first extent that can be | |
1815 | * NOCOW, following one which needs to be COW'ed | |
1816 | */ | |
80ff3856 | 1817 | if (cow_start != (u64)-1) { |
968322c8 | 1818 | ret = fallback_to_cow(inode, locked_page, |
8ba96f3d | 1819 | cow_start, found_key.offset - 1, |
467dc47e | 1820 | page_started, nr_written); |
230ed397 | 1821 | if (ret) |
79787eaa | 1822 | goto error; |
80ff3856 | 1823 | cow_start = (u64)-1; |
7ea394f1 | 1824 | } |
80ff3856 | 1825 | |
d899e052 | 1826 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
6f9994db | 1827 | u64 orig_start = found_key.offset - extent_offset; |
3e024846 | 1828 | struct extent_map *em; |
6f9994db | 1829 | |
968322c8 | 1830 | em = create_io_em(inode, cur_offset, num_bytes, |
6f9994db LB |
1831 | orig_start, |
1832 | disk_bytenr, /* block_start */ | |
1833 | num_bytes, /* block_len */ | |
1834 | disk_num_bytes, /* orig_block_len */ | |
1835 | ram_bytes, BTRFS_COMPRESS_NONE, | |
1836 | BTRFS_ORDERED_PREALLOC); | |
1837 | if (IS_ERR(em)) { | |
6f9994db LB |
1838 | ret = PTR_ERR(em); |
1839 | goto error; | |
d899e052 | 1840 | } |
6f9994db | 1841 | free_extent_map(em); |
968322c8 | 1842 | ret = btrfs_add_ordered_extent(inode, cur_offset, |
bb55f626 NB |
1843 | disk_bytenr, num_bytes, |
1844 | num_bytes, | |
1845 | BTRFS_ORDERED_PREALLOC); | |
762bf098 | 1846 | if (ret) { |
968322c8 | 1847 | btrfs_drop_extent_cache(inode, cur_offset, |
762bf098 NB |
1848 | cur_offset + num_bytes - 1, |
1849 | 0); | |
1850 | goto error; | |
1851 | } | |
d899e052 | 1852 | } else { |
968322c8 | 1853 | ret = btrfs_add_ordered_extent(inode, cur_offset, |
bb55f626 NB |
1854 | disk_bytenr, num_bytes, |
1855 | num_bytes, | |
1856 | BTRFS_ORDERED_NOCOW); | |
762bf098 NB |
1857 | if (ret) |
1858 | goto error; | |
d899e052 | 1859 | } |
80ff3856 | 1860 | |
f78c436c | 1861 | if (nocow) |
0b246afa | 1862 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
762bf098 | 1863 | nocow = false; |
771ed689 | 1864 | |
efa56464 | 1865 | if (root->root_key.objectid == |
4dbd80fb QW |
1866 | BTRFS_DATA_RELOC_TREE_OBJECTID) |
1867 | /* | |
1868 | * Error handled later, as we must prevent | |
1869 | * extent_clear_unlock_delalloc() in error handler | |
1870 | * from freeing metadata of created ordered extent. | |
1871 | */ | |
968322c8 | 1872 | ret = btrfs_reloc_clone_csums(inode, cur_offset, |
efa56464 | 1873 | num_bytes); |
efa56464 | 1874 | |
968322c8 | 1875 | extent_clear_unlock_delalloc(inode, cur_offset, |
74e9194a | 1876 | cur_offset + num_bytes - 1, |
c2790a2e | 1877 | locked_page, EXTENT_LOCKED | |
18513091 WX |
1878 | EXTENT_DELALLOC | |
1879 | EXTENT_CLEAR_DATA_RESV, | |
f57ad937 | 1880 | PAGE_UNLOCK | PAGE_SET_ORDERED); |
18513091 | 1881 | |
80ff3856 | 1882 | cur_offset = extent_end; |
4dbd80fb QW |
1883 | |
1884 | /* | |
1885 | * btrfs_reloc_clone_csums() error, now we're OK to call error | |
1886 | * handler, as metadata for created ordered extent will only | |
1887 | * be freed by btrfs_finish_ordered_io(). | |
1888 | */ | |
1889 | if (ret) | |
1890 | goto error; | |
80ff3856 YZ |
1891 | if (cur_offset > end) |
1892 | break; | |
be20aa9d | 1893 | } |
b3b4aa74 | 1894 | btrfs_release_path(path); |
80ff3856 | 1895 | |
506481b2 | 1896 | if (cur_offset <= end && cow_start == (u64)-1) |
80ff3856 | 1897 | cow_start = cur_offset; |
17ca04af | 1898 | |
80ff3856 | 1899 | if (cow_start != (u64)-1) { |
506481b2 | 1900 | cur_offset = end; |
968322c8 NB |
1901 | ret = fallback_to_cow(inode, locked_page, cow_start, end, |
1902 | page_started, nr_written); | |
d788a349 | 1903 | if (ret) |
79787eaa | 1904 | goto error; |
80ff3856 YZ |
1905 | } |
1906 | ||
79787eaa | 1907 | error: |
762bf098 NB |
1908 | if (nocow) |
1909 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); | |
1910 | ||
17ca04af | 1911 | if (ret && cur_offset < end) |
968322c8 | 1912 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
c2790a2e | 1913 | locked_page, EXTENT_LOCKED | |
151a41bc JB |
1914 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1915 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
6869b0a8 | 1916 | PAGE_START_WRITEBACK | |
c2790a2e | 1917 | PAGE_END_WRITEBACK); |
7ea394f1 | 1918 | btrfs_free_path(path); |
79787eaa | 1919 | return ret; |
be20aa9d CM |
1920 | } |
1921 | ||
6e65ae76 | 1922 | static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end) |
47059d93 | 1923 | { |
6e65ae76 GR |
1924 | if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) { |
1925 | if (inode->defrag_bytes && | |
1926 | test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG, | |
1927 | 0, NULL)) | |
1928 | return false; | |
1929 | return true; | |
1930 | } | |
1931 | return false; | |
47059d93 WS |
1932 | } |
1933 | ||
d352ac68 | 1934 | /* |
5eaad97a NB |
1935 | * Function to process delayed allocation (create CoW) for ranges which are |
1936 | * being touched for the first time. | |
d352ac68 | 1937 | */ |
98456b9c | 1938 | int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page, |
5eaad97a NB |
1939 | u64 start, u64 end, int *page_started, unsigned long *nr_written, |
1940 | struct writeback_control *wbc) | |
be20aa9d | 1941 | { |
be20aa9d | 1942 | int ret; |
42c01100 | 1943 | const bool zoned = btrfs_is_zoned(inode->root->fs_info); |
a2135011 | 1944 | |
6e65ae76 | 1945 | if (should_nocow(inode, start, end)) { |
42c01100 | 1946 | ASSERT(!zoned); |
98456b9c | 1947 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
6e65ae76 | 1948 | page_started, nr_written); |
98456b9c NB |
1949 | } else if (!inode_can_compress(inode) || |
1950 | !inode_need_compress(inode, start, end)) { | |
42c01100 NA |
1951 | if (zoned) |
1952 | ret = run_delalloc_zoned(inode, locked_page, start, end, | |
1953 | page_started, nr_written); | |
1954 | else | |
1955 | ret = cow_file_range(inode, locked_page, start, end, | |
1956 | page_started, nr_written, 1); | |
7ddf5a42 | 1957 | } else { |
98456b9c NB |
1958 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags); |
1959 | ret = cow_file_range_async(inode, wbc, locked_page, start, end, | |
fac07d2b | 1960 | page_started, nr_written); |
7ddf5a42 | 1961 | } |
52427260 | 1962 | if (ret) |
98456b9c | 1963 | btrfs_cleanup_ordered_extents(inode, locked_page, start, |
d1051d6e | 1964 | end - start + 1); |
b888db2b CM |
1965 | return ret; |
1966 | } | |
1967 | ||
abbb55f4 NB |
1968 | void btrfs_split_delalloc_extent(struct inode *inode, |
1969 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1970 | { |
dcab6a3b JB |
1971 | u64 size; |
1972 | ||
0ca1f7ce | 1973 | /* not delalloc, ignore it */ |
9ed74f2d | 1974 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1975 | return; |
9ed74f2d | 1976 | |
dcab6a3b JB |
1977 | size = orig->end - orig->start + 1; |
1978 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
823bb20a | 1979 | u32 num_extents; |
dcab6a3b JB |
1980 | u64 new_size; |
1981 | ||
1982 | /* | |
5c848198 | 1983 | * See the explanation in btrfs_merge_delalloc_extent, the same |
ba117213 | 1984 | * applies here, just in reverse. |
dcab6a3b JB |
1985 | */ |
1986 | new_size = orig->end - split + 1; | |
823bb20a | 1987 | num_extents = count_max_extents(new_size); |
ba117213 | 1988 | new_size = split - orig->start; |
823bb20a DS |
1989 | num_extents += count_max_extents(new_size); |
1990 | if (count_max_extents(size) >= num_extents) | |
dcab6a3b JB |
1991 | return; |
1992 | } | |
1993 | ||
9e0baf60 | 1994 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 1995 | btrfs_mod_outstanding_extents(BTRFS_I(inode), 1); |
9e0baf60 | 1996 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
1997 | } |
1998 | ||
1999 | /* | |
5c848198 NB |
2000 | * Handle merged delayed allocation extents so we can keep track of new extents |
2001 | * that are just merged onto old extents, such as when we are doing sequential | |
2002 | * writes, so we can properly account for the metadata space we'll need. | |
9ed74f2d | 2003 | */ |
5c848198 NB |
2004 | void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new, |
2005 | struct extent_state *other) | |
9ed74f2d | 2006 | { |
dcab6a3b | 2007 | u64 new_size, old_size; |
823bb20a | 2008 | u32 num_extents; |
dcab6a3b | 2009 | |
9ed74f2d JB |
2010 | /* not delalloc, ignore it */ |
2011 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 2012 | return; |
9ed74f2d | 2013 | |
8461a3de JB |
2014 | if (new->start > other->start) |
2015 | new_size = new->end - other->start + 1; | |
2016 | else | |
2017 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
2018 | |
2019 | /* we're not bigger than the max, unreserve the space and go */ | |
2020 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
2021 | spin_lock(&BTRFS_I(inode)->lock); | |
8b62f87b | 2022 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
dcab6a3b JB |
2023 | spin_unlock(&BTRFS_I(inode)->lock); |
2024 | return; | |
2025 | } | |
2026 | ||
2027 | /* | |
ba117213 JB |
2028 | * We have to add up either side to figure out how many extents were |
2029 | * accounted for before we merged into one big extent. If the number of | |
2030 | * extents we accounted for is <= the amount we need for the new range | |
2031 | * then we can return, otherwise drop. Think of it like this | |
2032 | * | |
2033 | * [ 4k][MAX_SIZE] | |
2034 | * | |
2035 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
2036 | * need 2 outstanding extents, on one side we have 1 and the other side | |
2037 | * we have 1 so they are == and we can return. But in this case | |
2038 | * | |
2039 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
2040 | * | |
2041 | * Each range on their own accounts for 2 extents, but merged together | |
2042 | * they are only 3 extents worth of accounting, so we need to drop in | |
2043 | * this case. | |
dcab6a3b | 2044 | */ |
ba117213 | 2045 | old_size = other->end - other->start + 1; |
823bb20a | 2046 | num_extents = count_max_extents(old_size); |
ba117213 | 2047 | old_size = new->end - new->start + 1; |
823bb20a DS |
2048 | num_extents += count_max_extents(old_size); |
2049 | if (count_max_extents(new_size) >= num_extents) | |
dcab6a3b JB |
2050 | return; |
2051 | ||
9e0baf60 | 2052 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 2053 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
9e0baf60 | 2054 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
2055 | } |
2056 | ||
eb73c1b7 MX |
2057 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
2058 | struct inode *inode) | |
2059 | { | |
0b246afa JM |
2060 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2061 | ||
eb73c1b7 MX |
2062 | spin_lock(&root->delalloc_lock); |
2063 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
2064 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
2065 | &root->delalloc_inodes); | |
2066 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
2067 | &BTRFS_I(inode)->runtime_flags); | |
2068 | root->nr_delalloc_inodes++; | |
2069 | if (root->nr_delalloc_inodes == 1) { | |
0b246afa | 2070 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2071 | BUG_ON(!list_empty(&root->delalloc_root)); |
2072 | list_add_tail(&root->delalloc_root, | |
0b246afa JM |
2073 | &fs_info->delalloc_roots); |
2074 | spin_unlock(&fs_info->delalloc_root_lock); | |
eb73c1b7 MX |
2075 | } |
2076 | } | |
2077 | spin_unlock(&root->delalloc_lock); | |
2078 | } | |
2079 | ||
2b877331 NB |
2080 | |
2081 | void __btrfs_del_delalloc_inode(struct btrfs_root *root, | |
2082 | struct btrfs_inode *inode) | |
eb73c1b7 | 2083 | { |
3ffbd68c | 2084 | struct btrfs_fs_info *fs_info = root->fs_info; |
0b246afa | 2085 | |
9e3e97f4 NB |
2086 | if (!list_empty(&inode->delalloc_inodes)) { |
2087 | list_del_init(&inode->delalloc_inodes); | |
eb73c1b7 | 2088 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 2089 | &inode->runtime_flags); |
eb73c1b7 MX |
2090 | root->nr_delalloc_inodes--; |
2091 | if (!root->nr_delalloc_inodes) { | |
7c8a0d36 | 2092 | ASSERT(list_empty(&root->delalloc_inodes)); |
0b246afa | 2093 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2094 | BUG_ON(list_empty(&root->delalloc_root)); |
2095 | list_del_init(&root->delalloc_root); | |
0b246afa | 2096 | spin_unlock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2097 | } |
2098 | } | |
2b877331 NB |
2099 | } |
2100 | ||
2101 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
2102 | struct btrfs_inode *inode) | |
2103 | { | |
2104 | spin_lock(&root->delalloc_lock); | |
2105 | __btrfs_del_delalloc_inode(root, inode); | |
eb73c1b7 MX |
2106 | spin_unlock(&root->delalloc_lock); |
2107 | } | |
2108 | ||
d352ac68 | 2109 | /* |
e06a1fc9 NB |
2110 | * Properly track delayed allocation bytes in the inode and to maintain the |
2111 | * list of inodes that have pending delalloc work to be done. | |
d352ac68 | 2112 | */ |
e06a1fc9 NB |
2113 | void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state, |
2114 | unsigned *bits) | |
291d673e | 2115 | { |
0b246afa JM |
2116 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2117 | ||
47059d93 WS |
2118 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
2119 | WARN_ON(1); | |
75eff68e CM |
2120 | /* |
2121 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 2122 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
2123 | * bit, which is only set or cleared with irqs on |
2124 | */ | |
0ca1f7ce | 2125 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 2126 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2127 | u64 len = state->end + 1 - state->start; |
8b62f87b | 2128 | u32 num_extents = count_max_extents(len); |
70ddc553 | 2129 | bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode)); |
9ed74f2d | 2130 | |
8b62f87b JB |
2131 | spin_lock(&BTRFS_I(inode)->lock); |
2132 | btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents); | |
2133 | spin_unlock(&BTRFS_I(inode)->lock); | |
287a0ab9 | 2134 | |
6a3891c5 | 2135 | /* For sanity tests */ |
0b246afa | 2136 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
2137 | return; |
2138 | ||
104b4e51 NB |
2139 | percpu_counter_add_batch(&fs_info->delalloc_bytes, len, |
2140 | fs_info->delalloc_batch); | |
df0af1a5 | 2141 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 2142 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
2143 | if (*bits & EXTENT_DEFRAG) |
2144 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 2145 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
2146 | &BTRFS_I(inode)->runtime_flags)) |
2147 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 2148 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 2149 | } |
a7e3b975 FM |
2150 | |
2151 | if (!(state->state & EXTENT_DELALLOC_NEW) && | |
2152 | (*bits & EXTENT_DELALLOC_NEW)) { | |
2153 | spin_lock(&BTRFS_I(inode)->lock); | |
2154 | BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 - | |
2155 | state->start; | |
2156 | spin_unlock(&BTRFS_I(inode)->lock); | |
2157 | } | |
291d673e CM |
2158 | } |
2159 | ||
d352ac68 | 2160 | /* |
a36bb5f9 NB |
2161 | * Once a range is no longer delalloc this function ensures that proper |
2162 | * accounting happens. | |
d352ac68 | 2163 | */ |
a36bb5f9 NB |
2164 | void btrfs_clear_delalloc_extent(struct inode *vfs_inode, |
2165 | struct extent_state *state, unsigned *bits) | |
291d673e | 2166 | { |
a36bb5f9 NB |
2167 | struct btrfs_inode *inode = BTRFS_I(vfs_inode); |
2168 | struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb); | |
47059d93 | 2169 | u64 len = state->end + 1 - state->start; |
823bb20a | 2170 | u32 num_extents = count_max_extents(len); |
47059d93 | 2171 | |
4a4b964f FM |
2172 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) { |
2173 | spin_lock(&inode->lock); | |
6fc0ef68 | 2174 | inode->defrag_bytes -= len; |
4a4b964f FM |
2175 | spin_unlock(&inode->lock); |
2176 | } | |
47059d93 | 2177 | |
75eff68e CM |
2178 | /* |
2179 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 2180 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
2181 | * bit, which is only set or cleared with irqs on |
2182 | */ | |
0ca1f7ce | 2183 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
6fc0ef68 | 2184 | struct btrfs_root *root = inode->root; |
83eea1f1 | 2185 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 2186 | |
8b62f87b JB |
2187 | spin_lock(&inode->lock); |
2188 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
2189 | spin_unlock(&inode->lock); | |
0ca1f7ce | 2190 | |
b6d08f06 JB |
2191 | /* |
2192 | * We don't reserve metadata space for space cache inodes so we | |
52042d8e | 2193 | * don't need to call delalloc_release_metadata if there is an |
b6d08f06 JB |
2194 | * error. |
2195 | */ | |
a315e68f | 2196 | if (*bits & EXTENT_CLEAR_META_RESV && |
0b246afa | 2197 | root != fs_info->tree_root) |
43b18595 | 2198 | btrfs_delalloc_release_metadata(inode, len, false); |
0ca1f7ce | 2199 | |
6a3891c5 | 2200 | /* For sanity tests. */ |
0b246afa | 2201 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
2202 | return; |
2203 | ||
a315e68f FM |
2204 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID && |
2205 | do_list && !(state->state & EXTENT_NORESERVE) && | |
2206 | (*bits & EXTENT_CLEAR_DATA_RESV)) | |
9db5d510 | 2207 | btrfs_free_reserved_data_space_noquota(fs_info, len); |
9ed74f2d | 2208 | |
104b4e51 NB |
2209 | percpu_counter_add_batch(&fs_info->delalloc_bytes, -len, |
2210 | fs_info->delalloc_batch); | |
6fc0ef68 NB |
2211 | spin_lock(&inode->lock); |
2212 | inode->delalloc_bytes -= len; | |
2213 | if (do_list && inode->delalloc_bytes == 0 && | |
df0af1a5 | 2214 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 2215 | &inode->runtime_flags)) |
eb73c1b7 | 2216 | btrfs_del_delalloc_inode(root, inode); |
6fc0ef68 | 2217 | spin_unlock(&inode->lock); |
291d673e | 2218 | } |
a7e3b975 FM |
2219 | |
2220 | if ((state->state & EXTENT_DELALLOC_NEW) && | |
2221 | (*bits & EXTENT_DELALLOC_NEW)) { | |
2222 | spin_lock(&inode->lock); | |
2223 | ASSERT(inode->new_delalloc_bytes >= len); | |
2224 | inode->new_delalloc_bytes -= len; | |
2766ff61 FM |
2225 | if (*bits & EXTENT_ADD_INODE_BYTES) |
2226 | inode_add_bytes(&inode->vfs_inode, len); | |
a7e3b975 FM |
2227 | spin_unlock(&inode->lock); |
2228 | } | |
291d673e CM |
2229 | } |
2230 | ||
d352ac68 | 2231 | /* |
da12fe54 NB |
2232 | * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit |
2233 | * in a chunk's stripe. This function ensures that bios do not span a | |
2234 | * stripe/chunk | |
6f034ece | 2235 | * |
da12fe54 NB |
2236 | * @page - The page we are about to add to the bio |
2237 | * @size - size we want to add to the bio | |
2238 | * @bio - bio we want to ensure is smaller than a stripe | |
2239 | * @bio_flags - flags of the bio | |
2240 | * | |
2241 | * return 1 if page cannot be added to the bio | |
2242 | * return 0 if page can be added to the bio | |
6f034ece | 2243 | * return error otherwise |
d352ac68 | 2244 | */ |
da12fe54 NB |
2245 | int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio, |
2246 | unsigned long bio_flags) | |
239b14b3 | 2247 | { |
0b246afa JM |
2248 | struct inode *inode = page->mapping->host; |
2249 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
1201b58b | 2250 | u64 logical = bio->bi_iter.bi_sector << 9; |
1a0b5c4d | 2251 | u32 bio_len = bio->bi_iter.bi_size; |
42034313 | 2252 | struct extent_map *em; |
42034313 | 2253 | int ret = 0; |
89b798ad | 2254 | struct btrfs_io_geometry geom; |
239b14b3 | 2255 | |
771ed689 CM |
2256 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
2257 | return 0; | |
2258 | ||
1a0b5c4d | 2259 | em = btrfs_get_chunk_map(fs_info, logical, fs_info->sectorsize); |
42034313 MR |
2260 | if (IS_ERR(em)) |
2261 | return PTR_ERR(em); | |
43c0d1a5 | 2262 | ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio), logical, &geom); |
6f034ece | 2263 | if (ret < 0) |
42034313 | 2264 | goto out; |
89b798ad | 2265 | |
1a0b5c4d | 2266 | if (geom.len < bio_len + size) |
42034313 MR |
2267 | ret = 1; |
2268 | out: | |
2269 | free_extent_map(em); | |
2270 | return ret; | |
239b14b3 CM |
2271 | } |
2272 | ||
d352ac68 CM |
2273 | /* |
2274 | * in order to insert checksums into the metadata in large chunks, | |
2275 | * we wait until bio submission time. All the pages in the bio are | |
2276 | * checksummed and sums are attached onto the ordered extent record. | |
2277 | * | |
2278 | * At IO completion time the cums attached on the ordered extent record | |
2279 | * are inserted into the btree | |
2280 | */ | |
8896a08d | 2281 | static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio, |
1941b64b | 2282 | u64 dio_file_offset) |
065631f6 | 2283 | { |
c965d640 | 2284 | return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0); |
4a69a410 | 2285 | } |
e015640f | 2286 | |
abb99cfd NA |
2287 | /* |
2288 | * Split an extent_map at [start, start + len] | |
2289 | * | |
2290 | * This function is intended to be used only for extract_ordered_extent(). | |
2291 | */ | |
2292 | static int split_zoned_em(struct btrfs_inode *inode, u64 start, u64 len, | |
2293 | u64 pre, u64 post) | |
2294 | { | |
2295 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
2296 | struct extent_map *em; | |
2297 | struct extent_map *split_pre = NULL; | |
2298 | struct extent_map *split_mid = NULL; | |
2299 | struct extent_map *split_post = NULL; | |
2300 | int ret = 0; | |
2301 | int modified; | |
2302 | unsigned long flags; | |
2303 | ||
2304 | /* Sanity check */ | |
2305 | if (pre == 0 && post == 0) | |
2306 | return 0; | |
2307 | ||
2308 | split_pre = alloc_extent_map(); | |
2309 | if (pre) | |
2310 | split_mid = alloc_extent_map(); | |
2311 | if (post) | |
2312 | split_post = alloc_extent_map(); | |
2313 | if (!split_pre || (pre && !split_mid) || (post && !split_post)) { | |
2314 | ret = -ENOMEM; | |
2315 | goto out; | |
2316 | } | |
2317 | ||
2318 | ASSERT(pre + post < len); | |
2319 | ||
2320 | lock_extent(&inode->io_tree, start, start + len - 1); | |
2321 | write_lock(&em_tree->lock); | |
2322 | em = lookup_extent_mapping(em_tree, start, len); | |
2323 | if (!em) { | |
2324 | ret = -EIO; | |
2325 | goto out_unlock; | |
2326 | } | |
2327 | ||
2328 | ASSERT(em->len == len); | |
2329 | ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)); | |
2330 | ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE); | |
2331 | ||
2332 | flags = em->flags; | |
2333 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); | |
2334 | clear_bit(EXTENT_FLAG_LOGGING, &flags); | |
2335 | modified = !list_empty(&em->list); | |
2336 | ||
2337 | /* First, replace the em with a new extent_map starting from * em->start */ | |
2338 | split_pre->start = em->start; | |
2339 | split_pre->len = (pre ? pre : em->len - post); | |
2340 | split_pre->orig_start = split_pre->start; | |
2341 | split_pre->block_start = em->block_start; | |
2342 | split_pre->block_len = split_pre->len; | |
2343 | split_pre->orig_block_len = split_pre->block_len; | |
2344 | split_pre->ram_bytes = split_pre->len; | |
2345 | split_pre->flags = flags; | |
2346 | split_pre->compress_type = em->compress_type; | |
2347 | split_pre->generation = em->generation; | |
2348 | ||
2349 | replace_extent_mapping(em_tree, em, split_pre, modified); | |
2350 | ||
2351 | /* | |
2352 | * Now we only have an extent_map at: | |
2353 | * [em->start, em->start + pre] if pre != 0 | |
2354 | * [em->start, em->start + em->len - post] if pre == 0 | |
2355 | */ | |
2356 | ||
2357 | if (pre) { | |
2358 | /* Insert the middle extent_map */ | |
2359 | split_mid->start = em->start + pre; | |
2360 | split_mid->len = em->len - pre - post; | |
2361 | split_mid->orig_start = split_mid->start; | |
2362 | split_mid->block_start = em->block_start + pre; | |
2363 | split_mid->block_len = split_mid->len; | |
2364 | split_mid->orig_block_len = split_mid->block_len; | |
2365 | split_mid->ram_bytes = split_mid->len; | |
2366 | split_mid->flags = flags; | |
2367 | split_mid->compress_type = em->compress_type; | |
2368 | split_mid->generation = em->generation; | |
2369 | add_extent_mapping(em_tree, split_mid, modified); | |
2370 | } | |
2371 | ||
2372 | if (post) { | |
2373 | split_post->start = em->start + em->len - post; | |
2374 | split_post->len = post; | |
2375 | split_post->orig_start = split_post->start; | |
2376 | split_post->block_start = em->block_start + em->len - post; | |
2377 | split_post->block_len = split_post->len; | |
2378 | split_post->orig_block_len = split_post->block_len; | |
2379 | split_post->ram_bytes = split_post->len; | |
2380 | split_post->flags = flags; | |
2381 | split_post->compress_type = em->compress_type; | |
2382 | split_post->generation = em->generation; | |
2383 | add_extent_mapping(em_tree, split_post, modified); | |
2384 | } | |
2385 | ||
2386 | /* Once for us */ | |
2387 | free_extent_map(em); | |
2388 | /* Once for the tree */ | |
2389 | free_extent_map(em); | |
2390 | ||
2391 | out_unlock: | |
2392 | write_unlock(&em_tree->lock); | |
2393 | unlock_extent(&inode->io_tree, start, start + len - 1); | |
2394 | out: | |
2395 | free_extent_map(split_pre); | |
2396 | free_extent_map(split_mid); | |
2397 | free_extent_map(split_post); | |
2398 | ||
2399 | return ret; | |
2400 | } | |
2401 | ||
d22002fd NA |
2402 | static blk_status_t extract_ordered_extent(struct btrfs_inode *inode, |
2403 | struct bio *bio, loff_t file_offset) | |
2404 | { | |
2405 | struct btrfs_ordered_extent *ordered; | |
d22002fd | 2406 | u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT; |
abb99cfd | 2407 | u64 file_len; |
d22002fd NA |
2408 | u64 len = bio->bi_iter.bi_size; |
2409 | u64 end = start + len; | |
2410 | u64 ordered_end; | |
2411 | u64 pre, post; | |
2412 | int ret = 0; | |
2413 | ||
2414 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); | |
2415 | if (WARN_ON_ONCE(!ordered)) | |
2416 | return BLK_STS_IOERR; | |
2417 | ||
2418 | /* No need to split */ | |
2419 | if (ordered->disk_num_bytes == len) | |
2420 | goto out; | |
2421 | ||
2422 | /* We cannot split once end_bio'd ordered extent */ | |
2423 | if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) { | |
2424 | ret = -EINVAL; | |
2425 | goto out; | |
2426 | } | |
2427 | ||
2428 | /* We cannot split a compressed ordered extent */ | |
2429 | if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) { | |
2430 | ret = -EINVAL; | |
2431 | goto out; | |
2432 | } | |
2433 | ||
2434 | ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes; | |
2435 | /* bio must be in one ordered extent */ | |
2436 | if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) { | |
2437 | ret = -EINVAL; | |
2438 | goto out; | |
2439 | } | |
2440 | ||
2441 | /* Checksum list should be empty */ | |
2442 | if (WARN_ON_ONCE(!list_empty(&ordered->list))) { | |
2443 | ret = -EINVAL; | |
2444 | goto out; | |
2445 | } | |
2446 | ||
abb99cfd | 2447 | file_len = ordered->num_bytes; |
d22002fd NA |
2448 | pre = start - ordered->disk_bytenr; |
2449 | post = ordered_end - end; | |
2450 | ||
2451 | ret = btrfs_split_ordered_extent(ordered, pre, post); | |
2452 | if (ret) | |
2453 | goto out; | |
abb99cfd | 2454 | ret = split_zoned_em(inode, file_offset, file_len, pre, post); |
d22002fd NA |
2455 | |
2456 | out: | |
d22002fd NA |
2457 | btrfs_put_ordered_extent(ordered); |
2458 | ||
2459 | return errno_to_blk_status(ret); | |
2460 | } | |
2461 | ||
d352ac68 | 2462 | /* |
cad321ad | 2463 | * extent_io.c submission hook. This does the right thing for csum calculation |
4c274bc6 LB |
2464 | * on write, or reading the csums from the tree before a read. |
2465 | * | |
2466 | * Rules about async/sync submit, | |
2467 | * a) read: sync submit | |
2468 | * | |
2469 | * b) write without checksum: sync submit | |
2470 | * | |
2471 | * c) write with checksum: | |
2472 | * c-1) if bio is issued by fsync: sync submit | |
2473 | * (sync_writers != 0) | |
2474 | * | |
2475 | * c-2) if root is reloc root: sync submit | |
2476 | * (only in case of buffered IO) | |
2477 | * | |
2478 | * c-3) otherwise: async submit | |
d352ac68 | 2479 | */ |
908930f3 NB |
2480 | blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio, |
2481 | int mirror_num, unsigned long bio_flags) | |
50489a57 | 2482 | |
44b8bd7e | 2483 | { |
0b246afa | 2484 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
44b8bd7e | 2485 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0d51e28a | 2486 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
4e4cbee9 | 2487 | blk_status_t ret = 0; |
19b9bdb0 | 2488 | int skip_sum; |
b812ce28 | 2489 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 2490 | |
42437a63 JB |
2491 | skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) || |
2492 | !fs_info->csum_root; | |
cad321ad | 2493 | |
70ddc553 | 2494 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) |
0d51e28a | 2495 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 2496 | |
d22002fd NA |
2497 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) { |
2498 | struct page *page = bio_first_bvec_all(bio)->bv_page; | |
2499 | loff_t file_offset = page_offset(page); | |
2500 | ||
2501 | ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset); | |
2502 | if (ret) | |
2503 | goto out; | |
2504 | } | |
2505 | ||
cfe94440 | 2506 | if (btrfs_op(bio) != BTRFS_MAP_WRITE) { |
0b246afa | 2507 | ret = btrfs_bio_wq_end_io(fs_info, bio, metadata); |
5fd02043 | 2508 | if (ret) |
61891923 | 2509 | goto out; |
5fd02043 | 2510 | |
d20f7043 | 2511 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
2512 | ret = btrfs_submit_compressed_read(inode, bio, |
2513 | mirror_num, | |
2514 | bio_flags); | |
2515 | goto out; | |
334c16d8 JB |
2516 | } else { |
2517 | /* | |
2518 | * Lookup bio sums does extra checks around whether we | |
2519 | * need to csum or not, which is why we ignore skip_sum | |
2520 | * here. | |
2521 | */ | |
6275193e | 2522 | ret = btrfs_lookup_bio_sums(inode, bio, NULL); |
c2db1073 | 2523 | if (ret) |
61891923 | 2524 | goto out; |
c2db1073 | 2525 | } |
4d1b5fb4 | 2526 | goto mapit; |
b812ce28 | 2527 | } else if (async && !skip_sum) { |
17d217fe YZ |
2528 | /* csum items have already been cloned */ |
2529 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
2530 | goto mapit; | |
19b9bdb0 | 2531 | /* we're doing a write, do the async checksumming */ |
8896a08d QW |
2532 | ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags, |
2533 | 0, btrfs_submit_bio_start); | |
61891923 | 2534 | goto out; |
b812ce28 | 2535 | } else if (!skip_sum) { |
bd242a08 | 2536 | ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0); |
b812ce28 JB |
2537 | if (ret) |
2538 | goto out; | |
19b9bdb0 CM |
2539 | } |
2540 | ||
0b86a832 | 2541 | mapit: |
08635bae | 2542 | ret = btrfs_map_bio(fs_info, bio, mirror_num); |
61891923 SB |
2543 | |
2544 | out: | |
4e4cbee9 CH |
2545 | if (ret) { |
2546 | bio->bi_status = ret; | |
4246a0b6 CH |
2547 | bio_endio(bio); |
2548 | } | |
61891923 | 2549 | return ret; |
065631f6 | 2550 | } |
6885f308 | 2551 | |
d352ac68 CM |
2552 | /* |
2553 | * given a list of ordered sums record them in the inode. This happens | |
2554 | * at IO completion time based on sums calculated at bio submission time. | |
2555 | */ | |
510f85ed NB |
2556 | static int add_pending_csums(struct btrfs_trans_handle *trans, |
2557 | struct list_head *list) | |
e6dcd2dc | 2558 | { |
e6dcd2dc | 2559 | struct btrfs_ordered_sum *sum; |
ac01f26a | 2560 | int ret; |
e6dcd2dc | 2561 | |
c6e30871 | 2562 | list_for_each_entry(sum, list, list) { |
7c2871a2 | 2563 | trans->adding_csums = true; |
510f85ed | 2564 | ret = btrfs_csum_file_blocks(trans, trans->fs_info->csum_root, sum); |
7c2871a2 | 2565 | trans->adding_csums = false; |
ac01f26a NB |
2566 | if (ret) |
2567 | return ret; | |
e6dcd2dc CM |
2568 | } |
2569 | return 0; | |
2570 | } | |
2571 | ||
c3347309 FM |
2572 | static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode, |
2573 | const u64 start, | |
2574 | const u64 len, | |
2575 | struct extent_state **cached_state) | |
2576 | { | |
2577 | u64 search_start = start; | |
2578 | const u64 end = start + len - 1; | |
2579 | ||
2580 | while (search_start < end) { | |
2581 | const u64 search_len = end - search_start + 1; | |
2582 | struct extent_map *em; | |
2583 | u64 em_len; | |
2584 | int ret = 0; | |
2585 | ||
2586 | em = btrfs_get_extent(inode, NULL, 0, search_start, search_len); | |
2587 | if (IS_ERR(em)) | |
2588 | return PTR_ERR(em); | |
2589 | ||
2590 | if (em->block_start != EXTENT_MAP_HOLE) | |
2591 | goto next; | |
2592 | ||
2593 | em_len = em->len; | |
2594 | if (em->start < search_start) | |
2595 | em_len -= search_start - em->start; | |
2596 | if (em_len > search_len) | |
2597 | em_len = search_len; | |
2598 | ||
2599 | ret = set_extent_bit(&inode->io_tree, search_start, | |
2600 | search_start + em_len - 1, | |
1cab5e72 NB |
2601 | EXTENT_DELALLOC_NEW, 0, NULL, cached_state, |
2602 | GFP_NOFS, NULL); | |
c3347309 FM |
2603 | next: |
2604 | search_start = extent_map_end(em); | |
2605 | free_extent_map(em); | |
2606 | if (ret) | |
2607 | return ret; | |
2608 | } | |
2609 | return 0; | |
2610 | } | |
2611 | ||
c2566f22 | 2612 | int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end, |
e3b8a485 | 2613 | unsigned int extra_bits, |
330a5827 | 2614 | struct extent_state **cached_state) |
ea8c2819 | 2615 | { |
fdb1e121 | 2616 | WARN_ON(PAGE_ALIGNED(end)); |
c3347309 FM |
2617 | |
2618 | if (start >= i_size_read(&inode->vfs_inode) && | |
2619 | !(inode->flags & BTRFS_INODE_PREALLOC)) { | |
2620 | /* | |
2621 | * There can't be any extents following eof in this case so just | |
2622 | * set the delalloc new bit for the range directly. | |
2623 | */ | |
2624 | extra_bits |= EXTENT_DELALLOC_NEW; | |
2625 | } else { | |
2626 | int ret; | |
2627 | ||
2628 | ret = btrfs_find_new_delalloc_bytes(inode, start, | |
2629 | end + 1 - start, | |
2630 | cached_state); | |
2631 | if (ret) | |
2632 | return ret; | |
2633 | } | |
2634 | ||
c2566f22 NB |
2635 | return set_extent_delalloc(&inode->io_tree, start, end, extra_bits, |
2636 | cached_state); | |
ea8c2819 CM |
2637 | } |
2638 | ||
d352ac68 | 2639 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
2640 | struct btrfs_writepage_fixup { |
2641 | struct page *page; | |
f4b1363c | 2642 | struct inode *inode; |
247e743c CM |
2643 | struct btrfs_work work; |
2644 | }; | |
2645 | ||
b2950863 | 2646 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
2647 | { |
2648 | struct btrfs_writepage_fixup *fixup; | |
2649 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 2650 | struct extent_state *cached_state = NULL; |
364ecf36 | 2651 | struct extent_changeset *data_reserved = NULL; |
247e743c | 2652 | struct page *page; |
65d87f79 | 2653 | struct btrfs_inode *inode; |
247e743c CM |
2654 | u64 page_start; |
2655 | u64 page_end; | |
25f3c502 | 2656 | int ret = 0; |
f4b1363c | 2657 | bool free_delalloc_space = true; |
247e743c CM |
2658 | |
2659 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2660 | page = fixup->page; | |
65d87f79 | 2661 | inode = BTRFS_I(fixup->inode); |
f4b1363c JB |
2662 | page_start = page_offset(page); |
2663 | page_end = page_offset(page) + PAGE_SIZE - 1; | |
2664 | ||
2665 | /* | |
2666 | * This is similar to page_mkwrite, we need to reserve the space before | |
2667 | * we take the page lock. | |
2668 | */ | |
65d87f79 NB |
2669 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
2670 | PAGE_SIZE); | |
4a096752 | 2671 | again: |
247e743c | 2672 | lock_page(page); |
25f3c502 CM |
2673 | |
2674 | /* | |
2675 | * Before we queued this fixup, we took a reference on the page. | |
2676 | * page->mapping may go NULL, but it shouldn't be moved to a different | |
2677 | * address space. | |
2678 | */ | |
f4b1363c JB |
2679 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { |
2680 | /* | |
2681 | * Unfortunately this is a little tricky, either | |
2682 | * | |
2683 | * 1) We got here and our page had already been dealt with and | |
2684 | * we reserved our space, thus ret == 0, so we need to just | |
2685 | * drop our space reservation and bail. This can happen the | |
2686 | * first time we come into the fixup worker, or could happen | |
2687 | * while waiting for the ordered extent. | |
2688 | * 2) Our page was already dealt with, but we happened to get an | |
2689 | * ENOSPC above from the btrfs_delalloc_reserve_space. In | |
2690 | * this case we obviously don't have anything to release, but | |
2691 | * because the page was already dealt with we don't want to | |
2692 | * mark the page with an error, so make sure we're resetting | |
2693 | * ret to 0. This is why we have this check _before_ the ret | |
2694 | * check, because we do not want to have a surprise ENOSPC | |
2695 | * when the page was already properly dealt with. | |
2696 | */ | |
2697 | if (!ret) { | |
65d87f79 NB |
2698 | btrfs_delalloc_release_extents(inode, PAGE_SIZE); |
2699 | btrfs_delalloc_release_space(inode, data_reserved, | |
f4b1363c JB |
2700 | page_start, PAGE_SIZE, |
2701 | true); | |
2702 | } | |
2703 | ret = 0; | |
247e743c | 2704 | goto out_page; |
f4b1363c | 2705 | } |
247e743c | 2706 | |
25f3c502 | 2707 | /* |
f4b1363c JB |
2708 | * We can't mess with the page state unless it is locked, so now that |
2709 | * it is locked bail if we failed to make our space reservation. | |
25f3c502 | 2710 | */ |
f4b1363c JB |
2711 | if (ret) |
2712 | goto out_page; | |
247e743c | 2713 | |
65d87f79 | 2714 | lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state); |
4a096752 CM |
2715 | |
2716 | /* already ordered? We're done */ | |
f57ad937 | 2717 | if (PageOrdered(page)) |
f4b1363c | 2718 | goto out_reserved; |
4a096752 | 2719 | |
65d87f79 | 2720 | ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE); |
4a096752 | 2721 | if (ordered) { |
65d87f79 NB |
2722 | unlock_extent_cached(&inode->io_tree, page_start, page_end, |
2723 | &cached_state); | |
4a096752 | 2724 | unlock_page(page); |
c0a43603 | 2725 | btrfs_start_ordered_extent(ordered, 1); |
87826df0 | 2726 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2727 | goto again; |
2728 | } | |
247e743c | 2729 | |
65d87f79 | 2730 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, |
330a5827 | 2731 | &cached_state); |
25f3c502 | 2732 | if (ret) |
53687007 | 2733 | goto out_reserved; |
f3038ee3 | 2734 | |
25f3c502 CM |
2735 | /* |
2736 | * Everything went as planned, we're now the owner of a dirty page with | |
2737 | * delayed allocation bits set and space reserved for our COW | |
2738 | * destination. | |
2739 | * | |
2740 | * The page was dirty when we started, nothing should have cleaned it. | |
2741 | */ | |
2742 | BUG_ON(!PageDirty(page)); | |
f4b1363c | 2743 | free_delalloc_space = false; |
53687007 | 2744 | out_reserved: |
65d87f79 | 2745 | btrfs_delalloc_release_extents(inode, PAGE_SIZE); |
f4b1363c | 2746 | if (free_delalloc_space) |
65d87f79 NB |
2747 | btrfs_delalloc_release_space(inode, data_reserved, page_start, |
2748 | PAGE_SIZE, true); | |
2749 | unlock_extent_cached(&inode->io_tree, page_start, page_end, | |
e43bbe5e | 2750 | &cached_state); |
247e743c | 2751 | out_page: |
25f3c502 CM |
2752 | if (ret) { |
2753 | /* | |
2754 | * We hit ENOSPC or other errors. Update the mapping and page | |
2755 | * to reflect the errors and clean the page. | |
2756 | */ | |
2757 | mapping_set_error(page->mapping, ret); | |
2758 | end_extent_writepage(page, ret, page_start, page_end); | |
2759 | clear_page_dirty_for_io(page); | |
2760 | SetPageError(page); | |
2761 | } | |
2762 | ClearPageChecked(page); | |
247e743c | 2763 | unlock_page(page); |
09cbfeaf | 2764 | put_page(page); |
b897abec | 2765 | kfree(fixup); |
364ecf36 | 2766 | extent_changeset_free(data_reserved); |
f4b1363c JB |
2767 | /* |
2768 | * As a precaution, do a delayed iput in case it would be the last iput | |
2769 | * that could need flushing space. Recursing back to fixup worker would | |
2770 | * deadlock. | |
2771 | */ | |
65d87f79 | 2772 | btrfs_add_delayed_iput(&inode->vfs_inode); |
247e743c CM |
2773 | } |
2774 | ||
2775 | /* | |
2776 | * There are a few paths in the higher layers of the kernel that directly | |
2777 | * set the page dirty bit without asking the filesystem if it is a | |
2778 | * good idea. This causes problems because we want to make sure COW | |
2779 | * properly happens and the data=ordered rules are followed. | |
2780 | * | |
c8b97818 | 2781 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2782 | * hasn't been properly setup for IO. We kick off an async process |
2783 | * to fix it up. The async helper will wait for ordered extents, set | |
2784 | * the delalloc bit and make it safe to write the page. | |
2785 | */ | |
a129ffb8 | 2786 | int btrfs_writepage_cow_fixup(struct page *page) |
247e743c CM |
2787 | { |
2788 | struct inode *inode = page->mapping->host; | |
0b246afa | 2789 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 2790 | struct btrfs_writepage_fixup *fixup; |
247e743c | 2791 | |
f57ad937 QW |
2792 | /* This page has ordered extent covering it already */ |
2793 | if (PageOrdered(page)) | |
247e743c CM |
2794 | return 0; |
2795 | ||
25f3c502 CM |
2796 | /* |
2797 | * PageChecked is set below when we create a fixup worker for this page, | |
2798 | * don't try to create another one if we're already PageChecked() | |
2799 | * | |
2800 | * The extent_io writepage code will redirty the page if we send back | |
2801 | * EAGAIN. | |
2802 | */ | |
247e743c CM |
2803 | if (PageChecked(page)) |
2804 | return -EAGAIN; | |
2805 | ||
2806 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2807 | if (!fixup) | |
2808 | return -EAGAIN; | |
f421950f | 2809 | |
f4b1363c JB |
2810 | /* |
2811 | * We are already holding a reference to this inode from | |
2812 | * write_cache_pages. We need to hold it because the space reservation | |
2813 | * takes place outside of the page lock, and we can't trust | |
2814 | * page->mapping outside of the page lock. | |
2815 | */ | |
2816 | ihold(inode); | |
247e743c | 2817 | SetPageChecked(page); |
09cbfeaf | 2818 | get_page(page); |
a0cac0ec | 2819 | btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL); |
247e743c | 2820 | fixup->page = page; |
f4b1363c | 2821 | fixup->inode = inode; |
0b246afa | 2822 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
25f3c502 CM |
2823 | |
2824 | return -EAGAIN; | |
247e743c CM |
2825 | } |
2826 | ||
d899e052 | 2827 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
c553f94d | 2828 | struct btrfs_inode *inode, u64 file_pos, |
9729f10a | 2829 | struct btrfs_file_extent_item *stack_fi, |
2766ff61 | 2830 | const bool update_inode_bytes, |
9729f10a | 2831 | u64 qgroup_reserved) |
d899e052 | 2832 | { |
c553f94d | 2833 | struct btrfs_root *root = inode->root; |
2766ff61 | 2834 | const u64 sectorsize = root->fs_info->sectorsize; |
d899e052 YZ |
2835 | struct btrfs_path *path; |
2836 | struct extent_buffer *leaf; | |
2837 | struct btrfs_key ins; | |
203f44c5 QW |
2838 | u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi); |
2839 | u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi); | |
2840 | u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi); | |
2841 | u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi); | |
5893dfb9 | 2842 | struct btrfs_drop_extents_args drop_args = { 0 }; |
d899e052 YZ |
2843 | int ret; |
2844 | ||
2845 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2846 | if (!path) |
2847 | return -ENOMEM; | |
d899e052 | 2848 | |
a1ed835e CM |
2849 | /* |
2850 | * we may be replacing one extent in the tree with another. | |
2851 | * The new extent is pinned in the extent map, and we don't want | |
2852 | * to drop it from the cache until it is completely in the btree. | |
2853 | * | |
2854 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2855 | * the caller is expected to unpin it and allow it to be merged | |
2856 | * with the others. | |
2857 | */ | |
5893dfb9 FM |
2858 | drop_args.path = path; |
2859 | drop_args.start = file_pos; | |
2860 | drop_args.end = file_pos + num_bytes; | |
2861 | drop_args.replace_extent = true; | |
2862 | drop_args.extent_item_size = sizeof(*stack_fi); | |
2863 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); | |
79787eaa JM |
2864 | if (ret) |
2865 | goto out; | |
d899e052 | 2866 | |
5893dfb9 | 2867 | if (!drop_args.extent_inserted) { |
c553f94d | 2868 | ins.objectid = btrfs_ino(inode); |
1acae57b FDBM |
2869 | ins.offset = file_pos; |
2870 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2871 | ||
1acae57b | 2872 | ret = btrfs_insert_empty_item(trans, root, path, &ins, |
203f44c5 | 2873 | sizeof(*stack_fi)); |
1acae57b FDBM |
2874 | if (ret) |
2875 | goto out; | |
2876 | } | |
d899e052 | 2877 | leaf = path->nodes[0]; |
203f44c5 QW |
2878 | btrfs_set_stack_file_extent_generation(stack_fi, trans->transid); |
2879 | write_extent_buffer(leaf, stack_fi, | |
2880 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
2881 | sizeof(struct btrfs_file_extent_item)); | |
b9473439 | 2882 | |
d899e052 | 2883 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2884 | btrfs_release_path(path); |
d899e052 | 2885 | |
2766ff61 FM |
2886 | /* |
2887 | * If we dropped an inline extent here, we know the range where it is | |
2888 | * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the | |
1a9fd417 | 2889 | * number of bytes only for that range containing the inline extent. |
2766ff61 FM |
2890 | * The remaining of the range will be processed when clearning the |
2891 | * EXTENT_DELALLOC_BIT bit through the ordered extent completion. | |
2892 | */ | |
2893 | if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) { | |
2894 | u64 inline_size = round_down(drop_args.bytes_found, sectorsize); | |
2895 | ||
2896 | inline_size = drop_args.bytes_found - inline_size; | |
2897 | btrfs_update_inode_bytes(inode, sectorsize, inline_size); | |
2898 | drop_args.bytes_found -= inline_size; | |
2899 | num_bytes -= sectorsize; | |
2900 | } | |
2901 | ||
2902 | if (update_inode_bytes) | |
2903 | btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found); | |
d899e052 YZ |
2904 | |
2905 | ins.objectid = disk_bytenr; | |
2906 | ins.offset = disk_num_bytes; | |
2907 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
a12b877b | 2908 | |
c553f94d | 2909 | ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes); |
9ddc959e JB |
2910 | if (ret) |
2911 | goto out; | |
2912 | ||
c553f94d | 2913 | ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode), |
9729f10a | 2914 | file_pos, qgroup_reserved, &ins); |
79787eaa | 2915 | out: |
d899e052 | 2916 | btrfs_free_path(path); |
b9473439 | 2917 | |
79787eaa | 2918 | return ret; |
d899e052 YZ |
2919 | } |
2920 | ||
2ff7e61e | 2921 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
2922 | u64 start, u64 len) |
2923 | { | |
32da5386 | 2924 | struct btrfs_block_group *cache; |
e570fd27 | 2925 | |
0b246afa | 2926 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
2927 | ASSERT(cache); |
2928 | ||
2929 | spin_lock(&cache->lock); | |
2930 | cache->delalloc_bytes -= len; | |
2931 | spin_unlock(&cache->lock); | |
2932 | ||
2933 | btrfs_put_block_group(cache); | |
2934 | } | |
2935 | ||
203f44c5 | 2936 | static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans, |
203f44c5 QW |
2937 | struct btrfs_ordered_extent *oe) |
2938 | { | |
2939 | struct btrfs_file_extent_item stack_fi; | |
2940 | u64 logical_len; | |
2766ff61 | 2941 | bool update_inode_bytes; |
203f44c5 QW |
2942 | |
2943 | memset(&stack_fi, 0, sizeof(stack_fi)); | |
2944 | btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG); | |
2945 | btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr); | |
2946 | btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, | |
2947 | oe->disk_num_bytes); | |
2948 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) | |
2949 | logical_len = oe->truncated_len; | |
2950 | else | |
2951 | logical_len = oe->num_bytes; | |
2952 | btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len); | |
2953 | btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len); | |
2954 | btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type); | |
2955 | /* Encryption and other encoding is reserved and all 0 */ | |
2956 | ||
2766ff61 FM |
2957 | /* |
2958 | * For delalloc, when completing an ordered extent we update the inode's | |
2959 | * bytes when clearing the range in the inode's io tree, so pass false | |
2960 | * as the argument 'update_inode_bytes' to insert_reserved_file_extent(), | |
2961 | * except if the ordered extent was truncated. | |
2962 | */ | |
2963 | update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) || | |
2964 | test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags); | |
2965 | ||
3c38c877 NB |
2966 | return insert_reserved_file_extent(trans, BTRFS_I(oe->inode), |
2967 | oe->file_offset, &stack_fi, | |
2766ff61 | 2968 | update_inode_bytes, oe->qgroup_rsv); |
203f44c5 QW |
2969 | } |
2970 | ||
2971 | /* | |
2972 | * As ordered data IO finishes, this gets called so we can finish | |
d352ac68 CM |
2973 | * an ordered extent if the range of bytes in the file it covers are |
2974 | * fully written. | |
2975 | */ | |
5fd02043 | 2976 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2977 | { |
72e7e6ed NB |
2978 | struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode); |
2979 | struct btrfs_root *root = inode->root; | |
2980 | struct btrfs_fs_info *fs_info = root->fs_info; | |
0ca1f7ce | 2981 | struct btrfs_trans_handle *trans = NULL; |
72e7e6ed | 2982 | struct extent_io_tree *io_tree = &inode->io_tree; |
2ac55d41 | 2983 | struct extent_state *cached_state = NULL; |
bffe633e | 2984 | u64 start, end; |
261507a0 | 2985 | int compress_type = 0; |
77cef2ec | 2986 | int ret = 0; |
bffe633e | 2987 | u64 logical_len = ordered_extent->num_bytes; |
8d510121 | 2988 | bool freespace_inode; |
77cef2ec | 2989 | bool truncated = false; |
49940bdd | 2990 | bool clear_reserved_extent = true; |
2766ff61 | 2991 | unsigned int clear_bits = EXTENT_DEFRAG; |
a7e3b975 | 2992 | |
bffe633e OS |
2993 | start = ordered_extent->file_offset; |
2994 | end = start + ordered_extent->num_bytes - 1; | |
2995 | ||
a7e3b975 FM |
2996 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
2997 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) && | |
2998 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags)) | |
2766ff61 | 2999 | clear_bits |= EXTENT_DELALLOC_NEW; |
e6dcd2dc | 3000 | |
72e7e6ed | 3001 | freespace_inode = btrfs_is_free_space_inode(inode); |
0cb59c99 | 3002 | |
5fd02043 JB |
3003 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
3004 | ret = -EIO; | |
3005 | goto out; | |
3006 | } | |
3007 | ||
c7c3a6dc | 3008 | if (ordered_extent->bdev) |
d8e3fb10 NA |
3009 | btrfs_rewrite_logical_zoned(ordered_extent); |
3010 | ||
72e7e6ed | 3011 | btrfs_free_io_failure_record(inode, start, end); |
f612496b | 3012 | |
77cef2ec JB |
3013 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
3014 | truncated = true; | |
3015 | logical_len = ordered_extent->truncated_len; | |
3016 | /* Truncated the entire extent, don't bother adding */ | |
3017 | if (!logical_len) | |
3018 | goto out; | |
3019 | } | |
3020 | ||
c2167754 | 3021 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 3022 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a | 3023 | |
72e7e6ed | 3024 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
8d510121 NB |
3025 | if (freespace_inode) |
3026 | trans = btrfs_join_transaction_spacecache(root); | |
6c760c07 JB |
3027 | else |
3028 | trans = btrfs_join_transaction(root); | |
3029 | if (IS_ERR(trans)) { | |
3030 | ret = PTR_ERR(trans); | |
3031 | trans = NULL; | |
3032 | goto out; | |
c2167754 | 3033 | } |
72e7e6ed | 3034 | trans->block_rsv = &inode->block_rsv; |
729f7961 | 3035 | ret = btrfs_update_inode_fallback(trans, root, inode); |
6c760c07 | 3036 | if (ret) /* -ENOMEM or corruption */ |
66642832 | 3037 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
3038 | goto out; |
3039 | } | |
e6dcd2dc | 3040 | |
2766ff61 | 3041 | clear_bits |= EXTENT_LOCKED; |
bffe633e | 3042 | lock_extent_bits(io_tree, start, end, &cached_state); |
e6dcd2dc | 3043 | |
8d510121 NB |
3044 | if (freespace_inode) |
3045 | trans = btrfs_join_transaction_spacecache(root); | |
0cb59c99 | 3046 | else |
7a7eaa40 | 3047 | trans = btrfs_join_transaction(root); |
79787eaa JM |
3048 | if (IS_ERR(trans)) { |
3049 | ret = PTR_ERR(trans); | |
3050 | trans = NULL; | |
a7e3b975 | 3051 | goto out; |
79787eaa | 3052 | } |
a79b7d4b | 3053 | |
72e7e6ed | 3054 | trans->block_rsv = &inode->block_rsv; |
c2167754 | 3055 | |
c8b97818 | 3056 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 3057 | compress_type = ordered_extent->compress_type; |
d899e052 | 3058 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 3059 | BUG_ON(compress_type); |
72e7e6ed | 3060 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
3061 | ordered_extent->file_offset, |
3062 | ordered_extent->file_offset + | |
77cef2ec | 3063 | logical_len); |
d899e052 | 3064 | } else { |
0b246afa | 3065 | BUG_ON(root == fs_info->tree_root); |
3c38c877 | 3066 | ret = insert_ordered_extent_file_extent(trans, ordered_extent); |
49940bdd JB |
3067 | if (!ret) { |
3068 | clear_reserved_extent = false; | |
2ff7e61e | 3069 | btrfs_release_delalloc_bytes(fs_info, |
bffe633e OS |
3070 | ordered_extent->disk_bytenr, |
3071 | ordered_extent->disk_num_bytes); | |
49940bdd | 3072 | } |
d899e052 | 3073 | } |
72e7e6ed | 3074 | unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset, |
bffe633e | 3075 | ordered_extent->num_bytes, trans->transid); |
79787eaa | 3076 | if (ret < 0) { |
66642832 | 3077 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3078 | goto out; |
79787eaa | 3079 | } |
2ac55d41 | 3080 | |
510f85ed | 3081 | ret = add_pending_csums(trans, &ordered_extent->list); |
ac01f26a NB |
3082 | if (ret) { |
3083 | btrfs_abort_transaction(trans, ret); | |
3084 | goto out; | |
3085 | } | |
e6dcd2dc | 3086 | |
2766ff61 FM |
3087 | /* |
3088 | * If this is a new delalloc range, clear its new delalloc flag to | |
3089 | * update the inode's number of bytes. This needs to be done first | |
3090 | * before updating the inode item. | |
3091 | */ | |
3092 | if ((clear_bits & EXTENT_DELALLOC_NEW) && | |
3093 | !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) | |
72e7e6ed | 3094 | clear_extent_bit(&inode->io_tree, start, end, |
2766ff61 FM |
3095 | EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES, |
3096 | 0, 0, &cached_state); | |
3097 | ||
72e7e6ed | 3098 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
729f7961 | 3099 | ret = btrfs_update_inode_fallback(trans, root, inode); |
6c760c07 | 3100 | if (ret) { /* -ENOMEM or corruption */ |
66642832 | 3101 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3102 | goto out; |
1ef30be1 JB |
3103 | } |
3104 | ret = 0; | |
c2167754 | 3105 | out: |
72e7e6ed | 3106 | clear_extent_bit(&inode->io_tree, start, end, clear_bits, |
bffe633e | 3107 | (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0, |
313facc5 | 3108 | &cached_state); |
a7e3b975 | 3109 | |
a698d075 | 3110 | if (trans) |
3a45bb20 | 3111 | btrfs_end_transaction(trans); |
0cb59c99 | 3112 | |
77cef2ec | 3113 | if (ret || truncated) { |
bffe633e | 3114 | u64 unwritten_start = start; |
77cef2ec | 3115 | |
d61bec08 JB |
3116 | /* |
3117 | * If we failed to finish this ordered extent for any reason we | |
3118 | * need to make sure BTRFS_ORDERED_IOERR is set on the ordered | |
3119 | * extent, and mark the inode with the error if it wasn't | |
3120 | * already set. Any error during writeback would have already | |
3121 | * set the mapping error, so we need to set it if we're the ones | |
3122 | * marking this ordered extent as failed. | |
3123 | */ | |
3124 | if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR, | |
3125 | &ordered_extent->flags)) | |
3126 | mapping_set_error(ordered_extent->inode->i_mapping, -EIO); | |
3127 | ||
77cef2ec | 3128 | if (truncated) |
bffe633e OS |
3129 | unwritten_start += logical_len; |
3130 | clear_extent_uptodate(io_tree, unwritten_start, end, NULL); | |
77cef2ec JB |
3131 | |
3132 | /* Drop the cache for the part of the extent we didn't write. */ | |
72e7e6ed | 3133 | btrfs_drop_extent_cache(inode, unwritten_start, end, 0); |
5fd02043 | 3134 | |
0bec9ef5 JB |
3135 | /* |
3136 | * If the ordered extent had an IOERR or something else went | |
3137 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
3138 | * back to the allocator. We only free the extent in the |
3139 | * truncated case if we didn't write out the extent at all. | |
49940bdd JB |
3140 | * |
3141 | * If we made it past insert_reserved_file_extent before we | |
3142 | * errored out then we don't need to do this as the accounting | |
3143 | * has already been done. | |
0bec9ef5 | 3144 | */ |
77cef2ec | 3145 | if ((ret || !logical_len) && |
49940bdd | 3146 | clear_reserved_extent && |
77cef2ec | 3147 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
4eaaec24 NB |
3148 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
3149 | /* | |
3150 | * Discard the range before returning it back to the | |
3151 | * free space pool | |
3152 | */ | |
46b27f50 | 3153 | if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC)) |
4eaaec24 | 3154 | btrfs_discard_extent(fs_info, |
bffe633e OS |
3155 | ordered_extent->disk_bytenr, |
3156 | ordered_extent->disk_num_bytes, | |
3157 | NULL); | |
2ff7e61e | 3158 | btrfs_free_reserved_extent(fs_info, |
bffe633e OS |
3159 | ordered_extent->disk_bytenr, |
3160 | ordered_extent->disk_num_bytes, 1); | |
4eaaec24 | 3161 | } |
0bec9ef5 JB |
3162 | } |
3163 | ||
5fd02043 | 3164 | /* |
8bad3c02 LB |
3165 | * This needs to be done to make sure anybody waiting knows we are done |
3166 | * updating everything for this ordered extent. | |
5fd02043 | 3167 | */ |
72e7e6ed | 3168 | btrfs_remove_ordered_extent(inode, ordered_extent); |
5fd02043 | 3169 | |
e6dcd2dc CM |
3170 | /* once for us */ |
3171 | btrfs_put_ordered_extent(ordered_extent); | |
3172 | /* once for the tree */ | |
3173 | btrfs_put_ordered_extent(ordered_extent); | |
3174 | ||
5fd02043 JB |
3175 | return ret; |
3176 | } | |
3177 | ||
3178 | static void finish_ordered_fn(struct btrfs_work *work) | |
3179 | { | |
3180 | struct btrfs_ordered_extent *ordered_extent; | |
3181 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
3182 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
3183 | } |
3184 | ||
38a39ac7 QW |
3185 | void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode, |
3186 | struct page *page, u64 start, | |
25c1252a | 3187 | u64 end, bool uptodate) |
211f90e6 | 3188 | { |
38a39ac7 | 3189 | trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate); |
1abe9b8a | 3190 | |
e65f152e QW |
3191 | btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start, |
3192 | finish_ordered_fn, uptodate); | |
211f90e6 CM |
3193 | } |
3194 | ||
265d4ac0 QW |
3195 | /* |
3196 | * check_data_csum - verify checksum of one sector of uncompressed data | |
7ffd27e3 | 3197 | * @inode: inode |
265d4ac0 | 3198 | * @io_bio: btrfs_io_bio which contains the csum |
7ffd27e3 | 3199 | * @bio_offset: offset to the beginning of the bio (in bytes) |
265d4ac0 QW |
3200 | * @page: page where is the data to be verified |
3201 | * @pgoff: offset inside the page | |
c1d6abda | 3202 | * @start: logical offset in the file |
265d4ac0 QW |
3203 | * |
3204 | * The length of such check is always one sector size. | |
3205 | */ | |
47df7765 | 3206 | static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio, |
c1d6abda OS |
3207 | u32 bio_offset, struct page *page, u32 pgoff, |
3208 | u64 start) | |
dc380aea | 3209 | { |
d5178578 JT |
3210 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
3211 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); | |
dc380aea | 3212 | char *kaddr; |
265d4ac0 | 3213 | u32 len = fs_info->sectorsize; |
223486c2 | 3214 | const u32 csum_size = fs_info->csum_size; |
7ffd27e3 | 3215 | unsigned int offset_sectors; |
d5178578 JT |
3216 | u8 *csum_expected; |
3217 | u8 csum[BTRFS_CSUM_SIZE]; | |
dc380aea | 3218 | |
265d4ac0 QW |
3219 | ASSERT(pgoff + len <= PAGE_SIZE); |
3220 | ||
7ffd27e3 QW |
3221 | offset_sectors = bio_offset >> fs_info->sectorsize_bits; |
3222 | csum_expected = ((u8 *)io_bio->csum) + offset_sectors * csum_size; | |
dc380aea MX |
3223 | |
3224 | kaddr = kmap_atomic(page); | |
d5178578 JT |
3225 | shash->tfm = fs_info->csum_shash; |
3226 | ||
fd08001f | 3227 | crypto_shash_digest(shash, kaddr + pgoff, len, csum); |
d5178578 JT |
3228 | |
3229 | if (memcmp(csum, csum_expected, csum_size)) | |
dc380aea MX |
3230 | goto zeroit; |
3231 | ||
3232 | kunmap_atomic(kaddr); | |
3233 | return 0; | |
3234 | zeroit: | |
c1d6abda OS |
3235 | btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected, |
3236 | io_bio->mirror_num); | |
814723e0 NB |
3237 | if (io_bio->device) |
3238 | btrfs_dev_stat_inc_and_print(io_bio->device, | |
3239 | BTRFS_DEV_STAT_CORRUPTION_ERRS); | |
dc380aea MX |
3240 | memset(kaddr + pgoff, 1, len); |
3241 | flush_dcache_page(page); | |
3242 | kunmap_atomic(kaddr); | |
dc380aea MX |
3243 | return -EIO; |
3244 | } | |
3245 | ||
d352ac68 | 3246 | /* |
7ffd27e3 | 3247 | * When reads are done, we need to check csums to verify the data is correct. |
4a54c8c1 JS |
3248 | * if there's a match, we allow the bio to finish. If not, the code in |
3249 | * extent_io.c will try to find good copies for us. | |
7ffd27e3 QW |
3250 | * |
3251 | * @bio_offset: offset to the beginning of the bio (in bytes) | |
3252 | * @start: file offset of the range start | |
3253 | * @end: file offset of the range end (inclusive) | |
08508fea QW |
3254 | * |
3255 | * Return a bitmap where bit set means a csum mismatch, and bit not set means | |
3256 | * csum match. | |
d352ac68 | 3257 | */ |
08508fea QW |
3258 | unsigned int btrfs_verify_data_csum(struct btrfs_io_bio *io_bio, u32 bio_offset, |
3259 | struct page *page, u64 start, u64 end) | |
07157aac | 3260 | { |
07157aac | 3261 | struct inode *inode = page->mapping->host; |
d1310b2e | 3262 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3263 | struct btrfs_root *root = BTRFS_I(inode)->root; |
f44cf410 QW |
3264 | const u32 sectorsize = root->fs_info->sectorsize; |
3265 | u32 pg_off; | |
08508fea | 3266 | unsigned int result = 0; |
d1310b2e | 3267 | |
d20f7043 CM |
3268 | if (PageChecked(page)) { |
3269 | ClearPageChecked(page); | |
dc380aea | 3270 | return 0; |
d20f7043 | 3271 | } |
6cbff00f | 3272 | |
3670e645 QW |
3273 | /* |
3274 | * For subpage case, above PageChecked is not safe as it's not subpage | |
3275 | * compatible. | |
3276 | * But for now only cow fixup and compressed read utilize PageChecked | |
3277 | * flag, while in this context we can easily use io_bio->csum to | |
3278 | * determine if we really need to do csum verification. | |
3279 | * | |
3280 | * So for now, just exit if io_bio->csum is NULL, as it means it's | |
3281 | * compressed read, and its compressed data csum has already been | |
3282 | * verified. | |
3283 | */ | |
3284 | if (io_bio->csum == NULL) | |
3285 | return 0; | |
3286 | ||
6cbff00f | 3287 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
dc380aea | 3288 | return 0; |
17d217fe | 3289 | |
42437a63 JB |
3290 | if (!root->fs_info->csum_root) |
3291 | return 0; | |
3292 | ||
17d217fe | 3293 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && |
9655d298 | 3294 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
91166212 | 3295 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM); |
b6cda9bc | 3296 | return 0; |
17d217fe | 3297 | } |
d20f7043 | 3298 | |
f44cf410 QW |
3299 | ASSERT(page_offset(page) <= start && |
3300 | end <= page_offset(page) + PAGE_SIZE - 1); | |
3301 | for (pg_off = offset_in_page(start); | |
3302 | pg_off < offset_in_page(end); | |
3303 | pg_off += sectorsize, bio_offset += sectorsize) { | |
3304 | int ret; | |
3305 | ||
c1d6abda OS |
3306 | ret = check_data_csum(inode, io_bio, bio_offset, page, pg_off, |
3307 | page_offset(page) + pg_off); | |
08508fea QW |
3308 | if (ret < 0) { |
3309 | const int nr_bit = (pg_off - offset_in_page(start)) >> | |
3310 | root->fs_info->sectorsize_bits; | |
3311 | ||
3312 | result |= (1U << nr_bit); | |
3313 | } | |
f44cf410 | 3314 | } |
08508fea | 3315 | return result; |
07157aac | 3316 | } |
b888db2b | 3317 | |
c1c3fac2 NB |
3318 | /* |
3319 | * btrfs_add_delayed_iput - perform a delayed iput on @inode | |
3320 | * | |
3321 | * @inode: The inode we want to perform iput on | |
3322 | * | |
3323 | * This function uses the generic vfs_inode::i_count to track whether we should | |
3324 | * just decrement it (in case it's > 1) or if this is the last iput then link | |
3325 | * the inode to the delayed iput machinery. Delayed iputs are processed at | |
3326 | * transaction commit time/superblock commit/cleaner kthread. | |
3327 | */ | |
24bbcf04 YZ |
3328 | void btrfs_add_delayed_iput(struct inode *inode) |
3329 | { | |
0b246afa | 3330 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8089fe62 | 3331 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3332 | |
3333 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3334 | return; | |
3335 | ||
034f784d | 3336 | atomic_inc(&fs_info->nr_delayed_iputs); |
24bbcf04 | 3337 | spin_lock(&fs_info->delayed_iput_lock); |
c1c3fac2 NB |
3338 | ASSERT(list_empty(&binode->delayed_iput)); |
3339 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
24bbcf04 | 3340 | spin_unlock(&fs_info->delayed_iput_lock); |
fd340d0f JB |
3341 | if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags)) |
3342 | wake_up_process(fs_info->cleaner_kthread); | |
24bbcf04 YZ |
3343 | } |
3344 | ||
63611e73 JB |
3345 | static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info, |
3346 | struct btrfs_inode *inode) | |
3347 | { | |
3348 | list_del_init(&inode->delayed_iput); | |
3349 | spin_unlock(&fs_info->delayed_iput_lock); | |
3350 | iput(&inode->vfs_inode); | |
3351 | if (atomic_dec_and_test(&fs_info->nr_delayed_iputs)) | |
3352 | wake_up(&fs_info->delayed_iputs_wait); | |
3353 | spin_lock(&fs_info->delayed_iput_lock); | |
3354 | } | |
3355 | ||
3356 | static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info, | |
3357 | struct btrfs_inode *inode) | |
3358 | { | |
3359 | if (!list_empty(&inode->delayed_iput)) { | |
3360 | spin_lock(&fs_info->delayed_iput_lock); | |
3361 | if (!list_empty(&inode->delayed_iput)) | |
3362 | run_delayed_iput_locked(fs_info, inode); | |
3363 | spin_unlock(&fs_info->delayed_iput_lock); | |
3364 | } | |
3365 | } | |
3366 | ||
2ff7e61e | 3367 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 3368 | { |
24bbcf04 | 3369 | |
24bbcf04 | 3370 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3371 | while (!list_empty(&fs_info->delayed_iputs)) { |
3372 | struct btrfs_inode *inode; | |
3373 | ||
3374 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3375 | struct btrfs_inode, delayed_iput); | |
63611e73 | 3376 | run_delayed_iput_locked(fs_info, inode); |
71795ee5 | 3377 | cond_resched_lock(&fs_info->delayed_iput_lock); |
24bbcf04 | 3378 | } |
8089fe62 | 3379 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3380 | } |
3381 | ||
034f784d | 3382 | /** |
2639631d NB |
3383 | * Wait for flushing all delayed iputs |
3384 | * | |
3385 | * @fs_info: the filesystem | |
034f784d JB |
3386 | * |
3387 | * This will wait on any delayed iputs that are currently running with KILLABLE | |
3388 | * set. Once they are all done running we will return, unless we are killed in | |
3389 | * which case we return EINTR. This helps in user operations like fallocate etc | |
3390 | * that might get blocked on the iputs. | |
2639631d NB |
3391 | * |
3392 | * Return EINTR if we were killed, 0 if nothing's pending | |
034f784d JB |
3393 | */ |
3394 | int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info) | |
3395 | { | |
3396 | int ret = wait_event_killable(fs_info->delayed_iputs_wait, | |
3397 | atomic_read(&fs_info->nr_delayed_iputs) == 0); | |
3398 | if (ret) | |
3399 | return -EINTR; | |
3400 | return 0; | |
3401 | } | |
3402 | ||
7b128766 | 3403 | /* |
f7e9e8fc OS |
3404 | * This creates an orphan entry for the given inode in case something goes wrong |
3405 | * in the middle of an unlink. | |
7b128766 | 3406 | */ |
73f2e545 | 3407 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, |
27919067 | 3408 | struct btrfs_inode *inode) |
7b128766 | 3409 | { |
d68fc57b | 3410 | int ret; |
7b128766 | 3411 | |
27919067 OS |
3412 | ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode)); |
3413 | if (ret && ret != -EEXIST) { | |
3414 | btrfs_abort_transaction(trans, ret); | |
3415 | return ret; | |
d68fc57b YZ |
3416 | } |
3417 | ||
d68fc57b | 3418 | return 0; |
7b128766 JB |
3419 | } |
3420 | ||
3421 | /* | |
f7e9e8fc OS |
3422 | * We have done the delete so we can go ahead and remove the orphan item for |
3423 | * this particular inode. | |
7b128766 | 3424 | */ |
48a3b636 | 3425 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3d6ae7bb | 3426 | struct btrfs_inode *inode) |
7b128766 | 3427 | { |
27919067 | 3428 | return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode)); |
7b128766 JB |
3429 | } |
3430 | ||
3431 | /* | |
3432 | * this cleans up any orphans that may be left on the list from the last use | |
3433 | * of this root. | |
3434 | */ | |
66b4ffd1 | 3435 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 3436 | { |
0b246afa | 3437 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
3438 | struct btrfs_path *path; |
3439 | struct extent_buffer *leaf; | |
7b128766 JB |
3440 | struct btrfs_key key, found_key; |
3441 | struct btrfs_trans_handle *trans; | |
3442 | struct inode *inode; | |
8f6d7f4f | 3443 | u64 last_objectid = 0; |
f7e9e8fc | 3444 | int ret = 0, nr_unlink = 0; |
7b128766 | 3445 | |
d68fc57b | 3446 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3447 | return 0; |
c71bf099 YZ |
3448 | |
3449 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3450 | if (!path) { |
3451 | ret = -ENOMEM; | |
3452 | goto out; | |
3453 | } | |
e4058b54 | 3454 | path->reada = READA_BACK; |
7b128766 JB |
3455 | |
3456 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3457 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3458 | key.offset = (u64)-1; |
3459 | ||
7b128766 JB |
3460 | while (1) { |
3461 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3462 | if (ret < 0) |
3463 | goto out; | |
7b128766 JB |
3464 | |
3465 | /* | |
3466 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3467 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3468 | * find the key and see if we have stuff that matches |
3469 | */ | |
3470 | if (ret > 0) { | |
66b4ffd1 | 3471 | ret = 0; |
7b128766 JB |
3472 | if (path->slots[0] == 0) |
3473 | break; | |
3474 | path->slots[0]--; | |
3475 | } | |
3476 | ||
3477 | /* pull out the item */ | |
3478 | leaf = path->nodes[0]; | |
7b128766 JB |
3479 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3480 | ||
3481 | /* make sure the item matches what we want */ | |
3482 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3483 | break; | |
962a298f | 3484 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3485 | break; |
3486 | ||
3487 | /* release the path since we're done with it */ | |
b3b4aa74 | 3488 | btrfs_release_path(path); |
7b128766 JB |
3489 | |
3490 | /* | |
3491 | * this is where we are basically btrfs_lookup, without the | |
3492 | * crossing root thing. we store the inode number in the | |
3493 | * offset of the orphan item. | |
3494 | */ | |
8f6d7f4f JB |
3495 | |
3496 | if (found_key.offset == last_objectid) { | |
0b246afa JM |
3497 | btrfs_err(fs_info, |
3498 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3499 | ret = -EINVAL; |
3500 | goto out; | |
3501 | } | |
3502 | ||
3503 | last_objectid = found_key.offset; | |
3504 | ||
5d4f98a2 YZ |
3505 | found_key.objectid = found_key.offset; |
3506 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3507 | found_key.offset = 0; | |
0202e83f | 3508 | inode = btrfs_iget(fs_info->sb, last_objectid, root); |
8c6ffba0 | 3509 | ret = PTR_ERR_OR_ZERO(inode); |
67710892 | 3510 | if (ret && ret != -ENOENT) |
66b4ffd1 | 3511 | goto out; |
7b128766 | 3512 | |
0b246afa | 3513 | if (ret == -ENOENT && root == fs_info->tree_root) { |
f8e9e0b0 | 3514 | struct btrfs_root *dead_root; |
f8e9e0b0 AJ |
3515 | int is_dead_root = 0; |
3516 | ||
3517 | /* | |
0c0218e9 | 3518 | * This is an orphan in the tree root. Currently these |
f8e9e0b0 | 3519 | * could come from 2 sources: |
0c0218e9 | 3520 | * a) a root (snapshot/subvolume) deletion in progress |
f8e9e0b0 | 3521 | * b) a free space cache inode |
0c0218e9 FM |
3522 | * We need to distinguish those two, as the orphan item |
3523 | * for a root must not get deleted before the deletion | |
3524 | * of the snapshot/subvolume's tree completes. | |
3525 | * | |
3526 | * btrfs_find_orphan_roots() ran before us, which has | |
3527 | * found all deleted roots and loaded them into | |
3528 | * fs_info->fs_roots_radix. So here we can find if an | |
3529 | * orphan item corresponds to a deleted root by looking | |
3530 | * up the root from that radix tree. | |
f8e9e0b0 | 3531 | */ |
a619b3c7 RK |
3532 | |
3533 | spin_lock(&fs_info->fs_roots_radix_lock); | |
3534 | dead_root = radix_tree_lookup(&fs_info->fs_roots_radix, | |
3535 | (unsigned long)found_key.objectid); | |
3536 | if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0) | |
3537 | is_dead_root = 1; | |
3538 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
3539 | ||
f8e9e0b0 AJ |
3540 | if (is_dead_root) { |
3541 | /* prevent this orphan from being found again */ | |
3542 | key.offset = found_key.objectid - 1; | |
3543 | continue; | |
3544 | } | |
f7e9e8fc | 3545 | |
f8e9e0b0 | 3546 | } |
f7e9e8fc | 3547 | |
7b128766 | 3548 | /* |
f7e9e8fc OS |
3549 | * If we have an inode with links, there are a couple of |
3550 | * possibilities. Old kernels (before v3.12) used to create an | |
3551 | * orphan item for truncate indicating that there were possibly | |
3552 | * extent items past i_size that needed to be deleted. In v3.12, | |
3553 | * truncate was changed to update i_size in sync with the extent | |
3554 | * items, but the (useless) orphan item was still created. Since | |
3555 | * v4.18, we don't create the orphan item for truncate at all. | |
3556 | * | |
3557 | * So, this item could mean that we need to do a truncate, but | |
3558 | * only if this filesystem was last used on a pre-v3.12 kernel | |
3559 | * and was not cleanly unmounted. The odds of that are quite | |
3560 | * slim, and it's a pain to do the truncate now, so just delete | |
3561 | * the orphan item. | |
3562 | * | |
3563 | * It's also possible that this orphan item was supposed to be | |
3564 | * deleted but wasn't. The inode number may have been reused, | |
3565 | * but either way, we can delete the orphan item. | |
7b128766 | 3566 | */ |
f7e9e8fc OS |
3567 | if (ret == -ENOENT || inode->i_nlink) { |
3568 | if (!ret) | |
3569 | iput(inode); | |
a8c9e576 | 3570 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3571 | if (IS_ERR(trans)) { |
3572 | ret = PTR_ERR(trans); | |
3573 | goto out; | |
3574 | } | |
0b246afa JM |
3575 | btrfs_debug(fs_info, "auto deleting %Lu", |
3576 | found_key.objectid); | |
a8c9e576 JB |
3577 | ret = btrfs_del_orphan_item(trans, root, |
3578 | found_key.objectid); | |
3a45bb20 | 3579 | btrfs_end_transaction(trans); |
4ef31a45 JB |
3580 | if (ret) |
3581 | goto out; | |
7b128766 JB |
3582 | continue; |
3583 | } | |
3584 | ||
f7e9e8fc | 3585 | nr_unlink++; |
7b128766 JB |
3586 | |
3587 | /* this will do delete_inode and everything for us */ | |
3588 | iput(inode); | |
3589 | } | |
3254c876 MX |
3590 | /* release the path since we're done with it */ |
3591 | btrfs_release_path(path); | |
3592 | ||
d68fc57b YZ |
3593 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3594 | ||
a575ceeb | 3595 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { |
7a7eaa40 | 3596 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3597 | if (!IS_ERR(trans)) |
3a45bb20 | 3598 | btrfs_end_transaction(trans); |
d68fc57b | 3599 | } |
7b128766 JB |
3600 | |
3601 | if (nr_unlink) | |
0b246afa | 3602 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
66b4ffd1 JB |
3603 | |
3604 | out: | |
3605 | if (ret) | |
0b246afa | 3606 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3607 | btrfs_free_path(path); |
3608 | return ret; | |
7b128766 JB |
3609 | } |
3610 | ||
46a53cca CM |
3611 | /* |
3612 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3613 | * don't find any xattrs, we know there can't be any acls. | |
3614 | * | |
3615 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3616 | */ | |
3617 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3618 | int slot, u64 objectid, |
3619 | int *first_xattr_slot) | |
46a53cca CM |
3620 | { |
3621 | u32 nritems = btrfs_header_nritems(leaf); | |
3622 | struct btrfs_key found_key; | |
f23b5a59 JB |
3623 | static u64 xattr_access = 0; |
3624 | static u64 xattr_default = 0; | |
46a53cca CM |
3625 | int scanned = 0; |
3626 | ||
f23b5a59 | 3627 | if (!xattr_access) { |
97d79299 AG |
3628 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3629 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3630 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3631 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3632 | } |
3633 | ||
46a53cca | 3634 | slot++; |
63541927 | 3635 | *first_xattr_slot = -1; |
46a53cca CM |
3636 | while (slot < nritems) { |
3637 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3638 | ||
3639 | /* we found a different objectid, there must not be acls */ | |
3640 | if (found_key.objectid != objectid) | |
3641 | return 0; | |
3642 | ||
3643 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3644 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3645 | if (*first_xattr_slot == -1) |
3646 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3647 | if (found_key.offset == xattr_access || |
3648 | found_key.offset == xattr_default) | |
3649 | return 1; | |
3650 | } | |
46a53cca CM |
3651 | |
3652 | /* | |
3653 | * we found a key greater than an xattr key, there can't | |
3654 | * be any acls later on | |
3655 | */ | |
3656 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3657 | return 0; | |
3658 | ||
3659 | slot++; | |
3660 | scanned++; | |
3661 | ||
3662 | /* | |
3663 | * it goes inode, inode backrefs, xattrs, extents, | |
3664 | * so if there are a ton of hard links to an inode there can | |
3665 | * be a lot of backrefs. Don't waste time searching too hard, | |
3666 | * this is just an optimization | |
3667 | */ | |
3668 | if (scanned >= 8) | |
3669 | break; | |
3670 | } | |
3671 | /* we hit the end of the leaf before we found an xattr or | |
3672 | * something larger than an xattr. We have to assume the inode | |
3673 | * has acls | |
3674 | */ | |
63541927 FDBM |
3675 | if (*first_xattr_slot == -1) |
3676 | *first_xattr_slot = slot; | |
46a53cca CM |
3677 | return 1; |
3678 | } | |
3679 | ||
d352ac68 CM |
3680 | /* |
3681 | * read an inode from the btree into the in-memory inode | |
3682 | */ | |
4222ea71 FM |
3683 | static int btrfs_read_locked_inode(struct inode *inode, |
3684 | struct btrfs_path *in_path) | |
39279cc3 | 3685 | { |
0b246afa | 3686 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4222ea71 | 3687 | struct btrfs_path *path = in_path; |
5f39d397 | 3688 | struct extent_buffer *leaf; |
39279cc3 CM |
3689 | struct btrfs_inode_item *inode_item; |
3690 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3691 | struct btrfs_key location; | |
67de1176 | 3692 | unsigned long ptr; |
46a53cca | 3693 | int maybe_acls; |
618e21d5 | 3694 | u32 rdev; |
39279cc3 | 3695 | int ret; |
2f7e33d4 | 3696 | bool filled = false; |
63541927 | 3697 | int first_xattr_slot; |
2f7e33d4 MX |
3698 | |
3699 | ret = btrfs_fill_inode(inode, &rdev); | |
3700 | if (!ret) | |
3701 | filled = true; | |
39279cc3 | 3702 | |
4222ea71 FM |
3703 | if (!path) { |
3704 | path = btrfs_alloc_path(); | |
3705 | if (!path) | |
3706 | return -ENOMEM; | |
3707 | } | |
1748f843 | 3708 | |
39279cc3 | 3709 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3710 | |
39279cc3 | 3711 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 | 3712 | if (ret) { |
4222ea71 FM |
3713 | if (path != in_path) |
3714 | btrfs_free_path(path); | |
f5b3a417 | 3715 | return ret; |
67710892 | 3716 | } |
39279cc3 | 3717 | |
5f39d397 | 3718 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3719 | |
3720 | if (filled) | |
67de1176 | 3721 | goto cache_index; |
2f7e33d4 | 3722 | |
5f39d397 CM |
3723 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3724 | struct btrfs_inode_item); | |
5f39d397 | 3725 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3726 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3727 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3728 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
6ef06d27 | 3729 | btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item)); |
41a2ee75 JB |
3730 | btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0, |
3731 | round_up(i_size_read(inode), fs_info->sectorsize)); | |
5f39d397 | 3732 | |
a937b979 DS |
3733 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3734 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3735 | |
a937b979 DS |
3736 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3737 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3738 | |
a937b979 DS |
3739 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3740 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3741 | |
9cc97d64 | 3742 | BTRFS_I(inode)->i_otime.tv_sec = |
3743 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3744 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3745 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3746 | |
a76a3cd4 | 3747 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3748 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3749 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3750 | ||
c7f88c4e JL |
3751 | inode_set_iversion_queried(inode, |
3752 | btrfs_inode_sequence(leaf, inode_item)); | |
6e17d30b YD |
3753 | inode->i_generation = BTRFS_I(inode)->generation; |
3754 | inode->i_rdev = 0; | |
3755 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3756 | ||
3757 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3758 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3759 | ||
3760 | cache_index: | |
5dc562c5 JB |
3761 | /* |
3762 | * If we were modified in the current generation and evicted from memory | |
3763 | * and then re-read we need to do a full sync since we don't have any | |
3764 | * idea about which extents were modified before we were evicted from | |
3765 | * cache. | |
6e17d30b YD |
3766 | * |
3767 | * This is required for both inode re-read from disk and delayed inode | |
3768 | * in delayed_nodes_tree. | |
5dc562c5 | 3769 | */ |
0b246afa | 3770 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
3771 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3772 | &BTRFS_I(inode)->runtime_flags); | |
3773 | ||
bde6c242 FM |
3774 | /* |
3775 | * We don't persist the id of the transaction where an unlink operation | |
3776 | * against the inode was last made. So here we assume the inode might | |
3777 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3778 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3779 | * between the inode and its parent if the inode is fsync'ed and the log | |
3780 | * replayed. For example, in the scenario: | |
3781 | * | |
3782 | * touch mydir/foo | |
3783 | * ln mydir/foo mydir/bar | |
3784 | * sync | |
3785 | * unlink mydir/bar | |
3786 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3787 | * xfs_io -c fsync mydir/foo | |
3788 | * <power failure> | |
3789 | * mount fs, triggers fsync log replay | |
3790 | * | |
3791 | * We must make sure that when we fsync our inode foo we also log its | |
3792 | * parent inode, otherwise after log replay the parent still has the | |
3793 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3794 | * and doesn't have an inode ref with the name "bar" anymore. | |
3795 | * | |
3796 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3797 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3798 | * transaction commits on fsync if our inode is a directory, or if our |
3799 | * inode is not a directory, logging its parent unnecessarily. | |
3800 | */ | |
3801 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3802 | ||
3ebac17c FM |
3803 | /* |
3804 | * Same logic as for last_unlink_trans. We don't persist the generation | |
3805 | * of the last transaction where this inode was used for a reflink | |
3806 | * operation, so after eviction and reloading the inode we must be | |
3807 | * pessimistic and assume the last transaction that modified the inode. | |
3808 | */ | |
3809 | BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans; | |
3810 | ||
67de1176 MX |
3811 | path->slots[0]++; |
3812 | if (inode->i_nlink != 1 || | |
3813 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3814 | goto cache_acl; | |
3815 | ||
3816 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 3817 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
3818 | goto cache_acl; |
3819 | ||
3820 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3821 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3822 | struct btrfs_inode_ref *ref; | |
3823 | ||
3824 | ref = (struct btrfs_inode_ref *)ptr; | |
3825 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3826 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3827 | struct btrfs_inode_extref *extref; | |
3828 | ||
3829 | extref = (struct btrfs_inode_extref *)ptr; | |
3830 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3831 | extref); | |
3832 | } | |
2f7e33d4 | 3833 | cache_acl: |
46a53cca CM |
3834 | /* |
3835 | * try to precache a NULL acl entry for files that don't have | |
3836 | * any xattrs or acls | |
3837 | */ | |
33345d01 | 3838 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 3839 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
3840 | if (first_xattr_slot != -1) { |
3841 | path->slots[0] = first_xattr_slot; | |
3842 | ret = btrfs_load_inode_props(inode, path); | |
3843 | if (ret) | |
0b246afa | 3844 | btrfs_err(fs_info, |
351fd353 | 3845 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 3846 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
3847 | root->root_key.objectid, ret); |
3848 | } | |
4222ea71 FM |
3849 | if (path != in_path) |
3850 | btrfs_free_path(path); | |
63541927 | 3851 | |
72c04902 AV |
3852 | if (!maybe_acls) |
3853 | cache_no_acl(inode); | |
46a53cca | 3854 | |
39279cc3 | 3855 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3856 | case S_IFREG: |
3857 | inode->i_mapping->a_ops = &btrfs_aops; | |
3858 | inode->i_fop = &btrfs_file_operations; | |
3859 | inode->i_op = &btrfs_file_inode_operations; | |
3860 | break; | |
3861 | case S_IFDIR: | |
3862 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 3863 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
3864 | break; |
3865 | case S_IFLNK: | |
3866 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3867 | inode_nohighmem(inode); |
4779cc04 | 3868 | inode->i_mapping->a_ops = &btrfs_aops; |
39279cc3 | 3869 | break; |
618e21d5 | 3870 | default: |
0279b4cd | 3871 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3872 | init_special_inode(inode, inode->i_mode, rdev); |
3873 | break; | |
39279cc3 | 3874 | } |
6cbff00f | 3875 | |
7b6a221e | 3876 | btrfs_sync_inode_flags_to_i_flags(inode); |
67710892 | 3877 | return 0; |
39279cc3 CM |
3878 | } |
3879 | ||
d352ac68 CM |
3880 | /* |
3881 | * given a leaf and an inode, copy the inode fields into the leaf | |
3882 | */ | |
e02119d5 CM |
3883 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3884 | struct extent_buffer *leaf, | |
5f39d397 | 3885 | struct btrfs_inode_item *item, |
39279cc3 CM |
3886 | struct inode *inode) |
3887 | { | |
51fab693 LB |
3888 | struct btrfs_map_token token; |
3889 | ||
c82f823c | 3890 | btrfs_init_map_token(&token, leaf); |
5f39d397 | 3891 | |
cc4c13d5 DS |
3892 | btrfs_set_token_inode_uid(&token, item, i_uid_read(inode)); |
3893 | btrfs_set_token_inode_gid(&token, item, i_gid_read(inode)); | |
3894 | btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size); | |
3895 | btrfs_set_token_inode_mode(&token, item, inode->i_mode); | |
3896 | btrfs_set_token_inode_nlink(&token, item, inode->i_nlink); | |
3897 | ||
3898 | btrfs_set_token_timespec_sec(&token, &item->atime, | |
3899 | inode->i_atime.tv_sec); | |
3900 | btrfs_set_token_timespec_nsec(&token, &item->atime, | |
3901 | inode->i_atime.tv_nsec); | |
3902 | ||
3903 | btrfs_set_token_timespec_sec(&token, &item->mtime, | |
3904 | inode->i_mtime.tv_sec); | |
3905 | btrfs_set_token_timespec_nsec(&token, &item->mtime, | |
3906 | inode->i_mtime.tv_nsec); | |
3907 | ||
3908 | btrfs_set_token_timespec_sec(&token, &item->ctime, | |
3909 | inode->i_ctime.tv_sec); | |
3910 | btrfs_set_token_timespec_nsec(&token, &item->ctime, | |
3911 | inode->i_ctime.tv_nsec); | |
3912 | ||
3913 | btrfs_set_token_timespec_sec(&token, &item->otime, | |
3914 | BTRFS_I(inode)->i_otime.tv_sec); | |
3915 | btrfs_set_token_timespec_nsec(&token, &item->otime, | |
3916 | BTRFS_I(inode)->i_otime.tv_nsec); | |
3917 | ||
3918 | btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode)); | |
3919 | btrfs_set_token_inode_generation(&token, item, | |
3920 | BTRFS_I(inode)->generation); | |
3921 | btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode)); | |
3922 | btrfs_set_token_inode_transid(&token, item, trans->transid); | |
3923 | btrfs_set_token_inode_rdev(&token, item, inode->i_rdev); | |
3924 | btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags); | |
3925 | btrfs_set_token_inode_block_group(&token, item, 0); | |
39279cc3 CM |
3926 | } |
3927 | ||
d352ac68 CM |
3928 | /* |
3929 | * copy everything in the in-memory inode into the btree. | |
3930 | */ | |
2115133f | 3931 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
dfeb9e7c NB |
3932 | struct btrfs_root *root, |
3933 | struct btrfs_inode *inode) | |
39279cc3 CM |
3934 | { |
3935 | struct btrfs_inode_item *inode_item; | |
3936 | struct btrfs_path *path; | |
5f39d397 | 3937 | struct extent_buffer *leaf; |
39279cc3 CM |
3938 | int ret; |
3939 | ||
3940 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3941 | if (!path) |
3942 | return -ENOMEM; | |
3943 | ||
dfeb9e7c | 3944 | ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1); |
39279cc3 CM |
3945 | if (ret) { |
3946 | if (ret > 0) | |
3947 | ret = -ENOENT; | |
3948 | goto failed; | |
3949 | } | |
3950 | ||
5f39d397 CM |
3951 | leaf = path->nodes[0]; |
3952 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3953 | struct btrfs_inode_item); |
39279cc3 | 3954 | |
dfeb9e7c | 3955 | fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode); |
5f39d397 | 3956 | btrfs_mark_buffer_dirty(leaf); |
dfeb9e7c | 3957 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3958 | ret = 0; |
3959 | failed: | |
39279cc3 CM |
3960 | btrfs_free_path(path); |
3961 | return ret; | |
3962 | } | |
3963 | ||
2115133f CM |
3964 | /* |
3965 | * copy everything in the in-memory inode into the btree. | |
3966 | */ | |
3967 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
9a56fcd1 NB |
3968 | struct btrfs_root *root, |
3969 | struct btrfs_inode *inode) | |
2115133f | 3970 | { |
0b246afa | 3971 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
3972 | int ret; |
3973 | ||
3974 | /* | |
3975 | * If the inode is a free space inode, we can deadlock during commit | |
3976 | * if we put it into the delayed code. | |
3977 | * | |
3978 | * The data relocation inode should also be directly updated | |
3979 | * without delay | |
3980 | */ | |
9a56fcd1 | 3981 | if (!btrfs_is_free_space_inode(inode) |
1d52c78a | 3982 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
0b246afa | 3983 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
3984 | btrfs_update_root_times(trans, root); |
3985 | ||
9a56fcd1 | 3986 | ret = btrfs_delayed_update_inode(trans, root, inode); |
2115133f | 3987 | if (!ret) |
9a56fcd1 | 3988 | btrfs_set_inode_last_trans(trans, inode); |
2115133f CM |
3989 | return ret; |
3990 | } | |
3991 | ||
9a56fcd1 | 3992 | return btrfs_update_inode_item(trans, root, inode); |
2115133f CM |
3993 | } |
3994 | ||
729f7961 NB |
3995 | int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3996 | struct btrfs_root *root, struct btrfs_inode *inode) | |
2115133f CM |
3997 | { |
3998 | int ret; | |
3999 | ||
729f7961 | 4000 | ret = btrfs_update_inode(trans, root, inode); |
2115133f | 4001 | if (ret == -ENOSPC) |
729f7961 | 4002 | return btrfs_update_inode_item(trans, root, inode); |
2115133f CM |
4003 | return ret; |
4004 | } | |
4005 | ||
d352ac68 CM |
4006 | /* |
4007 | * unlink helper that gets used here in inode.c and in the tree logging | |
4008 | * recovery code. It remove a link in a directory with a given name, and | |
4009 | * also drops the back refs in the inode to the directory | |
4010 | */ | |
92986796 AV |
4011 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4012 | struct btrfs_root *root, | |
4ec5934e NB |
4013 | struct btrfs_inode *dir, |
4014 | struct btrfs_inode *inode, | |
92986796 | 4015 | const char *name, int name_len) |
39279cc3 | 4016 | { |
0b246afa | 4017 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4018 | struct btrfs_path *path; |
39279cc3 | 4019 | int ret = 0; |
39279cc3 | 4020 | struct btrfs_dir_item *di; |
aec7477b | 4021 | u64 index; |
33345d01 LZ |
4022 | u64 ino = btrfs_ino(inode); |
4023 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
4024 | |
4025 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
4026 | if (!path) { |
4027 | ret = -ENOMEM; | |
554233a6 | 4028 | goto out; |
54aa1f4d CM |
4029 | } |
4030 | ||
33345d01 | 4031 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 | 4032 | name, name_len, -1); |
3cf5068f LB |
4033 | if (IS_ERR_OR_NULL(di)) { |
4034 | ret = di ? PTR_ERR(di) : -ENOENT; | |
39279cc3 CM |
4035 | goto err; |
4036 | } | |
39279cc3 | 4037 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
4038 | if (ret) |
4039 | goto err; | |
b3b4aa74 | 4040 | btrfs_release_path(path); |
39279cc3 | 4041 | |
67de1176 MX |
4042 | /* |
4043 | * If we don't have dir index, we have to get it by looking up | |
4044 | * the inode ref, since we get the inode ref, remove it directly, | |
4045 | * it is unnecessary to do delayed deletion. | |
4046 | * | |
4047 | * But if we have dir index, needn't search inode ref to get it. | |
4048 | * Since the inode ref is close to the inode item, it is better | |
4049 | * that we delay to delete it, and just do this deletion when | |
4050 | * we update the inode item. | |
4051 | */ | |
4ec5934e | 4052 | if (inode->dir_index) { |
67de1176 MX |
4053 | ret = btrfs_delayed_delete_inode_ref(inode); |
4054 | if (!ret) { | |
4ec5934e | 4055 | index = inode->dir_index; |
67de1176 MX |
4056 | goto skip_backref; |
4057 | } | |
4058 | } | |
4059 | ||
33345d01 LZ |
4060 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
4061 | dir_ino, &index); | |
aec7477b | 4062 | if (ret) { |
0b246afa | 4063 | btrfs_info(fs_info, |
c2cf52eb | 4064 | "failed to delete reference to %.*s, inode %llu parent %llu", |
c1c9ff7c | 4065 | name_len, name, ino, dir_ino); |
66642832 | 4066 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
4067 | goto err; |
4068 | } | |
67de1176 | 4069 | skip_backref: |
9add2945 | 4070 | ret = btrfs_delete_delayed_dir_index(trans, dir, index); |
79787eaa | 4071 | if (ret) { |
66642832 | 4072 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 4073 | goto err; |
79787eaa | 4074 | } |
39279cc3 | 4075 | |
4ec5934e NB |
4076 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode, |
4077 | dir_ino); | |
79787eaa | 4078 | if (ret != 0 && ret != -ENOENT) { |
66642832 | 4079 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4080 | goto err; |
4081 | } | |
e02119d5 | 4082 | |
4ec5934e NB |
4083 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, |
4084 | index); | |
6418c961 CM |
4085 | if (ret == -ENOENT) |
4086 | ret = 0; | |
d4e3991b | 4087 | else if (ret) |
66642832 | 4088 | btrfs_abort_transaction(trans, ret); |
63611e73 JB |
4089 | |
4090 | /* | |
4091 | * If we have a pending delayed iput we could end up with the final iput | |
4092 | * being run in btrfs-cleaner context. If we have enough of these built | |
4093 | * up we can end up burning a lot of time in btrfs-cleaner without any | |
4094 | * way to throttle the unlinks. Since we're currently holding a ref on | |
4095 | * the inode we can run the delayed iput here without any issues as the | |
4096 | * final iput won't be done until after we drop the ref we're currently | |
4097 | * holding. | |
4098 | */ | |
4099 | btrfs_run_delayed_iput(fs_info, inode); | |
39279cc3 CM |
4100 | err: |
4101 | btrfs_free_path(path); | |
e02119d5 CM |
4102 | if (ret) |
4103 | goto out; | |
4104 | ||
6ef06d27 | 4105 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2); |
4ec5934e NB |
4106 | inode_inc_iversion(&inode->vfs_inode); |
4107 | inode_inc_iversion(&dir->vfs_inode); | |
4108 | inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime = | |
4109 | dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode); | |
9a56fcd1 | 4110 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 4111 | out: |
39279cc3 CM |
4112 | return ret; |
4113 | } | |
4114 | ||
92986796 AV |
4115 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4116 | struct btrfs_root *root, | |
4ec5934e | 4117 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
92986796 AV |
4118 | const char *name, int name_len) |
4119 | { | |
4120 | int ret; | |
4121 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
4122 | if (!ret) { | |
4ec5934e | 4123 | drop_nlink(&inode->vfs_inode); |
9a56fcd1 | 4124 | ret = btrfs_update_inode(trans, root, inode); |
92986796 AV |
4125 | } |
4126 | return ret; | |
4127 | } | |
39279cc3 | 4128 | |
a22285a6 YZ |
4129 | /* |
4130 | * helper to start transaction for unlink and rmdir. | |
4131 | * | |
d52be818 JB |
4132 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4133 | * if we cannot make our reservations the normal way try and see if there is | |
4134 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4135 | * allow the unlink to occur. | |
a22285a6 | 4136 | */ |
d52be818 | 4137 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4138 | { |
a22285a6 | 4139 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 4140 | |
e70bea5f JB |
4141 | /* |
4142 | * 1 for the possible orphan item | |
4143 | * 1 for the dir item | |
4144 | * 1 for the dir index | |
4145 | * 1 for the inode ref | |
e70bea5f JB |
4146 | * 1 for the inode |
4147 | */ | |
7f9fe614 | 4148 | return btrfs_start_transaction_fallback_global_rsv(root, 5); |
a22285a6 YZ |
4149 | } |
4150 | ||
4151 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4152 | { | |
4153 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4154 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 4155 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4156 | int ret; |
a22285a6 | 4157 | |
d52be818 | 4158 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4159 | if (IS_ERR(trans)) |
4160 | return PTR_ERR(trans); | |
5f39d397 | 4161 | |
4ec5934e NB |
4162 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
4163 | 0); | |
12fcfd22 | 4164 | |
4ec5934e NB |
4165 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4166 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4167 | dentry->d_name.len); | |
b532402e TI |
4168 | if (ret) |
4169 | goto out; | |
7b128766 | 4170 | |
a22285a6 | 4171 | if (inode->i_nlink == 0) { |
73f2e545 | 4172 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
b532402e TI |
4173 | if (ret) |
4174 | goto out; | |
a22285a6 | 4175 | } |
7b128766 | 4176 | |
b532402e | 4177 | out: |
3a45bb20 | 4178 | btrfs_end_transaction(trans); |
2ff7e61e | 4179 | btrfs_btree_balance_dirty(root->fs_info); |
39279cc3 CM |
4180 | return ret; |
4181 | } | |
4182 | ||
f60a2364 | 4183 | static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
045d3967 | 4184 | struct inode *dir, struct dentry *dentry) |
4df27c4d | 4185 | { |
401b3b19 | 4186 | struct btrfs_root *root = BTRFS_I(dir)->root; |
045d3967 | 4187 | struct btrfs_inode *inode = BTRFS_I(d_inode(dentry)); |
4df27c4d YZ |
4188 | struct btrfs_path *path; |
4189 | struct extent_buffer *leaf; | |
4190 | struct btrfs_dir_item *di; | |
4191 | struct btrfs_key key; | |
045d3967 JB |
4192 | const char *name = dentry->d_name.name; |
4193 | int name_len = dentry->d_name.len; | |
4df27c4d YZ |
4194 | u64 index; |
4195 | int ret; | |
045d3967 | 4196 | u64 objectid; |
4a0cc7ca | 4197 | u64 dir_ino = btrfs_ino(BTRFS_I(dir)); |
4df27c4d | 4198 | |
045d3967 JB |
4199 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) { |
4200 | objectid = inode->root->root_key.objectid; | |
4201 | } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) { | |
4202 | objectid = inode->location.objectid; | |
4203 | } else { | |
4204 | WARN_ON(1); | |
4205 | return -EINVAL; | |
4206 | } | |
4207 | ||
4df27c4d YZ |
4208 | path = btrfs_alloc_path(); |
4209 | if (!path) | |
4210 | return -ENOMEM; | |
4211 | ||
33345d01 | 4212 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4213 | name, name_len, -1); |
79787eaa | 4214 | if (IS_ERR_OR_NULL(di)) { |
3cf5068f | 4215 | ret = di ? PTR_ERR(di) : -ENOENT; |
79787eaa JM |
4216 | goto out; |
4217 | } | |
4df27c4d YZ |
4218 | |
4219 | leaf = path->nodes[0]; | |
4220 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4221 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4222 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 4223 | if (ret) { |
66642832 | 4224 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4225 | goto out; |
4226 | } | |
b3b4aa74 | 4227 | btrfs_release_path(path); |
4df27c4d | 4228 | |
d49d3287 JB |
4229 | /* |
4230 | * This is a placeholder inode for a subvolume we didn't have a | |
4231 | * reference to at the time of the snapshot creation. In the meantime | |
4232 | * we could have renamed the real subvol link into our snapshot, so | |
1a9fd417 | 4233 | * depending on btrfs_del_root_ref to return -ENOENT here is incorrect. |
d49d3287 JB |
4234 | * Instead simply lookup the dir_index_item for this entry so we can |
4235 | * remove it. Otherwise we know we have a ref to the root and we can | |
4236 | * call btrfs_del_root_ref, and it _shouldn't_ fail. | |
4237 | */ | |
4238 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) { | |
33345d01 | 4239 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4240 | name, name_len); |
79787eaa JM |
4241 | if (IS_ERR_OR_NULL(di)) { |
4242 | if (!di) | |
4243 | ret = -ENOENT; | |
4244 | else | |
4245 | ret = PTR_ERR(di); | |
66642832 | 4246 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4247 | goto out; |
4248 | } | |
4df27c4d YZ |
4249 | |
4250 | leaf = path->nodes[0]; | |
4251 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4df27c4d | 4252 | index = key.offset; |
d49d3287 JB |
4253 | btrfs_release_path(path); |
4254 | } else { | |
4255 | ret = btrfs_del_root_ref(trans, objectid, | |
4256 | root->root_key.objectid, dir_ino, | |
4257 | &index, name, name_len); | |
4258 | if (ret) { | |
4259 | btrfs_abort_transaction(trans, ret); | |
4260 | goto out; | |
4261 | } | |
4df27c4d YZ |
4262 | } |
4263 | ||
9add2945 | 4264 | ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index); |
79787eaa | 4265 | if (ret) { |
66642832 | 4266 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4267 | goto out; |
4268 | } | |
4df27c4d | 4269 | |
6ef06d27 | 4270 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2); |
0c4d2d95 | 4271 | inode_inc_iversion(dir); |
c2050a45 | 4272 | dir->i_mtime = dir->i_ctime = current_time(dir); |
729f7961 | 4273 | ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir)); |
79787eaa | 4274 | if (ret) |
66642832 | 4275 | btrfs_abort_transaction(trans, ret); |
79787eaa | 4276 | out: |
71d7aed0 | 4277 | btrfs_free_path(path); |
79787eaa | 4278 | return ret; |
4df27c4d YZ |
4279 | } |
4280 | ||
ec42f167 MT |
4281 | /* |
4282 | * Helper to check if the subvolume references other subvolumes or if it's | |
4283 | * default. | |
4284 | */ | |
f60a2364 | 4285 | static noinline int may_destroy_subvol(struct btrfs_root *root) |
ec42f167 MT |
4286 | { |
4287 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4288 | struct btrfs_path *path; | |
4289 | struct btrfs_dir_item *di; | |
4290 | struct btrfs_key key; | |
4291 | u64 dir_id; | |
4292 | int ret; | |
4293 | ||
4294 | path = btrfs_alloc_path(); | |
4295 | if (!path) | |
4296 | return -ENOMEM; | |
4297 | ||
4298 | /* Make sure this root isn't set as the default subvol */ | |
4299 | dir_id = btrfs_super_root_dir(fs_info->super_copy); | |
4300 | di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path, | |
4301 | dir_id, "default", 7, 0); | |
4302 | if (di && !IS_ERR(di)) { | |
4303 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
4304 | if (key.objectid == root->root_key.objectid) { | |
4305 | ret = -EPERM; | |
4306 | btrfs_err(fs_info, | |
4307 | "deleting default subvolume %llu is not allowed", | |
4308 | key.objectid); | |
4309 | goto out; | |
4310 | } | |
4311 | btrfs_release_path(path); | |
4312 | } | |
4313 | ||
4314 | key.objectid = root->root_key.objectid; | |
4315 | key.type = BTRFS_ROOT_REF_KEY; | |
4316 | key.offset = (u64)-1; | |
4317 | ||
4318 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); | |
4319 | if (ret < 0) | |
4320 | goto out; | |
4321 | BUG_ON(ret == 0); | |
4322 | ||
4323 | ret = 0; | |
4324 | if (path->slots[0] > 0) { | |
4325 | path->slots[0]--; | |
4326 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
4327 | if (key.objectid == root->root_key.objectid && | |
4328 | key.type == BTRFS_ROOT_REF_KEY) | |
4329 | ret = -ENOTEMPTY; | |
4330 | } | |
4331 | out: | |
4332 | btrfs_free_path(path); | |
4333 | return ret; | |
4334 | } | |
4335 | ||
20a68004 NB |
4336 | /* Delete all dentries for inodes belonging to the root */ |
4337 | static void btrfs_prune_dentries(struct btrfs_root *root) | |
4338 | { | |
4339 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4340 | struct rb_node *node; | |
4341 | struct rb_node *prev; | |
4342 | struct btrfs_inode *entry; | |
4343 | struct inode *inode; | |
4344 | u64 objectid = 0; | |
4345 | ||
4346 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) | |
4347 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
4348 | ||
4349 | spin_lock(&root->inode_lock); | |
4350 | again: | |
4351 | node = root->inode_tree.rb_node; | |
4352 | prev = NULL; | |
4353 | while (node) { | |
4354 | prev = node; | |
4355 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4356 | ||
37508515 | 4357 | if (objectid < btrfs_ino(entry)) |
20a68004 | 4358 | node = node->rb_left; |
37508515 | 4359 | else if (objectid > btrfs_ino(entry)) |
20a68004 NB |
4360 | node = node->rb_right; |
4361 | else | |
4362 | break; | |
4363 | } | |
4364 | if (!node) { | |
4365 | while (prev) { | |
4366 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
37508515 | 4367 | if (objectid <= btrfs_ino(entry)) { |
20a68004 NB |
4368 | node = prev; |
4369 | break; | |
4370 | } | |
4371 | prev = rb_next(prev); | |
4372 | } | |
4373 | } | |
4374 | while (node) { | |
4375 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
37508515 | 4376 | objectid = btrfs_ino(entry) + 1; |
20a68004 NB |
4377 | inode = igrab(&entry->vfs_inode); |
4378 | if (inode) { | |
4379 | spin_unlock(&root->inode_lock); | |
4380 | if (atomic_read(&inode->i_count) > 1) | |
4381 | d_prune_aliases(inode); | |
4382 | /* | |
4383 | * btrfs_drop_inode will have it removed from the inode | |
4384 | * cache when its usage count hits zero. | |
4385 | */ | |
4386 | iput(inode); | |
4387 | cond_resched(); | |
4388 | spin_lock(&root->inode_lock); | |
4389 | goto again; | |
4390 | } | |
4391 | ||
4392 | if (cond_resched_lock(&root->inode_lock)) | |
4393 | goto again; | |
4394 | ||
4395 | node = rb_next(node); | |
4396 | } | |
4397 | spin_unlock(&root->inode_lock); | |
4398 | } | |
4399 | ||
f60a2364 MT |
4400 | int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry) |
4401 | { | |
4402 | struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb); | |
4403 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4404 | struct inode *inode = d_inode(dentry); | |
4405 | struct btrfs_root *dest = BTRFS_I(inode)->root; | |
4406 | struct btrfs_trans_handle *trans; | |
4407 | struct btrfs_block_rsv block_rsv; | |
4408 | u64 root_flags; | |
f60a2364 | 4409 | int ret; |
f60a2364 MT |
4410 | |
4411 | /* | |
4412 | * Don't allow to delete a subvolume with send in progress. This is | |
4413 | * inside the inode lock so the error handling that has to drop the bit | |
4414 | * again is not run concurrently. | |
4415 | */ | |
4416 | spin_lock(&dest->root_item_lock); | |
a7176f74 | 4417 | if (dest->send_in_progress) { |
f60a2364 MT |
4418 | spin_unlock(&dest->root_item_lock); |
4419 | btrfs_warn(fs_info, | |
4420 | "attempt to delete subvolume %llu during send", | |
4421 | dest->root_key.objectid); | |
4422 | return -EPERM; | |
4423 | } | |
a7176f74 LF |
4424 | root_flags = btrfs_root_flags(&dest->root_item); |
4425 | btrfs_set_root_flags(&dest->root_item, | |
4426 | root_flags | BTRFS_ROOT_SUBVOL_DEAD); | |
4427 | spin_unlock(&dest->root_item_lock); | |
f60a2364 MT |
4428 | |
4429 | down_write(&fs_info->subvol_sem); | |
4430 | ||
ee0d904f NB |
4431 | ret = may_destroy_subvol(dest); |
4432 | if (ret) | |
f60a2364 MT |
4433 | goto out_up_write; |
4434 | ||
4435 | btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); | |
4436 | /* | |
4437 | * One for dir inode, | |
4438 | * two for dir entries, | |
4439 | * two for root ref/backref. | |
4440 | */ | |
ee0d904f NB |
4441 | ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true); |
4442 | if (ret) | |
f60a2364 MT |
4443 | goto out_up_write; |
4444 | ||
4445 | trans = btrfs_start_transaction(root, 0); | |
4446 | if (IS_ERR(trans)) { | |
ee0d904f | 4447 | ret = PTR_ERR(trans); |
f60a2364 MT |
4448 | goto out_release; |
4449 | } | |
4450 | trans->block_rsv = &block_rsv; | |
4451 | trans->bytes_reserved = block_rsv.size; | |
4452 | ||
4453 | btrfs_record_snapshot_destroy(trans, BTRFS_I(dir)); | |
4454 | ||
045d3967 | 4455 | ret = btrfs_unlink_subvol(trans, dir, dentry); |
f60a2364 | 4456 | if (ret) { |
f60a2364 MT |
4457 | btrfs_abort_transaction(trans, ret); |
4458 | goto out_end_trans; | |
4459 | } | |
4460 | ||
2731f518 JB |
4461 | ret = btrfs_record_root_in_trans(trans, dest); |
4462 | if (ret) { | |
4463 | btrfs_abort_transaction(trans, ret); | |
4464 | goto out_end_trans; | |
4465 | } | |
f60a2364 MT |
4466 | |
4467 | memset(&dest->root_item.drop_progress, 0, | |
4468 | sizeof(dest->root_item.drop_progress)); | |
c8422684 | 4469 | btrfs_set_root_drop_level(&dest->root_item, 0); |
f60a2364 MT |
4470 | btrfs_set_root_refs(&dest->root_item, 0); |
4471 | ||
4472 | if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) { | |
4473 | ret = btrfs_insert_orphan_item(trans, | |
4474 | fs_info->tree_root, | |
4475 | dest->root_key.objectid); | |
4476 | if (ret) { | |
4477 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4478 | goto out_end_trans; |
4479 | } | |
4480 | } | |
4481 | ||
d1957791 | 4482 | ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid, |
f60a2364 MT |
4483 | BTRFS_UUID_KEY_SUBVOL, |
4484 | dest->root_key.objectid); | |
4485 | if (ret && ret != -ENOENT) { | |
4486 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4487 | goto out_end_trans; |
4488 | } | |
4489 | if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) { | |
d1957791 | 4490 | ret = btrfs_uuid_tree_remove(trans, |
f60a2364 MT |
4491 | dest->root_item.received_uuid, |
4492 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4493 | dest->root_key.objectid); | |
4494 | if (ret && ret != -ENOENT) { | |
4495 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4496 | goto out_end_trans; |
4497 | } | |
4498 | } | |
4499 | ||
082b6c97 QW |
4500 | free_anon_bdev(dest->anon_dev); |
4501 | dest->anon_dev = 0; | |
f60a2364 MT |
4502 | out_end_trans: |
4503 | trans->block_rsv = NULL; | |
4504 | trans->bytes_reserved = 0; | |
4505 | ret = btrfs_end_transaction(trans); | |
f60a2364 MT |
4506 | inode->i_flags |= S_DEAD; |
4507 | out_release: | |
e85fde51 | 4508 | btrfs_subvolume_release_metadata(root, &block_rsv); |
f60a2364 MT |
4509 | out_up_write: |
4510 | up_write(&fs_info->subvol_sem); | |
ee0d904f | 4511 | if (ret) { |
f60a2364 MT |
4512 | spin_lock(&dest->root_item_lock); |
4513 | root_flags = btrfs_root_flags(&dest->root_item); | |
4514 | btrfs_set_root_flags(&dest->root_item, | |
4515 | root_flags & ~BTRFS_ROOT_SUBVOL_DEAD); | |
4516 | spin_unlock(&dest->root_item_lock); | |
4517 | } else { | |
4518 | d_invalidate(dentry); | |
20a68004 | 4519 | btrfs_prune_dentries(dest); |
f60a2364 | 4520 | ASSERT(dest->send_in_progress == 0); |
f60a2364 MT |
4521 | } |
4522 | ||
ee0d904f | 4523 | return ret; |
f60a2364 MT |
4524 | } |
4525 | ||
39279cc3 CM |
4526 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4527 | { | |
2b0143b5 | 4528 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4529 | int err = 0; |
39279cc3 | 4530 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4531 | struct btrfs_trans_handle *trans; |
44f714da | 4532 | u64 last_unlink_trans; |
39279cc3 | 4533 | |
b3ae244e | 4534 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4535 | return -ENOTEMPTY; |
4a0cc7ca | 4536 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) |
a79a464d | 4537 | return btrfs_delete_subvolume(dir, dentry); |
134d4512 | 4538 | |
d52be818 | 4539 | trans = __unlink_start_trans(dir); |
a22285a6 | 4540 | if (IS_ERR(trans)) |
5df6a9f6 | 4541 | return PTR_ERR(trans); |
5df6a9f6 | 4542 | |
4a0cc7ca | 4543 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
045d3967 | 4544 | err = btrfs_unlink_subvol(trans, dir, dentry); |
4df27c4d YZ |
4545 | goto out; |
4546 | } | |
4547 | ||
73f2e545 | 4548 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
7b128766 | 4549 | if (err) |
4df27c4d | 4550 | goto out; |
7b128766 | 4551 | |
44f714da FM |
4552 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4553 | ||
39279cc3 | 4554 | /* now the directory is empty */ |
4ec5934e NB |
4555 | err = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4556 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4557 | dentry->d_name.len); | |
44f714da | 4558 | if (!err) { |
6ef06d27 | 4559 | btrfs_i_size_write(BTRFS_I(inode), 0); |
44f714da FM |
4560 | /* |
4561 | * Propagate the last_unlink_trans value of the deleted dir to | |
4562 | * its parent directory. This is to prevent an unrecoverable | |
4563 | * log tree in the case we do something like this: | |
4564 | * 1) create dir foo | |
4565 | * 2) create snapshot under dir foo | |
4566 | * 3) delete the snapshot | |
4567 | * 4) rmdir foo | |
4568 | * 5) mkdir foo | |
4569 | * 6) fsync foo or some file inside foo | |
4570 | */ | |
4571 | if (last_unlink_trans >= trans->transid) | |
4572 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4573 | } | |
4df27c4d | 4574 | out: |
3a45bb20 | 4575 | btrfs_end_transaction(trans); |
2ff7e61e | 4576 | btrfs_btree_balance_dirty(root->fs_info); |
3954401f | 4577 | |
39279cc3 CM |
4578 | return err; |
4579 | } | |
4580 | ||
ddfae63c JB |
4581 | /* |
4582 | * Return this if we need to call truncate_block for the last bit of the | |
4583 | * truncate. | |
4584 | */ | |
4585 | #define NEED_TRUNCATE_BLOCK 1 | |
0305cd5f | 4586 | |
39279cc3 | 4587 | /* |
4f7e6737 | 4588 | * Remove inode items from a given root. |
39279cc3 | 4589 | * |
4f7e6737 FM |
4590 | * @trans: A transaction handle. |
4591 | * @root: The root from which to remove items. | |
4592 | * @inode: The inode whose items we want to remove. | |
4593 | * @new_size: The new i_size for the inode. This is only applicable when | |
4594 | * @min_type is BTRFS_EXTENT_DATA_KEY, must be 0 otherwise. | |
4595 | * @min_type: The minimum key type to remove. All keys with a type | |
4596 | * greater than this value are removed and all keys with | |
4597 | * this type are removed only if their offset is >= @new_size. | |
0d7d3165 FM |
4598 | * @extents_found: Output parameter that will contain the number of file |
4599 | * extent items that were removed or adjusted to the new | |
4600 | * inode i_size. The caller is responsible for initializing | |
4601 | * the counter. Also, it can be NULL if the caller does not | |
4602 | * need this counter. | |
7b128766 | 4603 | * |
4f7e6737 FM |
4604 | * Remove all keys associated with the inode from the given root that have a key |
4605 | * with a type greater than or equals to @min_type. When @min_type has a value of | |
4606 | * BTRFS_EXTENT_DATA_KEY, only remove file extent items that have an offset value | |
4607 | * greater than or equals to @new_size. If a file extent item that starts before | |
4608 | * @new_size and ends after it is found, its length is adjusted. | |
4609 | * | |
4610 | * Returns: 0 on success, < 0 on error and NEED_TRUNCATE_BLOCK when @min_type is | |
4611 | * BTRFS_EXTENT_DATA_KEY and the caller must truncate the last block. | |
39279cc3 | 4612 | */ |
8082510e YZ |
4613 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4614 | struct btrfs_root *root, | |
50743398 | 4615 | struct btrfs_inode *inode, |
0d7d3165 FM |
4616 | u64 new_size, u32 min_type, |
4617 | u64 *extents_found) | |
39279cc3 | 4618 | { |
0b246afa | 4619 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4620 | struct btrfs_path *path; |
5f39d397 | 4621 | struct extent_buffer *leaf; |
39279cc3 | 4622 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4623 | struct btrfs_key key; |
4624 | struct btrfs_key found_key; | |
39279cc3 | 4625 | u64 extent_start = 0; |
db94535d | 4626 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4627 | u64 extent_offset = 0; |
39279cc3 | 4628 | u64 item_end = 0; |
c1aa4575 | 4629 | u64 last_size = new_size; |
8082510e | 4630 | u32 found_type = (u8)-1; |
39279cc3 CM |
4631 | int found_extent; |
4632 | int del_item; | |
85e21bac CM |
4633 | int pending_del_nr = 0; |
4634 | int pending_del_slot = 0; | |
179e29e4 | 4635 | int extent_type = -1; |
8082510e | 4636 | int ret; |
50743398 | 4637 | u64 ino = btrfs_ino(inode); |
28ed1345 | 4638 | u64 bytes_deleted = 0; |
897ca819 TM |
4639 | bool be_nice = false; |
4640 | bool should_throttle = false; | |
28553fa9 FM |
4641 | const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize); |
4642 | struct extent_state *cached_state = NULL; | |
8082510e YZ |
4643 | |
4644 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4645 | |
28ed1345 | 4646 | /* |
92a7cc42 QW |
4647 | * For non-free space inodes and non-shareable roots, we want to back |
4648 | * off from time to time. This means all inodes in subvolume roots, | |
4649 | * reloc roots, and data reloc roots. | |
28ed1345 | 4650 | */ |
50743398 | 4651 | if (!btrfs_is_free_space_inode(inode) && |
92a7cc42 | 4652 | test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
897ca819 | 4653 | be_nice = true; |
28ed1345 | 4654 | |
0eb0e19c MF |
4655 | path = btrfs_alloc_path(); |
4656 | if (!path) | |
4657 | return -ENOMEM; | |
e4058b54 | 4658 | path->reada = READA_BACK; |
0eb0e19c | 4659 | |
82028e0a | 4660 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
50743398 | 4661 | lock_extent_bits(&inode->io_tree, lock_start, (u64)-1, |
a5ae50de | 4662 | &cached_state); |
28553fa9 | 4663 | |
82028e0a QW |
4664 | /* |
4665 | * We want to drop from the next block forward in case this | |
4666 | * new size is not block aligned since we will be keeping the | |
4667 | * last block of the extent just the way it is. | |
4668 | */ | |
50743398 | 4669 | btrfs_drop_extent_cache(inode, ALIGN(new_size, |
0b246afa | 4670 | fs_info->sectorsize), |
da17066c | 4671 | (u64)-1, 0); |
82028e0a | 4672 | } |
8082510e | 4673 | |
16cdcec7 MX |
4674 | /* |
4675 | * This function is also used to drop the items in the log tree before | |
4676 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
52042d8e | 4677 | * it is used to drop the logged items. So we shouldn't kill the delayed |
16cdcec7 MX |
4678 | * items. |
4679 | */ | |
50743398 NB |
4680 | if (min_type == 0 && root == inode->root) |
4681 | btrfs_kill_delayed_inode_items(inode); | |
16cdcec7 | 4682 | |
33345d01 | 4683 | key.objectid = ino; |
39279cc3 | 4684 | key.offset = (u64)-1; |
5f39d397 CM |
4685 | key.type = (u8)-1; |
4686 | ||
85e21bac | 4687 | search_again: |
28ed1345 CM |
4688 | /* |
4689 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4690 | * up a huge file in a single leaf. Most of the time that | |
4691 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4692 | */ | |
fd86a3a3 OS |
4693 | if (be_nice && bytes_deleted > SZ_32M && |
4694 | btrfs_should_end_transaction(trans)) { | |
4695 | ret = -EAGAIN; | |
4696 | goto out; | |
28ed1345 CM |
4697 | } |
4698 | ||
85e21bac | 4699 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
fd86a3a3 | 4700 | if (ret < 0) |
8082510e | 4701 | goto out; |
d397712b | 4702 | |
85e21bac | 4703 | if (ret > 0) { |
fd86a3a3 | 4704 | ret = 0; |
e02119d5 CM |
4705 | /* there are no items in the tree for us to truncate, we're |
4706 | * done | |
4707 | */ | |
8082510e YZ |
4708 | if (path->slots[0] == 0) |
4709 | goto out; | |
85e21bac CM |
4710 | path->slots[0]--; |
4711 | } | |
4712 | ||
d397712b | 4713 | while (1) { |
9ddc959e JB |
4714 | u64 clear_start = 0, clear_len = 0; |
4715 | ||
39279cc3 | 4716 | fi = NULL; |
5f39d397 CM |
4717 | leaf = path->nodes[0]; |
4718 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4719 | found_type = found_key.type; |
39279cc3 | 4720 | |
33345d01 | 4721 | if (found_key.objectid != ino) |
39279cc3 | 4722 | break; |
5f39d397 | 4723 | |
85e21bac | 4724 | if (found_type < min_type) |
39279cc3 CM |
4725 | break; |
4726 | ||
5f39d397 | 4727 | item_end = found_key.offset; |
39279cc3 | 4728 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4729 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4730 | struct btrfs_file_extent_item); |
179e29e4 CM |
4731 | extent_type = btrfs_file_extent_type(leaf, fi); |
4732 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4733 | item_end += |
db94535d | 4734 | btrfs_file_extent_num_bytes(leaf, fi); |
09ed2f16 LB |
4735 | |
4736 | trace_btrfs_truncate_show_fi_regular( | |
50743398 | 4737 | inode, leaf, fi, found_key.offset); |
179e29e4 | 4738 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
e41ca589 QW |
4739 | item_end += btrfs_file_extent_ram_bytes(leaf, |
4740 | fi); | |
09ed2f16 LB |
4741 | |
4742 | trace_btrfs_truncate_show_fi_inline( | |
50743398 | 4743 | inode, leaf, fi, path->slots[0], |
09ed2f16 | 4744 | found_key.offset); |
39279cc3 | 4745 | } |
008630c1 | 4746 | item_end--; |
39279cc3 | 4747 | } |
8082510e YZ |
4748 | if (found_type > min_type) { |
4749 | del_item = 1; | |
4750 | } else { | |
76b42abb | 4751 | if (item_end < new_size) |
b888db2b | 4752 | break; |
8082510e YZ |
4753 | if (found_key.offset >= new_size) |
4754 | del_item = 1; | |
4755 | else | |
4756 | del_item = 0; | |
39279cc3 | 4757 | } |
39279cc3 | 4758 | found_extent = 0; |
39279cc3 | 4759 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4760 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4761 | goto delete; | |
4762 | ||
0d7d3165 FM |
4763 | if (extents_found != NULL) |
4764 | (*extents_found)++; | |
4765 | ||
179e29e4 | 4766 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
39279cc3 | 4767 | u64 num_dec; |
9ddc959e JB |
4768 | |
4769 | clear_start = found_key.offset; | |
db94535d | 4770 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4771 | if (!del_item) { |
db94535d CM |
4772 | u64 orig_num_bytes = |
4773 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4774 | extent_num_bytes = ALIGN(new_size - |
4775 | found_key.offset, | |
0b246afa | 4776 | fs_info->sectorsize); |
9ddc959e | 4777 | clear_start = ALIGN(new_size, fs_info->sectorsize); |
db94535d CM |
4778 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4779 | extent_num_bytes); | |
4780 | num_dec = (orig_num_bytes - | |
9069218d | 4781 | extent_num_bytes); |
92a7cc42 | 4782 | if (test_bit(BTRFS_ROOT_SHAREABLE, |
27cdeb70 MX |
4783 | &root->state) && |
4784 | extent_start != 0) | |
50743398 NB |
4785 | inode_sub_bytes(&inode->vfs_inode, |
4786 | num_dec); | |
5f39d397 | 4787 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4788 | } else { |
db94535d CM |
4789 | extent_num_bytes = |
4790 | btrfs_file_extent_disk_num_bytes(leaf, | |
4791 | fi); | |
5d4f98a2 YZ |
4792 | extent_offset = found_key.offset - |
4793 | btrfs_file_extent_offset(leaf, fi); | |
4794 | ||
39279cc3 | 4795 | /* FIXME blocksize != 4096 */ |
9069218d | 4796 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4797 | if (extent_start != 0) { |
4798 | found_extent = 1; | |
92a7cc42 | 4799 | if (test_bit(BTRFS_ROOT_SHAREABLE, |
27cdeb70 | 4800 | &root->state)) |
50743398 NB |
4801 | inode_sub_bytes(&inode->vfs_inode, |
4802 | num_dec); | |
e02119d5 | 4803 | } |
39279cc3 | 4804 | } |
9ddc959e | 4805 | clear_len = num_dec; |
9069218d | 4806 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4807 | /* |
4808 | * we can't truncate inline items that have had | |
4809 | * special encodings | |
4810 | */ | |
4811 | if (!del_item && | |
c8b97818 | 4812 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
ddfae63c JB |
4813 | btrfs_file_extent_other_encoding(leaf, fi) == 0 && |
4814 | btrfs_file_extent_compression(leaf, fi) == 0) { | |
4815 | u32 size = (u32)(new_size - found_key.offset); | |
4816 | ||
4817 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4818 | size = btrfs_file_extent_calc_inline_size(size); | |
78ac4f9e | 4819 | btrfs_truncate_item(path, size, 1); |
ddfae63c | 4820 | } else if (!del_item) { |
514ac8ad | 4821 | /* |
ddfae63c JB |
4822 | * We have to bail so the last_size is set to |
4823 | * just before this extent. | |
514ac8ad | 4824 | */ |
fd86a3a3 | 4825 | ret = NEED_TRUNCATE_BLOCK; |
ddfae63c | 4826 | break; |
9ddc959e JB |
4827 | } else { |
4828 | /* | |
4829 | * Inline extents are special, we just treat | |
4830 | * them as a full sector worth in the file | |
4831 | * extent tree just for simplicity sake. | |
4832 | */ | |
4833 | clear_len = fs_info->sectorsize; | |
ddfae63c | 4834 | } |
0305cd5f | 4835 | |
92a7cc42 | 4836 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
50743398 NB |
4837 | inode_sub_bytes(&inode->vfs_inode, |
4838 | item_end + 1 - new_size); | |
39279cc3 | 4839 | } |
179e29e4 | 4840 | delete: |
9ddc959e JB |
4841 | /* |
4842 | * We use btrfs_truncate_inode_items() to clean up log trees for | |
4843 | * multiple fsyncs, and in this case we don't want to clear the | |
4844 | * file extent range because it's just the log. | |
4845 | */ | |
50743398 NB |
4846 | if (root == inode->root) { |
4847 | ret = btrfs_inode_clear_file_extent_range(inode, | |
9ddc959e JB |
4848 | clear_start, clear_len); |
4849 | if (ret) { | |
4850 | btrfs_abort_transaction(trans, ret); | |
4851 | break; | |
4852 | } | |
4853 | } | |
4854 | ||
ddfae63c JB |
4855 | if (del_item) |
4856 | last_size = found_key.offset; | |
4857 | else | |
4858 | last_size = new_size; | |
39279cc3 | 4859 | if (del_item) { |
85e21bac CM |
4860 | if (!pending_del_nr) { |
4861 | /* no pending yet, add ourselves */ | |
4862 | pending_del_slot = path->slots[0]; | |
4863 | pending_del_nr = 1; | |
4864 | } else if (pending_del_nr && | |
4865 | path->slots[0] + 1 == pending_del_slot) { | |
4866 | /* hop on the pending chunk */ | |
4867 | pending_del_nr++; | |
4868 | pending_del_slot = path->slots[0]; | |
4869 | } else { | |
d397712b | 4870 | BUG(); |
85e21bac | 4871 | } |
39279cc3 CM |
4872 | } else { |
4873 | break; | |
4874 | } | |
897ca819 | 4875 | should_throttle = false; |
28f75a0e | 4876 | |
27cdeb70 | 4877 | if (found_extent && |
82028e0a | 4878 | root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
ffd4bb2a QW |
4879 | struct btrfs_ref ref = { 0 }; |
4880 | ||
28ed1345 | 4881 | bytes_deleted += extent_num_bytes; |
ffd4bb2a QW |
4882 | |
4883 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, | |
4884 | extent_start, extent_num_bytes, 0); | |
4885 | ref.real_root = root->root_key.objectid; | |
4886 | btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), | |
4887 | ino, extent_offset); | |
4888 | ret = btrfs_free_extent(trans, &ref); | |
05522109 OS |
4889 | if (ret) { |
4890 | btrfs_abort_transaction(trans, ret); | |
4891 | break; | |
4892 | } | |
28f75a0e | 4893 | if (be_nice) { |
7c861627 | 4894 | if (btrfs_should_throttle_delayed_refs(trans)) |
897ca819 | 4895 | should_throttle = true; |
28f75a0e | 4896 | } |
39279cc3 | 4897 | } |
85e21bac | 4898 | |
8082510e YZ |
4899 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4900 | break; | |
4901 | ||
4902 | if (path->slots[0] == 0 || | |
1262133b | 4903 | path->slots[0] != pending_del_slot || |
28bad212 | 4904 | should_throttle) { |
8082510e YZ |
4905 | if (pending_del_nr) { |
4906 | ret = btrfs_del_items(trans, root, path, | |
4907 | pending_del_slot, | |
4908 | pending_del_nr); | |
79787eaa | 4909 | if (ret) { |
66642832 | 4910 | btrfs_abort_transaction(trans, ret); |
fd86a3a3 | 4911 | break; |
79787eaa | 4912 | } |
8082510e YZ |
4913 | pending_del_nr = 0; |
4914 | } | |
b3b4aa74 | 4915 | btrfs_release_path(path); |
28bad212 | 4916 | |
28f75a0e | 4917 | /* |
28bad212 JB |
4918 | * We can generate a lot of delayed refs, so we need to |
4919 | * throttle every once and a while and make sure we're | |
4920 | * adding enough space to keep up with the work we are | |
4921 | * generating. Since we hold a transaction here we | |
4922 | * can't flush, and we don't want to FLUSH_LIMIT because | |
4923 | * we could have generated too many delayed refs to | |
4924 | * actually allocate, so just bail if we're short and | |
4925 | * let the normal reservation dance happen higher up. | |
28f75a0e | 4926 | */ |
28bad212 JB |
4927 | if (should_throttle) { |
4928 | ret = btrfs_delayed_refs_rsv_refill(fs_info, | |
4929 | BTRFS_RESERVE_NO_FLUSH); | |
4930 | if (ret) { | |
4931 | ret = -EAGAIN; | |
4932 | break; | |
4933 | } | |
28f75a0e | 4934 | } |
85e21bac | 4935 | goto search_again; |
8082510e YZ |
4936 | } else { |
4937 | path->slots[0]--; | |
85e21bac | 4938 | } |
39279cc3 | 4939 | } |
8082510e | 4940 | out: |
fd86a3a3 OS |
4941 | if (ret >= 0 && pending_del_nr) { |
4942 | int err; | |
4943 | ||
4944 | err = btrfs_del_items(trans, root, path, pending_del_slot, | |
85e21bac | 4945 | pending_del_nr); |
fd86a3a3 OS |
4946 | if (err) { |
4947 | btrfs_abort_transaction(trans, err); | |
4948 | ret = err; | |
4949 | } | |
85e21bac | 4950 | } |
76b42abb FM |
4951 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
4952 | ASSERT(last_size >= new_size); | |
fd86a3a3 | 4953 | if (!ret && last_size > new_size) |
76b42abb | 4954 | last_size = new_size; |
50743398 NB |
4955 | btrfs_inode_safe_disk_i_size_write(inode, last_size); |
4956 | unlock_extent_cached(&inode->io_tree, lock_start, (u64)-1, | |
4957 | &cached_state); | |
76b42abb | 4958 | } |
28ed1345 | 4959 | |
39279cc3 | 4960 | btrfs_free_path(path); |
fd86a3a3 | 4961 | return ret; |
39279cc3 CM |
4962 | } |
4963 | ||
4964 | /* | |
9703fefe | 4965 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4966 | * @inode - inode that we're zeroing |
4967 | * @from - the offset to start zeroing | |
4968 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4969 | * offset | |
4970 | * @front - zero up to the offset instead of from the offset on | |
4971 | * | |
9703fefe | 4972 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4973 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4974 | */ |
217f42eb NB |
4975 | int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len, |
4976 | int front) | |
39279cc3 | 4977 | { |
217f42eb NB |
4978 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
4979 | struct address_space *mapping = inode->vfs_inode.i_mapping; | |
4980 | struct extent_io_tree *io_tree = &inode->io_tree; | |
e6dcd2dc | 4981 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 4982 | struct extent_state *cached_state = NULL; |
364ecf36 | 4983 | struct extent_changeset *data_reserved = NULL; |
6d4572a9 | 4984 | bool only_release_metadata = false; |
0b246afa | 4985 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 4986 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4987 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4988 | struct page *page; |
3b16a4e3 | 4989 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
6d4572a9 | 4990 | size_t write_bytes = blocksize; |
39279cc3 | 4991 | int ret = 0; |
9703fefe CR |
4992 | u64 block_start; |
4993 | u64 block_end; | |
39279cc3 | 4994 | |
b03ebd99 NB |
4995 | if (IS_ALIGNED(offset, blocksize) && |
4996 | (!len || IS_ALIGNED(len, blocksize))) | |
39279cc3 | 4997 | goto out; |
9703fefe | 4998 | |
8b62f87b JB |
4999 | block_start = round_down(from, blocksize); |
5000 | block_end = block_start + blocksize - 1; | |
5001 | ||
217f42eb NB |
5002 | ret = btrfs_check_data_free_space(inode, &data_reserved, block_start, |
5003 | blocksize); | |
6d4572a9 | 5004 | if (ret < 0) { |
217f42eb | 5005 | if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) { |
6d4572a9 QW |
5006 | /* For nocow case, no need to reserve data space */ |
5007 | only_release_metadata = true; | |
5008 | } else { | |
5009 | goto out; | |
5010 | } | |
5011 | } | |
217f42eb | 5012 | ret = btrfs_delalloc_reserve_metadata(inode, blocksize); |
6d4572a9 QW |
5013 | if (ret < 0) { |
5014 | if (!only_release_metadata) | |
217f42eb NB |
5015 | btrfs_free_reserved_data_space(inode, data_reserved, |
5016 | block_start, blocksize); | |
6d4572a9 QW |
5017 | goto out; |
5018 | } | |
211c17f5 | 5019 | again: |
3b16a4e3 | 5020 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 5021 | if (!page) { |
217f42eb NB |
5022 | btrfs_delalloc_release_space(inode, data_reserved, block_start, |
5023 | blocksize, true); | |
5024 | btrfs_delalloc_release_extents(inode, blocksize); | |
ac6a2b36 | 5025 | ret = -ENOMEM; |
39279cc3 | 5026 | goto out; |
5d5e103a | 5027 | } |
32443de3 QW |
5028 | ret = set_page_extent_mapped(page); |
5029 | if (ret < 0) | |
5030 | goto out_unlock; | |
e6dcd2dc | 5031 | |
39279cc3 | 5032 | if (!PageUptodate(page)) { |
9ebefb18 | 5033 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 5034 | lock_page(page); |
211c17f5 CM |
5035 | if (page->mapping != mapping) { |
5036 | unlock_page(page); | |
09cbfeaf | 5037 | put_page(page); |
211c17f5 CM |
5038 | goto again; |
5039 | } | |
39279cc3 CM |
5040 | if (!PageUptodate(page)) { |
5041 | ret = -EIO; | |
89642229 | 5042 | goto out_unlock; |
39279cc3 CM |
5043 | } |
5044 | } | |
211c17f5 | 5045 | wait_on_page_writeback(page); |
e6dcd2dc | 5046 | |
9703fefe | 5047 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc | 5048 | |
217f42eb | 5049 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 5050 | if (ordered) { |
9703fefe | 5051 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 5052 | &cached_state); |
e6dcd2dc | 5053 | unlock_page(page); |
09cbfeaf | 5054 | put_page(page); |
c0a43603 | 5055 | btrfs_start_ordered_extent(ordered, 1); |
e6dcd2dc CM |
5056 | btrfs_put_ordered_extent(ordered); |
5057 | goto again; | |
5058 | } | |
5059 | ||
217f42eb | 5060 | clear_extent_bit(&inode->io_tree, block_start, block_end, |
e182163d OS |
5061 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, |
5062 | 0, 0, &cached_state); | |
5d5e103a | 5063 | |
217f42eb | 5064 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0, |
330a5827 | 5065 | &cached_state); |
9ed74f2d | 5066 | if (ret) { |
9703fefe | 5067 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 5068 | &cached_state); |
9ed74f2d JB |
5069 | goto out_unlock; |
5070 | } | |
5071 | ||
9703fefe | 5072 | if (offset != blocksize) { |
2aaa6655 | 5073 | if (!len) |
9703fefe | 5074 | len = blocksize - offset; |
2aaa6655 | 5075 | if (front) |
d048b9c2 IW |
5076 | memzero_page(page, (block_start - page_offset(page)), |
5077 | offset); | |
2aaa6655 | 5078 | else |
d048b9c2 IW |
5079 | memzero_page(page, (block_start - page_offset(page)) + offset, |
5080 | len); | |
e6dcd2dc | 5081 | flush_dcache_page(page); |
e6dcd2dc | 5082 | } |
247e743c | 5083 | ClearPageChecked(page); |
6c9ac8be | 5084 | btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start); |
e43bbe5e | 5085 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state); |
39279cc3 | 5086 | |
6d4572a9 | 5087 | if (only_release_metadata) |
217f42eb | 5088 | set_extent_bit(&inode->io_tree, block_start, block_end, |
1cab5e72 | 5089 | EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL); |
6d4572a9 | 5090 | |
89642229 | 5091 | out_unlock: |
6d4572a9 QW |
5092 | if (ret) { |
5093 | if (only_release_metadata) | |
217f42eb | 5094 | btrfs_delalloc_release_metadata(inode, blocksize, true); |
6d4572a9 | 5095 | else |
217f42eb | 5096 | btrfs_delalloc_release_space(inode, data_reserved, |
6d4572a9 QW |
5097 | block_start, blocksize, true); |
5098 | } | |
217f42eb | 5099 | btrfs_delalloc_release_extents(inode, blocksize); |
39279cc3 | 5100 | unlock_page(page); |
09cbfeaf | 5101 | put_page(page); |
39279cc3 | 5102 | out: |
6d4572a9 | 5103 | if (only_release_metadata) |
217f42eb | 5104 | btrfs_check_nocow_unlock(inode); |
364ecf36 | 5105 | extent_changeset_free(data_reserved); |
39279cc3 CM |
5106 | return ret; |
5107 | } | |
5108 | ||
a4ba6cc0 | 5109 | static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode, |
16e7549f JB |
5110 | u64 offset, u64 len) |
5111 | { | |
a4ba6cc0 | 5112 | struct btrfs_fs_info *fs_info = root->fs_info; |
16e7549f | 5113 | struct btrfs_trans_handle *trans; |
5893dfb9 | 5114 | struct btrfs_drop_extents_args drop_args = { 0 }; |
16e7549f JB |
5115 | int ret; |
5116 | ||
5117 | /* | |
cceaa89f FM |
5118 | * If NO_HOLES is enabled, we don't need to do anything. |
5119 | * Later, up in the call chain, either btrfs_set_inode_last_sub_trans() | |
5120 | * or btrfs_update_inode() will be called, which guarantee that the next | |
5121 | * fsync will know this inode was changed and needs to be logged. | |
16e7549f | 5122 | */ |
cceaa89f | 5123 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) |
16e7549f | 5124 | return 0; |
16e7549f JB |
5125 | |
5126 | /* | |
5127 | * 1 - for the one we're dropping | |
5128 | * 1 - for the one we're adding | |
5129 | * 1 - for updating the inode. | |
5130 | */ | |
5131 | trans = btrfs_start_transaction(root, 3); | |
5132 | if (IS_ERR(trans)) | |
5133 | return PTR_ERR(trans); | |
5134 | ||
5893dfb9 FM |
5135 | drop_args.start = offset; |
5136 | drop_args.end = offset + len; | |
5137 | drop_args.drop_cache = true; | |
5138 | ||
a4ba6cc0 | 5139 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
16e7549f | 5140 | if (ret) { |
66642832 | 5141 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 5142 | btrfs_end_transaction(trans); |
16e7549f JB |
5143 | return ret; |
5144 | } | |
5145 | ||
a4ba6cc0 | 5146 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), |
f85b7379 | 5147 | offset, 0, 0, len, 0, len, 0, 0, 0); |
2766ff61 | 5148 | if (ret) { |
66642832 | 5149 | btrfs_abort_transaction(trans, ret); |
2766ff61 | 5150 | } else { |
a4ba6cc0 NB |
5151 | btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found); |
5152 | btrfs_update_inode(trans, root, inode); | |
2766ff61 | 5153 | } |
3a45bb20 | 5154 | btrfs_end_transaction(trans); |
16e7549f JB |
5155 | return ret; |
5156 | } | |
5157 | ||
695a0d0d JB |
5158 | /* |
5159 | * This function puts in dummy file extents for the area we're creating a hole | |
5160 | * for. So if we are truncating this file to a larger size we need to insert | |
5161 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
5162 | * the range between oldsize and size | |
5163 | */ | |
b06359a3 | 5164 | int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 5165 | { |
b06359a3 NB |
5166 | struct btrfs_root *root = inode->root; |
5167 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5168 | struct extent_io_tree *io_tree = &inode->io_tree; | |
a22285a6 | 5169 | struct extent_map *em = NULL; |
2ac55d41 | 5170 | struct extent_state *cached_state = NULL; |
b06359a3 | 5171 | struct extent_map_tree *em_tree = &inode->extent_tree; |
0b246afa JM |
5172 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
5173 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
5174 | u64 last_byte; |
5175 | u64 cur_offset; | |
5176 | u64 hole_size; | |
9ed74f2d | 5177 | int err = 0; |
39279cc3 | 5178 | |
a71754fc | 5179 | /* |
9703fefe CR |
5180 | * If our size started in the middle of a block we need to zero out the |
5181 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
5182 | * expose stale data. |
5183 | */ | |
b06359a3 | 5184 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
5185 | if (err) |
5186 | return err; | |
5187 | ||
9036c102 YZ |
5188 | if (size <= hole_start) |
5189 | return 0; | |
5190 | ||
b06359a3 NB |
5191 | btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1, |
5192 | &cached_state); | |
9036c102 YZ |
5193 | cur_offset = hole_start; |
5194 | while (1) { | |
b06359a3 | 5195 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, |
39b07b5d | 5196 | block_end - cur_offset); |
79787eaa JM |
5197 | if (IS_ERR(em)) { |
5198 | err = PTR_ERR(em); | |
f2767956 | 5199 | em = NULL; |
79787eaa JM |
5200 | break; |
5201 | } | |
9036c102 | 5202 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 5203 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
9ddc959e JB |
5204 | hole_size = last_byte - cur_offset; |
5205 | ||
8082510e | 5206 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 5207 | struct extent_map *hole_em; |
9ed74f2d | 5208 | |
b06359a3 NB |
5209 | err = maybe_insert_hole(root, inode, cur_offset, |
5210 | hole_size); | |
16e7549f | 5211 | if (err) |
3893e33b | 5212 | break; |
9ddc959e | 5213 | |
b06359a3 | 5214 | err = btrfs_inode_set_file_extent_range(inode, |
9ddc959e JB |
5215 | cur_offset, hole_size); |
5216 | if (err) | |
5217 | break; | |
5218 | ||
b06359a3 | 5219 | btrfs_drop_extent_cache(inode, cur_offset, |
5dc562c5 JB |
5220 | cur_offset + hole_size - 1, 0); |
5221 | hole_em = alloc_extent_map(); | |
5222 | if (!hole_em) { | |
5223 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
b06359a3 | 5224 | &inode->runtime_flags); |
5dc562c5 JB |
5225 | goto next; |
5226 | } | |
5227 | hole_em->start = cur_offset; | |
5228 | hole_em->len = hole_size; | |
5229 | hole_em->orig_start = cur_offset; | |
8082510e | 5230 | |
5dc562c5 JB |
5231 | hole_em->block_start = EXTENT_MAP_HOLE; |
5232 | hole_em->block_len = 0; | |
b4939680 | 5233 | hole_em->orig_block_len = 0; |
cc95bef6 | 5234 | hole_em->ram_bytes = hole_size; |
5dc562c5 | 5235 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 5236 | hole_em->generation = fs_info->generation; |
8082510e | 5237 | |
5dc562c5 JB |
5238 | while (1) { |
5239 | write_lock(&em_tree->lock); | |
09a2a8f9 | 5240 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
5241 | write_unlock(&em_tree->lock); |
5242 | if (err != -EEXIST) | |
5243 | break; | |
b06359a3 | 5244 | btrfs_drop_extent_cache(inode, cur_offset, |
5dc562c5 JB |
5245 | cur_offset + |
5246 | hole_size - 1, 0); | |
5247 | } | |
5248 | free_extent_map(hole_em); | |
9ddc959e | 5249 | } else { |
b06359a3 | 5250 | err = btrfs_inode_set_file_extent_range(inode, |
9ddc959e JB |
5251 | cur_offset, hole_size); |
5252 | if (err) | |
5253 | break; | |
9036c102 | 5254 | } |
16e7549f | 5255 | next: |
9036c102 | 5256 | free_extent_map(em); |
a22285a6 | 5257 | em = NULL; |
9036c102 | 5258 | cur_offset = last_byte; |
8082510e | 5259 | if (cur_offset >= block_end) |
9036c102 YZ |
5260 | break; |
5261 | } | |
a22285a6 | 5262 | free_extent_map(em); |
e43bbe5e | 5263 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state); |
9036c102 YZ |
5264 | return err; |
5265 | } | |
39279cc3 | 5266 | |
3972f260 | 5267 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 5268 | { |
f4a2f4c5 MX |
5269 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5270 | struct btrfs_trans_handle *trans; | |
a41ad394 | 5271 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
5272 | loff_t newsize = attr->ia_size; |
5273 | int mask = attr->ia_valid; | |
8082510e YZ |
5274 | int ret; |
5275 | ||
3972f260 ES |
5276 | /* |
5277 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
5278 | * special case where we need to update the times despite not having | |
5279 | * these flags set. For all other operations the VFS set these flags | |
5280 | * explicitly if it wants a timestamp update. | |
5281 | */ | |
dff6efc3 CH |
5282 | if (newsize != oldsize) { |
5283 | inode_inc_iversion(inode); | |
5284 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
5285 | inode->i_ctime = inode->i_mtime = | |
c2050a45 | 5286 | current_time(inode); |
dff6efc3 | 5287 | } |
3972f260 | 5288 | |
a41ad394 | 5289 | if (newsize > oldsize) { |
9ea24bbe | 5290 | /* |
ea14b57f | 5291 | * Don't do an expanding truncate while snapshotting is ongoing. |
9ea24bbe FM |
5292 | * This is to ensure the snapshot captures a fully consistent |
5293 | * state of this file - if the snapshot captures this expanding | |
5294 | * truncation, it must capture all writes that happened before | |
5295 | * this truncation. | |
5296 | */ | |
dcc3eb96 | 5297 | btrfs_drew_write_lock(&root->snapshot_lock); |
b06359a3 | 5298 | ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize); |
9ea24bbe | 5299 | if (ret) { |
dcc3eb96 | 5300 | btrfs_drew_write_unlock(&root->snapshot_lock); |
8082510e | 5301 | return ret; |
9ea24bbe | 5302 | } |
8082510e | 5303 | |
f4a2f4c5 | 5304 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe | 5305 | if (IS_ERR(trans)) { |
dcc3eb96 | 5306 | btrfs_drew_write_unlock(&root->snapshot_lock); |
f4a2f4c5 | 5307 | return PTR_ERR(trans); |
9ea24bbe | 5308 | } |
f4a2f4c5 MX |
5309 | |
5310 | i_size_write(inode, newsize); | |
76aea537 | 5311 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
27772b68 | 5312 | pagecache_isize_extended(inode, oldsize, newsize); |
9a56fcd1 | 5313 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
dcc3eb96 | 5314 | btrfs_drew_write_unlock(&root->snapshot_lock); |
3a45bb20 | 5315 | btrfs_end_transaction(trans); |
a41ad394 | 5316 | } else { |
24c0a722 NA |
5317 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5318 | ||
5319 | if (btrfs_is_zoned(fs_info)) { | |
5320 | ret = btrfs_wait_ordered_range(inode, | |
5321 | ALIGN(newsize, fs_info->sectorsize), | |
5322 | (u64)-1); | |
5323 | if (ret) | |
5324 | return ret; | |
5325 | } | |
8082510e | 5326 | |
a41ad394 JB |
5327 | /* |
5328 | * We're truncating a file that used to have good data down to | |
1fd4033d NB |
5329 | * zero. Make sure any new writes to the file get on disk |
5330 | * on close. | |
a41ad394 JB |
5331 | */ |
5332 | if (newsize == 0) | |
1fd4033d | 5333 | set_bit(BTRFS_INODE_FLUSH_ON_CLOSE, |
72ac3c0d | 5334 | &BTRFS_I(inode)->runtime_flags); |
8082510e | 5335 | |
a41ad394 | 5336 | truncate_setsize(inode, newsize); |
2e60a51e | 5337 | |
2e60a51e | 5338 | inode_dio_wait(inode); |
2e60a51e | 5339 | |
213e8c55 | 5340 | ret = btrfs_truncate(inode, newsize == oldsize); |
7f4f6e0a JB |
5341 | if (ret && inode->i_nlink) { |
5342 | int err; | |
5343 | ||
5344 | /* | |
f7e9e8fc OS |
5345 | * Truncate failed, so fix up the in-memory size. We |
5346 | * adjusted disk_i_size down as we removed extents, so | |
5347 | * wait for disk_i_size to be stable and then update the | |
5348 | * in-memory size to match. | |
7f4f6e0a | 5349 | */ |
f7e9e8fc | 5350 | err = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
7f4f6e0a | 5351 | if (err) |
f7e9e8fc OS |
5352 | return err; |
5353 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
7f4f6e0a | 5354 | } |
8082510e YZ |
5355 | } |
5356 | ||
a41ad394 | 5357 | return ret; |
8082510e YZ |
5358 | } |
5359 | ||
549c7297 CB |
5360 | static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, |
5361 | struct iattr *attr) | |
9036c102 | 5362 | { |
2b0143b5 | 5363 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5364 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5365 | int err; |
39279cc3 | 5366 | |
b83cc969 LZ |
5367 | if (btrfs_root_readonly(root)) |
5368 | return -EROFS; | |
5369 | ||
2f221d6f | 5370 | err = setattr_prepare(&init_user_ns, dentry, attr); |
9036c102 YZ |
5371 | if (err) |
5372 | return err; | |
2bf5a725 | 5373 | |
5a3f23d5 | 5374 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5375 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5376 | if (err) |
5377 | return err; | |
39279cc3 | 5378 | } |
9036c102 | 5379 | |
1025774c | 5380 | if (attr->ia_valid) { |
2f221d6f | 5381 | setattr_copy(&init_user_ns, inode, attr); |
0c4d2d95 | 5382 | inode_inc_iversion(inode); |
22c44fe6 | 5383 | err = btrfs_dirty_inode(inode); |
1025774c | 5384 | |
22c44fe6 | 5385 | if (!err && attr->ia_valid & ATTR_MODE) |
e65ce2a5 CB |
5386 | err = posix_acl_chmod(&init_user_ns, inode, |
5387 | inode->i_mode); | |
1025774c | 5388 | } |
33268eaf | 5389 | |
39279cc3 CM |
5390 | return err; |
5391 | } | |
61295eb8 | 5392 | |
131e404a FDBM |
5393 | /* |
5394 | * While truncating the inode pages during eviction, we get the VFS calling | |
5395 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
5396 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
5397 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
5398 | * extent_state structures over and over, wasting lots of time. | |
5399 | * | |
5400 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
5401 | * those expensive operations on a per page basis and do only the ordered io | |
5402 | * finishing, while we release here the extent_map and extent_state structures, | |
5403 | * without the excessive merging and splitting. | |
5404 | */ | |
5405 | static void evict_inode_truncate_pages(struct inode *inode) | |
5406 | { | |
5407 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5408 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5409 | struct rb_node *node; | |
5410 | ||
5411 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5412 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5413 | |
5414 | write_lock(&map_tree->lock); | |
07e1ce09 | 5415 | while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) { |
131e404a FDBM |
5416 | struct extent_map *em; |
5417 | ||
07e1ce09 | 5418 | node = rb_first_cached(&map_tree->map); |
131e404a | 5419 | em = rb_entry(node, struct extent_map, rb_node); |
180589ef WS |
5420 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5421 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5422 | remove_extent_mapping(map_tree, em); |
5423 | free_extent_map(em); | |
7064dd5c FM |
5424 | if (need_resched()) { |
5425 | write_unlock(&map_tree->lock); | |
5426 | cond_resched(); | |
5427 | write_lock(&map_tree->lock); | |
5428 | } | |
131e404a FDBM |
5429 | } |
5430 | write_unlock(&map_tree->lock); | |
5431 | ||
6ca07097 FM |
5432 | /* |
5433 | * Keep looping until we have no more ranges in the io tree. | |
ba206a02 MWO |
5434 | * We can have ongoing bios started by readahead that have |
5435 | * their endio callback (extent_io.c:end_bio_extent_readpage) | |
9c6429d9 FM |
5436 | * still in progress (unlocked the pages in the bio but did not yet |
5437 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5438 | * ranges can still be locked and eviction started because before |
5439 | * submitting those bios, which are executed by a separate task (work | |
5440 | * queue kthread), inode references (inode->i_count) were not taken | |
5441 | * (which would be dropped in the end io callback of each bio). | |
5442 | * Therefore here we effectively end up waiting for those bios and | |
5443 | * anyone else holding locked ranges without having bumped the inode's | |
5444 | * reference count - if we don't do it, when they access the inode's | |
5445 | * io_tree to unlock a range it may be too late, leading to an | |
5446 | * use-after-free issue. | |
5447 | */ | |
131e404a FDBM |
5448 | spin_lock(&io_tree->lock); |
5449 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5450 | struct extent_state *state; | |
5451 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5452 | u64 start; |
5453 | u64 end; | |
421f0922 | 5454 | unsigned state_flags; |
131e404a FDBM |
5455 | |
5456 | node = rb_first(&io_tree->state); | |
5457 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5458 | start = state->start; |
5459 | end = state->end; | |
421f0922 | 5460 | state_flags = state->state; |
131e404a FDBM |
5461 | spin_unlock(&io_tree->lock); |
5462 | ||
ff13db41 | 5463 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5464 | |
5465 | /* | |
5466 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5467 | * and its reserved space won't be freed by delayed_ref. | |
5468 | * So we need to free its reserved space here. | |
5469 | * (Refer to comment in btrfs_invalidatepage, case 2) | |
5470 | * | |
5471 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5472 | */ | |
421f0922 | 5473 | if (state_flags & EXTENT_DELALLOC) |
8b8a979f NB |
5474 | btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start, |
5475 | end - start + 1); | |
b9d0b389 | 5476 | |
6ca07097 | 5477 | clear_extent_bit(io_tree, start, end, |
e182163d OS |
5478 | EXTENT_LOCKED | EXTENT_DELALLOC | |
5479 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, | |
5480 | &cached_state); | |
131e404a | 5481 | |
7064dd5c | 5482 | cond_resched(); |
131e404a FDBM |
5483 | spin_lock(&io_tree->lock); |
5484 | } | |
5485 | spin_unlock(&io_tree->lock); | |
5486 | } | |
5487 | ||
4b9d7b59 | 5488 | static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root, |
ad80cf50 | 5489 | struct btrfs_block_rsv *rsv) |
4b9d7b59 OS |
5490 | { |
5491 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5492 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d3984c90 | 5493 | struct btrfs_trans_handle *trans; |
2bd36e7b | 5494 | u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1); |
d3984c90 | 5495 | int ret; |
4b9d7b59 | 5496 | |
d3984c90 JB |
5497 | /* |
5498 | * Eviction should be taking place at some place safe because of our | |
5499 | * delayed iputs. However the normal flushing code will run delayed | |
5500 | * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock. | |
5501 | * | |
5502 | * We reserve the delayed_refs_extra here again because we can't use | |
5503 | * btrfs_start_transaction(root, 0) for the same deadlocky reason as | |
5504 | * above. We reserve our extra bit here because we generate a ton of | |
5505 | * delayed refs activity by truncating. | |
5506 | * | |
5507 | * If we cannot make our reservation we'll attempt to steal from the | |
5508 | * global reserve, because we really want to be able to free up space. | |
5509 | */ | |
5510 | ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra, | |
5511 | BTRFS_RESERVE_FLUSH_EVICT); | |
5512 | if (ret) { | |
4b9d7b59 OS |
5513 | /* |
5514 | * Try to steal from the global reserve if there is space for | |
5515 | * it. | |
5516 | */ | |
d3984c90 JB |
5517 | if (btrfs_check_space_for_delayed_refs(fs_info) || |
5518 | btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) { | |
5519 | btrfs_warn(fs_info, | |
5520 | "could not allocate space for delete; will truncate on mount"); | |
5521 | return ERR_PTR(-ENOSPC); | |
5522 | } | |
5523 | delayed_refs_extra = 0; | |
5524 | } | |
4b9d7b59 | 5525 | |
d3984c90 JB |
5526 | trans = btrfs_join_transaction(root); |
5527 | if (IS_ERR(trans)) | |
5528 | return trans; | |
5529 | ||
5530 | if (delayed_refs_extra) { | |
5531 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5532 | trans->bytes_reserved = delayed_refs_extra; | |
5533 | btrfs_block_rsv_migrate(rsv, trans->block_rsv, | |
5534 | delayed_refs_extra, 1); | |
4b9d7b59 | 5535 | } |
d3984c90 | 5536 | return trans; |
4b9d7b59 OS |
5537 | } |
5538 | ||
bd555975 | 5539 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5540 | { |
0b246afa | 5541 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5542 | struct btrfs_trans_handle *trans; |
5543 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b9d7b59 | 5544 | struct btrfs_block_rsv *rsv; |
39279cc3 CM |
5545 | int ret; |
5546 | ||
1abe9b8a | 5547 | trace_btrfs_inode_evict(inode); |
5548 | ||
3d48d981 | 5549 | if (!root) { |
e8f1bc14 | 5550 | clear_inode(inode); |
3d48d981 NB |
5551 | return; |
5552 | } | |
5553 | ||
131e404a FDBM |
5554 | evict_inode_truncate_pages(inode); |
5555 | ||
69e9c6c6 SB |
5556 | if (inode->i_nlink && |
5557 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5558 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
70ddc553 | 5559 | btrfs_is_free_space_inode(BTRFS_I(inode)))) |
bd555975 AV |
5560 | goto no_delete; |
5561 | ||
27919067 | 5562 | if (is_bad_inode(inode)) |
39279cc3 | 5563 | goto no_delete; |
5f39d397 | 5564 | |
7ab7956e | 5565 | btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1); |
f612496b | 5566 | |
7b40b695 | 5567 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
c71bf099 | 5568 | goto no_delete; |
c71bf099 | 5569 | |
76dda93c | 5570 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5571 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5572 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5573 | goto no_delete; |
5574 | } | |
5575 | ||
aa79021f | 5576 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
27919067 | 5577 | if (ret) |
0e8c36a9 | 5578 | goto no_delete; |
0e8c36a9 | 5579 | |
2ff7e61e | 5580 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
27919067 | 5581 | if (!rsv) |
4289a667 | 5582 | goto no_delete; |
2bd36e7b | 5583 | rsv->size = btrfs_calc_metadata_size(fs_info, 1); |
ca7e70f5 | 5584 | rsv->failfast = 1; |
4289a667 | 5585 | |
6ef06d27 | 5586 | btrfs_i_size_write(BTRFS_I(inode), 0); |
5f39d397 | 5587 | |
8082510e | 5588 | while (1) { |
ad80cf50 | 5589 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5590 | if (IS_ERR(trans)) |
5591 | goto free_rsv; | |
7b128766 | 5592 | |
4289a667 JB |
5593 | trans->block_rsv = rsv; |
5594 | ||
50743398 | 5595 | ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode), |
0d7d3165 | 5596 | 0, 0, NULL); |
27919067 OS |
5597 | trans->block_rsv = &fs_info->trans_block_rsv; |
5598 | btrfs_end_transaction(trans); | |
5599 | btrfs_btree_balance_dirty(fs_info); | |
5600 | if (ret && ret != -ENOSPC && ret != -EAGAIN) | |
5601 | goto free_rsv; | |
5602 | else if (!ret) | |
8082510e | 5603 | break; |
8082510e | 5604 | } |
5f39d397 | 5605 | |
4ef31a45 | 5606 | /* |
27919067 OS |
5607 | * Errors here aren't a big deal, it just means we leave orphan items in |
5608 | * the tree. They will be cleaned up on the next mount. If the inode | |
5609 | * number gets reused, cleanup deletes the orphan item without doing | |
5610 | * anything, and unlink reuses the existing orphan item. | |
5611 | * | |
5612 | * If it turns out that we are dropping too many of these, we might want | |
5613 | * to add a mechanism for retrying these after a commit. | |
4ef31a45 | 5614 | */ |
ad80cf50 | 5615 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5616 | if (!IS_ERR(trans)) { |
5617 | trans->block_rsv = rsv; | |
5618 | btrfs_orphan_del(trans, BTRFS_I(inode)); | |
5619 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5620 | btrfs_end_transaction(trans); | |
5621 | } | |
54aa1f4d | 5622 | |
27919067 OS |
5623 | free_rsv: |
5624 | btrfs_free_block_rsv(fs_info, rsv); | |
39279cc3 | 5625 | no_delete: |
27919067 OS |
5626 | /* |
5627 | * If we didn't successfully delete, the orphan item will still be in | |
5628 | * the tree and we'll retry on the next mount. Again, we might also want | |
5629 | * to retry these periodically in the future. | |
5630 | */ | |
f48d1cf5 | 5631 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
dbd5768f | 5632 | clear_inode(inode); |
39279cc3 CM |
5633 | } |
5634 | ||
5635 | /* | |
6bf9e4bd QW |
5636 | * Return the key found in the dir entry in the location pointer, fill @type |
5637 | * with BTRFS_FT_*, and return 0. | |
5638 | * | |
005d6712 SY |
5639 | * If no dir entries were found, returns -ENOENT. |
5640 | * If found a corrupted location in dir entry, returns -EUCLEAN. | |
39279cc3 CM |
5641 | */ |
5642 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
6bf9e4bd | 5643 | struct btrfs_key *location, u8 *type) |
39279cc3 CM |
5644 | { |
5645 | const char *name = dentry->d_name.name; | |
5646 | int namelen = dentry->d_name.len; | |
5647 | struct btrfs_dir_item *di; | |
5648 | struct btrfs_path *path; | |
5649 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5650 | int ret = 0; |
39279cc3 CM |
5651 | |
5652 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5653 | if (!path) |
5654 | return -ENOMEM; | |
3954401f | 5655 | |
f85b7379 DS |
5656 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)), |
5657 | name, namelen, 0); | |
3cf5068f LB |
5658 | if (IS_ERR_OR_NULL(di)) { |
5659 | ret = di ? PTR_ERR(di) : -ENOENT; | |
005d6712 SY |
5660 | goto out; |
5661 | } | |
d397712b | 5662 | |
5f39d397 | 5663 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
56a0e706 LB |
5664 | if (location->type != BTRFS_INODE_ITEM_KEY && |
5665 | location->type != BTRFS_ROOT_ITEM_KEY) { | |
005d6712 | 5666 | ret = -EUCLEAN; |
56a0e706 LB |
5667 | btrfs_warn(root->fs_info, |
5668 | "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))", | |
5669 | __func__, name, btrfs_ino(BTRFS_I(dir)), | |
5670 | location->objectid, location->type, location->offset); | |
56a0e706 | 5671 | } |
6bf9e4bd QW |
5672 | if (!ret) |
5673 | *type = btrfs_dir_type(path->nodes[0], di); | |
39279cc3 | 5674 | out: |
39279cc3 CM |
5675 | btrfs_free_path(path); |
5676 | return ret; | |
5677 | } | |
5678 | ||
5679 | /* | |
5680 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5681 | * needs to be changed to reflect the root directory of the tree root. This | |
5682 | * is kind of like crossing a mount point. | |
5683 | */ | |
2ff7e61e | 5684 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
4df27c4d YZ |
5685 | struct inode *dir, |
5686 | struct dentry *dentry, | |
5687 | struct btrfs_key *location, | |
5688 | struct btrfs_root **sub_root) | |
39279cc3 | 5689 | { |
4df27c4d YZ |
5690 | struct btrfs_path *path; |
5691 | struct btrfs_root *new_root; | |
5692 | struct btrfs_root_ref *ref; | |
5693 | struct extent_buffer *leaf; | |
1d4c08e0 | 5694 | struct btrfs_key key; |
4df27c4d YZ |
5695 | int ret; |
5696 | int err = 0; | |
39279cc3 | 5697 | |
4df27c4d YZ |
5698 | path = btrfs_alloc_path(); |
5699 | if (!path) { | |
5700 | err = -ENOMEM; | |
5701 | goto out; | |
5702 | } | |
39279cc3 | 5703 | |
4df27c4d | 5704 | err = -ENOENT; |
1d4c08e0 DS |
5705 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5706 | key.type = BTRFS_ROOT_REF_KEY; | |
5707 | key.offset = location->objectid; | |
5708 | ||
0b246afa | 5709 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5710 | if (ret) { |
5711 | if (ret < 0) | |
5712 | err = ret; | |
5713 | goto out; | |
5714 | } | |
39279cc3 | 5715 | |
4df27c4d YZ |
5716 | leaf = path->nodes[0]; |
5717 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
4a0cc7ca | 5718 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) || |
4df27c4d YZ |
5719 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5720 | goto out; | |
39279cc3 | 5721 | |
4df27c4d YZ |
5722 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5723 | (unsigned long)(ref + 1), | |
5724 | dentry->d_name.len); | |
5725 | if (ret) | |
5726 | goto out; | |
5727 | ||
b3b4aa74 | 5728 | btrfs_release_path(path); |
4df27c4d | 5729 | |
56e9357a | 5730 | new_root = btrfs_get_fs_root(fs_info, location->objectid, true); |
4df27c4d YZ |
5731 | if (IS_ERR(new_root)) { |
5732 | err = PTR_ERR(new_root); | |
5733 | goto out; | |
5734 | } | |
5735 | ||
4df27c4d YZ |
5736 | *sub_root = new_root; |
5737 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5738 | location->type = BTRFS_INODE_ITEM_KEY; | |
5739 | location->offset = 0; | |
5740 | err = 0; | |
5741 | out: | |
5742 | btrfs_free_path(path); | |
5743 | return err; | |
39279cc3 CM |
5744 | } |
5745 | ||
5d4f98a2 YZ |
5746 | static void inode_tree_add(struct inode *inode) |
5747 | { | |
5748 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5749 | struct btrfs_inode *entry; | |
03e860bd FNP |
5750 | struct rb_node **p; |
5751 | struct rb_node *parent; | |
cef21937 | 5752 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
4a0cc7ca | 5753 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 5754 | |
1d3382cb | 5755 | if (inode_unhashed(inode)) |
76dda93c | 5756 | return; |
e1409cef | 5757 | parent = NULL; |
5d4f98a2 | 5758 | spin_lock(&root->inode_lock); |
e1409cef | 5759 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5760 | while (*p) { |
5761 | parent = *p; | |
5762 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5763 | ||
37508515 | 5764 | if (ino < btrfs_ino(entry)) |
03e860bd | 5765 | p = &parent->rb_left; |
37508515 | 5766 | else if (ino > btrfs_ino(entry)) |
03e860bd | 5767 | p = &parent->rb_right; |
5d4f98a2 YZ |
5768 | else { |
5769 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5770 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5771 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5772 | RB_CLEAR_NODE(parent); |
5773 | spin_unlock(&root->inode_lock); | |
cef21937 | 5774 | return; |
5d4f98a2 YZ |
5775 | } |
5776 | } | |
cef21937 FDBM |
5777 | rb_link_node(new, parent, p); |
5778 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5779 | spin_unlock(&root->inode_lock); |
5780 | } | |
5781 | ||
b79b7249 | 5782 | static void inode_tree_del(struct btrfs_inode *inode) |
5d4f98a2 | 5783 | { |
b79b7249 | 5784 | struct btrfs_root *root = inode->root; |
76dda93c | 5785 | int empty = 0; |
5d4f98a2 | 5786 | |
03e860bd | 5787 | spin_lock(&root->inode_lock); |
b79b7249 NB |
5788 | if (!RB_EMPTY_NODE(&inode->rb_node)) { |
5789 | rb_erase(&inode->rb_node, &root->inode_tree); | |
5790 | RB_CLEAR_NODE(&inode->rb_node); | |
76dda93c | 5791 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5792 | } |
03e860bd | 5793 | spin_unlock(&root->inode_lock); |
76dda93c | 5794 | |
69e9c6c6 | 5795 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5796 | spin_lock(&root->inode_lock); |
5797 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5798 | spin_unlock(&root->inode_lock); | |
5799 | if (empty) | |
5800 | btrfs_add_dead_root(root); | |
5801 | } | |
5802 | } | |
5803 | ||
5d4f98a2 | 5804 | |
e02119d5 CM |
5805 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5806 | { | |
5807 | struct btrfs_iget_args *args = p; | |
0202e83f DS |
5808 | |
5809 | inode->i_ino = args->ino; | |
5810 | BTRFS_I(inode)->location.objectid = args->ino; | |
5811 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
5812 | BTRFS_I(inode)->location.offset = 0; | |
5c8fd99f JB |
5813 | BTRFS_I(inode)->root = btrfs_grab_root(args->root); |
5814 | BUG_ON(args->root && !BTRFS_I(inode)->root); | |
39279cc3 CM |
5815 | return 0; |
5816 | } | |
5817 | ||
5818 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5819 | { | |
5820 | struct btrfs_iget_args *args = opaque; | |
0202e83f DS |
5821 | |
5822 | return args->ino == BTRFS_I(inode)->location.objectid && | |
d397712b | 5823 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5824 | } |
5825 | ||
0202e83f | 5826 | static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino, |
5d4f98a2 | 5827 | struct btrfs_root *root) |
39279cc3 CM |
5828 | { |
5829 | struct inode *inode; | |
5830 | struct btrfs_iget_args args; | |
0202e83f | 5831 | unsigned long hashval = btrfs_inode_hash(ino, root); |
778ba82b | 5832 | |
0202e83f | 5833 | args.ino = ino; |
39279cc3 CM |
5834 | args.root = root; |
5835 | ||
778ba82b | 5836 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5837 | btrfs_init_locked_inode, |
5838 | (void *)&args); | |
5839 | return inode; | |
5840 | } | |
5841 | ||
4c66e0d4 | 5842 | /* |
0202e83f | 5843 | * Get an inode object given its inode number and corresponding root. |
4c66e0d4 DS |
5844 | * Path can be preallocated to prevent recursing back to iget through |
5845 | * allocator. NULL is also valid but may require an additional allocation | |
5846 | * later. | |
1a54ef8c | 5847 | */ |
0202e83f | 5848 | struct inode *btrfs_iget_path(struct super_block *s, u64 ino, |
4c66e0d4 | 5849 | struct btrfs_root *root, struct btrfs_path *path) |
1a54ef8c BR |
5850 | { |
5851 | struct inode *inode; | |
5852 | ||
0202e83f | 5853 | inode = btrfs_iget_locked(s, ino, root); |
1a54ef8c | 5854 | if (!inode) |
5d4f98a2 | 5855 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5856 | |
5857 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5858 | int ret; |
5859 | ||
4222ea71 | 5860 | ret = btrfs_read_locked_inode(inode, path); |
9bc2ceff | 5861 | if (!ret) { |
1748f843 MF |
5862 | inode_tree_add(inode); |
5863 | unlock_new_inode(inode); | |
1748f843 | 5864 | } else { |
f5b3a417 AV |
5865 | iget_failed(inode); |
5866 | /* | |
5867 | * ret > 0 can come from btrfs_search_slot called by | |
5868 | * btrfs_read_locked_inode, this means the inode item | |
5869 | * was not found. | |
5870 | */ | |
5871 | if (ret > 0) | |
5872 | ret = -ENOENT; | |
5873 | inode = ERR_PTR(ret); | |
1748f843 MF |
5874 | } |
5875 | } | |
5876 | ||
1a54ef8c BR |
5877 | return inode; |
5878 | } | |
5879 | ||
0202e83f | 5880 | struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root) |
4222ea71 | 5881 | { |
0202e83f | 5882 | return btrfs_iget_path(s, ino, root, NULL); |
4222ea71 FM |
5883 | } |
5884 | ||
4df27c4d YZ |
5885 | static struct inode *new_simple_dir(struct super_block *s, |
5886 | struct btrfs_key *key, | |
5887 | struct btrfs_root *root) | |
5888 | { | |
5889 | struct inode *inode = new_inode(s); | |
5890 | ||
5891 | if (!inode) | |
5892 | return ERR_PTR(-ENOMEM); | |
5893 | ||
5c8fd99f | 5894 | BTRFS_I(inode)->root = btrfs_grab_root(root); |
4df27c4d | 5895 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); |
72ac3c0d | 5896 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5897 | |
5898 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
6bb6b514 OS |
5899 | /* |
5900 | * We only need lookup, the rest is read-only and there's no inode | |
5901 | * associated with the dentry | |
5902 | */ | |
5903 | inode->i_op = &simple_dir_inode_operations; | |
1fdf4194 | 5904 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5905 | inode->i_fop = &simple_dir_operations; |
5906 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
c2050a45 | 5907 | inode->i_mtime = current_time(inode); |
9cc97d64 | 5908 | inode->i_atime = inode->i_mtime; |
5909 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 5910 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
4df27c4d YZ |
5911 | |
5912 | return inode; | |
5913 | } | |
5914 | ||
6bf9e4bd QW |
5915 | static inline u8 btrfs_inode_type(struct inode *inode) |
5916 | { | |
5917 | /* | |
5918 | * Compile-time asserts that generic FT_* types still match | |
5919 | * BTRFS_FT_* types | |
5920 | */ | |
5921 | BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN); | |
5922 | BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE); | |
5923 | BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR); | |
5924 | BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV); | |
5925 | BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV); | |
5926 | BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO); | |
5927 | BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK); | |
5928 | BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK); | |
5929 | ||
5930 | return fs_umode_to_ftype(inode->i_mode); | |
5931 | } | |
5932 | ||
3de4586c | 5933 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5934 | { |
0b246afa | 5935 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5936 | struct inode *inode; |
4df27c4d | 5937 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5938 | struct btrfs_root *sub_root = root; |
5939 | struct btrfs_key location; | |
6bf9e4bd | 5940 | u8 di_type = 0; |
b4aff1f8 | 5941 | int ret = 0; |
39279cc3 CM |
5942 | |
5943 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5944 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5945 | |
6bf9e4bd | 5946 | ret = btrfs_inode_by_name(dir, dentry, &location, &di_type); |
39279cc3 CM |
5947 | if (ret < 0) |
5948 | return ERR_PTR(ret); | |
5f39d397 | 5949 | |
4df27c4d | 5950 | if (location.type == BTRFS_INODE_ITEM_KEY) { |
0202e83f | 5951 | inode = btrfs_iget(dir->i_sb, location.objectid, root); |
6bf9e4bd QW |
5952 | if (IS_ERR(inode)) |
5953 | return inode; | |
5954 | ||
5955 | /* Do extra check against inode mode with di_type */ | |
5956 | if (btrfs_inode_type(inode) != di_type) { | |
5957 | btrfs_crit(fs_info, | |
5958 | "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u", | |
5959 | inode->i_mode, btrfs_inode_type(inode), | |
5960 | di_type); | |
5961 | iput(inode); | |
5962 | return ERR_PTR(-EUCLEAN); | |
5963 | } | |
4df27c4d YZ |
5964 | return inode; |
5965 | } | |
5966 | ||
2ff7e61e | 5967 | ret = fixup_tree_root_location(fs_info, dir, dentry, |
4df27c4d YZ |
5968 | &location, &sub_root); |
5969 | if (ret < 0) { | |
5970 | if (ret != -ENOENT) | |
5971 | inode = ERR_PTR(ret); | |
5972 | else | |
5973 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5974 | } else { | |
0202e83f | 5975 | inode = btrfs_iget(dir->i_sb, location.objectid, sub_root); |
39279cc3 | 5976 | } |
8727002f | 5977 | if (root != sub_root) |
00246528 | 5978 | btrfs_put_root(sub_root); |
76dda93c | 5979 | |
34d19bad | 5980 | if (!IS_ERR(inode) && root != sub_root) { |
0b246afa | 5981 | down_read(&fs_info->cleanup_work_sem); |
bc98a42c | 5982 | if (!sb_rdonly(inode->i_sb)) |
66b4ffd1 | 5983 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5984 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5985 | if (ret) { |
5986 | iput(inode); | |
66b4ffd1 | 5987 | inode = ERR_PTR(ret); |
01cd3367 | 5988 | } |
c71bf099 YZ |
5989 | } |
5990 | ||
3de4586c CM |
5991 | return inode; |
5992 | } | |
5993 | ||
fe15ce44 | 5994 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5995 | { |
5996 | struct btrfs_root *root; | |
2b0143b5 | 5997 | struct inode *inode = d_inode(dentry); |
76dda93c | 5998 | |
848cce0d | 5999 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 6000 | inode = d_inode(dentry->d_parent); |
76dda93c | 6001 | |
848cce0d LZ |
6002 | if (inode) { |
6003 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
6004 | if (btrfs_root_refs(&root->root_item) == 0) |
6005 | return 1; | |
848cce0d | 6006 | |
4a0cc7ca | 6007 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 6008 | return 1; |
efefb143 | 6009 | } |
76dda93c YZ |
6010 | return 0; |
6011 | } | |
6012 | ||
3de4586c | 6013 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 6014 | unsigned int flags) |
3de4586c | 6015 | { |
3837d208 | 6016 | struct inode *inode = btrfs_lookup_dentry(dir, dentry); |
5662344b | 6017 | |
3837d208 AV |
6018 | if (inode == ERR_PTR(-ENOENT)) |
6019 | inode = NULL; | |
41d28bca | 6020 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
6021 | } |
6022 | ||
23b5ec74 JB |
6023 | /* |
6024 | * All this infrastructure exists because dir_emit can fault, and we are holding | |
6025 | * the tree lock when doing readdir. For now just allocate a buffer and copy | |
6026 | * our information into that, and then dir_emit from the buffer. This is | |
6027 | * similar to what NFS does, only we don't keep the buffer around in pagecache | |
6028 | * because I'm afraid I'll mess that up. Long term we need to make filldir do | |
6029 | * copy_to_user_inatomic so we don't have to worry about page faulting under the | |
6030 | * tree lock. | |
6031 | */ | |
6032 | static int btrfs_opendir(struct inode *inode, struct file *file) | |
6033 | { | |
6034 | struct btrfs_file_private *private; | |
6035 | ||
6036 | private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL); | |
6037 | if (!private) | |
6038 | return -ENOMEM; | |
6039 | private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
6040 | if (!private->filldir_buf) { | |
6041 | kfree(private); | |
6042 | return -ENOMEM; | |
6043 | } | |
6044 | file->private_data = private; | |
6045 | return 0; | |
6046 | } | |
6047 | ||
6048 | struct dir_entry { | |
6049 | u64 ino; | |
6050 | u64 offset; | |
6051 | unsigned type; | |
6052 | int name_len; | |
6053 | }; | |
6054 | ||
6055 | static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx) | |
6056 | { | |
6057 | while (entries--) { | |
6058 | struct dir_entry *entry = addr; | |
6059 | char *name = (char *)(entry + 1); | |
6060 | ||
92d32170 DS |
6061 | ctx->pos = get_unaligned(&entry->offset); |
6062 | if (!dir_emit(ctx, name, get_unaligned(&entry->name_len), | |
6063 | get_unaligned(&entry->ino), | |
6064 | get_unaligned(&entry->type))) | |
23b5ec74 | 6065 | return 1; |
92d32170 DS |
6066 | addr += sizeof(struct dir_entry) + |
6067 | get_unaligned(&entry->name_len); | |
23b5ec74 JB |
6068 | ctx->pos++; |
6069 | } | |
6070 | return 0; | |
6071 | } | |
6072 | ||
9cdda8d3 | 6073 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 6074 | { |
9cdda8d3 | 6075 | struct inode *inode = file_inode(file); |
39279cc3 | 6076 | struct btrfs_root *root = BTRFS_I(inode)->root; |
23b5ec74 | 6077 | struct btrfs_file_private *private = file->private_data; |
39279cc3 CM |
6078 | struct btrfs_dir_item *di; |
6079 | struct btrfs_key key; | |
5f39d397 | 6080 | struct btrfs_key found_key; |
39279cc3 | 6081 | struct btrfs_path *path; |
23b5ec74 | 6082 | void *addr; |
16cdcec7 MX |
6083 | struct list_head ins_list; |
6084 | struct list_head del_list; | |
39279cc3 | 6085 | int ret; |
5f39d397 | 6086 | struct extent_buffer *leaf; |
39279cc3 | 6087 | int slot; |
5f39d397 CM |
6088 | char *name_ptr; |
6089 | int name_len; | |
23b5ec74 JB |
6090 | int entries = 0; |
6091 | int total_len = 0; | |
02dbfc99 | 6092 | bool put = false; |
c2951f32 | 6093 | struct btrfs_key location; |
5f39d397 | 6094 | |
9cdda8d3 AV |
6095 | if (!dir_emit_dots(file, ctx)) |
6096 | return 0; | |
6097 | ||
49593bfa | 6098 | path = btrfs_alloc_path(); |
16cdcec7 MX |
6099 | if (!path) |
6100 | return -ENOMEM; | |
ff5714cc | 6101 | |
23b5ec74 | 6102 | addr = private->filldir_buf; |
e4058b54 | 6103 | path->reada = READA_FORWARD; |
49593bfa | 6104 | |
c2951f32 JM |
6105 | INIT_LIST_HEAD(&ins_list); |
6106 | INIT_LIST_HEAD(&del_list); | |
6107 | put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list); | |
16cdcec7 | 6108 | |
23b5ec74 | 6109 | again: |
c2951f32 | 6110 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 6111 | key.offset = ctx->pos; |
4a0cc7ca | 6112 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 6113 | |
39279cc3 CM |
6114 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6115 | if (ret < 0) | |
6116 | goto err; | |
49593bfa DW |
6117 | |
6118 | while (1) { | |
23b5ec74 JB |
6119 | struct dir_entry *entry; |
6120 | ||
5f39d397 | 6121 | leaf = path->nodes[0]; |
39279cc3 | 6122 | slot = path->slots[0]; |
b9e03af0 LZ |
6123 | if (slot >= btrfs_header_nritems(leaf)) { |
6124 | ret = btrfs_next_leaf(root, path); | |
6125 | if (ret < 0) | |
6126 | goto err; | |
6127 | else if (ret > 0) | |
6128 | break; | |
6129 | continue; | |
39279cc3 | 6130 | } |
3de4586c | 6131 | |
5f39d397 CM |
6132 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
6133 | ||
6134 | if (found_key.objectid != key.objectid) | |
39279cc3 | 6135 | break; |
c2951f32 | 6136 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 6137 | break; |
9cdda8d3 | 6138 | if (found_key.offset < ctx->pos) |
b9e03af0 | 6139 | goto next; |
c2951f32 | 6140 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
16cdcec7 | 6141 | goto next; |
39279cc3 | 6142 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
c2951f32 | 6143 | name_len = btrfs_dir_name_len(leaf, di); |
23b5ec74 JB |
6144 | if ((total_len + sizeof(struct dir_entry) + name_len) >= |
6145 | PAGE_SIZE) { | |
6146 | btrfs_release_path(path); | |
6147 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
6148 | if (ret) | |
6149 | goto nopos; | |
6150 | addr = private->filldir_buf; | |
6151 | entries = 0; | |
6152 | total_len = 0; | |
6153 | goto again; | |
c2951f32 | 6154 | } |
23b5ec74 JB |
6155 | |
6156 | entry = addr; | |
92d32170 | 6157 | put_unaligned(name_len, &entry->name_len); |
23b5ec74 | 6158 | name_ptr = (char *)(entry + 1); |
c2951f32 JM |
6159 | read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1), |
6160 | name_len); | |
7d157c3d | 6161 | put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)), |
92d32170 | 6162 | &entry->type); |
c2951f32 | 6163 | btrfs_dir_item_key_to_cpu(leaf, di, &location); |
92d32170 DS |
6164 | put_unaligned(location.objectid, &entry->ino); |
6165 | put_unaligned(found_key.offset, &entry->offset); | |
23b5ec74 JB |
6166 | entries++; |
6167 | addr += sizeof(struct dir_entry) + name_len; | |
6168 | total_len += sizeof(struct dir_entry) + name_len; | |
b9e03af0 LZ |
6169 | next: |
6170 | path->slots[0]++; | |
39279cc3 | 6171 | } |
23b5ec74 JB |
6172 | btrfs_release_path(path); |
6173 | ||
6174 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
6175 | if (ret) | |
6176 | goto nopos; | |
49593bfa | 6177 | |
d2fbb2b5 | 6178 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 6179 | if (ret) |
bc4ef759 DS |
6180 | goto nopos; |
6181 | ||
db62efbb ZB |
6182 | /* |
6183 | * Stop new entries from being returned after we return the last | |
6184 | * entry. | |
6185 | * | |
6186 | * New directory entries are assigned a strictly increasing | |
6187 | * offset. This means that new entries created during readdir | |
6188 | * are *guaranteed* to be seen in the future by that readdir. | |
6189 | * This has broken buggy programs which operate on names as | |
6190 | * they're returned by readdir. Until we re-use freed offsets | |
6191 | * we have this hack to stop new entries from being returned | |
6192 | * under the assumption that they'll never reach this huge | |
6193 | * offset. | |
6194 | * | |
6195 | * This is being careful not to overflow 32bit loff_t unless the | |
6196 | * last entry requires it because doing so has broken 32bit apps | |
6197 | * in the past. | |
6198 | */ | |
c2951f32 JM |
6199 | if (ctx->pos >= INT_MAX) |
6200 | ctx->pos = LLONG_MAX; | |
6201 | else | |
6202 | ctx->pos = INT_MAX; | |
39279cc3 CM |
6203 | nopos: |
6204 | ret = 0; | |
6205 | err: | |
02dbfc99 OS |
6206 | if (put) |
6207 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 6208 | btrfs_free_path(path); |
39279cc3 CM |
6209 | return ret; |
6210 | } | |
6211 | ||
39279cc3 | 6212 | /* |
54aa1f4d | 6213 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
6214 | * inode changes. But, it is most likely to find the inode in cache. |
6215 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
6216 | * to keep or drop this code. | |
6217 | */ | |
48a3b636 | 6218 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 | 6219 | { |
2ff7e61e | 6220 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
6221 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6222 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
6223 | int ret; |
6224 | ||
72ac3c0d | 6225 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 6226 | return 0; |
39279cc3 | 6227 | |
7a7eaa40 | 6228 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
6229 | if (IS_ERR(trans)) |
6230 | return PTR_ERR(trans); | |
8929ecfa | 6231 | |
9a56fcd1 | 6232 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
4d14c5cd | 6233 | if (ret && (ret == -ENOSPC || ret == -EDQUOT)) { |
94b60442 | 6234 | /* whoops, lets try again with the full transaction */ |
3a45bb20 | 6235 | btrfs_end_transaction(trans); |
94b60442 | 6236 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
6237 | if (IS_ERR(trans)) |
6238 | return PTR_ERR(trans); | |
8929ecfa | 6239 | |
9a56fcd1 | 6240 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
94b60442 | 6241 | } |
3a45bb20 | 6242 | btrfs_end_transaction(trans); |
16cdcec7 | 6243 | if (BTRFS_I(inode)->delayed_node) |
2ff7e61e | 6244 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
6245 | |
6246 | return ret; | |
6247 | } | |
6248 | ||
6249 | /* | |
6250 | * This is a copy of file_update_time. We need this so we can return error on | |
6251 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
6252 | */ | |
95582b00 | 6253 | static int btrfs_update_time(struct inode *inode, struct timespec64 *now, |
e41f941a | 6254 | int flags) |
22c44fe6 | 6255 | { |
2bc55652 | 6256 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3a8c7231 | 6257 | bool dirty = flags & ~S_VERSION; |
2bc55652 AB |
6258 | |
6259 | if (btrfs_root_readonly(root)) | |
6260 | return -EROFS; | |
6261 | ||
e41f941a | 6262 | if (flags & S_VERSION) |
3a8c7231 | 6263 | dirty |= inode_maybe_inc_iversion(inode, dirty); |
e41f941a JB |
6264 | if (flags & S_CTIME) |
6265 | inode->i_ctime = *now; | |
6266 | if (flags & S_MTIME) | |
6267 | inode->i_mtime = *now; | |
6268 | if (flags & S_ATIME) | |
6269 | inode->i_atime = *now; | |
3a8c7231 | 6270 | return dirty ? btrfs_dirty_inode(inode) : 0; |
39279cc3 CM |
6271 | } |
6272 | ||
d352ac68 CM |
6273 | /* |
6274 | * find the highest existing sequence number in a directory | |
6275 | * and then set the in-memory index_cnt variable to reflect | |
6276 | * free sequence numbers | |
6277 | */ | |
4c570655 | 6278 | static int btrfs_set_inode_index_count(struct btrfs_inode *inode) |
aec7477b | 6279 | { |
4c570655 | 6280 | struct btrfs_root *root = inode->root; |
aec7477b JB |
6281 | struct btrfs_key key, found_key; |
6282 | struct btrfs_path *path; | |
6283 | struct extent_buffer *leaf; | |
6284 | int ret; | |
6285 | ||
4c570655 | 6286 | key.objectid = btrfs_ino(inode); |
962a298f | 6287 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
6288 | key.offset = (u64)-1; |
6289 | ||
6290 | path = btrfs_alloc_path(); | |
6291 | if (!path) | |
6292 | return -ENOMEM; | |
6293 | ||
6294 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6295 | if (ret < 0) | |
6296 | goto out; | |
6297 | /* FIXME: we should be able to handle this */ | |
6298 | if (ret == 0) | |
6299 | goto out; | |
6300 | ret = 0; | |
6301 | ||
6302 | /* | |
6303 | * MAGIC NUMBER EXPLANATION: | |
6304 | * since we search a directory based on f_pos we have to start at 2 | |
6305 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
6306 | * else has to start at 2 | |
6307 | */ | |
6308 | if (path->slots[0] == 0) { | |
4c570655 | 6309 | inode->index_cnt = 2; |
aec7477b JB |
6310 | goto out; |
6311 | } | |
6312 | ||
6313 | path->slots[0]--; | |
6314 | ||
6315 | leaf = path->nodes[0]; | |
6316 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6317 | ||
4c570655 | 6318 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 6319 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
4c570655 | 6320 | inode->index_cnt = 2; |
aec7477b JB |
6321 | goto out; |
6322 | } | |
6323 | ||
4c570655 | 6324 | inode->index_cnt = found_key.offset + 1; |
aec7477b JB |
6325 | out: |
6326 | btrfs_free_path(path); | |
6327 | return ret; | |
6328 | } | |
6329 | ||
d352ac68 CM |
6330 | /* |
6331 | * helper to find a free sequence number in a given directory. This current | |
6332 | * code is very simple, later versions will do smarter things in the btree | |
6333 | */ | |
877574e2 | 6334 | int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index) |
aec7477b JB |
6335 | { |
6336 | int ret = 0; | |
6337 | ||
877574e2 NB |
6338 | if (dir->index_cnt == (u64)-1) { |
6339 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
16cdcec7 MX |
6340 | if (ret) { |
6341 | ret = btrfs_set_inode_index_count(dir); | |
6342 | if (ret) | |
6343 | return ret; | |
6344 | } | |
aec7477b JB |
6345 | } |
6346 | ||
877574e2 NB |
6347 | *index = dir->index_cnt; |
6348 | dir->index_cnt++; | |
aec7477b JB |
6349 | |
6350 | return ret; | |
6351 | } | |
6352 | ||
b0d5d10f CM |
6353 | static int btrfs_insert_inode_locked(struct inode *inode) |
6354 | { | |
6355 | struct btrfs_iget_args args; | |
0202e83f DS |
6356 | |
6357 | args.ino = BTRFS_I(inode)->location.objectid; | |
b0d5d10f CM |
6358 | args.root = BTRFS_I(inode)->root; |
6359 | ||
6360 | return insert_inode_locked4(inode, | |
6361 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6362 | btrfs_find_actor, &args); | |
6363 | } | |
6364 | ||
19aee8de AJ |
6365 | /* |
6366 | * Inherit flags from the parent inode. | |
6367 | * | |
6368 | * Currently only the compression flags and the cow flags are inherited. | |
6369 | */ | |
6370 | static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) | |
6371 | { | |
6372 | unsigned int flags; | |
6373 | ||
6374 | if (!dir) | |
6375 | return; | |
6376 | ||
6377 | flags = BTRFS_I(dir)->flags; | |
6378 | ||
6379 | if (flags & BTRFS_INODE_NOCOMPRESS) { | |
6380 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS; | |
6381 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; | |
6382 | } else if (flags & BTRFS_INODE_COMPRESS) { | |
6383 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS; | |
6384 | BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS; | |
6385 | } | |
6386 | ||
6387 | if (flags & BTRFS_INODE_NODATACOW) { | |
6388 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
6389 | if (S_ISREG(inode->i_mode)) | |
6390 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
6391 | } | |
6392 | ||
7b6a221e | 6393 | btrfs_sync_inode_flags_to_i_flags(inode); |
19aee8de AJ |
6394 | } |
6395 | ||
39279cc3 CM |
6396 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6397 | struct btrfs_root *root, | |
aec7477b | 6398 | struct inode *dir, |
9c58309d | 6399 | const char *name, int name_len, |
175a4eb7 AV |
6400 | u64 ref_objectid, u64 objectid, |
6401 | umode_t mode, u64 *index) | |
39279cc3 | 6402 | { |
0b246afa | 6403 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 6404 | struct inode *inode; |
5f39d397 | 6405 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6406 | struct btrfs_key *location; |
5f39d397 | 6407 | struct btrfs_path *path; |
9c58309d CM |
6408 | struct btrfs_inode_ref *ref; |
6409 | struct btrfs_key key[2]; | |
6410 | u32 sizes[2]; | |
ef3b9af5 | 6411 | int nitems = name ? 2 : 1; |
9c58309d | 6412 | unsigned long ptr; |
11a19a90 | 6413 | unsigned int nofs_flag; |
39279cc3 | 6414 | int ret; |
39279cc3 | 6415 | |
5f39d397 | 6416 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6417 | if (!path) |
6418 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6419 | |
11a19a90 | 6420 | nofs_flag = memalloc_nofs_save(); |
0b246afa | 6421 | inode = new_inode(fs_info->sb); |
11a19a90 | 6422 | memalloc_nofs_restore(nofs_flag); |
8fb27640 YS |
6423 | if (!inode) { |
6424 | btrfs_free_path(path); | |
39279cc3 | 6425 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6426 | } |
39279cc3 | 6427 | |
5762b5c9 FM |
6428 | /* |
6429 | * O_TMPFILE, set link count to 0, so that after this point, | |
6430 | * we fill in an inode item with the correct link count. | |
6431 | */ | |
6432 | if (!name) | |
6433 | set_nlink(inode, 0); | |
6434 | ||
581bb050 LZ |
6435 | /* |
6436 | * we have to initialize this early, so we can reclaim the inode | |
6437 | * number if we fail afterwards in this function. | |
6438 | */ | |
6439 | inode->i_ino = objectid; | |
6440 | ||
ef3b9af5 | 6441 | if (dir && name) { |
1abe9b8a | 6442 | trace_btrfs_inode_request(dir); |
6443 | ||
877574e2 | 6444 | ret = btrfs_set_inode_index(BTRFS_I(dir), index); |
09771430 | 6445 | if (ret) { |
8fb27640 | 6446 | btrfs_free_path(path); |
09771430 | 6447 | iput(inode); |
aec7477b | 6448 | return ERR_PTR(ret); |
09771430 | 6449 | } |
ef3b9af5 FM |
6450 | } else if (dir) { |
6451 | *index = 0; | |
aec7477b JB |
6452 | } |
6453 | /* | |
6454 | * index_cnt is ignored for everything but a dir, | |
df6703e1 | 6455 | * btrfs_set_inode_index_count has an explanation for the magic |
aec7477b JB |
6456 | * number |
6457 | */ | |
6458 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6459 | BTRFS_I(inode)->dir_index = *index; |
5c8fd99f | 6460 | BTRFS_I(inode)->root = btrfs_grab_root(root); |
e02119d5 | 6461 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6462 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6463 | |
5dc562c5 JB |
6464 | /* |
6465 | * We could have gotten an inode number from somebody who was fsynced | |
6466 | * and then removed in this same transaction, so let's just set full | |
6467 | * sync since it will be a full sync anyway and this will blow away the | |
6468 | * old info in the log. | |
6469 | */ | |
6470 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
6471 | ||
9c58309d | 6472 | key[0].objectid = objectid; |
962a298f | 6473 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6474 | key[0].offset = 0; |
6475 | ||
9c58309d | 6476 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6477 | |
6478 | if (name) { | |
6479 | /* | |
6480 | * Start new inodes with an inode_ref. This is slightly more | |
6481 | * efficient for small numbers of hard links since they will | |
6482 | * be packed into one item. Extended refs will kick in if we | |
6483 | * add more hard links than can fit in the ref item. | |
6484 | */ | |
6485 | key[1].objectid = objectid; | |
962a298f | 6486 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
6487 | key[1].offset = ref_objectid; |
6488 | ||
6489 | sizes[1] = name_len + sizeof(*ref); | |
6490 | } | |
9c58309d | 6491 | |
b0d5d10f CM |
6492 | location = &BTRFS_I(inode)->location; |
6493 | location->objectid = objectid; | |
6494 | location->offset = 0; | |
962a298f | 6495 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6496 | |
6497 | ret = btrfs_insert_inode_locked(inode); | |
32955c54 AV |
6498 | if (ret < 0) { |
6499 | iput(inode); | |
b0d5d10f | 6500 | goto fail; |
32955c54 | 6501 | } |
b0d5d10f | 6502 | |
ef3b9af5 | 6503 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6504 | if (ret != 0) |
b0d5d10f | 6505 | goto fail_unlock; |
5f39d397 | 6506 | |
21cb47be | 6507 | inode_init_owner(&init_user_ns, inode, dir, mode); |
a76a3cd4 | 6508 | inode_set_bytes(inode, 0); |
9cc97d64 | 6509 | |
c2050a45 | 6510 | inode->i_mtime = current_time(inode); |
9cc97d64 | 6511 | inode->i_atime = inode->i_mtime; |
6512 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 6513 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
9cc97d64 | 6514 | |
5f39d397 CM |
6515 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6516 | struct btrfs_inode_item); | |
b159fa28 | 6517 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6518 | sizeof(*inode_item)); |
e02119d5 | 6519 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6520 | |
ef3b9af5 FM |
6521 | if (name) { |
6522 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6523 | struct btrfs_inode_ref); | |
6524 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6525 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6526 | ptr = (unsigned long)(ref + 1); | |
6527 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6528 | } | |
9c58309d | 6529 | |
5f39d397 CM |
6530 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6531 | btrfs_free_path(path); | |
6532 | ||
6cbff00f CH |
6533 | btrfs_inherit_iflags(inode, dir); |
6534 | ||
569254b0 | 6535 | if (S_ISREG(mode)) { |
0b246afa | 6536 | if (btrfs_test_opt(fs_info, NODATASUM)) |
94272164 | 6537 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; |
0b246afa | 6538 | if (btrfs_test_opt(fs_info, NODATACOW)) |
f2bdf9a8 JB |
6539 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6540 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6541 | } |
6542 | ||
5d4f98a2 | 6543 | inode_tree_add(inode); |
1abe9b8a | 6544 | |
6545 | trace_btrfs_inode_new(inode); | |
d9094414 | 6546 | btrfs_set_inode_last_trans(trans, BTRFS_I(inode)); |
1abe9b8a | 6547 | |
8ea05e3a AB |
6548 | btrfs_update_root_times(trans, root); |
6549 | ||
63541927 FDBM |
6550 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6551 | if (ret) | |
0b246afa | 6552 | btrfs_err(fs_info, |
63541927 | 6553 | "error inheriting props for ino %llu (root %llu): %d", |
f85b7379 | 6554 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret); |
63541927 | 6555 | |
39279cc3 | 6556 | return inode; |
b0d5d10f CM |
6557 | |
6558 | fail_unlock: | |
32955c54 | 6559 | discard_new_inode(inode); |
5f39d397 | 6560 | fail: |
ef3b9af5 | 6561 | if (dir && name) |
aec7477b | 6562 | BTRFS_I(dir)->index_cnt--; |
5f39d397 CM |
6563 | btrfs_free_path(path); |
6564 | return ERR_PTR(ret); | |
39279cc3 CM |
6565 | } |
6566 | ||
d352ac68 CM |
6567 | /* |
6568 | * utility function to add 'inode' into 'parent_inode' with | |
6569 | * a give name and a given sequence number. | |
6570 | * if 'add_backref' is true, also insert a backref from the | |
6571 | * inode to the parent directory. | |
6572 | */ | |
e02119d5 | 6573 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
db0a669f | 6574 | struct btrfs_inode *parent_inode, struct btrfs_inode *inode, |
e02119d5 | 6575 | const char *name, int name_len, int add_backref, u64 index) |
39279cc3 | 6576 | { |
4df27c4d | 6577 | int ret = 0; |
39279cc3 | 6578 | struct btrfs_key key; |
db0a669f NB |
6579 | struct btrfs_root *root = parent_inode->root; |
6580 | u64 ino = btrfs_ino(inode); | |
6581 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6582 | |
33345d01 | 6583 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
db0a669f | 6584 | memcpy(&key, &inode->root->root_key, sizeof(key)); |
4df27c4d | 6585 | } else { |
33345d01 | 6586 | key.objectid = ino; |
962a298f | 6587 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6588 | key.offset = 0; |
6589 | } | |
6590 | ||
33345d01 | 6591 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
6025c19f | 6592 | ret = btrfs_add_root_ref(trans, key.objectid, |
0b246afa JM |
6593 | root->root_key.objectid, parent_ino, |
6594 | index, name, name_len); | |
4df27c4d | 6595 | } else if (add_backref) { |
33345d01 LZ |
6596 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6597 | parent_ino, index); | |
4df27c4d | 6598 | } |
39279cc3 | 6599 | |
79787eaa JM |
6600 | /* Nothing to clean up yet */ |
6601 | if (ret) | |
6602 | return ret; | |
4df27c4d | 6603 | |
684572df | 6604 | ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key, |
db0a669f | 6605 | btrfs_inode_type(&inode->vfs_inode), index); |
9c52057c | 6606 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6607 | goto fail_dir_item; |
6608 | else if (ret) { | |
66642832 | 6609 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6610 | return ret; |
39279cc3 | 6611 | } |
79787eaa | 6612 | |
db0a669f | 6613 | btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size + |
79787eaa | 6614 | name_len * 2); |
db0a669f | 6615 | inode_inc_iversion(&parent_inode->vfs_inode); |
5338e43a FM |
6616 | /* |
6617 | * If we are replaying a log tree, we do not want to update the mtime | |
6618 | * and ctime of the parent directory with the current time, since the | |
6619 | * log replay procedure is responsible for setting them to their correct | |
6620 | * values (the ones it had when the fsync was done). | |
6621 | */ | |
6622 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) { | |
6623 | struct timespec64 now = current_time(&parent_inode->vfs_inode); | |
6624 | ||
6625 | parent_inode->vfs_inode.i_mtime = now; | |
6626 | parent_inode->vfs_inode.i_ctime = now; | |
6627 | } | |
9a56fcd1 | 6628 | ret = btrfs_update_inode(trans, root, parent_inode); |
79787eaa | 6629 | if (ret) |
66642832 | 6630 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6631 | return ret; |
fe66a05a CM |
6632 | |
6633 | fail_dir_item: | |
6634 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6635 | u64 local_index; | |
6636 | int err; | |
3ee1c553 | 6637 | err = btrfs_del_root_ref(trans, key.objectid, |
0b246afa JM |
6638 | root->root_key.objectid, parent_ino, |
6639 | &local_index, name, name_len); | |
1690dd41 JT |
6640 | if (err) |
6641 | btrfs_abort_transaction(trans, err); | |
fe66a05a CM |
6642 | } else if (add_backref) { |
6643 | u64 local_index; | |
6644 | int err; | |
6645 | ||
6646 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6647 | ino, parent_ino, &local_index); | |
1690dd41 JT |
6648 | if (err) |
6649 | btrfs_abort_transaction(trans, err); | |
fe66a05a | 6650 | } |
1690dd41 JT |
6651 | |
6652 | /* Return the original error code */ | |
fe66a05a | 6653 | return ret; |
39279cc3 CM |
6654 | } |
6655 | ||
6656 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
cef415af NB |
6657 | struct btrfs_inode *dir, struct dentry *dentry, |
6658 | struct btrfs_inode *inode, int backref, u64 index) | |
39279cc3 | 6659 | { |
a1b075d2 JB |
6660 | int err = btrfs_add_link(trans, dir, inode, |
6661 | dentry->d_name.name, dentry->d_name.len, | |
6662 | backref, index); | |
39279cc3 CM |
6663 | if (err > 0) |
6664 | err = -EEXIST; | |
6665 | return err; | |
6666 | } | |
6667 | ||
549c7297 CB |
6668 | static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, |
6669 | struct dentry *dentry, umode_t mode, dev_t rdev) | |
618e21d5 | 6670 | { |
2ff7e61e | 6671 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 JB |
6672 | struct btrfs_trans_handle *trans; |
6673 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6674 | struct inode *inode = NULL; |
618e21d5 | 6675 | int err; |
618e21d5 | 6676 | u64 objectid; |
00e4e6b3 | 6677 | u64 index = 0; |
618e21d5 | 6678 | |
9ed74f2d JB |
6679 | /* |
6680 | * 2 for inode item and ref | |
6681 | * 2 for dir items | |
6682 | * 1 for xattr if selinux is on | |
6683 | */ | |
a22285a6 YZ |
6684 | trans = btrfs_start_transaction(root, 5); |
6685 | if (IS_ERR(trans)) | |
6686 | return PTR_ERR(trans); | |
1832a6d5 | 6687 | |
543068a2 | 6688 | err = btrfs_get_free_objectid(root, &objectid); |
581bb050 LZ |
6689 | if (err) |
6690 | goto out_unlock; | |
6691 | ||
aec7477b | 6692 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6693 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6694 | mode, &index); | |
7cf96da3 TI |
6695 | if (IS_ERR(inode)) { |
6696 | err = PTR_ERR(inode); | |
32955c54 | 6697 | inode = NULL; |
618e21d5 | 6698 | goto out_unlock; |
7cf96da3 | 6699 | } |
618e21d5 | 6700 | |
ad19db71 CS |
6701 | /* |
6702 | * If the active LSM wants to access the inode during | |
6703 | * d_instantiate it needs these. Smack checks to see | |
6704 | * if the filesystem supports xattrs by looking at the | |
6705 | * ops vector. | |
6706 | */ | |
ad19db71 | 6707 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6708 | init_special_inode(inode, inode->i_mode, rdev); |
6709 | ||
6710 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6711 | if (err) |
32955c54 | 6712 | goto out_unlock; |
b0d5d10f | 6713 | |
cef415af NB |
6714 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6715 | 0, index); | |
32955c54 AV |
6716 | if (err) |
6717 | goto out_unlock; | |
6718 | ||
9a56fcd1 | 6719 | btrfs_update_inode(trans, root, BTRFS_I(inode)); |
32955c54 | 6720 | d_instantiate_new(dentry, inode); |
b0d5d10f | 6721 | |
618e21d5 | 6722 | out_unlock: |
3a45bb20 | 6723 | btrfs_end_transaction(trans); |
2ff7e61e | 6724 | btrfs_btree_balance_dirty(fs_info); |
32955c54 | 6725 | if (err && inode) { |
618e21d5 | 6726 | inode_dec_link_count(inode); |
32955c54 | 6727 | discard_new_inode(inode); |
618e21d5 | 6728 | } |
618e21d5 JB |
6729 | return err; |
6730 | } | |
6731 | ||
549c7297 CB |
6732 | static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir, |
6733 | struct dentry *dentry, umode_t mode, bool excl) | |
39279cc3 | 6734 | { |
2ff7e61e | 6735 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
6736 | struct btrfs_trans_handle *trans; |
6737 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6738 | struct inode *inode = NULL; |
a22285a6 | 6739 | int err; |
39279cc3 | 6740 | u64 objectid; |
00e4e6b3 | 6741 | u64 index = 0; |
39279cc3 | 6742 | |
9ed74f2d JB |
6743 | /* |
6744 | * 2 for inode item and ref | |
6745 | * 2 for dir items | |
6746 | * 1 for xattr if selinux is on | |
6747 | */ | |
a22285a6 YZ |
6748 | trans = btrfs_start_transaction(root, 5); |
6749 | if (IS_ERR(trans)) | |
6750 | return PTR_ERR(trans); | |
9ed74f2d | 6751 | |
543068a2 | 6752 | err = btrfs_get_free_objectid(root, &objectid); |
581bb050 LZ |
6753 | if (err) |
6754 | goto out_unlock; | |
6755 | ||
aec7477b | 6756 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6757 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6758 | mode, &index); | |
7cf96da3 TI |
6759 | if (IS_ERR(inode)) { |
6760 | err = PTR_ERR(inode); | |
32955c54 | 6761 | inode = NULL; |
39279cc3 | 6762 | goto out_unlock; |
7cf96da3 | 6763 | } |
ad19db71 CS |
6764 | /* |
6765 | * If the active LSM wants to access the inode during | |
6766 | * d_instantiate it needs these. Smack checks to see | |
6767 | * if the filesystem supports xattrs by looking at the | |
6768 | * ops vector. | |
6769 | */ | |
6770 | inode->i_fop = &btrfs_file_operations; | |
6771 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6772 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6773 | |
6774 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6775 | if (err) | |
32955c54 | 6776 | goto out_unlock; |
b0d5d10f | 6777 | |
9a56fcd1 | 6778 | err = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
b0d5d10f | 6779 | if (err) |
32955c54 | 6780 | goto out_unlock; |
ad19db71 | 6781 | |
cef415af NB |
6782 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6783 | 0, index); | |
39279cc3 | 6784 | if (err) |
32955c54 | 6785 | goto out_unlock; |
43baa579 | 6786 | |
1e2e547a | 6787 | d_instantiate_new(dentry, inode); |
43baa579 | 6788 | |
39279cc3 | 6789 | out_unlock: |
3a45bb20 | 6790 | btrfs_end_transaction(trans); |
32955c54 | 6791 | if (err && inode) { |
39279cc3 | 6792 | inode_dec_link_count(inode); |
32955c54 | 6793 | discard_new_inode(inode); |
39279cc3 | 6794 | } |
2ff7e61e | 6795 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6796 | return err; |
6797 | } | |
6798 | ||
6799 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6800 | struct dentry *dentry) | |
6801 | { | |
271dba45 | 6802 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6803 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6804 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6805 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00e4e6b3 | 6806 | u64 index; |
39279cc3 CM |
6807 | int err; |
6808 | int drop_inode = 0; | |
6809 | ||
4a8be425 | 6810 | /* do not allow sys_link's with other subvols of the same device */ |
4fd786e6 | 6811 | if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid) |
3ab3564f | 6812 | return -EXDEV; |
4a8be425 | 6813 | |
f186373f | 6814 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6815 | return -EMLINK; |
4a8be425 | 6816 | |
877574e2 | 6817 | err = btrfs_set_inode_index(BTRFS_I(dir), &index); |
aec7477b JB |
6818 | if (err) |
6819 | goto fail; | |
6820 | ||
a22285a6 | 6821 | /* |
7e6b6465 | 6822 | * 2 items for inode and inode ref |
a22285a6 | 6823 | * 2 items for dir items |
7e6b6465 | 6824 | * 1 item for parent inode |
399b0bbf | 6825 | * 1 item for orphan item deletion if O_TMPFILE |
a22285a6 | 6826 | */ |
399b0bbf | 6827 | trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6); |
a22285a6 YZ |
6828 | if (IS_ERR(trans)) { |
6829 | err = PTR_ERR(trans); | |
271dba45 | 6830 | trans = NULL; |
a22285a6 YZ |
6831 | goto fail; |
6832 | } | |
5f39d397 | 6833 | |
67de1176 MX |
6834 | /* There are several dir indexes for this inode, clear the cache. */ |
6835 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6836 | inc_nlink(inode); |
0c4d2d95 | 6837 | inode_inc_iversion(inode); |
c2050a45 | 6838 | inode->i_ctime = current_time(inode); |
7de9c6ee | 6839 | ihold(inode); |
e9976151 | 6840 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6841 | |
cef415af NB |
6842 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6843 | 1, index); | |
5f39d397 | 6844 | |
a5719521 | 6845 | if (err) { |
54aa1f4d | 6846 | drop_inode = 1; |
a5719521 | 6847 | } else { |
10d9f309 | 6848 | struct dentry *parent = dentry->d_parent; |
d4682ba0 | 6849 | |
9a56fcd1 | 6850 | err = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
79787eaa JM |
6851 | if (err) |
6852 | goto fail; | |
ef3b9af5 FM |
6853 | if (inode->i_nlink == 1) { |
6854 | /* | |
6855 | * If new hard link count is 1, it's a file created | |
6856 | * with open(2) O_TMPFILE flag. | |
6857 | */ | |
3d6ae7bb | 6858 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
ef3b9af5 FM |
6859 | if (err) |
6860 | goto fail; | |
6861 | } | |
08c422c2 | 6862 | d_instantiate(dentry, inode); |
75b463d2 | 6863 | btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent); |
a5719521 | 6864 | } |
39279cc3 | 6865 | |
1832a6d5 | 6866 | fail: |
271dba45 | 6867 | if (trans) |
3a45bb20 | 6868 | btrfs_end_transaction(trans); |
39279cc3 CM |
6869 | if (drop_inode) { |
6870 | inode_dec_link_count(inode); | |
6871 | iput(inode); | |
6872 | } | |
2ff7e61e | 6873 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6874 | return err; |
6875 | } | |
6876 | ||
549c7297 CB |
6877 | static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, |
6878 | struct dentry *dentry, umode_t mode) | |
39279cc3 | 6879 | { |
2ff7e61e | 6880 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
b9d86667 | 6881 | struct inode *inode = NULL; |
39279cc3 CM |
6882 | struct btrfs_trans_handle *trans; |
6883 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6884 | int err = 0; | |
b9d86667 | 6885 | u64 objectid = 0; |
00e4e6b3 | 6886 | u64 index = 0; |
39279cc3 | 6887 | |
9ed74f2d JB |
6888 | /* |
6889 | * 2 items for inode and ref | |
6890 | * 2 items for dir items | |
6891 | * 1 for xattr if selinux is on | |
6892 | */ | |
a22285a6 YZ |
6893 | trans = btrfs_start_transaction(root, 5); |
6894 | if (IS_ERR(trans)) | |
6895 | return PTR_ERR(trans); | |
39279cc3 | 6896 | |
543068a2 | 6897 | err = btrfs_get_free_objectid(root, &objectid); |
581bb050 LZ |
6898 | if (err) |
6899 | goto out_fail; | |
6900 | ||
aec7477b | 6901 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6902 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6903 | S_IFDIR | mode, &index); | |
39279cc3 CM |
6904 | if (IS_ERR(inode)) { |
6905 | err = PTR_ERR(inode); | |
32955c54 | 6906 | inode = NULL; |
39279cc3 CM |
6907 | goto out_fail; |
6908 | } | |
5f39d397 | 6909 | |
b0d5d10f CM |
6910 | /* these must be set before we unlock the inode */ |
6911 | inode->i_op = &btrfs_dir_inode_operations; | |
6912 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6913 | |
2a7dba39 | 6914 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6915 | if (err) |
32955c54 | 6916 | goto out_fail; |
39279cc3 | 6917 | |
6ef06d27 | 6918 | btrfs_i_size_write(BTRFS_I(inode), 0); |
9a56fcd1 | 6919 | err = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
39279cc3 | 6920 | if (err) |
32955c54 | 6921 | goto out_fail; |
5f39d397 | 6922 | |
db0a669f NB |
6923 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6924 | dentry->d_name.name, | |
6925 | dentry->d_name.len, 0, index); | |
39279cc3 | 6926 | if (err) |
32955c54 | 6927 | goto out_fail; |
5f39d397 | 6928 | |
1e2e547a | 6929 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
6930 | |
6931 | out_fail: | |
3a45bb20 | 6932 | btrfs_end_transaction(trans); |
32955c54 | 6933 | if (err && inode) { |
c7cfb8a5 | 6934 | inode_dec_link_count(inode); |
32955c54 | 6935 | discard_new_inode(inode); |
c7cfb8a5 | 6936 | } |
2ff7e61e | 6937 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6938 | return err; |
6939 | } | |
6940 | ||
c8b97818 | 6941 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6942 | struct page *page, |
c8b97818 CM |
6943 | size_t pg_offset, u64 extent_offset, |
6944 | struct btrfs_file_extent_item *item) | |
6945 | { | |
6946 | int ret; | |
6947 | struct extent_buffer *leaf = path->nodes[0]; | |
6948 | char *tmp; | |
6949 | size_t max_size; | |
6950 | unsigned long inline_size; | |
6951 | unsigned long ptr; | |
261507a0 | 6952 | int compress_type; |
c8b97818 CM |
6953 | |
6954 | WARN_ON(pg_offset != 0); | |
261507a0 | 6955 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6956 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6957 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6958 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6959 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6960 | if (!tmp) |
6961 | return -ENOMEM; | |
c8b97818 CM |
6962 | ptr = btrfs_file_extent_inline_start(item); |
6963 | ||
6964 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6965 | ||
09cbfeaf | 6966 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6967 | ret = btrfs_decompress(compress_type, tmp, page, |
6968 | extent_offset, inline_size, max_size); | |
e1699d2d ZB |
6969 | |
6970 | /* | |
6971 | * decompression code contains a memset to fill in any space between the end | |
6972 | * of the uncompressed data and the end of max_size in case the decompressed | |
6973 | * data ends up shorter than ram_bytes. That doesn't cover the hole between | |
6974 | * the end of an inline extent and the beginning of the next block, so we | |
6975 | * cover that region here. | |
6976 | */ | |
6977 | ||
d048b9c2 IW |
6978 | if (max_size + pg_offset < PAGE_SIZE) |
6979 | memzero_page(page, pg_offset + max_size, | |
6980 | PAGE_SIZE - max_size - pg_offset); | |
c8b97818 | 6981 | kfree(tmp); |
166ae5a4 | 6982 | return ret; |
c8b97818 CM |
6983 | } |
6984 | ||
39b07b5d OS |
6985 | /** |
6986 | * btrfs_get_extent - Lookup the first extent overlapping a range in a file. | |
6987 | * @inode: file to search in | |
6988 | * @page: page to read extent data into if the extent is inline | |
6989 | * @pg_offset: offset into @page to copy to | |
6990 | * @start: file offset | |
6991 | * @len: length of range starting at @start | |
6992 | * | |
6993 | * This returns the first &struct extent_map which overlaps with the given | |
6994 | * range, reading it from the B-tree and caching it if necessary. Note that | |
6995 | * there may be more extents which overlap the given range after the returned | |
6996 | * extent_map. | |
d352ac68 | 6997 | * |
39b07b5d OS |
6998 | * If @page is not NULL and the extent is inline, this also reads the extent |
6999 | * data directly into the page and marks the extent up to date in the io_tree. | |
7000 | * | |
7001 | * Return: ERR_PTR on error, non-NULL extent_map on success. | |
d352ac68 | 7002 | */ |
fc4f21b1 | 7003 | struct extent_map *btrfs_get_extent(struct btrfs_inode *inode, |
39b07b5d OS |
7004 | struct page *page, size_t pg_offset, |
7005 | u64 start, u64 len) | |
a52d9a80 | 7006 | { |
3ffbd68c | 7007 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
1028d1c4 | 7008 | int ret = 0; |
a52d9a80 CM |
7009 | u64 extent_start = 0; |
7010 | u64 extent_end = 0; | |
fc4f21b1 | 7011 | u64 objectid = btrfs_ino(inode); |
7e74e235 | 7012 | int extent_type = -1; |
f421950f | 7013 | struct btrfs_path *path = NULL; |
fc4f21b1 | 7014 | struct btrfs_root *root = inode->root; |
a52d9a80 | 7015 | struct btrfs_file_extent_item *item; |
5f39d397 CM |
7016 | struct extent_buffer *leaf; |
7017 | struct btrfs_key found_key; | |
a52d9a80 | 7018 | struct extent_map *em = NULL; |
fc4f21b1 NB |
7019 | struct extent_map_tree *em_tree = &inode->extent_tree; |
7020 | struct extent_io_tree *io_tree = &inode->io_tree; | |
a52d9a80 | 7021 | |
890871be | 7022 | read_lock(&em_tree->lock); |
d1310b2e | 7023 | em = lookup_extent_mapping(em_tree, start, len); |
890871be | 7024 | read_unlock(&em_tree->lock); |
d1310b2e | 7025 | |
a52d9a80 | 7026 | if (em) { |
e1c4b745 CM |
7027 | if (em->start > start || em->start + em->len <= start) |
7028 | free_extent_map(em); | |
7029 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
7030 | free_extent_map(em); |
7031 | else | |
7032 | goto out; | |
a52d9a80 | 7033 | } |
172ddd60 | 7034 | em = alloc_extent_map(); |
a52d9a80 | 7035 | if (!em) { |
1028d1c4 | 7036 | ret = -ENOMEM; |
d1310b2e | 7037 | goto out; |
a52d9a80 | 7038 | } |
d1310b2e | 7039 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 7040 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 7041 | em->len = (u64)-1; |
c8b97818 | 7042 | em->block_len = (u64)-1; |
f421950f | 7043 | |
bee6ec82 | 7044 | path = btrfs_alloc_path(); |
f421950f | 7045 | if (!path) { |
1028d1c4 | 7046 | ret = -ENOMEM; |
bee6ec82 | 7047 | goto out; |
f421950f CM |
7048 | } |
7049 | ||
bee6ec82 LB |
7050 | /* Chances are we'll be called again, so go ahead and do readahead */ |
7051 | path->reada = READA_FORWARD; | |
4d7240f0 JB |
7052 | |
7053 | /* | |
7054 | * The same explanation in load_free_space_cache applies here as well, | |
7055 | * we only read when we're loading the free space cache, and at that | |
7056 | * point the commit_root has everything we need. | |
7057 | */ | |
7058 | if (btrfs_is_free_space_inode(inode)) { | |
7059 | path->search_commit_root = 1; | |
7060 | path->skip_locking = 1; | |
7061 | } | |
51899412 | 7062 | |
5c9a702e | 7063 | ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0); |
a52d9a80 | 7064 | if (ret < 0) { |
a52d9a80 | 7065 | goto out; |
b8eeab7f | 7066 | } else if (ret > 0) { |
a52d9a80 CM |
7067 | if (path->slots[0] == 0) |
7068 | goto not_found; | |
7069 | path->slots[0]--; | |
1028d1c4 | 7070 | ret = 0; |
a52d9a80 CM |
7071 | } |
7072 | ||
5f39d397 CM |
7073 | leaf = path->nodes[0]; |
7074 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 7075 | struct btrfs_file_extent_item); |
5f39d397 | 7076 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5f39d397 | 7077 | if (found_key.objectid != objectid || |
694c12ed | 7078 | found_key.type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
7079 | /* |
7080 | * If we backup past the first extent we want to move forward | |
7081 | * and see if there is an extent in front of us, otherwise we'll | |
7082 | * say there is a hole for our whole search range which can | |
7083 | * cause problems. | |
7084 | */ | |
7085 | extent_end = start; | |
7086 | goto next; | |
a52d9a80 CM |
7087 | } |
7088 | ||
694c12ed | 7089 | extent_type = btrfs_file_extent_type(leaf, item); |
5f39d397 | 7090 | extent_start = found_key.offset; |
a5eeb3d1 | 7091 | extent_end = btrfs_file_extent_end(path); |
694c12ed NB |
7092 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
7093 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
6bf9e4bd QW |
7094 | /* Only regular file could have regular/prealloc extent */ |
7095 | if (!S_ISREG(inode->vfs_inode.i_mode)) { | |
1028d1c4 | 7096 | ret = -EUCLEAN; |
6bf9e4bd QW |
7097 | btrfs_crit(fs_info, |
7098 | "regular/prealloc extent found for non-regular inode %llu", | |
7099 | btrfs_ino(inode)); | |
7100 | goto out; | |
7101 | } | |
09ed2f16 LB |
7102 | trace_btrfs_get_extent_show_fi_regular(inode, leaf, item, |
7103 | extent_start); | |
694c12ed | 7104 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
09ed2f16 LB |
7105 | trace_btrfs_get_extent_show_fi_inline(inode, leaf, item, |
7106 | path->slots[0], | |
7107 | extent_start); | |
9036c102 | 7108 | } |
25a50341 | 7109 | next: |
9036c102 YZ |
7110 | if (start >= extent_end) { |
7111 | path->slots[0]++; | |
7112 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
7113 | ret = btrfs_next_leaf(root, path); | |
1028d1c4 | 7114 | if (ret < 0) |
9036c102 | 7115 | goto out; |
1028d1c4 | 7116 | else if (ret > 0) |
9036c102 | 7117 | goto not_found; |
1028d1c4 | 7118 | |
9036c102 | 7119 | leaf = path->nodes[0]; |
a52d9a80 | 7120 | } |
9036c102 YZ |
7121 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
7122 | if (found_key.objectid != objectid || | |
7123 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
7124 | goto not_found; | |
7125 | if (start + len <= found_key.offset) | |
7126 | goto not_found; | |
e2eca69d WS |
7127 | if (start > found_key.offset) |
7128 | goto next; | |
02a033df NB |
7129 | |
7130 | /* New extent overlaps with existing one */ | |
9036c102 | 7131 | em->start = start; |
70c8a91c | 7132 | em->orig_start = start; |
9036c102 | 7133 | em->len = found_key.offset - start; |
02a033df NB |
7134 | em->block_start = EXTENT_MAP_HOLE; |
7135 | goto insert; | |
9036c102 YZ |
7136 | } |
7137 | ||
39b07b5d | 7138 | btrfs_extent_item_to_extent_map(inode, path, item, !page, em); |
7ffbb598 | 7139 | |
694c12ed NB |
7140 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
7141 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 7142 | goto insert; |
694c12ed | 7143 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 7144 | unsigned long ptr; |
a52d9a80 | 7145 | char *map; |
3326d1b0 CM |
7146 | size_t size; |
7147 | size_t extent_offset; | |
7148 | size_t copy_size; | |
a52d9a80 | 7149 | |
39b07b5d | 7150 | if (!page) |
689f9346 | 7151 | goto out; |
5f39d397 | 7152 | |
e41ca589 | 7153 | size = btrfs_file_extent_ram_bytes(leaf, item); |
9036c102 | 7154 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
7155 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
7156 | size - extent_offset); | |
3326d1b0 | 7157 | em->start = extent_start + extent_offset; |
0b246afa | 7158 | em->len = ALIGN(copy_size, fs_info->sectorsize); |
b4939680 | 7159 | em->orig_block_len = em->len; |
70c8a91c | 7160 | em->orig_start = em->start; |
689f9346 | 7161 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
e49aabd9 | 7162 | |
bf46f52d | 7163 | if (!PageUptodate(page)) { |
261507a0 LZ |
7164 | if (btrfs_file_extent_compression(leaf, item) != |
7165 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 7166 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 7167 | extent_offset, item); |
1028d1c4 | 7168 | if (ret) |
166ae5a4 | 7169 | goto out; |
c8b97818 | 7170 | } else { |
58c1a35c | 7171 | map = kmap_local_page(page); |
c8b97818 CM |
7172 | read_extent_buffer(leaf, map + pg_offset, ptr, |
7173 | copy_size); | |
09cbfeaf | 7174 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 7175 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 7176 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
7177 | copy_size); |
7178 | } | |
58c1a35c | 7179 | kunmap_local(map); |
c8b97818 | 7180 | } |
179e29e4 | 7181 | flush_dcache_page(page); |
a52d9a80 | 7182 | } |
d1310b2e | 7183 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 7184 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 7185 | goto insert; |
a52d9a80 CM |
7186 | } |
7187 | not_found: | |
7188 | em->start = start; | |
70c8a91c | 7189 | em->orig_start = start; |
d1310b2e | 7190 | em->len = len; |
5f39d397 | 7191 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 | 7192 | insert: |
1028d1c4 | 7193 | ret = 0; |
b3b4aa74 | 7194 | btrfs_release_path(path); |
d1310b2e | 7195 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 7196 | btrfs_err(fs_info, |
5d163e0e JM |
7197 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
7198 | em->start, em->len, start, len); | |
1028d1c4 | 7199 | ret = -EIO; |
a52d9a80 CM |
7200 | goto out; |
7201 | } | |
d1310b2e | 7202 | |
890871be | 7203 | write_lock(&em_tree->lock); |
1028d1c4 | 7204 | ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len); |
890871be | 7205 | write_unlock(&em_tree->lock); |
a52d9a80 | 7206 | out: |
c6414280 | 7207 | btrfs_free_path(path); |
1abe9b8a | 7208 | |
fc4f21b1 | 7209 | trace_btrfs_get_extent(root, inode, em); |
1abe9b8a | 7210 | |
1028d1c4 | 7211 | if (ret) { |
a52d9a80 | 7212 | free_extent_map(em); |
1028d1c4 | 7213 | return ERR_PTR(ret); |
a52d9a80 CM |
7214 | } |
7215 | return em; | |
7216 | } | |
7217 | ||
fc4f21b1 | 7218 | struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode, |
4ab47a8d | 7219 | u64 start, u64 len) |
ec29ed5b CM |
7220 | { |
7221 | struct extent_map *em; | |
7222 | struct extent_map *hole_em = NULL; | |
f3714ef4 | 7223 | u64 delalloc_start = start; |
ec29ed5b | 7224 | u64 end; |
f3714ef4 NB |
7225 | u64 delalloc_len; |
7226 | u64 delalloc_end; | |
ec29ed5b CM |
7227 | int err = 0; |
7228 | ||
39b07b5d | 7229 | em = btrfs_get_extent(inode, NULL, 0, start, len); |
ec29ed5b CM |
7230 | if (IS_ERR(em)) |
7231 | return em; | |
9986277e DC |
7232 | /* |
7233 | * If our em maps to: | |
7234 | * - a hole or | |
7235 | * - a pre-alloc extent, | |
7236 | * there might actually be delalloc bytes behind it. | |
7237 | */ | |
7238 | if (em->block_start != EXTENT_MAP_HOLE && | |
7239 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7240 | return em; | |
7241 | else | |
7242 | hole_em = em; | |
ec29ed5b CM |
7243 | |
7244 | /* check to see if we've wrapped (len == -1 or similar) */ | |
7245 | end = start + len; | |
7246 | if (end < start) | |
7247 | end = (u64)-1; | |
7248 | else | |
7249 | end -= 1; | |
7250 | ||
7251 | em = NULL; | |
7252 | ||
7253 | /* ok, we didn't find anything, lets look for delalloc */ | |
f3714ef4 | 7254 | delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start, |
ec29ed5b | 7255 | end, len, EXTENT_DELALLOC, 1); |
f3714ef4 NB |
7256 | delalloc_end = delalloc_start + delalloc_len; |
7257 | if (delalloc_end < delalloc_start) | |
7258 | delalloc_end = (u64)-1; | |
ec29ed5b CM |
7259 | |
7260 | /* | |
f3714ef4 NB |
7261 | * We didn't find anything useful, return the original results from |
7262 | * get_extent() | |
ec29ed5b | 7263 | */ |
f3714ef4 | 7264 | if (delalloc_start > end || delalloc_end <= start) { |
ec29ed5b CM |
7265 | em = hole_em; |
7266 | hole_em = NULL; | |
7267 | goto out; | |
7268 | } | |
7269 | ||
f3714ef4 NB |
7270 | /* |
7271 | * Adjust the delalloc_start to make sure it doesn't go backwards from | |
7272 | * the start they passed in | |
ec29ed5b | 7273 | */ |
f3714ef4 NB |
7274 | delalloc_start = max(start, delalloc_start); |
7275 | delalloc_len = delalloc_end - delalloc_start; | |
ec29ed5b | 7276 | |
f3714ef4 NB |
7277 | if (delalloc_len > 0) { |
7278 | u64 hole_start; | |
02950af4 | 7279 | u64 hole_len; |
f3714ef4 | 7280 | const u64 hole_end = extent_map_end(hole_em); |
ec29ed5b | 7281 | |
172ddd60 | 7282 | em = alloc_extent_map(); |
ec29ed5b CM |
7283 | if (!em) { |
7284 | err = -ENOMEM; | |
7285 | goto out; | |
7286 | } | |
f3714ef4 NB |
7287 | |
7288 | ASSERT(hole_em); | |
ec29ed5b | 7289 | /* |
f3714ef4 NB |
7290 | * When btrfs_get_extent can't find anything it returns one |
7291 | * huge hole | |
ec29ed5b | 7292 | * |
f3714ef4 NB |
7293 | * Make sure what it found really fits our range, and adjust to |
7294 | * make sure it is based on the start from the caller | |
ec29ed5b | 7295 | */ |
f3714ef4 NB |
7296 | if (hole_end <= start || hole_em->start > end) { |
7297 | free_extent_map(hole_em); | |
7298 | hole_em = NULL; | |
7299 | } else { | |
7300 | hole_start = max(hole_em->start, start); | |
7301 | hole_len = hole_end - hole_start; | |
ec29ed5b | 7302 | } |
f3714ef4 NB |
7303 | |
7304 | if (hole_em && delalloc_start > hole_start) { | |
7305 | /* | |
7306 | * Our hole starts before our delalloc, so we have to | |
7307 | * return just the parts of the hole that go until the | |
7308 | * delalloc starts | |
ec29ed5b | 7309 | */ |
f3714ef4 | 7310 | em->len = min(hole_len, delalloc_start - hole_start); |
ec29ed5b CM |
7311 | em->start = hole_start; |
7312 | em->orig_start = hole_start; | |
7313 | /* | |
f3714ef4 NB |
7314 | * Don't adjust block start at all, it is fixed at |
7315 | * EXTENT_MAP_HOLE | |
ec29ed5b CM |
7316 | */ |
7317 | em->block_start = hole_em->block_start; | |
7318 | em->block_len = hole_len; | |
f9e4fb53 LB |
7319 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
7320 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b | 7321 | } else { |
f3714ef4 NB |
7322 | /* |
7323 | * Hole is out of passed range or it starts after | |
7324 | * delalloc range | |
7325 | */ | |
7326 | em->start = delalloc_start; | |
7327 | em->len = delalloc_len; | |
7328 | em->orig_start = delalloc_start; | |
ec29ed5b | 7329 | em->block_start = EXTENT_MAP_DELALLOC; |
f3714ef4 | 7330 | em->block_len = delalloc_len; |
ec29ed5b | 7331 | } |
bf8d32b9 | 7332 | } else { |
ec29ed5b CM |
7333 | return hole_em; |
7334 | } | |
7335 | out: | |
7336 | ||
7337 | free_extent_map(hole_em); | |
7338 | if (err) { | |
7339 | free_extent_map(em); | |
7340 | return ERR_PTR(err); | |
7341 | } | |
7342 | return em; | |
7343 | } | |
7344 | ||
64f54188 | 7345 | static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode, |
5f9a8a51 FM |
7346 | const u64 start, |
7347 | const u64 len, | |
7348 | const u64 orig_start, | |
7349 | const u64 block_start, | |
7350 | const u64 block_len, | |
7351 | const u64 orig_block_len, | |
7352 | const u64 ram_bytes, | |
7353 | const int type) | |
7354 | { | |
7355 | struct extent_map *em = NULL; | |
7356 | int ret; | |
7357 | ||
5f9a8a51 | 7358 | if (type != BTRFS_ORDERED_NOCOW) { |
64f54188 NB |
7359 | em = create_io_em(inode, start, len, orig_start, block_start, |
7360 | block_len, orig_block_len, ram_bytes, | |
6f9994db LB |
7361 | BTRFS_COMPRESS_NONE, /* compress_type */ |
7362 | type); | |
5f9a8a51 FM |
7363 | if (IS_ERR(em)) |
7364 | goto out; | |
7365 | } | |
64f54188 NB |
7366 | ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len, |
7367 | block_len, type); | |
5f9a8a51 FM |
7368 | if (ret) { |
7369 | if (em) { | |
7370 | free_extent_map(em); | |
64f54188 | 7371 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); |
5f9a8a51 FM |
7372 | } |
7373 | em = ERR_PTR(ret); | |
7374 | } | |
7375 | out: | |
5f9a8a51 FM |
7376 | |
7377 | return em; | |
7378 | } | |
7379 | ||
9fc6f911 | 7380 | static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode, |
4b46fce2 JB |
7381 | u64 start, u64 len) |
7382 | { | |
9fc6f911 NB |
7383 | struct btrfs_root *root = inode->root; |
7384 | struct btrfs_fs_info *fs_info = root->fs_info; | |
70c8a91c | 7385 | struct extent_map *em; |
4b46fce2 JB |
7386 | struct btrfs_key ins; |
7387 | u64 alloc_hint; | |
7388 | int ret; | |
4b46fce2 | 7389 | |
9fc6f911 | 7390 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 7391 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 7392 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
7393 | if (ret) |
7394 | return ERR_PTR(ret); | |
4b46fce2 | 7395 | |
9fc6f911 | 7396 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
5f9a8a51 | 7397 | ins.objectid, ins.offset, ins.offset, |
6288d6ea | 7398 | ins.offset, BTRFS_ORDERED_REGULAR); |
0b246afa | 7399 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 7400 | if (IS_ERR(em)) |
9fc6f911 NB |
7401 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, |
7402 | 1); | |
de0ee0ed | 7403 | |
4b46fce2 JB |
7404 | return em; |
7405 | } | |
7406 | ||
f4639636 | 7407 | static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
05947ae1 AJ |
7408 | { |
7409 | struct btrfs_block_group *block_group; | |
f4639636 | 7410 | bool readonly = false; |
05947ae1 AJ |
7411 | |
7412 | block_group = btrfs_lookup_block_group(fs_info, bytenr); | |
7413 | if (!block_group || block_group->ro) | |
f4639636 | 7414 | readonly = true; |
05947ae1 AJ |
7415 | if (block_group) |
7416 | btrfs_put_block_group(block_group); | |
7417 | return readonly; | |
7418 | } | |
7419 | ||
46bfbb5c | 7420 | /* |
e4ecaf90 QW |
7421 | * Check if we can do nocow write into the range [@offset, @offset + @len) |
7422 | * | |
7423 | * @offset: File offset | |
7424 | * @len: The length to write, will be updated to the nocow writeable | |
7425 | * range | |
7426 | * @orig_start: (optional) Return the original file offset of the file extent | |
7427 | * @orig_len: (optional) Return the original on-disk length of the file extent | |
7428 | * @ram_bytes: (optional) Return the ram_bytes of the file extent | |
a84d5d42 BB |
7429 | * @strict: if true, omit optimizations that might force us into unnecessary |
7430 | * cow. e.g., don't trust generation number. | |
e4ecaf90 | 7431 | * |
e4ecaf90 QW |
7432 | * Return: |
7433 | * >0 and update @len if we can do nocow write | |
7434 | * 0 if we can't do nocow write | |
7435 | * <0 if error happened | |
7436 | * | |
7437 | * NOTE: This only checks the file extents, caller is responsible to wait for | |
7438 | * any ordered extents. | |
46bfbb5c | 7439 | */ |
00361589 | 7440 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 | 7441 | u64 *orig_start, u64 *orig_block_len, |
a84d5d42 | 7442 | u64 *ram_bytes, bool strict) |
46bfbb5c | 7443 | { |
2ff7e61e | 7444 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
46bfbb5c CM |
7445 | struct btrfs_path *path; |
7446 | int ret; | |
7447 | struct extent_buffer *leaf; | |
7448 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7449 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7450 | struct btrfs_file_extent_item *fi; |
7451 | struct btrfs_key key; | |
7452 | u64 disk_bytenr; | |
7453 | u64 backref_offset; | |
7454 | u64 extent_end; | |
7455 | u64 num_bytes; | |
7456 | int slot; | |
7457 | int found_type; | |
7ee9e440 | 7458 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7459 | |
46bfbb5c CM |
7460 | path = btrfs_alloc_path(); |
7461 | if (!path) | |
7462 | return -ENOMEM; | |
7463 | ||
f85b7379 DS |
7464 | ret = btrfs_lookup_file_extent(NULL, root, path, |
7465 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
7466 | if (ret < 0) |
7467 | goto out; | |
7468 | ||
7469 | slot = path->slots[0]; | |
7470 | if (ret == 1) { | |
7471 | if (slot == 0) { | |
7472 | /* can't find the item, must cow */ | |
7473 | ret = 0; | |
7474 | goto out; | |
7475 | } | |
7476 | slot--; | |
7477 | } | |
7478 | ret = 0; | |
7479 | leaf = path->nodes[0]; | |
7480 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4a0cc7ca | 7481 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
7482 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7483 | /* not our file or wrong item type, must cow */ | |
7484 | goto out; | |
7485 | } | |
7486 | ||
7487 | if (key.offset > offset) { | |
7488 | /* Wrong offset, must cow */ | |
7489 | goto out; | |
7490 | } | |
7491 | ||
7492 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7493 | found_type = btrfs_file_extent_type(leaf, fi); | |
7494 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7495 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7496 | /* not a regular extent, must cow */ | |
7497 | goto out; | |
7498 | } | |
7ee9e440 JB |
7499 | |
7500 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7501 | goto out; | |
7502 | ||
e77751aa MX |
7503 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7504 | if (extent_end <= offset) | |
7505 | goto out; | |
7506 | ||
46bfbb5c | 7507 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7508 | if (disk_bytenr == 0) |
7509 | goto out; | |
7510 | ||
7511 | if (btrfs_file_extent_compression(leaf, fi) || | |
7512 | btrfs_file_extent_encryption(leaf, fi) || | |
7513 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7514 | goto out; | |
7515 | ||
78d4295b EL |
7516 | /* |
7517 | * Do the same check as in btrfs_cross_ref_exist but without the | |
7518 | * unnecessary search. | |
7519 | */ | |
a84d5d42 BB |
7520 | if (!strict && |
7521 | (btrfs_file_extent_generation(leaf, fi) <= | |
7522 | btrfs_root_last_snapshot(&root->root_item))) | |
78d4295b EL |
7523 | goto out; |
7524 | ||
46bfbb5c CM |
7525 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7526 | ||
7ee9e440 JB |
7527 | if (orig_start) { |
7528 | *orig_start = key.offset - backref_offset; | |
7529 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7530 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7531 | } | |
eb384b55 | 7532 | |
2ff7e61e | 7533 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
46bfbb5c | 7534 | goto out; |
7b2b7085 MX |
7535 | |
7536 | num_bytes = min(offset + *len, extent_end) - offset; | |
7537 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7538 | u64 range_end; | |
7539 | ||
da17066c JM |
7540 | range_end = round_up(offset + num_bytes, |
7541 | root->fs_info->sectorsize) - 1; | |
7b2b7085 MX |
7542 | ret = test_range_bit(io_tree, offset, range_end, |
7543 | EXTENT_DELALLOC, 0, NULL); | |
7544 | if (ret) { | |
7545 | ret = -EAGAIN; | |
7546 | goto out; | |
7547 | } | |
7548 | } | |
7549 | ||
1bda19eb | 7550 | btrfs_release_path(path); |
46bfbb5c CM |
7551 | |
7552 | /* | |
7553 | * look for other files referencing this extent, if we | |
7554 | * find any we must cow | |
7555 | */ | |
00361589 | 7556 | |
e4c3b2dc | 7557 | ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)), |
a84d5d42 BB |
7558 | key.offset - backref_offset, disk_bytenr, |
7559 | strict); | |
00361589 JB |
7560 | if (ret) { |
7561 | ret = 0; | |
7562 | goto out; | |
7563 | } | |
46bfbb5c CM |
7564 | |
7565 | /* | |
7566 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7567 | * in this extent we are about to write. If there | |
7568 | * are any csums in that range we have to cow in order | |
7569 | * to keep the csums correct | |
7570 | */ | |
7571 | disk_bytenr += backref_offset; | |
7572 | disk_bytenr += offset - key.offset; | |
2ff7e61e JM |
7573 | if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes)) |
7574 | goto out; | |
46bfbb5c CM |
7575 | /* |
7576 | * all of the above have passed, it is safe to overwrite this extent | |
7577 | * without cow | |
7578 | */ | |
eb384b55 | 7579 | *len = num_bytes; |
46bfbb5c CM |
7580 | ret = 1; |
7581 | out: | |
7582 | btrfs_free_path(path); | |
7583 | return ret; | |
7584 | } | |
7585 | ||
eb838e73 | 7586 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
f85781fb | 7587 | struct extent_state **cached_state, bool writing) |
eb838e73 JB |
7588 | { |
7589 | struct btrfs_ordered_extent *ordered; | |
7590 | int ret = 0; | |
7591 | ||
7592 | while (1) { | |
7593 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7594 | cached_state); |
eb838e73 JB |
7595 | /* |
7596 | * We're concerned with the entire range that we're going to be | |
01327610 | 7597 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7598 | * extents in this range. |
7599 | */ | |
a776c6fa | 7600 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart, |
eb838e73 JB |
7601 | lockend - lockstart + 1); |
7602 | ||
7603 | /* | |
7604 | * We need to make sure there are no buffered pages in this | |
7605 | * range either, we could have raced between the invalidate in | |
7606 | * generic_file_direct_write and locking the extent. The | |
7607 | * invalidate needs to happen so that reads after a write do not | |
7608 | * get stale data. | |
7609 | */ | |
fc4adbff | 7610 | if (!ordered && |
051c98eb DS |
7611 | (!writing || !filemap_range_has_page(inode->i_mapping, |
7612 | lockstart, lockend))) | |
eb838e73 JB |
7613 | break; |
7614 | ||
7615 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
e43bbe5e | 7616 | cached_state); |
eb838e73 JB |
7617 | |
7618 | if (ordered) { | |
ade77029 FM |
7619 | /* |
7620 | * If we are doing a DIO read and the ordered extent we | |
7621 | * found is for a buffered write, we can not wait for it | |
7622 | * to complete and retry, because if we do so we can | |
7623 | * deadlock with concurrent buffered writes on page | |
7624 | * locks. This happens only if our DIO read covers more | |
7625 | * than one extent map, if at this point has already | |
7626 | * created an ordered extent for a previous extent map | |
7627 | * and locked its range in the inode's io tree, and a | |
7628 | * concurrent write against that previous extent map's | |
7629 | * range and this range started (we unlock the ranges | |
7630 | * in the io tree only when the bios complete and | |
7631 | * buffered writes always lock pages before attempting | |
7632 | * to lock range in the io tree). | |
7633 | */ | |
7634 | if (writing || | |
7635 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
c0a43603 | 7636 | btrfs_start_ordered_extent(ordered, 1); |
ade77029 FM |
7637 | else |
7638 | ret = -ENOTBLK; | |
eb838e73 JB |
7639 | btrfs_put_ordered_extent(ordered); |
7640 | } else { | |
eb838e73 | 7641 | /* |
b850ae14 FM |
7642 | * We could trigger writeback for this range (and wait |
7643 | * for it to complete) and then invalidate the pages for | |
7644 | * this range (through invalidate_inode_pages2_range()), | |
7645 | * but that can lead us to a deadlock with a concurrent | |
ba206a02 | 7646 | * call to readahead (a buffered read or a defrag call |
b850ae14 FM |
7647 | * triggered a readahead) on a page lock due to an |
7648 | * ordered dio extent we created before but did not have | |
7649 | * yet a corresponding bio submitted (whence it can not | |
ba206a02 | 7650 | * complete), which makes readahead wait for that |
b850ae14 FM |
7651 | * ordered extent to complete while holding a lock on |
7652 | * that page. | |
eb838e73 | 7653 | */ |
b850ae14 | 7654 | ret = -ENOTBLK; |
eb838e73 JB |
7655 | } |
7656 | ||
ade77029 FM |
7657 | if (ret) |
7658 | break; | |
7659 | ||
eb838e73 JB |
7660 | cond_resched(); |
7661 | } | |
7662 | ||
7663 | return ret; | |
7664 | } | |
7665 | ||
6f9994db | 7666 | /* The callers of this must take lock_extent() */ |
4b67c11d NB |
7667 | static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start, |
7668 | u64 len, u64 orig_start, u64 block_start, | |
6f9994db LB |
7669 | u64 block_len, u64 orig_block_len, |
7670 | u64 ram_bytes, int compress_type, | |
7671 | int type) | |
69ffb543 JB |
7672 | { |
7673 | struct extent_map_tree *em_tree; | |
7674 | struct extent_map *em; | |
69ffb543 JB |
7675 | int ret; |
7676 | ||
6f9994db LB |
7677 | ASSERT(type == BTRFS_ORDERED_PREALLOC || |
7678 | type == BTRFS_ORDERED_COMPRESSED || | |
7679 | type == BTRFS_ORDERED_NOCOW || | |
1af4a0aa | 7680 | type == BTRFS_ORDERED_REGULAR); |
6f9994db | 7681 | |
4b67c11d | 7682 | em_tree = &inode->extent_tree; |
69ffb543 JB |
7683 | em = alloc_extent_map(); |
7684 | if (!em) | |
7685 | return ERR_PTR(-ENOMEM); | |
7686 | ||
7687 | em->start = start; | |
7688 | em->orig_start = orig_start; | |
7689 | em->len = len; | |
7690 | em->block_len = block_len; | |
7691 | em->block_start = block_start; | |
b4939680 | 7692 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7693 | em->ram_bytes = ram_bytes; |
70c8a91c | 7694 | em->generation = -1; |
69ffb543 | 7695 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
1af4a0aa | 7696 | if (type == BTRFS_ORDERED_PREALLOC) { |
b11e234d | 7697 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
1af4a0aa | 7698 | } else if (type == BTRFS_ORDERED_COMPRESSED) { |
6f9994db LB |
7699 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
7700 | em->compress_type = compress_type; | |
7701 | } | |
69ffb543 JB |
7702 | |
7703 | do { | |
4b67c11d NB |
7704 | btrfs_drop_extent_cache(inode, em->start, |
7705 | em->start + em->len - 1, 0); | |
69ffb543 | 7706 | write_lock(&em_tree->lock); |
09a2a8f9 | 7707 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 | 7708 | write_unlock(&em_tree->lock); |
6f9994db LB |
7709 | /* |
7710 | * The caller has taken lock_extent(), who could race with us | |
7711 | * to add em? | |
7712 | */ | |
69ffb543 JB |
7713 | } while (ret == -EEXIST); |
7714 | ||
7715 | if (ret) { | |
7716 | free_extent_map(em); | |
7717 | return ERR_PTR(ret); | |
7718 | } | |
7719 | ||
6f9994db | 7720 | /* em got 2 refs now, callers needs to do free_extent_map once. */ |
69ffb543 JB |
7721 | return em; |
7722 | } | |
7723 | ||
1c8d0175 | 7724 | |
c5794e51 | 7725 | static int btrfs_get_blocks_direct_write(struct extent_map **map, |
c5794e51 NB |
7726 | struct inode *inode, |
7727 | struct btrfs_dio_data *dio_data, | |
7728 | u64 start, u64 len) | |
7729 | { | |
7730 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
7731 | struct extent_map *em = *map; | |
7732 | int ret = 0; | |
7733 | ||
7734 | /* | |
7735 | * We don't allocate a new extent in the following cases | |
7736 | * | |
7737 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7738 | * existing extent. | |
7739 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7740 | * just use the extent. | |
7741 | * | |
7742 | */ | |
7743 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7744 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7745 | em->block_start != EXTENT_MAP_HOLE)) { | |
7746 | int type; | |
7747 | u64 block_start, orig_start, orig_block_len, ram_bytes; | |
7748 | ||
7749 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7750 | type = BTRFS_ORDERED_PREALLOC; | |
7751 | else | |
7752 | type = BTRFS_ORDERED_NOCOW; | |
7753 | len = min(len, em->len - (start - em->start)); | |
7754 | block_start = em->block_start + (start - em->start); | |
7755 | ||
7756 | if (can_nocow_extent(inode, start, &len, &orig_start, | |
a84d5d42 | 7757 | &orig_block_len, &ram_bytes, false) == 1 && |
c5794e51 NB |
7758 | btrfs_inc_nocow_writers(fs_info, block_start)) { |
7759 | struct extent_map *em2; | |
7760 | ||
64f54188 | 7761 | em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len, |
c5794e51 NB |
7762 | orig_start, block_start, |
7763 | len, orig_block_len, | |
7764 | ram_bytes, type); | |
7765 | btrfs_dec_nocow_writers(fs_info, block_start); | |
7766 | if (type == BTRFS_ORDERED_PREALLOC) { | |
7767 | free_extent_map(em); | |
7768 | *map = em = em2; | |
7769 | } | |
7770 | ||
7771 | if (em2 && IS_ERR(em2)) { | |
7772 | ret = PTR_ERR(em2); | |
7773 | goto out; | |
7774 | } | |
7775 | /* | |
7776 | * For inode marked NODATACOW or extent marked PREALLOC, | |
7777 | * use the existing or preallocated extent, so does not | |
7778 | * need to adjust btrfs_space_info's bytes_may_use. | |
7779 | */ | |
9db5d510 | 7780 | btrfs_free_reserved_data_space_noquota(fs_info, len); |
c5794e51 NB |
7781 | goto skip_cow; |
7782 | } | |
7783 | } | |
7784 | ||
7785 | /* this will cow the extent */ | |
c5794e51 | 7786 | free_extent_map(em); |
9fc6f911 | 7787 | *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len); |
c5794e51 NB |
7788 | if (IS_ERR(em)) { |
7789 | ret = PTR_ERR(em); | |
7790 | goto out; | |
7791 | } | |
7792 | ||
7793 | len = min(len, em->len - (start - em->start)); | |
7794 | ||
7795 | skip_cow: | |
c5794e51 NB |
7796 | /* |
7797 | * Need to update the i_size under the extent lock so buffered | |
7798 | * readers will get the updated i_size when we unlock. | |
7799 | */ | |
f85781fb | 7800 | if (start + len > i_size_read(inode)) |
c5794e51 NB |
7801 | i_size_write(inode, start + len); |
7802 | ||
c5794e51 | 7803 | dio_data->reserve -= len; |
c5794e51 NB |
7804 | out: |
7805 | return ret; | |
7806 | } | |
7807 | ||
f85781fb GR |
7808 | static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start, |
7809 | loff_t length, unsigned int flags, struct iomap *iomap, | |
7810 | struct iomap *srcmap) | |
4b46fce2 | 7811 | { |
0b246afa | 7812 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7813 | struct extent_map *em; |
eb838e73 | 7814 | struct extent_state *cached_state = NULL; |
50745b0a | 7815 | struct btrfs_dio_data *dio_data = NULL; |
eb838e73 | 7816 | u64 lockstart, lockend; |
f85781fb | 7817 | const bool write = !!(flags & IOMAP_WRITE); |
0934856d | 7818 | int ret = 0; |
f85781fb GR |
7819 | u64 len = length; |
7820 | bool unlock_extents = false; | |
eb838e73 | 7821 | |
f85781fb | 7822 | if (!write) |
0b246afa | 7823 | len = min_t(u64, len, fs_info->sectorsize); |
eb838e73 | 7824 | |
c329861d JB |
7825 | lockstart = start; |
7826 | lockend = start + len - 1; | |
7827 | ||
f85781fb GR |
7828 | /* |
7829 | * The generic stuff only does filemap_write_and_wait_range, which | |
7830 | * isn't enough if we've written compressed pages to this area, so we | |
7831 | * need to flush the dirty pages again to make absolutely sure that any | |
7832 | * outstanding dirty pages are on disk. | |
7833 | */ | |
7834 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
7835 | &BTRFS_I(inode)->runtime_flags)) { | |
7836 | ret = filemap_fdatawrite_range(inode->i_mapping, start, | |
7837 | start + length - 1); | |
7838 | if (ret) | |
7839 | return ret; | |
7840 | } | |
7841 | ||
7842 | dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS); | |
7843 | if (!dio_data) | |
7844 | return -ENOMEM; | |
7845 | ||
7846 | dio_data->length = length; | |
7847 | if (write) { | |
7848 | dio_data->reserve = round_up(length, fs_info->sectorsize); | |
7849 | ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), | |
7850 | &dio_data->data_reserved, | |
7851 | start, dio_data->reserve); | |
7852 | if (ret) { | |
7853 | extent_changeset_free(dio_data->data_reserved); | |
7854 | kfree(dio_data); | |
7855 | return ret; | |
7856 | } | |
e1cbbfa5 | 7857 | } |
f85781fb GR |
7858 | iomap->private = dio_data; |
7859 | ||
e1cbbfa5 | 7860 | |
eb838e73 JB |
7861 | /* |
7862 | * If this errors out it's because we couldn't invalidate pagecache for | |
7863 | * this range and we need to fallback to buffered. | |
7864 | */ | |
f85781fb | 7865 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) { |
9c9464cc FM |
7866 | ret = -ENOTBLK; |
7867 | goto err; | |
7868 | } | |
eb838e73 | 7869 | |
39b07b5d | 7870 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len); |
eb838e73 JB |
7871 | if (IS_ERR(em)) { |
7872 | ret = PTR_ERR(em); | |
7873 | goto unlock_err; | |
7874 | } | |
4b46fce2 JB |
7875 | |
7876 | /* | |
7877 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7878 | * io. INLINE is special, and we could probably kludge it in here, but | |
7879 | * it's still buffered so for safety lets just fall back to the generic | |
7880 | * buffered path. | |
7881 | * | |
7882 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7883 | * decompress it, so there will be buffering required no matter what we | |
7884 | * do, so go ahead and fallback to buffered. | |
7885 | * | |
01327610 | 7886 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7887 | * to buffered IO. Don't blame me, this is the price we pay for using |
7888 | * the generic code. | |
7889 | */ | |
7890 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7891 | em->block_start == EXTENT_MAP_INLINE) { | |
7892 | free_extent_map(em); | |
eb838e73 JB |
7893 | ret = -ENOTBLK; |
7894 | goto unlock_err; | |
4b46fce2 JB |
7895 | } |
7896 | ||
f85781fb GR |
7897 | len = min(len, em->len - (start - em->start)); |
7898 | if (write) { | |
7899 | ret = btrfs_get_blocks_direct_write(&em, inode, dio_data, | |
7900 | start, len); | |
c5794e51 NB |
7901 | if (ret < 0) |
7902 | goto unlock_err; | |
f85781fb GR |
7903 | unlock_extents = true; |
7904 | /* Recalc len in case the new em is smaller than requested */ | |
7905 | len = min(len, em->len - (start - em->start)); | |
c5794e51 | 7906 | } else { |
1c8d0175 NB |
7907 | /* |
7908 | * We need to unlock only the end area that we aren't using. | |
7909 | * The rest is going to be unlocked by the endio routine. | |
7910 | */ | |
f85781fb GR |
7911 | lockstart = start + len; |
7912 | if (lockstart < lockend) | |
7913 | unlock_extents = true; | |
7914 | } | |
7915 | ||
7916 | if (unlock_extents) | |
7917 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, | |
7918 | lockstart, lockend, &cached_state); | |
7919 | else | |
7920 | free_extent_state(cached_state); | |
7921 | ||
7922 | /* | |
7923 | * Translate extent map information to iomap. | |
7924 | * We trim the extents (and move the addr) even though iomap code does | |
7925 | * that, since we have locked only the parts we are performing I/O in. | |
7926 | */ | |
7927 | if ((em->block_start == EXTENT_MAP_HOLE) || | |
7928 | (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) { | |
7929 | iomap->addr = IOMAP_NULL_ADDR; | |
7930 | iomap->type = IOMAP_HOLE; | |
7931 | } else { | |
7932 | iomap->addr = em->block_start + (start - em->start); | |
7933 | iomap->type = IOMAP_MAPPED; | |
a43a67a2 | 7934 | } |
f85781fb GR |
7935 | iomap->offset = start; |
7936 | iomap->bdev = fs_info->fs_devices->latest_bdev; | |
7937 | iomap->length = len; | |
a43a67a2 | 7938 | |
e380adfc | 7939 | if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start)) |
544d24f9 NA |
7940 | iomap->flags |= IOMAP_F_ZONE_APPEND; |
7941 | ||
4b46fce2 JB |
7942 | free_extent_map(em); |
7943 | ||
7944 | return 0; | |
eb838e73 JB |
7945 | |
7946 | unlock_err: | |
e182163d OS |
7947 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7948 | &cached_state); | |
9c9464cc | 7949 | err: |
f85781fb GR |
7950 | if (dio_data) { |
7951 | btrfs_delalloc_release_space(BTRFS_I(inode), | |
7952 | dio_data->data_reserved, start, | |
7953 | dio_data->reserve, true); | |
7954 | btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->reserve); | |
7955 | extent_changeset_free(dio_data->data_reserved); | |
7956 | kfree(dio_data); | |
7957 | } | |
7958 | return ret; | |
7959 | } | |
7960 | ||
7961 | static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length, | |
7962 | ssize_t written, unsigned int flags, struct iomap *iomap) | |
7963 | { | |
7964 | int ret = 0; | |
7965 | struct btrfs_dio_data *dio_data = iomap->private; | |
7966 | size_t submitted = dio_data->submitted; | |
7967 | const bool write = !!(flags & IOMAP_WRITE); | |
7968 | ||
7969 | if (!write && (iomap->type == IOMAP_HOLE)) { | |
7970 | /* If reading from a hole, unlock and return */ | |
7971 | unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1); | |
7972 | goto out; | |
7973 | } | |
7974 | ||
7975 | if (submitted < length) { | |
7976 | pos += submitted; | |
7977 | length -= submitted; | |
7978 | if (write) | |
7979 | __endio_write_update_ordered(BTRFS_I(inode), pos, | |
7980 | length, false); | |
7981 | else | |
7982 | unlock_extent(&BTRFS_I(inode)->io_tree, pos, | |
7983 | pos + length - 1); | |
7984 | ret = -ENOTBLK; | |
7985 | } | |
7986 | ||
7987 | if (write) { | |
7988 | if (dio_data->reserve) | |
7989 | btrfs_delalloc_release_space(BTRFS_I(inode), | |
7990 | dio_data->data_reserved, pos, | |
7991 | dio_data->reserve, true); | |
7992 | btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->length); | |
7993 | extent_changeset_free(dio_data->data_reserved); | |
7994 | } | |
7995 | out: | |
7996 | kfree(dio_data); | |
7997 | iomap->private = NULL; | |
7998 | ||
8b110e39 MX |
7999 | return ret; |
8000 | } | |
8001 | ||
769b4f24 | 8002 | static void btrfs_dio_private_put(struct btrfs_dio_private *dip) |
8b110e39 | 8003 | { |
769b4f24 OS |
8004 | /* |
8005 | * This implies a barrier so that stores to dio_bio->bi_status before | |
8006 | * this and loads of dio_bio->bi_status after this are fully ordered. | |
8007 | */ | |
8008 | if (!refcount_dec_and_test(&dip->refs)) | |
8009 | return; | |
8b110e39 | 8010 | |
cfe94440 | 8011 | if (btrfs_op(dip->dio_bio) == BTRFS_MAP_WRITE) { |
b672b5c1 NB |
8012 | __endio_write_update_ordered(BTRFS_I(dip->inode), |
8013 | dip->logical_offset, | |
769b4f24 OS |
8014 | dip->bytes, |
8015 | !dip->dio_bio->bi_status); | |
8016 | } else { | |
8017 | unlock_extent(&BTRFS_I(dip->inode)->io_tree, | |
8018 | dip->logical_offset, | |
8019 | dip->logical_offset + dip->bytes - 1); | |
8b110e39 MX |
8020 | } |
8021 | ||
f85781fb | 8022 | bio_endio(dip->dio_bio); |
769b4f24 | 8023 | kfree(dip); |
8b110e39 MX |
8024 | } |
8025 | ||
77d5d689 OS |
8026 | static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio, |
8027 | int mirror_num, | |
8028 | unsigned long bio_flags) | |
8b110e39 | 8029 | { |
77d5d689 | 8030 | struct btrfs_dio_private *dip = bio->bi_private; |
2ff7e61e | 8031 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
58efbc9f | 8032 | blk_status_t ret; |
8b110e39 | 8033 | |
37226b21 | 8034 | BUG_ON(bio_op(bio) == REQ_OP_WRITE); |
8b110e39 | 8035 | |
5c047a69 | 8036 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
8b110e39 | 8037 | if (ret) |
ea057f6d | 8038 | return ret; |
8b110e39 | 8039 | |
77d5d689 | 8040 | refcount_inc(&dip->refs); |
08635bae | 8041 | ret = btrfs_map_bio(fs_info, bio, mirror_num); |
8b110e39 | 8042 | if (ret) |
fd9d6670 | 8043 | refcount_dec(&dip->refs); |
77d5d689 | 8044 | return ret; |
8b110e39 MX |
8045 | } |
8046 | ||
fd9d6670 OS |
8047 | static blk_status_t btrfs_check_read_dio_bio(struct inode *inode, |
8048 | struct btrfs_io_bio *io_bio, | |
8049 | const bool uptodate) | |
4b46fce2 | 8050 | { |
fd9d6670 OS |
8051 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
8052 | const u32 sectorsize = fs_info->sectorsize; | |
8053 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8054 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
8055 | const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM); | |
17347cec LB |
8056 | struct bio_vec bvec; |
8057 | struct bvec_iter iter; | |
fd9d6670 | 8058 | u64 start = io_bio->logical; |
7ffd27e3 | 8059 | u32 bio_offset = 0; |
58efbc9f | 8060 | blk_status_t err = BLK_STS_OK; |
4b46fce2 | 8061 | |
fd9d6670 OS |
8062 | __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) { |
8063 | unsigned int i, nr_sectors, pgoff; | |
8b110e39 | 8064 | |
17347cec LB |
8065 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); |
8066 | pgoff = bvec.bv_offset; | |
fd9d6670 | 8067 | for (i = 0; i < nr_sectors; i++) { |
97bf5a55 | 8068 | ASSERT(pgoff < PAGE_SIZE); |
fd9d6670 | 8069 | if (uptodate && |
7ffd27e3 | 8070 | (!csum || !check_data_csum(inode, io_bio, |
c1d6abda OS |
8071 | bio_offset, bvec.bv_page, |
8072 | pgoff, start))) { | |
fd9d6670 OS |
8073 | clean_io_failure(fs_info, failure_tree, io_tree, |
8074 | start, bvec.bv_page, | |
8075 | btrfs_ino(BTRFS_I(inode)), | |
8076 | pgoff); | |
8077 | } else { | |
150e4b05 | 8078 | int ret; |
fd9d6670 | 8079 | |
7ffd27e3 | 8080 | ASSERT((start - io_bio->logical) < UINT_MAX); |
150e4b05 QW |
8081 | ret = btrfs_repair_one_sector(inode, |
8082 | &io_bio->bio, | |
8083 | start - io_bio->logical, | |
8084 | bvec.bv_page, pgoff, | |
8085 | start, io_bio->mirror_num, | |
8086 | submit_dio_repair_bio); | |
8087 | if (ret) | |
8088 | err = errno_to_blk_status(ret); | |
fd9d6670 OS |
8089 | } |
8090 | start += sectorsize; | |
7ffd27e3 QW |
8091 | ASSERT(bio_offset + sectorsize > bio_offset); |
8092 | bio_offset += sectorsize; | |
2dabb324 | 8093 | pgoff += sectorsize; |
2dabb324 | 8094 | } |
2c30c71b | 8095 | } |
c1dc0896 MX |
8096 | return err; |
8097 | } | |
8098 | ||
b672b5c1 | 8099 | static void __endio_write_update_ordered(struct btrfs_inode *inode, |
52427260 QW |
8100 | const u64 offset, const u64 bytes, |
8101 | const bool uptodate) | |
4b46fce2 | 8102 | { |
e65f152e QW |
8103 | btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, |
8104 | finish_ordered_fn, uptodate); | |
14543774 FM |
8105 | } |
8106 | ||
8896a08d | 8107 | static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode, |
1941b64b QW |
8108 | struct bio *bio, |
8109 | u64 dio_file_offset) | |
eaf25d93 | 8110 | { |
1941b64b | 8111 | return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, 1); |
eaf25d93 CM |
8112 | } |
8113 | ||
4246a0b6 | 8114 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
8115 | { |
8116 | struct btrfs_dio_private *dip = bio->bi_private; | |
4e4cbee9 | 8117 | blk_status_t err = bio->bi_status; |
e65e1535 | 8118 | |
8b110e39 MX |
8119 | if (err) |
8120 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
6296b960 | 8121 | "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d", |
f85b7379 | 8122 | btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio), |
1201b58b | 8123 | bio->bi_opf, bio->bi_iter.bi_sector, |
8b110e39 MX |
8124 | bio->bi_iter.bi_size, err); |
8125 | ||
769b4f24 OS |
8126 | if (bio_op(bio) == REQ_OP_READ) { |
8127 | err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio), | |
fd9d6670 | 8128 | !err); |
e65e1535 MX |
8129 | } |
8130 | ||
769b4f24 OS |
8131 | if (err) |
8132 | dip->dio_bio->bi_status = err; | |
e65e1535 | 8133 | |
544d24f9 NA |
8134 | btrfs_record_physical_zoned(dip->inode, dip->logical_offset, bio); |
8135 | ||
e65e1535 | 8136 | bio_put(bio); |
769b4f24 | 8137 | btrfs_dio_private_put(dip); |
c1dc0896 MX |
8138 | } |
8139 | ||
d0ee3934 DS |
8140 | static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio, |
8141 | struct inode *inode, u64 file_offset, int async_submit) | |
e65e1535 | 8142 | { |
0b246afa | 8143 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
facc8a22 | 8144 | struct btrfs_dio_private *dip = bio->bi_private; |
cfe94440 | 8145 | bool write = btrfs_op(bio) == BTRFS_MAP_WRITE; |
4e4cbee9 | 8146 | blk_status_t ret; |
e65e1535 | 8147 | |
4c274bc6 | 8148 | /* Check btrfs_submit_bio_hook() for rules about async submit. */ |
b812ce28 JB |
8149 | if (async_submit) |
8150 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
8151 | ||
5fd02043 | 8152 | if (!write) { |
0b246afa | 8153 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
5fd02043 JB |
8154 | if (ret) |
8155 | goto err; | |
8156 | } | |
e65e1535 | 8157 | |
e6961cac | 8158 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
1ae39938 JB |
8159 | goto map; |
8160 | ||
8161 | if (write && async_submit) { | |
1941b64b | 8162 | ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset, |
e288c080 | 8163 | btrfs_submit_bio_start_direct_io); |
e65e1535 | 8164 | goto err; |
1ae39938 JB |
8165 | } else if (write) { |
8166 | /* | |
8167 | * If we aren't doing async submit, calculate the csum of the | |
8168 | * bio now. | |
8169 | */ | |
bd242a08 | 8170 | ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1); |
1ae39938 JB |
8171 | if (ret) |
8172 | goto err; | |
23ea8e5a | 8173 | } else { |
85879573 OS |
8174 | u64 csum_offset; |
8175 | ||
8176 | csum_offset = file_offset - dip->logical_offset; | |
265fdfa6 | 8177 | csum_offset >>= fs_info->sectorsize_bits; |
55fc29be | 8178 | csum_offset *= fs_info->csum_size; |
85879573 | 8179 | btrfs_io_bio(bio)->csum = dip->csums + csum_offset; |
c2db1073 | 8180 | } |
1ae39938 | 8181 | map: |
08635bae | 8182 | ret = btrfs_map_bio(fs_info, bio, 0); |
e65e1535 | 8183 | err: |
e65e1535 MX |
8184 | return ret; |
8185 | } | |
8186 | ||
c36cac28 OS |
8187 | /* |
8188 | * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked | |
8189 | * or ordered extents whether or not we submit any bios. | |
8190 | */ | |
8191 | static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio, | |
8192 | struct inode *inode, | |
8193 | loff_t file_offset) | |
e65e1535 | 8194 | { |
cfe94440 | 8195 | const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE); |
85879573 OS |
8196 | const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM); |
8197 | size_t dip_size; | |
c36cac28 | 8198 | struct btrfs_dio_private *dip; |
c36cac28 | 8199 | |
85879573 OS |
8200 | dip_size = sizeof(*dip); |
8201 | if (!write && csum) { | |
8202 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
85879573 OS |
8203 | size_t nblocks; |
8204 | ||
265fdfa6 | 8205 | nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits; |
223486c2 | 8206 | dip_size += fs_info->csum_size * nblocks; |
85879573 OS |
8207 | } |
8208 | ||
8209 | dip = kzalloc(dip_size, GFP_NOFS); | |
c36cac28 OS |
8210 | if (!dip) |
8211 | return NULL; | |
8212 | ||
c36cac28 OS |
8213 | dip->inode = inode; |
8214 | dip->logical_offset = file_offset; | |
8215 | dip->bytes = dio_bio->bi_iter.bi_size; | |
1201b58b | 8216 | dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9; |
c36cac28 | 8217 | dip->dio_bio = dio_bio; |
e3b318d1 | 8218 | refcount_set(&dip->refs, 1); |
c36cac28 OS |
8219 | return dip; |
8220 | } | |
8221 | ||
f85781fb GR |
8222 | static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap, |
8223 | struct bio *dio_bio, loff_t file_offset) | |
c36cac28 | 8224 | { |
cfe94440 | 8225 | const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE); |
0b246afa | 8226 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
769b4f24 OS |
8227 | const bool raid56 = (btrfs_data_alloc_profile(fs_info) & |
8228 | BTRFS_BLOCK_GROUP_RAID56_MASK); | |
c36cac28 | 8229 | struct btrfs_dio_private *dip; |
e65e1535 | 8230 | struct bio *bio; |
c36cac28 | 8231 | u64 start_sector; |
1ae39938 | 8232 | int async_submit = 0; |
725130ba LB |
8233 | u64 submit_len; |
8234 | int clone_offset = 0; | |
8235 | int clone_len; | |
42034313 | 8236 | u64 logical; |
5f4dc8fc | 8237 | int ret; |
58efbc9f | 8238 | blk_status_t status; |
89b798ad | 8239 | struct btrfs_io_geometry geom; |
f85781fb | 8240 | struct btrfs_dio_data *dio_data = iomap->private; |
42034313 | 8241 | struct extent_map *em = NULL; |
e65e1535 | 8242 | |
c36cac28 OS |
8243 | dip = btrfs_create_dio_private(dio_bio, inode, file_offset); |
8244 | if (!dip) { | |
8245 | if (!write) { | |
8246 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, | |
8247 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
8248 | } | |
8249 | dio_bio->bi_status = BLK_STS_RESOURCE; | |
f85781fb GR |
8250 | bio_endio(dio_bio); |
8251 | return BLK_QC_T_NONE; | |
c36cac28 | 8252 | } |
facc8a22 | 8253 | |
334c16d8 | 8254 | if (!write) { |
85879573 OS |
8255 | /* |
8256 | * Load the csums up front to reduce csum tree searches and | |
8257 | * contention when submitting bios. | |
334c16d8 JB |
8258 | * |
8259 | * If we have csums disabled this will do nothing. | |
85879573 | 8260 | */ |
6275193e | 8261 | status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums); |
85879573 OS |
8262 | if (status != BLK_STS_OK) |
8263 | goto out_err; | |
02f57c7a JB |
8264 | } |
8265 | ||
769b4f24 OS |
8266 | start_sector = dio_bio->bi_iter.bi_sector; |
8267 | submit_len = dio_bio->bi_iter.bi_size; | |
53b381b3 | 8268 | |
3c91ee69 | 8269 | do { |
42034313 MR |
8270 | logical = start_sector << 9; |
8271 | em = btrfs_get_chunk_map(fs_info, logical, submit_len); | |
8272 | if (IS_ERR(em)) { | |
8273 | status = errno_to_blk_status(PTR_ERR(em)); | |
8274 | em = NULL; | |
8275 | goto out_err_em; | |
8276 | } | |
8277 | ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio), | |
43c0d1a5 | 8278 | logical, &geom); |
769b4f24 OS |
8279 | if (ret) { |
8280 | status = errno_to_blk_status(ret); | |
42034313 | 8281 | goto out_err_em; |
769b4f24 OS |
8282 | } |
8283 | ASSERT(geom.len <= INT_MAX); | |
8284 | ||
89b798ad | 8285 | clone_len = min_t(int, submit_len, geom.len); |
02f57c7a | 8286 | |
725130ba LB |
8287 | /* |
8288 | * This will never fail as it's passing GPF_NOFS and | |
8289 | * the allocation is backed by btrfs_bioset. | |
8290 | */ | |
769b4f24 | 8291 | bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len); |
725130ba LB |
8292 | bio->bi_private = dip; |
8293 | bio->bi_end_io = btrfs_end_dio_bio; | |
8294 | btrfs_io_bio(bio)->logical = file_offset; | |
8295 | ||
544d24f9 NA |
8296 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) { |
8297 | status = extract_ordered_extent(BTRFS_I(inode), bio, | |
8298 | file_offset); | |
8299 | if (status) { | |
8300 | bio_put(bio); | |
8301 | goto out_err; | |
8302 | } | |
8303 | } | |
8304 | ||
725130ba LB |
8305 | ASSERT(submit_len >= clone_len); |
8306 | submit_len -= clone_len; | |
e65e1535 | 8307 | |
725130ba LB |
8308 | /* |
8309 | * Increase the count before we submit the bio so we know | |
8310 | * the end IO handler won't happen before we increase the | |
8311 | * count. Otherwise, the dip might get freed before we're | |
8312 | * done setting it up. | |
769b4f24 OS |
8313 | * |
8314 | * We transfer the initial reference to the last bio, so we | |
8315 | * don't need to increment the reference count for the last one. | |
725130ba | 8316 | */ |
769b4f24 OS |
8317 | if (submit_len > 0) { |
8318 | refcount_inc(&dip->refs); | |
8319 | /* | |
8320 | * If we are submitting more than one bio, submit them | |
8321 | * all asynchronously. The exception is RAID 5 or 6, as | |
8322 | * asynchronous checksums make it difficult to collect | |
8323 | * full stripe writes. | |
8324 | */ | |
8325 | if (!raid56) | |
8326 | async_submit = 1; | |
8327 | } | |
e65e1535 | 8328 | |
d0ee3934 | 8329 | status = btrfs_submit_dio_bio(bio, inode, file_offset, |
58efbc9f OS |
8330 | async_submit); |
8331 | if (status) { | |
725130ba | 8332 | bio_put(bio); |
769b4f24 OS |
8333 | if (submit_len > 0) |
8334 | refcount_dec(&dip->refs); | |
42034313 | 8335 | goto out_err_em; |
725130ba | 8336 | } |
e65e1535 | 8337 | |
f85781fb | 8338 | dio_data->submitted += clone_len; |
725130ba LB |
8339 | clone_offset += clone_len; |
8340 | start_sector += clone_len >> 9; | |
8341 | file_offset += clone_len; | |
42034313 MR |
8342 | |
8343 | free_extent_map(em); | |
3c91ee69 | 8344 | } while (submit_len > 0); |
f85781fb | 8345 | return BLK_QC_T_NONE; |
e65e1535 | 8346 | |
42034313 MR |
8347 | out_err_em: |
8348 | free_extent_map(em); | |
e65e1535 | 8349 | out_err: |
769b4f24 OS |
8350 | dip->dio_bio->bi_status = status; |
8351 | btrfs_dio_private_put(dip); | |
42034313 | 8352 | |
f85781fb | 8353 | return BLK_QC_T_NONE; |
4b46fce2 JB |
8354 | } |
8355 | ||
4e4cabec | 8356 | const struct iomap_ops btrfs_dio_iomap_ops = { |
f85781fb GR |
8357 | .iomap_begin = btrfs_dio_iomap_begin, |
8358 | .iomap_end = btrfs_dio_iomap_end, | |
8359 | }; | |
8360 | ||
4e4cabec | 8361 | const struct iomap_dio_ops btrfs_dio_ops = { |
f85781fb GR |
8362 | .submit_io = btrfs_submit_direct, |
8363 | }; | |
8364 | ||
1506fcc8 | 8365 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
bab16e21 | 8366 | u64 start, u64 len) |
1506fcc8 | 8367 | { |
05dadc09 TI |
8368 | int ret; |
8369 | ||
45dd052e | 8370 | ret = fiemap_prep(inode, fieinfo, start, &len, 0); |
05dadc09 TI |
8371 | if (ret) |
8372 | return ret; | |
8373 | ||
facee0a0 | 8374 | return extent_fiemap(BTRFS_I(inode), fieinfo, start, len); |
1506fcc8 YS |
8375 | } |
8376 | ||
a52d9a80 | 8377 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8378 | { |
0f208812 NB |
8379 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
8380 | u64 start = page_offset(page); | |
8381 | u64 end = start + PAGE_SIZE - 1; | |
390ed29b | 8382 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; |
c1be9c1a NB |
8383 | int ret; |
8384 | ||
0f208812 NB |
8385 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); |
8386 | ||
390ed29b QW |
8387 | ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL); |
8388 | if (bio_ctrl.bio) | |
8389 | ret = submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.bio_flags); | |
c1be9c1a | 8390 | return ret; |
9ebefb18 | 8391 | } |
1832a6d5 | 8392 | |
a52d9a80 | 8393 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8394 | { |
be7bd730 JB |
8395 | struct inode *inode = page->mapping->host; |
8396 | int ret; | |
b888db2b CM |
8397 | |
8398 | if (current->flags & PF_MEMALLOC) { | |
8399 | redirty_page_for_writepage(wbc, page); | |
8400 | unlock_page(page); | |
8401 | return 0; | |
8402 | } | |
be7bd730 JB |
8403 | |
8404 | /* | |
8405 | * If we are under memory pressure we will call this directly from the | |
8406 | * VM, we need to make sure we have the inode referenced for the ordered | |
8407 | * extent. If not just return like we didn't do anything. | |
8408 | */ | |
8409 | if (!igrab(inode)) { | |
8410 | redirty_page_for_writepage(wbc, page); | |
8411 | return AOP_WRITEPAGE_ACTIVATE; | |
8412 | } | |
0a9b0e53 | 8413 | ret = extent_write_full_page(page, wbc); |
be7bd730 JB |
8414 | btrfs_add_delayed_iput(inode); |
8415 | return ret; | |
9ebefb18 CM |
8416 | } |
8417 | ||
48a3b636 ES |
8418 | static int btrfs_writepages(struct address_space *mapping, |
8419 | struct writeback_control *wbc) | |
b293f02e | 8420 | { |
8ae225a8 | 8421 | return extent_writepages(mapping, wbc); |
b293f02e CM |
8422 | } |
8423 | ||
ba206a02 | 8424 | static void btrfs_readahead(struct readahead_control *rac) |
3ab2fb5a | 8425 | { |
ba206a02 | 8426 | extent_readahead(rac); |
3ab2fb5a | 8427 | } |
2a3ff0ad | 8428 | |
7c11d0ae QW |
8429 | /* |
8430 | * For releasepage() and invalidatepage() we have a race window where | |
8431 | * end_page_writeback() is called but the subpage spinlock is not yet released. | |
8432 | * If we continue to release/invalidate the page, we could cause use-after-free | |
8433 | * for subpage spinlock. So this function is to spin and wait for subpage | |
8434 | * spinlock. | |
8435 | */ | |
8436 | static void wait_subpage_spinlock(struct page *page) | |
8437 | { | |
8438 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
8439 | struct btrfs_subpage *subpage; | |
8440 | ||
8441 | if (fs_info->sectorsize == PAGE_SIZE) | |
8442 | return; | |
8443 | ||
8444 | ASSERT(PagePrivate(page) && page->private); | |
8445 | subpage = (struct btrfs_subpage *)page->private; | |
8446 | ||
8447 | /* | |
8448 | * This may look insane as we just acquire the spinlock and release it, | |
8449 | * without doing anything. But we just want to make sure no one is | |
8450 | * still holding the subpage spinlock. | |
8451 | * And since the page is not dirty nor writeback, and we have page | |
8452 | * locked, the only possible way to hold a spinlock is from the endio | |
8453 | * function to clear page writeback. | |
8454 | * | |
8455 | * Here we just acquire the spinlock so that all existing callers | |
8456 | * should exit and we're safe to release/invalidate the page. | |
8457 | */ | |
8458 | spin_lock_irq(&subpage->lock); | |
8459 | spin_unlock_irq(&subpage->lock); | |
8460 | } | |
8461 | ||
e6dcd2dc | 8462 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8463 | { |
477a30ba | 8464 | int ret = try_release_extent_mapping(page, gfp_flags); |
7c11d0ae QW |
8465 | |
8466 | if (ret == 1) { | |
8467 | wait_subpage_spinlock(page); | |
32443de3 | 8468 | clear_page_extent_mapped(page); |
7c11d0ae | 8469 | } |
a52d9a80 | 8470 | return ret; |
39279cc3 CM |
8471 | } |
8472 | ||
e6dcd2dc CM |
8473 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8474 | { | |
98509cfc CM |
8475 | if (PageWriteback(page) || PageDirty(page)) |
8476 | return 0; | |
3ba7ab22 | 8477 | return __btrfs_releasepage(page, gfp_flags); |
e6dcd2dc CM |
8478 | } |
8479 | ||
f8e66081 RG |
8480 | #ifdef CONFIG_MIGRATION |
8481 | static int btrfs_migratepage(struct address_space *mapping, | |
8482 | struct page *newpage, struct page *page, | |
8483 | enum migrate_mode mode) | |
8484 | { | |
8485 | int ret; | |
8486 | ||
8487 | ret = migrate_page_move_mapping(mapping, newpage, page, 0); | |
8488 | if (ret != MIGRATEPAGE_SUCCESS) | |
8489 | return ret; | |
8490 | ||
d1b89bc0 GJ |
8491 | if (page_has_private(page)) |
8492 | attach_page_private(newpage, detach_page_private(page)); | |
f8e66081 | 8493 | |
f57ad937 QW |
8494 | if (PageOrdered(page)) { |
8495 | ClearPageOrdered(page); | |
8496 | SetPageOrdered(newpage); | |
f8e66081 RG |
8497 | } |
8498 | ||
8499 | if (mode != MIGRATE_SYNC_NO_COPY) | |
8500 | migrate_page_copy(newpage, page); | |
8501 | else | |
8502 | migrate_page_states(newpage, page); | |
8503 | return MIGRATEPAGE_SUCCESS; | |
8504 | } | |
8505 | #endif | |
8506 | ||
d47992f8 LC |
8507 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8508 | unsigned int length) | |
39279cc3 | 8509 | { |
53ac7ead | 8510 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
b945a463 | 8511 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
53ac7ead | 8512 | struct extent_io_tree *tree = &inode->io_tree; |
2ac55d41 | 8513 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 8514 | u64 page_start = page_offset(page); |
09cbfeaf | 8515 | u64 page_end = page_start + PAGE_SIZE - 1; |
3b835840 | 8516 | u64 cur; |
53ac7ead | 8517 | int inode_evicting = inode->vfs_inode.i_state & I_FREEING; |
39279cc3 | 8518 | |
8b62b72b | 8519 | /* |
266a2586 QW |
8520 | * We have page locked so no new ordered extent can be created on this |
8521 | * page, nor bio can be submitted for this page. | |
8b62b72b | 8522 | * |
266a2586 | 8523 | * But already submitted bio can still be finished on this page. |
f57ad937 QW |
8524 | * Furthermore, endio function won't skip page which has Ordered |
8525 | * (Private2) already cleared, so it's possible for endio and | |
8526 | * invalidatepage to do the same ordered extent accounting twice | |
8527 | * on one page. | |
266a2586 QW |
8528 | * |
8529 | * So here we wait for any submitted bios to finish, so that we won't | |
8530 | * do double ordered extent accounting on the same page. | |
8b62b72b | 8531 | */ |
e6dcd2dc | 8532 | wait_on_page_writeback(page); |
7c11d0ae | 8533 | wait_subpage_spinlock(page); |
8b62b72b | 8534 | |
bcd77455 QW |
8535 | /* |
8536 | * For subpage case, we have call sites like | |
8537 | * btrfs_punch_hole_lock_range() which passes range not aligned to | |
8538 | * sectorsize. | |
8539 | * If the range doesn't cover the full page, we don't need to and | |
8540 | * shouldn't clear page extent mapped, as page->private can still | |
8541 | * record subpage dirty bits for other part of the range. | |
8542 | * | |
8543 | * For cases that can invalidate the full even the range doesn't | |
8544 | * cover the full page, like invalidating the last page, we're | |
8545 | * still safe to wait for ordered extent to finish. | |
8546 | */ | |
8547 | if (!(offset == 0 && length == PAGE_SIZE)) { | |
e6dcd2dc CM |
8548 | btrfs_releasepage(page, GFP_NOFS); |
8549 | return; | |
8550 | } | |
131e404a FDBM |
8551 | |
8552 | if (!inode_evicting) | |
ff13db41 | 8553 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
951c80f8 | 8554 | |
3b835840 QW |
8555 | cur = page_start; |
8556 | while (cur < page_end) { | |
8557 | struct btrfs_ordered_extent *ordered; | |
8558 | bool delete_states; | |
8559 | u64 range_end; | |
b945a463 | 8560 | u32 range_len; |
3b835840 QW |
8561 | |
8562 | ordered = btrfs_lookup_first_ordered_range(inode, cur, | |
8563 | page_end + 1 - cur); | |
8564 | if (!ordered) { | |
8565 | range_end = page_end; | |
8566 | /* | |
8567 | * No ordered extent covering this range, we are safe | |
8568 | * to delete all extent states in the range. | |
8569 | */ | |
8570 | delete_states = true; | |
8571 | goto next; | |
8572 | } | |
8573 | if (ordered->file_offset > cur) { | |
8574 | /* | |
8575 | * There is a range between [cur, oe->file_offset) not | |
8576 | * covered by any ordered extent. | |
8577 | * We are safe to delete all extent states, and handle | |
8578 | * the ordered extent in the next iteration. | |
8579 | */ | |
8580 | range_end = ordered->file_offset - 1; | |
8581 | delete_states = true; | |
8582 | goto next; | |
8583 | } | |
8584 | ||
8585 | range_end = min(ordered->file_offset + ordered->num_bytes - 1, | |
8586 | page_end); | |
b945a463 QW |
8587 | ASSERT(range_end + 1 - cur < U32_MAX); |
8588 | range_len = range_end + 1 - cur; | |
8589 | if (!btrfs_page_test_ordered(fs_info, page, cur, range_len)) { | |
3b835840 | 8590 | /* |
f57ad937 QW |
8591 | * If Ordered (Private2) is cleared, it means endio has |
8592 | * already been executed for the range. | |
3b835840 QW |
8593 | * We can't delete the extent states as |
8594 | * btrfs_finish_ordered_io() may still use some of them. | |
8595 | */ | |
8596 | delete_states = false; | |
8597 | goto next; | |
8598 | } | |
b945a463 | 8599 | btrfs_page_clear_ordered(fs_info, page, cur, range_len); |
3b835840 | 8600 | |
eb84ae03 | 8601 | /* |
2766ff61 FM |
8602 | * IO on this page will never be started, so we need to account |
8603 | * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW | |
8604 | * here, must leave that up for the ordered extent completion. | |
3b835840 QW |
8605 | * |
8606 | * This will also unlock the range for incoming | |
8607 | * btrfs_finish_ordered_io(). | |
eb84ae03 | 8608 | */ |
131e404a | 8609 | if (!inode_evicting) |
3b835840 | 8610 | clear_extent_bit(tree, cur, range_end, |
2766ff61 | 8611 | EXTENT_DELALLOC | |
131e404a | 8612 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
ae0f1625 | 8613 | EXTENT_DEFRAG, 1, 0, &cached_state); |
3b835840 QW |
8614 | |
8615 | spin_lock_irq(&inode->ordered_tree.lock); | |
8616 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
8617 | ordered->truncated_len = min(ordered->truncated_len, | |
8618 | cur - ordered->file_offset); | |
8619 | spin_unlock_irq(&inode->ordered_tree.lock); | |
8620 | ||
8621 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
f41b6ba9 | 8622 | cur, range_end + 1 - cur)) { |
3b835840 QW |
8623 | btrfs_finish_ordered_io(ordered); |
8624 | /* | |
8625 | * The ordered extent has finished, now we're again | |
8626 | * safe to delete all extent states of the range. | |
8627 | */ | |
8628 | delete_states = true; | |
8629 | } else { | |
8630 | /* | |
8631 | * btrfs_finish_ordered_io() will get executed by endio | |
8632 | * of other pages, thus we can't delete extent states | |
8633 | * anymore | |
8634 | */ | |
8635 | delete_states = false; | |
8636 | } | |
8637 | next: | |
8638 | if (ordered) | |
8639 | btrfs_put_ordered_extent(ordered); | |
8b62b72b | 8640 | /* |
3b835840 QW |
8641 | * Qgroup reserved space handler |
8642 | * Sector(s) here will be either: | |
266a2586 | 8643 | * |
3b835840 QW |
8644 | * 1) Already written to disk or bio already finished |
8645 | * Then its QGROUP_RESERVED bit in io_tree is already cleared. | |
8646 | * Qgroup will be handled by its qgroup_record then. | |
8647 | * btrfs_qgroup_free_data() call will do nothing here. | |
8648 | * | |
8649 | * 2) Not written to disk yet | |
8650 | * Then btrfs_qgroup_free_data() call will clear the | |
8651 | * QGROUP_RESERVED bit of its io_tree, and free the qgroup | |
8652 | * reserved data space. | |
8653 | * Since the IO will never happen for this page. | |
8b62b72b | 8654 | */ |
3b835840 | 8655 | btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur); |
131e404a | 8656 | if (!inode_evicting) { |
3b835840 QW |
8657 | clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED | |
8658 | EXTENT_DELALLOC | EXTENT_UPTODATE | | |
8659 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, | |
8660 | delete_states, &cached_state); | |
131e404a | 8661 | } |
3b835840 | 8662 | cur = range_end + 1; |
131e404a | 8663 | } |
b9d0b389 | 8664 | /* |
3b835840 | 8665 | * We have iterated through all ordered extents of the page, the page |
f57ad937 QW |
8666 | * should not have Ordered (Private2) anymore, or the above iteration |
8667 | * did something wrong. | |
b9d0b389 | 8668 | */ |
f57ad937 | 8669 | ASSERT(!PageOrdered(page)); |
3b835840 | 8670 | if (!inode_evicting) |
131e404a | 8671 | __btrfs_releasepage(page, GFP_NOFS); |
4a096752 | 8672 | ClearPageChecked(page); |
32443de3 | 8673 | clear_page_extent_mapped(page); |
39279cc3 CM |
8674 | } |
8675 | ||
9ebefb18 CM |
8676 | /* |
8677 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8678 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8679 | * be careful to check for EOF conditions here. We set the page up correctly | |
8680 | * for a written page which means we get ENOSPC checking when writing into | |
8681 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8682 | * support these features. | |
8683 | * | |
8684 | * We are not allowed to take the i_mutex here so we have to play games to | |
8685 | * protect against truncate races as the page could now be beyond EOF. Because | |
d1342aad OS |
8686 | * truncate_setsize() writes the inode size before removing pages, once we have |
8687 | * the page lock we can determine safely if the page is beyond EOF. If it is not | |
9ebefb18 CM |
8688 | * beyond EOF, then the page is guaranteed safe against truncation until we |
8689 | * unlock the page. | |
8690 | */ | |
a528a241 | 8691 | vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf) |
9ebefb18 | 8692 | { |
c2ec175c | 8693 | struct page *page = vmf->page; |
11bac800 | 8694 | struct inode *inode = file_inode(vmf->vma->vm_file); |
0b246afa | 8695 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
8696 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8697 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8698 | struct extent_state *cached_state = NULL; |
364ecf36 | 8699 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc | 8700 | unsigned long zero_start; |
9ebefb18 | 8701 | loff_t size; |
a528a241 SJ |
8702 | vm_fault_t ret; |
8703 | int ret2; | |
9998eb70 | 8704 | int reserved = 0; |
d0b7da88 | 8705 | u64 reserved_space; |
a52d9a80 | 8706 | u64 page_start; |
e6dcd2dc | 8707 | u64 page_end; |
d0b7da88 CR |
8708 | u64 end; |
8709 | ||
09cbfeaf | 8710 | reserved_space = PAGE_SIZE; |
9ebefb18 | 8711 | |
b2b5ef5c | 8712 | sb_start_pagefault(inode->i_sb); |
df480633 | 8713 | page_start = page_offset(page); |
09cbfeaf | 8714 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 8715 | end = page_end; |
df480633 | 8716 | |
d0b7da88 CR |
8717 | /* |
8718 | * Reserving delalloc space after obtaining the page lock can lead to | |
8719 | * deadlock. For example, if a dirty page is locked by this function | |
8720 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
8721 | * dirty page write out, then the btrfs_writepage() function could | |
8722 | * end up waiting indefinitely to get a lock on the page currently | |
8723 | * being processed by btrfs_page_mkwrite() function. | |
8724 | */ | |
e5b7231e NB |
8725 | ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved, |
8726 | page_start, reserved_space); | |
a528a241 SJ |
8727 | if (!ret2) { |
8728 | ret2 = file_update_time(vmf->vma->vm_file); | |
9998eb70 CM |
8729 | reserved = 1; |
8730 | } | |
a528a241 SJ |
8731 | if (ret2) { |
8732 | ret = vmf_error(ret2); | |
9998eb70 CM |
8733 | if (reserved) |
8734 | goto out; | |
8735 | goto out_noreserve; | |
56a76f82 | 8736 | } |
1832a6d5 | 8737 | |
56a76f82 | 8738 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8739 | again: |
8318ba79 | 8740 | down_read(&BTRFS_I(inode)->i_mmap_lock); |
9ebefb18 | 8741 | lock_page(page); |
9ebefb18 | 8742 | size = i_size_read(inode); |
a52d9a80 | 8743 | |
9ebefb18 | 8744 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8745 | (page_start >= size)) { |
9ebefb18 CM |
8746 | /* page got truncated out from underneath us */ |
8747 | goto out_unlock; | |
8748 | } | |
e6dcd2dc CM |
8749 | wait_on_page_writeback(page); |
8750 | ||
ff13db41 | 8751 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
32443de3 QW |
8752 | ret2 = set_page_extent_mapped(page); |
8753 | if (ret2 < 0) { | |
8754 | ret = vmf_error(ret2); | |
8755 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state); | |
8756 | goto out_unlock; | |
8757 | } | |
e6dcd2dc | 8758 | |
eb84ae03 CM |
8759 | /* |
8760 | * we can't set the delalloc bits if there are pending ordered | |
8761 | * extents. Drop our locks and wait for them to finish | |
8762 | */ | |
a776c6fa NB |
8763 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
8764 | PAGE_SIZE); | |
e6dcd2dc | 8765 | if (ordered) { |
2ac55d41 | 8766 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 8767 | &cached_state); |
e6dcd2dc | 8768 | unlock_page(page); |
8318ba79 | 8769 | up_read(&BTRFS_I(inode)->i_mmap_lock); |
c0a43603 | 8770 | btrfs_start_ordered_extent(ordered, 1); |
e6dcd2dc CM |
8771 | btrfs_put_ordered_extent(ordered); |
8772 | goto again; | |
8773 | } | |
8774 | ||
09cbfeaf | 8775 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 8776 | reserved_space = round_up(size - page_start, |
0b246afa | 8777 | fs_info->sectorsize); |
09cbfeaf | 8778 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 | 8779 | end = page_start + reserved_space - 1; |
86d52921 NB |
8780 | btrfs_delalloc_release_space(BTRFS_I(inode), |
8781 | data_reserved, page_start, | |
8782 | PAGE_SIZE - reserved_space, true); | |
d0b7da88 CR |
8783 | } |
8784 | } | |
8785 | ||
fbf19087 | 8786 | /* |
5416034f LB |
8787 | * page_mkwrite gets called when the page is firstly dirtied after it's |
8788 | * faulted in, but write(2) could also dirty a page and set delalloc | |
8789 | * bits, thus in this case for space account reason, we still need to | |
8790 | * clear any delalloc bits within this page range since we have to | |
8791 | * reserve data&meta space before lock_page() (see above comments). | |
fbf19087 | 8792 | */ |
d0b7da88 | 8793 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
e182163d OS |
8794 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | |
8795 | EXTENT_DEFRAG, 0, 0, &cached_state); | |
fbf19087 | 8796 | |
c2566f22 | 8797 | ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0, |
330a5827 | 8798 | &cached_state); |
a528a241 | 8799 | if (ret2) { |
2ac55d41 | 8800 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 8801 | &cached_state); |
9ed74f2d JB |
8802 | ret = VM_FAULT_SIGBUS; |
8803 | goto out_unlock; | |
8804 | } | |
9ebefb18 CM |
8805 | |
8806 | /* page is wholly or partially inside EOF */ | |
09cbfeaf | 8807 | if (page_start + PAGE_SIZE > size) |
7073017a | 8808 | zero_start = offset_in_page(size); |
9ebefb18 | 8809 | else |
09cbfeaf | 8810 | zero_start = PAGE_SIZE; |
9ebefb18 | 8811 | |
09cbfeaf | 8812 | if (zero_start != PAGE_SIZE) { |
d048b9c2 | 8813 | memzero_page(page, zero_start, PAGE_SIZE - zero_start); |
e6dcd2dc | 8814 | flush_dcache_page(page); |
e6dcd2dc | 8815 | } |
247e743c | 8816 | ClearPageChecked(page); |
2d8ec40e QW |
8817 | btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start); |
8818 | btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start); | |
5a3f23d5 | 8819 | |
bc0939fc | 8820 | btrfs_set_inode_last_sub_trans(BTRFS_I(inode)); |
257c62e1 | 8821 | |
e43bbe5e | 8822 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state); |
8318ba79 | 8823 | up_read(&BTRFS_I(inode)->i_mmap_lock); |
9ebefb18 | 8824 | |
76de60ed YY |
8825 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
8826 | sb_end_pagefault(inode->i_sb); | |
8827 | extent_changeset_free(data_reserved); | |
8828 | return VM_FAULT_LOCKED; | |
717beb96 CM |
8829 | |
8830 | out_unlock: | |
9ebefb18 | 8831 | unlock_page(page); |
8318ba79 | 8832 | up_read(&BTRFS_I(inode)->i_mmap_lock); |
1832a6d5 | 8833 | out: |
8702ba93 | 8834 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
86d52921 | 8835 | btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start, |
43b18595 | 8836 | reserved_space, (ret != 0)); |
9998eb70 | 8837 | out_noreserve: |
b2b5ef5c | 8838 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 8839 | extent_changeset_free(data_reserved); |
9ebefb18 CM |
8840 | return ret; |
8841 | } | |
8842 | ||
213e8c55 | 8843 | static int btrfs_truncate(struct inode *inode, bool skip_writeback) |
39279cc3 | 8844 | { |
0b246afa | 8845 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 8846 | struct btrfs_root *root = BTRFS_I(inode)->root; |
fcb80c2a | 8847 | struct btrfs_block_rsv *rsv; |
ad7e1a74 | 8848 | int ret; |
39279cc3 | 8849 | struct btrfs_trans_handle *trans; |
0b246afa | 8850 | u64 mask = fs_info->sectorsize - 1; |
2bd36e7b | 8851 | u64 min_size = btrfs_calc_metadata_size(fs_info, 1); |
0d7d3165 | 8852 | u64 extents_found = 0; |
39279cc3 | 8853 | |
213e8c55 FM |
8854 | if (!skip_writeback) { |
8855 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), | |
8856 | (u64)-1); | |
8857 | if (ret) | |
8858 | return ret; | |
8859 | } | |
39279cc3 | 8860 | |
fcb80c2a | 8861 | /* |
f7e9e8fc OS |
8862 | * Yes ladies and gentlemen, this is indeed ugly. We have a couple of |
8863 | * things going on here: | |
fcb80c2a | 8864 | * |
f7e9e8fc | 8865 | * 1) We need to reserve space to update our inode. |
fcb80c2a | 8866 | * |
f7e9e8fc | 8867 | * 2) We need to have something to cache all the space that is going to |
fcb80c2a JB |
8868 | * be free'd up by the truncate operation, but also have some slack |
8869 | * space reserved in case it uses space during the truncate (thank you | |
8870 | * very much snapshotting). | |
8871 | * | |
f7e9e8fc | 8872 | * And we need these to be separate. The fact is we can use a lot of |
fcb80c2a | 8873 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 8874 | * we will use, so we need the truncate reservation to be separate so it |
f7e9e8fc OS |
8875 | * doesn't end up using space reserved for updating the inode. We also |
8876 | * need to be able to stop the transaction and start a new one, which | |
8877 | * means we need to be able to update the inode several times, and we | |
8878 | * have no idea of knowing how many times that will be, so we can't just | |
8879 | * reserve 1 item for the entirety of the operation, so that has to be | |
8880 | * done separately as well. | |
fcb80c2a JB |
8881 | * |
8882 | * So that leaves us with | |
8883 | * | |
f7e9e8fc | 8884 | * 1) rsv - for the truncate reservation, which we will steal from the |
fcb80c2a | 8885 | * transaction reservation. |
f7e9e8fc | 8886 | * 2) fs_info->trans_block_rsv - this will have 1 items worth left for |
fcb80c2a JB |
8887 | * updating the inode. |
8888 | */ | |
2ff7e61e | 8889 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
8890 | if (!rsv) |
8891 | return -ENOMEM; | |
4a338542 | 8892 | rsv->size = min_size; |
ca7e70f5 | 8893 | rsv->failfast = 1; |
f0cd846e | 8894 | |
907cbceb | 8895 | /* |
07127184 | 8896 | * 1 for the truncate slack space |
907cbceb JB |
8897 | * 1 for updating the inode. |
8898 | */ | |
f3fe820c | 8899 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a | 8900 | if (IS_ERR(trans)) { |
ad7e1a74 | 8901 | ret = PTR_ERR(trans); |
fcb80c2a JB |
8902 | goto out; |
8903 | } | |
f0cd846e | 8904 | |
907cbceb | 8905 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 8906 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
3a584174 | 8907 | min_size, false); |
fcb80c2a | 8908 | BUG_ON(ret); |
f0cd846e | 8909 | |
ca7e70f5 | 8910 | trans->block_rsv = rsv; |
907cbceb | 8911 | |
8082510e | 8912 | while (1) { |
50743398 | 8913 | ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode), |
8082510e | 8914 | inode->i_size, |
0d7d3165 FM |
8915 | BTRFS_EXTENT_DATA_KEY, |
8916 | &extents_found); | |
ddfae63c | 8917 | trans->block_rsv = &fs_info->trans_block_rsv; |
ad7e1a74 | 8918 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 8919 | break; |
39279cc3 | 8920 | |
9a56fcd1 | 8921 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
ad7e1a74 | 8922 | if (ret) |
3893e33b | 8923 | break; |
ca7e70f5 | 8924 | |
3a45bb20 | 8925 | btrfs_end_transaction(trans); |
2ff7e61e | 8926 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
8927 | |
8928 | trans = btrfs_start_transaction(root, 2); | |
8929 | if (IS_ERR(trans)) { | |
ad7e1a74 | 8930 | ret = PTR_ERR(trans); |
ca7e70f5 JB |
8931 | trans = NULL; |
8932 | break; | |
8933 | } | |
8934 | ||
63f018be | 8935 | btrfs_block_rsv_release(fs_info, rsv, -1, NULL); |
0b246afa | 8936 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
3a584174 | 8937 | rsv, min_size, false); |
ca7e70f5 JB |
8938 | BUG_ON(ret); /* shouldn't happen */ |
8939 | trans->block_rsv = rsv; | |
8082510e YZ |
8940 | } |
8941 | ||
ddfae63c JB |
8942 | /* |
8943 | * We can't call btrfs_truncate_block inside a trans handle as we could | |
8944 | * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know | |
8945 | * we've truncated everything except the last little bit, and can do | |
8946 | * btrfs_truncate_block and then update the disk_i_size. | |
8947 | */ | |
8948 | if (ret == NEED_TRUNCATE_BLOCK) { | |
8949 | btrfs_end_transaction(trans); | |
8950 | btrfs_btree_balance_dirty(fs_info); | |
8951 | ||
217f42eb | 8952 | ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0); |
ddfae63c JB |
8953 | if (ret) |
8954 | goto out; | |
8955 | trans = btrfs_start_transaction(root, 1); | |
8956 | if (IS_ERR(trans)) { | |
8957 | ret = PTR_ERR(trans); | |
8958 | goto out; | |
8959 | } | |
76aea537 | 8960 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
ddfae63c JB |
8961 | } |
8962 | ||
917c16b2 | 8963 | if (trans) { |
ad7e1a74 OS |
8964 | int ret2; |
8965 | ||
0b246afa | 8966 | trans->block_rsv = &fs_info->trans_block_rsv; |
9a56fcd1 | 8967 | ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
ad7e1a74 OS |
8968 | if (ret2 && !ret) |
8969 | ret = ret2; | |
7b128766 | 8970 | |
ad7e1a74 OS |
8971 | ret2 = btrfs_end_transaction(trans); |
8972 | if (ret2 && !ret) | |
8973 | ret = ret2; | |
2ff7e61e | 8974 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 8975 | } |
fcb80c2a | 8976 | out: |
2ff7e61e | 8977 | btrfs_free_block_rsv(fs_info, rsv); |
0d7d3165 FM |
8978 | /* |
8979 | * So if we truncate and then write and fsync we normally would just | |
8980 | * write the extents that changed, which is a problem if we need to | |
8981 | * first truncate that entire inode. So set this flag so we write out | |
8982 | * all of the extents in the inode to the sync log so we're completely | |
8983 | * safe. | |
8984 | * | |
8985 | * If no extents were dropped or trimmed we don't need to force the next | |
8986 | * fsync to truncate all the inode's items from the log and re-log them | |
8987 | * all. This means the truncate operation did not change the file size, | |
8988 | * or changed it to a smaller size but there was only an implicit hole | |
8989 | * between the old i_size and the new i_size, and there were no prealloc | |
8990 | * extents beyond i_size to drop. | |
8991 | */ | |
8992 | if (extents_found > 0) | |
8993 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
fcb80c2a | 8994 | |
ad7e1a74 | 8995 | return ret; |
39279cc3 CM |
8996 | } |
8997 | ||
d352ac68 CM |
8998 | /* |
8999 | * create a new subvolume directory/inode (helper for the ioctl). | |
9000 | */ | |
d2fb3437 | 9001 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 | 9002 | struct btrfs_root *new_root, |
69948022 | 9003 | struct btrfs_root *parent_root) |
39279cc3 | 9004 | { |
39279cc3 | 9005 | struct inode *inode; |
76dda93c | 9006 | int err; |
00e4e6b3 | 9007 | u64 index = 0; |
23125104 NB |
9008 | u64 ino; |
9009 | ||
9010 | err = btrfs_get_free_objectid(new_root, &ino); | |
9011 | if (err < 0) | |
9012 | return err; | |
39279cc3 | 9013 | |
23125104 | 9014 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, ino, ino, |
12fc9d09 FA |
9015 | S_IFDIR | (~current_umask() & S_IRWXUGO), |
9016 | &index); | |
54aa1f4d | 9017 | if (IS_ERR(inode)) |
f46b5a66 | 9018 | return PTR_ERR(inode); |
39279cc3 CM |
9019 | inode->i_op = &btrfs_dir_inode_operations; |
9020 | inode->i_fop = &btrfs_dir_file_operations; | |
9021 | ||
bfe86848 | 9022 | set_nlink(inode, 1); |
6ef06d27 | 9023 | btrfs_i_size_write(BTRFS_I(inode), 0); |
b0d5d10f | 9024 | unlock_new_inode(inode); |
3b96362c | 9025 | |
63541927 FDBM |
9026 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
9027 | if (err) | |
9028 | btrfs_err(new_root->fs_info, | |
351fd353 | 9029 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
9030 | new_root->root_key.objectid, err); |
9031 | ||
9a56fcd1 | 9032 | err = btrfs_update_inode(trans, new_root, BTRFS_I(inode)); |
cb8e7090 | 9033 | |
76dda93c | 9034 | iput(inode); |
ce598979 | 9035 | return err; |
39279cc3 CM |
9036 | } |
9037 | ||
39279cc3 CM |
9038 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
9039 | { | |
69fe2d75 | 9040 | struct btrfs_fs_info *fs_info = btrfs_sb(sb); |
39279cc3 | 9041 | struct btrfs_inode *ei; |
2ead6ae7 | 9042 | struct inode *inode; |
39279cc3 | 9043 | |
712e36c5 | 9044 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL); |
39279cc3 CM |
9045 | if (!ei) |
9046 | return NULL; | |
2ead6ae7 YZ |
9047 | |
9048 | ei->root = NULL; | |
2ead6ae7 | 9049 | ei->generation = 0; |
15ee9bc7 | 9050 | ei->last_trans = 0; |
257c62e1 | 9051 | ei->last_sub_trans = 0; |
e02119d5 | 9052 | ei->logged_trans = 0; |
2ead6ae7 | 9053 | ei->delalloc_bytes = 0; |
a7e3b975 | 9054 | ei->new_delalloc_bytes = 0; |
47059d93 | 9055 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
9056 | ei->disk_i_size = 0; |
9057 | ei->flags = 0; | |
7709cde3 | 9058 | ei->csum_bytes = 0; |
2ead6ae7 | 9059 | ei->index_cnt = (u64)-1; |
67de1176 | 9060 | ei->dir_index = 0; |
2ead6ae7 | 9061 | ei->last_unlink_trans = 0; |
3ebac17c | 9062 | ei->last_reflink_trans = 0; |
46d8bc34 | 9063 | ei->last_log_commit = 0; |
2ead6ae7 | 9064 | |
9e0baf60 JB |
9065 | spin_lock_init(&ei->lock); |
9066 | ei->outstanding_extents = 0; | |
69fe2d75 JB |
9067 | if (sb->s_magic != BTRFS_TEST_MAGIC) |
9068 | btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv, | |
9069 | BTRFS_BLOCK_RSV_DELALLOC); | |
72ac3c0d | 9070 | ei->runtime_flags = 0; |
b52aa8c9 | 9071 | ei->prop_compress = BTRFS_COMPRESS_NONE; |
eec63c65 | 9072 | ei->defrag_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 9073 | |
16cdcec7 MX |
9074 | ei->delayed_node = NULL; |
9075 | ||
9cc97d64 | 9076 | ei->i_otime.tv_sec = 0; |
9077 | ei->i_otime.tv_nsec = 0; | |
9078 | ||
2ead6ae7 | 9079 | inode = &ei->vfs_inode; |
a8067e02 | 9080 | extent_map_tree_init(&ei->extent_tree); |
43eb5f29 QW |
9081 | extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode); |
9082 | extent_io_tree_init(fs_info, &ei->io_failure_tree, | |
9083 | IO_TREE_INODE_IO_FAILURE, inode); | |
41a2ee75 JB |
9084 | extent_io_tree_init(fs_info, &ei->file_extent_tree, |
9085 | IO_TREE_INODE_FILE_EXTENT, inode); | |
7b439738 DS |
9086 | ei->io_tree.track_uptodate = true; |
9087 | ei->io_failure_tree.track_uptodate = true; | |
b812ce28 | 9088 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 9089 | mutex_init(&ei->log_mutex); |
e6dcd2dc | 9090 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 9091 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 9092 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 9093 | RB_CLEAR_NODE(&ei->rb_node); |
8318ba79 | 9094 | init_rwsem(&ei->i_mmap_lock); |
2ead6ae7 YZ |
9095 | |
9096 | return inode; | |
39279cc3 CM |
9097 | } |
9098 | ||
aaedb55b JB |
9099 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
9100 | void btrfs_test_destroy_inode(struct inode *inode) | |
9101 | { | |
dcdbc059 | 9102 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
aaedb55b JB |
9103 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9104 | } | |
9105 | #endif | |
9106 | ||
26602cab | 9107 | void btrfs_free_inode(struct inode *inode) |
fa0d7e3d | 9108 | { |
fa0d7e3d NP |
9109 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9110 | } | |
9111 | ||
633cc816 | 9112 | void btrfs_destroy_inode(struct inode *vfs_inode) |
39279cc3 | 9113 | { |
e6dcd2dc | 9114 | struct btrfs_ordered_extent *ordered; |
633cc816 NB |
9115 | struct btrfs_inode *inode = BTRFS_I(vfs_inode); |
9116 | struct btrfs_root *root = inode->root; | |
5a3f23d5 | 9117 | |
633cc816 NB |
9118 | WARN_ON(!hlist_empty(&vfs_inode->i_dentry)); |
9119 | WARN_ON(vfs_inode->i_data.nrpages); | |
9120 | WARN_ON(inode->block_rsv.reserved); | |
9121 | WARN_ON(inode->block_rsv.size); | |
9122 | WARN_ON(inode->outstanding_extents); | |
9123 | WARN_ON(inode->delalloc_bytes); | |
9124 | WARN_ON(inode->new_delalloc_bytes); | |
9125 | WARN_ON(inode->csum_bytes); | |
9126 | WARN_ON(inode->defrag_bytes); | |
39279cc3 | 9127 | |
a6dbd429 JB |
9128 | /* |
9129 | * This can happen where we create an inode, but somebody else also | |
9130 | * created the same inode and we need to destroy the one we already | |
9131 | * created. | |
9132 | */ | |
9133 | if (!root) | |
26602cab | 9134 | return; |
a6dbd429 | 9135 | |
d397712b | 9136 | while (1) { |
633cc816 | 9137 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
e6dcd2dc CM |
9138 | if (!ordered) |
9139 | break; | |
9140 | else { | |
633cc816 | 9141 | btrfs_err(root->fs_info, |
5d163e0e | 9142 | "found ordered extent %llu %llu on inode cleanup", |
bffe633e | 9143 | ordered->file_offset, ordered->num_bytes); |
71fe0a55 | 9144 | btrfs_remove_ordered_extent(inode, ordered); |
e6dcd2dc CM |
9145 | btrfs_put_ordered_extent(ordered); |
9146 | btrfs_put_ordered_extent(ordered); | |
9147 | } | |
9148 | } | |
633cc816 NB |
9149 | btrfs_qgroup_check_reserved_leak(inode); |
9150 | inode_tree_del(inode); | |
9151 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); | |
9152 | btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1); | |
9153 | btrfs_put_root(inode->root); | |
39279cc3 CM |
9154 | } |
9155 | ||
45321ac5 | 9156 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
9157 | { |
9158 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 9159 | |
6379ef9f NA |
9160 | if (root == NULL) |
9161 | return 1; | |
9162 | ||
fa6ac876 | 9163 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 9164 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 9165 | return 1; |
76dda93c | 9166 | else |
45321ac5 | 9167 | return generic_drop_inode(inode); |
76dda93c YZ |
9168 | } |
9169 | ||
0ee0fda0 | 9170 | static void init_once(void *foo) |
39279cc3 CM |
9171 | { |
9172 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
9173 | ||
9174 | inode_init_once(&ei->vfs_inode); | |
9175 | } | |
9176 | ||
e67c718b | 9177 | void __cold btrfs_destroy_cachep(void) |
39279cc3 | 9178 | { |
8c0a8537 KS |
9179 | /* |
9180 | * Make sure all delayed rcu free inodes are flushed before we | |
9181 | * destroy cache. | |
9182 | */ | |
9183 | rcu_barrier(); | |
5598e900 KM |
9184 | kmem_cache_destroy(btrfs_inode_cachep); |
9185 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5598e900 KM |
9186 | kmem_cache_destroy(btrfs_path_cachep); |
9187 | kmem_cache_destroy(btrfs_free_space_cachep); | |
3acd4850 | 9188 | kmem_cache_destroy(btrfs_free_space_bitmap_cachep); |
39279cc3 CM |
9189 | } |
9190 | ||
f5c29bd9 | 9191 | int __init btrfs_init_cachep(void) |
39279cc3 | 9192 | { |
837e1972 | 9193 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 9194 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
9195 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
9196 | init_once); | |
39279cc3 CM |
9197 | if (!btrfs_inode_cachep) |
9198 | goto fail; | |
9601e3f6 | 9199 | |
837e1972 | 9200 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 | 9201 | sizeof(struct btrfs_trans_handle), 0, |
fba4b697 | 9202 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9203 | if (!btrfs_trans_handle_cachep) |
9204 | goto fail; | |
9601e3f6 | 9205 | |
837e1972 | 9206 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 | 9207 | sizeof(struct btrfs_path), 0, |
fba4b697 | 9208 | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9209 | if (!btrfs_path_cachep) |
9210 | goto fail; | |
9601e3f6 | 9211 | |
837e1972 | 9212 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 | 9213 | sizeof(struct btrfs_free_space), 0, |
fba4b697 | 9214 | SLAB_MEM_SPREAD, NULL); |
dc89e982 JB |
9215 | if (!btrfs_free_space_cachep) |
9216 | goto fail; | |
9217 | ||
3acd4850 CL |
9218 | btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap", |
9219 | PAGE_SIZE, PAGE_SIZE, | |
34e49994 | 9220 | SLAB_MEM_SPREAD, NULL); |
3acd4850 CL |
9221 | if (!btrfs_free_space_bitmap_cachep) |
9222 | goto fail; | |
9223 | ||
39279cc3 CM |
9224 | return 0; |
9225 | fail: | |
9226 | btrfs_destroy_cachep(); | |
9227 | return -ENOMEM; | |
9228 | } | |
9229 | ||
549c7297 CB |
9230 | static int btrfs_getattr(struct user_namespace *mnt_userns, |
9231 | const struct path *path, struct kstat *stat, | |
a528d35e | 9232 | u32 request_mask, unsigned int flags) |
39279cc3 | 9233 | { |
df0af1a5 | 9234 | u64 delalloc_bytes; |
2766ff61 | 9235 | u64 inode_bytes; |
a528d35e | 9236 | struct inode *inode = d_inode(path->dentry); |
fadc0d8b | 9237 | u32 blocksize = inode->i_sb->s_blocksize; |
04a87e34 YS |
9238 | u32 bi_flags = BTRFS_I(inode)->flags; |
9239 | ||
9240 | stat->result_mask |= STATX_BTIME; | |
9241 | stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec; | |
9242 | stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec; | |
9243 | if (bi_flags & BTRFS_INODE_APPEND) | |
9244 | stat->attributes |= STATX_ATTR_APPEND; | |
9245 | if (bi_flags & BTRFS_INODE_COMPRESS) | |
9246 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
9247 | if (bi_flags & BTRFS_INODE_IMMUTABLE) | |
9248 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
9249 | if (bi_flags & BTRFS_INODE_NODUMP) | |
9250 | stat->attributes |= STATX_ATTR_NODUMP; | |
9251 | ||
9252 | stat->attributes_mask |= (STATX_ATTR_APPEND | | |
9253 | STATX_ATTR_COMPRESSED | | |
9254 | STATX_ATTR_IMMUTABLE | | |
9255 | STATX_ATTR_NODUMP); | |
fadc0d8b | 9256 | |
0d56a451 | 9257 | generic_fillattr(&init_user_ns, inode, stat); |
0ee5dc67 | 9258 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
9259 | |
9260 | spin_lock(&BTRFS_I(inode)->lock); | |
a7e3b975 | 9261 | delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes; |
2766ff61 | 9262 | inode_bytes = inode_get_bytes(inode); |
df0af1a5 | 9263 | spin_unlock(&BTRFS_I(inode)->lock); |
2766ff61 | 9264 | stat->blocks = (ALIGN(inode_bytes, blocksize) + |
df0af1a5 | 9265 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
9266 | return 0; |
9267 | } | |
9268 | ||
cdd1fedf DF |
9269 | static int btrfs_rename_exchange(struct inode *old_dir, |
9270 | struct dentry *old_dentry, | |
9271 | struct inode *new_dir, | |
9272 | struct dentry *new_dentry) | |
9273 | { | |
0b246afa | 9274 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf DF |
9275 | struct btrfs_trans_handle *trans; |
9276 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
9277 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
9278 | struct inode *new_inode = new_dentry->d_inode; | |
9279 | struct inode *old_inode = old_dentry->d_inode; | |
95582b00 | 9280 | struct timespec64 ctime = current_time(old_inode); |
4a0cc7ca NB |
9281 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
9282 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
9283 | u64 old_idx = 0; |
9284 | u64 new_idx = 0; | |
cdd1fedf | 9285 | int ret; |
75b463d2 | 9286 | int ret2; |
86e8aa0e FM |
9287 | bool root_log_pinned = false; |
9288 | bool dest_log_pinned = false; | |
dc09ef35 | 9289 | bool need_abort = false; |
cdd1fedf | 9290 | |
3f79f6f6 N |
9291 | /* |
9292 | * For non-subvolumes allow exchange only within one subvolume, in the | |
9293 | * same inode namespace. Two subvolumes (represented as directory) can | |
9294 | * be exchanged as they're a logical link and have a fixed inode number. | |
9295 | */ | |
9296 | if (root != dest && | |
9297 | (old_ino != BTRFS_FIRST_FREE_OBJECTID || | |
9298 | new_ino != BTRFS_FIRST_FREE_OBJECTID)) | |
cdd1fedf DF |
9299 | return -EXDEV; |
9300 | ||
9301 | /* close the race window with snapshot create/destroy ioctl */ | |
943eb3bf JB |
9302 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID || |
9303 | new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9304 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9305 | |
9306 | /* | |
9307 | * We want to reserve the absolute worst case amount of items. So if | |
9308 | * both inodes are subvols and we need to unlink them then that would | |
9309 | * require 4 item modifications, but if they are both normal inodes it | |
9310 | * would require 5 item modifications, so we'll assume their normal | |
9311 | * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items | |
9312 | * should cover the worst case number of items we'll modify. | |
9313 | */ | |
9314 | trans = btrfs_start_transaction(root, 12); | |
9315 | if (IS_ERR(trans)) { | |
9316 | ret = PTR_ERR(trans); | |
9317 | goto out_notrans; | |
9318 | } | |
9319 | ||
00aa8e87 JB |
9320 | if (dest != root) { |
9321 | ret = btrfs_record_root_in_trans(trans, dest); | |
9322 | if (ret) | |
9323 | goto out_fail; | |
9324 | } | |
3e174099 | 9325 | |
cdd1fedf DF |
9326 | /* |
9327 | * We need to find a free sequence number both in the source and | |
9328 | * in the destination directory for the exchange. | |
9329 | */ | |
877574e2 | 9330 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx); |
cdd1fedf DF |
9331 | if (ret) |
9332 | goto out_fail; | |
877574e2 | 9333 | ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx); |
cdd1fedf DF |
9334 | if (ret) |
9335 | goto out_fail; | |
9336 | ||
9337 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
9338 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
9339 | ||
9340 | /* Reference for the source. */ | |
9341 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9342 | /* force full log commit if subvolume involved. */ | |
90787766 | 9343 | btrfs_set_log_full_commit(trans); |
cdd1fedf | 9344 | } else { |
376e5a57 FM |
9345 | btrfs_pin_log_trans(root); |
9346 | root_log_pinned = true; | |
cdd1fedf DF |
9347 | ret = btrfs_insert_inode_ref(trans, dest, |
9348 | new_dentry->d_name.name, | |
9349 | new_dentry->d_name.len, | |
9350 | old_ino, | |
f85b7379 DS |
9351 | btrfs_ino(BTRFS_I(new_dir)), |
9352 | old_idx); | |
cdd1fedf DF |
9353 | if (ret) |
9354 | goto out_fail; | |
dc09ef35 | 9355 | need_abort = true; |
cdd1fedf DF |
9356 | } |
9357 | ||
9358 | /* And now for the dest. */ | |
9359 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9360 | /* force full log commit if subvolume involved. */ | |
90787766 | 9361 | btrfs_set_log_full_commit(trans); |
cdd1fedf | 9362 | } else { |
376e5a57 FM |
9363 | btrfs_pin_log_trans(dest); |
9364 | dest_log_pinned = true; | |
cdd1fedf DF |
9365 | ret = btrfs_insert_inode_ref(trans, root, |
9366 | old_dentry->d_name.name, | |
9367 | old_dentry->d_name.len, | |
9368 | new_ino, | |
f85b7379 DS |
9369 | btrfs_ino(BTRFS_I(old_dir)), |
9370 | new_idx); | |
dc09ef35 JB |
9371 | if (ret) { |
9372 | if (need_abort) | |
9373 | btrfs_abort_transaction(trans, ret); | |
cdd1fedf | 9374 | goto out_fail; |
dc09ef35 | 9375 | } |
cdd1fedf DF |
9376 | } |
9377 | ||
9378 | /* Update inode version and ctime/mtime. */ | |
9379 | inode_inc_iversion(old_dir); | |
9380 | inode_inc_iversion(new_dir); | |
9381 | inode_inc_iversion(old_inode); | |
9382 | inode_inc_iversion(new_inode); | |
9383 | old_dir->i_ctime = old_dir->i_mtime = ctime; | |
9384 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
9385 | old_inode->i_ctime = ctime; | |
9386 | new_inode->i_ctime = ctime; | |
9387 | ||
9388 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 DS |
9389 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9390 | BTRFS_I(old_inode), 1); | |
9391 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), | |
9392 | BTRFS_I(new_inode), 1); | |
cdd1fedf DF |
9393 | } |
9394 | ||
9395 | /* src is a subvolume */ | |
9396 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
045d3967 | 9397 | ret = btrfs_unlink_subvol(trans, old_dir, old_dentry); |
cdd1fedf | 9398 | } else { /* src is an inode */ |
4ec5934e NB |
9399 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9400 | BTRFS_I(old_dentry->d_inode), | |
cdd1fedf DF |
9401 | old_dentry->d_name.name, |
9402 | old_dentry->d_name.len); | |
9403 | if (!ret) | |
9a56fcd1 | 9404 | ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode)); |
cdd1fedf DF |
9405 | } |
9406 | if (ret) { | |
66642832 | 9407 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9408 | goto out_fail; |
9409 | } | |
9410 | ||
9411 | /* dest is a subvolume */ | |
9412 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
045d3967 | 9413 | ret = btrfs_unlink_subvol(trans, new_dir, new_dentry); |
cdd1fedf | 9414 | } else { /* dest is an inode */ |
4ec5934e NB |
9415 | ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9416 | BTRFS_I(new_dentry->d_inode), | |
cdd1fedf DF |
9417 | new_dentry->d_name.name, |
9418 | new_dentry->d_name.len); | |
9419 | if (!ret) | |
9a56fcd1 | 9420 | ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode)); |
cdd1fedf DF |
9421 | } |
9422 | if (ret) { | |
66642832 | 9423 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9424 | goto out_fail; |
9425 | } | |
9426 | ||
db0a669f | 9427 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
cdd1fedf DF |
9428 | new_dentry->d_name.name, |
9429 | new_dentry->d_name.len, 0, old_idx); | |
9430 | if (ret) { | |
66642832 | 9431 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9432 | goto out_fail; |
9433 | } | |
9434 | ||
db0a669f | 9435 | ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode), |
cdd1fedf DF |
9436 | old_dentry->d_name.name, |
9437 | old_dentry->d_name.len, 0, new_idx); | |
9438 | if (ret) { | |
66642832 | 9439 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9440 | goto out_fail; |
9441 | } | |
9442 | ||
9443 | if (old_inode->i_nlink == 1) | |
9444 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9445 | if (new_inode->i_nlink == 1) | |
9446 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9447 | ||
86e8aa0e | 9448 | if (root_log_pinned) { |
75b463d2 FM |
9449 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
9450 | new_dentry->d_parent); | |
cdd1fedf | 9451 | btrfs_end_log_trans(root); |
86e8aa0e | 9452 | root_log_pinned = false; |
cdd1fedf | 9453 | } |
86e8aa0e | 9454 | if (dest_log_pinned) { |
75b463d2 FM |
9455 | btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir), |
9456 | old_dentry->d_parent); | |
cdd1fedf | 9457 | btrfs_end_log_trans(dest); |
86e8aa0e | 9458 | dest_log_pinned = false; |
cdd1fedf DF |
9459 | } |
9460 | out_fail: | |
86e8aa0e FM |
9461 | /* |
9462 | * If we have pinned a log and an error happened, we unpin tasks | |
9463 | * trying to sync the log and force them to fallback to a transaction | |
9464 | * commit if the log currently contains any of the inodes involved in | |
9465 | * this rename operation (to ensure we do not persist a log with an | |
9466 | * inconsistent state for any of these inodes or leading to any | |
9467 | * inconsistencies when replayed). If the transaction was aborted, the | |
9468 | * abortion reason is propagated to userspace when attempting to commit | |
9469 | * the transaction. If the log does not contain any of these inodes, we | |
9470 | * allow the tasks to sync it. | |
9471 | */ | |
9472 | if (ret && (root_log_pinned || dest_log_pinned)) { | |
0f8939b8 NB |
9473 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9474 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9475 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
86e8aa0e | 9476 | (new_inode && |
0f8939b8 | 9477 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
90787766 | 9478 | btrfs_set_log_full_commit(trans); |
86e8aa0e FM |
9479 | |
9480 | if (root_log_pinned) { | |
9481 | btrfs_end_log_trans(root); | |
9482 | root_log_pinned = false; | |
9483 | } | |
9484 | if (dest_log_pinned) { | |
9485 | btrfs_end_log_trans(dest); | |
9486 | dest_log_pinned = false; | |
9487 | } | |
9488 | } | |
75b463d2 FM |
9489 | ret2 = btrfs_end_transaction(trans); |
9490 | ret = ret ? ret : ret2; | |
cdd1fedf | 9491 | out_notrans: |
943eb3bf JB |
9492 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID || |
9493 | old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9494 | up_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9495 | |
9496 | return ret; | |
9497 | } | |
9498 | ||
9499 | static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans, | |
9500 | struct btrfs_root *root, | |
9501 | struct inode *dir, | |
9502 | struct dentry *dentry) | |
9503 | { | |
9504 | int ret; | |
9505 | struct inode *inode; | |
9506 | u64 objectid; | |
9507 | u64 index; | |
9508 | ||
543068a2 | 9509 | ret = btrfs_get_free_objectid(root, &objectid); |
cdd1fedf DF |
9510 | if (ret) |
9511 | return ret; | |
9512 | ||
9513 | inode = btrfs_new_inode(trans, root, dir, | |
9514 | dentry->d_name.name, | |
9515 | dentry->d_name.len, | |
4a0cc7ca | 9516 | btrfs_ino(BTRFS_I(dir)), |
cdd1fedf DF |
9517 | objectid, |
9518 | S_IFCHR | WHITEOUT_MODE, | |
9519 | &index); | |
9520 | ||
9521 | if (IS_ERR(inode)) { | |
9522 | ret = PTR_ERR(inode); | |
9523 | return ret; | |
9524 | } | |
9525 | ||
9526 | inode->i_op = &btrfs_special_inode_operations; | |
9527 | init_special_inode(inode, inode->i_mode, | |
9528 | WHITEOUT_DEV); | |
9529 | ||
9530 | ret = btrfs_init_inode_security(trans, inode, dir, | |
9531 | &dentry->d_name); | |
9532 | if (ret) | |
c9901618 | 9533 | goto out; |
cdd1fedf | 9534 | |
cef415af NB |
9535 | ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
9536 | BTRFS_I(inode), 0, index); | |
cdd1fedf | 9537 | if (ret) |
c9901618 | 9538 | goto out; |
cdd1fedf | 9539 | |
9a56fcd1 | 9540 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
c9901618 | 9541 | out: |
cdd1fedf | 9542 | unlock_new_inode(inode); |
c9901618 FM |
9543 | if (ret) |
9544 | inode_dec_link_count(inode); | |
cdd1fedf DF |
9545 | iput(inode); |
9546 | ||
c9901618 | 9547 | return ret; |
cdd1fedf DF |
9548 | } |
9549 | ||
d397712b | 9550 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
cdd1fedf DF |
9551 | struct inode *new_dir, struct dentry *new_dentry, |
9552 | unsigned int flags) | |
39279cc3 | 9553 | { |
0b246afa | 9554 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
39279cc3 | 9555 | struct btrfs_trans_handle *trans; |
5062af35 | 9556 | unsigned int trans_num_items; |
39279cc3 | 9557 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9558 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9559 | struct inode *new_inode = d_inode(new_dentry); |
9560 | struct inode *old_inode = d_inode(old_dentry); | |
00e4e6b3 | 9561 | u64 index = 0; |
39279cc3 | 9562 | int ret; |
75b463d2 | 9563 | int ret2; |
4a0cc7ca | 9564 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
3dc9e8f7 | 9565 | bool log_pinned = false; |
39279cc3 | 9566 | |
4a0cc7ca | 9567 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9568 | return -EPERM; |
9569 | ||
4df27c4d | 9570 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9571 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9572 | return -EXDEV; |
9573 | ||
33345d01 | 9574 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 9575 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 9576 | return -ENOTEMPTY; |
5f39d397 | 9577 | |
4df27c4d YZ |
9578 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9579 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9580 | return -ENOTEMPTY; | |
9c52057c CM |
9581 | |
9582 | ||
9583 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9584 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9585 | new_dentry->d_name.name, |
9586 | new_dentry->d_name.len); | |
9587 | ||
9588 | if (ret) { | |
9589 | if (ret == -EEXIST) { | |
9590 | /* we shouldn't get | |
9591 | * eexist without a new_inode */ | |
fae7f21c | 9592 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9593 | return ret; |
9594 | } | |
9595 | } else { | |
9596 | /* maybe -EOVERFLOW */ | |
9597 | return ret; | |
9598 | } | |
9599 | } | |
9600 | ret = 0; | |
9601 | ||
5a3f23d5 | 9602 | /* |
8d875f95 CM |
9603 | * we're using rename to replace one file with another. Start IO on it |
9604 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9605 | */ |
8d875f95 | 9606 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9607 | filemap_flush(old_inode->i_mapping); |
9608 | ||
76dda93c | 9609 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 9610 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9611 | down_read(&fs_info->subvol_sem); |
a22285a6 YZ |
9612 | /* |
9613 | * We want to reserve the absolute worst case amount of items. So if | |
9614 | * both inodes are subvols and we need to unlink them then that would | |
9615 | * require 4 item modifications, but if they are both normal inodes it | |
cdd1fedf | 9616 | * would require 5 item modifications, so we'll assume they are normal |
a22285a6 YZ |
9617 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items |
9618 | * should cover the worst case number of items we'll modify. | |
5062af35 FM |
9619 | * If our rename has the whiteout flag, we need more 5 units for the |
9620 | * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item | |
9621 | * when selinux is enabled). | |
a22285a6 | 9622 | */ |
5062af35 FM |
9623 | trans_num_items = 11; |
9624 | if (flags & RENAME_WHITEOUT) | |
9625 | trans_num_items += 5; | |
9626 | trans = btrfs_start_transaction(root, trans_num_items); | |
b44c59a8 | 9627 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9628 | ret = PTR_ERR(trans); |
9629 | goto out_notrans; | |
9630 | } | |
76dda93c | 9631 | |
b0fec6fd JB |
9632 | if (dest != root) { |
9633 | ret = btrfs_record_root_in_trans(trans, dest); | |
9634 | if (ret) | |
9635 | goto out_fail; | |
9636 | } | |
5f39d397 | 9637 | |
877574e2 | 9638 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index); |
a5719521 YZ |
9639 | if (ret) |
9640 | goto out_fail; | |
5a3f23d5 | 9641 | |
67de1176 | 9642 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9643 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9644 | /* force full log commit if subvolume involved. */ |
90787766 | 9645 | btrfs_set_log_full_commit(trans); |
4df27c4d | 9646 | } else { |
c4aba954 FM |
9647 | btrfs_pin_log_trans(root); |
9648 | log_pinned = true; | |
a5719521 YZ |
9649 | ret = btrfs_insert_inode_ref(trans, dest, |
9650 | new_dentry->d_name.name, | |
9651 | new_dentry->d_name.len, | |
33345d01 | 9652 | old_ino, |
4a0cc7ca | 9653 | btrfs_ino(BTRFS_I(new_dir)), index); |
a5719521 YZ |
9654 | if (ret) |
9655 | goto out_fail; | |
4df27c4d | 9656 | } |
5a3f23d5 | 9657 | |
0c4d2d95 JB |
9658 | inode_inc_iversion(old_dir); |
9659 | inode_inc_iversion(new_dir); | |
9660 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
9661 | old_dir->i_ctime = old_dir->i_mtime = |
9662 | new_dir->i_ctime = new_dir->i_mtime = | |
c2050a45 | 9663 | old_inode->i_ctime = current_time(old_dir); |
5f39d397 | 9664 | |
12fcfd22 | 9665 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 DS |
9666 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9667 | BTRFS_I(old_inode), 1); | |
12fcfd22 | 9668 | |
33345d01 | 9669 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
045d3967 | 9670 | ret = btrfs_unlink_subvol(trans, old_dir, old_dentry); |
4df27c4d | 9671 | } else { |
4ec5934e NB |
9672 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9673 | BTRFS_I(d_inode(old_dentry)), | |
92986796 AV |
9674 | old_dentry->d_name.name, |
9675 | old_dentry->d_name.len); | |
9676 | if (!ret) | |
9a56fcd1 | 9677 | ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode)); |
4df27c4d | 9678 | } |
79787eaa | 9679 | if (ret) { |
66642832 | 9680 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9681 | goto out_fail; |
9682 | } | |
39279cc3 CM |
9683 | |
9684 | if (new_inode) { | |
0c4d2d95 | 9685 | inode_inc_iversion(new_inode); |
c2050a45 | 9686 | new_inode->i_ctime = current_time(new_inode); |
4a0cc7ca | 9687 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d | 9688 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
045d3967 | 9689 | ret = btrfs_unlink_subvol(trans, new_dir, new_dentry); |
4df27c4d YZ |
9690 | BUG_ON(new_inode->i_nlink == 0); |
9691 | } else { | |
4ec5934e NB |
9692 | ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9693 | BTRFS_I(d_inode(new_dentry)), | |
4df27c4d YZ |
9694 | new_dentry->d_name.name, |
9695 | new_dentry->d_name.len); | |
9696 | } | |
4ef31a45 | 9697 | if (!ret && new_inode->i_nlink == 0) |
73f2e545 NB |
9698 | ret = btrfs_orphan_add(trans, |
9699 | BTRFS_I(d_inode(new_dentry))); | |
79787eaa | 9700 | if (ret) { |
66642832 | 9701 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9702 | goto out_fail; |
9703 | } | |
39279cc3 | 9704 | } |
aec7477b | 9705 | |
db0a669f | 9706 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
4df27c4d | 9707 | new_dentry->d_name.name, |
a5719521 | 9708 | new_dentry->d_name.len, 0, index); |
79787eaa | 9709 | if (ret) { |
66642832 | 9710 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9711 | goto out_fail; |
9712 | } | |
39279cc3 | 9713 | |
67de1176 MX |
9714 | if (old_inode->i_nlink == 1) |
9715 | BTRFS_I(old_inode)->dir_index = index; | |
9716 | ||
3dc9e8f7 | 9717 | if (log_pinned) { |
75b463d2 FM |
9718 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
9719 | new_dentry->d_parent); | |
4df27c4d | 9720 | btrfs_end_log_trans(root); |
3dc9e8f7 | 9721 | log_pinned = false; |
4df27c4d | 9722 | } |
cdd1fedf DF |
9723 | |
9724 | if (flags & RENAME_WHITEOUT) { | |
9725 | ret = btrfs_whiteout_for_rename(trans, root, old_dir, | |
9726 | old_dentry); | |
9727 | ||
9728 | if (ret) { | |
66642832 | 9729 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9730 | goto out_fail; |
9731 | } | |
4df27c4d | 9732 | } |
39279cc3 | 9733 | out_fail: |
3dc9e8f7 FM |
9734 | /* |
9735 | * If we have pinned the log and an error happened, we unpin tasks | |
9736 | * trying to sync the log and force them to fallback to a transaction | |
9737 | * commit if the log currently contains any of the inodes involved in | |
9738 | * this rename operation (to ensure we do not persist a log with an | |
9739 | * inconsistent state for any of these inodes or leading to any | |
9740 | * inconsistencies when replayed). If the transaction was aborted, the | |
9741 | * abortion reason is propagated to userspace when attempting to commit | |
9742 | * the transaction. If the log does not contain any of these inodes, we | |
9743 | * allow the tasks to sync it. | |
9744 | */ | |
9745 | if (ret && log_pinned) { | |
0f8939b8 NB |
9746 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9747 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9748 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
3dc9e8f7 | 9749 | (new_inode && |
0f8939b8 | 9750 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
90787766 | 9751 | btrfs_set_log_full_commit(trans); |
3dc9e8f7 FM |
9752 | |
9753 | btrfs_end_log_trans(root); | |
9754 | log_pinned = false; | |
9755 | } | |
75b463d2 FM |
9756 | ret2 = btrfs_end_transaction(trans); |
9757 | ret = ret ? ret : ret2; | |
b44c59a8 | 9758 | out_notrans: |
33345d01 | 9759 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9760 | up_read(&fs_info->subvol_sem); |
9ed74f2d | 9761 | |
39279cc3 CM |
9762 | return ret; |
9763 | } | |
9764 | ||
549c7297 CB |
9765 | static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir, |
9766 | struct dentry *old_dentry, struct inode *new_dir, | |
9767 | struct dentry *new_dentry, unsigned int flags) | |
80ace85c | 9768 | { |
cdd1fedf | 9769 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
9770 | return -EINVAL; |
9771 | ||
cdd1fedf DF |
9772 | if (flags & RENAME_EXCHANGE) |
9773 | return btrfs_rename_exchange(old_dir, old_dentry, new_dir, | |
9774 | new_dentry); | |
9775 | ||
9776 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags); | |
80ace85c MS |
9777 | } |
9778 | ||
3a2f8c07 NB |
9779 | struct btrfs_delalloc_work { |
9780 | struct inode *inode; | |
9781 | struct completion completion; | |
9782 | struct list_head list; | |
9783 | struct btrfs_work work; | |
9784 | }; | |
9785 | ||
8ccf6f19 MX |
9786 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
9787 | { | |
9788 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 9789 | struct inode *inode; |
8ccf6f19 MX |
9790 | |
9791 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
9792 | work); | |
9f23e289 | 9793 | inode = delalloc_work->inode; |
30424601 DS |
9794 | filemap_flush(inode->i_mapping); |
9795 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9796 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 9797 | filemap_flush(inode->i_mapping); |
8ccf6f19 | 9798 | |
076da91c | 9799 | iput(inode); |
8ccf6f19 MX |
9800 | complete(&delalloc_work->completion); |
9801 | } | |
9802 | ||
3a2f8c07 | 9803 | static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode) |
8ccf6f19 MX |
9804 | { |
9805 | struct btrfs_delalloc_work *work; | |
9806 | ||
100d5702 | 9807 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
9808 | if (!work) |
9809 | return NULL; | |
9810 | ||
9811 | init_completion(&work->completion); | |
9812 | INIT_LIST_HEAD(&work->list); | |
9813 | work->inode = inode; | |
a0cac0ec | 9814 | btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL); |
8ccf6f19 MX |
9815 | |
9816 | return work; | |
9817 | } | |
9818 | ||
d352ac68 CM |
9819 | /* |
9820 | * some fairly slow code that needs optimization. This walks the list | |
9821 | * of all the inodes with pending delalloc and forces them to disk. | |
9822 | */ | |
e076ab2a JB |
9823 | static int start_delalloc_inodes(struct btrfs_root *root, |
9824 | struct writeback_control *wbc, bool snapshot, | |
3d45f221 | 9825 | bool in_reclaim_context) |
ea8c2819 | 9826 | { |
ea8c2819 | 9827 | struct btrfs_inode *binode; |
5b21f2ed | 9828 | struct inode *inode; |
8ccf6f19 MX |
9829 | struct btrfs_delalloc_work *work, *next; |
9830 | struct list_head works; | |
1eafa6c7 | 9831 | struct list_head splice; |
8ccf6f19 | 9832 | int ret = 0; |
e076ab2a | 9833 | bool full_flush = wbc->nr_to_write == LONG_MAX; |
ea8c2819 | 9834 | |
8ccf6f19 | 9835 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 9836 | INIT_LIST_HEAD(&splice); |
63607cc8 | 9837 | |
573bfb72 | 9838 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
9839 | spin_lock(&root->delalloc_lock); |
9840 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
9841 | while (!list_empty(&splice)) { |
9842 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 9843 | delalloc_inodes); |
1eafa6c7 | 9844 | |
eb73c1b7 MX |
9845 | list_move_tail(&binode->delalloc_inodes, |
9846 | &root->delalloc_inodes); | |
3d45f221 FM |
9847 | |
9848 | if (in_reclaim_context && | |
9849 | test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags)) | |
9850 | continue; | |
9851 | ||
5b21f2ed | 9852 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 9853 | if (!inode) { |
eb73c1b7 | 9854 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 9855 | continue; |
df0af1a5 | 9856 | } |
eb73c1b7 | 9857 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 9858 | |
3cd24c69 EL |
9859 | if (snapshot) |
9860 | set_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
9861 | &binode->runtime_flags); | |
e076ab2a JB |
9862 | if (full_flush) { |
9863 | work = btrfs_alloc_delalloc_work(inode); | |
9864 | if (!work) { | |
9865 | iput(inode); | |
9866 | ret = -ENOMEM; | |
9867 | goto out; | |
9868 | } | |
9869 | list_add_tail(&work->list, &works); | |
9870 | btrfs_queue_work(root->fs_info->flush_workers, | |
9871 | &work->work); | |
9872 | } else { | |
9873 | ret = sync_inode(inode, wbc); | |
9874 | if (!ret && | |
9875 | test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9876 | &BTRFS_I(inode)->runtime_flags)) | |
9877 | ret = sync_inode(inode, wbc); | |
9878 | btrfs_add_delayed_iput(inode); | |
9879 | if (ret || wbc->nr_to_write <= 0) | |
b4912139 JB |
9880 | goto out; |
9881 | } | |
5b21f2ed | 9882 | cond_resched(); |
eb73c1b7 | 9883 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 9884 | } |
eb73c1b7 | 9885 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 9886 | |
a1ecaabb | 9887 | out: |
eb73c1b7 MX |
9888 | list_for_each_entry_safe(work, next, &works, list) { |
9889 | list_del_init(&work->list); | |
40012f96 NB |
9890 | wait_for_completion(&work->completion); |
9891 | kfree(work); | |
eb73c1b7 MX |
9892 | } |
9893 | ||
81f1d390 | 9894 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
9895 | spin_lock(&root->delalloc_lock); |
9896 | list_splice_tail(&splice, &root->delalloc_inodes); | |
9897 | spin_unlock(&root->delalloc_lock); | |
9898 | } | |
573bfb72 | 9899 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
9900 | return ret; |
9901 | } | |
1eafa6c7 | 9902 | |
f9baa501 | 9903 | int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context) |
eb73c1b7 | 9904 | { |
e076ab2a JB |
9905 | struct writeback_control wbc = { |
9906 | .nr_to_write = LONG_MAX, | |
9907 | .sync_mode = WB_SYNC_NONE, | |
9908 | .range_start = 0, | |
9909 | .range_end = LLONG_MAX, | |
9910 | }; | |
0b246afa | 9911 | struct btrfs_fs_info *fs_info = root->fs_info; |
1eafa6c7 | 9912 | |
0b246afa | 9913 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
9914 | return -EROFS; |
9915 | ||
f9baa501 | 9916 | return start_delalloc_inodes(root, &wbc, true, in_reclaim_context); |
eb73c1b7 MX |
9917 | } |
9918 | ||
9db4dc24 | 9919 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr, |
3d45f221 | 9920 | bool in_reclaim_context) |
eb73c1b7 | 9921 | { |
e076ab2a | 9922 | struct writeback_control wbc = { |
9db4dc24 | 9923 | .nr_to_write = nr, |
e076ab2a JB |
9924 | .sync_mode = WB_SYNC_NONE, |
9925 | .range_start = 0, | |
9926 | .range_end = LLONG_MAX, | |
9927 | }; | |
eb73c1b7 MX |
9928 | struct btrfs_root *root; |
9929 | struct list_head splice; | |
9930 | int ret; | |
9931 | ||
2c21b4d7 | 9932 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
9933 | return -EROFS; |
9934 | ||
9935 | INIT_LIST_HEAD(&splice); | |
9936 | ||
573bfb72 | 9937 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
9938 | spin_lock(&fs_info->delalloc_root_lock); |
9939 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
d7830b71 | 9940 | while (!list_empty(&splice)) { |
e076ab2a JB |
9941 | /* |
9942 | * Reset nr_to_write here so we know that we're doing a full | |
9943 | * flush. | |
9944 | */ | |
9db4dc24 | 9945 | if (nr == LONG_MAX) |
e076ab2a JB |
9946 | wbc.nr_to_write = LONG_MAX; |
9947 | ||
eb73c1b7 MX |
9948 | root = list_first_entry(&splice, struct btrfs_root, |
9949 | delalloc_root); | |
00246528 | 9950 | root = btrfs_grab_root(root); |
eb73c1b7 MX |
9951 | BUG_ON(!root); |
9952 | list_move_tail(&root->delalloc_root, | |
9953 | &fs_info->delalloc_roots); | |
9954 | spin_unlock(&fs_info->delalloc_root_lock); | |
9955 | ||
e076ab2a | 9956 | ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context); |
00246528 | 9957 | btrfs_put_root(root); |
e076ab2a | 9958 | if (ret < 0 || wbc.nr_to_write <= 0) |
eb73c1b7 | 9959 | goto out; |
eb73c1b7 | 9960 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 9961 | } |
eb73c1b7 | 9962 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 9963 | |
6c255e67 | 9964 | ret = 0; |
eb73c1b7 | 9965 | out: |
81f1d390 | 9966 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
9967 | spin_lock(&fs_info->delalloc_root_lock); |
9968 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
9969 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 9970 | } |
573bfb72 | 9971 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 9972 | return ret; |
ea8c2819 CM |
9973 | } |
9974 | ||
549c7297 CB |
9975 | static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, |
9976 | struct dentry *dentry, const char *symname) | |
39279cc3 | 9977 | { |
0b246afa | 9978 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
9979 | struct btrfs_trans_handle *trans; |
9980 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9981 | struct btrfs_path *path; | |
9982 | struct btrfs_key key; | |
1832a6d5 | 9983 | struct inode *inode = NULL; |
39279cc3 | 9984 | int err; |
39279cc3 | 9985 | u64 objectid; |
67871254 | 9986 | u64 index = 0; |
39279cc3 CM |
9987 | int name_len; |
9988 | int datasize; | |
5f39d397 | 9989 | unsigned long ptr; |
39279cc3 | 9990 | struct btrfs_file_extent_item *ei; |
5f39d397 | 9991 | struct extent_buffer *leaf; |
39279cc3 | 9992 | |
f06becc4 | 9993 | name_len = strlen(symname); |
0b246afa | 9994 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 9995 | return -ENAMETOOLONG; |
1832a6d5 | 9996 | |
9ed74f2d JB |
9997 | /* |
9998 | * 2 items for inode item and ref | |
9999 | * 2 items for dir items | |
9269d12b FM |
10000 | * 1 item for updating parent inode item |
10001 | * 1 item for the inline extent item | |
9ed74f2d JB |
10002 | * 1 item for xattr if selinux is on |
10003 | */ | |
9269d12b | 10004 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
10005 | if (IS_ERR(trans)) |
10006 | return PTR_ERR(trans); | |
1832a6d5 | 10007 | |
543068a2 | 10008 | err = btrfs_get_free_objectid(root, &objectid); |
581bb050 LZ |
10009 | if (err) |
10010 | goto out_unlock; | |
10011 | ||
aec7477b | 10012 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
10013 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), |
10014 | objectid, S_IFLNK|S_IRWXUGO, &index); | |
7cf96da3 TI |
10015 | if (IS_ERR(inode)) { |
10016 | err = PTR_ERR(inode); | |
32955c54 | 10017 | inode = NULL; |
39279cc3 | 10018 | goto out_unlock; |
7cf96da3 | 10019 | } |
39279cc3 | 10020 | |
ad19db71 CS |
10021 | /* |
10022 | * If the active LSM wants to access the inode during | |
10023 | * d_instantiate it needs these. Smack checks to see | |
10024 | * if the filesystem supports xattrs by looking at the | |
10025 | * ops vector. | |
10026 | */ | |
10027 | inode->i_fop = &btrfs_file_operations; | |
10028 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 10029 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
10030 | |
10031 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
10032 | if (err) | |
32955c54 | 10033 | goto out_unlock; |
ad19db71 | 10034 | |
39279cc3 | 10035 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10036 | if (!path) { |
10037 | err = -ENOMEM; | |
32955c54 | 10038 | goto out_unlock; |
d8926bb3 | 10039 | } |
4a0cc7ca | 10040 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 10041 | key.offset = 0; |
962a298f | 10042 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
10043 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
10044 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
10045 | datasize); | |
54aa1f4d | 10046 | if (err) { |
b0839166 | 10047 | btrfs_free_path(path); |
32955c54 | 10048 | goto out_unlock; |
54aa1f4d | 10049 | } |
5f39d397 CM |
10050 | leaf = path->nodes[0]; |
10051 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
10052 | struct btrfs_file_extent_item); | |
10053 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
10054 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 10055 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
10056 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
10057 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
10058 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
10059 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
10060 | ||
39279cc3 | 10061 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
10062 | write_extent_buffer(leaf, symname, ptr, name_len); |
10063 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 10064 | btrfs_free_path(path); |
5f39d397 | 10065 | |
39279cc3 | 10066 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 10067 | inode_nohighmem(inode); |
d899e052 | 10068 | inode_set_bytes(inode, name_len); |
6ef06d27 | 10069 | btrfs_i_size_write(BTRFS_I(inode), name_len); |
9a56fcd1 | 10070 | err = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
d50866d0 FM |
10071 | /* |
10072 | * Last step, add directory indexes for our symlink inode. This is the | |
10073 | * last step to avoid extra cleanup of these indexes if an error happens | |
10074 | * elsewhere above. | |
10075 | */ | |
10076 | if (!err) | |
cef415af NB |
10077 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
10078 | BTRFS_I(inode), 0, index); | |
32955c54 AV |
10079 | if (err) |
10080 | goto out_unlock; | |
b0d5d10f | 10081 | |
1e2e547a | 10082 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
10083 | |
10084 | out_unlock: | |
3a45bb20 | 10085 | btrfs_end_transaction(trans); |
32955c54 | 10086 | if (err && inode) { |
39279cc3 | 10087 | inode_dec_link_count(inode); |
32955c54 | 10088 | discard_new_inode(inode); |
39279cc3 | 10089 | } |
2ff7e61e | 10090 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
10091 | return err; |
10092 | } | |
16432985 | 10093 | |
8fccebfa FM |
10094 | static struct btrfs_trans_handle *insert_prealloc_file_extent( |
10095 | struct btrfs_trans_handle *trans_in, | |
90dffd0c NB |
10096 | struct btrfs_inode *inode, |
10097 | struct btrfs_key *ins, | |
203f44c5 QW |
10098 | u64 file_offset) |
10099 | { | |
10100 | struct btrfs_file_extent_item stack_fi; | |
bf385648 | 10101 | struct btrfs_replace_extent_info extent_info; |
8fccebfa FM |
10102 | struct btrfs_trans_handle *trans = trans_in; |
10103 | struct btrfs_path *path; | |
203f44c5 QW |
10104 | u64 start = ins->objectid; |
10105 | u64 len = ins->offset; | |
fbf48bb0 | 10106 | int qgroup_released; |
9729f10a | 10107 | int ret; |
203f44c5 QW |
10108 | |
10109 | memset(&stack_fi, 0, sizeof(stack_fi)); | |
10110 | ||
10111 | btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC); | |
10112 | btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start); | |
10113 | btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len); | |
10114 | btrfs_set_stack_file_extent_num_bytes(&stack_fi, len); | |
10115 | btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len); | |
10116 | btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE); | |
10117 | /* Encryption and other encoding is reserved and all 0 */ | |
10118 | ||
fbf48bb0 QW |
10119 | qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len); |
10120 | if (qgroup_released < 0) | |
10121 | return ERR_PTR(qgroup_released); | |
8fccebfa FM |
10122 | |
10123 | if (trans) { | |
90dffd0c | 10124 | ret = insert_reserved_file_extent(trans, inode, |
2766ff61 | 10125 | file_offset, &stack_fi, |
fbf48bb0 | 10126 | true, qgroup_released); |
8fccebfa | 10127 | if (ret) |
a3ee79bd | 10128 | goto free_qgroup; |
8fccebfa FM |
10129 | return trans; |
10130 | } | |
10131 | ||
10132 | extent_info.disk_offset = start; | |
10133 | extent_info.disk_len = len; | |
10134 | extent_info.data_offset = 0; | |
10135 | extent_info.data_len = len; | |
10136 | extent_info.file_offset = file_offset; | |
10137 | extent_info.extent_buf = (char *)&stack_fi; | |
8fccebfa | 10138 | extent_info.is_new_extent = true; |
fbf48bb0 | 10139 | extent_info.qgroup_reserved = qgroup_released; |
8fccebfa FM |
10140 | extent_info.insertions = 0; |
10141 | ||
10142 | path = btrfs_alloc_path(); | |
a3ee79bd QW |
10143 | if (!path) { |
10144 | ret = -ENOMEM; | |
10145 | goto free_qgroup; | |
10146 | } | |
8fccebfa | 10147 | |
bfc78479 | 10148 | ret = btrfs_replace_file_extents(inode, path, file_offset, |
8fccebfa FM |
10149 | file_offset + len - 1, &extent_info, |
10150 | &trans); | |
10151 | btrfs_free_path(path); | |
10152 | if (ret) | |
a3ee79bd | 10153 | goto free_qgroup; |
8fccebfa | 10154 | return trans; |
a3ee79bd QW |
10155 | |
10156 | free_qgroup: | |
10157 | /* | |
10158 | * We have released qgroup data range at the beginning of the function, | |
10159 | * and normally qgroup_released bytes will be freed when committing | |
10160 | * transaction. | |
10161 | * But if we error out early, we have to free what we have released | |
10162 | * or we leak qgroup data reservation. | |
10163 | */ | |
10164 | btrfs_qgroup_free_refroot(inode->root->fs_info, | |
10165 | inode->root->root_key.objectid, qgroup_released, | |
10166 | BTRFS_QGROUP_RSV_DATA); | |
10167 | return ERR_PTR(ret); | |
203f44c5 | 10168 | } |
8fccebfa | 10169 | |
0af3d00b JB |
10170 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
10171 | u64 start, u64 num_bytes, u64 min_size, | |
10172 | loff_t actual_len, u64 *alloc_hint, | |
10173 | struct btrfs_trans_handle *trans) | |
d899e052 | 10174 | { |
0b246afa | 10175 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 JB |
10176 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
10177 | struct extent_map *em; | |
d899e052 YZ |
10178 | struct btrfs_root *root = BTRFS_I(inode)->root; |
10179 | struct btrfs_key ins; | |
d899e052 | 10180 | u64 cur_offset = start; |
b778cf96 | 10181 | u64 clear_offset = start; |
55a61d1d | 10182 | u64 i_size; |
154ea289 | 10183 | u64 cur_bytes; |
0b670dc4 | 10184 | u64 last_alloc = (u64)-1; |
d899e052 | 10185 | int ret = 0; |
0af3d00b | 10186 | bool own_trans = true; |
18513091 | 10187 | u64 end = start + num_bytes - 1; |
d899e052 | 10188 | |
0af3d00b JB |
10189 | if (trans) |
10190 | own_trans = false; | |
d899e052 | 10191 | while (num_bytes > 0) { |
ee22184b | 10192 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 10193 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
10194 | /* |
10195 | * If we are severely fragmented we could end up with really | |
10196 | * small allocations, so if the allocator is returning small | |
10197 | * chunks lets make its job easier by only searching for those | |
10198 | * sized chunks. | |
10199 | */ | |
10200 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
10201 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
10202 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
8fccebfa | 10203 | if (ret) |
a22285a6 | 10204 | break; |
b778cf96 JB |
10205 | |
10206 | /* | |
10207 | * We've reserved this space, and thus converted it from | |
10208 | * ->bytes_may_use to ->bytes_reserved. Any error that happens | |
10209 | * from here on out we will only need to clear our reservation | |
10210 | * for the remaining unreserved area, so advance our | |
10211 | * clear_offset by our extent size. | |
10212 | */ | |
10213 | clear_offset += ins.offset; | |
5a303d5d | 10214 | |
0b670dc4 | 10215 | last_alloc = ins.offset; |
90dffd0c NB |
10216 | trans = insert_prealloc_file_extent(trans, BTRFS_I(inode), |
10217 | &ins, cur_offset); | |
1afc708d FM |
10218 | /* |
10219 | * Now that we inserted the prealloc extent we can finally | |
10220 | * decrement the number of reservations in the block group. | |
10221 | * If we did it before, we could race with relocation and have | |
10222 | * relocation miss the reserved extent, making it fail later. | |
10223 | */ | |
10224 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); | |
8fccebfa FM |
10225 | if (IS_ERR(trans)) { |
10226 | ret = PTR_ERR(trans); | |
2ff7e61e | 10227 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 10228 | ins.offset, 0); |
79787eaa JM |
10229 | break; |
10230 | } | |
31193213 | 10231 | |
dcdbc059 | 10232 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
a1ed835e | 10233 | cur_offset + ins.offset -1, 0); |
5a303d5d | 10234 | |
5dc562c5 JB |
10235 | em = alloc_extent_map(); |
10236 | if (!em) { | |
10237 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
10238 | &BTRFS_I(inode)->runtime_flags); | |
10239 | goto next; | |
10240 | } | |
10241 | ||
10242 | em->start = cur_offset; | |
10243 | em->orig_start = cur_offset; | |
10244 | em->len = ins.offset; | |
10245 | em->block_start = ins.objectid; | |
10246 | em->block_len = ins.offset; | |
b4939680 | 10247 | em->orig_block_len = ins.offset; |
cc95bef6 | 10248 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
10249 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
10250 | em->generation = trans->transid; | |
10251 | ||
10252 | while (1) { | |
10253 | write_lock(&em_tree->lock); | |
09a2a8f9 | 10254 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
10255 | write_unlock(&em_tree->lock); |
10256 | if (ret != -EEXIST) | |
10257 | break; | |
dcdbc059 | 10258 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
10259 | cur_offset + ins.offset - 1, |
10260 | 0); | |
10261 | } | |
10262 | free_extent_map(em); | |
10263 | next: | |
d899e052 YZ |
10264 | num_bytes -= ins.offset; |
10265 | cur_offset += ins.offset; | |
efa56464 | 10266 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 10267 | |
0c4d2d95 | 10268 | inode_inc_iversion(inode); |
c2050a45 | 10269 | inode->i_ctime = current_time(inode); |
6cbff00f | 10270 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 10271 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
10272 | (actual_len > inode->i_size) && |
10273 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 10274 | if (cur_offset > actual_len) |
55a61d1d | 10275 | i_size = actual_len; |
d1ea6a61 | 10276 | else |
55a61d1d JB |
10277 | i_size = cur_offset; |
10278 | i_size_write(inode, i_size); | |
76aea537 | 10279 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
5a303d5d YZ |
10280 | } |
10281 | ||
9a56fcd1 | 10282 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
79787eaa JM |
10283 | |
10284 | if (ret) { | |
66642832 | 10285 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10286 | if (own_trans) |
3a45bb20 | 10287 | btrfs_end_transaction(trans); |
79787eaa JM |
10288 | break; |
10289 | } | |
d899e052 | 10290 | |
8fccebfa | 10291 | if (own_trans) { |
3a45bb20 | 10292 | btrfs_end_transaction(trans); |
8fccebfa FM |
10293 | trans = NULL; |
10294 | } | |
5a303d5d | 10295 | } |
b778cf96 | 10296 | if (clear_offset < end) |
25ce28ca | 10297 | btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset, |
b778cf96 | 10298 | end - clear_offset + 1); |
d899e052 YZ |
10299 | return ret; |
10300 | } | |
10301 | ||
0af3d00b JB |
10302 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
10303 | u64 start, u64 num_bytes, u64 min_size, | |
10304 | loff_t actual_len, u64 *alloc_hint) | |
10305 | { | |
10306 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10307 | min_size, actual_len, alloc_hint, | |
10308 | NULL); | |
10309 | } | |
10310 | ||
10311 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
10312 | struct btrfs_trans_handle *trans, int mode, | |
10313 | u64 start, u64 num_bytes, u64 min_size, | |
10314 | loff_t actual_len, u64 *alloc_hint) | |
10315 | { | |
10316 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10317 | min_size, actual_len, alloc_hint, trans); | |
10318 | } | |
10319 | ||
e6dcd2dc CM |
10320 | static int btrfs_set_page_dirty(struct page *page) |
10321 | { | |
e6dcd2dc CM |
10322 | return __set_page_dirty_nobuffers(page); |
10323 | } | |
10324 | ||
549c7297 CB |
10325 | static int btrfs_permission(struct user_namespace *mnt_userns, |
10326 | struct inode *inode, int mask) | |
fdebe2bd | 10327 | { |
b83cc969 | 10328 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 10329 | umode_t mode = inode->i_mode; |
b83cc969 | 10330 | |
cb6db4e5 JM |
10331 | if (mask & MAY_WRITE && |
10332 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
10333 | if (btrfs_root_readonly(root)) | |
10334 | return -EROFS; | |
10335 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
10336 | return -EACCES; | |
10337 | } | |
47291baa | 10338 | return generic_permission(&init_user_ns, inode, mask); |
fdebe2bd | 10339 | } |
39279cc3 | 10340 | |
549c7297 CB |
10341 | static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir, |
10342 | struct dentry *dentry, umode_t mode) | |
ef3b9af5 | 10343 | { |
2ff7e61e | 10344 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
10345 | struct btrfs_trans_handle *trans; |
10346 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10347 | struct inode *inode = NULL; | |
10348 | u64 objectid; | |
10349 | u64 index; | |
10350 | int ret = 0; | |
10351 | ||
10352 | /* | |
10353 | * 5 units required for adding orphan entry | |
10354 | */ | |
10355 | trans = btrfs_start_transaction(root, 5); | |
10356 | if (IS_ERR(trans)) | |
10357 | return PTR_ERR(trans); | |
10358 | ||
543068a2 | 10359 | ret = btrfs_get_free_objectid(root, &objectid); |
ef3b9af5 FM |
10360 | if (ret) |
10361 | goto out; | |
10362 | ||
10363 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
f85b7379 | 10364 | btrfs_ino(BTRFS_I(dir)), objectid, mode, &index); |
ef3b9af5 FM |
10365 | if (IS_ERR(inode)) { |
10366 | ret = PTR_ERR(inode); | |
10367 | inode = NULL; | |
10368 | goto out; | |
10369 | } | |
10370 | ||
ef3b9af5 FM |
10371 | inode->i_fop = &btrfs_file_operations; |
10372 | inode->i_op = &btrfs_file_inode_operations; | |
10373 | ||
10374 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 | 10375 | |
b0d5d10f CM |
10376 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
10377 | if (ret) | |
32955c54 | 10378 | goto out; |
b0d5d10f | 10379 | |
9a56fcd1 | 10380 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
b0d5d10f | 10381 | if (ret) |
32955c54 | 10382 | goto out; |
73f2e545 | 10383 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
ef3b9af5 | 10384 | if (ret) |
32955c54 | 10385 | goto out; |
ef3b9af5 | 10386 | |
5762b5c9 FM |
10387 | /* |
10388 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
10389 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
10390 | * through: | |
10391 | * | |
10392 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
10393 | */ | |
10394 | set_nlink(inode, 1); | |
ef3b9af5 | 10395 | d_tmpfile(dentry, inode); |
32955c54 | 10396 | unlock_new_inode(inode); |
ef3b9af5 | 10397 | mark_inode_dirty(inode); |
ef3b9af5 | 10398 | out: |
3a45bb20 | 10399 | btrfs_end_transaction(trans); |
32955c54 AV |
10400 | if (ret && inode) |
10401 | discard_new_inode(inode); | |
2ff7e61e | 10402 | btrfs_btree_balance_dirty(fs_info); |
ef3b9af5 FM |
10403 | return ret; |
10404 | } | |
10405 | ||
d2a91064 | 10406 | void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end) |
c6100a4b | 10407 | { |
d2a91064 | 10408 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
c6100a4b JB |
10409 | unsigned long index = start >> PAGE_SHIFT; |
10410 | unsigned long end_index = end >> PAGE_SHIFT; | |
10411 | struct page *page; | |
d2a91064 | 10412 | u32 len; |
c6100a4b | 10413 | |
d2a91064 QW |
10414 | ASSERT(end + 1 - start <= U32_MAX); |
10415 | len = end + 1 - start; | |
c6100a4b | 10416 | while (index <= end_index) { |
d2a91064 | 10417 | page = find_get_page(inode->vfs_inode.i_mapping, index); |
c6100a4b | 10418 | ASSERT(page); /* Pages should be in the extent_io_tree */ |
d2a91064 QW |
10419 | |
10420 | btrfs_page_set_writeback(fs_info, page, start, len); | |
c6100a4b JB |
10421 | put_page(page); |
10422 | index++; | |
10423 | } | |
10424 | } | |
10425 | ||
ed46ff3d OS |
10426 | #ifdef CONFIG_SWAP |
10427 | /* | |
10428 | * Add an entry indicating a block group or device which is pinned by a | |
10429 | * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a | |
10430 | * negative errno on failure. | |
10431 | */ | |
10432 | static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr, | |
10433 | bool is_block_group) | |
10434 | { | |
10435 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10436 | struct btrfs_swapfile_pin *sp, *entry; | |
10437 | struct rb_node **p; | |
10438 | struct rb_node *parent = NULL; | |
10439 | ||
10440 | sp = kmalloc(sizeof(*sp), GFP_NOFS); | |
10441 | if (!sp) | |
10442 | return -ENOMEM; | |
10443 | sp->ptr = ptr; | |
10444 | sp->inode = inode; | |
10445 | sp->is_block_group = is_block_group; | |
195a49ea | 10446 | sp->bg_extent_count = 1; |
ed46ff3d OS |
10447 | |
10448 | spin_lock(&fs_info->swapfile_pins_lock); | |
10449 | p = &fs_info->swapfile_pins.rb_node; | |
10450 | while (*p) { | |
10451 | parent = *p; | |
10452 | entry = rb_entry(parent, struct btrfs_swapfile_pin, node); | |
10453 | if (sp->ptr < entry->ptr || | |
10454 | (sp->ptr == entry->ptr && sp->inode < entry->inode)) { | |
10455 | p = &(*p)->rb_left; | |
10456 | } else if (sp->ptr > entry->ptr || | |
10457 | (sp->ptr == entry->ptr && sp->inode > entry->inode)) { | |
10458 | p = &(*p)->rb_right; | |
10459 | } else { | |
195a49ea FM |
10460 | if (is_block_group) |
10461 | entry->bg_extent_count++; | |
ed46ff3d OS |
10462 | spin_unlock(&fs_info->swapfile_pins_lock); |
10463 | kfree(sp); | |
10464 | return 1; | |
10465 | } | |
10466 | } | |
10467 | rb_link_node(&sp->node, parent, p); | |
10468 | rb_insert_color(&sp->node, &fs_info->swapfile_pins); | |
10469 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10470 | return 0; | |
10471 | } | |
10472 | ||
10473 | /* Free all of the entries pinned by this swapfile. */ | |
10474 | static void btrfs_free_swapfile_pins(struct inode *inode) | |
10475 | { | |
10476 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10477 | struct btrfs_swapfile_pin *sp; | |
10478 | struct rb_node *node, *next; | |
10479 | ||
10480 | spin_lock(&fs_info->swapfile_pins_lock); | |
10481 | node = rb_first(&fs_info->swapfile_pins); | |
10482 | while (node) { | |
10483 | next = rb_next(node); | |
10484 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
10485 | if (sp->inode == inode) { | |
10486 | rb_erase(&sp->node, &fs_info->swapfile_pins); | |
195a49ea FM |
10487 | if (sp->is_block_group) { |
10488 | btrfs_dec_block_group_swap_extents(sp->ptr, | |
10489 | sp->bg_extent_count); | |
ed46ff3d | 10490 | btrfs_put_block_group(sp->ptr); |
195a49ea | 10491 | } |
ed46ff3d OS |
10492 | kfree(sp); |
10493 | } | |
10494 | node = next; | |
10495 | } | |
10496 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10497 | } | |
10498 | ||
10499 | struct btrfs_swap_info { | |
10500 | u64 start; | |
10501 | u64 block_start; | |
10502 | u64 block_len; | |
10503 | u64 lowest_ppage; | |
10504 | u64 highest_ppage; | |
10505 | unsigned long nr_pages; | |
10506 | int nr_extents; | |
10507 | }; | |
10508 | ||
10509 | static int btrfs_add_swap_extent(struct swap_info_struct *sis, | |
10510 | struct btrfs_swap_info *bsi) | |
10511 | { | |
10512 | unsigned long nr_pages; | |
10513 | u64 first_ppage, first_ppage_reported, next_ppage; | |
10514 | int ret; | |
10515 | ||
10516 | first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT; | |
10517 | next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len, | |
10518 | PAGE_SIZE) >> PAGE_SHIFT; | |
10519 | ||
10520 | if (first_ppage >= next_ppage) | |
10521 | return 0; | |
10522 | nr_pages = next_ppage - first_ppage; | |
10523 | ||
10524 | first_ppage_reported = first_ppage; | |
10525 | if (bsi->start == 0) | |
10526 | first_ppage_reported++; | |
10527 | if (bsi->lowest_ppage > first_ppage_reported) | |
10528 | bsi->lowest_ppage = first_ppage_reported; | |
10529 | if (bsi->highest_ppage < (next_ppage - 1)) | |
10530 | bsi->highest_ppage = next_ppage - 1; | |
10531 | ||
10532 | ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage); | |
10533 | if (ret < 0) | |
10534 | return ret; | |
10535 | bsi->nr_extents += ret; | |
10536 | bsi->nr_pages += nr_pages; | |
10537 | return 0; | |
10538 | } | |
10539 | ||
10540 | static void btrfs_swap_deactivate(struct file *file) | |
10541 | { | |
10542 | struct inode *inode = file_inode(file); | |
10543 | ||
10544 | btrfs_free_swapfile_pins(inode); | |
10545 | atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles); | |
10546 | } | |
10547 | ||
10548 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10549 | sector_t *span) | |
10550 | { | |
10551 | struct inode *inode = file_inode(file); | |
dd0734f2 FM |
10552 | struct btrfs_root *root = BTRFS_I(inode)->root; |
10553 | struct btrfs_fs_info *fs_info = root->fs_info; | |
ed46ff3d OS |
10554 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
10555 | struct extent_state *cached_state = NULL; | |
10556 | struct extent_map *em = NULL; | |
10557 | struct btrfs_device *device = NULL; | |
10558 | struct btrfs_swap_info bsi = { | |
10559 | .lowest_ppage = (sector_t)-1ULL, | |
10560 | }; | |
10561 | int ret = 0; | |
10562 | u64 isize; | |
10563 | u64 start; | |
10564 | ||
10565 | /* | |
10566 | * If the swap file was just created, make sure delalloc is done. If the | |
10567 | * file changes again after this, the user is doing something stupid and | |
10568 | * we don't really care. | |
10569 | */ | |
10570 | ret = btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
10571 | if (ret) | |
10572 | return ret; | |
10573 | ||
10574 | /* | |
10575 | * The inode is locked, so these flags won't change after we check them. | |
10576 | */ | |
10577 | if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) { | |
10578 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10579 | return -EINVAL; | |
10580 | } | |
10581 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) { | |
10582 | btrfs_warn(fs_info, "swapfile must not be copy-on-write"); | |
10583 | return -EINVAL; | |
10584 | } | |
10585 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
10586 | btrfs_warn(fs_info, "swapfile must not be checksummed"); | |
10587 | return -EINVAL; | |
10588 | } | |
10589 | ||
10590 | /* | |
10591 | * Balance or device remove/replace/resize can move stuff around from | |
c3e1f96c GR |
10592 | * under us. The exclop protection makes sure they aren't running/won't |
10593 | * run concurrently while we are mapping the swap extents, and | |
10594 | * fs_info->swapfile_pins prevents them from running while the swap | |
10595 | * file is active and moving the extents. Note that this also prevents | |
10596 | * a concurrent device add which isn't actually necessary, but it's not | |
ed46ff3d OS |
10597 | * really worth the trouble to allow it. |
10598 | */ | |
c3e1f96c | 10599 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) { |
ed46ff3d OS |
10600 | btrfs_warn(fs_info, |
10601 | "cannot activate swapfile while exclusive operation is running"); | |
10602 | return -EBUSY; | |
10603 | } | |
dd0734f2 FM |
10604 | |
10605 | /* | |
10606 | * Prevent snapshot creation while we are activating the swap file. | |
10607 | * We do not want to race with snapshot creation. If snapshot creation | |
10608 | * already started before we bumped nr_swapfiles from 0 to 1 and | |
10609 | * completes before the first write into the swap file after it is | |
10610 | * activated, than that write would fallback to COW. | |
10611 | */ | |
10612 | if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) { | |
10613 | btrfs_exclop_finish(fs_info); | |
10614 | btrfs_warn(fs_info, | |
10615 | "cannot activate swapfile because snapshot creation is in progress"); | |
10616 | return -EINVAL; | |
10617 | } | |
ed46ff3d OS |
10618 | /* |
10619 | * Snapshots can create extents which require COW even if NODATACOW is | |
10620 | * set. We use this counter to prevent snapshots. We must increment it | |
10621 | * before walking the extents because we don't want a concurrent | |
10622 | * snapshot to run after we've already checked the extents. | |
10623 | */ | |
dd0734f2 | 10624 | atomic_inc(&root->nr_swapfiles); |
ed46ff3d OS |
10625 | |
10626 | isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize); | |
10627 | ||
10628 | lock_extent_bits(io_tree, 0, isize - 1, &cached_state); | |
10629 | start = 0; | |
10630 | while (start < isize) { | |
10631 | u64 logical_block_start, physical_block_start; | |
32da5386 | 10632 | struct btrfs_block_group *bg; |
ed46ff3d OS |
10633 | u64 len = isize - start; |
10634 | ||
39b07b5d | 10635 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len); |
ed46ff3d OS |
10636 | if (IS_ERR(em)) { |
10637 | ret = PTR_ERR(em); | |
10638 | goto out; | |
10639 | } | |
10640 | ||
10641 | if (em->block_start == EXTENT_MAP_HOLE) { | |
10642 | btrfs_warn(fs_info, "swapfile must not have holes"); | |
10643 | ret = -EINVAL; | |
10644 | goto out; | |
10645 | } | |
10646 | if (em->block_start == EXTENT_MAP_INLINE) { | |
10647 | /* | |
10648 | * It's unlikely we'll ever actually find ourselves | |
10649 | * here, as a file small enough to fit inline won't be | |
10650 | * big enough to store more than the swap header, but in | |
10651 | * case something changes in the future, let's catch it | |
10652 | * here rather than later. | |
10653 | */ | |
10654 | btrfs_warn(fs_info, "swapfile must not be inline"); | |
10655 | ret = -EINVAL; | |
10656 | goto out; | |
10657 | } | |
10658 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
10659 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10660 | ret = -EINVAL; | |
10661 | goto out; | |
10662 | } | |
10663 | ||
10664 | logical_block_start = em->block_start + (start - em->start); | |
10665 | len = min(len, em->len - (start - em->start)); | |
10666 | free_extent_map(em); | |
10667 | em = NULL; | |
10668 | ||
a84d5d42 | 10669 | ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true); |
ed46ff3d OS |
10670 | if (ret < 0) { |
10671 | goto out; | |
10672 | } else if (ret) { | |
10673 | ret = 0; | |
10674 | } else { | |
10675 | btrfs_warn(fs_info, | |
10676 | "swapfile must not be copy-on-write"); | |
10677 | ret = -EINVAL; | |
10678 | goto out; | |
10679 | } | |
10680 | ||
10681 | em = btrfs_get_chunk_map(fs_info, logical_block_start, len); | |
10682 | if (IS_ERR(em)) { | |
10683 | ret = PTR_ERR(em); | |
10684 | goto out; | |
10685 | } | |
10686 | ||
10687 | if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
10688 | btrfs_warn(fs_info, | |
10689 | "swapfile must have single data profile"); | |
10690 | ret = -EINVAL; | |
10691 | goto out; | |
10692 | } | |
10693 | ||
10694 | if (device == NULL) { | |
10695 | device = em->map_lookup->stripes[0].dev; | |
10696 | ret = btrfs_add_swapfile_pin(inode, device, false); | |
10697 | if (ret == 1) | |
10698 | ret = 0; | |
10699 | else if (ret) | |
10700 | goto out; | |
10701 | } else if (device != em->map_lookup->stripes[0].dev) { | |
10702 | btrfs_warn(fs_info, "swapfile must be on one device"); | |
10703 | ret = -EINVAL; | |
10704 | goto out; | |
10705 | } | |
10706 | ||
10707 | physical_block_start = (em->map_lookup->stripes[0].physical + | |
10708 | (logical_block_start - em->start)); | |
10709 | len = min(len, em->len - (logical_block_start - em->start)); | |
10710 | free_extent_map(em); | |
10711 | em = NULL; | |
10712 | ||
10713 | bg = btrfs_lookup_block_group(fs_info, logical_block_start); | |
10714 | if (!bg) { | |
10715 | btrfs_warn(fs_info, | |
10716 | "could not find block group containing swapfile"); | |
10717 | ret = -EINVAL; | |
10718 | goto out; | |
10719 | } | |
10720 | ||
195a49ea FM |
10721 | if (!btrfs_inc_block_group_swap_extents(bg)) { |
10722 | btrfs_warn(fs_info, | |
10723 | "block group for swapfile at %llu is read-only%s", | |
10724 | bg->start, | |
10725 | atomic_read(&fs_info->scrubs_running) ? | |
10726 | " (scrub running)" : ""); | |
10727 | btrfs_put_block_group(bg); | |
10728 | ret = -EINVAL; | |
10729 | goto out; | |
10730 | } | |
10731 | ||
ed46ff3d OS |
10732 | ret = btrfs_add_swapfile_pin(inode, bg, true); |
10733 | if (ret) { | |
10734 | btrfs_put_block_group(bg); | |
10735 | if (ret == 1) | |
10736 | ret = 0; | |
10737 | else | |
10738 | goto out; | |
10739 | } | |
10740 | ||
10741 | if (bsi.block_len && | |
10742 | bsi.block_start + bsi.block_len == physical_block_start) { | |
10743 | bsi.block_len += len; | |
10744 | } else { | |
10745 | if (bsi.block_len) { | |
10746 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10747 | if (ret) | |
10748 | goto out; | |
10749 | } | |
10750 | bsi.start = start; | |
10751 | bsi.block_start = physical_block_start; | |
10752 | bsi.block_len = len; | |
10753 | } | |
10754 | ||
10755 | start += len; | |
10756 | } | |
10757 | ||
10758 | if (bsi.block_len) | |
10759 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10760 | ||
10761 | out: | |
10762 | if (!IS_ERR_OR_NULL(em)) | |
10763 | free_extent_map(em); | |
10764 | ||
10765 | unlock_extent_cached(io_tree, 0, isize - 1, &cached_state); | |
10766 | ||
10767 | if (ret) | |
10768 | btrfs_swap_deactivate(file); | |
10769 | ||
dd0734f2 FM |
10770 | btrfs_drew_write_unlock(&root->snapshot_lock); |
10771 | ||
c3e1f96c | 10772 | btrfs_exclop_finish(fs_info); |
ed46ff3d OS |
10773 | |
10774 | if (ret) | |
10775 | return ret; | |
10776 | ||
10777 | if (device) | |
10778 | sis->bdev = device->bdev; | |
10779 | *span = bsi.highest_ppage - bsi.lowest_ppage + 1; | |
10780 | sis->max = bsi.nr_pages; | |
10781 | sis->pages = bsi.nr_pages - 1; | |
10782 | sis->highest_bit = bsi.nr_pages - 1; | |
10783 | return bsi.nr_extents; | |
10784 | } | |
10785 | #else | |
10786 | static void btrfs_swap_deactivate(struct file *file) | |
10787 | { | |
10788 | } | |
10789 | ||
10790 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10791 | sector_t *span) | |
10792 | { | |
10793 | return -EOPNOTSUPP; | |
10794 | } | |
10795 | #endif | |
10796 | ||
2766ff61 FM |
10797 | /* |
10798 | * Update the number of bytes used in the VFS' inode. When we replace extents in | |
10799 | * a range (clone, dedupe, fallocate's zero range), we must update the number of | |
10800 | * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls | |
10801 | * always get a correct value. | |
10802 | */ | |
10803 | void btrfs_update_inode_bytes(struct btrfs_inode *inode, | |
10804 | const u64 add_bytes, | |
10805 | const u64 del_bytes) | |
10806 | { | |
10807 | if (add_bytes == del_bytes) | |
10808 | return; | |
10809 | ||
10810 | spin_lock(&inode->lock); | |
10811 | if (del_bytes > 0) | |
10812 | inode_sub_bytes(&inode->vfs_inode, del_bytes); | |
10813 | if (add_bytes > 0) | |
10814 | inode_add_bytes(&inode->vfs_inode, add_bytes); | |
10815 | spin_unlock(&inode->lock); | |
10816 | } | |
10817 | ||
6e1d5dcc | 10818 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10819 | .getattr = btrfs_getattr, |
39279cc3 CM |
10820 | .lookup = btrfs_lookup, |
10821 | .create = btrfs_create, | |
10822 | .unlink = btrfs_unlink, | |
10823 | .link = btrfs_link, | |
10824 | .mkdir = btrfs_mkdir, | |
10825 | .rmdir = btrfs_rmdir, | |
2773bf00 | 10826 | .rename = btrfs_rename2, |
39279cc3 CM |
10827 | .symlink = btrfs_symlink, |
10828 | .setattr = btrfs_setattr, | |
618e21d5 | 10829 | .mknod = btrfs_mknod, |
5103e947 | 10830 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10831 | .permission = btrfs_permission, |
4e34e719 | 10832 | .get_acl = btrfs_get_acl, |
996a710d | 10833 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10834 | .update_time = btrfs_update_time, |
ef3b9af5 | 10835 | .tmpfile = btrfs_tmpfile, |
97fc2977 MS |
10836 | .fileattr_get = btrfs_fileattr_get, |
10837 | .fileattr_set = btrfs_fileattr_set, | |
39279cc3 | 10838 | }; |
76dda93c | 10839 | |
828c0950 | 10840 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10841 | .llseek = generic_file_llseek, |
10842 | .read = generic_read_dir, | |
02dbfc99 | 10843 | .iterate_shared = btrfs_real_readdir, |
23b5ec74 | 10844 | .open = btrfs_opendir, |
34287aa3 | 10845 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10846 | #ifdef CONFIG_COMPAT |
4c63c245 | 10847 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 10848 | #endif |
6bf13c0c | 10849 | .release = btrfs_release_file, |
e02119d5 | 10850 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10851 | }; |
10852 | ||
35054394 CM |
10853 | /* |
10854 | * btrfs doesn't support the bmap operation because swapfiles | |
10855 | * use bmap to make a mapping of extents in the file. They assume | |
10856 | * these extents won't change over the life of the file and they | |
10857 | * use the bmap result to do IO directly to the drive. | |
10858 | * | |
10859 | * the btrfs bmap call would return logical addresses that aren't | |
10860 | * suitable for IO and they also will change frequently as COW | |
10861 | * operations happen. So, swapfile + btrfs == corruption. | |
10862 | * | |
10863 | * For now we're avoiding this by dropping bmap. | |
10864 | */ | |
7f09410b | 10865 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
10866 | .readpage = btrfs_readpage, |
10867 | .writepage = btrfs_writepage, | |
b293f02e | 10868 | .writepages = btrfs_writepages, |
ba206a02 | 10869 | .readahead = btrfs_readahead, |
f85781fb | 10870 | .direct_IO = noop_direct_IO, |
a52d9a80 CM |
10871 | .invalidatepage = btrfs_invalidatepage, |
10872 | .releasepage = btrfs_releasepage, | |
f8e66081 RG |
10873 | #ifdef CONFIG_MIGRATION |
10874 | .migratepage = btrfs_migratepage, | |
10875 | #endif | |
e6dcd2dc | 10876 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 10877 | .error_remove_page = generic_error_remove_page, |
ed46ff3d OS |
10878 | .swap_activate = btrfs_swap_activate, |
10879 | .swap_deactivate = btrfs_swap_deactivate, | |
39279cc3 CM |
10880 | }; |
10881 | ||
6e1d5dcc | 10882 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10883 | .getattr = btrfs_getattr, |
10884 | .setattr = btrfs_setattr, | |
5103e947 | 10885 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10886 | .permission = btrfs_permission, |
1506fcc8 | 10887 | .fiemap = btrfs_fiemap, |
4e34e719 | 10888 | .get_acl = btrfs_get_acl, |
996a710d | 10889 | .set_acl = btrfs_set_acl, |
e41f941a | 10890 | .update_time = btrfs_update_time, |
97fc2977 MS |
10891 | .fileattr_get = btrfs_fileattr_get, |
10892 | .fileattr_set = btrfs_fileattr_set, | |
39279cc3 | 10893 | }; |
6e1d5dcc | 10894 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10895 | .getattr = btrfs_getattr, |
10896 | .setattr = btrfs_setattr, | |
fdebe2bd | 10897 | .permission = btrfs_permission, |
33268eaf | 10898 | .listxattr = btrfs_listxattr, |
4e34e719 | 10899 | .get_acl = btrfs_get_acl, |
996a710d | 10900 | .set_acl = btrfs_set_acl, |
e41f941a | 10901 | .update_time = btrfs_update_time, |
618e21d5 | 10902 | }; |
6e1d5dcc | 10903 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 10904 | .get_link = page_get_link, |
f209561a | 10905 | .getattr = btrfs_getattr, |
22c44fe6 | 10906 | .setattr = btrfs_setattr, |
fdebe2bd | 10907 | .permission = btrfs_permission, |
0279b4cd | 10908 | .listxattr = btrfs_listxattr, |
e41f941a | 10909 | .update_time = btrfs_update_time, |
39279cc3 | 10910 | }; |
76dda93c | 10911 | |
82d339d9 | 10912 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
10913 | .d_delete = btrfs_dentry_delete, |
10914 | }; |