]>
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" |
39279cc3 CM |
53 | |
54 | struct btrfs_iget_args { | |
0202e83f | 55 | u64 ino; |
39279cc3 CM |
56 | struct btrfs_root *root; |
57 | }; | |
58 | ||
f28a4928 | 59 | struct btrfs_dio_data { |
f28a4928 | 60 | u64 reserve; |
f85781fb GR |
61 | loff_t length; |
62 | ssize_t submitted; | |
63 | struct extent_changeset *data_reserved; | |
f28a4928 FM |
64 | }; |
65 | ||
6e1d5dcc AD |
66 | static const struct inode_operations btrfs_dir_inode_operations; |
67 | static const struct inode_operations btrfs_symlink_inode_operations; | |
6e1d5dcc AD |
68 | static const struct inode_operations btrfs_special_inode_operations; |
69 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b | 70 | static const struct address_space_operations btrfs_aops; |
828c0950 | 71 | static const struct file_operations btrfs_dir_file_operations; |
39279cc3 CM |
72 | |
73 | static struct kmem_cache *btrfs_inode_cachep; | |
74 | struct kmem_cache *btrfs_trans_handle_cachep; | |
39279cc3 | 75 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 76 | struct kmem_cache *btrfs_free_space_cachep; |
3acd4850 | 77 | struct kmem_cache *btrfs_free_space_bitmap_cachep; |
39279cc3 | 78 | |
3972f260 | 79 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
213e8c55 | 80 | static int btrfs_truncate(struct inode *inode, bool skip_writeback); |
5fd02043 | 81 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
6e26c442 | 82 | static noinline int cow_file_range(struct btrfs_inode *inode, |
771ed689 | 83 | struct page *locked_page, |
74e9194a | 84 | u64 start, u64 end, int *page_started, |
330a5827 | 85 | unsigned long *nr_written, int unlock); |
4b67c11d NB |
86 | static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start, |
87 | u64 len, u64 orig_start, u64 block_start, | |
6f9994db LB |
88 | u64 block_len, u64 orig_block_len, |
89 | u64 ram_bytes, int compress_type, | |
90 | int type); | |
7b128766 | 91 | |
b672b5c1 | 92 | static void __endio_write_update_ordered(struct btrfs_inode *inode, |
52427260 QW |
93 | const u64 offset, const u64 bytes, |
94 | const bool uptodate); | |
95 | ||
a14b78ad GR |
96 | /* |
97 | * btrfs_inode_lock - lock inode i_rwsem based on arguments passed | |
98 | * | |
99 | * ilock_flags can have the following bit set: | |
100 | * | |
101 | * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode | |
102 | * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt | |
103 | * return -EAGAIN | |
104 | */ | |
105 | int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags) | |
106 | { | |
107 | if (ilock_flags & BTRFS_ILOCK_SHARED) { | |
108 | if (ilock_flags & BTRFS_ILOCK_TRY) { | |
109 | if (!inode_trylock_shared(inode)) | |
110 | return -EAGAIN; | |
111 | else | |
112 | return 0; | |
113 | } | |
114 | inode_lock_shared(inode); | |
115 | } else { | |
116 | if (ilock_flags & BTRFS_ILOCK_TRY) { | |
117 | if (!inode_trylock(inode)) | |
118 | return -EAGAIN; | |
119 | else | |
120 | return 0; | |
121 | } | |
122 | inode_lock(inode); | |
123 | } | |
124 | return 0; | |
125 | } | |
126 | ||
127 | /* | |
128 | * btrfs_inode_unlock - unock inode i_rwsem | |
129 | * | |
130 | * ilock_flags should contain the same bits set as passed to btrfs_inode_lock() | |
131 | * to decide whether the lock acquired is shared or exclusive. | |
132 | */ | |
133 | void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags) | |
134 | { | |
135 | if (ilock_flags & BTRFS_ILOCK_SHARED) | |
136 | inode_unlock_shared(inode); | |
137 | else | |
138 | inode_unlock(inode); | |
139 | } | |
140 | ||
52427260 QW |
141 | /* |
142 | * Cleanup all submitted ordered extents in specified range to handle errors | |
52042d8e | 143 | * from the btrfs_run_delalloc_range() callback. |
52427260 QW |
144 | * |
145 | * NOTE: caller must ensure that when an error happens, it can not call | |
146 | * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING | |
147 | * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata | |
148 | * to be released, which we want to happen only when finishing the ordered | |
d1051d6e | 149 | * extent (btrfs_finish_ordered_io()). |
52427260 | 150 | */ |
64e1db56 | 151 | static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode, |
d1051d6e NB |
152 | struct page *locked_page, |
153 | u64 offset, u64 bytes) | |
52427260 | 154 | { |
63d71450 NA |
155 | unsigned long index = offset >> PAGE_SHIFT; |
156 | unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT; | |
d1051d6e NB |
157 | u64 page_start = page_offset(locked_page); |
158 | u64 page_end = page_start + PAGE_SIZE - 1; | |
159 | ||
63d71450 NA |
160 | struct page *page; |
161 | ||
162 | while (index <= end_index) { | |
64e1db56 | 163 | page = find_get_page(inode->vfs_inode.i_mapping, index); |
63d71450 NA |
164 | index++; |
165 | if (!page) | |
166 | continue; | |
167 | ClearPagePrivate2(page); | |
168 | put_page(page); | |
169 | } | |
d1051d6e NB |
170 | |
171 | /* | |
172 | * In case this page belongs to the delalloc range being instantiated | |
173 | * then skip it, since the first page of a range is going to be | |
174 | * properly cleaned up by the caller of run_delalloc_range | |
175 | */ | |
176 | if (page_start >= offset && page_end <= (offset + bytes - 1)) { | |
177 | offset += PAGE_SIZE; | |
178 | bytes -= PAGE_SIZE; | |
179 | } | |
180 | ||
64e1db56 | 181 | return __endio_write_update_ordered(inode, offset, bytes, false); |
52427260 QW |
182 | } |
183 | ||
48a3b636 | 184 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 185 | |
f34f57a3 | 186 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
187 | struct inode *inode, struct inode *dir, |
188 | const struct qstr *qstr) | |
0279b4cd JO |
189 | { |
190 | int err; | |
191 | ||
f34f57a3 | 192 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 193 | if (!err) |
2a7dba39 | 194 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
195 | return err; |
196 | } | |
197 | ||
c8b97818 CM |
198 | /* |
199 | * this does all the hard work for inserting an inline extent into | |
200 | * the btree. The caller should have done a btrfs_drop_extents so that | |
201 | * no overlapping inline items exist in the btree | |
202 | */ | |
40f76580 | 203 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
5893dfb9 | 204 | struct btrfs_path *path, bool extent_inserted, |
c8b97818 CM |
205 | struct btrfs_root *root, struct inode *inode, |
206 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 207 | int compress_type, |
c8b97818 CM |
208 | struct page **compressed_pages) |
209 | { | |
c8b97818 CM |
210 | struct extent_buffer *leaf; |
211 | struct page *page = NULL; | |
212 | char *kaddr; | |
213 | unsigned long ptr; | |
214 | struct btrfs_file_extent_item *ei; | |
c8b97818 CM |
215 | int ret; |
216 | size_t cur_size = size; | |
c8b97818 | 217 | unsigned long offset; |
c8b97818 | 218 | |
982f1f5d JJB |
219 | ASSERT((compressed_size > 0 && compressed_pages) || |
220 | (compressed_size == 0 && !compressed_pages)); | |
221 | ||
fe3f566c | 222 | if (compressed_size && compressed_pages) |
c8b97818 | 223 | cur_size = compressed_size; |
c8b97818 | 224 | |
1acae57b FDBM |
225 | if (!extent_inserted) { |
226 | struct btrfs_key key; | |
227 | size_t datasize; | |
c8b97818 | 228 | |
4a0cc7ca | 229 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b | 230 | key.offset = start; |
962a298f | 231 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 232 | |
1acae57b | 233 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
1acae57b FDBM |
234 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
235 | datasize); | |
79b4f4c6 | 236 | if (ret) |
1acae57b | 237 | goto fail; |
c8b97818 CM |
238 | } |
239 | leaf = path->nodes[0]; | |
240 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
241 | struct btrfs_file_extent_item); | |
242 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
243 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
244 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
245 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
246 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
247 | ptr = btrfs_file_extent_inline_start(ei); | |
248 | ||
261507a0 | 249 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
250 | struct page *cpage; |
251 | int i = 0; | |
d397712b | 252 | while (compressed_size > 0) { |
c8b97818 | 253 | cpage = compressed_pages[i]; |
5b050f04 | 254 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 255 | PAGE_SIZE); |
c8b97818 | 256 | |
7ac687d9 | 257 | kaddr = kmap_atomic(cpage); |
c8b97818 | 258 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 259 | kunmap_atomic(kaddr); |
c8b97818 CM |
260 | |
261 | i++; | |
262 | ptr += cur_size; | |
263 | compressed_size -= cur_size; | |
264 | } | |
265 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 266 | compress_type); |
c8b97818 CM |
267 | } else { |
268 | page = find_get_page(inode->i_mapping, | |
09cbfeaf | 269 | start >> PAGE_SHIFT); |
c8b97818 | 270 | btrfs_set_file_extent_compression(leaf, ei, 0); |
7ac687d9 | 271 | kaddr = kmap_atomic(page); |
7073017a | 272 | offset = offset_in_page(start); |
c8b97818 | 273 | write_extent_buffer(leaf, kaddr + offset, ptr, size); |
7ac687d9 | 274 | kunmap_atomic(kaddr); |
09cbfeaf | 275 | put_page(page); |
c8b97818 CM |
276 | } |
277 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 278 | btrfs_release_path(path); |
c8b97818 | 279 | |
9ddc959e JB |
280 | /* |
281 | * We align size to sectorsize for inline extents just for simplicity | |
282 | * sake. | |
283 | */ | |
284 | size = ALIGN(size, root->fs_info->sectorsize); | |
285 | ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size); | |
286 | if (ret) | |
287 | goto fail; | |
288 | ||
c2167754 YZ |
289 | /* |
290 | * we're an inline extent, so nobody can | |
291 | * extend the file past i_size without locking | |
292 | * a page we already have locked. | |
293 | * | |
294 | * We must do any isize and inode updates | |
295 | * before we unlock the pages. Otherwise we | |
296 | * could end up racing with unlink. | |
297 | */ | |
c8b97818 | 298 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
c8b97818 | 299 | fail: |
79b4f4c6 | 300 | return ret; |
c8b97818 CM |
301 | } |
302 | ||
303 | ||
304 | /* | |
305 | * conditionally insert an inline extent into the file. This | |
306 | * does the checks required to make sure the data is small enough | |
307 | * to fit as an inline extent. | |
308 | */ | |
a0349401 | 309 | static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start, |
00361589 JB |
310 | u64 end, size_t compressed_size, |
311 | int compress_type, | |
312 | struct page **compressed_pages) | |
c8b97818 | 313 | { |
5893dfb9 | 314 | struct btrfs_drop_extents_args drop_args = { 0 }; |
a0349401 | 315 | struct btrfs_root *root = inode->root; |
0b246afa | 316 | struct btrfs_fs_info *fs_info = root->fs_info; |
00361589 | 317 | struct btrfs_trans_handle *trans; |
a0349401 | 318 | u64 isize = i_size_read(&inode->vfs_inode); |
c8b97818 CM |
319 | u64 actual_end = min(end + 1, isize); |
320 | u64 inline_len = actual_end - start; | |
0b246afa | 321 | u64 aligned_end = ALIGN(end, fs_info->sectorsize); |
c8b97818 CM |
322 | u64 data_len = inline_len; |
323 | int ret; | |
1acae57b | 324 | struct btrfs_path *path; |
c8b97818 CM |
325 | |
326 | if (compressed_size) | |
327 | data_len = compressed_size; | |
328 | ||
329 | if (start > 0 || | |
0b246afa JM |
330 | actual_end > fs_info->sectorsize || |
331 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) || | |
c8b97818 | 332 | (!compressed_size && |
0b246afa | 333 | (actual_end & (fs_info->sectorsize - 1)) == 0) || |
c8b97818 | 334 | end + 1 < isize || |
0b246afa | 335 | data_len > fs_info->max_inline) { |
c8b97818 CM |
336 | return 1; |
337 | } | |
338 | ||
1acae57b FDBM |
339 | path = btrfs_alloc_path(); |
340 | if (!path) | |
341 | return -ENOMEM; | |
342 | ||
00361589 | 343 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
344 | if (IS_ERR(trans)) { |
345 | btrfs_free_path(path); | |
00361589 | 346 | return PTR_ERR(trans); |
1acae57b | 347 | } |
a0349401 | 348 | trans->block_rsv = &inode->block_rsv; |
00361589 | 349 | |
5893dfb9 FM |
350 | drop_args.path = path; |
351 | drop_args.start = start; | |
352 | drop_args.end = aligned_end; | |
353 | drop_args.drop_cache = true; | |
354 | drop_args.replace_extent = true; | |
355 | ||
1acae57b | 356 | if (compressed_size && compressed_pages) |
5893dfb9 | 357 | drop_args.extent_item_size = btrfs_file_extent_calc_inline_size( |
1acae57b FDBM |
358 | compressed_size); |
359 | else | |
5893dfb9 | 360 | drop_args.extent_item_size = btrfs_file_extent_calc_inline_size( |
1acae57b FDBM |
361 | inline_len); |
362 | ||
5893dfb9 | 363 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
00361589 | 364 | if (ret) { |
66642832 | 365 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
366 | goto out; |
367 | } | |
c8b97818 CM |
368 | |
369 | if (isize > actual_end) | |
370 | inline_len = min_t(u64, isize, actual_end); | |
5893dfb9 | 371 | ret = insert_inline_extent(trans, path, drop_args.extent_inserted, |
a0349401 | 372 | root, &inode->vfs_inode, start, |
c8b97818 | 373 | inline_len, compressed_size, |
fe3f566c | 374 | compress_type, compressed_pages); |
2adcac1a | 375 | if (ret && ret != -ENOSPC) { |
66642832 | 376 | btrfs_abort_transaction(trans, ret); |
00361589 | 377 | goto out; |
2adcac1a | 378 | } else if (ret == -ENOSPC) { |
00361589 JB |
379 | ret = 1; |
380 | goto out; | |
79787eaa | 381 | } |
2adcac1a | 382 | |
2766ff61 | 383 | btrfs_update_inode_bytes(inode, inline_len, drop_args.bytes_found); |
9a56fcd1 | 384 | ret = btrfs_update_inode(trans, root, inode); |
2766ff61 FM |
385 | if (ret && ret != -ENOSPC) { |
386 | btrfs_abort_transaction(trans, ret); | |
387 | goto out; | |
388 | } else if (ret == -ENOSPC) { | |
389 | ret = 1; | |
390 | goto out; | |
391 | } | |
392 | ||
a0349401 | 393 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags); |
00361589 | 394 | out: |
94ed938a QW |
395 | /* |
396 | * Don't forget to free the reserved space, as for inlined extent | |
397 | * it won't count as data extent, free them directly here. | |
398 | * And at reserve time, it's always aligned to page size, so | |
399 | * just free one page here. | |
400 | */ | |
a0349401 | 401 | btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE); |
1acae57b | 402 | btrfs_free_path(path); |
3a45bb20 | 403 | btrfs_end_transaction(trans); |
00361589 | 404 | return ret; |
c8b97818 CM |
405 | } |
406 | ||
771ed689 CM |
407 | struct async_extent { |
408 | u64 start; | |
409 | u64 ram_size; | |
410 | u64 compressed_size; | |
411 | struct page **pages; | |
412 | unsigned long nr_pages; | |
261507a0 | 413 | int compress_type; |
771ed689 CM |
414 | struct list_head list; |
415 | }; | |
416 | ||
97db1204 | 417 | struct async_chunk { |
771ed689 | 418 | struct inode *inode; |
771ed689 CM |
419 | struct page *locked_page; |
420 | u64 start; | |
421 | u64 end; | |
f82b7359 | 422 | unsigned int write_flags; |
771ed689 | 423 | struct list_head extents; |
ec39f769 | 424 | struct cgroup_subsys_state *blkcg_css; |
771ed689 | 425 | struct btrfs_work work; |
97db1204 | 426 | atomic_t *pending; |
771ed689 CM |
427 | }; |
428 | ||
97db1204 NB |
429 | struct async_cow { |
430 | /* Number of chunks in flight; must be first in the structure */ | |
431 | atomic_t num_chunks; | |
432 | struct async_chunk chunks[]; | |
771ed689 CM |
433 | }; |
434 | ||
97db1204 | 435 | static noinline int add_async_extent(struct async_chunk *cow, |
771ed689 CM |
436 | u64 start, u64 ram_size, |
437 | u64 compressed_size, | |
438 | struct page **pages, | |
261507a0 LZ |
439 | unsigned long nr_pages, |
440 | int compress_type) | |
771ed689 CM |
441 | { |
442 | struct async_extent *async_extent; | |
443 | ||
444 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 445 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
446 | async_extent->start = start; |
447 | async_extent->ram_size = ram_size; | |
448 | async_extent->compressed_size = compressed_size; | |
449 | async_extent->pages = pages; | |
450 | async_extent->nr_pages = nr_pages; | |
261507a0 | 451 | async_extent->compress_type = compress_type; |
771ed689 CM |
452 | list_add_tail(&async_extent->list, &cow->extents); |
453 | return 0; | |
454 | } | |
455 | ||
42c16da6 QW |
456 | /* |
457 | * Check if the inode has flags compatible with compression | |
458 | */ | |
99c88dc7 | 459 | static inline bool inode_can_compress(struct btrfs_inode *inode) |
42c16da6 | 460 | { |
99c88dc7 NB |
461 | if (inode->flags & BTRFS_INODE_NODATACOW || |
462 | inode->flags & BTRFS_INODE_NODATASUM) | |
42c16da6 QW |
463 | return false; |
464 | return true; | |
465 | } | |
466 | ||
467 | /* | |
468 | * Check if the inode needs to be submitted to compression, based on mount | |
469 | * options, defragmentation, properties or heuristics. | |
470 | */ | |
808a1292 NB |
471 | static inline int inode_need_compress(struct btrfs_inode *inode, u64 start, |
472 | u64 end) | |
f79707b0 | 473 | { |
808a1292 | 474 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
f79707b0 | 475 | |
808a1292 | 476 | if (!inode_can_compress(inode)) { |
42c16da6 QW |
477 | WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), |
478 | KERN_ERR "BTRFS: unexpected compression for ino %llu\n", | |
808a1292 | 479 | btrfs_ino(inode)); |
42c16da6 QW |
480 | return 0; |
481 | } | |
f79707b0 | 482 | /* force compress */ |
0b246afa | 483 | if (btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
f79707b0 | 484 | return 1; |
eec63c65 | 485 | /* defrag ioctl */ |
808a1292 | 486 | if (inode->defrag_compress) |
eec63c65 | 487 | return 1; |
f79707b0 | 488 | /* bad compression ratios */ |
808a1292 | 489 | if (inode->flags & BTRFS_INODE_NOCOMPRESS) |
f79707b0 | 490 | return 0; |
0b246afa | 491 | if (btrfs_test_opt(fs_info, COMPRESS) || |
808a1292 NB |
492 | inode->flags & BTRFS_INODE_COMPRESS || |
493 | inode->prop_compress) | |
494 | return btrfs_compress_heuristic(&inode->vfs_inode, start, end); | |
f79707b0 WS |
495 | return 0; |
496 | } | |
497 | ||
6158e1ce | 498 | static inline void inode_should_defrag(struct btrfs_inode *inode, |
26d30f85 AJ |
499 | u64 start, u64 end, u64 num_bytes, u64 small_write) |
500 | { | |
501 | /* If this is a small write inside eof, kick off a defrag */ | |
502 | if (num_bytes < small_write && | |
6158e1ce | 503 | (start > 0 || end + 1 < inode->disk_i_size)) |
26d30f85 AJ |
504 | btrfs_add_inode_defrag(NULL, inode); |
505 | } | |
506 | ||
d352ac68 | 507 | /* |
771ed689 CM |
508 | * we create compressed extents in two phases. The first |
509 | * phase compresses a range of pages that have already been | |
510 | * locked (both pages and state bits are locked). | |
c8b97818 | 511 | * |
771ed689 CM |
512 | * This is done inside an ordered work queue, and the compression |
513 | * is spread across many cpus. The actual IO submission is step | |
514 | * two, and the ordered work queue takes care of making sure that | |
515 | * happens in the same order things were put onto the queue by | |
516 | * writepages and friends. | |
c8b97818 | 517 | * |
771ed689 CM |
518 | * If this code finds it can't get good compression, it puts an |
519 | * entry onto the work queue to write the uncompressed bytes. This | |
520 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
521 | * are written in the same order that the flusher thread sent them |
522 | * down. | |
d352ac68 | 523 | */ |
ac3e9933 | 524 | static noinline int compress_file_range(struct async_chunk *async_chunk) |
b888db2b | 525 | { |
1368c6da | 526 | struct inode *inode = async_chunk->inode; |
0b246afa | 527 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
0b246afa | 528 | u64 blocksize = fs_info->sectorsize; |
1368c6da NB |
529 | u64 start = async_chunk->start; |
530 | u64 end = async_chunk->end; | |
c8b97818 | 531 | u64 actual_end; |
d98da499 | 532 | u64 i_size; |
e6dcd2dc | 533 | int ret = 0; |
c8b97818 CM |
534 | struct page **pages = NULL; |
535 | unsigned long nr_pages; | |
c8b97818 CM |
536 | unsigned long total_compressed = 0; |
537 | unsigned long total_in = 0; | |
c8b97818 CM |
538 | int i; |
539 | int will_compress; | |
0b246afa | 540 | int compress_type = fs_info->compress_type; |
ac3e9933 | 541 | int compressed_extents = 0; |
4adaa611 | 542 | int redirty = 0; |
b888db2b | 543 | |
6158e1ce NB |
544 | inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1, |
545 | SZ_16K); | |
4cb5300b | 546 | |
d98da499 JB |
547 | /* |
548 | * We need to save i_size before now because it could change in between | |
549 | * us evaluating the size and assigning it. This is because we lock and | |
550 | * unlock the page in truncate and fallocate, and then modify the i_size | |
551 | * later on. | |
552 | * | |
553 | * The barriers are to emulate READ_ONCE, remove that once i_size_read | |
554 | * does that for us. | |
555 | */ | |
556 | barrier(); | |
557 | i_size = i_size_read(inode); | |
558 | barrier(); | |
559 | actual_end = min_t(u64, i_size, end + 1); | |
c8b97818 CM |
560 | again: |
561 | will_compress = 0; | |
09cbfeaf | 562 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
069eac78 DS |
563 | BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0); |
564 | nr_pages = min_t(unsigned long, nr_pages, | |
565 | BTRFS_MAX_COMPRESSED / PAGE_SIZE); | |
be20aa9d | 566 | |
f03d9301 CM |
567 | /* |
568 | * we don't want to send crud past the end of i_size through | |
569 | * compression, that's just a waste of CPU time. So, if the | |
570 | * end of the file is before the start of our current | |
571 | * requested range of bytes, we bail out to the uncompressed | |
572 | * cleanup code that can deal with all of this. | |
573 | * | |
574 | * It isn't really the fastest way to fix things, but this is a | |
575 | * very uncommon corner. | |
576 | */ | |
577 | if (actual_end <= start) | |
578 | goto cleanup_and_bail_uncompressed; | |
579 | ||
c8b97818 CM |
580 | total_compressed = actual_end - start; |
581 | ||
4bcbb332 SW |
582 | /* |
583 | * skip compression for a small file range(<=blocksize) that | |
01327610 | 584 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
585 | */ |
586 | if (total_compressed <= blocksize && | |
587 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
588 | goto cleanup_and_bail_uncompressed; | |
589 | ||
069eac78 DS |
590 | total_compressed = min_t(unsigned long, total_compressed, |
591 | BTRFS_MAX_UNCOMPRESSED); | |
c8b97818 CM |
592 | total_in = 0; |
593 | ret = 0; | |
db94535d | 594 | |
771ed689 CM |
595 | /* |
596 | * we do compression for mount -o compress and when the | |
597 | * inode has not been flagged as nocompress. This flag can | |
598 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 599 | */ |
808a1292 | 600 | if (inode_need_compress(BTRFS_I(inode), start, end)) { |
c8b97818 | 601 | WARN_ON(pages); |
31e818fe | 602 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
603 | if (!pages) { |
604 | /* just bail out to the uncompressed code */ | |
3527a018 | 605 | nr_pages = 0; |
560f7d75 LZ |
606 | goto cont; |
607 | } | |
c8b97818 | 608 | |
eec63c65 DS |
609 | if (BTRFS_I(inode)->defrag_compress) |
610 | compress_type = BTRFS_I(inode)->defrag_compress; | |
611 | else if (BTRFS_I(inode)->prop_compress) | |
b52aa8c9 | 612 | compress_type = BTRFS_I(inode)->prop_compress; |
261507a0 | 613 | |
4adaa611 CM |
614 | /* |
615 | * we need to call clear_page_dirty_for_io on each | |
616 | * page in the range. Otherwise applications with the file | |
617 | * mmap'd can wander in and change the page contents while | |
618 | * we are compressing them. | |
619 | * | |
620 | * If the compression fails for any reason, we set the pages | |
621 | * dirty again later on. | |
e9679de3 TT |
622 | * |
623 | * Note that the remaining part is redirtied, the start pointer | |
624 | * has moved, the end is the original one. | |
4adaa611 | 625 | */ |
e9679de3 TT |
626 | if (!redirty) { |
627 | extent_range_clear_dirty_for_io(inode, start, end); | |
628 | redirty = 1; | |
629 | } | |
f51d2b59 DS |
630 | |
631 | /* Compression level is applied here and only here */ | |
632 | ret = btrfs_compress_pages( | |
633 | compress_type | (fs_info->compress_level << 4), | |
261507a0 | 634 | inode->i_mapping, start, |
38c31464 | 635 | pages, |
4d3a800e | 636 | &nr_pages, |
261507a0 | 637 | &total_in, |
e5d74902 | 638 | &total_compressed); |
c8b97818 CM |
639 | |
640 | if (!ret) { | |
7073017a | 641 | unsigned long offset = offset_in_page(total_compressed); |
4d3a800e | 642 | struct page *page = pages[nr_pages - 1]; |
c8b97818 CM |
643 | char *kaddr; |
644 | ||
645 | /* zero the tail end of the last page, we might be | |
646 | * sending it down to disk | |
647 | */ | |
648 | if (offset) { | |
7ac687d9 | 649 | kaddr = kmap_atomic(page); |
c8b97818 | 650 | memset(kaddr + offset, 0, |
09cbfeaf | 651 | PAGE_SIZE - offset); |
7ac687d9 | 652 | kunmap_atomic(kaddr); |
c8b97818 CM |
653 | } |
654 | will_compress = 1; | |
655 | } | |
656 | } | |
560f7d75 | 657 | cont: |
c8b97818 CM |
658 | if (start == 0) { |
659 | /* lets try to make an inline extent */ | |
6018ba0a | 660 | if (ret || total_in < actual_end) { |
c8b97818 | 661 | /* we didn't compress the entire range, try |
771ed689 | 662 | * to make an uncompressed inline extent. |
c8b97818 | 663 | */ |
a0349401 NB |
664 | ret = cow_file_range_inline(BTRFS_I(inode), start, end, |
665 | 0, BTRFS_COMPRESS_NONE, | |
666 | NULL); | |
c8b97818 | 667 | } else { |
771ed689 | 668 | /* try making a compressed inline extent */ |
a0349401 | 669 | ret = cow_file_range_inline(BTRFS_I(inode), start, end, |
fe3f566c LZ |
670 | total_compressed, |
671 | compress_type, pages); | |
c8b97818 | 672 | } |
79787eaa | 673 | if (ret <= 0) { |
151a41bc | 674 | unsigned long clear_flags = EXTENT_DELALLOC | |
8b62f87b JB |
675 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
676 | EXTENT_DO_ACCOUNTING; | |
e6eb4314 FM |
677 | unsigned long page_error_op; |
678 | ||
e6eb4314 | 679 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 680 | |
771ed689 | 681 | /* |
79787eaa JM |
682 | * inline extent creation worked or returned error, |
683 | * we don't need to create any more async work items. | |
684 | * Unlock and free up our temp pages. | |
8b62f87b JB |
685 | * |
686 | * We use DO_ACCOUNTING here because we need the | |
687 | * delalloc_release_metadata to be done _after_ we drop | |
688 | * our outstanding extent for clearing delalloc for this | |
689 | * range. | |
771ed689 | 690 | */ |
ad7ff17b NB |
691 | extent_clear_unlock_delalloc(BTRFS_I(inode), start, end, |
692 | NULL, | |
74e9194a | 693 | clear_flags, |
ba8b04c1 | 694 | PAGE_UNLOCK | |
c2790a2e JB |
695 | PAGE_CLEAR_DIRTY | |
696 | PAGE_SET_WRITEBACK | | |
e6eb4314 | 697 | page_error_op | |
c2790a2e | 698 | PAGE_END_WRITEBACK); |
cecc8d90 | 699 | |
1e6e238c QW |
700 | /* |
701 | * Ensure we only free the compressed pages if we have | |
702 | * them allocated, as we can still reach here with | |
703 | * inode_need_compress() == false. | |
704 | */ | |
705 | if (pages) { | |
706 | for (i = 0; i < nr_pages; i++) { | |
707 | WARN_ON(pages[i]->mapping); | |
708 | put_page(pages[i]); | |
709 | } | |
710 | kfree(pages); | |
cecc8d90 | 711 | } |
cecc8d90 | 712 | return 0; |
c8b97818 CM |
713 | } |
714 | } | |
715 | ||
716 | if (will_compress) { | |
717 | /* | |
718 | * we aren't doing an inline extent round the compressed size | |
719 | * up to a block size boundary so the allocator does sane | |
720 | * things | |
721 | */ | |
fda2832f | 722 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
723 | |
724 | /* | |
725 | * one last check to make sure the compression is really a | |
170607eb TT |
726 | * win, compare the page count read with the blocks on disk, |
727 | * compression must free at least one sector size | |
c8b97818 | 728 | */ |
09cbfeaf | 729 | total_in = ALIGN(total_in, PAGE_SIZE); |
170607eb | 730 | if (total_compressed + blocksize <= total_in) { |
ac3e9933 | 731 | compressed_extents++; |
c8bb0c8b AS |
732 | |
733 | /* | |
734 | * The async work queues will take care of doing actual | |
735 | * allocation on disk for these compressed pages, and | |
736 | * will submit them to the elevator. | |
737 | */ | |
b5326271 | 738 | add_async_extent(async_chunk, start, total_in, |
4d3a800e | 739 | total_compressed, pages, nr_pages, |
c8bb0c8b AS |
740 | compress_type); |
741 | ||
1170862d TT |
742 | if (start + total_in < end) { |
743 | start += total_in; | |
c8bb0c8b AS |
744 | pages = NULL; |
745 | cond_resched(); | |
746 | goto again; | |
747 | } | |
ac3e9933 | 748 | return compressed_extents; |
c8b97818 CM |
749 | } |
750 | } | |
c8bb0c8b | 751 | if (pages) { |
c8b97818 CM |
752 | /* |
753 | * the compression code ran but failed to make things smaller, | |
754 | * free any pages it allocated and our page pointer array | |
755 | */ | |
4d3a800e | 756 | for (i = 0; i < nr_pages; i++) { |
70b99e69 | 757 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 758 | put_page(pages[i]); |
c8b97818 CM |
759 | } |
760 | kfree(pages); | |
761 | pages = NULL; | |
762 | total_compressed = 0; | |
4d3a800e | 763 | nr_pages = 0; |
c8b97818 CM |
764 | |
765 | /* flag the file so we don't compress in the future */ | |
0b246afa | 766 | if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && |
b52aa8c9 | 767 | !(BTRFS_I(inode)->prop_compress)) { |
a555f810 | 768 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 769 | } |
c8b97818 | 770 | } |
f03d9301 | 771 | cleanup_and_bail_uncompressed: |
c8bb0c8b AS |
772 | /* |
773 | * No compression, but we still need to write the pages in the file | |
774 | * we've been given so far. redirty the locked page if it corresponds | |
775 | * to our extent and set things up for the async work queue to run | |
776 | * cow_file_range to do the normal delalloc dance. | |
777 | */ | |
1d53c9e6 CM |
778 | if (async_chunk->locked_page && |
779 | (page_offset(async_chunk->locked_page) >= start && | |
780 | page_offset(async_chunk->locked_page)) <= end) { | |
1368c6da | 781 | __set_page_dirty_nobuffers(async_chunk->locked_page); |
c8bb0c8b | 782 | /* unlocked later on in the async handlers */ |
1d53c9e6 | 783 | } |
c8bb0c8b AS |
784 | |
785 | if (redirty) | |
786 | extent_range_redirty_for_io(inode, start, end); | |
b5326271 | 787 | add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0, |
c8bb0c8b | 788 | BTRFS_COMPRESS_NONE); |
ac3e9933 | 789 | compressed_extents++; |
3b951516 | 790 | |
ac3e9933 | 791 | return compressed_extents; |
771ed689 | 792 | } |
771ed689 | 793 | |
40ae837b FM |
794 | static void free_async_extent_pages(struct async_extent *async_extent) |
795 | { | |
796 | int i; | |
797 | ||
798 | if (!async_extent->pages) | |
799 | return; | |
800 | ||
801 | for (i = 0; i < async_extent->nr_pages; i++) { | |
802 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 803 | put_page(async_extent->pages[i]); |
40ae837b FM |
804 | } |
805 | kfree(async_extent->pages); | |
806 | async_extent->nr_pages = 0; | |
807 | async_extent->pages = NULL; | |
771ed689 CM |
808 | } |
809 | ||
810 | /* | |
811 | * phase two of compressed writeback. This is the ordered portion | |
812 | * of the code, which only gets called in the order the work was | |
813 | * queued. We walk all the async extents created by compress_file_range | |
814 | * and send them down to the disk. | |
815 | */ | |
b5326271 | 816 | static noinline void submit_compressed_extents(struct async_chunk *async_chunk) |
771ed689 | 817 | { |
a0ff10dc NB |
818 | struct btrfs_inode *inode = BTRFS_I(async_chunk->inode); |
819 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
771ed689 CM |
820 | struct async_extent *async_extent; |
821 | u64 alloc_hint = 0; | |
771ed689 CM |
822 | struct btrfs_key ins; |
823 | struct extent_map *em; | |
a0ff10dc NB |
824 | struct btrfs_root *root = inode->root; |
825 | struct extent_io_tree *io_tree = &inode->io_tree; | |
f5a84ee3 | 826 | int ret = 0; |
771ed689 | 827 | |
3e04e7f1 | 828 | again: |
b5326271 NB |
829 | while (!list_empty(&async_chunk->extents)) { |
830 | async_extent = list_entry(async_chunk->extents.next, | |
771ed689 CM |
831 | struct async_extent, list); |
832 | list_del(&async_extent->list); | |
c8b97818 | 833 | |
f5a84ee3 | 834 | retry: |
7447555f NB |
835 | lock_extent(io_tree, async_extent->start, |
836 | async_extent->start + async_extent->ram_size - 1); | |
771ed689 CM |
837 | /* did the compression code fall back to uncompressed IO? */ |
838 | if (!async_extent->pages) { | |
839 | int page_started = 0; | |
840 | unsigned long nr_written = 0; | |
841 | ||
771ed689 | 842 | /* allocate blocks */ |
a0ff10dc | 843 | ret = cow_file_range(inode, async_chunk->locked_page, |
f5a84ee3 JB |
844 | async_extent->start, |
845 | async_extent->start + | |
846 | async_extent->ram_size - 1, | |
330a5827 | 847 | &page_started, &nr_written, 0); |
771ed689 | 848 | |
79787eaa JM |
849 | /* JDM XXX */ |
850 | ||
771ed689 CM |
851 | /* |
852 | * if page_started, cow_file_range inserted an | |
853 | * inline extent and took care of all the unlocking | |
854 | * and IO for us. Otherwise, we need to submit | |
855 | * all those pages down to the drive. | |
856 | */ | |
f5a84ee3 | 857 | if (!page_started && !ret) |
a0ff10dc | 858 | extent_write_locked_range(&inode->vfs_inode, |
5e3ee236 | 859 | async_extent->start, |
d397712b | 860 | async_extent->start + |
771ed689 | 861 | async_extent->ram_size - 1, |
771ed689 | 862 | WB_SYNC_ALL); |
1d53c9e6 | 863 | else if (ret && async_chunk->locked_page) |
b5326271 | 864 | unlock_page(async_chunk->locked_page); |
771ed689 CM |
865 | kfree(async_extent); |
866 | cond_resched(); | |
867 | continue; | |
868 | } | |
869 | ||
18513091 | 870 | ret = btrfs_reserve_extent(root, async_extent->ram_size, |
771ed689 CM |
871 | async_extent->compressed_size, |
872 | async_extent->compressed_size, | |
e570fd27 | 873 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 874 | if (ret) { |
40ae837b | 875 | free_async_extent_pages(async_extent); |
3e04e7f1 | 876 | |
fdf8e2ea JB |
877 | if (ret == -ENOSPC) { |
878 | unlock_extent(io_tree, async_extent->start, | |
879 | async_extent->start + | |
880 | async_extent->ram_size - 1); | |
ce62003f LB |
881 | |
882 | /* | |
883 | * we need to redirty the pages if we decide to | |
884 | * fallback to uncompressed IO, otherwise we | |
885 | * will not submit these pages down to lower | |
886 | * layers. | |
887 | */ | |
a0ff10dc | 888 | extent_range_redirty_for_io(&inode->vfs_inode, |
ce62003f LB |
889 | async_extent->start, |
890 | async_extent->start + | |
891 | async_extent->ram_size - 1); | |
892 | ||
79787eaa | 893 | goto retry; |
fdf8e2ea | 894 | } |
3e04e7f1 | 895 | goto out_free; |
f5a84ee3 | 896 | } |
c2167754 YZ |
897 | /* |
898 | * here we're doing allocation and writeback of the | |
899 | * compressed pages | |
900 | */ | |
a0ff10dc | 901 | em = create_io_em(inode, async_extent->start, |
6f9994db LB |
902 | async_extent->ram_size, /* len */ |
903 | async_extent->start, /* orig_start */ | |
904 | ins.objectid, /* block_start */ | |
905 | ins.offset, /* block_len */ | |
906 | ins.offset, /* orig_block_len */ | |
907 | async_extent->ram_size, /* ram_bytes */ | |
908 | async_extent->compress_type, | |
909 | BTRFS_ORDERED_COMPRESSED); | |
910 | if (IS_ERR(em)) | |
911 | /* ret value is not necessary due to void function */ | |
3e04e7f1 | 912 | goto out_free_reserve; |
6f9994db | 913 | free_extent_map(em); |
3e04e7f1 | 914 | |
a0ff10dc | 915 | ret = btrfs_add_ordered_extent_compress(inode, |
261507a0 LZ |
916 | async_extent->start, |
917 | ins.objectid, | |
918 | async_extent->ram_size, | |
919 | ins.offset, | |
920 | BTRFS_ORDERED_COMPRESSED, | |
921 | async_extent->compress_type); | |
d9f85963 | 922 | if (ret) { |
a0ff10dc | 923 | btrfs_drop_extent_cache(inode, async_extent->start, |
d9f85963 FM |
924 | async_extent->start + |
925 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 926 | goto out_free_reserve; |
d9f85963 | 927 | } |
0b246afa | 928 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
771ed689 | 929 | |
771ed689 CM |
930 | /* |
931 | * clear dirty, set writeback and unlock the pages. | |
932 | */ | |
a0ff10dc | 933 | extent_clear_unlock_delalloc(inode, async_extent->start, |
a791e35e CM |
934 | async_extent->start + |
935 | async_extent->ram_size - 1, | |
151a41bc JB |
936 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
937 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 938 | PAGE_SET_WRITEBACK); |
a0ff10dc | 939 | if (btrfs_submit_compressed_write(inode, async_extent->start, |
d397712b CM |
940 | async_extent->ram_size, |
941 | ins.objectid, | |
942 | ins.offset, async_extent->pages, | |
f82b7359 | 943 | async_extent->nr_pages, |
ec39f769 CM |
944 | async_chunk->write_flags, |
945 | async_chunk->blkcg_css)) { | |
fce2a4e6 FM |
946 | struct page *p = async_extent->pages[0]; |
947 | const u64 start = async_extent->start; | |
948 | const u64 end = start + async_extent->ram_size - 1; | |
949 | ||
a0ff10dc | 950 | p->mapping = inode->vfs_inode.i_mapping; |
c629732d | 951 | btrfs_writepage_endio_finish_ordered(p, start, end, 0); |
7087a9d8 | 952 | |
fce2a4e6 | 953 | p->mapping = NULL; |
a0ff10dc | 954 | extent_clear_unlock_delalloc(inode, start, end, NULL, 0, |
fce2a4e6 FM |
955 | PAGE_END_WRITEBACK | |
956 | PAGE_SET_ERROR); | |
40ae837b | 957 | free_async_extent_pages(async_extent); |
fce2a4e6 | 958 | } |
771ed689 CM |
959 | alloc_hint = ins.objectid + ins.offset; |
960 | kfree(async_extent); | |
961 | cond_resched(); | |
962 | } | |
dec8f175 | 963 | return; |
3e04e7f1 | 964 | out_free_reserve: |
0b246afa | 965 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 966 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 967 | out_free: |
a0ff10dc | 968 | extent_clear_unlock_delalloc(inode, async_extent->start, |
3e04e7f1 JB |
969 | async_extent->start + |
970 | async_extent->ram_size - 1, | |
c2790a2e | 971 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
a7e3b975 | 972 | EXTENT_DELALLOC_NEW | |
151a41bc JB |
973 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
974 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
975 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
976 | PAGE_SET_ERROR); | |
40ae837b | 977 | free_async_extent_pages(async_extent); |
79787eaa | 978 | kfree(async_extent); |
3e04e7f1 | 979 | goto again; |
771ed689 CM |
980 | } |
981 | ||
43c69849 | 982 | static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start, |
4b46fce2 JB |
983 | u64 num_bytes) |
984 | { | |
43c69849 | 985 | struct extent_map_tree *em_tree = &inode->extent_tree; |
4b46fce2 JB |
986 | struct extent_map *em; |
987 | u64 alloc_hint = 0; | |
988 | ||
989 | read_lock(&em_tree->lock); | |
990 | em = search_extent_mapping(em_tree, start, num_bytes); | |
991 | if (em) { | |
992 | /* | |
993 | * if block start isn't an actual block number then find the | |
994 | * first block in this inode and use that as a hint. If that | |
995 | * block is also bogus then just don't worry about it. | |
996 | */ | |
997 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
998 | free_extent_map(em); | |
999 | em = search_extent_mapping(em_tree, 0, 0); | |
1000 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
1001 | alloc_hint = em->block_start; | |
1002 | if (em) | |
1003 | free_extent_map(em); | |
1004 | } else { | |
1005 | alloc_hint = em->block_start; | |
1006 | free_extent_map(em); | |
1007 | } | |
1008 | } | |
1009 | read_unlock(&em_tree->lock); | |
1010 | ||
1011 | return alloc_hint; | |
1012 | } | |
1013 | ||
771ed689 CM |
1014 | /* |
1015 | * when extent_io.c finds a delayed allocation range in the file, | |
1016 | * the call backs end up in this code. The basic idea is to | |
1017 | * allocate extents on disk for the range, and create ordered data structs | |
1018 | * in ram to track those extents. | |
1019 | * | |
1020 | * locked_page is the page that writepage had locked already. We use | |
1021 | * it to make sure we don't do extra locks or unlocks. | |
1022 | * | |
1023 | * *page_started is set to one if we unlock locked_page and do everything | |
1024 | * required to start IO on it. It may be clean and already done with | |
1025 | * IO when we return. | |
1026 | */ | |
6e26c442 | 1027 | static noinline int cow_file_range(struct btrfs_inode *inode, |
00361589 | 1028 | struct page *locked_page, |
74e9194a | 1029 | u64 start, u64 end, int *page_started, |
330a5827 | 1030 | unsigned long *nr_written, int unlock) |
771ed689 | 1031 | { |
6e26c442 NB |
1032 | struct btrfs_root *root = inode->root; |
1033 | struct btrfs_fs_info *fs_info = root->fs_info; | |
771ed689 CM |
1034 | u64 alloc_hint = 0; |
1035 | u64 num_bytes; | |
1036 | unsigned long ram_size; | |
a315e68f | 1037 | u64 cur_alloc_size = 0; |
432cd2a1 | 1038 | u64 min_alloc_size; |
0b246afa | 1039 | u64 blocksize = fs_info->sectorsize; |
771ed689 CM |
1040 | struct btrfs_key ins; |
1041 | struct extent_map *em; | |
a315e68f FM |
1042 | unsigned clear_bits; |
1043 | unsigned long page_ops; | |
1044 | bool extent_reserved = false; | |
771ed689 CM |
1045 | int ret = 0; |
1046 | ||
6e26c442 | 1047 | if (btrfs_is_free_space_inode(inode)) { |
02ecd2c2 | 1048 | WARN_ON_ONCE(1); |
29bce2f3 JB |
1049 | ret = -EINVAL; |
1050 | goto out_unlock; | |
02ecd2c2 | 1051 | } |
771ed689 | 1052 | |
fda2832f | 1053 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 | 1054 | num_bytes = max(blocksize, num_bytes); |
566b1760 | 1055 | ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy)); |
771ed689 | 1056 | |
6e26c442 | 1057 | inode_should_defrag(inode, start, end, num_bytes, SZ_64K); |
4cb5300b | 1058 | |
771ed689 CM |
1059 | if (start == 0) { |
1060 | /* lets try to make an inline extent */ | |
6e26c442 | 1061 | ret = cow_file_range_inline(inode, start, end, 0, |
d02c0e20 | 1062 | BTRFS_COMPRESS_NONE, NULL); |
771ed689 | 1063 | if (ret == 0) { |
8b62f87b JB |
1064 | /* |
1065 | * We use DO_ACCOUNTING here because we need the | |
1066 | * delalloc_release_metadata to be run _after_ we drop | |
1067 | * our outstanding extent for clearing delalloc for this | |
1068 | * range. | |
1069 | */ | |
6e26c442 | 1070 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
c2790a2e | 1071 | EXTENT_LOCKED | EXTENT_DELALLOC | |
8b62f87b JB |
1072 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
1073 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
c2790a2e JB |
1074 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
1075 | PAGE_END_WRITEBACK); | |
771ed689 | 1076 | *nr_written = *nr_written + |
09cbfeaf | 1077 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 1078 | *page_started = 1; |
771ed689 | 1079 | goto out; |
79787eaa | 1080 | } else if (ret < 0) { |
79787eaa | 1081 | goto out_unlock; |
771ed689 CM |
1082 | } |
1083 | } | |
1084 | ||
6e26c442 NB |
1085 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
1086 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); | |
771ed689 | 1087 | |
432cd2a1 FM |
1088 | /* |
1089 | * Relocation relies on the relocated extents to have exactly the same | |
1090 | * size as the original extents. Normally writeback for relocation data | |
1091 | * extents follows a NOCOW path because relocation preallocates the | |
1092 | * extents. However, due to an operation such as scrub turning a block | |
1093 | * group to RO mode, it may fallback to COW mode, so we must make sure | |
1094 | * an extent allocated during COW has exactly the requested size and can | |
1095 | * not be split into smaller extents, otherwise relocation breaks and | |
1096 | * fails during the stage where it updates the bytenr of file extent | |
1097 | * items. | |
1098 | */ | |
1099 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1100 | min_alloc_size = num_bytes; | |
1101 | else | |
1102 | min_alloc_size = fs_info->sectorsize; | |
1103 | ||
3752d22f AJ |
1104 | while (num_bytes > 0) { |
1105 | cur_alloc_size = num_bytes; | |
18513091 | 1106 | ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size, |
432cd2a1 | 1107 | min_alloc_size, 0, alloc_hint, |
e570fd27 | 1108 | &ins, 1, 1); |
00361589 | 1109 | if (ret < 0) |
79787eaa | 1110 | goto out_unlock; |
a315e68f FM |
1111 | cur_alloc_size = ins.offset; |
1112 | extent_reserved = true; | |
d397712b | 1113 | |
771ed689 | 1114 | ram_size = ins.offset; |
6e26c442 | 1115 | em = create_io_em(inode, start, ins.offset, /* len */ |
6f9994db LB |
1116 | start, /* orig_start */ |
1117 | ins.objectid, /* block_start */ | |
1118 | ins.offset, /* block_len */ | |
1119 | ins.offset, /* orig_block_len */ | |
1120 | ram_size, /* ram_bytes */ | |
1121 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
1af4a0aa | 1122 | BTRFS_ORDERED_REGULAR /* type */); |
090a127a SY |
1123 | if (IS_ERR(em)) { |
1124 | ret = PTR_ERR(em); | |
ace68bac | 1125 | goto out_reserve; |
090a127a | 1126 | } |
6f9994db | 1127 | free_extent_map(em); |
e6dcd2dc | 1128 | |
6e26c442 NB |
1129 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
1130 | ram_size, cur_alloc_size, 0); | |
ace68bac | 1131 | if (ret) |
d9f85963 | 1132 | goto out_drop_extent_cache; |
c8b97818 | 1133 | |
17d217fe YZ |
1134 | if (root->root_key.objectid == |
1135 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
6e26c442 | 1136 | ret = btrfs_reloc_clone_csums(inode, start, |
17d217fe | 1137 | cur_alloc_size); |
4dbd80fb QW |
1138 | /* |
1139 | * Only drop cache here, and process as normal. | |
1140 | * | |
1141 | * We must not allow extent_clear_unlock_delalloc() | |
1142 | * at out_unlock label to free meta of this ordered | |
1143 | * extent, as its meta should be freed by | |
1144 | * btrfs_finish_ordered_io(). | |
1145 | * | |
1146 | * So we must continue until @start is increased to | |
1147 | * skip current ordered extent. | |
1148 | */ | |
00361589 | 1149 | if (ret) |
6e26c442 | 1150 | btrfs_drop_extent_cache(inode, start, |
4dbd80fb | 1151 | start + ram_size - 1, 0); |
17d217fe YZ |
1152 | } |
1153 | ||
0b246afa | 1154 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
9cfa3e34 | 1155 | |
c8b97818 CM |
1156 | /* we're not doing compressed IO, don't unlock the first |
1157 | * page (which the caller expects to stay locked), don't | |
1158 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1159 | * |
1160 | * Do set the Private2 bit so we know this page was properly | |
1161 | * setup for writepage | |
c8b97818 | 1162 | */ |
a315e68f FM |
1163 | page_ops = unlock ? PAGE_UNLOCK : 0; |
1164 | page_ops |= PAGE_SET_PRIVATE2; | |
a791e35e | 1165 | |
6e26c442 | 1166 | extent_clear_unlock_delalloc(inode, start, start + ram_size - 1, |
74e9194a | 1167 | locked_page, |
c2790a2e | 1168 | EXTENT_LOCKED | EXTENT_DELALLOC, |
a315e68f | 1169 | page_ops); |
3752d22f AJ |
1170 | if (num_bytes < cur_alloc_size) |
1171 | num_bytes = 0; | |
4dbd80fb | 1172 | else |
3752d22f | 1173 | num_bytes -= cur_alloc_size; |
c59f8951 CM |
1174 | alloc_hint = ins.objectid + ins.offset; |
1175 | start += cur_alloc_size; | |
a315e68f | 1176 | extent_reserved = false; |
4dbd80fb QW |
1177 | |
1178 | /* | |
1179 | * btrfs_reloc_clone_csums() error, since start is increased | |
1180 | * extent_clear_unlock_delalloc() at out_unlock label won't | |
1181 | * free metadata of current ordered extent, we're OK to exit. | |
1182 | */ | |
1183 | if (ret) | |
1184 | goto out_unlock; | |
b888db2b | 1185 | } |
79787eaa | 1186 | out: |
be20aa9d | 1187 | return ret; |
b7d5b0a8 | 1188 | |
d9f85963 | 1189 | out_drop_extent_cache: |
6e26c442 | 1190 | btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0); |
ace68bac | 1191 | out_reserve: |
0b246afa | 1192 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1193 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1194 | out_unlock: |
a7e3b975 FM |
1195 | clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
1196 | EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV; | |
a315e68f FM |
1197 | page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
1198 | PAGE_END_WRITEBACK; | |
1199 | /* | |
1200 | * If we reserved an extent for our delalloc range (or a subrange) and | |
1201 | * failed to create the respective ordered extent, then it means that | |
1202 | * when we reserved the extent we decremented the extent's size from | |
1203 | * the data space_info's bytes_may_use counter and incremented the | |
1204 | * space_info's bytes_reserved counter by the same amount. We must make | |
1205 | * sure extent_clear_unlock_delalloc() does not try to decrement again | |
1206 | * the data space_info's bytes_may_use counter, therefore we do not pass | |
1207 | * it the flag EXTENT_CLEAR_DATA_RESV. | |
1208 | */ | |
1209 | if (extent_reserved) { | |
6e26c442 | 1210 | extent_clear_unlock_delalloc(inode, start, |
e2c8e92d | 1211 | start + cur_alloc_size - 1, |
a315e68f FM |
1212 | locked_page, |
1213 | clear_bits, | |
1214 | page_ops); | |
1215 | start += cur_alloc_size; | |
1216 | if (start >= end) | |
1217 | goto out; | |
1218 | } | |
6e26c442 | 1219 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
a315e68f FM |
1220 | clear_bits | EXTENT_CLEAR_DATA_RESV, |
1221 | page_ops); | |
79787eaa | 1222 | goto out; |
771ed689 | 1223 | } |
c8b97818 | 1224 | |
771ed689 CM |
1225 | /* |
1226 | * work queue call back to started compression on a file and pages | |
1227 | */ | |
1228 | static noinline void async_cow_start(struct btrfs_work *work) | |
1229 | { | |
b5326271 | 1230 | struct async_chunk *async_chunk; |
ac3e9933 | 1231 | int compressed_extents; |
771ed689 | 1232 | |
b5326271 | 1233 | async_chunk = container_of(work, struct async_chunk, work); |
771ed689 | 1234 | |
ac3e9933 NB |
1235 | compressed_extents = compress_file_range(async_chunk); |
1236 | if (compressed_extents == 0) { | |
b5326271 NB |
1237 | btrfs_add_delayed_iput(async_chunk->inode); |
1238 | async_chunk->inode = NULL; | |
8180ef88 | 1239 | } |
771ed689 CM |
1240 | } |
1241 | ||
1242 | /* | |
1243 | * work queue call back to submit previously compressed pages | |
1244 | */ | |
1245 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1246 | { | |
c5a68aec NB |
1247 | struct async_chunk *async_chunk = container_of(work, struct async_chunk, |
1248 | work); | |
1249 | struct btrfs_fs_info *fs_info = btrfs_work_owner(work); | |
771ed689 CM |
1250 | unsigned long nr_pages; |
1251 | ||
b5326271 | 1252 | nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >> |
09cbfeaf | 1253 | PAGE_SHIFT; |
771ed689 | 1254 | |
093258e6 | 1255 | /* atomic_sub_return implies a barrier */ |
0b246afa | 1256 | if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) < |
093258e6 DS |
1257 | 5 * SZ_1M) |
1258 | cond_wake_up_nomb(&fs_info->async_submit_wait); | |
771ed689 | 1259 | |
4546d178 | 1260 | /* |
b5326271 | 1261 | * ->inode could be NULL if async_chunk_start has failed to compress, |
4546d178 NB |
1262 | * in which case we don't have anything to submit, yet we need to |
1263 | * always adjust ->async_delalloc_pages as its paired with the init | |
1264 | * happening in cow_file_range_async | |
1265 | */ | |
b5326271 NB |
1266 | if (async_chunk->inode) |
1267 | submit_compressed_extents(async_chunk); | |
771ed689 | 1268 | } |
c8b97818 | 1269 | |
771ed689 CM |
1270 | static noinline void async_cow_free(struct btrfs_work *work) |
1271 | { | |
b5326271 | 1272 | struct async_chunk *async_chunk; |
97db1204 | 1273 | |
b5326271 NB |
1274 | async_chunk = container_of(work, struct async_chunk, work); |
1275 | if (async_chunk->inode) | |
1276 | btrfs_add_delayed_iput(async_chunk->inode); | |
ec39f769 CM |
1277 | if (async_chunk->blkcg_css) |
1278 | css_put(async_chunk->blkcg_css); | |
97db1204 NB |
1279 | /* |
1280 | * Since the pointer to 'pending' is at the beginning of the array of | |
b5326271 | 1281 | * async_chunk's, freeing it ensures the whole array has been freed. |
97db1204 | 1282 | */ |
b5326271 | 1283 | if (atomic_dec_and_test(async_chunk->pending)) |
b1c16ac9 | 1284 | kvfree(async_chunk->pending); |
771ed689 CM |
1285 | } |
1286 | ||
751b6431 | 1287 | static int cow_file_range_async(struct btrfs_inode *inode, |
ec39f769 CM |
1288 | struct writeback_control *wbc, |
1289 | struct page *locked_page, | |
771ed689 | 1290 | u64 start, u64 end, int *page_started, |
fac07d2b | 1291 | unsigned long *nr_written) |
771ed689 | 1292 | { |
751b6431 | 1293 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
ec39f769 | 1294 | struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc); |
97db1204 NB |
1295 | struct async_cow *ctx; |
1296 | struct async_chunk *async_chunk; | |
771ed689 CM |
1297 | unsigned long nr_pages; |
1298 | u64 cur_end; | |
97db1204 NB |
1299 | u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K); |
1300 | int i; | |
1301 | bool should_compress; | |
b1c16ac9 | 1302 | unsigned nofs_flag; |
fac07d2b | 1303 | const unsigned int write_flags = wbc_to_write_flags(wbc); |
771ed689 | 1304 | |
751b6431 | 1305 | unlock_extent(&inode->io_tree, start, end); |
97db1204 | 1306 | |
751b6431 | 1307 | if (inode->flags & BTRFS_INODE_NOCOMPRESS && |
97db1204 NB |
1308 | !btrfs_test_opt(fs_info, FORCE_COMPRESS)) { |
1309 | num_chunks = 1; | |
1310 | should_compress = false; | |
1311 | } else { | |
1312 | should_compress = true; | |
1313 | } | |
1314 | ||
b1c16ac9 NB |
1315 | nofs_flag = memalloc_nofs_save(); |
1316 | ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL); | |
1317 | memalloc_nofs_restore(nofs_flag); | |
1318 | ||
97db1204 NB |
1319 | if (!ctx) { |
1320 | unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | | |
1321 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | | |
1322 | EXTENT_DO_ACCOUNTING; | |
1323 | unsigned long page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
1324 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | | |
1325 | PAGE_SET_ERROR; | |
1326 | ||
751b6431 NB |
1327 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1328 | clear_bits, page_ops); | |
97db1204 NB |
1329 | return -ENOMEM; |
1330 | } | |
1331 | ||
1332 | async_chunk = ctx->chunks; | |
1333 | atomic_set(&ctx->num_chunks, num_chunks); | |
1334 | ||
1335 | for (i = 0; i < num_chunks; i++) { | |
1336 | if (should_compress) | |
1337 | cur_end = min(end, start + SZ_512K - 1); | |
1338 | else | |
1339 | cur_end = end; | |
771ed689 | 1340 | |
bd4691a0 NB |
1341 | /* |
1342 | * igrab is called higher up in the call chain, take only the | |
1343 | * lightweight reference for the callback lifetime | |
1344 | */ | |
751b6431 | 1345 | ihold(&inode->vfs_inode); |
97db1204 | 1346 | async_chunk[i].pending = &ctx->num_chunks; |
751b6431 | 1347 | async_chunk[i].inode = &inode->vfs_inode; |
97db1204 NB |
1348 | async_chunk[i].start = start; |
1349 | async_chunk[i].end = cur_end; | |
97db1204 NB |
1350 | async_chunk[i].write_flags = write_flags; |
1351 | INIT_LIST_HEAD(&async_chunk[i].extents); | |
1352 | ||
1d53c9e6 CM |
1353 | /* |
1354 | * The locked_page comes all the way from writepage and its | |
1355 | * the original page we were actually given. As we spread | |
1356 | * this large delalloc region across multiple async_chunk | |
1357 | * structs, only the first struct needs a pointer to locked_page | |
1358 | * | |
1359 | * This way we don't need racey decisions about who is supposed | |
1360 | * to unlock it. | |
1361 | */ | |
1362 | if (locked_page) { | |
ec39f769 CM |
1363 | /* |
1364 | * Depending on the compressibility, the pages might or | |
1365 | * might not go through async. We want all of them to | |
1366 | * be accounted against wbc once. Let's do it here | |
1367 | * before the paths diverge. wbc accounting is used | |
1368 | * only for foreign writeback detection and doesn't | |
1369 | * need full accuracy. Just account the whole thing | |
1370 | * against the first page. | |
1371 | */ | |
1372 | wbc_account_cgroup_owner(wbc, locked_page, | |
1373 | cur_end - start); | |
1d53c9e6 CM |
1374 | async_chunk[i].locked_page = locked_page; |
1375 | locked_page = NULL; | |
1376 | } else { | |
1377 | async_chunk[i].locked_page = NULL; | |
1378 | } | |
1379 | ||
ec39f769 CM |
1380 | if (blkcg_css != blkcg_root_css) { |
1381 | css_get(blkcg_css); | |
1382 | async_chunk[i].blkcg_css = blkcg_css; | |
1383 | } else { | |
1384 | async_chunk[i].blkcg_css = NULL; | |
1385 | } | |
1386 | ||
a0cac0ec OS |
1387 | btrfs_init_work(&async_chunk[i].work, async_cow_start, |
1388 | async_cow_submit, async_cow_free); | |
771ed689 | 1389 | |
97db1204 | 1390 | nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE); |
0b246afa | 1391 | atomic_add(nr_pages, &fs_info->async_delalloc_pages); |
771ed689 | 1392 | |
97db1204 | 1393 | btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work); |
771ed689 | 1394 | |
771ed689 CM |
1395 | *nr_written += nr_pages; |
1396 | start = cur_end + 1; | |
1397 | } | |
1398 | *page_started = 1; | |
1399 | return 0; | |
be20aa9d CM |
1400 | } |
1401 | ||
2ff7e61e | 1402 | static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, |
17d217fe YZ |
1403 | u64 bytenr, u64 num_bytes) |
1404 | { | |
1405 | int ret; | |
1406 | struct btrfs_ordered_sum *sums; | |
1407 | LIST_HEAD(list); | |
1408 | ||
0b246afa | 1409 | ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr, |
a2de733c | 1410 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1411 | if (ret == 0 && list_empty(&list)) |
1412 | return 0; | |
1413 | ||
1414 | while (!list_empty(&list)) { | |
1415 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1416 | list_del(&sums->list); | |
1417 | kfree(sums); | |
1418 | } | |
58113753 LB |
1419 | if (ret < 0) |
1420 | return ret; | |
17d217fe YZ |
1421 | return 1; |
1422 | } | |
1423 | ||
8ba96f3d | 1424 | static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page, |
467dc47e FM |
1425 | const u64 start, const u64 end, |
1426 | int *page_started, unsigned long *nr_written) | |
1427 | { | |
8ba96f3d NB |
1428 | const bool is_space_ino = btrfs_is_free_space_inode(inode); |
1429 | const bool is_reloc_ino = (inode->root->root_key.objectid == | |
6bd335b4 | 1430 | BTRFS_DATA_RELOC_TREE_OBJECTID); |
2166e5ed | 1431 | const u64 range_bytes = end + 1 - start; |
8ba96f3d | 1432 | struct extent_io_tree *io_tree = &inode->io_tree; |
467dc47e FM |
1433 | u64 range_start = start; |
1434 | u64 count; | |
1435 | ||
1436 | /* | |
1437 | * If EXTENT_NORESERVE is set it means that when the buffered write was | |
1438 | * made we had not enough available data space and therefore we did not | |
1439 | * reserve data space for it, since we though we could do NOCOW for the | |
1440 | * respective file range (either there is prealloc extent or the inode | |
1441 | * has the NOCOW bit set). | |
1442 | * | |
1443 | * However when we need to fallback to COW mode (because for example the | |
1444 | * block group for the corresponding extent was turned to RO mode by a | |
1445 | * scrub or relocation) we need to do the following: | |
1446 | * | |
1447 | * 1) We increment the bytes_may_use counter of the data space info. | |
1448 | * If COW succeeds, it allocates a new data extent and after doing | |
1449 | * that it decrements the space info's bytes_may_use counter and | |
1450 | * increments its bytes_reserved counter by the same amount (we do | |
1451 | * this at btrfs_add_reserved_bytes()). So we need to increment the | |
1452 | * bytes_may_use counter to compensate (when space is reserved at | |
1453 | * buffered write time, the bytes_may_use counter is incremented); | |
1454 | * | |
1455 | * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so | |
1456 | * that if the COW path fails for any reason, it decrements (through | |
1457 | * extent_clear_unlock_delalloc()) the bytes_may_use counter of the | |
1458 | * data space info, which we incremented in the step above. | |
2166e5ed FM |
1459 | * |
1460 | * If we need to fallback to cow and the inode corresponds to a free | |
6bd335b4 FM |
1461 | * space cache inode or an inode of the data relocation tree, we must |
1462 | * also increment bytes_may_use of the data space_info for the same | |
1463 | * reason. Space caches and relocated data extents always get a prealloc | |
2166e5ed | 1464 | * extent for them, however scrub or balance may have set the block |
6bd335b4 FM |
1465 | * group that contains that extent to RO mode and therefore force COW |
1466 | * when starting writeback. | |
467dc47e | 1467 | */ |
2166e5ed | 1468 | count = count_range_bits(io_tree, &range_start, end, range_bytes, |
467dc47e | 1469 | EXTENT_NORESERVE, 0); |
6bd335b4 FM |
1470 | if (count > 0 || is_space_ino || is_reloc_ino) { |
1471 | u64 bytes = count; | |
8ba96f3d | 1472 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
467dc47e FM |
1473 | struct btrfs_space_info *sinfo = fs_info->data_sinfo; |
1474 | ||
6bd335b4 FM |
1475 | if (is_space_ino || is_reloc_ino) |
1476 | bytes = range_bytes; | |
1477 | ||
467dc47e | 1478 | spin_lock(&sinfo->lock); |
2166e5ed | 1479 | btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes); |
467dc47e FM |
1480 | spin_unlock(&sinfo->lock); |
1481 | ||
2166e5ed FM |
1482 | if (count > 0) |
1483 | clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE, | |
1484 | 0, 0, NULL); | |
467dc47e FM |
1485 | } |
1486 | ||
8ba96f3d NB |
1487 | return cow_file_range(inode, locked_page, start, end, page_started, |
1488 | nr_written, 1); | |
467dc47e FM |
1489 | } |
1490 | ||
d352ac68 CM |
1491 | /* |
1492 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1493 | * of the extents that exist in the file, and COWs the file as required. | |
1494 | * | |
1495 | * If no cow copies or snapshots exist, we write directly to the existing | |
1496 | * blocks on disk | |
1497 | */ | |
968322c8 | 1498 | static noinline int run_delalloc_nocow(struct btrfs_inode *inode, |
7f366cfe | 1499 | struct page *locked_page, |
3e024846 NB |
1500 | const u64 start, const u64 end, |
1501 | int *page_started, int force, | |
1502 | unsigned long *nr_written) | |
be20aa9d | 1503 | { |
968322c8 NB |
1504 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
1505 | struct btrfs_root *root = inode->root; | |
be20aa9d | 1506 | struct btrfs_path *path; |
3e024846 NB |
1507 | u64 cow_start = (u64)-1; |
1508 | u64 cur_offset = start; | |
8ecebf4d | 1509 | int ret; |
3e024846 | 1510 | bool check_prev = true; |
968322c8 NB |
1511 | const bool freespace_inode = btrfs_is_free_space_inode(inode); |
1512 | u64 ino = btrfs_ino(inode); | |
762bf098 NB |
1513 | bool nocow = false; |
1514 | u64 disk_bytenr = 0; | |
be20aa9d CM |
1515 | |
1516 | path = btrfs_alloc_path(); | |
17ca04af | 1517 | if (!path) { |
968322c8 | 1518 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
c2790a2e | 1519 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
1520 | EXTENT_DO_ACCOUNTING | |
1521 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1522 | PAGE_CLEAR_DIRTY | |
1523 | PAGE_SET_WRITEBACK | | |
1524 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1525 | return -ENOMEM; |
17ca04af | 1526 | } |
82d5902d | 1527 | |
80ff3856 | 1528 | while (1) { |
3e024846 NB |
1529 | struct btrfs_key found_key; |
1530 | struct btrfs_file_extent_item *fi; | |
1531 | struct extent_buffer *leaf; | |
1532 | u64 extent_end; | |
1533 | u64 extent_offset; | |
3e024846 NB |
1534 | u64 num_bytes = 0; |
1535 | u64 disk_num_bytes; | |
3e024846 NB |
1536 | u64 ram_bytes; |
1537 | int extent_type; | |
762bf098 NB |
1538 | |
1539 | nocow = false; | |
3e024846 | 1540 | |
e4c3b2dc | 1541 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, |
80ff3856 | 1542 | cur_offset, 0); |
d788a349 | 1543 | if (ret < 0) |
79787eaa | 1544 | goto error; |
a6bd9cd1 NB |
1545 | |
1546 | /* | |
1547 | * If there is no extent for our range when doing the initial | |
1548 | * search, then go back to the previous slot as it will be the | |
1549 | * one containing the search offset | |
1550 | */ | |
80ff3856 YZ |
1551 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1552 | leaf = path->nodes[0]; | |
1553 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1554 | path->slots[0] - 1); | |
33345d01 | 1555 | if (found_key.objectid == ino && |
80ff3856 YZ |
1556 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1557 | path->slots[0]--; | |
1558 | } | |
3e024846 | 1559 | check_prev = false; |
80ff3856 | 1560 | next_slot: |
a6bd9cd1 | 1561 | /* Go to next leaf if we have exhausted the current one */ |
80ff3856 YZ |
1562 | leaf = path->nodes[0]; |
1563 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1564 | ret = btrfs_next_leaf(root, path); | |
e8916699 LB |
1565 | if (ret < 0) { |
1566 | if (cow_start != (u64)-1) | |
1567 | cur_offset = cow_start; | |
79787eaa | 1568 | goto error; |
e8916699 | 1569 | } |
80ff3856 YZ |
1570 | if (ret > 0) |
1571 | break; | |
1572 | leaf = path->nodes[0]; | |
1573 | } | |
be20aa9d | 1574 | |
80ff3856 YZ |
1575 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1576 | ||
a6bd9cd1 | 1577 | /* Didn't find anything for our INO */ |
1d512cb7 FM |
1578 | if (found_key.objectid > ino) |
1579 | break; | |
a6bd9cd1 NB |
1580 | /* |
1581 | * Keep searching until we find an EXTENT_ITEM or there are no | |
1582 | * more extents for this inode | |
1583 | */ | |
1d512cb7 FM |
1584 | if (WARN_ON_ONCE(found_key.objectid < ino) || |
1585 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1586 | path->slots[0]++; | |
1587 | goto next_slot; | |
1588 | } | |
a6bd9cd1 NB |
1589 | |
1590 | /* Found key is not EXTENT_DATA_KEY or starts after req range */ | |
1d512cb7 | 1591 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || |
80ff3856 YZ |
1592 | found_key.offset > end) |
1593 | break; | |
1594 | ||
a6bd9cd1 NB |
1595 | /* |
1596 | * If the found extent starts after requested offset, then | |
1597 | * adjust extent_end to be right before this extent begins | |
1598 | */ | |
80ff3856 YZ |
1599 | if (found_key.offset > cur_offset) { |
1600 | extent_end = found_key.offset; | |
e9061e21 | 1601 | extent_type = 0; |
80ff3856 YZ |
1602 | goto out_check; |
1603 | } | |
1604 | ||
a6bd9cd1 NB |
1605 | /* |
1606 | * Found extent which begins before our range and potentially | |
1607 | * intersect it | |
1608 | */ | |
80ff3856 YZ |
1609 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1610 | struct btrfs_file_extent_item); | |
1611 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1612 | ||
cc95bef6 | 1613 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1614 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1615 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1616 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1617 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1618 | extent_end = found_key.offset + |
1619 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1620 | disk_num_bytes = |
1621 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
a6bd9cd1 | 1622 | /* |
de7999af FM |
1623 | * If the extent we got ends before our current offset, |
1624 | * skip to the next extent. | |
a6bd9cd1 | 1625 | */ |
de7999af | 1626 | if (extent_end <= cur_offset) { |
80ff3856 YZ |
1627 | path->slots[0]++; |
1628 | goto next_slot; | |
1629 | } | |
a6bd9cd1 | 1630 | /* Skip holes */ |
17d217fe YZ |
1631 | if (disk_bytenr == 0) |
1632 | goto out_check; | |
a6bd9cd1 | 1633 | /* Skip compressed/encrypted/encoded extents */ |
80ff3856 YZ |
1634 | if (btrfs_file_extent_compression(leaf, fi) || |
1635 | btrfs_file_extent_encryption(leaf, fi) || | |
1636 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1637 | goto out_check; | |
78d4295b | 1638 | /* |
a6bd9cd1 NB |
1639 | * If extent is created before the last volume's snapshot |
1640 | * this implies the extent is shared, hence we can't do | |
1641 | * nocow. This is the same check as in | |
1642 | * btrfs_cross_ref_exist but without calling | |
1643 | * btrfs_search_slot. | |
78d4295b | 1644 | */ |
3e024846 | 1645 | if (!freespace_inode && |
27a7ff55 | 1646 | btrfs_file_extent_generation(leaf, fi) <= |
78d4295b EL |
1647 | btrfs_root_last_snapshot(&root->root_item)) |
1648 | goto out_check; | |
d899e052 YZ |
1649 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1650 | goto out_check; | |
c65ca98f FM |
1651 | |
1652 | /* | |
1653 | * The following checks can be expensive, as they need to | |
1654 | * take other locks and do btree or rbtree searches, so | |
1655 | * release the path to avoid blocking other tasks for too | |
1656 | * long. | |
1657 | */ | |
1658 | btrfs_release_path(path); | |
1659 | ||
a6bd9cd1 | 1660 | /* If extent is RO, we must COW it */ |
2ff7e61e | 1661 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
80ff3856 | 1662 | goto out_check; |
58113753 LB |
1663 | ret = btrfs_cross_ref_exist(root, ino, |
1664 | found_key.offset - | |
a84d5d42 | 1665 | extent_offset, disk_bytenr, false); |
58113753 LB |
1666 | if (ret) { |
1667 | /* | |
1668 | * ret could be -EIO if the above fails to read | |
1669 | * metadata. | |
1670 | */ | |
1671 | if (ret < 0) { | |
1672 | if (cow_start != (u64)-1) | |
1673 | cur_offset = cow_start; | |
1674 | goto error; | |
1675 | } | |
1676 | ||
3e024846 | 1677 | WARN_ON_ONCE(freespace_inode); |
17d217fe | 1678 | goto out_check; |
58113753 | 1679 | } |
5d4f98a2 | 1680 | disk_bytenr += extent_offset; |
17d217fe YZ |
1681 | disk_bytenr += cur_offset - found_key.offset; |
1682 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 | 1683 | /* |
a6bd9cd1 NB |
1684 | * If there are pending snapshots for this root, we |
1685 | * fall into common COW way | |
e9894fd3 | 1686 | */ |
3e024846 | 1687 | if (!freespace_inode && atomic_read(&root->snapshot_force_cow)) |
8ecebf4d | 1688 | goto out_check; |
17d217fe YZ |
1689 | /* |
1690 | * force cow if csum exists in the range. | |
1691 | * this ensure that csum for a given extent are | |
1692 | * either valid or do not exist. | |
1693 | */ | |
58113753 LB |
1694 | ret = csum_exist_in_range(fs_info, disk_bytenr, |
1695 | num_bytes); | |
1696 | if (ret) { | |
58113753 LB |
1697 | /* |
1698 | * ret could be -EIO if the above fails to read | |
1699 | * metadata. | |
1700 | */ | |
1701 | if (ret < 0) { | |
1702 | if (cow_start != (u64)-1) | |
1703 | cur_offset = cow_start; | |
1704 | goto error; | |
1705 | } | |
3e024846 | 1706 | WARN_ON_ONCE(freespace_inode); |
17d217fe | 1707 | goto out_check; |
91e1f56a | 1708 | } |
8ecebf4d | 1709 | if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) |
f78c436c | 1710 | goto out_check; |
3e024846 | 1711 | nocow = true; |
80ff3856 | 1712 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
e8e21007 NB |
1713 | extent_end = found_key.offset + ram_bytes; |
1714 | extent_end = ALIGN(extent_end, fs_info->sectorsize); | |
922f0518 NB |
1715 | /* Skip extents outside of our requested range */ |
1716 | if (extent_end <= start) { | |
1717 | path->slots[0]++; | |
1718 | goto next_slot; | |
1719 | } | |
80ff3856 | 1720 | } else { |
e8e21007 | 1721 | /* If this triggers then we have a memory corruption */ |
290342f6 | 1722 | BUG(); |
80ff3856 YZ |
1723 | } |
1724 | out_check: | |
a6bd9cd1 NB |
1725 | /* |
1726 | * If nocow is false then record the beginning of the range | |
1727 | * that needs to be COWed | |
1728 | */ | |
80ff3856 YZ |
1729 | if (!nocow) { |
1730 | if (cow_start == (u64)-1) | |
1731 | cow_start = cur_offset; | |
1732 | cur_offset = extent_end; | |
1733 | if (cur_offset > end) | |
1734 | break; | |
c65ca98f FM |
1735 | if (!path->nodes[0]) |
1736 | continue; | |
80ff3856 YZ |
1737 | path->slots[0]++; |
1738 | goto next_slot; | |
7ea394f1 YZ |
1739 | } |
1740 | ||
a6bd9cd1 NB |
1741 | /* |
1742 | * COW range from cow_start to found_key.offset - 1. As the key | |
1743 | * will contain the beginning of the first extent that can be | |
1744 | * NOCOW, following one which needs to be COW'ed | |
1745 | */ | |
80ff3856 | 1746 | if (cow_start != (u64)-1) { |
968322c8 | 1747 | ret = fallback_to_cow(inode, locked_page, |
8ba96f3d | 1748 | cow_start, found_key.offset - 1, |
467dc47e | 1749 | page_started, nr_written); |
230ed397 | 1750 | if (ret) |
79787eaa | 1751 | goto error; |
80ff3856 | 1752 | cow_start = (u64)-1; |
7ea394f1 | 1753 | } |
80ff3856 | 1754 | |
d899e052 | 1755 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
6f9994db | 1756 | u64 orig_start = found_key.offset - extent_offset; |
3e024846 | 1757 | struct extent_map *em; |
6f9994db | 1758 | |
968322c8 | 1759 | em = create_io_em(inode, cur_offset, num_bytes, |
6f9994db LB |
1760 | orig_start, |
1761 | disk_bytenr, /* block_start */ | |
1762 | num_bytes, /* block_len */ | |
1763 | disk_num_bytes, /* orig_block_len */ | |
1764 | ram_bytes, BTRFS_COMPRESS_NONE, | |
1765 | BTRFS_ORDERED_PREALLOC); | |
1766 | if (IS_ERR(em)) { | |
6f9994db LB |
1767 | ret = PTR_ERR(em); |
1768 | goto error; | |
d899e052 | 1769 | } |
6f9994db | 1770 | free_extent_map(em); |
968322c8 | 1771 | ret = btrfs_add_ordered_extent(inode, cur_offset, |
bb55f626 NB |
1772 | disk_bytenr, num_bytes, |
1773 | num_bytes, | |
1774 | BTRFS_ORDERED_PREALLOC); | |
762bf098 | 1775 | if (ret) { |
968322c8 | 1776 | btrfs_drop_extent_cache(inode, cur_offset, |
762bf098 NB |
1777 | cur_offset + num_bytes - 1, |
1778 | 0); | |
1779 | goto error; | |
1780 | } | |
d899e052 | 1781 | } else { |
968322c8 | 1782 | ret = btrfs_add_ordered_extent(inode, cur_offset, |
bb55f626 NB |
1783 | disk_bytenr, num_bytes, |
1784 | num_bytes, | |
1785 | BTRFS_ORDERED_NOCOW); | |
762bf098 NB |
1786 | if (ret) |
1787 | goto error; | |
d899e052 | 1788 | } |
80ff3856 | 1789 | |
f78c436c | 1790 | if (nocow) |
0b246afa | 1791 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
762bf098 | 1792 | nocow = false; |
771ed689 | 1793 | |
efa56464 | 1794 | if (root->root_key.objectid == |
4dbd80fb QW |
1795 | BTRFS_DATA_RELOC_TREE_OBJECTID) |
1796 | /* | |
1797 | * Error handled later, as we must prevent | |
1798 | * extent_clear_unlock_delalloc() in error handler | |
1799 | * from freeing metadata of created ordered extent. | |
1800 | */ | |
968322c8 | 1801 | ret = btrfs_reloc_clone_csums(inode, cur_offset, |
efa56464 | 1802 | num_bytes); |
efa56464 | 1803 | |
968322c8 | 1804 | extent_clear_unlock_delalloc(inode, cur_offset, |
74e9194a | 1805 | cur_offset + num_bytes - 1, |
c2790a2e | 1806 | locked_page, EXTENT_LOCKED | |
18513091 WX |
1807 | EXTENT_DELALLOC | |
1808 | EXTENT_CLEAR_DATA_RESV, | |
1809 | PAGE_UNLOCK | PAGE_SET_PRIVATE2); | |
1810 | ||
80ff3856 | 1811 | cur_offset = extent_end; |
4dbd80fb QW |
1812 | |
1813 | /* | |
1814 | * btrfs_reloc_clone_csums() error, now we're OK to call error | |
1815 | * handler, as metadata for created ordered extent will only | |
1816 | * be freed by btrfs_finish_ordered_io(). | |
1817 | */ | |
1818 | if (ret) | |
1819 | goto error; | |
80ff3856 YZ |
1820 | if (cur_offset > end) |
1821 | break; | |
be20aa9d | 1822 | } |
b3b4aa74 | 1823 | btrfs_release_path(path); |
80ff3856 | 1824 | |
506481b2 | 1825 | if (cur_offset <= end && cow_start == (u64)-1) |
80ff3856 | 1826 | cow_start = cur_offset; |
17ca04af | 1827 | |
80ff3856 | 1828 | if (cow_start != (u64)-1) { |
506481b2 | 1829 | cur_offset = end; |
968322c8 NB |
1830 | ret = fallback_to_cow(inode, locked_page, cow_start, end, |
1831 | page_started, nr_written); | |
d788a349 | 1832 | if (ret) |
79787eaa | 1833 | goto error; |
80ff3856 YZ |
1834 | } |
1835 | ||
79787eaa | 1836 | error: |
762bf098 NB |
1837 | if (nocow) |
1838 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); | |
1839 | ||
17ca04af | 1840 | if (ret && cur_offset < end) |
968322c8 | 1841 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
c2790a2e | 1842 | locked_page, EXTENT_LOCKED | |
151a41bc JB |
1843 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1844 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1845 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1846 | PAGE_SET_WRITEBACK | |
1847 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1848 | btrfs_free_path(path); |
79787eaa | 1849 | return ret; |
be20aa9d CM |
1850 | } |
1851 | ||
0c494225 | 1852 | static inline int need_force_cow(struct btrfs_inode *inode, u64 start, u64 end) |
47059d93 WS |
1853 | { |
1854 | ||
0c494225 NB |
1855 | if (!(inode->flags & BTRFS_INODE_NODATACOW) && |
1856 | !(inode->flags & BTRFS_INODE_PREALLOC)) | |
47059d93 WS |
1857 | return 0; |
1858 | ||
1859 | /* | |
1860 | * @defrag_bytes is a hint value, no spinlock held here, | |
1861 | * if is not zero, it means the file is defragging. | |
1862 | * Force cow if given extent needs to be defragged. | |
1863 | */ | |
0c494225 NB |
1864 | if (inode->defrag_bytes && |
1865 | test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG, 0, NULL)) | |
47059d93 WS |
1866 | return 1; |
1867 | ||
1868 | return 0; | |
1869 | } | |
1870 | ||
d352ac68 | 1871 | /* |
5eaad97a NB |
1872 | * Function to process delayed allocation (create CoW) for ranges which are |
1873 | * being touched for the first time. | |
d352ac68 | 1874 | */ |
98456b9c | 1875 | int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page, |
5eaad97a NB |
1876 | u64 start, u64 end, int *page_started, unsigned long *nr_written, |
1877 | struct writeback_control *wbc) | |
be20aa9d | 1878 | { |
be20aa9d | 1879 | int ret; |
98456b9c | 1880 | int force_cow = need_force_cow(inode, start, end); |
a2135011 | 1881 | |
98456b9c NB |
1882 | if (inode->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
1883 | ret = run_delalloc_nocow(inode, locked_page, start, end, | |
d397712b | 1884 | page_started, 1, nr_written); |
98456b9c NB |
1885 | } else if (inode->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
1886 | ret = run_delalloc_nocow(inode, locked_page, start, end, | |
d397712b | 1887 | page_started, 0, nr_written); |
98456b9c NB |
1888 | } else if (!inode_can_compress(inode) || |
1889 | !inode_need_compress(inode, start, end)) { | |
1890 | ret = cow_file_range(inode, locked_page, start, end, | |
1891 | page_started, nr_written, 1); | |
7ddf5a42 | 1892 | } else { |
98456b9c NB |
1893 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags); |
1894 | ret = cow_file_range_async(inode, wbc, locked_page, start, end, | |
fac07d2b | 1895 | page_started, nr_written); |
7ddf5a42 | 1896 | } |
52427260 | 1897 | if (ret) |
98456b9c | 1898 | btrfs_cleanup_ordered_extents(inode, locked_page, start, |
d1051d6e | 1899 | end - start + 1); |
b888db2b CM |
1900 | return ret; |
1901 | } | |
1902 | ||
abbb55f4 NB |
1903 | void btrfs_split_delalloc_extent(struct inode *inode, |
1904 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1905 | { |
dcab6a3b JB |
1906 | u64 size; |
1907 | ||
0ca1f7ce | 1908 | /* not delalloc, ignore it */ |
9ed74f2d | 1909 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1910 | return; |
9ed74f2d | 1911 | |
dcab6a3b JB |
1912 | size = orig->end - orig->start + 1; |
1913 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
823bb20a | 1914 | u32 num_extents; |
dcab6a3b JB |
1915 | u64 new_size; |
1916 | ||
1917 | /* | |
5c848198 | 1918 | * See the explanation in btrfs_merge_delalloc_extent, the same |
ba117213 | 1919 | * applies here, just in reverse. |
dcab6a3b JB |
1920 | */ |
1921 | new_size = orig->end - split + 1; | |
823bb20a | 1922 | num_extents = count_max_extents(new_size); |
ba117213 | 1923 | new_size = split - orig->start; |
823bb20a DS |
1924 | num_extents += count_max_extents(new_size); |
1925 | if (count_max_extents(size) >= num_extents) | |
dcab6a3b JB |
1926 | return; |
1927 | } | |
1928 | ||
9e0baf60 | 1929 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 1930 | btrfs_mod_outstanding_extents(BTRFS_I(inode), 1); |
9e0baf60 | 1931 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
1932 | } |
1933 | ||
1934 | /* | |
5c848198 NB |
1935 | * Handle merged delayed allocation extents so we can keep track of new extents |
1936 | * that are just merged onto old extents, such as when we are doing sequential | |
1937 | * writes, so we can properly account for the metadata space we'll need. | |
9ed74f2d | 1938 | */ |
5c848198 NB |
1939 | void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new, |
1940 | struct extent_state *other) | |
9ed74f2d | 1941 | { |
dcab6a3b | 1942 | u64 new_size, old_size; |
823bb20a | 1943 | u32 num_extents; |
dcab6a3b | 1944 | |
9ed74f2d JB |
1945 | /* not delalloc, ignore it */ |
1946 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1947 | return; |
9ed74f2d | 1948 | |
8461a3de JB |
1949 | if (new->start > other->start) |
1950 | new_size = new->end - other->start + 1; | |
1951 | else | |
1952 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
1953 | |
1954 | /* we're not bigger than the max, unreserve the space and go */ | |
1955 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
1956 | spin_lock(&BTRFS_I(inode)->lock); | |
8b62f87b | 1957 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
dcab6a3b JB |
1958 | spin_unlock(&BTRFS_I(inode)->lock); |
1959 | return; | |
1960 | } | |
1961 | ||
1962 | /* | |
ba117213 JB |
1963 | * We have to add up either side to figure out how many extents were |
1964 | * accounted for before we merged into one big extent. If the number of | |
1965 | * extents we accounted for is <= the amount we need for the new range | |
1966 | * then we can return, otherwise drop. Think of it like this | |
1967 | * | |
1968 | * [ 4k][MAX_SIZE] | |
1969 | * | |
1970 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
1971 | * need 2 outstanding extents, on one side we have 1 and the other side | |
1972 | * we have 1 so they are == and we can return. But in this case | |
1973 | * | |
1974 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
1975 | * | |
1976 | * Each range on their own accounts for 2 extents, but merged together | |
1977 | * they are only 3 extents worth of accounting, so we need to drop in | |
1978 | * this case. | |
dcab6a3b | 1979 | */ |
ba117213 | 1980 | old_size = other->end - other->start + 1; |
823bb20a | 1981 | num_extents = count_max_extents(old_size); |
ba117213 | 1982 | old_size = new->end - new->start + 1; |
823bb20a DS |
1983 | num_extents += count_max_extents(old_size); |
1984 | if (count_max_extents(new_size) >= num_extents) | |
dcab6a3b JB |
1985 | return; |
1986 | ||
9e0baf60 | 1987 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 1988 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
9e0baf60 | 1989 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
1990 | } |
1991 | ||
eb73c1b7 MX |
1992 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1993 | struct inode *inode) | |
1994 | { | |
0b246afa JM |
1995 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1996 | ||
eb73c1b7 MX |
1997 | spin_lock(&root->delalloc_lock); |
1998 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1999 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
2000 | &root->delalloc_inodes); | |
2001 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
2002 | &BTRFS_I(inode)->runtime_flags); | |
2003 | root->nr_delalloc_inodes++; | |
2004 | if (root->nr_delalloc_inodes == 1) { | |
0b246afa | 2005 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2006 | BUG_ON(!list_empty(&root->delalloc_root)); |
2007 | list_add_tail(&root->delalloc_root, | |
0b246afa JM |
2008 | &fs_info->delalloc_roots); |
2009 | spin_unlock(&fs_info->delalloc_root_lock); | |
eb73c1b7 MX |
2010 | } |
2011 | } | |
2012 | spin_unlock(&root->delalloc_lock); | |
2013 | } | |
2014 | ||
2b877331 NB |
2015 | |
2016 | void __btrfs_del_delalloc_inode(struct btrfs_root *root, | |
2017 | struct btrfs_inode *inode) | |
eb73c1b7 | 2018 | { |
3ffbd68c | 2019 | struct btrfs_fs_info *fs_info = root->fs_info; |
0b246afa | 2020 | |
9e3e97f4 NB |
2021 | if (!list_empty(&inode->delalloc_inodes)) { |
2022 | list_del_init(&inode->delalloc_inodes); | |
eb73c1b7 | 2023 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 2024 | &inode->runtime_flags); |
eb73c1b7 MX |
2025 | root->nr_delalloc_inodes--; |
2026 | if (!root->nr_delalloc_inodes) { | |
7c8a0d36 | 2027 | ASSERT(list_empty(&root->delalloc_inodes)); |
0b246afa | 2028 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2029 | BUG_ON(list_empty(&root->delalloc_root)); |
2030 | list_del_init(&root->delalloc_root); | |
0b246afa | 2031 | spin_unlock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
2032 | } |
2033 | } | |
2b877331 NB |
2034 | } |
2035 | ||
2036 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
2037 | struct btrfs_inode *inode) | |
2038 | { | |
2039 | spin_lock(&root->delalloc_lock); | |
2040 | __btrfs_del_delalloc_inode(root, inode); | |
eb73c1b7 MX |
2041 | spin_unlock(&root->delalloc_lock); |
2042 | } | |
2043 | ||
d352ac68 | 2044 | /* |
e06a1fc9 NB |
2045 | * Properly track delayed allocation bytes in the inode and to maintain the |
2046 | * list of inodes that have pending delalloc work to be done. | |
d352ac68 | 2047 | */ |
e06a1fc9 NB |
2048 | void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state, |
2049 | unsigned *bits) | |
291d673e | 2050 | { |
0b246afa JM |
2051 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2052 | ||
47059d93 WS |
2053 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
2054 | WARN_ON(1); | |
75eff68e CM |
2055 | /* |
2056 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 2057 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
2058 | * bit, which is only set or cleared with irqs on |
2059 | */ | |
0ca1f7ce | 2060 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 2061 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2062 | u64 len = state->end + 1 - state->start; |
8b62f87b | 2063 | u32 num_extents = count_max_extents(len); |
70ddc553 | 2064 | bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode)); |
9ed74f2d | 2065 | |
8b62f87b JB |
2066 | spin_lock(&BTRFS_I(inode)->lock); |
2067 | btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents); | |
2068 | spin_unlock(&BTRFS_I(inode)->lock); | |
287a0ab9 | 2069 | |
6a3891c5 | 2070 | /* For sanity tests */ |
0b246afa | 2071 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
2072 | return; |
2073 | ||
104b4e51 NB |
2074 | percpu_counter_add_batch(&fs_info->delalloc_bytes, len, |
2075 | fs_info->delalloc_batch); | |
df0af1a5 | 2076 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 2077 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
2078 | if (*bits & EXTENT_DEFRAG) |
2079 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 2080 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
2081 | &BTRFS_I(inode)->runtime_flags)) |
2082 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 2083 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 2084 | } |
a7e3b975 FM |
2085 | |
2086 | if (!(state->state & EXTENT_DELALLOC_NEW) && | |
2087 | (*bits & EXTENT_DELALLOC_NEW)) { | |
2088 | spin_lock(&BTRFS_I(inode)->lock); | |
2089 | BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 - | |
2090 | state->start; | |
2091 | spin_unlock(&BTRFS_I(inode)->lock); | |
2092 | } | |
291d673e CM |
2093 | } |
2094 | ||
d352ac68 | 2095 | /* |
a36bb5f9 NB |
2096 | * Once a range is no longer delalloc this function ensures that proper |
2097 | * accounting happens. | |
d352ac68 | 2098 | */ |
a36bb5f9 NB |
2099 | void btrfs_clear_delalloc_extent(struct inode *vfs_inode, |
2100 | struct extent_state *state, unsigned *bits) | |
291d673e | 2101 | { |
a36bb5f9 NB |
2102 | struct btrfs_inode *inode = BTRFS_I(vfs_inode); |
2103 | struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb); | |
47059d93 | 2104 | u64 len = state->end + 1 - state->start; |
823bb20a | 2105 | u32 num_extents = count_max_extents(len); |
47059d93 | 2106 | |
4a4b964f FM |
2107 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) { |
2108 | spin_lock(&inode->lock); | |
6fc0ef68 | 2109 | inode->defrag_bytes -= len; |
4a4b964f FM |
2110 | spin_unlock(&inode->lock); |
2111 | } | |
47059d93 | 2112 | |
75eff68e CM |
2113 | /* |
2114 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 2115 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
2116 | * bit, which is only set or cleared with irqs on |
2117 | */ | |
0ca1f7ce | 2118 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
6fc0ef68 | 2119 | struct btrfs_root *root = inode->root; |
83eea1f1 | 2120 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 2121 | |
8b62f87b JB |
2122 | spin_lock(&inode->lock); |
2123 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
2124 | spin_unlock(&inode->lock); | |
0ca1f7ce | 2125 | |
b6d08f06 JB |
2126 | /* |
2127 | * We don't reserve metadata space for space cache inodes so we | |
52042d8e | 2128 | * don't need to call delalloc_release_metadata if there is an |
b6d08f06 JB |
2129 | * error. |
2130 | */ | |
a315e68f | 2131 | if (*bits & EXTENT_CLEAR_META_RESV && |
0b246afa | 2132 | root != fs_info->tree_root) |
43b18595 | 2133 | btrfs_delalloc_release_metadata(inode, len, false); |
0ca1f7ce | 2134 | |
6a3891c5 | 2135 | /* For sanity tests. */ |
0b246afa | 2136 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
2137 | return; |
2138 | ||
a315e68f FM |
2139 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID && |
2140 | do_list && !(state->state & EXTENT_NORESERVE) && | |
2141 | (*bits & EXTENT_CLEAR_DATA_RESV)) | |
9db5d510 | 2142 | btrfs_free_reserved_data_space_noquota(fs_info, len); |
9ed74f2d | 2143 | |
104b4e51 NB |
2144 | percpu_counter_add_batch(&fs_info->delalloc_bytes, -len, |
2145 | fs_info->delalloc_batch); | |
6fc0ef68 NB |
2146 | spin_lock(&inode->lock); |
2147 | inode->delalloc_bytes -= len; | |
2148 | if (do_list && inode->delalloc_bytes == 0 && | |
df0af1a5 | 2149 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 2150 | &inode->runtime_flags)) |
eb73c1b7 | 2151 | btrfs_del_delalloc_inode(root, inode); |
6fc0ef68 | 2152 | spin_unlock(&inode->lock); |
291d673e | 2153 | } |
a7e3b975 FM |
2154 | |
2155 | if ((state->state & EXTENT_DELALLOC_NEW) && | |
2156 | (*bits & EXTENT_DELALLOC_NEW)) { | |
2157 | spin_lock(&inode->lock); | |
2158 | ASSERT(inode->new_delalloc_bytes >= len); | |
2159 | inode->new_delalloc_bytes -= len; | |
2766ff61 FM |
2160 | if (*bits & EXTENT_ADD_INODE_BYTES) |
2161 | inode_add_bytes(&inode->vfs_inode, len); | |
a7e3b975 FM |
2162 | spin_unlock(&inode->lock); |
2163 | } | |
291d673e CM |
2164 | } |
2165 | ||
d352ac68 | 2166 | /* |
da12fe54 NB |
2167 | * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit |
2168 | * in a chunk's stripe. This function ensures that bios do not span a | |
2169 | * stripe/chunk | |
6f034ece | 2170 | * |
da12fe54 NB |
2171 | * @page - The page we are about to add to the bio |
2172 | * @size - size we want to add to the bio | |
2173 | * @bio - bio we want to ensure is smaller than a stripe | |
2174 | * @bio_flags - flags of the bio | |
2175 | * | |
2176 | * return 1 if page cannot be added to the bio | |
2177 | * return 0 if page can be added to the bio | |
6f034ece | 2178 | * return error otherwise |
d352ac68 | 2179 | */ |
da12fe54 NB |
2180 | int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio, |
2181 | unsigned long bio_flags) | |
239b14b3 | 2182 | { |
0b246afa JM |
2183 | struct inode *inode = page->mapping->host; |
2184 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
1201b58b | 2185 | u64 logical = bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
2186 | u64 length = 0; |
2187 | u64 map_length; | |
239b14b3 | 2188 | int ret; |
89b798ad | 2189 | struct btrfs_io_geometry geom; |
239b14b3 | 2190 | |
771ed689 CM |
2191 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
2192 | return 0; | |
2193 | ||
4f024f37 | 2194 | length = bio->bi_iter.bi_size; |
239b14b3 | 2195 | map_length = length; |
89b798ad NB |
2196 | ret = btrfs_get_io_geometry(fs_info, btrfs_op(bio), logical, map_length, |
2197 | &geom); | |
6f034ece LB |
2198 | if (ret < 0) |
2199 | return ret; | |
89b798ad NB |
2200 | |
2201 | if (geom.len < length + size) | |
239b14b3 | 2202 | return 1; |
3444a972 | 2203 | return 0; |
239b14b3 CM |
2204 | } |
2205 | ||
d352ac68 CM |
2206 | /* |
2207 | * in order to insert checksums into the metadata in large chunks, | |
2208 | * we wait until bio submission time. All the pages in the bio are | |
2209 | * checksummed and sums are attached onto the ordered extent record. | |
2210 | * | |
2211 | * At IO completion time the cums attached on the ordered extent record | |
2212 | * are inserted into the btree | |
2213 | */ | |
8896a08d | 2214 | static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio, |
1941b64b | 2215 | u64 dio_file_offset) |
065631f6 | 2216 | { |
c965d640 | 2217 | return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0); |
4a69a410 | 2218 | } |
e015640f | 2219 | |
d352ac68 | 2220 | /* |
cad321ad | 2221 | * extent_io.c submission hook. This does the right thing for csum calculation |
4c274bc6 LB |
2222 | * on write, or reading the csums from the tree before a read. |
2223 | * | |
2224 | * Rules about async/sync submit, | |
2225 | * a) read: sync submit | |
2226 | * | |
2227 | * b) write without checksum: sync submit | |
2228 | * | |
2229 | * c) write with checksum: | |
2230 | * c-1) if bio is issued by fsync: sync submit | |
2231 | * (sync_writers != 0) | |
2232 | * | |
2233 | * c-2) if root is reloc root: sync submit | |
2234 | * (only in case of buffered IO) | |
2235 | * | |
2236 | * c-3) otherwise: async submit | |
d352ac68 | 2237 | */ |
908930f3 NB |
2238 | blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio, |
2239 | int mirror_num, unsigned long bio_flags) | |
50489a57 | 2240 | |
44b8bd7e | 2241 | { |
0b246afa | 2242 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
44b8bd7e | 2243 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0d51e28a | 2244 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
4e4cbee9 | 2245 | blk_status_t ret = 0; |
19b9bdb0 | 2246 | int skip_sum; |
b812ce28 | 2247 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 2248 | |
42437a63 JB |
2249 | skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) || |
2250 | !fs_info->csum_root; | |
cad321ad | 2251 | |
70ddc553 | 2252 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) |
0d51e28a | 2253 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 2254 | |
37226b21 | 2255 | if (bio_op(bio) != REQ_OP_WRITE) { |
0b246afa | 2256 | ret = btrfs_bio_wq_end_io(fs_info, bio, metadata); |
5fd02043 | 2257 | if (ret) |
61891923 | 2258 | goto out; |
5fd02043 | 2259 | |
d20f7043 | 2260 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
2261 | ret = btrfs_submit_compressed_read(inode, bio, |
2262 | mirror_num, | |
2263 | bio_flags); | |
2264 | goto out; | |
334c16d8 JB |
2265 | } else { |
2266 | /* | |
2267 | * Lookup bio sums does extra checks around whether we | |
2268 | * need to csum or not, which is why we ignore skip_sum | |
2269 | * here. | |
2270 | */ | |
6275193e | 2271 | ret = btrfs_lookup_bio_sums(inode, bio, NULL); |
c2db1073 | 2272 | if (ret) |
61891923 | 2273 | goto out; |
c2db1073 | 2274 | } |
4d1b5fb4 | 2275 | goto mapit; |
b812ce28 | 2276 | } else if (async && !skip_sum) { |
17d217fe YZ |
2277 | /* csum items have already been cloned */ |
2278 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
2279 | goto mapit; | |
19b9bdb0 | 2280 | /* we're doing a write, do the async checksumming */ |
8896a08d QW |
2281 | ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags, |
2282 | 0, btrfs_submit_bio_start); | |
61891923 | 2283 | goto out; |
b812ce28 | 2284 | } else if (!skip_sum) { |
bd242a08 | 2285 | ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0); |
b812ce28 JB |
2286 | if (ret) |
2287 | goto out; | |
19b9bdb0 CM |
2288 | } |
2289 | ||
0b86a832 | 2290 | mapit: |
08635bae | 2291 | ret = btrfs_map_bio(fs_info, bio, mirror_num); |
61891923 SB |
2292 | |
2293 | out: | |
4e4cbee9 CH |
2294 | if (ret) { |
2295 | bio->bi_status = ret; | |
4246a0b6 CH |
2296 | bio_endio(bio); |
2297 | } | |
61891923 | 2298 | return ret; |
065631f6 | 2299 | } |
6885f308 | 2300 | |
d352ac68 CM |
2301 | /* |
2302 | * given a list of ordered sums record them in the inode. This happens | |
2303 | * at IO completion time based on sums calculated at bio submission time. | |
2304 | */ | |
510f85ed NB |
2305 | static int add_pending_csums(struct btrfs_trans_handle *trans, |
2306 | struct list_head *list) | |
e6dcd2dc | 2307 | { |
e6dcd2dc | 2308 | struct btrfs_ordered_sum *sum; |
ac01f26a | 2309 | int ret; |
e6dcd2dc | 2310 | |
c6e30871 | 2311 | list_for_each_entry(sum, list, list) { |
7c2871a2 | 2312 | trans->adding_csums = true; |
510f85ed | 2313 | ret = btrfs_csum_file_blocks(trans, trans->fs_info->csum_root, sum); |
7c2871a2 | 2314 | trans->adding_csums = false; |
ac01f26a NB |
2315 | if (ret) |
2316 | return ret; | |
e6dcd2dc CM |
2317 | } |
2318 | return 0; | |
2319 | } | |
2320 | ||
c3347309 FM |
2321 | static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode, |
2322 | const u64 start, | |
2323 | const u64 len, | |
2324 | struct extent_state **cached_state) | |
2325 | { | |
2326 | u64 search_start = start; | |
2327 | const u64 end = start + len - 1; | |
2328 | ||
2329 | while (search_start < end) { | |
2330 | const u64 search_len = end - search_start + 1; | |
2331 | struct extent_map *em; | |
2332 | u64 em_len; | |
2333 | int ret = 0; | |
2334 | ||
2335 | em = btrfs_get_extent(inode, NULL, 0, search_start, search_len); | |
2336 | if (IS_ERR(em)) | |
2337 | return PTR_ERR(em); | |
2338 | ||
2339 | if (em->block_start != EXTENT_MAP_HOLE) | |
2340 | goto next; | |
2341 | ||
2342 | em_len = em->len; | |
2343 | if (em->start < search_start) | |
2344 | em_len -= search_start - em->start; | |
2345 | if (em_len > search_len) | |
2346 | em_len = search_len; | |
2347 | ||
2348 | ret = set_extent_bit(&inode->io_tree, search_start, | |
2349 | search_start + em_len - 1, | |
1cab5e72 NB |
2350 | EXTENT_DELALLOC_NEW, 0, NULL, cached_state, |
2351 | GFP_NOFS, NULL); | |
c3347309 FM |
2352 | next: |
2353 | search_start = extent_map_end(em); | |
2354 | free_extent_map(em); | |
2355 | if (ret) | |
2356 | return ret; | |
2357 | } | |
2358 | return 0; | |
2359 | } | |
2360 | ||
c2566f22 | 2361 | int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end, |
e3b8a485 | 2362 | unsigned int extra_bits, |
330a5827 | 2363 | struct extent_state **cached_state) |
ea8c2819 | 2364 | { |
fdb1e121 | 2365 | WARN_ON(PAGE_ALIGNED(end)); |
c3347309 FM |
2366 | |
2367 | if (start >= i_size_read(&inode->vfs_inode) && | |
2368 | !(inode->flags & BTRFS_INODE_PREALLOC)) { | |
2369 | /* | |
2370 | * There can't be any extents following eof in this case so just | |
2371 | * set the delalloc new bit for the range directly. | |
2372 | */ | |
2373 | extra_bits |= EXTENT_DELALLOC_NEW; | |
2374 | } else { | |
2375 | int ret; | |
2376 | ||
2377 | ret = btrfs_find_new_delalloc_bytes(inode, start, | |
2378 | end + 1 - start, | |
2379 | cached_state); | |
2380 | if (ret) | |
2381 | return ret; | |
2382 | } | |
2383 | ||
c2566f22 NB |
2384 | return set_extent_delalloc(&inode->io_tree, start, end, extra_bits, |
2385 | cached_state); | |
ea8c2819 CM |
2386 | } |
2387 | ||
d352ac68 | 2388 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
2389 | struct btrfs_writepage_fixup { |
2390 | struct page *page; | |
f4b1363c | 2391 | struct inode *inode; |
247e743c CM |
2392 | struct btrfs_work work; |
2393 | }; | |
2394 | ||
b2950863 | 2395 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
2396 | { |
2397 | struct btrfs_writepage_fixup *fixup; | |
2398 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 2399 | struct extent_state *cached_state = NULL; |
364ecf36 | 2400 | struct extent_changeset *data_reserved = NULL; |
247e743c | 2401 | struct page *page; |
65d87f79 | 2402 | struct btrfs_inode *inode; |
247e743c CM |
2403 | u64 page_start; |
2404 | u64 page_end; | |
25f3c502 | 2405 | int ret = 0; |
f4b1363c | 2406 | bool free_delalloc_space = true; |
247e743c CM |
2407 | |
2408 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2409 | page = fixup->page; | |
65d87f79 | 2410 | inode = BTRFS_I(fixup->inode); |
f4b1363c JB |
2411 | page_start = page_offset(page); |
2412 | page_end = page_offset(page) + PAGE_SIZE - 1; | |
2413 | ||
2414 | /* | |
2415 | * This is similar to page_mkwrite, we need to reserve the space before | |
2416 | * we take the page lock. | |
2417 | */ | |
65d87f79 NB |
2418 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
2419 | PAGE_SIZE); | |
4a096752 | 2420 | again: |
247e743c | 2421 | lock_page(page); |
25f3c502 CM |
2422 | |
2423 | /* | |
2424 | * Before we queued this fixup, we took a reference on the page. | |
2425 | * page->mapping may go NULL, but it shouldn't be moved to a different | |
2426 | * address space. | |
2427 | */ | |
f4b1363c JB |
2428 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { |
2429 | /* | |
2430 | * Unfortunately this is a little tricky, either | |
2431 | * | |
2432 | * 1) We got here and our page had already been dealt with and | |
2433 | * we reserved our space, thus ret == 0, so we need to just | |
2434 | * drop our space reservation and bail. This can happen the | |
2435 | * first time we come into the fixup worker, or could happen | |
2436 | * while waiting for the ordered extent. | |
2437 | * 2) Our page was already dealt with, but we happened to get an | |
2438 | * ENOSPC above from the btrfs_delalloc_reserve_space. In | |
2439 | * this case we obviously don't have anything to release, but | |
2440 | * because the page was already dealt with we don't want to | |
2441 | * mark the page with an error, so make sure we're resetting | |
2442 | * ret to 0. This is why we have this check _before_ the ret | |
2443 | * check, because we do not want to have a surprise ENOSPC | |
2444 | * when the page was already properly dealt with. | |
2445 | */ | |
2446 | if (!ret) { | |
65d87f79 NB |
2447 | btrfs_delalloc_release_extents(inode, PAGE_SIZE); |
2448 | btrfs_delalloc_release_space(inode, data_reserved, | |
f4b1363c JB |
2449 | page_start, PAGE_SIZE, |
2450 | true); | |
2451 | } | |
2452 | ret = 0; | |
247e743c | 2453 | goto out_page; |
f4b1363c | 2454 | } |
247e743c | 2455 | |
25f3c502 | 2456 | /* |
f4b1363c JB |
2457 | * We can't mess with the page state unless it is locked, so now that |
2458 | * it is locked bail if we failed to make our space reservation. | |
25f3c502 | 2459 | */ |
f4b1363c JB |
2460 | if (ret) |
2461 | goto out_page; | |
247e743c | 2462 | |
65d87f79 | 2463 | lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state); |
4a096752 CM |
2464 | |
2465 | /* already ordered? We're done */ | |
8b62b72b | 2466 | if (PagePrivate2(page)) |
f4b1363c | 2467 | goto out_reserved; |
4a096752 | 2468 | |
65d87f79 | 2469 | ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE); |
4a096752 | 2470 | if (ordered) { |
65d87f79 NB |
2471 | unlock_extent_cached(&inode->io_tree, page_start, page_end, |
2472 | &cached_state); | |
4a096752 | 2473 | unlock_page(page); |
c0a43603 | 2474 | btrfs_start_ordered_extent(ordered, 1); |
87826df0 | 2475 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2476 | goto again; |
2477 | } | |
247e743c | 2478 | |
65d87f79 | 2479 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, |
330a5827 | 2480 | &cached_state); |
25f3c502 | 2481 | if (ret) |
53687007 | 2482 | goto out_reserved; |
f3038ee3 | 2483 | |
25f3c502 CM |
2484 | /* |
2485 | * Everything went as planned, we're now the owner of a dirty page with | |
2486 | * delayed allocation bits set and space reserved for our COW | |
2487 | * destination. | |
2488 | * | |
2489 | * The page was dirty when we started, nothing should have cleaned it. | |
2490 | */ | |
2491 | BUG_ON(!PageDirty(page)); | |
f4b1363c | 2492 | free_delalloc_space = false; |
53687007 | 2493 | out_reserved: |
65d87f79 | 2494 | btrfs_delalloc_release_extents(inode, PAGE_SIZE); |
f4b1363c | 2495 | if (free_delalloc_space) |
65d87f79 NB |
2496 | btrfs_delalloc_release_space(inode, data_reserved, page_start, |
2497 | PAGE_SIZE, true); | |
2498 | unlock_extent_cached(&inode->io_tree, page_start, page_end, | |
e43bbe5e | 2499 | &cached_state); |
247e743c | 2500 | out_page: |
25f3c502 CM |
2501 | if (ret) { |
2502 | /* | |
2503 | * We hit ENOSPC or other errors. Update the mapping and page | |
2504 | * to reflect the errors and clean the page. | |
2505 | */ | |
2506 | mapping_set_error(page->mapping, ret); | |
2507 | end_extent_writepage(page, ret, page_start, page_end); | |
2508 | clear_page_dirty_for_io(page); | |
2509 | SetPageError(page); | |
2510 | } | |
2511 | ClearPageChecked(page); | |
247e743c | 2512 | unlock_page(page); |
09cbfeaf | 2513 | put_page(page); |
b897abec | 2514 | kfree(fixup); |
364ecf36 | 2515 | extent_changeset_free(data_reserved); |
f4b1363c JB |
2516 | /* |
2517 | * As a precaution, do a delayed iput in case it would be the last iput | |
2518 | * that could need flushing space. Recursing back to fixup worker would | |
2519 | * deadlock. | |
2520 | */ | |
65d87f79 | 2521 | btrfs_add_delayed_iput(&inode->vfs_inode); |
247e743c CM |
2522 | } |
2523 | ||
2524 | /* | |
2525 | * There are a few paths in the higher layers of the kernel that directly | |
2526 | * set the page dirty bit without asking the filesystem if it is a | |
2527 | * good idea. This causes problems because we want to make sure COW | |
2528 | * properly happens and the data=ordered rules are followed. | |
2529 | * | |
c8b97818 | 2530 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2531 | * hasn't been properly setup for IO. We kick off an async process |
2532 | * to fix it up. The async helper will wait for ordered extents, set | |
2533 | * the delalloc bit and make it safe to write the page. | |
2534 | */ | |
d75855b4 | 2535 | int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end) |
247e743c CM |
2536 | { |
2537 | struct inode *inode = page->mapping->host; | |
0b246afa | 2538 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 2539 | struct btrfs_writepage_fixup *fixup; |
247e743c | 2540 | |
8b62b72b CM |
2541 | /* this page is properly in the ordered list */ |
2542 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2543 | return 0; |
2544 | ||
25f3c502 CM |
2545 | /* |
2546 | * PageChecked is set below when we create a fixup worker for this page, | |
2547 | * don't try to create another one if we're already PageChecked() | |
2548 | * | |
2549 | * The extent_io writepage code will redirty the page if we send back | |
2550 | * EAGAIN. | |
2551 | */ | |
247e743c CM |
2552 | if (PageChecked(page)) |
2553 | return -EAGAIN; | |
2554 | ||
2555 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2556 | if (!fixup) | |
2557 | return -EAGAIN; | |
f421950f | 2558 | |
f4b1363c JB |
2559 | /* |
2560 | * We are already holding a reference to this inode from | |
2561 | * write_cache_pages. We need to hold it because the space reservation | |
2562 | * takes place outside of the page lock, and we can't trust | |
2563 | * page->mapping outside of the page lock. | |
2564 | */ | |
2565 | ihold(inode); | |
247e743c | 2566 | SetPageChecked(page); |
09cbfeaf | 2567 | get_page(page); |
a0cac0ec | 2568 | btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL); |
247e743c | 2569 | fixup->page = page; |
f4b1363c | 2570 | fixup->inode = inode; |
0b246afa | 2571 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
25f3c502 CM |
2572 | |
2573 | return -EAGAIN; | |
247e743c CM |
2574 | } |
2575 | ||
d899e052 | 2576 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
c553f94d | 2577 | struct btrfs_inode *inode, u64 file_pos, |
9729f10a | 2578 | struct btrfs_file_extent_item *stack_fi, |
2766ff61 | 2579 | const bool update_inode_bytes, |
9729f10a | 2580 | u64 qgroup_reserved) |
d899e052 | 2581 | { |
c553f94d | 2582 | struct btrfs_root *root = inode->root; |
2766ff61 | 2583 | const u64 sectorsize = root->fs_info->sectorsize; |
d899e052 YZ |
2584 | struct btrfs_path *path; |
2585 | struct extent_buffer *leaf; | |
2586 | struct btrfs_key ins; | |
203f44c5 QW |
2587 | u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi); |
2588 | u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi); | |
2589 | u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi); | |
2590 | u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi); | |
5893dfb9 | 2591 | struct btrfs_drop_extents_args drop_args = { 0 }; |
d899e052 YZ |
2592 | int ret; |
2593 | ||
2594 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2595 | if (!path) |
2596 | return -ENOMEM; | |
d899e052 | 2597 | |
a1ed835e CM |
2598 | /* |
2599 | * we may be replacing one extent in the tree with another. | |
2600 | * The new extent is pinned in the extent map, and we don't want | |
2601 | * to drop it from the cache until it is completely in the btree. | |
2602 | * | |
2603 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2604 | * the caller is expected to unpin it and allow it to be merged | |
2605 | * with the others. | |
2606 | */ | |
5893dfb9 FM |
2607 | drop_args.path = path; |
2608 | drop_args.start = file_pos; | |
2609 | drop_args.end = file_pos + num_bytes; | |
2610 | drop_args.replace_extent = true; | |
2611 | drop_args.extent_item_size = sizeof(*stack_fi); | |
2612 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); | |
79787eaa JM |
2613 | if (ret) |
2614 | goto out; | |
d899e052 | 2615 | |
5893dfb9 | 2616 | if (!drop_args.extent_inserted) { |
c553f94d | 2617 | ins.objectid = btrfs_ino(inode); |
1acae57b FDBM |
2618 | ins.offset = file_pos; |
2619 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2620 | ||
1acae57b | 2621 | ret = btrfs_insert_empty_item(trans, root, path, &ins, |
203f44c5 | 2622 | sizeof(*stack_fi)); |
1acae57b FDBM |
2623 | if (ret) |
2624 | goto out; | |
2625 | } | |
d899e052 | 2626 | leaf = path->nodes[0]; |
203f44c5 QW |
2627 | btrfs_set_stack_file_extent_generation(stack_fi, trans->transid); |
2628 | write_extent_buffer(leaf, stack_fi, | |
2629 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
2630 | sizeof(struct btrfs_file_extent_item)); | |
b9473439 | 2631 | |
d899e052 | 2632 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2633 | btrfs_release_path(path); |
d899e052 | 2634 | |
2766ff61 FM |
2635 | /* |
2636 | * If we dropped an inline extent here, we know the range where it is | |
2637 | * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the | |
2638 | * number of bytes only for that range contaning the inline extent. | |
2639 | * The remaining of the range will be processed when clearning the | |
2640 | * EXTENT_DELALLOC_BIT bit through the ordered extent completion. | |
2641 | */ | |
2642 | if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) { | |
2643 | u64 inline_size = round_down(drop_args.bytes_found, sectorsize); | |
2644 | ||
2645 | inline_size = drop_args.bytes_found - inline_size; | |
2646 | btrfs_update_inode_bytes(inode, sectorsize, inline_size); | |
2647 | drop_args.bytes_found -= inline_size; | |
2648 | num_bytes -= sectorsize; | |
2649 | } | |
2650 | ||
2651 | if (update_inode_bytes) | |
2652 | btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found); | |
d899e052 YZ |
2653 | |
2654 | ins.objectid = disk_bytenr; | |
2655 | ins.offset = disk_num_bytes; | |
2656 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
a12b877b | 2657 | |
c553f94d | 2658 | ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes); |
9ddc959e JB |
2659 | if (ret) |
2660 | goto out; | |
2661 | ||
c553f94d | 2662 | ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode), |
9729f10a | 2663 | file_pos, qgroup_reserved, &ins); |
79787eaa | 2664 | out: |
d899e052 | 2665 | btrfs_free_path(path); |
b9473439 | 2666 | |
79787eaa | 2667 | return ret; |
d899e052 YZ |
2668 | } |
2669 | ||
2ff7e61e | 2670 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
2671 | u64 start, u64 len) |
2672 | { | |
32da5386 | 2673 | struct btrfs_block_group *cache; |
e570fd27 | 2674 | |
0b246afa | 2675 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
2676 | ASSERT(cache); |
2677 | ||
2678 | spin_lock(&cache->lock); | |
2679 | cache->delalloc_bytes -= len; | |
2680 | spin_unlock(&cache->lock); | |
2681 | ||
2682 | btrfs_put_block_group(cache); | |
2683 | } | |
2684 | ||
203f44c5 | 2685 | static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans, |
203f44c5 QW |
2686 | struct btrfs_ordered_extent *oe) |
2687 | { | |
2688 | struct btrfs_file_extent_item stack_fi; | |
2689 | u64 logical_len; | |
2766ff61 | 2690 | bool update_inode_bytes; |
203f44c5 QW |
2691 | |
2692 | memset(&stack_fi, 0, sizeof(stack_fi)); | |
2693 | btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG); | |
2694 | btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr); | |
2695 | btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, | |
2696 | oe->disk_num_bytes); | |
2697 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) | |
2698 | logical_len = oe->truncated_len; | |
2699 | else | |
2700 | logical_len = oe->num_bytes; | |
2701 | btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len); | |
2702 | btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len); | |
2703 | btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type); | |
2704 | /* Encryption and other encoding is reserved and all 0 */ | |
2705 | ||
2766ff61 FM |
2706 | /* |
2707 | * For delalloc, when completing an ordered extent we update the inode's | |
2708 | * bytes when clearing the range in the inode's io tree, so pass false | |
2709 | * as the argument 'update_inode_bytes' to insert_reserved_file_extent(), | |
2710 | * except if the ordered extent was truncated. | |
2711 | */ | |
2712 | update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) || | |
2713 | test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags); | |
2714 | ||
3c38c877 NB |
2715 | return insert_reserved_file_extent(trans, BTRFS_I(oe->inode), |
2716 | oe->file_offset, &stack_fi, | |
2766ff61 | 2717 | update_inode_bytes, oe->qgroup_rsv); |
203f44c5 QW |
2718 | } |
2719 | ||
2720 | /* | |
2721 | * As ordered data IO finishes, this gets called so we can finish | |
d352ac68 CM |
2722 | * an ordered extent if the range of bytes in the file it covers are |
2723 | * fully written. | |
2724 | */ | |
5fd02043 | 2725 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2726 | { |
72e7e6ed NB |
2727 | struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode); |
2728 | struct btrfs_root *root = inode->root; | |
2729 | struct btrfs_fs_info *fs_info = root->fs_info; | |
0ca1f7ce | 2730 | struct btrfs_trans_handle *trans = NULL; |
72e7e6ed | 2731 | struct extent_io_tree *io_tree = &inode->io_tree; |
2ac55d41 | 2732 | struct extent_state *cached_state = NULL; |
bffe633e | 2733 | u64 start, end; |
261507a0 | 2734 | int compress_type = 0; |
77cef2ec | 2735 | int ret = 0; |
bffe633e | 2736 | u64 logical_len = ordered_extent->num_bytes; |
8d510121 | 2737 | bool freespace_inode; |
77cef2ec | 2738 | bool truncated = false; |
49940bdd | 2739 | bool clear_reserved_extent = true; |
2766ff61 | 2740 | unsigned int clear_bits = EXTENT_DEFRAG; |
a7e3b975 | 2741 | |
bffe633e OS |
2742 | start = ordered_extent->file_offset; |
2743 | end = start + ordered_extent->num_bytes - 1; | |
2744 | ||
a7e3b975 FM |
2745 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
2746 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) && | |
2747 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags)) | |
2766ff61 | 2748 | clear_bits |= EXTENT_DELALLOC_NEW; |
e6dcd2dc | 2749 | |
72e7e6ed | 2750 | freespace_inode = btrfs_is_free_space_inode(inode); |
0cb59c99 | 2751 | |
5fd02043 JB |
2752 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2753 | ret = -EIO; | |
2754 | goto out; | |
2755 | } | |
2756 | ||
72e7e6ed | 2757 | btrfs_free_io_failure_record(inode, start, end); |
f612496b | 2758 | |
77cef2ec JB |
2759 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2760 | truncated = true; | |
2761 | logical_len = ordered_extent->truncated_len; | |
2762 | /* Truncated the entire extent, don't bother adding */ | |
2763 | if (!logical_len) | |
2764 | goto out; | |
2765 | } | |
2766 | ||
c2167754 | 2767 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2768 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a | 2769 | |
72e7e6ed | 2770 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
8d510121 NB |
2771 | if (freespace_inode) |
2772 | trans = btrfs_join_transaction_spacecache(root); | |
6c760c07 JB |
2773 | else |
2774 | trans = btrfs_join_transaction(root); | |
2775 | if (IS_ERR(trans)) { | |
2776 | ret = PTR_ERR(trans); | |
2777 | trans = NULL; | |
2778 | goto out; | |
c2167754 | 2779 | } |
72e7e6ed | 2780 | trans->block_rsv = &inode->block_rsv; |
729f7961 | 2781 | ret = btrfs_update_inode_fallback(trans, root, inode); |
6c760c07 | 2782 | if (ret) /* -ENOMEM or corruption */ |
66642832 | 2783 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
2784 | goto out; |
2785 | } | |
e6dcd2dc | 2786 | |
2766ff61 | 2787 | clear_bits |= EXTENT_LOCKED; |
bffe633e | 2788 | lock_extent_bits(io_tree, start, end, &cached_state); |
e6dcd2dc | 2789 | |
8d510121 NB |
2790 | if (freespace_inode) |
2791 | trans = btrfs_join_transaction_spacecache(root); | |
0cb59c99 | 2792 | else |
7a7eaa40 | 2793 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2794 | if (IS_ERR(trans)) { |
2795 | ret = PTR_ERR(trans); | |
2796 | trans = NULL; | |
a7e3b975 | 2797 | goto out; |
79787eaa | 2798 | } |
a79b7d4b | 2799 | |
72e7e6ed | 2800 | trans->block_rsv = &inode->block_rsv; |
c2167754 | 2801 | |
c8b97818 | 2802 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 2803 | compress_type = ordered_extent->compress_type; |
d899e052 | 2804 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 2805 | BUG_ON(compress_type); |
72e7e6ed | 2806 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
2807 | ordered_extent->file_offset, |
2808 | ordered_extent->file_offset + | |
77cef2ec | 2809 | logical_len); |
d899e052 | 2810 | } else { |
0b246afa | 2811 | BUG_ON(root == fs_info->tree_root); |
3c38c877 | 2812 | ret = insert_ordered_extent_file_extent(trans, ordered_extent); |
49940bdd JB |
2813 | if (!ret) { |
2814 | clear_reserved_extent = false; | |
2ff7e61e | 2815 | btrfs_release_delalloc_bytes(fs_info, |
bffe633e OS |
2816 | ordered_extent->disk_bytenr, |
2817 | ordered_extent->disk_num_bytes); | |
49940bdd | 2818 | } |
d899e052 | 2819 | } |
72e7e6ed | 2820 | unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset, |
bffe633e | 2821 | ordered_extent->num_bytes, trans->transid); |
79787eaa | 2822 | if (ret < 0) { |
66642832 | 2823 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 2824 | goto out; |
79787eaa | 2825 | } |
2ac55d41 | 2826 | |
510f85ed | 2827 | ret = add_pending_csums(trans, &ordered_extent->list); |
ac01f26a NB |
2828 | if (ret) { |
2829 | btrfs_abort_transaction(trans, ret); | |
2830 | goto out; | |
2831 | } | |
e6dcd2dc | 2832 | |
2766ff61 FM |
2833 | /* |
2834 | * If this is a new delalloc range, clear its new delalloc flag to | |
2835 | * update the inode's number of bytes. This needs to be done first | |
2836 | * before updating the inode item. | |
2837 | */ | |
2838 | if ((clear_bits & EXTENT_DELALLOC_NEW) && | |
2839 | !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) | |
72e7e6ed | 2840 | clear_extent_bit(&inode->io_tree, start, end, |
2766ff61 FM |
2841 | EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES, |
2842 | 0, 0, &cached_state); | |
2843 | ||
72e7e6ed | 2844 | btrfs_inode_safe_disk_i_size_write(inode, 0); |
729f7961 | 2845 | ret = btrfs_update_inode_fallback(trans, root, inode); |
6c760c07 | 2846 | if (ret) { /* -ENOMEM or corruption */ |
66642832 | 2847 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 2848 | goto out; |
1ef30be1 JB |
2849 | } |
2850 | ret = 0; | |
c2167754 | 2851 | out: |
72e7e6ed | 2852 | clear_extent_bit(&inode->io_tree, start, end, clear_bits, |
bffe633e | 2853 | (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0, |
313facc5 | 2854 | &cached_state); |
a7e3b975 | 2855 | |
a698d075 | 2856 | if (trans) |
3a45bb20 | 2857 | btrfs_end_transaction(trans); |
0cb59c99 | 2858 | |
77cef2ec | 2859 | if (ret || truncated) { |
bffe633e | 2860 | u64 unwritten_start = start; |
77cef2ec JB |
2861 | |
2862 | if (truncated) | |
bffe633e OS |
2863 | unwritten_start += logical_len; |
2864 | clear_extent_uptodate(io_tree, unwritten_start, end, NULL); | |
77cef2ec JB |
2865 | |
2866 | /* Drop the cache for the part of the extent we didn't write. */ | |
72e7e6ed | 2867 | btrfs_drop_extent_cache(inode, unwritten_start, end, 0); |
5fd02043 | 2868 | |
0bec9ef5 JB |
2869 | /* |
2870 | * If the ordered extent had an IOERR or something else went | |
2871 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
2872 | * back to the allocator. We only free the extent in the |
2873 | * truncated case if we didn't write out the extent at all. | |
49940bdd JB |
2874 | * |
2875 | * If we made it past insert_reserved_file_extent before we | |
2876 | * errored out then we don't need to do this as the accounting | |
2877 | * has already been done. | |
0bec9ef5 | 2878 | */ |
77cef2ec | 2879 | if ((ret || !logical_len) && |
49940bdd | 2880 | clear_reserved_extent && |
77cef2ec | 2881 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
4eaaec24 NB |
2882 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
2883 | /* | |
2884 | * Discard the range before returning it back to the | |
2885 | * free space pool | |
2886 | */ | |
46b27f50 | 2887 | if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC)) |
4eaaec24 | 2888 | btrfs_discard_extent(fs_info, |
bffe633e OS |
2889 | ordered_extent->disk_bytenr, |
2890 | ordered_extent->disk_num_bytes, | |
2891 | NULL); | |
2ff7e61e | 2892 | btrfs_free_reserved_extent(fs_info, |
bffe633e OS |
2893 | ordered_extent->disk_bytenr, |
2894 | ordered_extent->disk_num_bytes, 1); | |
4eaaec24 | 2895 | } |
0bec9ef5 JB |
2896 | } |
2897 | ||
5fd02043 | 2898 | /* |
8bad3c02 LB |
2899 | * This needs to be done to make sure anybody waiting knows we are done |
2900 | * updating everything for this ordered extent. | |
5fd02043 | 2901 | */ |
72e7e6ed | 2902 | btrfs_remove_ordered_extent(inode, ordered_extent); |
5fd02043 | 2903 | |
e6dcd2dc CM |
2904 | /* once for us */ |
2905 | btrfs_put_ordered_extent(ordered_extent); | |
2906 | /* once for the tree */ | |
2907 | btrfs_put_ordered_extent(ordered_extent); | |
2908 | ||
5fd02043 JB |
2909 | return ret; |
2910 | } | |
2911 | ||
2912 | static void finish_ordered_fn(struct btrfs_work *work) | |
2913 | { | |
2914 | struct btrfs_ordered_extent *ordered_extent; | |
2915 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
2916 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
2917 | } |
2918 | ||
c629732d NB |
2919 | void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start, |
2920 | u64 end, int uptodate) | |
211f90e6 | 2921 | { |
3347c48f NB |
2922 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
2923 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
5fd02043 | 2924 | struct btrfs_ordered_extent *ordered_extent = NULL; |
9e0af237 | 2925 | struct btrfs_workqueue *wq; |
5fd02043 | 2926 | |
1abe9b8a | 2927 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
2928 | ||
8b62b72b | 2929 | ClearPagePrivate2(page); |
3347c48f NB |
2930 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
2931 | end - start + 1, uptodate)) | |
c3988d63 | 2932 | return; |
5fd02043 | 2933 | |
3347c48f | 2934 | if (btrfs_is_free_space_inode(inode)) |
0b246afa | 2935 | wq = fs_info->endio_freespace_worker; |
a0cac0ec | 2936 | else |
0b246afa | 2937 | wq = fs_info->endio_write_workers; |
5fd02043 | 2938 | |
a0cac0ec | 2939 | btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL); |
9e0af237 | 2940 | btrfs_queue_work(wq, &ordered_extent->work); |
211f90e6 CM |
2941 | } |
2942 | ||
265d4ac0 QW |
2943 | /* |
2944 | * check_data_csum - verify checksum of one sector of uncompressed data | |
7ffd27e3 | 2945 | * @inode: inode |
265d4ac0 | 2946 | * @io_bio: btrfs_io_bio which contains the csum |
7ffd27e3 | 2947 | * @bio_offset: offset to the beginning of the bio (in bytes) |
265d4ac0 QW |
2948 | * @page: page where is the data to be verified |
2949 | * @pgoff: offset inside the page | |
2950 | * | |
2951 | * The length of such check is always one sector size. | |
2952 | */ | |
47df7765 | 2953 | static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio, |
f44cf410 | 2954 | u32 bio_offset, struct page *page, u32 pgoff) |
dc380aea | 2955 | { |
d5178578 JT |
2956 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2957 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); | |
dc380aea | 2958 | char *kaddr; |
265d4ac0 | 2959 | u32 len = fs_info->sectorsize; |
223486c2 | 2960 | const u32 csum_size = fs_info->csum_size; |
7ffd27e3 | 2961 | unsigned int offset_sectors; |
d5178578 JT |
2962 | u8 *csum_expected; |
2963 | u8 csum[BTRFS_CSUM_SIZE]; | |
dc380aea | 2964 | |
265d4ac0 QW |
2965 | ASSERT(pgoff + len <= PAGE_SIZE); |
2966 | ||
7ffd27e3 QW |
2967 | offset_sectors = bio_offset >> fs_info->sectorsize_bits; |
2968 | csum_expected = ((u8 *)io_bio->csum) + offset_sectors * csum_size; | |
dc380aea MX |
2969 | |
2970 | kaddr = kmap_atomic(page); | |
d5178578 JT |
2971 | shash->tfm = fs_info->csum_shash; |
2972 | ||
fd08001f | 2973 | crypto_shash_digest(shash, kaddr + pgoff, len, csum); |
d5178578 JT |
2974 | |
2975 | if (memcmp(csum, csum_expected, csum_size)) | |
dc380aea MX |
2976 | goto zeroit; |
2977 | ||
2978 | kunmap_atomic(kaddr); | |
2979 | return 0; | |
2980 | zeroit: | |
265d4ac0 QW |
2981 | btrfs_print_data_csum_error(BTRFS_I(inode), page_offset(page) + pgoff, |
2982 | csum, csum_expected, io_bio->mirror_num); | |
814723e0 NB |
2983 | if (io_bio->device) |
2984 | btrfs_dev_stat_inc_and_print(io_bio->device, | |
2985 | BTRFS_DEV_STAT_CORRUPTION_ERRS); | |
dc380aea MX |
2986 | memset(kaddr + pgoff, 1, len); |
2987 | flush_dcache_page(page); | |
2988 | kunmap_atomic(kaddr); | |
dc380aea MX |
2989 | return -EIO; |
2990 | } | |
2991 | ||
d352ac68 | 2992 | /* |
7ffd27e3 | 2993 | * When reads are done, we need to check csums to verify the data is correct. |
4a54c8c1 JS |
2994 | * if there's a match, we allow the bio to finish. If not, the code in |
2995 | * extent_io.c will try to find good copies for us. | |
7ffd27e3 QW |
2996 | * |
2997 | * @bio_offset: offset to the beginning of the bio (in bytes) | |
2998 | * @start: file offset of the range start | |
2999 | * @end: file offset of the range end (inclusive) | |
3000 | * @mirror: mirror number | |
d352ac68 | 3001 | */ |
7ffd27e3 | 3002 | int btrfs_verify_data_csum(struct btrfs_io_bio *io_bio, u32 bio_offset, |
9a446d6a | 3003 | struct page *page, u64 start, u64 end, int mirror) |
07157aac | 3004 | { |
07157aac | 3005 | struct inode *inode = page->mapping->host; |
d1310b2e | 3006 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3007 | struct btrfs_root *root = BTRFS_I(inode)->root; |
f44cf410 QW |
3008 | const u32 sectorsize = root->fs_info->sectorsize; |
3009 | u32 pg_off; | |
d1310b2e | 3010 | |
d20f7043 CM |
3011 | if (PageChecked(page)) { |
3012 | ClearPageChecked(page); | |
dc380aea | 3013 | return 0; |
d20f7043 | 3014 | } |
6cbff00f CH |
3015 | |
3016 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 3017 | return 0; |
17d217fe | 3018 | |
42437a63 JB |
3019 | if (!root->fs_info->csum_root) |
3020 | return 0; | |
3021 | ||
17d217fe | 3022 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && |
9655d298 | 3023 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
91166212 | 3024 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM); |
b6cda9bc | 3025 | return 0; |
17d217fe | 3026 | } |
d20f7043 | 3027 | |
f44cf410 QW |
3028 | ASSERT(page_offset(page) <= start && |
3029 | end <= page_offset(page) + PAGE_SIZE - 1); | |
3030 | for (pg_off = offset_in_page(start); | |
3031 | pg_off < offset_in_page(end); | |
3032 | pg_off += sectorsize, bio_offset += sectorsize) { | |
3033 | int ret; | |
3034 | ||
3035 | ret = check_data_csum(inode, io_bio, bio_offset, page, pg_off); | |
3036 | if (ret < 0) | |
3037 | return -EIO; | |
3038 | } | |
3039 | return 0; | |
07157aac | 3040 | } |
b888db2b | 3041 | |
c1c3fac2 NB |
3042 | /* |
3043 | * btrfs_add_delayed_iput - perform a delayed iput on @inode | |
3044 | * | |
3045 | * @inode: The inode we want to perform iput on | |
3046 | * | |
3047 | * This function uses the generic vfs_inode::i_count to track whether we should | |
3048 | * just decrement it (in case it's > 1) or if this is the last iput then link | |
3049 | * the inode to the delayed iput machinery. Delayed iputs are processed at | |
3050 | * transaction commit time/superblock commit/cleaner kthread. | |
3051 | */ | |
24bbcf04 YZ |
3052 | void btrfs_add_delayed_iput(struct inode *inode) |
3053 | { | |
0b246afa | 3054 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8089fe62 | 3055 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3056 | |
3057 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3058 | return; | |
3059 | ||
034f784d | 3060 | atomic_inc(&fs_info->nr_delayed_iputs); |
24bbcf04 | 3061 | spin_lock(&fs_info->delayed_iput_lock); |
c1c3fac2 NB |
3062 | ASSERT(list_empty(&binode->delayed_iput)); |
3063 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
24bbcf04 | 3064 | spin_unlock(&fs_info->delayed_iput_lock); |
fd340d0f JB |
3065 | if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags)) |
3066 | wake_up_process(fs_info->cleaner_kthread); | |
24bbcf04 YZ |
3067 | } |
3068 | ||
63611e73 JB |
3069 | static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info, |
3070 | struct btrfs_inode *inode) | |
3071 | { | |
3072 | list_del_init(&inode->delayed_iput); | |
3073 | spin_unlock(&fs_info->delayed_iput_lock); | |
3074 | iput(&inode->vfs_inode); | |
3075 | if (atomic_dec_and_test(&fs_info->nr_delayed_iputs)) | |
3076 | wake_up(&fs_info->delayed_iputs_wait); | |
3077 | spin_lock(&fs_info->delayed_iput_lock); | |
3078 | } | |
3079 | ||
3080 | static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info, | |
3081 | struct btrfs_inode *inode) | |
3082 | { | |
3083 | if (!list_empty(&inode->delayed_iput)) { | |
3084 | spin_lock(&fs_info->delayed_iput_lock); | |
3085 | if (!list_empty(&inode->delayed_iput)) | |
3086 | run_delayed_iput_locked(fs_info, inode); | |
3087 | spin_unlock(&fs_info->delayed_iput_lock); | |
3088 | } | |
3089 | } | |
3090 | ||
2ff7e61e | 3091 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 3092 | { |
24bbcf04 | 3093 | |
24bbcf04 | 3094 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3095 | while (!list_empty(&fs_info->delayed_iputs)) { |
3096 | struct btrfs_inode *inode; | |
3097 | ||
3098 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3099 | struct btrfs_inode, delayed_iput); | |
63611e73 | 3100 | run_delayed_iput_locked(fs_info, inode); |
24bbcf04 | 3101 | } |
8089fe62 | 3102 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3103 | } |
3104 | ||
034f784d JB |
3105 | /** |
3106 | * btrfs_wait_on_delayed_iputs - wait on the delayed iputs to be done running | |
3107 | * @fs_info - the fs_info for this fs | |
3108 | * @return - EINTR if we were killed, 0 if nothing's pending | |
3109 | * | |
3110 | * This will wait on any delayed iputs that are currently running with KILLABLE | |
3111 | * set. Once they are all done running we will return, unless we are killed in | |
3112 | * which case we return EINTR. This helps in user operations like fallocate etc | |
3113 | * that might get blocked on the iputs. | |
3114 | */ | |
3115 | int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info) | |
3116 | { | |
3117 | int ret = wait_event_killable(fs_info->delayed_iputs_wait, | |
3118 | atomic_read(&fs_info->nr_delayed_iputs) == 0); | |
3119 | if (ret) | |
3120 | return -EINTR; | |
3121 | return 0; | |
3122 | } | |
3123 | ||
7b128766 | 3124 | /* |
f7e9e8fc OS |
3125 | * This creates an orphan entry for the given inode in case something goes wrong |
3126 | * in the middle of an unlink. | |
7b128766 | 3127 | */ |
73f2e545 | 3128 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, |
27919067 | 3129 | struct btrfs_inode *inode) |
7b128766 | 3130 | { |
d68fc57b | 3131 | int ret; |
7b128766 | 3132 | |
27919067 OS |
3133 | ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode)); |
3134 | if (ret && ret != -EEXIST) { | |
3135 | btrfs_abort_transaction(trans, ret); | |
3136 | return ret; | |
d68fc57b YZ |
3137 | } |
3138 | ||
d68fc57b | 3139 | return 0; |
7b128766 JB |
3140 | } |
3141 | ||
3142 | /* | |
f7e9e8fc OS |
3143 | * We have done the delete so we can go ahead and remove the orphan item for |
3144 | * this particular inode. | |
7b128766 | 3145 | */ |
48a3b636 | 3146 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3d6ae7bb | 3147 | struct btrfs_inode *inode) |
7b128766 | 3148 | { |
27919067 | 3149 | return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode)); |
7b128766 JB |
3150 | } |
3151 | ||
3152 | /* | |
3153 | * this cleans up any orphans that may be left on the list from the last use | |
3154 | * of this root. | |
3155 | */ | |
66b4ffd1 | 3156 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 3157 | { |
0b246afa | 3158 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
3159 | struct btrfs_path *path; |
3160 | struct extent_buffer *leaf; | |
7b128766 JB |
3161 | struct btrfs_key key, found_key; |
3162 | struct btrfs_trans_handle *trans; | |
3163 | struct inode *inode; | |
8f6d7f4f | 3164 | u64 last_objectid = 0; |
f7e9e8fc | 3165 | int ret = 0, nr_unlink = 0; |
7b128766 | 3166 | |
d68fc57b | 3167 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3168 | return 0; |
c71bf099 YZ |
3169 | |
3170 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3171 | if (!path) { |
3172 | ret = -ENOMEM; | |
3173 | goto out; | |
3174 | } | |
e4058b54 | 3175 | path->reada = READA_BACK; |
7b128766 JB |
3176 | |
3177 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3178 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3179 | key.offset = (u64)-1; |
3180 | ||
7b128766 JB |
3181 | while (1) { |
3182 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3183 | if (ret < 0) |
3184 | goto out; | |
7b128766 JB |
3185 | |
3186 | /* | |
3187 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3188 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3189 | * find the key and see if we have stuff that matches |
3190 | */ | |
3191 | if (ret > 0) { | |
66b4ffd1 | 3192 | ret = 0; |
7b128766 JB |
3193 | if (path->slots[0] == 0) |
3194 | break; | |
3195 | path->slots[0]--; | |
3196 | } | |
3197 | ||
3198 | /* pull out the item */ | |
3199 | leaf = path->nodes[0]; | |
7b128766 JB |
3200 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3201 | ||
3202 | /* make sure the item matches what we want */ | |
3203 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3204 | break; | |
962a298f | 3205 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3206 | break; |
3207 | ||
3208 | /* release the path since we're done with it */ | |
b3b4aa74 | 3209 | btrfs_release_path(path); |
7b128766 JB |
3210 | |
3211 | /* | |
3212 | * this is where we are basically btrfs_lookup, without the | |
3213 | * crossing root thing. we store the inode number in the | |
3214 | * offset of the orphan item. | |
3215 | */ | |
8f6d7f4f JB |
3216 | |
3217 | if (found_key.offset == last_objectid) { | |
0b246afa JM |
3218 | btrfs_err(fs_info, |
3219 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3220 | ret = -EINVAL; |
3221 | goto out; | |
3222 | } | |
3223 | ||
3224 | last_objectid = found_key.offset; | |
3225 | ||
5d4f98a2 YZ |
3226 | found_key.objectid = found_key.offset; |
3227 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3228 | found_key.offset = 0; | |
0202e83f | 3229 | inode = btrfs_iget(fs_info->sb, last_objectid, root); |
8c6ffba0 | 3230 | ret = PTR_ERR_OR_ZERO(inode); |
67710892 | 3231 | if (ret && ret != -ENOENT) |
66b4ffd1 | 3232 | goto out; |
7b128766 | 3233 | |
0b246afa | 3234 | if (ret == -ENOENT && root == fs_info->tree_root) { |
f8e9e0b0 | 3235 | struct btrfs_root *dead_root; |
f8e9e0b0 AJ |
3236 | int is_dead_root = 0; |
3237 | ||
3238 | /* | |
3239 | * this is an orphan in the tree root. Currently these | |
3240 | * could come from 2 sources: | |
3241 | * a) a snapshot deletion in progress | |
3242 | * b) a free space cache inode | |
3243 | * We need to distinguish those two, as the snapshot | |
3244 | * orphan must not get deleted. | |
3245 | * find_dead_roots already ran before us, so if this | |
3246 | * is a snapshot deletion, we should find the root | |
a619b3c7 | 3247 | * in the fs_roots radix tree. |
f8e9e0b0 | 3248 | */ |
a619b3c7 RK |
3249 | |
3250 | spin_lock(&fs_info->fs_roots_radix_lock); | |
3251 | dead_root = radix_tree_lookup(&fs_info->fs_roots_radix, | |
3252 | (unsigned long)found_key.objectid); | |
3253 | if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0) | |
3254 | is_dead_root = 1; | |
3255 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
3256 | ||
f8e9e0b0 AJ |
3257 | if (is_dead_root) { |
3258 | /* prevent this orphan from being found again */ | |
3259 | key.offset = found_key.objectid - 1; | |
3260 | continue; | |
3261 | } | |
f7e9e8fc | 3262 | |
f8e9e0b0 | 3263 | } |
f7e9e8fc | 3264 | |
7b128766 | 3265 | /* |
f7e9e8fc OS |
3266 | * If we have an inode with links, there are a couple of |
3267 | * possibilities. Old kernels (before v3.12) used to create an | |
3268 | * orphan item for truncate indicating that there were possibly | |
3269 | * extent items past i_size that needed to be deleted. In v3.12, | |
3270 | * truncate was changed to update i_size in sync with the extent | |
3271 | * items, but the (useless) orphan item was still created. Since | |
3272 | * v4.18, we don't create the orphan item for truncate at all. | |
3273 | * | |
3274 | * So, this item could mean that we need to do a truncate, but | |
3275 | * only if this filesystem was last used on a pre-v3.12 kernel | |
3276 | * and was not cleanly unmounted. The odds of that are quite | |
3277 | * slim, and it's a pain to do the truncate now, so just delete | |
3278 | * the orphan item. | |
3279 | * | |
3280 | * It's also possible that this orphan item was supposed to be | |
3281 | * deleted but wasn't. The inode number may have been reused, | |
3282 | * but either way, we can delete the orphan item. | |
7b128766 | 3283 | */ |
f7e9e8fc OS |
3284 | if (ret == -ENOENT || inode->i_nlink) { |
3285 | if (!ret) | |
3286 | iput(inode); | |
a8c9e576 | 3287 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3288 | if (IS_ERR(trans)) { |
3289 | ret = PTR_ERR(trans); | |
3290 | goto out; | |
3291 | } | |
0b246afa JM |
3292 | btrfs_debug(fs_info, "auto deleting %Lu", |
3293 | found_key.objectid); | |
a8c9e576 JB |
3294 | ret = btrfs_del_orphan_item(trans, root, |
3295 | found_key.objectid); | |
3a45bb20 | 3296 | btrfs_end_transaction(trans); |
4ef31a45 JB |
3297 | if (ret) |
3298 | goto out; | |
7b128766 JB |
3299 | continue; |
3300 | } | |
3301 | ||
f7e9e8fc | 3302 | nr_unlink++; |
7b128766 JB |
3303 | |
3304 | /* this will do delete_inode and everything for us */ | |
3305 | iput(inode); | |
3306 | } | |
3254c876 MX |
3307 | /* release the path since we're done with it */ |
3308 | btrfs_release_path(path); | |
3309 | ||
d68fc57b YZ |
3310 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3311 | ||
a575ceeb | 3312 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { |
7a7eaa40 | 3313 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3314 | if (!IS_ERR(trans)) |
3a45bb20 | 3315 | btrfs_end_transaction(trans); |
d68fc57b | 3316 | } |
7b128766 JB |
3317 | |
3318 | if (nr_unlink) | |
0b246afa | 3319 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
66b4ffd1 JB |
3320 | |
3321 | out: | |
3322 | if (ret) | |
0b246afa | 3323 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3324 | btrfs_free_path(path); |
3325 | return ret; | |
7b128766 JB |
3326 | } |
3327 | ||
46a53cca CM |
3328 | /* |
3329 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3330 | * don't find any xattrs, we know there can't be any acls. | |
3331 | * | |
3332 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3333 | */ | |
3334 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3335 | int slot, u64 objectid, |
3336 | int *first_xattr_slot) | |
46a53cca CM |
3337 | { |
3338 | u32 nritems = btrfs_header_nritems(leaf); | |
3339 | struct btrfs_key found_key; | |
f23b5a59 JB |
3340 | static u64 xattr_access = 0; |
3341 | static u64 xattr_default = 0; | |
46a53cca CM |
3342 | int scanned = 0; |
3343 | ||
f23b5a59 | 3344 | if (!xattr_access) { |
97d79299 AG |
3345 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3346 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3347 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3348 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3349 | } |
3350 | ||
46a53cca | 3351 | slot++; |
63541927 | 3352 | *first_xattr_slot = -1; |
46a53cca CM |
3353 | while (slot < nritems) { |
3354 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3355 | ||
3356 | /* we found a different objectid, there must not be acls */ | |
3357 | if (found_key.objectid != objectid) | |
3358 | return 0; | |
3359 | ||
3360 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3361 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3362 | if (*first_xattr_slot == -1) |
3363 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3364 | if (found_key.offset == xattr_access || |
3365 | found_key.offset == xattr_default) | |
3366 | return 1; | |
3367 | } | |
46a53cca CM |
3368 | |
3369 | /* | |
3370 | * we found a key greater than an xattr key, there can't | |
3371 | * be any acls later on | |
3372 | */ | |
3373 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3374 | return 0; | |
3375 | ||
3376 | slot++; | |
3377 | scanned++; | |
3378 | ||
3379 | /* | |
3380 | * it goes inode, inode backrefs, xattrs, extents, | |
3381 | * so if there are a ton of hard links to an inode there can | |
3382 | * be a lot of backrefs. Don't waste time searching too hard, | |
3383 | * this is just an optimization | |
3384 | */ | |
3385 | if (scanned >= 8) | |
3386 | break; | |
3387 | } | |
3388 | /* we hit the end of the leaf before we found an xattr or | |
3389 | * something larger than an xattr. We have to assume the inode | |
3390 | * has acls | |
3391 | */ | |
63541927 FDBM |
3392 | if (*first_xattr_slot == -1) |
3393 | *first_xattr_slot = slot; | |
46a53cca CM |
3394 | return 1; |
3395 | } | |
3396 | ||
d352ac68 CM |
3397 | /* |
3398 | * read an inode from the btree into the in-memory inode | |
3399 | */ | |
4222ea71 FM |
3400 | static int btrfs_read_locked_inode(struct inode *inode, |
3401 | struct btrfs_path *in_path) | |
39279cc3 | 3402 | { |
0b246afa | 3403 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4222ea71 | 3404 | struct btrfs_path *path = in_path; |
5f39d397 | 3405 | struct extent_buffer *leaf; |
39279cc3 CM |
3406 | struct btrfs_inode_item *inode_item; |
3407 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3408 | struct btrfs_key location; | |
67de1176 | 3409 | unsigned long ptr; |
46a53cca | 3410 | int maybe_acls; |
618e21d5 | 3411 | u32 rdev; |
39279cc3 | 3412 | int ret; |
2f7e33d4 | 3413 | bool filled = false; |
63541927 | 3414 | int first_xattr_slot; |
2f7e33d4 MX |
3415 | |
3416 | ret = btrfs_fill_inode(inode, &rdev); | |
3417 | if (!ret) | |
3418 | filled = true; | |
39279cc3 | 3419 | |
4222ea71 FM |
3420 | if (!path) { |
3421 | path = btrfs_alloc_path(); | |
3422 | if (!path) | |
3423 | return -ENOMEM; | |
3424 | } | |
1748f843 | 3425 | |
39279cc3 | 3426 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3427 | |
39279cc3 | 3428 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 | 3429 | if (ret) { |
4222ea71 FM |
3430 | if (path != in_path) |
3431 | btrfs_free_path(path); | |
f5b3a417 | 3432 | return ret; |
67710892 | 3433 | } |
39279cc3 | 3434 | |
5f39d397 | 3435 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3436 | |
3437 | if (filled) | |
67de1176 | 3438 | goto cache_index; |
2f7e33d4 | 3439 | |
5f39d397 CM |
3440 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3441 | struct btrfs_inode_item); | |
5f39d397 | 3442 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3443 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3444 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3445 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
6ef06d27 | 3446 | btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item)); |
41a2ee75 JB |
3447 | btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0, |
3448 | round_up(i_size_read(inode), fs_info->sectorsize)); | |
5f39d397 | 3449 | |
a937b979 DS |
3450 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3451 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3452 | |
a937b979 DS |
3453 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3454 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3455 | |
a937b979 DS |
3456 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3457 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3458 | |
9cc97d64 | 3459 | BTRFS_I(inode)->i_otime.tv_sec = |
3460 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3461 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3462 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3463 | |
a76a3cd4 | 3464 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3465 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3466 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3467 | ||
c7f88c4e JL |
3468 | inode_set_iversion_queried(inode, |
3469 | btrfs_inode_sequence(leaf, inode_item)); | |
6e17d30b YD |
3470 | inode->i_generation = BTRFS_I(inode)->generation; |
3471 | inode->i_rdev = 0; | |
3472 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3473 | ||
3474 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3475 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3476 | ||
3477 | cache_index: | |
5dc562c5 JB |
3478 | /* |
3479 | * If we were modified in the current generation and evicted from memory | |
3480 | * and then re-read we need to do a full sync since we don't have any | |
3481 | * idea about which extents were modified before we were evicted from | |
3482 | * cache. | |
6e17d30b YD |
3483 | * |
3484 | * This is required for both inode re-read from disk and delayed inode | |
3485 | * in delayed_nodes_tree. | |
5dc562c5 | 3486 | */ |
0b246afa | 3487 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
3488 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3489 | &BTRFS_I(inode)->runtime_flags); | |
3490 | ||
bde6c242 FM |
3491 | /* |
3492 | * We don't persist the id of the transaction where an unlink operation | |
3493 | * against the inode was last made. So here we assume the inode might | |
3494 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3495 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3496 | * between the inode and its parent if the inode is fsync'ed and the log | |
3497 | * replayed. For example, in the scenario: | |
3498 | * | |
3499 | * touch mydir/foo | |
3500 | * ln mydir/foo mydir/bar | |
3501 | * sync | |
3502 | * unlink mydir/bar | |
3503 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3504 | * xfs_io -c fsync mydir/foo | |
3505 | * <power failure> | |
3506 | * mount fs, triggers fsync log replay | |
3507 | * | |
3508 | * We must make sure that when we fsync our inode foo we also log its | |
3509 | * parent inode, otherwise after log replay the parent still has the | |
3510 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3511 | * and doesn't have an inode ref with the name "bar" anymore. | |
3512 | * | |
3513 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3514 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3515 | * transaction commits on fsync if our inode is a directory, or if our |
3516 | * inode is not a directory, logging its parent unnecessarily. | |
3517 | */ | |
3518 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3519 | ||
3ebac17c FM |
3520 | /* |
3521 | * Same logic as for last_unlink_trans. We don't persist the generation | |
3522 | * of the last transaction where this inode was used for a reflink | |
3523 | * operation, so after eviction and reloading the inode we must be | |
3524 | * pessimistic and assume the last transaction that modified the inode. | |
3525 | */ | |
3526 | BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans; | |
3527 | ||
67de1176 MX |
3528 | path->slots[0]++; |
3529 | if (inode->i_nlink != 1 || | |
3530 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3531 | goto cache_acl; | |
3532 | ||
3533 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 3534 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
3535 | goto cache_acl; |
3536 | ||
3537 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3538 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3539 | struct btrfs_inode_ref *ref; | |
3540 | ||
3541 | ref = (struct btrfs_inode_ref *)ptr; | |
3542 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3543 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3544 | struct btrfs_inode_extref *extref; | |
3545 | ||
3546 | extref = (struct btrfs_inode_extref *)ptr; | |
3547 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3548 | extref); | |
3549 | } | |
2f7e33d4 | 3550 | cache_acl: |
46a53cca CM |
3551 | /* |
3552 | * try to precache a NULL acl entry for files that don't have | |
3553 | * any xattrs or acls | |
3554 | */ | |
33345d01 | 3555 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 3556 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
3557 | if (first_xattr_slot != -1) { |
3558 | path->slots[0] = first_xattr_slot; | |
3559 | ret = btrfs_load_inode_props(inode, path); | |
3560 | if (ret) | |
0b246afa | 3561 | btrfs_err(fs_info, |
351fd353 | 3562 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 3563 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
3564 | root->root_key.objectid, ret); |
3565 | } | |
4222ea71 FM |
3566 | if (path != in_path) |
3567 | btrfs_free_path(path); | |
63541927 | 3568 | |
72c04902 AV |
3569 | if (!maybe_acls) |
3570 | cache_no_acl(inode); | |
46a53cca | 3571 | |
39279cc3 | 3572 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3573 | case S_IFREG: |
3574 | inode->i_mapping->a_ops = &btrfs_aops; | |
3575 | inode->i_fop = &btrfs_file_operations; | |
3576 | inode->i_op = &btrfs_file_inode_operations; | |
3577 | break; | |
3578 | case S_IFDIR: | |
3579 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 3580 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
3581 | break; |
3582 | case S_IFLNK: | |
3583 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3584 | inode_nohighmem(inode); |
4779cc04 | 3585 | inode->i_mapping->a_ops = &btrfs_aops; |
39279cc3 | 3586 | break; |
618e21d5 | 3587 | default: |
0279b4cd | 3588 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3589 | init_special_inode(inode, inode->i_mode, rdev); |
3590 | break; | |
39279cc3 | 3591 | } |
6cbff00f | 3592 | |
7b6a221e | 3593 | btrfs_sync_inode_flags_to_i_flags(inode); |
67710892 | 3594 | return 0; |
39279cc3 CM |
3595 | } |
3596 | ||
d352ac68 CM |
3597 | /* |
3598 | * given a leaf and an inode, copy the inode fields into the leaf | |
3599 | */ | |
e02119d5 CM |
3600 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3601 | struct extent_buffer *leaf, | |
5f39d397 | 3602 | struct btrfs_inode_item *item, |
39279cc3 CM |
3603 | struct inode *inode) |
3604 | { | |
51fab693 LB |
3605 | struct btrfs_map_token token; |
3606 | ||
c82f823c | 3607 | btrfs_init_map_token(&token, leaf); |
5f39d397 | 3608 | |
cc4c13d5 DS |
3609 | btrfs_set_token_inode_uid(&token, item, i_uid_read(inode)); |
3610 | btrfs_set_token_inode_gid(&token, item, i_gid_read(inode)); | |
3611 | btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size); | |
3612 | btrfs_set_token_inode_mode(&token, item, inode->i_mode); | |
3613 | btrfs_set_token_inode_nlink(&token, item, inode->i_nlink); | |
3614 | ||
3615 | btrfs_set_token_timespec_sec(&token, &item->atime, | |
3616 | inode->i_atime.tv_sec); | |
3617 | btrfs_set_token_timespec_nsec(&token, &item->atime, | |
3618 | inode->i_atime.tv_nsec); | |
3619 | ||
3620 | btrfs_set_token_timespec_sec(&token, &item->mtime, | |
3621 | inode->i_mtime.tv_sec); | |
3622 | btrfs_set_token_timespec_nsec(&token, &item->mtime, | |
3623 | inode->i_mtime.tv_nsec); | |
3624 | ||
3625 | btrfs_set_token_timespec_sec(&token, &item->ctime, | |
3626 | inode->i_ctime.tv_sec); | |
3627 | btrfs_set_token_timespec_nsec(&token, &item->ctime, | |
3628 | inode->i_ctime.tv_nsec); | |
3629 | ||
3630 | btrfs_set_token_timespec_sec(&token, &item->otime, | |
3631 | BTRFS_I(inode)->i_otime.tv_sec); | |
3632 | btrfs_set_token_timespec_nsec(&token, &item->otime, | |
3633 | BTRFS_I(inode)->i_otime.tv_nsec); | |
3634 | ||
3635 | btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode)); | |
3636 | btrfs_set_token_inode_generation(&token, item, | |
3637 | BTRFS_I(inode)->generation); | |
3638 | btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode)); | |
3639 | btrfs_set_token_inode_transid(&token, item, trans->transid); | |
3640 | btrfs_set_token_inode_rdev(&token, item, inode->i_rdev); | |
3641 | btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags); | |
3642 | btrfs_set_token_inode_block_group(&token, item, 0); | |
39279cc3 CM |
3643 | } |
3644 | ||
d352ac68 CM |
3645 | /* |
3646 | * copy everything in the in-memory inode into the btree. | |
3647 | */ | |
2115133f | 3648 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
dfeb9e7c NB |
3649 | struct btrfs_root *root, |
3650 | struct btrfs_inode *inode) | |
39279cc3 CM |
3651 | { |
3652 | struct btrfs_inode_item *inode_item; | |
3653 | struct btrfs_path *path; | |
5f39d397 | 3654 | struct extent_buffer *leaf; |
39279cc3 CM |
3655 | int ret; |
3656 | ||
3657 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3658 | if (!path) |
3659 | return -ENOMEM; | |
3660 | ||
dfeb9e7c | 3661 | ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1); |
39279cc3 CM |
3662 | if (ret) { |
3663 | if (ret > 0) | |
3664 | ret = -ENOENT; | |
3665 | goto failed; | |
3666 | } | |
3667 | ||
5f39d397 CM |
3668 | leaf = path->nodes[0]; |
3669 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3670 | struct btrfs_inode_item); |
39279cc3 | 3671 | |
dfeb9e7c | 3672 | fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode); |
5f39d397 | 3673 | btrfs_mark_buffer_dirty(leaf); |
dfeb9e7c | 3674 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3675 | ret = 0; |
3676 | failed: | |
39279cc3 CM |
3677 | btrfs_free_path(path); |
3678 | return ret; | |
3679 | } | |
3680 | ||
2115133f CM |
3681 | /* |
3682 | * copy everything in the in-memory inode into the btree. | |
3683 | */ | |
3684 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
9a56fcd1 NB |
3685 | struct btrfs_root *root, |
3686 | struct btrfs_inode *inode) | |
2115133f | 3687 | { |
0b246afa | 3688 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
3689 | int ret; |
3690 | ||
3691 | /* | |
3692 | * If the inode is a free space inode, we can deadlock during commit | |
3693 | * if we put it into the delayed code. | |
3694 | * | |
3695 | * The data relocation inode should also be directly updated | |
3696 | * without delay | |
3697 | */ | |
9a56fcd1 | 3698 | if (!btrfs_is_free_space_inode(inode) |
1d52c78a | 3699 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
0b246afa | 3700 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
3701 | btrfs_update_root_times(trans, root); |
3702 | ||
9a56fcd1 | 3703 | ret = btrfs_delayed_update_inode(trans, root, inode); |
2115133f | 3704 | if (!ret) |
9a56fcd1 | 3705 | btrfs_set_inode_last_trans(trans, inode); |
2115133f CM |
3706 | return ret; |
3707 | } | |
3708 | ||
9a56fcd1 | 3709 | return btrfs_update_inode_item(trans, root, inode); |
2115133f CM |
3710 | } |
3711 | ||
729f7961 NB |
3712 | int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3713 | struct btrfs_root *root, struct btrfs_inode *inode) | |
2115133f CM |
3714 | { |
3715 | int ret; | |
3716 | ||
729f7961 | 3717 | ret = btrfs_update_inode(trans, root, inode); |
2115133f | 3718 | if (ret == -ENOSPC) |
729f7961 | 3719 | return btrfs_update_inode_item(trans, root, inode); |
2115133f CM |
3720 | return ret; |
3721 | } | |
3722 | ||
d352ac68 CM |
3723 | /* |
3724 | * unlink helper that gets used here in inode.c and in the tree logging | |
3725 | * recovery code. It remove a link in a directory with a given name, and | |
3726 | * also drops the back refs in the inode to the directory | |
3727 | */ | |
92986796 AV |
3728 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3729 | struct btrfs_root *root, | |
4ec5934e NB |
3730 | struct btrfs_inode *dir, |
3731 | struct btrfs_inode *inode, | |
92986796 | 3732 | const char *name, int name_len) |
39279cc3 | 3733 | { |
0b246afa | 3734 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 3735 | struct btrfs_path *path; |
39279cc3 | 3736 | int ret = 0; |
39279cc3 | 3737 | struct btrfs_dir_item *di; |
aec7477b | 3738 | u64 index; |
33345d01 LZ |
3739 | u64 ino = btrfs_ino(inode); |
3740 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3741 | |
3742 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3743 | if (!path) { |
3744 | ret = -ENOMEM; | |
554233a6 | 3745 | goto out; |
54aa1f4d CM |
3746 | } |
3747 | ||
33345d01 | 3748 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 | 3749 | name, name_len, -1); |
3cf5068f LB |
3750 | if (IS_ERR_OR_NULL(di)) { |
3751 | ret = di ? PTR_ERR(di) : -ENOENT; | |
39279cc3 CM |
3752 | goto err; |
3753 | } | |
39279cc3 | 3754 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3755 | if (ret) |
3756 | goto err; | |
b3b4aa74 | 3757 | btrfs_release_path(path); |
39279cc3 | 3758 | |
67de1176 MX |
3759 | /* |
3760 | * If we don't have dir index, we have to get it by looking up | |
3761 | * the inode ref, since we get the inode ref, remove it directly, | |
3762 | * it is unnecessary to do delayed deletion. | |
3763 | * | |
3764 | * But if we have dir index, needn't search inode ref to get it. | |
3765 | * Since the inode ref is close to the inode item, it is better | |
3766 | * that we delay to delete it, and just do this deletion when | |
3767 | * we update the inode item. | |
3768 | */ | |
4ec5934e | 3769 | if (inode->dir_index) { |
67de1176 MX |
3770 | ret = btrfs_delayed_delete_inode_ref(inode); |
3771 | if (!ret) { | |
4ec5934e | 3772 | index = inode->dir_index; |
67de1176 MX |
3773 | goto skip_backref; |
3774 | } | |
3775 | } | |
3776 | ||
33345d01 LZ |
3777 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
3778 | dir_ino, &index); | |
aec7477b | 3779 | if (ret) { |
0b246afa | 3780 | btrfs_info(fs_info, |
c2cf52eb | 3781 | "failed to delete reference to %.*s, inode %llu parent %llu", |
c1c9ff7c | 3782 | name_len, name, ino, dir_ino); |
66642832 | 3783 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
3784 | goto err; |
3785 | } | |
67de1176 | 3786 | skip_backref: |
9add2945 | 3787 | ret = btrfs_delete_delayed_dir_index(trans, dir, index); |
79787eaa | 3788 | if (ret) { |
66642832 | 3789 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 3790 | goto err; |
79787eaa | 3791 | } |
39279cc3 | 3792 | |
4ec5934e NB |
3793 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode, |
3794 | dir_ino); | |
79787eaa | 3795 | if (ret != 0 && ret != -ENOENT) { |
66642832 | 3796 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3797 | goto err; |
3798 | } | |
e02119d5 | 3799 | |
4ec5934e NB |
3800 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, |
3801 | index); | |
6418c961 CM |
3802 | if (ret == -ENOENT) |
3803 | ret = 0; | |
d4e3991b | 3804 | else if (ret) |
66642832 | 3805 | btrfs_abort_transaction(trans, ret); |
63611e73 JB |
3806 | |
3807 | /* | |
3808 | * If we have a pending delayed iput we could end up with the final iput | |
3809 | * being run in btrfs-cleaner context. If we have enough of these built | |
3810 | * up we can end up burning a lot of time in btrfs-cleaner without any | |
3811 | * way to throttle the unlinks. Since we're currently holding a ref on | |
3812 | * the inode we can run the delayed iput here without any issues as the | |
3813 | * final iput won't be done until after we drop the ref we're currently | |
3814 | * holding. | |
3815 | */ | |
3816 | btrfs_run_delayed_iput(fs_info, inode); | |
39279cc3 CM |
3817 | err: |
3818 | btrfs_free_path(path); | |
e02119d5 CM |
3819 | if (ret) |
3820 | goto out; | |
3821 | ||
6ef06d27 | 3822 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2); |
4ec5934e NB |
3823 | inode_inc_iversion(&inode->vfs_inode); |
3824 | inode_inc_iversion(&dir->vfs_inode); | |
3825 | inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime = | |
3826 | dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode); | |
9a56fcd1 | 3827 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 3828 | out: |
39279cc3 CM |
3829 | return ret; |
3830 | } | |
3831 | ||
92986796 AV |
3832 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3833 | struct btrfs_root *root, | |
4ec5934e | 3834 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
92986796 AV |
3835 | const char *name, int name_len) |
3836 | { | |
3837 | int ret; | |
3838 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
3839 | if (!ret) { | |
4ec5934e | 3840 | drop_nlink(&inode->vfs_inode); |
9a56fcd1 | 3841 | ret = btrfs_update_inode(trans, root, inode); |
92986796 AV |
3842 | } |
3843 | return ret; | |
3844 | } | |
39279cc3 | 3845 | |
a22285a6 YZ |
3846 | /* |
3847 | * helper to start transaction for unlink and rmdir. | |
3848 | * | |
d52be818 JB |
3849 | * unlink and rmdir are special in btrfs, they do not always free space, so |
3850 | * if we cannot make our reservations the normal way try and see if there is | |
3851 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
3852 | * allow the unlink to occur. | |
a22285a6 | 3853 | */ |
d52be818 | 3854 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 3855 | { |
a22285a6 | 3856 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 3857 | |
e70bea5f JB |
3858 | /* |
3859 | * 1 for the possible orphan item | |
3860 | * 1 for the dir item | |
3861 | * 1 for the dir index | |
3862 | * 1 for the inode ref | |
e70bea5f JB |
3863 | * 1 for the inode |
3864 | */ | |
7f9fe614 | 3865 | return btrfs_start_transaction_fallback_global_rsv(root, 5); |
a22285a6 YZ |
3866 | } |
3867 | ||
3868 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
3869 | { | |
3870 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3871 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 3872 | struct inode *inode = d_inode(dentry); |
a22285a6 | 3873 | int ret; |
a22285a6 | 3874 | |
d52be818 | 3875 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
3876 | if (IS_ERR(trans)) |
3877 | return PTR_ERR(trans); | |
5f39d397 | 3878 | |
4ec5934e NB |
3879 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
3880 | 0); | |
12fcfd22 | 3881 | |
4ec5934e NB |
3882 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
3883 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
3884 | dentry->d_name.len); | |
b532402e TI |
3885 | if (ret) |
3886 | goto out; | |
7b128766 | 3887 | |
a22285a6 | 3888 | if (inode->i_nlink == 0) { |
73f2e545 | 3889 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
b532402e TI |
3890 | if (ret) |
3891 | goto out; | |
a22285a6 | 3892 | } |
7b128766 | 3893 | |
b532402e | 3894 | out: |
3a45bb20 | 3895 | btrfs_end_transaction(trans); |
2ff7e61e | 3896 | btrfs_btree_balance_dirty(root->fs_info); |
39279cc3 CM |
3897 | return ret; |
3898 | } | |
3899 | ||
f60a2364 | 3900 | static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
045d3967 | 3901 | struct inode *dir, struct dentry *dentry) |
4df27c4d | 3902 | { |
401b3b19 | 3903 | struct btrfs_root *root = BTRFS_I(dir)->root; |
045d3967 | 3904 | struct btrfs_inode *inode = BTRFS_I(d_inode(dentry)); |
4df27c4d YZ |
3905 | struct btrfs_path *path; |
3906 | struct extent_buffer *leaf; | |
3907 | struct btrfs_dir_item *di; | |
3908 | struct btrfs_key key; | |
045d3967 JB |
3909 | const char *name = dentry->d_name.name; |
3910 | int name_len = dentry->d_name.len; | |
4df27c4d YZ |
3911 | u64 index; |
3912 | int ret; | |
045d3967 | 3913 | u64 objectid; |
4a0cc7ca | 3914 | u64 dir_ino = btrfs_ino(BTRFS_I(dir)); |
4df27c4d | 3915 | |
045d3967 JB |
3916 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) { |
3917 | objectid = inode->root->root_key.objectid; | |
3918 | } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) { | |
3919 | objectid = inode->location.objectid; | |
3920 | } else { | |
3921 | WARN_ON(1); | |
3922 | return -EINVAL; | |
3923 | } | |
3924 | ||
4df27c4d YZ |
3925 | path = btrfs_alloc_path(); |
3926 | if (!path) | |
3927 | return -ENOMEM; | |
3928 | ||
33345d01 | 3929 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3930 | name, name_len, -1); |
79787eaa | 3931 | if (IS_ERR_OR_NULL(di)) { |
3cf5068f | 3932 | ret = di ? PTR_ERR(di) : -ENOENT; |
79787eaa JM |
3933 | goto out; |
3934 | } | |
4df27c4d YZ |
3935 | |
3936 | leaf = path->nodes[0]; | |
3937 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3938 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3939 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 3940 | if (ret) { |
66642832 | 3941 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3942 | goto out; |
3943 | } | |
b3b4aa74 | 3944 | btrfs_release_path(path); |
4df27c4d | 3945 | |
d49d3287 JB |
3946 | /* |
3947 | * This is a placeholder inode for a subvolume we didn't have a | |
3948 | * reference to at the time of the snapshot creation. In the meantime | |
3949 | * we could have renamed the real subvol link into our snapshot, so | |
3950 | * depending on btrfs_del_root_ref to return -ENOENT here is incorret. | |
3951 | * Instead simply lookup the dir_index_item for this entry so we can | |
3952 | * remove it. Otherwise we know we have a ref to the root and we can | |
3953 | * call btrfs_del_root_ref, and it _shouldn't_ fail. | |
3954 | */ | |
3955 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) { | |
33345d01 | 3956 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3957 | name, name_len); |
79787eaa JM |
3958 | if (IS_ERR_OR_NULL(di)) { |
3959 | if (!di) | |
3960 | ret = -ENOENT; | |
3961 | else | |
3962 | ret = PTR_ERR(di); | |
66642832 | 3963 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3964 | goto out; |
3965 | } | |
4df27c4d YZ |
3966 | |
3967 | leaf = path->nodes[0]; | |
3968 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4df27c4d | 3969 | index = key.offset; |
d49d3287 JB |
3970 | btrfs_release_path(path); |
3971 | } else { | |
3972 | ret = btrfs_del_root_ref(trans, objectid, | |
3973 | root->root_key.objectid, dir_ino, | |
3974 | &index, name, name_len); | |
3975 | if (ret) { | |
3976 | btrfs_abort_transaction(trans, ret); | |
3977 | goto out; | |
3978 | } | |
4df27c4d YZ |
3979 | } |
3980 | ||
9add2945 | 3981 | ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index); |
79787eaa | 3982 | if (ret) { |
66642832 | 3983 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3984 | goto out; |
3985 | } | |
4df27c4d | 3986 | |
6ef06d27 | 3987 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2); |
0c4d2d95 | 3988 | inode_inc_iversion(dir); |
c2050a45 | 3989 | dir->i_mtime = dir->i_ctime = current_time(dir); |
729f7961 | 3990 | ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir)); |
79787eaa | 3991 | if (ret) |
66642832 | 3992 | btrfs_abort_transaction(trans, ret); |
79787eaa | 3993 | out: |
71d7aed0 | 3994 | btrfs_free_path(path); |
79787eaa | 3995 | return ret; |
4df27c4d YZ |
3996 | } |
3997 | ||
ec42f167 MT |
3998 | /* |
3999 | * Helper to check if the subvolume references other subvolumes or if it's | |
4000 | * default. | |
4001 | */ | |
f60a2364 | 4002 | static noinline int may_destroy_subvol(struct btrfs_root *root) |
ec42f167 MT |
4003 | { |
4004 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4005 | struct btrfs_path *path; | |
4006 | struct btrfs_dir_item *di; | |
4007 | struct btrfs_key key; | |
4008 | u64 dir_id; | |
4009 | int ret; | |
4010 | ||
4011 | path = btrfs_alloc_path(); | |
4012 | if (!path) | |
4013 | return -ENOMEM; | |
4014 | ||
4015 | /* Make sure this root isn't set as the default subvol */ | |
4016 | dir_id = btrfs_super_root_dir(fs_info->super_copy); | |
4017 | di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path, | |
4018 | dir_id, "default", 7, 0); | |
4019 | if (di && !IS_ERR(di)) { | |
4020 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
4021 | if (key.objectid == root->root_key.objectid) { | |
4022 | ret = -EPERM; | |
4023 | btrfs_err(fs_info, | |
4024 | "deleting default subvolume %llu is not allowed", | |
4025 | key.objectid); | |
4026 | goto out; | |
4027 | } | |
4028 | btrfs_release_path(path); | |
4029 | } | |
4030 | ||
4031 | key.objectid = root->root_key.objectid; | |
4032 | key.type = BTRFS_ROOT_REF_KEY; | |
4033 | key.offset = (u64)-1; | |
4034 | ||
4035 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); | |
4036 | if (ret < 0) | |
4037 | goto out; | |
4038 | BUG_ON(ret == 0); | |
4039 | ||
4040 | ret = 0; | |
4041 | if (path->slots[0] > 0) { | |
4042 | path->slots[0]--; | |
4043 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
4044 | if (key.objectid == root->root_key.objectid && | |
4045 | key.type == BTRFS_ROOT_REF_KEY) | |
4046 | ret = -ENOTEMPTY; | |
4047 | } | |
4048 | out: | |
4049 | btrfs_free_path(path); | |
4050 | return ret; | |
4051 | } | |
4052 | ||
20a68004 NB |
4053 | /* Delete all dentries for inodes belonging to the root */ |
4054 | static void btrfs_prune_dentries(struct btrfs_root *root) | |
4055 | { | |
4056 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4057 | struct rb_node *node; | |
4058 | struct rb_node *prev; | |
4059 | struct btrfs_inode *entry; | |
4060 | struct inode *inode; | |
4061 | u64 objectid = 0; | |
4062 | ||
4063 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) | |
4064 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
4065 | ||
4066 | spin_lock(&root->inode_lock); | |
4067 | again: | |
4068 | node = root->inode_tree.rb_node; | |
4069 | prev = NULL; | |
4070 | while (node) { | |
4071 | prev = node; | |
4072 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4073 | ||
37508515 | 4074 | if (objectid < btrfs_ino(entry)) |
20a68004 | 4075 | node = node->rb_left; |
37508515 | 4076 | else if (objectid > btrfs_ino(entry)) |
20a68004 NB |
4077 | node = node->rb_right; |
4078 | else | |
4079 | break; | |
4080 | } | |
4081 | if (!node) { | |
4082 | while (prev) { | |
4083 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
37508515 | 4084 | if (objectid <= btrfs_ino(entry)) { |
20a68004 NB |
4085 | node = prev; |
4086 | break; | |
4087 | } | |
4088 | prev = rb_next(prev); | |
4089 | } | |
4090 | } | |
4091 | while (node) { | |
4092 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
37508515 | 4093 | objectid = btrfs_ino(entry) + 1; |
20a68004 NB |
4094 | inode = igrab(&entry->vfs_inode); |
4095 | if (inode) { | |
4096 | spin_unlock(&root->inode_lock); | |
4097 | if (atomic_read(&inode->i_count) > 1) | |
4098 | d_prune_aliases(inode); | |
4099 | /* | |
4100 | * btrfs_drop_inode will have it removed from the inode | |
4101 | * cache when its usage count hits zero. | |
4102 | */ | |
4103 | iput(inode); | |
4104 | cond_resched(); | |
4105 | spin_lock(&root->inode_lock); | |
4106 | goto again; | |
4107 | } | |
4108 | ||
4109 | if (cond_resched_lock(&root->inode_lock)) | |
4110 | goto again; | |
4111 | ||
4112 | node = rb_next(node); | |
4113 | } | |
4114 | spin_unlock(&root->inode_lock); | |
4115 | } | |
4116 | ||
f60a2364 MT |
4117 | int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry) |
4118 | { | |
4119 | struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb); | |
4120 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4121 | struct inode *inode = d_inode(dentry); | |
4122 | struct btrfs_root *dest = BTRFS_I(inode)->root; | |
4123 | struct btrfs_trans_handle *trans; | |
4124 | struct btrfs_block_rsv block_rsv; | |
4125 | u64 root_flags; | |
f60a2364 | 4126 | int ret; |
f60a2364 MT |
4127 | |
4128 | /* | |
4129 | * Don't allow to delete a subvolume with send in progress. This is | |
4130 | * inside the inode lock so the error handling that has to drop the bit | |
4131 | * again is not run concurrently. | |
4132 | */ | |
4133 | spin_lock(&dest->root_item_lock); | |
a7176f74 | 4134 | if (dest->send_in_progress) { |
f60a2364 MT |
4135 | spin_unlock(&dest->root_item_lock); |
4136 | btrfs_warn(fs_info, | |
4137 | "attempt to delete subvolume %llu during send", | |
4138 | dest->root_key.objectid); | |
4139 | return -EPERM; | |
4140 | } | |
a7176f74 LF |
4141 | root_flags = btrfs_root_flags(&dest->root_item); |
4142 | btrfs_set_root_flags(&dest->root_item, | |
4143 | root_flags | BTRFS_ROOT_SUBVOL_DEAD); | |
4144 | spin_unlock(&dest->root_item_lock); | |
f60a2364 MT |
4145 | |
4146 | down_write(&fs_info->subvol_sem); | |
4147 | ||
ee0d904f NB |
4148 | ret = may_destroy_subvol(dest); |
4149 | if (ret) | |
f60a2364 MT |
4150 | goto out_up_write; |
4151 | ||
4152 | btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); | |
4153 | /* | |
4154 | * One for dir inode, | |
4155 | * two for dir entries, | |
4156 | * two for root ref/backref. | |
4157 | */ | |
ee0d904f NB |
4158 | ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true); |
4159 | if (ret) | |
f60a2364 MT |
4160 | goto out_up_write; |
4161 | ||
4162 | trans = btrfs_start_transaction(root, 0); | |
4163 | if (IS_ERR(trans)) { | |
ee0d904f | 4164 | ret = PTR_ERR(trans); |
f60a2364 MT |
4165 | goto out_release; |
4166 | } | |
4167 | trans->block_rsv = &block_rsv; | |
4168 | trans->bytes_reserved = block_rsv.size; | |
4169 | ||
4170 | btrfs_record_snapshot_destroy(trans, BTRFS_I(dir)); | |
4171 | ||
045d3967 | 4172 | ret = btrfs_unlink_subvol(trans, dir, dentry); |
f60a2364 | 4173 | if (ret) { |
f60a2364 MT |
4174 | btrfs_abort_transaction(trans, ret); |
4175 | goto out_end_trans; | |
4176 | } | |
4177 | ||
4178 | btrfs_record_root_in_trans(trans, dest); | |
4179 | ||
4180 | memset(&dest->root_item.drop_progress, 0, | |
4181 | sizeof(dest->root_item.drop_progress)); | |
c8422684 | 4182 | btrfs_set_root_drop_level(&dest->root_item, 0); |
f60a2364 MT |
4183 | btrfs_set_root_refs(&dest->root_item, 0); |
4184 | ||
4185 | if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) { | |
4186 | ret = btrfs_insert_orphan_item(trans, | |
4187 | fs_info->tree_root, | |
4188 | dest->root_key.objectid); | |
4189 | if (ret) { | |
4190 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4191 | goto out_end_trans; |
4192 | } | |
4193 | } | |
4194 | ||
d1957791 | 4195 | ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid, |
f60a2364 MT |
4196 | BTRFS_UUID_KEY_SUBVOL, |
4197 | dest->root_key.objectid); | |
4198 | if (ret && ret != -ENOENT) { | |
4199 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4200 | goto out_end_trans; |
4201 | } | |
4202 | if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) { | |
d1957791 | 4203 | ret = btrfs_uuid_tree_remove(trans, |
f60a2364 MT |
4204 | dest->root_item.received_uuid, |
4205 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4206 | dest->root_key.objectid); | |
4207 | if (ret && ret != -ENOENT) { | |
4208 | btrfs_abort_transaction(trans, ret); | |
f60a2364 MT |
4209 | goto out_end_trans; |
4210 | } | |
4211 | } | |
4212 | ||
082b6c97 QW |
4213 | free_anon_bdev(dest->anon_dev); |
4214 | dest->anon_dev = 0; | |
f60a2364 MT |
4215 | out_end_trans: |
4216 | trans->block_rsv = NULL; | |
4217 | trans->bytes_reserved = 0; | |
4218 | ret = btrfs_end_transaction(trans); | |
f60a2364 MT |
4219 | inode->i_flags |= S_DEAD; |
4220 | out_release: | |
e85fde51 | 4221 | btrfs_subvolume_release_metadata(root, &block_rsv); |
f60a2364 MT |
4222 | out_up_write: |
4223 | up_write(&fs_info->subvol_sem); | |
ee0d904f | 4224 | if (ret) { |
f60a2364 MT |
4225 | spin_lock(&dest->root_item_lock); |
4226 | root_flags = btrfs_root_flags(&dest->root_item); | |
4227 | btrfs_set_root_flags(&dest->root_item, | |
4228 | root_flags & ~BTRFS_ROOT_SUBVOL_DEAD); | |
4229 | spin_unlock(&dest->root_item_lock); | |
4230 | } else { | |
4231 | d_invalidate(dentry); | |
20a68004 | 4232 | btrfs_prune_dentries(dest); |
f60a2364 | 4233 | ASSERT(dest->send_in_progress == 0); |
f60a2364 MT |
4234 | } |
4235 | ||
ee0d904f | 4236 | return ret; |
f60a2364 MT |
4237 | } |
4238 | ||
39279cc3 CM |
4239 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4240 | { | |
2b0143b5 | 4241 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4242 | int err = 0; |
39279cc3 | 4243 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4244 | struct btrfs_trans_handle *trans; |
44f714da | 4245 | u64 last_unlink_trans; |
39279cc3 | 4246 | |
b3ae244e | 4247 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4248 | return -ENOTEMPTY; |
4a0cc7ca | 4249 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) |
a79a464d | 4250 | return btrfs_delete_subvolume(dir, dentry); |
134d4512 | 4251 | |
d52be818 | 4252 | trans = __unlink_start_trans(dir); |
a22285a6 | 4253 | if (IS_ERR(trans)) |
5df6a9f6 | 4254 | return PTR_ERR(trans); |
5df6a9f6 | 4255 | |
4a0cc7ca | 4256 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
045d3967 | 4257 | err = btrfs_unlink_subvol(trans, dir, dentry); |
4df27c4d YZ |
4258 | goto out; |
4259 | } | |
4260 | ||
73f2e545 | 4261 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
7b128766 | 4262 | if (err) |
4df27c4d | 4263 | goto out; |
7b128766 | 4264 | |
44f714da FM |
4265 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4266 | ||
39279cc3 | 4267 | /* now the directory is empty */ |
4ec5934e NB |
4268 | err = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4269 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4270 | dentry->d_name.len); | |
44f714da | 4271 | if (!err) { |
6ef06d27 | 4272 | btrfs_i_size_write(BTRFS_I(inode), 0); |
44f714da FM |
4273 | /* |
4274 | * Propagate the last_unlink_trans value of the deleted dir to | |
4275 | * its parent directory. This is to prevent an unrecoverable | |
4276 | * log tree in the case we do something like this: | |
4277 | * 1) create dir foo | |
4278 | * 2) create snapshot under dir foo | |
4279 | * 3) delete the snapshot | |
4280 | * 4) rmdir foo | |
4281 | * 5) mkdir foo | |
4282 | * 6) fsync foo or some file inside foo | |
4283 | */ | |
4284 | if (last_unlink_trans >= trans->transid) | |
4285 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4286 | } | |
4df27c4d | 4287 | out: |
3a45bb20 | 4288 | btrfs_end_transaction(trans); |
2ff7e61e | 4289 | btrfs_btree_balance_dirty(root->fs_info); |
3954401f | 4290 | |
39279cc3 CM |
4291 | return err; |
4292 | } | |
4293 | ||
ddfae63c JB |
4294 | /* |
4295 | * Return this if we need to call truncate_block for the last bit of the | |
4296 | * truncate. | |
4297 | */ | |
4298 | #define NEED_TRUNCATE_BLOCK 1 | |
0305cd5f | 4299 | |
39279cc3 CM |
4300 | /* |
4301 | * this can truncate away extent items, csum items and directory items. | |
4302 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4303 | * any higher than new_size |
39279cc3 CM |
4304 | * |
4305 | * csum items that cross the new i_size are truncated to the new size | |
4306 | * as well. | |
7b128766 JB |
4307 | * |
4308 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4309 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4310 | */ |
8082510e YZ |
4311 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4312 | struct btrfs_root *root, | |
50743398 | 4313 | struct btrfs_inode *inode, |
8082510e | 4314 | u64 new_size, u32 min_type) |
39279cc3 | 4315 | { |
0b246afa | 4316 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4317 | struct btrfs_path *path; |
5f39d397 | 4318 | struct extent_buffer *leaf; |
39279cc3 | 4319 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4320 | struct btrfs_key key; |
4321 | struct btrfs_key found_key; | |
39279cc3 | 4322 | u64 extent_start = 0; |
db94535d | 4323 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4324 | u64 extent_offset = 0; |
39279cc3 | 4325 | u64 item_end = 0; |
c1aa4575 | 4326 | u64 last_size = new_size; |
8082510e | 4327 | u32 found_type = (u8)-1; |
39279cc3 CM |
4328 | int found_extent; |
4329 | int del_item; | |
85e21bac CM |
4330 | int pending_del_nr = 0; |
4331 | int pending_del_slot = 0; | |
179e29e4 | 4332 | int extent_type = -1; |
8082510e | 4333 | int ret; |
50743398 | 4334 | u64 ino = btrfs_ino(inode); |
28ed1345 | 4335 | u64 bytes_deleted = 0; |
897ca819 TM |
4336 | bool be_nice = false; |
4337 | bool should_throttle = false; | |
28553fa9 FM |
4338 | const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize); |
4339 | struct extent_state *cached_state = NULL; | |
8082510e YZ |
4340 | |
4341 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4342 | |
28ed1345 | 4343 | /* |
92a7cc42 QW |
4344 | * For non-free space inodes and non-shareable roots, we want to back |
4345 | * off from time to time. This means all inodes in subvolume roots, | |
4346 | * reloc roots, and data reloc roots. | |
28ed1345 | 4347 | */ |
50743398 | 4348 | if (!btrfs_is_free_space_inode(inode) && |
92a7cc42 | 4349 | test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
897ca819 | 4350 | be_nice = true; |
28ed1345 | 4351 | |
0eb0e19c MF |
4352 | path = btrfs_alloc_path(); |
4353 | if (!path) | |
4354 | return -ENOMEM; | |
e4058b54 | 4355 | path->reada = READA_BACK; |
0eb0e19c | 4356 | |
82028e0a | 4357 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
50743398 | 4358 | lock_extent_bits(&inode->io_tree, lock_start, (u64)-1, |
a5ae50de | 4359 | &cached_state); |
28553fa9 | 4360 | |
82028e0a QW |
4361 | /* |
4362 | * We want to drop from the next block forward in case this | |
4363 | * new size is not block aligned since we will be keeping the | |
4364 | * last block of the extent just the way it is. | |
4365 | */ | |
50743398 | 4366 | btrfs_drop_extent_cache(inode, ALIGN(new_size, |
0b246afa | 4367 | fs_info->sectorsize), |
da17066c | 4368 | (u64)-1, 0); |
82028e0a | 4369 | } |
8082510e | 4370 | |
16cdcec7 MX |
4371 | /* |
4372 | * This function is also used to drop the items in the log tree before | |
4373 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
52042d8e | 4374 | * it is used to drop the logged items. So we shouldn't kill the delayed |
16cdcec7 MX |
4375 | * items. |
4376 | */ | |
50743398 NB |
4377 | if (min_type == 0 && root == inode->root) |
4378 | btrfs_kill_delayed_inode_items(inode); | |
16cdcec7 | 4379 | |
33345d01 | 4380 | key.objectid = ino; |
39279cc3 | 4381 | key.offset = (u64)-1; |
5f39d397 CM |
4382 | key.type = (u8)-1; |
4383 | ||
85e21bac | 4384 | search_again: |
28ed1345 CM |
4385 | /* |
4386 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4387 | * up a huge file in a single leaf. Most of the time that | |
4388 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4389 | */ | |
fd86a3a3 OS |
4390 | if (be_nice && bytes_deleted > SZ_32M && |
4391 | btrfs_should_end_transaction(trans)) { | |
4392 | ret = -EAGAIN; | |
4393 | goto out; | |
28ed1345 CM |
4394 | } |
4395 | ||
85e21bac | 4396 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
fd86a3a3 | 4397 | if (ret < 0) |
8082510e | 4398 | goto out; |
d397712b | 4399 | |
85e21bac | 4400 | if (ret > 0) { |
fd86a3a3 | 4401 | ret = 0; |
e02119d5 CM |
4402 | /* there are no items in the tree for us to truncate, we're |
4403 | * done | |
4404 | */ | |
8082510e YZ |
4405 | if (path->slots[0] == 0) |
4406 | goto out; | |
85e21bac CM |
4407 | path->slots[0]--; |
4408 | } | |
4409 | ||
d397712b | 4410 | while (1) { |
9ddc959e JB |
4411 | u64 clear_start = 0, clear_len = 0; |
4412 | ||
39279cc3 | 4413 | fi = NULL; |
5f39d397 CM |
4414 | leaf = path->nodes[0]; |
4415 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4416 | found_type = found_key.type; |
39279cc3 | 4417 | |
33345d01 | 4418 | if (found_key.objectid != ino) |
39279cc3 | 4419 | break; |
5f39d397 | 4420 | |
85e21bac | 4421 | if (found_type < min_type) |
39279cc3 CM |
4422 | break; |
4423 | ||
5f39d397 | 4424 | item_end = found_key.offset; |
39279cc3 | 4425 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4426 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4427 | struct btrfs_file_extent_item); |
179e29e4 CM |
4428 | extent_type = btrfs_file_extent_type(leaf, fi); |
4429 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4430 | item_end += |
db94535d | 4431 | btrfs_file_extent_num_bytes(leaf, fi); |
09ed2f16 LB |
4432 | |
4433 | trace_btrfs_truncate_show_fi_regular( | |
50743398 | 4434 | inode, leaf, fi, found_key.offset); |
179e29e4 | 4435 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
e41ca589 QW |
4436 | item_end += btrfs_file_extent_ram_bytes(leaf, |
4437 | fi); | |
09ed2f16 LB |
4438 | |
4439 | trace_btrfs_truncate_show_fi_inline( | |
50743398 | 4440 | inode, leaf, fi, path->slots[0], |
09ed2f16 | 4441 | found_key.offset); |
39279cc3 | 4442 | } |
008630c1 | 4443 | item_end--; |
39279cc3 | 4444 | } |
8082510e YZ |
4445 | if (found_type > min_type) { |
4446 | del_item = 1; | |
4447 | } else { | |
76b42abb | 4448 | if (item_end < new_size) |
b888db2b | 4449 | break; |
8082510e YZ |
4450 | if (found_key.offset >= new_size) |
4451 | del_item = 1; | |
4452 | else | |
4453 | del_item = 0; | |
39279cc3 | 4454 | } |
39279cc3 | 4455 | found_extent = 0; |
39279cc3 | 4456 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4457 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4458 | goto delete; | |
4459 | ||
4460 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 4461 | u64 num_dec; |
9ddc959e JB |
4462 | |
4463 | clear_start = found_key.offset; | |
db94535d | 4464 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4465 | if (!del_item) { |
db94535d CM |
4466 | u64 orig_num_bytes = |
4467 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4468 | extent_num_bytes = ALIGN(new_size - |
4469 | found_key.offset, | |
0b246afa | 4470 | fs_info->sectorsize); |
9ddc959e | 4471 | clear_start = ALIGN(new_size, fs_info->sectorsize); |
db94535d CM |
4472 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4473 | extent_num_bytes); | |
4474 | num_dec = (orig_num_bytes - | |
9069218d | 4475 | extent_num_bytes); |
92a7cc42 | 4476 | if (test_bit(BTRFS_ROOT_SHAREABLE, |
27cdeb70 MX |
4477 | &root->state) && |
4478 | extent_start != 0) | |
50743398 NB |
4479 | inode_sub_bytes(&inode->vfs_inode, |
4480 | num_dec); | |
5f39d397 | 4481 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4482 | } else { |
db94535d CM |
4483 | extent_num_bytes = |
4484 | btrfs_file_extent_disk_num_bytes(leaf, | |
4485 | fi); | |
5d4f98a2 YZ |
4486 | extent_offset = found_key.offset - |
4487 | btrfs_file_extent_offset(leaf, fi); | |
4488 | ||
39279cc3 | 4489 | /* FIXME blocksize != 4096 */ |
9069218d | 4490 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4491 | if (extent_start != 0) { |
4492 | found_extent = 1; | |
92a7cc42 | 4493 | if (test_bit(BTRFS_ROOT_SHAREABLE, |
27cdeb70 | 4494 | &root->state)) |
50743398 NB |
4495 | inode_sub_bytes(&inode->vfs_inode, |
4496 | num_dec); | |
e02119d5 | 4497 | } |
39279cc3 | 4498 | } |
9ddc959e | 4499 | clear_len = num_dec; |
9069218d | 4500 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4501 | /* |
4502 | * we can't truncate inline items that have had | |
4503 | * special encodings | |
4504 | */ | |
4505 | if (!del_item && | |
c8b97818 | 4506 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
ddfae63c JB |
4507 | btrfs_file_extent_other_encoding(leaf, fi) == 0 && |
4508 | btrfs_file_extent_compression(leaf, fi) == 0) { | |
4509 | u32 size = (u32)(new_size - found_key.offset); | |
4510 | ||
4511 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4512 | size = btrfs_file_extent_calc_inline_size(size); | |
78ac4f9e | 4513 | btrfs_truncate_item(path, size, 1); |
ddfae63c | 4514 | } else if (!del_item) { |
514ac8ad | 4515 | /* |
ddfae63c JB |
4516 | * We have to bail so the last_size is set to |
4517 | * just before this extent. | |
514ac8ad | 4518 | */ |
fd86a3a3 | 4519 | ret = NEED_TRUNCATE_BLOCK; |
ddfae63c | 4520 | break; |
9ddc959e JB |
4521 | } else { |
4522 | /* | |
4523 | * Inline extents are special, we just treat | |
4524 | * them as a full sector worth in the file | |
4525 | * extent tree just for simplicity sake. | |
4526 | */ | |
4527 | clear_len = fs_info->sectorsize; | |
ddfae63c | 4528 | } |
0305cd5f | 4529 | |
92a7cc42 | 4530 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
50743398 NB |
4531 | inode_sub_bytes(&inode->vfs_inode, |
4532 | item_end + 1 - new_size); | |
39279cc3 | 4533 | } |
179e29e4 | 4534 | delete: |
9ddc959e JB |
4535 | /* |
4536 | * We use btrfs_truncate_inode_items() to clean up log trees for | |
4537 | * multiple fsyncs, and in this case we don't want to clear the | |
4538 | * file extent range because it's just the log. | |
4539 | */ | |
50743398 NB |
4540 | if (root == inode->root) { |
4541 | ret = btrfs_inode_clear_file_extent_range(inode, | |
9ddc959e JB |
4542 | clear_start, clear_len); |
4543 | if (ret) { | |
4544 | btrfs_abort_transaction(trans, ret); | |
4545 | break; | |
4546 | } | |
4547 | } | |
4548 | ||
ddfae63c JB |
4549 | if (del_item) |
4550 | last_size = found_key.offset; | |
4551 | else | |
4552 | last_size = new_size; | |
39279cc3 | 4553 | if (del_item) { |
85e21bac CM |
4554 | if (!pending_del_nr) { |
4555 | /* no pending yet, add ourselves */ | |
4556 | pending_del_slot = path->slots[0]; | |
4557 | pending_del_nr = 1; | |
4558 | } else if (pending_del_nr && | |
4559 | path->slots[0] + 1 == pending_del_slot) { | |
4560 | /* hop on the pending chunk */ | |
4561 | pending_del_nr++; | |
4562 | pending_del_slot = path->slots[0]; | |
4563 | } else { | |
d397712b | 4564 | BUG(); |
85e21bac | 4565 | } |
39279cc3 CM |
4566 | } else { |
4567 | break; | |
4568 | } | |
897ca819 | 4569 | should_throttle = false; |
28f75a0e | 4570 | |
27cdeb70 | 4571 | if (found_extent && |
82028e0a | 4572 | root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
ffd4bb2a QW |
4573 | struct btrfs_ref ref = { 0 }; |
4574 | ||
28ed1345 | 4575 | bytes_deleted += extent_num_bytes; |
ffd4bb2a QW |
4576 | |
4577 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, | |
4578 | extent_start, extent_num_bytes, 0); | |
4579 | ref.real_root = root->root_key.objectid; | |
4580 | btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), | |
4581 | ino, extent_offset); | |
4582 | ret = btrfs_free_extent(trans, &ref); | |
05522109 OS |
4583 | if (ret) { |
4584 | btrfs_abort_transaction(trans, ret); | |
4585 | break; | |
4586 | } | |
28f75a0e | 4587 | if (be_nice) { |
7c861627 | 4588 | if (btrfs_should_throttle_delayed_refs(trans)) |
897ca819 | 4589 | should_throttle = true; |
28f75a0e | 4590 | } |
39279cc3 | 4591 | } |
85e21bac | 4592 | |
8082510e YZ |
4593 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4594 | break; | |
4595 | ||
4596 | if (path->slots[0] == 0 || | |
1262133b | 4597 | path->slots[0] != pending_del_slot || |
28bad212 | 4598 | should_throttle) { |
8082510e YZ |
4599 | if (pending_del_nr) { |
4600 | ret = btrfs_del_items(trans, root, path, | |
4601 | pending_del_slot, | |
4602 | pending_del_nr); | |
79787eaa | 4603 | if (ret) { |
66642832 | 4604 | btrfs_abort_transaction(trans, ret); |
fd86a3a3 | 4605 | break; |
79787eaa | 4606 | } |
8082510e YZ |
4607 | pending_del_nr = 0; |
4608 | } | |
b3b4aa74 | 4609 | btrfs_release_path(path); |
28bad212 | 4610 | |
28f75a0e | 4611 | /* |
28bad212 JB |
4612 | * We can generate a lot of delayed refs, so we need to |
4613 | * throttle every once and a while and make sure we're | |
4614 | * adding enough space to keep up with the work we are | |
4615 | * generating. Since we hold a transaction here we | |
4616 | * can't flush, and we don't want to FLUSH_LIMIT because | |
4617 | * we could have generated too many delayed refs to | |
4618 | * actually allocate, so just bail if we're short and | |
4619 | * let the normal reservation dance happen higher up. | |
28f75a0e | 4620 | */ |
28bad212 JB |
4621 | if (should_throttle) { |
4622 | ret = btrfs_delayed_refs_rsv_refill(fs_info, | |
4623 | BTRFS_RESERVE_NO_FLUSH); | |
4624 | if (ret) { | |
4625 | ret = -EAGAIN; | |
4626 | break; | |
4627 | } | |
28f75a0e | 4628 | } |
85e21bac | 4629 | goto search_again; |
8082510e YZ |
4630 | } else { |
4631 | path->slots[0]--; | |
85e21bac | 4632 | } |
39279cc3 | 4633 | } |
8082510e | 4634 | out: |
fd86a3a3 OS |
4635 | if (ret >= 0 && pending_del_nr) { |
4636 | int err; | |
4637 | ||
4638 | err = btrfs_del_items(trans, root, path, pending_del_slot, | |
85e21bac | 4639 | pending_del_nr); |
fd86a3a3 OS |
4640 | if (err) { |
4641 | btrfs_abort_transaction(trans, err); | |
4642 | ret = err; | |
4643 | } | |
85e21bac | 4644 | } |
76b42abb FM |
4645 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
4646 | ASSERT(last_size >= new_size); | |
fd86a3a3 | 4647 | if (!ret && last_size > new_size) |
76b42abb | 4648 | last_size = new_size; |
50743398 NB |
4649 | btrfs_inode_safe_disk_i_size_write(inode, last_size); |
4650 | unlock_extent_cached(&inode->io_tree, lock_start, (u64)-1, | |
4651 | &cached_state); | |
76b42abb | 4652 | } |
28ed1345 | 4653 | |
39279cc3 | 4654 | btrfs_free_path(path); |
fd86a3a3 | 4655 | return ret; |
39279cc3 CM |
4656 | } |
4657 | ||
4658 | /* | |
9703fefe | 4659 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4660 | * @inode - inode that we're zeroing |
4661 | * @from - the offset to start zeroing | |
4662 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4663 | * offset | |
4664 | * @front - zero up to the offset instead of from the offset on | |
4665 | * | |
9703fefe | 4666 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4667 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4668 | */ |
217f42eb NB |
4669 | int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len, |
4670 | int front) | |
39279cc3 | 4671 | { |
217f42eb NB |
4672 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
4673 | struct address_space *mapping = inode->vfs_inode.i_mapping; | |
4674 | struct extent_io_tree *io_tree = &inode->io_tree; | |
e6dcd2dc | 4675 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 4676 | struct extent_state *cached_state = NULL; |
364ecf36 | 4677 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc | 4678 | char *kaddr; |
6d4572a9 | 4679 | bool only_release_metadata = false; |
0b246afa | 4680 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 4681 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4682 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4683 | struct page *page; |
3b16a4e3 | 4684 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
6d4572a9 | 4685 | size_t write_bytes = blocksize; |
39279cc3 | 4686 | int ret = 0; |
9703fefe CR |
4687 | u64 block_start; |
4688 | u64 block_end; | |
39279cc3 | 4689 | |
b03ebd99 NB |
4690 | if (IS_ALIGNED(offset, blocksize) && |
4691 | (!len || IS_ALIGNED(len, blocksize))) | |
39279cc3 | 4692 | goto out; |
9703fefe | 4693 | |
8b62f87b JB |
4694 | block_start = round_down(from, blocksize); |
4695 | block_end = block_start + blocksize - 1; | |
4696 | ||
217f42eb NB |
4697 | ret = btrfs_check_data_free_space(inode, &data_reserved, block_start, |
4698 | blocksize); | |
6d4572a9 | 4699 | if (ret < 0) { |
217f42eb | 4700 | if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) { |
6d4572a9 QW |
4701 | /* For nocow case, no need to reserve data space */ |
4702 | only_release_metadata = true; | |
4703 | } else { | |
4704 | goto out; | |
4705 | } | |
4706 | } | |
217f42eb | 4707 | ret = btrfs_delalloc_reserve_metadata(inode, blocksize); |
6d4572a9 QW |
4708 | if (ret < 0) { |
4709 | if (!only_release_metadata) | |
217f42eb NB |
4710 | btrfs_free_reserved_data_space(inode, data_reserved, |
4711 | block_start, blocksize); | |
6d4572a9 QW |
4712 | goto out; |
4713 | } | |
211c17f5 | 4714 | again: |
3b16a4e3 | 4715 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4716 | if (!page) { |
217f42eb NB |
4717 | btrfs_delalloc_release_space(inode, data_reserved, block_start, |
4718 | blocksize, true); | |
4719 | btrfs_delalloc_release_extents(inode, blocksize); | |
ac6a2b36 | 4720 | ret = -ENOMEM; |
39279cc3 | 4721 | goto out; |
5d5e103a | 4722 | } |
e6dcd2dc | 4723 | |
39279cc3 | 4724 | if (!PageUptodate(page)) { |
9ebefb18 | 4725 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4726 | lock_page(page); |
211c17f5 CM |
4727 | if (page->mapping != mapping) { |
4728 | unlock_page(page); | |
09cbfeaf | 4729 | put_page(page); |
211c17f5 CM |
4730 | goto again; |
4731 | } | |
39279cc3 CM |
4732 | if (!PageUptodate(page)) { |
4733 | ret = -EIO; | |
89642229 | 4734 | goto out_unlock; |
39279cc3 CM |
4735 | } |
4736 | } | |
211c17f5 | 4737 | wait_on_page_writeback(page); |
e6dcd2dc | 4738 | |
9703fefe | 4739 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc CM |
4740 | set_page_extent_mapped(page); |
4741 | ||
217f42eb | 4742 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4743 | if (ordered) { |
9703fefe | 4744 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 4745 | &cached_state); |
e6dcd2dc | 4746 | unlock_page(page); |
09cbfeaf | 4747 | put_page(page); |
c0a43603 | 4748 | btrfs_start_ordered_extent(ordered, 1); |
e6dcd2dc CM |
4749 | btrfs_put_ordered_extent(ordered); |
4750 | goto again; | |
4751 | } | |
4752 | ||
217f42eb | 4753 | clear_extent_bit(&inode->io_tree, block_start, block_end, |
e182163d OS |
4754 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, |
4755 | 0, 0, &cached_state); | |
5d5e103a | 4756 | |
217f42eb | 4757 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0, |
330a5827 | 4758 | &cached_state); |
9ed74f2d | 4759 | if (ret) { |
9703fefe | 4760 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 4761 | &cached_state); |
9ed74f2d JB |
4762 | goto out_unlock; |
4763 | } | |
4764 | ||
9703fefe | 4765 | if (offset != blocksize) { |
2aaa6655 | 4766 | if (!len) |
9703fefe | 4767 | len = blocksize - offset; |
e6dcd2dc | 4768 | kaddr = kmap(page); |
2aaa6655 | 4769 | if (front) |
9703fefe CR |
4770 | memset(kaddr + (block_start - page_offset(page)), |
4771 | 0, offset); | |
2aaa6655 | 4772 | else |
9703fefe CR |
4773 | memset(kaddr + (block_start - page_offset(page)) + offset, |
4774 | 0, len); | |
e6dcd2dc CM |
4775 | flush_dcache_page(page); |
4776 | kunmap(page); | |
4777 | } | |
247e743c | 4778 | ClearPageChecked(page); |
e6dcd2dc | 4779 | set_page_dirty(page); |
e43bbe5e | 4780 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state); |
39279cc3 | 4781 | |
6d4572a9 | 4782 | if (only_release_metadata) |
217f42eb | 4783 | set_extent_bit(&inode->io_tree, block_start, block_end, |
1cab5e72 | 4784 | EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL); |
6d4572a9 | 4785 | |
89642229 | 4786 | out_unlock: |
6d4572a9 QW |
4787 | if (ret) { |
4788 | if (only_release_metadata) | |
217f42eb | 4789 | btrfs_delalloc_release_metadata(inode, blocksize, true); |
6d4572a9 | 4790 | else |
217f42eb | 4791 | btrfs_delalloc_release_space(inode, data_reserved, |
6d4572a9 QW |
4792 | block_start, blocksize, true); |
4793 | } | |
217f42eb | 4794 | btrfs_delalloc_release_extents(inode, blocksize); |
39279cc3 | 4795 | unlock_page(page); |
09cbfeaf | 4796 | put_page(page); |
39279cc3 | 4797 | out: |
6d4572a9 | 4798 | if (only_release_metadata) |
217f42eb | 4799 | btrfs_check_nocow_unlock(inode); |
364ecf36 | 4800 | extent_changeset_free(data_reserved); |
39279cc3 CM |
4801 | return ret; |
4802 | } | |
4803 | ||
a4ba6cc0 | 4804 | static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode, |
16e7549f JB |
4805 | u64 offset, u64 len) |
4806 | { | |
a4ba6cc0 | 4807 | struct btrfs_fs_info *fs_info = root->fs_info; |
16e7549f | 4808 | struct btrfs_trans_handle *trans; |
5893dfb9 | 4809 | struct btrfs_drop_extents_args drop_args = { 0 }; |
16e7549f JB |
4810 | int ret; |
4811 | ||
4812 | /* | |
4813 | * Still need to make sure the inode looks like it's been updated so | |
4814 | * that any holes get logged if we fsync. | |
4815 | */ | |
0b246afa | 4816 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) { |
a4ba6cc0 NB |
4817 | inode->last_trans = fs_info->generation; |
4818 | inode->last_sub_trans = root->log_transid; | |
4819 | inode->last_log_commit = root->last_log_commit; | |
16e7549f JB |
4820 | return 0; |
4821 | } | |
4822 | ||
4823 | /* | |
4824 | * 1 - for the one we're dropping | |
4825 | * 1 - for the one we're adding | |
4826 | * 1 - for updating the inode. | |
4827 | */ | |
4828 | trans = btrfs_start_transaction(root, 3); | |
4829 | if (IS_ERR(trans)) | |
4830 | return PTR_ERR(trans); | |
4831 | ||
5893dfb9 FM |
4832 | drop_args.start = offset; |
4833 | drop_args.end = offset + len; | |
4834 | drop_args.drop_cache = true; | |
4835 | ||
a4ba6cc0 | 4836 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
16e7549f | 4837 | if (ret) { |
66642832 | 4838 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4839 | btrfs_end_transaction(trans); |
16e7549f JB |
4840 | return ret; |
4841 | } | |
4842 | ||
a4ba6cc0 | 4843 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), |
f85b7379 | 4844 | offset, 0, 0, len, 0, len, 0, 0, 0); |
2766ff61 | 4845 | if (ret) { |
66642832 | 4846 | btrfs_abort_transaction(trans, ret); |
2766ff61 | 4847 | } else { |
a4ba6cc0 NB |
4848 | btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found); |
4849 | btrfs_update_inode(trans, root, inode); | |
2766ff61 | 4850 | } |
3a45bb20 | 4851 | btrfs_end_transaction(trans); |
16e7549f JB |
4852 | return ret; |
4853 | } | |
4854 | ||
695a0d0d JB |
4855 | /* |
4856 | * This function puts in dummy file extents for the area we're creating a hole | |
4857 | * for. So if we are truncating this file to a larger size we need to insert | |
4858 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4859 | * the range between oldsize and size | |
4860 | */ | |
b06359a3 | 4861 | int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4862 | { |
b06359a3 NB |
4863 | struct btrfs_root *root = inode->root; |
4864 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4865 | struct extent_io_tree *io_tree = &inode->io_tree; | |
a22285a6 | 4866 | struct extent_map *em = NULL; |
2ac55d41 | 4867 | struct extent_state *cached_state = NULL; |
b06359a3 | 4868 | struct extent_map_tree *em_tree = &inode->extent_tree; |
0b246afa JM |
4869 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
4870 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
4871 | u64 last_byte; |
4872 | u64 cur_offset; | |
4873 | u64 hole_size; | |
9ed74f2d | 4874 | int err = 0; |
39279cc3 | 4875 | |
a71754fc | 4876 | /* |
9703fefe CR |
4877 | * If our size started in the middle of a block we need to zero out the |
4878 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
4879 | * expose stale data. |
4880 | */ | |
b06359a3 | 4881 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
4882 | if (err) |
4883 | return err; | |
4884 | ||
9036c102 YZ |
4885 | if (size <= hole_start) |
4886 | return 0; | |
4887 | ||
b06359a3 NB |
4888 | btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1, |
4889 | &cached_state); | |
9036c102 YZ |
4890 | cur_offset = hole_start; |
4891 | while (1) { | |
b06359a3 | 4892 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, |
39b07b5d | 4893 | block_end - cur_offset); |
79787eaa JM |
4894 | if (IS_ERR(em)) { |
4895 | err = PTR_ERR(em); | |
f2767956 | 4896 | em = NULL; |
79787eaa JM |
4897 | break; |
4898 | } | |
9036c102 | 4899 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 4900 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
9ddc959e JB |
4901 | hole_size = last_byte - cur_offset; |
4902 | ||
8082510e | 4903 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4904 | struct extent_map *hole_em; |
9ed74f2d | 4905 | |
b06359a3 NB |
4906 | err = maybe_insert_hole(root, inode, cur_offset, |
4907 | hole_size); | |
16e7549f | 4908 | if (err) |
3893e33b | 4909 | break; |
9ddc959e | 4910 | |
b06359a3 | 4911 | err = btrfs_inode_set_file_extent_range(inode, |
9ddc959e JB |
4912 | cur_offset, hole_size); |
4913 | if (err) | |
4914 | break; | |
4915 | ||
b06359a3 | 4916 | btrfs_drop_extent_cache(inode, cur_offset, |
5dc562c5 JB |
4917 | cur_offset + hole_size - 1, 0); |
4918 | hole_em = alloc_extent_map(); | |
4919 | if (!hole_em) { | |
4920 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
b06359a3 | 4921 | &inode->runtime_flags); |
5dc562c5 JB |
4922 | goto next; |
4923 | } | |
4924 | hole_em->start = cur_offset; | |
4925 | hole_em->len = hole_size; | |
4926 | hole_em->orig_start = cur_offset; | |
8082510e | 4927 | |
5dc562c5 JB |
4928 | hole_em->block_start = EXTENT_MAP_HOLE; |
4929 | hole_em->block_len = 0; | |
b4939680 | 4930 | hole_em->orig_block_len = 0; |
cc95bef6 | 4931 | hole_em->ram_bytes = hole_size; |
5dc562c5 | 4932 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 4933 | hole_em->generation = fs_info->generation; |
8082510e | 4934 | |
5dc562c5 JB |
4935 | while (1) { |
4936 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4937 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4938 | write_unlock(&em_tree->lock); |
4939 | if (err != -EEXIST) | |
4940 | break; | |
b06359a3 | 4941 | btrfs_drop_extent_cache(inode, cur_offset, |
5dc562c5 JB |
4942 | cur_offset + |
4943 | hole_size - 1, 0); | |
4944 | } | |
4945 | free_extent_map(hole_em); | |
9ddc959e | 4946 | } else { |
b06359a3 | 4947 | err = btrfs_inode_set_file_extent_range(inode, |
9ddc959e JB |
4948 | cur_offset, hole_size); |
4949 | if (err) | |
4950 | break; | |
9036c102 | 4951 | } |
16e7549f | 4952 | next: |
9036c102 | 4953 | free_extent_map(em); |
a22285a6 | 4954 | em = NULL; |
9036c102 | 4955 | cur_offset = last_byte; |
8082510e | 4956 | if (cur_offset >= block_end) |
9036c102 YZ |
4957 | break; |
4958 | } | |
a22285a6 | 4959 | free_extent_map(em); |
e43bbe5e | 4960 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state); |
9036c102 YZ |
4961 | return err; |
4962 | } | |
39279cc3 | 4963 | |
3972f260 | 4964 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4965 | { |
f4a2f4c5 MX |
4966 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4967 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4968 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4969 | loff_t newsize = attr->ia_size; |
4970 | int mask = attr->ia_valid; | |
8082510e YZ |
4971 | int ret; |
4972 | ||
3972f260 ES |
4973 | /* |
4974 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4975 | * special case where we need to update the times despite not having | |
4976 | * these flags set. For all other operations the VFS set these flags | |
4977 | * explicitly if it wants a timestamp update. | |
4978 | */ | |
dff6efc3 CH |
4979 | if (newsize != oldsize) { |
4980 | inode_inc_iversion(inode); | |
4981 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
4982 | inode->i_ctime = inode->i_mtime = | |
c2050a45 | 4983 | current_time(inode); |
dff6efc3 | 4984 | } |
3972f260 | 4985 | |
a41ad394 | 4986 | if (newsize > oldsize) { |
9ea24bbe | 4987 | /* |
ea14b57f | 4988 | * Don't do an expanding truncate while snapshotting is ongoing. |
9ea24bbe FM |
4989 | * This is to ensure the snapshot captures a fully consistent |
4990 | * state of this file - if the snapshot captures this expanding | |
4991 | * truncation, it must capture all writes that happened before | |
4992 | * this truncation. | |
4993 | */ | |
dcc3eb96 | 4994 | btrfs_drew_write_lock(&root->snapshot_lock); |
b06359a3 | 4995 | ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize); |
9ea24bbe | 4996 | if (ret) { |
dcc3eb96 | 4997 | btrfs_drew_write_unlock(&root->snapshot_lock); |
8082510e | 4998 | return ret; |
9ea24bbe | 4999 | } |
8082510e | 5000 | |
f4a2f4c5 | 5001 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe | 5002 | if (IS_ERR(trans)) { |
dcc3eb96 | 5003 | btrfs_drew_write_unlock(&root->snapshot_lock); |
f4a2f4c5 | 5004 | return PTR_ERR(trans); |
9ea24bbe | 5005 | } |
f4a2f4c5 MX |
5006 | |
5007 | i_size_write(inode, newsize); | |
76aea537 | 5008 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
27772b68 | 5009 | pagecache_isize_extended(inode, oldsize, newsize); |
9a56fcd1 | 5010 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
dcc3eb96 | 5011 | btrfs_drew_write_unlock(&root->snapshot_lock); |
3a45bb20 | 5012 | btrfs_end_transaction(trans); |
a41ad394 | 5013 | } else { |
8082510e | 5014 | |
a41ad394 JB |
5015 | /* |
5016 | * We're truncating a file that used to have good data down to | |
1fd4033d NB |
5017 | * zero. Make sure any new writes to the file get on disk |
5018 | * on close. | |
a41ad394 JB |
5019 | */ |
5020 | if (newsize == 0) | |
1fd4033d | 5021 | set_bit(BTRFS_INODE_FLUSH_ON_CLOSE, |
72ac3c0d | 5022 | &BTRFS_I(inode)->runtime_flags); |
8082510e | 5023 | |
a41ad394 | 5024 | truncate_setsize(inode, newsize); |
2e60a51e | 5025 | |
2e60a51e | 5026 | inode_dio_wait(inode); |
2e60a51e | 5027 | |
213e8c55 | 5028 | ret = btrfs_truncate(inode, newsize == oldsize); |
7f4f6e0a JB |
5029 | if (ret && inode->i_nlink) { |
5030 | int err; | |
5031 | ||
5032 | /* | |
f7e9e8fc OS |
5033 | * Truncate failed, so fix up the in-memory size. We |
5034 | * adjusted disk_i_size down as we removed extents, so | |
5035 | * wait for disk_i_size to be stable and then update the | |
5036 | * in-memory size to match. | |
7f4f6e0a | 5037 | */ |
f7e9e8fc | 5038 | err = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
7f4f6e0a | 5039 | if (err) |
f7e9e8fc OS |
5040 | return err; |
5041 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
7f4f6e0a | 5042 | } |
8082510e YZ |
5043 | } |
5044 | ||
a41ad394 | 5045 | return ret; |
8082510e YZ |
5046 | } |
5047 | ||
9036c102 YZ |
5048 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
5049 | { | |
2b0143b5 | 5050 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5051 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5052 | int err; |
39279cc3 | 5053 | |
b83cc969 LZ |
5054 | if (btrfs_root_readonly(root)) |
5055 | return -EROFS; | |
5056 | ||
31051c85 | 5057 | err = setattr_prepare(dentry, attr); |
9036c102 YZ |
5058 | if (err) |
5059 | return err; | |
2bf5a725 | 5060 | |
5a3f23d5 | 5061 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5062 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5063 | if (err) |
5064 | return err; | |
39279cc3 | 5065 | } |
9036c102 | 5066 | |
1025774c CH |
5067 | if (attr->ia_valid) { |
5068 | setattr_copy(inode, attr); | |
0c4d2d95 | 5069 | inode_inc_iversion(inode); |
22c44fe6 | 5070 | err = btrfs_dirty_inode(inode); |
1025774c | 5071 | |
22c44fe6 | 5072 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 5073 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 5074 | } |
33268eaf | 5075 | |
39279cc3 CM |
5076 | return err; |
5077 | } | |
61295eb8 | 5078 | |
131e404a FDBM |
5079 | /* |
5080 | * While truncating the inode pages during eviction, we get the VFS calling | |
5081 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
5082 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
5083 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
5084 | * extent_state structures over and over, wasting lots of time. | |
5085 | * | |
5086 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
5087 | * those expensive operations on a per page basis and do only the ordered io | |
5088 | * finishing, while we release here the extent_map and extent_state structures, | |
5089 | * without the excessive merging and splitting. | |
5090 | */ | |
5091 | static void evict_inode_truncate_pages(struct inode *inode) | |
5092 | { | |
5093 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5094 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5095 | struct rb_node *node; | |
5096 | ||
5097 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5098 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5099 | |
5100 | write_lock(&map_tree->lock); | |
07e1ce09 | 5101 | while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) { |
131e404a FDBM |
5102 | struct extent_map *em; |
5103 | ||
07e1ce09 | 5104 | node = rb_first_cached(&map_tree->map); |
131e404a | 5105 | em = rb_entry(node, struct extent_map, rb_node); |
180589ef WS |
5106 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5107 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5108 | remove_extent_mapping(map_tree, em); |
5109 | free_extent_map(em); | |
7064dd5c FM |
5110 | if (need_resched()) { |
5111 | write_unlock(&map_tree->lock); | |
5112 | cond_resched(); | |
5113 | write_lock(&map_tree->lock); | |
5114 | } | |
131e404a FDBM |
5115 | } |
5116 | write_unlock(&map_tree->lock); | |
5117 | ||
6ca07097 FM |
5118 | /* |
5119 | * Keep looping until we have no more ranges in the io tree. | |
ba206a02 MWO |
5120 | * We can have ongoing bios started by readahead that have |
5121 | * their endio callback (extent_io.c:end_bio_extent_readpage) | |
9c6429d9 FM |
5122 | * still in progress (unlocked the pages in the bio but did not yet |
5123 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5124 | * ranges can still be locked and eviction started because before |
5125 | * submitting those bios, which are executed by a separate task (work | |
5126 | * queue kthread), inode references (inode->i_count) were not taken | |
5127 | * (which would be dropped in the end io callback of each bio). | |
5128 | * Therefore here we effectively end up waiting for those bios and | |
5129 | * anyone else holding locked ranges without having bumped the inode's | |
5130 | * reference count - if we don't do it, when they access the inode's | |
5131 | * io_tree to unlock a range it may be too late, leading to an | |
5132 | * use-after-free issue. | |
5133 | */ | |
131e404a FDBM |
5134 | spin_lock(&io_tree->lock); |
5135 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5136 | struct extent_state *state; | |
5137 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5138 | u64 start; |
5139 | u64 end; | |
421f0922 | 5140 | unsigned state_flags; |
131e404a FDBM |
5141 | |
5142 | node = rb_first(&io_tree->state); | |
5143 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5144 | start = state->start; |
5145 | end = state->end; | |
421f0922 | 5146 | state_flags = state->state; |
131e404a FDBM |
5147 | spin_unlock(&io_tree->lock); |
5148 | ||
ff13db41 | 5149 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5150 | |
5151 | /* | |
5152 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5153 | * and its reserved space won't be freed by delayed_ref. | |
5154 | * So we need to free its reserved space here. | |
5155 | * (Refer to comment in btrfs_invalidatepage, case 2) | |
5156 | * | |
5157 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5158 | */ | |
421f0922 | 5159 | if (state_flags & EXTENT_DELALLOC) |
8b8a979f NB |
5160 | btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start, |
5161 | end - start + 1); | |
b9d0b389 | 5162 | |
6ca07097 | 5163 | clear_extent_bit(io_tree, start, end, |
e182163d OS |
5164 | EXTENT_LOCKED | EXTENT_DELALLOC | |
5165 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, | |
5166 | &cached_state); | |
131e404a | 5167 | |
7064dd5c | 5168 | cond_resched(); |
131e404a FDBM |
5169 | spin_lock(&io_tree->lock); |
5170 | } | |
5171 | spin_unlock(&io_tree->lock); | |
5172 | } | |
5173 | ||
4b9d7b59 | 5174 | static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root, |
ad80cf50 | 5175 | struct btrfs_block_rsv *rsv) |
4b9d7b59 OS |
5176 | { |
5177 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5178 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d3984c90 | 5179 | struct btrfs_trans_handle *trans; |
2bd36e7b | 5180 | u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1); |
d3984c90 | 5181 | int ret; |
4b9d7b59 | 5182 | |
d3984c90 JB |
5183 | /* |
5184 | * Eviction should be taking place at some place safe because of our | |
5185 | * delayed iputs. However the normal flushing code will run delayed | |
5186 | * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock. | |
5187 | * | |
5188 | * We reserve the delayed_refs_extra here again because we can't use | |
5189 | * btrfs_start_transaction(root, 0) for the same deadlocky reason as | |
5190 | * above. We reserve our extra bit here because we generate a ton of | |
5191 | * delayed refs activity by truncating. | |
5192 | * | |
5193 | * If we cannot make our reservation we'll attempt to steal from the | |
5194 | * global reserve, because we really want to be able to free up space. | |
5195 | */ | |
5196 | ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra, | |
5197 | BTRFS_RESERVE_FLUSH_EVICT); | |
5198 | if (ret) { | |
4b9d7b59 OS |
5199 | /* |
5200 | * Try to steal from the global reserve if there is space for | |
5201 | * it. | |
5202 | */ | |
d3984c90 JB |
5203 | if (btrfs_check_space_for_delayed_refs(fs_info) || |
5204 | btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) { | |
5205 | btrfs_warn(fs_info, | |
5206 | "could not allocate space for delete; will truncate on mount"); | |
5207 | return ERR_PTR(-ENOSPC); | |
5208 | } | |
5209 | delayed_refs_extra = 0; | |
5210 | } | |
4b9d7b59 | 5211 | |
d3984c90 JB |
5212 | trans = btrfs_join_transaction(root); |
5213 | if (IS_ERR(trans)) | |
5214 | return trans; | |
5215 | ||
5216 | if (delayed_refs_extra) { | |
5217 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5218 | trans->bytes_reserved = delayed_refs_extra; | |
5219 | btrfs_block_rsv_migrate(rsv, trans->block_rsv, | |
5220 | delayed_refs_extra, 1); | |
4b9d7b59 | 5221 | } |
d3984c90 | 5222 | return trans; |
4b9d7b59 OS |
5223 | } |
5224 | ||
bd555975 | 5225 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5226 | { |
0b246afa | 5227 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5228 | struct btrfs_trans_handle *trans; |
5229 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b9d7b59 | 5230 | struct btrfs_block_rsv *rsv; |
39279cc3 CM |
5231 | int ret; |
5232 | ||
1abe9b8a | 5233 | trace_btrfs_inode_evict(inode); |
5234 | ||
3d48d981 | 5235 | if (!root) { |
e8f1bc14 | 5236 | clear_inode(inode); |
3d48d981 NB |
5237 | return; |
5238 | } | |
5239 | ||
131e404a FDBM |
5240 | evict_inode_truncate_pages(inode); |
5241 | ||
69e9c6c6 SB |
5242 | if (inode->i_nlink && |
5243 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5244 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
70ddc553 | 5245 | btrfs_is_free_space_inode(BTRFS_I(inode)))) |
bd555975 AV |
5246 | goto no_delete; |
5247 | ||
27919067 | 5248 | if (is_bad_inode(inode)) |
39279cc3 | 5249 | goto no_delete; |
5f39d397 | 5250 | |
7ab7956e | 5251 | btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1); |
f612496b | 5252 | |
7b40b695 | 5253 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
c71bf099 | 5254 | goto no_delete; |
c71bf099 | 5255 | |
76dda93c | 5256 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5257 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5258 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5259 | goto no_delete; |
5260 | } | |
5261 | ||
aa79021f | 5262 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
27919067 | 5263 | if (ret) |
0e8c36a9 | 5264 | goto no_delete; |
0e8c36a9 | 5265 | |
2ff7e61e | 5266 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
27919067 | 5267 | if (!rsv) |
4289a667 | 5268 | goto no_delete; |
2bd36e7b | 5269 | rsv->size = btrfs_calc_metadata_size(fs_info, 1); |
ca7e70f5 | 5270 | rsv->failfast = 1; |
4289a667 | 5271 | |
6ef06d27 | 5272 | btrfs_i_size_write(BTRFS_I(inode), 0); |
5f39d397 | 5273 | |
8082510e | 5274 | while (1) { |
ad80cf50 | 5275 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5276 | if (IS_ERR(trans)) |
5277 | goto free_rsv; | |
7b128766 | 5278 | |
4289a667 JB |
5279 | trans->block_rsv = rsv; |
5280 | ||
50743398 NB |
5281 | ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode), |
5282 | 0, 0); | |
27919067 OS |
5283 | trans->block_rsv = &fs_info->trans_block_rsv; |
5284 | btrfs_end_transaction(trans); | |
5285 | btrfs_btree_balance_dirty(fs_info); | |
5286 | if (ret && ret != -ENOSPC && ret != -EAGAIN) | |
5287 | goto free_rsv; | |
5288 | else if (!ret) | |
8082510e | 5289 | break; |
8082510e | 5290 | } |
5f39d397 | 5291 | |
4ef31a45 | 5292 | /* |
27919067 OS |
5293 | * Errors here aren't a big deal, it just means we leave orphan items in |
5294 | * the tree. They will be cleaned up on the next mount. If the inode | |
5295 | * number gets reused, cleanup deletes the orphan item without doing | |
5296 | * anything, and unlink reuses the existing orphan item. | |
5297 | * | |
5298 | * If it turns out that we are dropping too many of these, we might want | |
5299 | * to add a mechanism for retrying these after a commit. | |
4ef31a45 | 5300 | */ |
ad80cf50 | 5301 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5302 | if (!IS_ERR(trans)) { |
5303 | trans->block_rsv = rsv; | |
5304 | btrfs_orphan_del(trans, BTRFS_I(inode)); | |
5305 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5306 | btrfs_end_transaction(trans); | |
5307 | } | |
54aa1f4d | 5308 | |
27919067 OS |
5309 | free_rsv: |
5310 | btrfs_free_block_rsv(fs_info, rsv); | |
39279cc3 | 5311 | no_delete: |
27919067 OS |
5312 | /* |
5313 | * If we didn't successfully delete, the orphan item will still be in | |
5314 | * the tree and we'll retry on the next mount. Again, we might also want | |
5315 | * to retry these periodically in the future. | |
5316 | */ | |
f48d1cf5 | 5317 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
dbd5768f | 5318 | clear_inode(inode); |
39279cc3 CM |
5319 | } |
5320 | ||
5321 | /* | |
6bf9e4bd QW |
5322 | * Return the key found in the dir entry in the location pointer, fill @type |
5323 | * with BTRFS_FT_*, and return 0. | |
5324 | * | |
005d6712 SY |
5325 | * If no dir entries were found, returns -ENOENT. |
5326 | * If found a corrupted location in dir entry, returns -EUCLEAN. | |
39279cc3 CM |
5327 | */ |
5328 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
6bf9e4bd | 5329 | struct btrfs_key *location, u8 *type) |
39279cc3 CM |
5330 | { |
5331 | const char *name = dentry->d_name.name; | |
5332 | int namelen = dentry->d_name.len; | |
5333 | struct btrfs_dir_item *di; | |
5334 | struct btrfs_path *path; | |
5335 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5336 | int ret = 0; |
39279cc3 CM |
5337 | |
5338 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5339 | if (!path) |
5340 | return -ENOMEM; | |
3954401f | 5341 | |
f85b7379 DS |
5342 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)), |
5343 | name, namelen, 0); | |
3cf5068f LB |
5344 | if (IS_ERR_OR_NULL(di)) { |
5345 | ret = di ? PTR_ERR(di) : -ENOENT; | |
005d6712 SY |
5346 | goto out; |
5347 | } | |
d397712b | 5348 | |
5f39d397 | 5349 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
56a0e706 LB |
5350 | if (location->type != BTRFS_INODE_ITEM_KEY && |
5351 | location->type != BTRFS_ROOT_ITEM_KEY) { | |
005d6712 | 5352 | ret = -EUCLEAN; |
56a0e706 LB |
5353 | btrfs_warn(root->fs_info, |
5354 | "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))", | |
5355 | __func__, name, btrfs_ino(BTRFS_I(dir)), | |
5356 | location->objectid, location->type, location->offset); | |
56a0e706 | 5357 | } |
6bf9e4bd QW |
5358 | if (!ret) |
5359 | *type = btrfs_dir_type(path->nodes[0], di); | |
39279cc3 | 5360 | out: |
39279cc3 CM |
5361 | btrfs_free_path(path); |
5362 | return ret; | |
5363 | } | |
5364 | ||
5365 | /* | |
5366 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5367 | * needs to be changed to reflect the root directory of the tree root. This | |
5368 | * is kind of like crossing a mount point. | |
5369 | */ | |
2ff7e61e | 5370 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
4df27c4d YZ |
5371 | struct inode *dir, |
5372 | struct dentry *dentry, | |
5373 | struct btrfs_key *location, | |
5374 | struct btrfs_root **sub_root) | |
39279cc3 | 5375 | { |
4df27c4d YZ |
5376 | struct btrfs_path *path; |
5377 | struct btrfs_root *new_root; | |
5378 | struct btrfs_root_ref *ref; | |
5379 | struct extent_buffer *leaf; | |
1d4c08e0 | 5380 | struct btrfs_key key; |
4df27c4d YZ |
5381 | int ret; |
5382 | int err = 0; | |
39279cc3 | 5383 | |
4df27c4d YZ |
5384 | path = btrfs_alloc_path(); |
5385 | if (!path) { | |
5386 | err = -ENOMEM; | |
5387 | goto out; | |
5388 | } | |
39279cc3 | 5389 | |
4df27c4d | 5390 | err = -ENOENT; |
1d4c08e0 DS |
5391 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5392 | key.type = BTRFS_ROOT_REF_KEY; | |
5393 | key.offset = location->objectid; | |
5394 | ||
0b246afa | 5395 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5396 | if (ret) { |
5397 | if (ret < 0) | |
5398 | err = ret; | |
5399 | goto out; | |
5400 | } | |
39279cc3 | 5401 | |
4df27c4d YZ |
5402 | leaf = path->nodes[0]; |
5403 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
4a0cc7ca | 5404 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) || |
4df27c4d YZ |
5405 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5406 | goto out; | |
39279cc3 | 5407 | |
4df27c4d YZ |
5408 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5409 | (unsigned long)(ref + 1), | |
5410 | dentry->d_name.len); | |
5411 | if (ret) | |
5412 | goto out; | |
5413 | ||
b3b4aa74 | 5414 | btrfs_release_path(path); |
4df27c4d | 5415 | |
56e9357a | 5416 | new_root = btrfs_get_fs_root(fs_info, location->objectid, true); |
4df27c4d YZ |
5417 | if (IS_ERR(new_root)) { |
5418 | err = PTR_ERR(new_root); | |
5419 | goto out; | |
5420 | } | |
5421 | ||
4df27c4d YZ |
5422 | *sub_root = new_root; |
5423 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5424 | location->type = BTRFS_INODE_ITEM_KEY; | |
5425 | location->offset = 0; | |
5426 | err = 0; | |
5427 | out: | |
5428 | btrfs_free_path(path); | |
5429 | return err; | |
39279cc3 CM |
5430 | } |
5431 | ||
5d4f98a2 YZ |
5432 | static void inode_tree_add(struct inode *inode) |
5433 | { | |
5434 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5435 | struct btrfs_inode *entry; | |
03e860bd FNP |
5436 | struct rb_node **p; |
5437 | struct rb_node *parent; | |
cef21937 | 5438 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
4a0cc7ca | 5439 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 5440 | |
1d3382cb | 5441 | if (inode_unhashed(inode)) |
76dda93c | 5442 | return; |
e1409cef | 5443 | parent = NULL; |
5d4f98a2 | 5444 | spin_lock(&root->inode_lock); |
e1409cef | 5445 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5446 | while (*p) { |
5447 | parent = *p; | |
5448 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5449 | ||
37508515 | 5450 | if (ino < btrfs_ino(entry)) |
03e860bd | 5451 | p = &parent->rb_left; |
37508515 | 5452 | else if (ino > btrfs_ino(entry)) |
03e860bd | 5453 | p = &parent->rb_right; |
5d4f98a2 YZ |
5454 | else { |
5455 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5456 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5457 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5458 | RB_CLEAR_NODE(parent); |
5459 | spin_unlock(&root->inode_lock); | |
cef21937 | 5460 | return; |
5d4f98a2 YZ |
5461 | } |
5462 | } | |
cef21937 FDBM |
5463 | rb_link_node(new, parent, p); |
5464 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5465 | spin_unlock(&root->inode_lock); |
5466 | } | |
5467 | ||
b79b7249 | 5468 | static void inode_tree_del(struct btrfs_inode *inode) |
5d4f98a2 | 5469 | { |
b79b7249 | 5470 | struct btrfs_root *root = inode->root; |
76dda93c | 5471 | int empty = 0; |
5d4f98a2 | 5472 | |
03e860bd | 5473 | spin_lock(&root->inode_lock); |
b79b7249 NB |
5474 | if (!RB_EMPTY_NODE(&inode->rb_node)) { |
5475 | rb_erase(&inode->rb_node, &root->inode_tree); | |
5476 | RB_CLEAR_NODE(&inode->rb_node); | |
76dda93c | 5477 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5478 | } |
03e860bd | 5479 | spin_unlock(&root->inode_lock); |
76dda93c | 5480 | |
69e9c6c6 | 5481 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5482 | spin_lock(&root->inode_lock); |
5483 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5484 | spin_unlock(&root->inode_lock); | |
5485 | if (empty) | |
5486 | btrfs_add_dead_root(root); | |
5487 | } | |
5488 | } | |
5489 | ||
5d4f98a2 | 5490 | |
e02119d5 CM |
5491 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5492 | { | |
5493 | struct btrfs_iget_args *args = p; | |
0202e83f DS |
5494 | |
5495 | inode->i_ino = args->ino; | |
5496 | BTRFS_I(inode)->location.objectid = args->ino; | |
5497 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
5498 | BTRFS_I(inode)->location.offset = 0; | |
5c8fd99f JB |
5499 | BTRFS_I(inode)->root = btrfs_grab_root(args->root); |
5500 | BUG_ON(args->root && !BTRFS_I(inode)->root); | |
39279cc3 CM |
5501 | return 0; |
5502 | } | |
5503 | ||
5504 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5505 | { | |
5506 | struct btrfs_iget_args *args = opaque; | |
0202e83f DS |
5507 | |
5508 | return args->ino == BTRFS_I(inode)->location.objectid && | |
d397712b | 5509 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5510 | } |
5511 | ||
0202e83f | 5512 | static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino, |
5d4f98a2 | 5513 | struct btrfs_root *root) |
39279cc3 CM |
5514 | { |
5515 | struct inode *inode; | |
5516 | struct btrfs_iget_args args; | |
0202e83f | 5517 | unsigned long hashval = btrfs_inode_hash(ino, root); |
778ba82b | 5518 | |
0202e83f | 5519 | args.ino = ino; |
39279cc3 CM |
5520 | args.root = root; |
5521 | ||
778ba82b | 5522 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5523 | btrfs_init_locked_inode, |
5524 | (void *)&args); | |
5525 | return inode; | |
5526 | } | |
5527 | ||
4c66e0d4 | 5528 | /* |
0202e83f | 5529 | * Get an inode object given its inode number and corresponding root. |
4c66e0d4 DS |
5530 | * Path can be preallocated to prevent recursing back to iget through |
5531 | * allocator. NULL is also valid but may require an additional allocation | |
5532 | * later. | |
1a54ef8c | 5533 | */ |
0202e83f | 5534 | struct inode *btrfs_iget_path(struct super_block *s, u64 ino, |
4c66e0d4 | 5535 | struct btrfs_root *root, struct btrfs_path *path) |
1a54ef8c BR |
5536 | { |
5537 | struct inode *inode; | |
5538 | ||
0202e83f | 5539 | inode = btrfs_iget_locked(s, ino, root); |
1a54ef8c | 5540 | if (!inode) |
5d4f98a2 | 5541 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5542 | |
5543 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5544 | int ret; |
5545 | ||
4222ea71 | 5546 | ret = btrfs_read_locked_inode(inode, path); |
9bc2ceff | 5547 | if (!ret) { |
1748f843 MF |
5548 | inode_tree_add(inode); |
5549 | unlock_new_inode(inode); | |
1748f843 | 5550 | } else { |
f5b3a417 AV |
5551 | iget_failed(inode); |
5552 | /* | |
5553 | * ret > 0 can come from btrfs_search_slot called by | |
5554 | * btrfs_read_locked_inode, this means the inode item | |
5555 | * was not found. | |
5556 | */ | |
5557 | if (ret > 0) | |
5558 | ret = -ENOENT; | |
5559 | inode = ERR_PTR(ret); | |
1748f843 MF |
5560 | } |
5561 | } | |
5562 | ||
1a54ef8c BR |
5563 | return inode; |
5564 | } | |
5565 | ||
0202e83f | 5566 | struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root) |
4222ea71 | 5567 | { |
0202e83f | 5568 | return btrfs_iget_path(s, ino, root, NULL); |
4222ea71 FM |
5569 | } |
5570 | ||
4df27c4d YZ |
5571 | static struct inode *new_simple_dir(struct super_block *s, |
5572 | struct btrfs_key *key, | |
5573 | struct btrfs_root *root) | |
5574 | { | |
5575 | struct inode *inode = new_inode(s); | |
5576 | ||
5577 | if (!inode) | |
5578 | return ERR_PTR(-ENOMEM); | |
5579 | ||
5c8fd99f | 5580 | BTRFS_I(inode)->root = btrfs_grab_root(root); |
4df27c4d | 5581 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); |
72ac3c0d | 5582 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5583 | |
5584 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
6bb6b514 OS |
5585 | /* |
5586 | * We only need lookup, the rest is read-only and there's no inode | |
5587 | * associated with the dentry | |
5588 | */ | |
5589 | inode->i_op = &simple_dir_inode_operations; | |
1fdf4194 | 5590 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5591 | inode->i_fop = &simple_dir_operations; |
5592 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
c2050a45 | 5593 | inode->i_mtime = current_time(inode); |
9cc97d64 | 5594 | inode->i_atime = inode->i_mtime; |
5595 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 5596 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
4df27c4d YZ |
5597 | |
5598 | return inode; | |
5599 | } | |
5600 | ||
6bf9e4bd QW |
5601 | static inline u8 btrfs_inode_type(struct inode *inode) |
5602 | { | |
5603 | /* | |
5604 | * Compile-time asserts that generic FT_* types still match | |
5605 | * BTRFS_FT_* types | |
5606 | */ | |
5607 | BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN); | |
5608 | BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE); | |
5609 | BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR); | |
5610 | BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV); | |
5611 | BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV); | |
5612 | BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO); | |
5613 | BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK); | |
5614 | BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK); | |
5615 | ||
5616 | return fs_umode_to_ftype(inode->i_mode); | |
5617 | } | |
5618 | ||
3de4586c | 5619 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5620 | { |
0b246afa | 5621 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5622 | struct inode *inode; |
4df27c4d | 5623 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5624 | struct btrfs_root *sub_root = root; |
5625 | struct btrfs_key location; | |
6bf9e4bd | 5626 | u8 di_type = 0; |
b4aff1f8 | 5627 | int ret = 0; |
39279cc3 CM |
5628 | |
5629 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5630 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5631 | |
6bf9e4bd | 5632 | ret = btrfs_inode_by_name(dir, dentry, &location, &di_type); |
39279cc3 CM |
5633 | if (ret < 0) |
5634 | return ERR_PTR(ret); | |
5f39d397 | 5635 | |
4df27c4d | 5636 | if (location.type == BTRFS_INODE_ITEM_KEY) { |
0202e83f | 5637 | inode = btrfs_iget(dir->i_sb, location.objectid, root); |
6bf9e4bd QW |
5638 | if (IS_ERR(inode)) |
5639 | return inode; | |
5640 | ||
5641 | /* Do extra check against inode mode with di_type */ | |
5642 | if (btrfs_inode_type(inode) != di_type) { | |
5643 | btrfs_crit(fs_info, | |
5644 | "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u", | |
5645 | inode->i_mode, btrfs_inode_type(inode), | |
5646 | di_type); | |
5647 | iput(inode); | |
5648 | return ERR_PTR(-EUCLEAN); | |
5649 | } | |
4df27c4d YZ |
5650 | return inode; |
5651 | } | |
5652 | ||
2ff7e61e | 5653 | ret = fixup_tree_root_location(fs_info, dir, dentry, |
4df27c4d YZ |
5654 | &location, &sub_root); |
5655 | if (ret < 0) { | |
5656 | if (ret != -ENOENT) | |
5657 | inode = ERR_PTR(ret); | |
5658 | else | |
5659 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5660 | } else { | |
0202e83f | 5661 | inode = btrfs_iget(dir->i_sb, location.objectid, sub_root); |
39279cc3 | 5662 | } |
8727002f | 5663 | if (root != sub_root) |
00246528 | 5664 | btrfs_put_root(sub_root); |
76dda93c | 5665 | |
34d19bad | 5666 | if (!IS_ERR(inode) && root != sub_root) { |
0b246afa | 5667 | down_read(&fs_info->cleanup_work_sem); |
bc98a42c | 5668 | if (!sb_rdonly(inode->i_sb)) |
66b4ffd1 | 5669 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5670 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5671 | if (ret) { |
5672 | iput(inode); | |
66b4ffd1 | 5673 | inode = ERR_PTR(ret); |
01cd3367 | 5674 | } |
c71bf099 YZ |
5675 | } |
5676 | ||
3de4586c CM |
5677 | return inode; |
5678 | } | |
5679 | ||
fe15ce44 | 5680 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5681 | { |
5682 | struct btrfs_root *root; | |
2b0143b5 | 5683 | struct inode *inode = d_inode(dentry); |
76dda93c | 5684 | |
848cce0d | 5685 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5686 | inode = d_inode(dentry->d_parent); |
76dda93c | 5687 | |
848cce0d LZ |
5688 | if (inode) { |
5689 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5690 | if (btrfs_root_refs(&root->root_item) == 0) |
5691 | return 1; | |
848cce0d | 5692 | |
4a0cc7ca | 5693 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 5694 | return 1; |
efefb143 | 5695 | } |
76dda93c YZ |
5696 | return 0; |
5697 | } | |
5698 | ||
3de4586c | 5699 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5700 | unsigned int flags) |
3de4586c | 5701 | { |
3837d208 | 5702 | struct inode *inode = btrfs_lookup_dentry(dir, dentry); |
5662344b | 5703 | |
3837d208 AV |
5704 | if (inode == ERR_PTR(-ENOENT)) |
5705 | inode = NULL; | |
41d28bca | 5706 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5707 | } |
5708 | ||
23b5ec74 JB |
5709 | /* |
5710 | * All this infrastructure exists because dir_emit can fault, and we are holding | |
5711 | * the tree lock when doing readdir. For now just allocate a buffer and copy | |
5712 | * our information into that, and then dir_emit from the buffer. This is | |
5713 | * similar to what NFS does, only we don't keep the buffer around in pagecache | |
5714 | * because I'm afraid I'll mess that up. Long term we need to make filldir do | |
5715 | * copy_to_user_inatomic so we don't have to worry about page faulting under the | |
5716 | * tree lock. | |
5717 | */ | |
5718 | static int btrfs_opendir(struct inode *inode, struct file *file) | |
5719 | { | |
5720 | struct btrfs_file_private *private; | |
5721 | ||
5722 | private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL); | |
5723 | if (!private) | |
5724 | return -ENOMEM; | |
5725 | private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
5726 | if (!private->filldir_buf) { | |
5727 | kfree(private); | |
5728 | return -ENOMEM; | |
5729 | } | |
5730 | file->private_data = private; | |
5731 | return 0; | |
5732 | } | |
5733 | ||
5734 | struct dir_entry { | |
5735 | u64 ino; | |
5736 | u64 offset; | |
5737 | unsigned type; | |
5738 | int name_len; | |
5739 | }; | |
5740 | ||
5741 | static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx) | |
5742 | { | |
5743 | while (entries--) { | |
5744 | struct dir_entry *entry = addr; | |
5745 | char *name = (char *)(entry + 1); | |
5746 | ||
92d32170 DS |
5747 | ctx->pos = get_unaligned(&entry->offset); |
5748 | if (!dir_emit(ctx, name, get_unaligned(&entry->name_len), | |
5749 | get_unaligned(&entry->ino), | |
5750 | get_unaligned(&entry->type))) | |
23b5ec74 | 5751 | return 1; |
92d32170 DS |
5752 | addr += sizeof(struct dir_entry) + |
5753 | get_unaligned(&entry->name_len); | |
23b5ec74 JB |
5754 | ctx->pos++; |
5755 | } | |
5756 | return 0; | |
5757 | } | |
5758 | ||
9cdda8d3 | 5759 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5760 | { |
9cdda8d3 | 5761 | struct inode *inode = file_inode(file); |
39279cc3 | 5762 | struct btrfs_root *root = BTRFS_I(inode)->root; |
23b5ec74 | 5763 | struct btrfs_file_private *private = file->private_data; |
39279cc3 CM |
5764 | struct btrfs_dir_item *di; |
5765 | struct btrfs_key key; | |
5f39d397 | 5766 | struct btrfs_key found_key; |
39279cc3 | 5767 | struct btrfs_path *path; |
23b5ec74 | 5768 | void *addr; |
16cdcec7 MX |
5769 | struct list_head ins_list; |
5770 | struct list_head del_list; | |
39279cc3 | 5771 | int ret; |
5f39d397 | 5772 | struct extent_buffer *leaf; |
39279cc3 | 5773 | int slot; |
5f39d397 CM |
5774 | char *name_ptr; |
5775 | int name_len; | |
23b5ec74 JB |
5776 | int entries = 0; |
5777 | int total_len = 0; | |
02dbfc99 | 5778 | bool put = false; |
c2951f32 | 5779 | struct btrfs_key location; |
5f39d397 | 5780 | |
9cdda8d3 AV |
5781 | if (!dir_emit_dots(file, ctx)) |
5782 | return 0; | |
5783 | ||
49593bfa | 5784 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5785 | if (!path) |
5786 | return -ENOMEM; | |
ff5714cc | 5787 | |
23b5ec74 | 5788 | addr = private->filldir_buf; |
e4058b54 | 5789 | path->reada = READA_FORWARD; |
49593bfa | 5790 | |
c2951f32 JM |
5791 | INIT_LIST_HEAD(&ins_list); |
5792 | INIT_LIST_HEAD(&del_list); | |
5793 | put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list); | |
16cdcec7 | 5794 | |
23b5ec74 | 5795 | again: |
c2951f32 | 5796 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 5797 | key.offset = ctx->pos; |
4a0cc7ca | 5798 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 5799 | |
39279cc3 CM |
5800 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5801 | if (ret < 0) | |
5802 | goto err; | |
49593bfa DW |
5803 | |
5804 | while (1) { | |
23b5ec74 JB |
5805 | struct dir_entry *entry; |
5806 | ||
5f39d397 | 5807 | leaf = path->nodes[0]; |
39279cc3 | 5808 | slot = path->slots[0]; |
b9e03af0 LZ |
5809 | if (slot >= btrfs_header_nritems(leaf)) { |
5810 | ret = btrfs_next_leaf(root, path); | |
5811 | if (ret < 0) | |
5812 | goto err; | |
5813 | else if (ret > 0) | |
5814 | break; | |
5815 | continue; | |
39279cc3 | 5816 | } |
3de4586c | 5817 | |
5f39d397 CM |
5818 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5819 | ||
5820 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5821 | break; |
c2951f32 | 5822 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 5823 | break; |
9cdda8d3 | 5824 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5825 | goto next; |
c2951f32 | 5826 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
16cdcec7 | 5827 | goto next; |
39279cc3 | 5828 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
c2951f32 | 5829 | name_len = btrfs_dir_name_len(leaf, di); |
23b5ec74 JB |
5830 | if ((total_len + sizeof(struct dir_entry) + name_len) >= |
5831 | PAGE_SIZE) { | |
5832 | btrfs_release_path(path); | |
5833 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
5834 | if (ret) | |
5835 | goto nopos; | |
5836 | addr = private->filldir_buf; | |
5837 | entries = 0; | |
5838 | total_len = 0; | |
5839 | goto again; | |
c2951f32 | 5840 | } |
23b5ec74 JB |
5841 | |
5842 | entry = addr; | |
92d32170 | 5843 | put_unaligned(name_len, &entry->name_len); |
23b5ec74 | 5844 | name_ptr = (char *)(entry + 1); |
c2951f32 JM |
5845 | read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1), |
5846 | name_len); | |
7d157c3d | 5847 | put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)), |
92d32170 | 5848 | &entry->type); |
c2951f32 | 5849 | btrfs_dir_item_key_to_cpu(leaf, di, &location); |
92d32170 DS |
5850 | put_unaligned(location.objectid, &entry->ino); |
5851 | put_unaligned(found_key.offset, &entry->offset); | |
23b5ec74 JB |
5852 | entries++; |
5853 | addr += sizeof(struct dir_entry) + name_len; | |
5854 | total_len += sizeof(struct dir_entry) + name_len; | |
b9e03af0 LZ |
5855 | next: |
5856 | path->slots[0]++; | |
39279cc3 | 5857 | } |
23b5ec74 JB |
5858 | btrfs_release_path(path); |
5859 | ||
5860 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
5861 | if (ret) | |
5862 | goto nopos; | |
49593bfa | 5863 | |
d2fbb2b5 | 5864 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 5865 | if (ret) |
bc4ef759 DS |
5866 | goto nopos; |
5867 | ||
db62efbb ZB |
5868 | /* |
5869 | * Stop new entries from being returned after we return the last | |
5870 | * entry. | |
5871 | * | |
5872 | * New directory entries are assigned a strictly increasing | |
5873 | * offset. This means that new entries created during readdir | |
5874 | * are *guaranteed* to be seen in the future by that readdir. | |
5875 | * This has broken buggy programs which operate on names as | |
5876 | * they're returned by readdir. Until we re-use freed offsets | |
5877 | * we have this hack to stop new entries from being returned | |
5878 | * under the assumption that they'll never reach this huge | |
5879 | * offset. | |
5880 | * | |
5881 | * This is being careful not to overflow 32bit loff_t unless the | |
5882 | * last entry requires it because doing so has broken 32bit apps | |
5883 | * in the past. | |
5884 | */ | |
c2951f32 JM |
5885 | if (ctx->pos >= INT_MAX) |
5886 | ctx->pos = LLONG_MAX; | |
5887 | else | |
5888 | ctx->pos = INT_MAX; | |
39279cc3 CM |
5889 | nopos: |
5890 | ret = 0; | |
5891 | err: | |
02dbfc99 OS |
5892 | if (put) |
5893 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 5894 | btrfs_free_path(path); |
39279cc3 CM |
5895 | return ret; |
5896 | } | |
5897 | ||
39279cc3 | 5898 | /* |
54aa1f4d | 5899 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5900 | * inode changes. But, it is most likely to find the inode in cache. |
5901 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5902 | * to keep or drop this code. | |
5903 | */ | |
48a3b636 | 5904 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 | 5905 | { |
2ff7e61e | 5906 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5907 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5908 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5909 | int ret; |
5910 | ||
72ac3c0d | 5911 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5912 | return 0; |
39279cc3 | 5913 | |
7a7eaa40 | 5914 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5915 | if (IS_ERR(trans)) |
5916 | return PTR_ERR(trans); | |
8929ecfa | 5917 | |
9a56fcd1 | 5918 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
94b60442 CM |
5919 | if (ret && ret == -ENOSPC) { |
5920 | /* whoops, lets try again with the full transaction */ | |
3a45bb20 | 5921 | btrfs_end_transaction(trans); |
94b60442 | 5922 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
5923 | if (IS_ERR(trans)) |
5924 | return PTR_ERR(trans); | |
8929ecfa | 5925 | |
9a56fcd1 | 5926 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
94b60442 | 5927 | } |
3a45bb20 | 5928 | btrfs_end_transaction(trans); |
16cdcec7 | 5929 | if (BTRFS_I(inode)->delayed_node) |
2ff7e61e | 5930 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
5931 | |
5932 | return ret; | |
5933 | } | |
5934 | ||
5935 | /* | |
5936 | * This is a copy of file_update_time. We need this so we can return error on | |
5937 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
5938 | */ | |
95582b00 | 5939 | static int btrfs_update_time(struct inode *inode, struct timespec64 *now, |
e41f941a | 5940 | int flags) |
22c44fe6 | 5941 | { |
2bc55652 | 5942 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3a8c7231 | 5943 | bool dirty = flags & ~S_VERSION; |
2bc55652 AB |
5944 | |
5945 | if (btrfs_root_readonly(root)) | |
5946 | return -EROFS; | |
5947 | ||
e41f941a | 5948 | if (flags & S_VERSION) |
3a8c7231 | 5949 | dirty |= inode_maybe_inc_iversion(inode, dirty); |
e41f941a JB |
5950 | if (flags & S_CTIME) |
5951 | inode->i_ctime = *now; | |
5952 | if (flags & S_MTIME) | |
5953 | inode->i_mtime = *now; | |
5954 | if (flags & S_ATIME) | |
5955 | inode->i_atime = *now; | |
3a8c7231 | 5956 | return dirty ? btrfs_dirty_inode(inode) : 0; |
39279cc3 CM |
5957 | } |
5958 | ||
d352ac68 CM |
5959 | /* |
5960 | * find the highest existing sequence number in a directory | |
5961 | * and then set the in-memory index_cnt variable to reflect | |
5962 | * free sequence numbers | |
5963 | */ | |
4c570655 | 5964 | static int btrfs_set_inode_index_count(struct btrfs_inode *inode) |
aec7477b | 5965 | { |
4c570655 | 5966 | struct btrfs_root *root = inode->root; |
aec7477b JB |
5967 | struct btrfs_key key, found_key; |
5968 | struct btrfs_path *path; | |
5969 | struct extent_buffer *leaf; | |
5970 | int ret; | |
5971 | ||
4c570655 | 5972 | key.objectid = btrfs_ino(inode); |
962a298f | 5973 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
5974 | key.offset = (u64)-1; |
5975 | ||
5976 | path = btrfs_alloc_path(); | |
5977 | if (!path) | |
5978 | return -ENOMEM; | |
5979 | ||
5980 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5981 | if (ret < 0) | |
5982 | goto out; | |
5983 | /* FIXME: we should be able to handle this */ | |
5984 | if (ret == 0) | |
5985 | goto out; | |
5986 | ret = 0; | |
5987 | ||
5988 | /* | |
5989 | * MAGIC NUMBER EXPLANATION: | |
5990 | * since we search a directory based on f_pos we have to start at 2 | |
5991 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
5992 | * else has to start at 2 | |
5993 | */ | |
5994 | if (path->slots[0] == 0) { | |
4c570655 | 5995 | inode->index_cnt = 2; |
aec7477b JB |
5996 | goto out; |
5997 | } | |
5998 | ||
5999 | path->slots[0]--; | |
6000 | ||
6001 | leaf = path->nodes[0]; | |
6002 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6003 | ||
4c570655 | 6004 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 6005 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
4c570655 | 6006 | inode->index_cnt = 2; |
aec7477b JB |
6007 | goto out; |
6008 | } | |
6009 | ||
4c570655 | 6010 | inode->index_cnt = found_key.offset + 1; |
aec7477b JB |
6011 | out: |
6012 | btrfs_free_path(path); | |
6013 | return ret; | |
6014 | } | |
6015 | ||
d352ac68 CM |
6016 | /* |
6017 | * helper to find a free sequence number in a given directory. This current | |
6018 | * code is very simple, later versions will do smarter things in the btree | |
6019 | */ | |
877574e2 | 6020 | int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index) |
aec7477b JB |
6021 | { |
6022 | int ret = 0; | |
6023 | ||
877574e2 NB |
6024 | if (dir->index_cnt == (u64)-1) { |
6025 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
16cdcec7 MX |
6026 | if (ret) { |
6027 | ret = btrfs_set_inode_index_count(dir); | |
6028 | if (ret) | |
6029 | return ret; | |
6030 | } | |
aec7477b JB |
6031 | } |
6032 | ||
877574e2 NB |
6033 | *index = dir->index_cnt; |
6034 | dir->index_cnt++; | |
aec7477b JB |
6035 | |
6036 | return ret; | |
6037 | } | |
6038 | ||
b0d5d10f CM |
6039 | static int btrfs_insert_inode_locked(struct inode *inode) |
6040 | { | |
6041 | struct btrfs_iget_args args; | |
0202e83f DS |
6042 | |
6043 | args.ino = BTRFS_I(inode)->location.objectid; | |
b0d5d10f CM |
6044 | args.root = BTRFS_I(inode)->root; |
6045 | ||
6046 | return insert_inode_locked4(inode, | |
6047 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6048 | btrfs_find_actor, &args); | |
6049 | } | |
6050 | ||
19aee8de AJ |
6051 | /* |
6052 | * Inherit flags from the parent inode. | |
6053 | * | |
6054 | * Currently only the compression flags and the cow flags are inherited. | |
6055 | */ | |
6056 | static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) | |
6057 | { | |
6058 | unsigned int flags; | |
6059 | ||
6060 | if (!dir) | |
6061 | return; | |
6062 | ||
6063 | flags = BTRFS_I(dir)->flags; | |
6064 | ||
6065 | if (flags & BTRFS_INODE_NOCOMPRESS) { | |
6066 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS; | |
6067 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; | |
6068 | } else if (flags & BTRFS_INODE_COMPRESS) { | |
6069 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS; | |
6070 | BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS; | |
6071 | } | |
6072 | ||
6073 | if (flags & BTRFS_INODE_NODATACOW) { | |
6074 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
6075 | if (S_ISREG(inode->i_mode)) | |
6076 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
6077 | } | |
6078 | ||
7b6a221e | 6079 | btrfs_sync_inode_flags_to_i_flags(inode); |
19aee8de AJ |
6080 | } |
6081 | ||
39279cc3 CM |
6082 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6083 | struct btrfs_root *root, | |
aec7477b | 6084 | struct inode *dir, |
9c58309d | 6085 | const char *name, int name_len, |
175a4eb7 AV |
6086 | u64 ref_objectid, u64 objectid, |
6087 | umode_t mode, u64 *index) | |
39279cc3 | 6088 | { |
0b246afa | 6089 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 6090 | struct inode *inode; |
5f39d397 | 6091 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6092 | struct btrfs_key *location; |
5f39d397 | 6093 | struct btrfs_path *path; |
9c58309d CM |
6094 | struct btrfs_inode_ref *ref; |
6095 | struct btrfs_key key[2]; | |
6096 | u32 sizes[2]; | |
ef3b9af5 | 6097 | int nitems = name ? 2 : 1; |
9c58309d | 6098 | unsigned long ptr; |
11a19a90 | 6099 | unsigned int nofs_flag; |
39279cc3 | 6100 | int ret; |
39279cc3 | 6101 | |
5f39d397 | 6102 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6103 | if (!path) |
6104 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6105 | |
11a19a90 | 6106 | nofs_flag = memalloc_nofs_save(); |
0b246afa | 6107 | inode = new_inode(fs_info->sb); |
11a19a90 | 6108 | memalloc_nofs_restore(nofs_flag); |
8fb27640 YS |
6109 | if (!inode) { |
6110 | btrfs_free_path(path); | |
39279cc3 | 6111 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6112 | } |
39279cc3 | 6113 | |
5762b5c9 FM |
6114 | /* |
6115 | * O_TMPFILE, set link count to 0, so that after this point, | |
6116 | * we fill in an inode item with the correct link count. | |
6117 | */ | |
6118 | if (!name) | |
6119 | set_nlink(inode, 0); | |
6120 | ||
581bb050 LZ |
6121 | /* |
6122 | * we have to initialize this early, so we can reclaim the inode | |
6123 | * number if we fail afterwards in this function. | |
6124 | */ | |
6125 | inode->i_ino = objectid; | |
6126 | ||
ef3b9af5 | 6127 | if (dir && name) { |
1abe9b8a | 6128 | trace_btrfs_inode_request(dir); |
6129 | ||
877574e2 | 6130 | ret = btrfs_set_inode_index(BTRFS_I(dir), index); |
09771430 | 6131 | if (ret) { |
8fb27640 | 6132 | btrfs_free_path(path); |
09771430 | 6133 | iput(inode); |
aec7477b | 6134 | return ERR_PTR(ret); |
09771430 | 6135 | } |
ef3b9af5 FM |
6136 | } else if (dir) { |
6137 | *index = 0; | |
aec7477b JB |
6138 | } |
6139 | /* | |
6140 | * index_cnt is ignored for everything but a dir, | |
df6703e1 | 6141 | * btrfs_set_inode_index_count has an explanation for the magic |
aec7477b JB |
6142 | * number |
6143 | */ | |
6144 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6145 | BTRFS_I(inode)->dir_index = *index; |
5c8fd99f | 6146 | BTRFS_I(inode)->root = btrfs_grab_root(root); |
e02119d5 | 6147 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6148 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6149 | |
5dc562c5 JB |
6150 | /* |
6151 | * We could have gotten an inode number from somebody who was fsynced | |
6152 | * and then removed in this same transaction, so let's just set full | |
6153 | * sync since it will be a full sync anyway and this will blow away the | |
6154 | * old info in the log. | |
6155 | */ | |
6156 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
6157 | ||
9c58309d | 6158 | key[0].objectid = objectid; |
962a298f | 6159 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6160 | key[0].offset = 0; |
6161 | ||
9c58309d | 6162 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6163 | |
6164 | if (name) { | |
6165 | /* | |
6166 | * Start new inodes with an inode_ref. This is slightly more | |
6167 | * efficient for small numbers of hard links since they will | |
6168 | * be packed into one item. Extended refs will kick in if we | |
6169 | * add more hard links than can fit in the ref item. | |
6170 | */ | |
6171 | key[1].objectid = objectid; | |
962a298f | 6172 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
6173 | key[1].offset = ref_objectid; |
6174 | ||
6175 | sizes[1] = name_len + sizeof(*ref); | |
6176 | } | |
9c58309d | 6177 | |
b0d5d10f CM |
6178 | location = &BTRFS_I(inode)->location; |
6179 | location->objectid = objectid; | |
6180 | location->offset = 0; | |
962a298f | 6181 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6182 | |
6183 | ret = btrfs_insert_inode_locked(inode); | |
32955c54 AV |
6184 | if (ret < 0) { |
6185 | iput(inode); | |
b0d5d10f | 6186 | goto fail; |
32955c54 | 6187 | } |
b0d5d10f | 6188 | |
ef3b9af5 | 6189 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6190 | if (ret != 0) |
b0d5d10f | 6191 | goto fail_unlock; |
5f39d397 | 6192 | |
ecc11fab | 6193 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 6194 | inode_set_bytes(inode, 0); |
9cc97d64 | 6195 | |
c2050a45 | 6196 | inode->i_mtime = current_time(inode); |
9cc97d64 | 6197 | inode->i_atime = inode->i_mtime; |
6198 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 6199 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
9cc97d64 | 6200 | |
5f39d397 CM |
6201 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6202 | struct btrfs_inode_item); | |
b159fa28 | 6203 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6204 | sizeof(*inode_item)); |
e02119d5 | 6205 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6206 | |
ef3b9af5 FM |
6207 | if (name) { |
6208 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6209 | struct btrfs_inode_ref); | |
6210 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6211 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6212 | ptr = (unsigned long)(ref + 1); | |
6213 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6214 | } | |
9c58309d | 6215 | |
5f39d397 CM |
6216 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6217 | btrfs_free_path(path); | |
6218 | ||
6cbff00f CH |
6219 | btrfs_inherit_iflags(inode, dir); |
6220 | ||
569254b0 | 6221 | if (S_ISREG(mode)) { |
0b246afa | 6222 | if (btrfs_test_opt(fs_info, NODATASUM)) |
94272164 | 6223 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; |
0b246afa | 6224 | if (btrfs_test_opt(fs_info, NODATACOW)) |
f2bdf9a8 JB |
6225 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6226 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6227 | } |
6228 | ||
5d4f98a2 | 6229 | inode_tree_add(inode); |
1abe9b8a | 6230 | |
6231 | trace_btrfs_inode_new(inode); | |
d9094414 | 6232 | btrfs_set_inode_last_trans(trans, BTRFS_I(inode)); |
1abe9b8a | 6233 | |
8ea05e3a AB |
6234 | btrfs_update_root_times(trans, root); |
6235 | ||
63541927 FDBM |
6236 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6237 | if (ret) | |
0b246afa | 6238 | btrfs_err(fs_info, |
63541927 | 6239 | "error inheriting props for ino %llu (root %llu): %d", |
f85b7379 | 6240 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret); |
63541927 | 6241 | |
39279cc3 | 6242 | return inode; |
b0d5d10f CM |
6243 | |
6244 | fail_unlock: | |
32955c54 | 6245 | discard_new_inode(inode); |
5f39d397 | 6246 | fail: |
ef3b9af5 | 6247 | if (dir && name) |
aec7477b | 6248 | BTRFS_I(dir)->index_cnt--; |
5f39d397 CM |
6249 | btrfs_free_path(path); |
6250 | return ERR_PTR(ret); | |
39279cc3 CM |
6251 | } |
6252 | ||
d352ac68 CM |
6253 | /* |
6254 | * utility function to add 'inode' into 'parent_inode' with | |
6255 | * a give name and a given sequence number. | |
6256 | * if 'add_backref' is true, also insert a backref from the | |
6257 | * inode to the parent directory. | |
6258 | */ | |
e02119d5 | 6259 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
db0a669f | 6260 | struct btrfs_inode *parent_inode, struct btrfs_inode *inode, |
e02119d5 | 6261 | const char *name, int name_len, int add_backref, u64 index) |
39279cc3 | 6262 | { |
4df27c4d | 6263 | int ret = 0; |
39279cc3 | 6264 | struct btrfs_key key; |
db0a669f NB |
6265 | struct btrfs_root *root = parent_inode->root; |
6266 | u64 ino = btrfs_ino(inode); | |
6267 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6268 | |
33345d01 | 6269 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
db0a669f | 6270 | memcpy(&key, &inode->root->root_key, sizeof(key)); |
4df27c4d | 6271 | } else { |
33345d01 | 6272 | key.objectid = ino; |
962a298f | 6273 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6274 | key.offset = 0; |
6275 | } | |
6276 | ||
33345d01 | 6277 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
6025c19f | 6278 | ret = btrfs_add_root_ref(trans, key.objectid, |
0b246afa JM |
6279 | root->root_key.objectid, parent_ino, |
6280 | index, name, name_len); | |
4df27c4d | 6281 | } else if (add_backref) { |
33345d01 LZ |
6282 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6283 | parent_ino, index); | |
4df27c4d | 6284 | } |
39279cc3 | 6285 | |
79787eaa JM |
6286 | /* Nothing to clean up yet */ |
6287 | if (ret) | |
6288 | return ret; | |
4df27c4d | 6289 | |
684572df | 6290 | ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key, |
db0a669f | 6291 | btrfs_inode_type(&inode->vfs_inode), index); |
9c52057c | 6292 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6293 | goto fail_dir_item; |
6294 | else if (ret) { | |
66642832 | 6295 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6296 | return ret; |
39279cc3 | 6297 | } |
79787eaa | 6298 | |
db0a669f | 6299 | btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size + |
79787eaa | 6300 | name_len * 2); |
db0a669f | 6301 | inode_inc_iversion(&parent_inode->vfs_inode); |
5338e43a FM |
6302 | /* |
6303 | * If we are replaying a log tree, we do not want to update the mtime | |
6304 | * and ctime of the parent directory with the current time, since the | |
6305 | * log replay procedure is responsible for setting them to their correct | |
6306 | * values (the ones it had when the fsync was done). | |
6307 | */ | |
6308 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) { | |
6309 | struct timespec64 now = current_time(&parent_inode->vfs_inode); | |
6310 | ||
6311 | parent_inode->vfs_inode.i_mtime = now; | |
6312 | parent_inode->vfs_inode.i_ctime = now; | |
6313 | } | |
9a56fcd1 | 6314 | ret = btrfs_update_inode(trans, root, parent_inode); |
79787eaa | 6315 | if (ret) |
66642832 | 6316 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6317 | return ret; |
fe66a05a CM |
6318 | |
6319 | fail_dir_item: | |
6320 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6321 | u64 local_index; | |
6322 | int err; | |
3ee1c553 | 6323 | err = btrfs_del_root_ref(trans, key.objectid, |
0b246afa JM |
6324 | root->root_key.objectid, parent_ino, |
6325 | &local_index, name, name_len); | |
1690dd41 JT |
6326 | if (err) |
6327 | btrfs_abort_transaction(trans, err); | |
fe66a05a CM |
6328 | } else if (add_backref) { |
6329 | u64 local_index; | |
6330 | int err; | |
6331 | ||
6332 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6333 | ino, parent_ino, &local_index); | |
1690dd41 JT |
6334 | if (err) |
6335 | btrfs_abort_transaction(trans, err); | |
fe66a05a | 6336 | } |
1690dd41 JT |
6337 | |
6338 | /* Return the original error code */ | |
fe66a05a | 6339 | return ret; |
39279cc3 CM |
6340 | } |
6341 | ||
6342 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
cef415af NB |
6343 | struct btrfs_inode *dir, struct dentry *dentry, |
6344 | struct btrfs_inode *inode, int backref, u64 index) | |
39279cc3 | 6345 | { |
a1b075d2 JB |
6346 | int err = btrfs_add_link(trans, dir, inode, |
6347 | dentry->d_name.name, dentry->d_name.len, | |
6348 | backref, index); | |
39279cc3 CM |
6349 | if (err > 0) |
6350 | err = -EEXIST; | |
6351 | return err; | |
6352 | } | |
6353 | ||
618e21d5 | 6354 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6355 | umode_t mode, dev_t rdev) |
618e21d5 | 6356 | { |
2ff7e61e | 6357 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 JB |
6358 | struct btrfs_trans_handle *trans; |
6359 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6360 | struct inode *inode = NULL; |
618e21d5 | 6361 | int err; |
618e21d5 | 6362 | u64 objectid; |
00e4e6b3 | 6363 | u64 index = 0; |
618e21d5 | 6364 | |
9ed74f2d JB |
6365 | /* |
6366 | * 2 for inode item and ref | |
6367 | * 2 for dir items | |
6368 | * 1 for xattr if selinux is on | |
6369 | */ | |
a22285a6 YZ |
6370 | trans = btrfs_start_transaction(root, 5); |
6371 | if (IS_ERR(trans)) | |
6372 | return PTR_ERR(trans); | |
1832a6d5 | 6373 | |
abadc1fc | 6374 | err = btrfs_find_free_objectid(root, &objectid); |
581bb050 LZ |
6375 | if (err) |
6376 | goto out_unlock; | |
6377 | ||
aec7477b | 6378 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6379 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6380 | mode, &index); | |
7cf96da3 TI |
6381 | if (IS_ERR(inode)) { |
6382 | err = PTR_ERR(inode); | |
32955c54 | 6383 | inode = NULL; |
618e21d5 | 6384 | goto out_unlock; |
7cf96da3 | 6385 | } |
618e21d5 | 6386 | |
ad19db71 CS |
6387 | /* |
6388 | * If the active LSM wants to access the inode during | |
6389 | * d_instantiate it needs these. Smack checks to see | |
6390 | * if the filesystem supports xattrs by looking at the | |
6391 | * ops vector. | |
6392 | */ | |
ad19db71 | 6393 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6394 | init_special_inode(inode, inode->i_mode, rdev); |
6395 | ||
6396 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6397 | if (err) |
32955c54 | 6398 | goto out_unlock; |
b0d5d10f | 6399 | |
cef415af NB |
6400 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6401 | 0, index); | |
32955c54 AV |
6402 | if (err) |
6403 | goto out_unlock; | |
6404 | ||
9a56fcd1 | 6405 | btrfs_update_inode(trans, root, BTRFS_I(inode)); |
32955c54 | 6406 | d_instantiate_new(dentry, inode); |
b0d5d10f | 6407 | |
618e21d5 | 6408 | out_unlock: |
3a45bb20 | 6409 | btrfs_end_transaction(trans); |
2ff7e61e | 6410 | btrfs_btree_balance_dirty(fs_info); |
32955c54 | 6411 | if (err && inode) { |
618e21d5 | 6412 | inode_dec_link_count(inode); |
32955c54 | 6413 | discard_new_inode(inode); |
618e21d5 | 6414 | } |
618e21d5 JB |
6415 | return err; |
6416 | } | |
6417 | ||
39279cc3 | 6418 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6419 | umode_t mode, bool excl) |
39279cc3 | 6420 | { |
2ff7e61e | 6421 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
6422 | struct btrfs_trans_handle *trans; |
6423 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6424 | struct inode *inode = NULL; |
a22285a6 | 6425 | int err; |
39279cc3 | 6426 | u64 objectid; |
00e4e6b3 | 6427 | u64 index = 0; |
39279cc3 | 6428 | |
9ed74f2d JB |
6429 | /* |
6430 | * 2 for inode item and ref | |
6431 | * 2 for dir items | |
6432 | * 1 for xattr if selinux is on | |
6433 | */ | |
a22285a6 YZ |
6434 | trans = btrfs_start_transaction(root, 5); |
6435 | if (IS_ERR(trans)) | |
6436 | return PTR_ERR(trans); | |
9ed74f2d | 6437 | |
abadc1fc | 6438 | err = btrfs_find_free_objectid(root, &objectid); |
581bb050 LZ |
6439 | if (err) |
6440 | goto out_unlock; | |
6441 | ||
aec7477b | 6442 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6443 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6444 | mode, &index); | |
7cf96da3 TI |
6445 | if (IS_ERR(inode)) { |
6446 | err = PTR_ERR(inode); | |
32955c54 | 6447 | inode = NULL; |
39279cc3 | 6448 | goto out_unlock; |
7cf96da3 | 6449 | } |
ad19db71 CS |
6450 | /* |
6451 | * If the active LSM wants to access the inode during | |
6452 | * d_instantiate it needs these. Smack checks to see | |
6453 | * if the filesystem supports xattrs by looking at the | |
6454 | * ops vector. | |
6455 | */ | |
6456 | inode->i_fop = &btrfs_file_operations; | |
6457 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6458 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6459 | |
6460 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6461 | if (err) | |
32955c54 | 6462 | goto out_unlock; |
b0d5d10f | 6463 | |
9a56fcd1 | 6464 | err = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
b0d5d10f | 6465 | if (err) |
32955c54 | 6466 | goto out_unlock; |
ad19db71 | 6467 | |
cef415af NB |
6468 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6469 | 0, index); | |
39279cc3 | 6470 | if (err) |
32955c54 | 6471 | goto out_unlock; |
43baa579 | 6472 | |
1e2e547a | 6473 | d_instantiate_new(dentry, inode); |
43baa579 | 6474 | |
39279cc3 | 6475 | out_unlock: |
3a45bb20 | 6476 | btrfs_end_transaction(trans); |
32955c54 | 6477 | if (err && inode) { |
39279cc3 | 6478 | inode_dec_link_count(inode); |
32955c54 | 6479 | discard_new_inode(inode); |
39279cc3 | 6480 | } |
2ff7e61e | 6481 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6482 | return err; |
6483 | } | |
6484 | ||
6485 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6486 | struct dentry *dentry) | |
6487 | { | |
271dba45 | 6488 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6489 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6490 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6491 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00e4e6b3 | 6492 | u64 index; |
39279cc3 CM |
6493 | int err; |
6494 | int drop_inode = 0; | |
6495 | ||
4a8be425 | 6496 | /* do not allow sys_link's with other subvols of the same device */ |
4fd786e6 | 6497 | if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid) |
3ab3564f | 6498 | return -EXDEV; |
4a8be425 | 6499 | |
f186373f | 6500 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6501 | return -EMLINK; |
4a8be425 | 6502 | |
877574e2 | 6503 | err = btrfs_set_inode_index(BTRFS_I(dir), &index); |
aec7477b JB |
6504 | if (err) |
6505 | goto fail; | |
6506 | ||
a22285a6 | 6507 | /* |
7e6b6465 | 6508 | * 2 items for inode and inode ref |
a22285a6 | 6509 | * 2 items for dir items |
7e6b6465 | 6510 | * 1 item for parent inode |
399b0bbf | 6511 | * 1 item for orphan item deletion if O_TMPFILE |
a22285a6 | 6512 | */ |
399b0bbf | 6513 | trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6); |
a22285a6 YZ |
6514 | if (IS_ERR(trans)) { |
6515 | err = PTR_ERR(trans); | |
271dba45 | 6516 | trans = NULL; |
a22285a6 YZ |
6517 | goto fail; |
6518 | } | |
5f39d397 | 6519 | |
67de1176 MX |
6520 | /* There are several dir indexes for this inode, clear the cache. */ |
6521 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6522 | inc_nlink(inode); |
0c4d2d95 | 6523 | inode_inc_iversion(inode); |
c2050a45 | 6524 | inode->i_ctime = current_time(inode); |
7de9c6ee | 6525 | ihold(inode); |
e9976151 | 6526 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6527 | |
cef415af NB |
6528 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6529 | 1, index); | |
5f39d397 | 6530 | |
a5719521 | 6531 | if (err) { |
54aa1f4d | 6532 | drop_inode = 1; |
a5719521 | 6533 | } else { |
10d9f309 | 6534 | struct dentry *parent = dentry->d_parent; |
d4682ba0 | 6535 | |
9a56fcd1 | 6536 | err = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
79787eaa JM |
6537 | if (err) |
6538 | goto fail; | |
ef3b9af5 FM |
6539 | if (inode->i_nlink == 1) { |
6540 | /* | |
6541 | * If new hard link count is 1, it's a file created | |
6542 | * with open(2) O_TMPFILE flag. | |
6543 | */ | |
3d6ae7bb | 6544 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
ef3b9af5 FM |
6545 | if (err) |
6546 | goto fail; | |
6547 | } | |
08c422c2 | 6548 | d_instantiate(dentry, inode); |
75b463d2 | 6549 | btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent); |
a5719521 | 6550 | } |
39279cc3 | 6551 | |
1832a6d5 | 6552 | fail: |
271dba45 | 6553 | if (trans) |
3a45bb20 | 6554 | btrfs_end_transaction(trans); |
39279cc3 CM |
6555 | if (drop_inode) { |
6556 | inode_dec_link_count(inode); | |
6557 | iput(inode); | |
6558 | } | |
2ff7e61e | 6559 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6560 | return err; |
6561 | } | |
6562 | ||
18bb1db3 | 6563 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6564 | { |
2ff7e61e | 6565 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
b9d86667 | 6566 | struct inode *inode = NULL; |
39279cc3 CM |
6567 | struct btrfs_trans_handle *trans; |
6568 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6569 | int err = 0; | |
b9d86667 | 6570 | u64 objectid = 0; |
00e4e6b3 | 6571 | u64 index = 0; |
39279cc3 | 6572 | |
9ed74f2d JB |
6573 | /* |
6574 | * 2 items for inode and ref | |
6575 | * 2 items for dir items | |
6576 | * 1 for xattr if selinux is on | |
6577 | */ | |
a22285a6 YZ |
6578 | trans = btrfs_start_transaction(root, 5); |
6579 | if (IS_ERR(trans)) | |
6580 | return PTR_ERR(trans); | |
39279cc3 | 6581 | |
abadc1fc | 6582 | err = btrfs_find_free_objectid(root, &objectid); |
581bb050 LZ |
6583 | if (err) |
6584 | goto out_fail; | |
6585 | ||
aec7477b | 6586 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6587 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6588 | S_IFDIR | mode, &index); | |
39279cc3 CM |
6589 | if (IS_ERR(inode)) { |
6590 | err = PTR_ERR(inode); | |
32955c54 | 6591 | inode = NULL; |
39279cc3 CM |
6592 | goto out_fail; |
6593 | } | |
5f39d397 | 6594 | |
b0d5d10f CM |
6595 | /* these must be set before we unlock the inode */ |
6596 | inode->i_op = &btrfs_dir_inode_operations; | |
6597 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6598 | |
2a7dba39 | 6599 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6600 | if (err) |
32955c54 | 6601 | goto out_fail; |
39279cc3 | 6602 | |
6ef06d27 | 6603 | btrfs_i_size_write(BTRFS_I(inode), 0); |
9a56fcd1 | 6604 | err = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
39279cc3 | 6605 | if (err) |
32955c54 | 6606 | goto out_fail; |
5f39d397 | 6607 | |
db0a669f NB |
6608 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6609 | dentry->d_name.name, | |
6610 | dentry->d_name.len, 0, index); | |
39279cc3 | 6611 | if (err) |
32955c54 | 6612 | goto out_fail; |
5f39d397 | 6613 | |
1e2e547a | 6614 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
6615 | |
6616 | out_fail: | |
3a45bb20 | 6617 | btrfs_end_transaction(trans); |
32955c54 | 6618 | if (err && inode) { |
c7cfb8a5 | 6619 | inode_dec_link_count(inode); |
32955c54 | 6620 | discard_new_inode(inode); |
c7cfb8a5 | 6621 | } |
2ff7e61e | 6622 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6623 | return err; |
6624 | } | |
6625 | ||
c8b97818 | 6626 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6627 | struct page *page, |
c8b97818 CM |
6628 | size_t pg_offset, u64 extent_offset, |
6629 | struct btrfs_file_extent_item *item) | |
6630 | { | |
6631 | int ret; | |
6632 | struct extent_buffer *leaf = path->nodes[0]; | |
6633 | char *tmp; | |
6634 | size_t max_size; | |
6635 | unsigned long inline_size; | |
6636 | unsigned long ptr; | |
261507a0 | 6637 | int compress_type; |
c8b97818 CM |
6638 | |
6639 | WARN_ON(pg_offset != 0); | |
261507a0 | 6640 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6641 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6642 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6643 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6644 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6645 | if (!tmp) |
6646 | return -ENOMEM; | |
c8b97818 CM |
6647 | ptr = btrfs_file_extent_inline_start(item); |
6648 | ||
6649 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6650 | ||
09cbfeaf | 6651 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6652 | ret = btrfs_decompress(compress_type, tmp, page, |
6653 | extent_offset, inline_size, max_size); | |
e1699d2d ZB |
6654 | |
6655 | /* | |
6656 | * decompression code contains a memset to fill in any space between the end | |
6657 | * of the uncompressed data and the end of max_size in case the decompressed | |
6658 | * data ends up shorter than ram_bytes. That doesn't cover the hole between | |
6659 | * the end of an inline extent and the beginning of the next block, so we | |
6660 | * cover that region here. | |
6661 | */ | |
6662 | ||
6663 | if (max_size + pg_offset < PAGE_SIZE) { | |
6664 | char *map = kmap(page); | |
6665 | memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset); | |
6666 | kunmap(page); | |
6667 | } | |
c8b97818 | 6668 | kfree(tmp); |
166ae5a4 | 6669 | return ret; |
c8b97818 CM |
6670 | } |
6671 | ||
39b07b5d OS |
6672 | /** |
6673 | * btrfs_get_extent - Lookup the first extent overlapping a range in a file. | |
6674 | * @inode: file to search in | |
6675 | * @page: page to read extent data into if the extent is inline | |
6676 | * @pg_offset: offset into @page to copy to | |
6677 | * @start: file offset | |
6678 | * @len: length of range starting at @start | |
6679 | * | |
6680 | * This returns the first &struct extent_map which overlaps with the given | |
6681 | * range, reading it from the B-tree and caching it if necessary. Note that | |
6682 | * there may be more extents which overlap the given range after the returned | |
6683 | * extent_map. | |
d352ac68 | 6684 | * |
39b07b5d OS |
6685 | * If @page is not NULL and the extent is inline, this also reads the extent |
6686 | * data directly into the page and marks the extent up to date in the io_tree. | |
6687 | * | |
6688 | * Return: ERR_PTR on error, non-NULL extent_map on success. | |
d352ac68 | 6689 | */ |
fc4f21b1 | 6690 | struct extent_map *btrfs_get_extent(struct btrfs_inode *inode, |
39b07b5d OS |
6691 | struct page *page, size_t pg_offset, |
6692 | u64 start, u64 len) | |
a52d9a80 | 6693 | { |
3ffbd68c | 6694 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
1028d1c4 | 6695 | int ret = 0; |
a52d9a80 CM |
6696 | u64 extent_start = 0; |
6697 | u64 extent_end = 0; | |
fc4f21b1 | 6698 | u64 objectid = btrfs_ino(inode); |
7e74e235 | 6699 | int extent_type = -1; |
f421950f | 6700 | struct btrfs_path *path = NULL; |
fc4f21b1 | 6701 | struct btrfs_root *root = inode->root; |
a52d9a80 | 6702 | struct btrfs_file_extent_item *item; |
5f39d397 CM |
6703 | struct extent_buffer *leaf; |
6704 | struct btrfs_key found_key; | |
a52d9a80 | 6705 | struct extent_map *em = NULL; |
fc4f21b1 NB |
6706 | struct extent_map_tree *em_tree = &inode->extent_tree; |
6707 | struct extent_io_tree *io_tree = &inode->io_tree; | |
a52d9a80 | 6708 | |
890871be | 6709 | read_lock(&em_tree->lock); |
d1310b2e | 6710 | em = lookup_extent_mapping(em_tree, start, len); |
890871be | 6711 | read_unlock(&em_tree->lock); |
d1310b2e | 6712 | |
a52d9a80 | 6713 | if (em) { |
e1c4b745 CM |
6714 | if (em->start > start || em->start + em->len <= start) |
6715 | free_extent_map(em); | |
6716 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6717 | free_extent_map(em); |
6718 | else | |
6719 | goto out; | |
a52d9a80 | 6720 | } |
172ddd60 | 6721 | em = alloc_extent_map(); |
a52d9a80 | 6722 | if (!em) { |
1028d1c4 | 6723 | ret = -ENOMEM; |
d1310b2e | 6724 | goto out; |
a52d9a80 | 6725 | } |
d1310b2e | 6726 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6727 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6728 | em->len = (u64)-1; |
c8b97818 | 6729 | em->block_len = (u64)-1; |
f421950f | 6730 | |
bee6ec82 | 6731 | path = btrfs_alloc_path(); |
f421950f | 6732 | if (!path) { |
1028d1c4 | 6733 | ret = -ENOMEM; |
bee6ec82 | 6734 | goto out; |
f421950f CM |
6735 | } |
6736 | ||
bee6ec82 LB |
6737 | /* Chances are we'll be called again, so go ahead and do readahead */ |
6738 | path->reada = READA_FORWARD; | |
4d7240f0 JB |
6739 | |
6740 | /* | |
6741 | * The same explanation in load_free_space_cache applies here as well, | |
6742 | * we only read when we're loading the free space cache, and at that | |
6743 | * point the commit_root has everything we need. | |
6744 | */ | |
6745 | if (btrfs_is_free_space_inode(inode)) { | |
6746 | path->search_commit_root = 1; | |
6747 | path->skip_locking = 1; | |
6748 | } | |
51899412 | 6749 | |
5c9a702e | 6750 | ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0); |
a52d9a80 | 6751 | if (ret < 0) { |
a52d9a80 | 6752 | goto out; |
b8eeab7f | 6753 | } else if (ret > 0) { |
a52d9a80 CM |
6754 | if (path->slots[0] == 0) |
6755 | goto not_found; | |
6756 | path->slots[0]--; | |
1028d1c4 | 6757 | ret = 0; |
a52d9a80 CM |
6758 | } |
6759 | ||
5f39d397 CM |
6760 | leaf = path->nodes[0]; |
6761 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6762 | struct btrfs_file_extent_item); |
5f39d397 | 6763 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5f39d397 | 6764 | if (found_key.objectid != objectid || |
694c12ed | 6765 | found_key.type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6766 | /* |
6767 | * If we backup past the first extent we want to move forward | |
6768 | * and see if there is an extent in front of us, otherwise we'll | |
6769 | * say there is a hole for our whole search range which can | |
6770 | * cause problems. | |
6771 | */ | |
6772 | extent_end = start; | |
6773 | goto next; | |
a52d9a80 CM |
6774 | } |
6775 | ||
694c12ed | 6776 | extent_type = btrfs_file_extent_type(leaf, item); |
5f39d397 | 6777 | extent_start = found_key.offset; |
a5eeb3d1 | 6778 | extent_end = btrfs_file_extent_end(path); |
694c12ed NB |
6779 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
6780 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
6bf9e4bd QW |
6781 | /* Only regular file could have regular/prealloc extent */ |
6782 | if (!S_ISREG(inode->vfs_inode.i_mode)) { | |
1028d1c4 | 6783 | ret = -EUCLEAN; |
6bf9e4bd QW |
6784 | btrfs_crit(fs_info, |
6785 | "regular/prealloc extent found for non-regular inode %llu", | |
6786 | btrfs_ino(inode)); | |
6787 | goto out; | |
6788 | } | |
09ed2f16 LB |
6789 | trace_btrfs_get_extent_show_fi_regular(inode, leaf, item, |
6790 | extent_start); | |
694c12ed | 6791 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
09ed2f16 LB |
6792 | trace_btrfs_get_extent_show_fi_inline(inode, leaf, item, |
6793 | path->slots[0], | |
6794 | extent_start); | |
9036c102 | 6795 | } |
25a50341 | 6796 | next: |
9036c102 YZ |
6797 | if (start >= extent_end) { |
6798 | path->slots[0]++; | |
6799 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6800 | ret = btrfs_next_leaf(root, path); | |
1028d1c4 | 6801 | if (ret < 0) |
9036c102 | 6802 | goto out; |
1028d1c4 | 6803 | else if (ret > 0) |
9036c102 | 6804 | goto not_found; |
1028d1c4 | 6805 | |
9036c102 | 6806 | leaf = path->nodes[0]; |
a52d9a80 | 6807 | } |
9036c102 YZ |
6808 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6809 | if (found_key.objectid != objectid || | |
6810 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6811 | goto not_found; | |
6812 | if (start + len <= found_key.offset) | |
6813 | goto not_found; | |
e2eca69d WS |
6814 | if (start > found_key.offset) |
6815 | goto next; | |
02a033df NB |
6816 | |
6817 | /* New extent overlaps with existing one */ | |
9036c102 | 6818 | em->start = start; |
70c8a91c | 6819 | em->orig_start = start; |
9036c102 | 6820 | em->len = found_key.offset - start; |
02a033df NB |
6821 | em->block_start = EXTENT_MAP_HOLE; |
6822 | goto insert; | |
9036c102 YZ |
6823 | } |
6824 | ||
39b07b5d | 6825 | btrfs_extent_item_to_extent_map(inode, path, item, !page, em); |
7ffbb598 | 6826 | |
694c12ed NB |
6827 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
6828 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6829 | goto insert; |
694c12ed | 6830 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 6831 | unsigned long ptr; |
a52d9a80 | 6832 | char *map; |
3326d1b0 CM |
6833 | size_t size; |
6834 | size_t extent_offset; | |
6835 | size_t copy_size; | |
a52d9a80 | 6836 | |
39b07b5d | 6837 | if (!page) |
689f9346 | 6838 | goto out; |
5f39d397 | 6839 | |
e41ca589 | 6840 | size = btrfs_file_extent_ram_bytes(leaf, item); |
9036c102 | 6841 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
6842 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
6843 | size - extent_offset); | |
3326d1b0 | 6844 | em->start = extent_start + extent_offset; |
0b246afa | 6845 | em->len = ALIGN(copy_size, fs_info->sectorsize); |
b4939680 | 6846 | em->orig_block_len = em->len; |
70c8a91c | 6847 | em->orig_start = em->start; |
689f9346 | 6848 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
e49aabd9 | 6849 | |
bf46f52d | 6850 | if (!PageUptodate(page)) { |
261507a0 LZ |
6851 | if (btrfs_file_extent_compression(leaf, item) != |
6852 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 6853 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 6854 | extent_offset, item); |
1028d1c4 | 6855 | if (ret) |
166ae5a4 | 6856 | goto out; |
c8b97818 CM |
6857 | } else { |
6858 | map = kmap(page); | |
6859 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6860 | copy_size); | |
09cbfeaf | 6861 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 6862 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 6863 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
6864 | copy_size); |
6865 | } | |
c8b97818 CM |
6866 | kunmap(page); |
6867 | } | |
179e29e4 | 6868 | flush_dcache_page(page); |
a52d9a80 | 6869 | } |
d1310b2e | 6870 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6871 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 6872 | goto insert; |
a52d9a80 CM |
6873 | } |
6874 | not_found: | |
6875 | em->start = start; | |
70c8a91c | 6876 | em->orig_start = start; |
d1310b2e | 6877 | em->len = len; |
5f39d397 | 6878 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 | 6879 | insert: |
1028d1c4 | 6880 | ret = 0; |
b3b4aa74 | 6881 | btrfs_release_path(path); |
d1310b2e | 6882 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 6883 | btrfs_err(fs_info, |
5d163e0e JM |
6884 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
6885 | em->start, em->len, start, len); | |
1028d1c4 | 6886 | ret = -EIO; |
a52d9a80 CM |
6887 | goto out; |
6888 | } | |
d1310b2e | 6889 | |
890871be | 6890 | write_lock(&em_tree->lock); |
1028d1c4 | 6891 | ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len); |
890871be | 6892 | write_unlock(&em_tree->lock); |
a52d9a80 | 6893 | out: |
c6414280 | 6894 | btrfs_free_path(path); |
1abe9b8a | 6895 | |
fc4f21b1 | 6896 | trace_btrfs_get_extent(root, inode, em); |
1abe9b8a | 6897 | |
1028d1c4 | 6898 | if (ret) { |
a52d9a80 | 6899 | free_extent_map(em); |
1028d1c4 | 6900 | return ERR_PTR(ret); |
a52d9a80 CM |
6901 | } |
6902 | return em; | |
6903 | } | |
6904 | ||
fc4f21b1 | 6905 | struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode, |
4ab47a8d | 6906 | u64 start, u64 len) |
ec29ed5b CM |
6907 | { |
6908 | struct extent_map *em; | |
6909 | struct extent_map *hole_em = NULL; | |
f3714ef4 | 6910 | u64 delalloc_start = start; |
ec29ed5b | 6911 | u64 end; |
f3714ef4 NB |
6912 | u64 delalloc_len; |
6913 | u64 delalloc_end; | |
ec29ed5b CM |
6914 | int err = 0; |
6915 | ||
39b07b5d | 6916 | em = btrfs_get_extent(inode, NULL, 0, start, len); |
ec29ed5b CM |
6917 | if (IS_ERR(em)) |
6918 | return em; | |
9986277e DC |
6919 | /* |
6920 | * If our em maps to: | |
6921 | * - a hole or | |
6922 | * - a pre-alloc extent, | |
6923 | * there might actually be delalloc bytes behind it. | |
6924 | */ | |
6925 | if (em->block_start != EXTENT_MAP_HOLE && | |
6926 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6927 | return em; | |
6928 | else | |
6929 | hole_em = em; | |
ec29ed5b CM |
6930 | |
6931 | /* check to see if we've wrapped (len == -1 or similar) */ | |
6932 | end = start + len; | |
6933 | if (end < start) | |
6934 | end = (u64)-1; | |
6935 | else | |
6936 | end -= 1; | |
6937 | ||
6938 | em = NULL; | |
6939 | ||
6940 | /* ok, we didn't find anything, lets look for delalloc */ | |
f3714ef4 | 6941 | delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start, |
ec29ed5b | 6942 | end, len, EXTENT_DELALLOC, 1); |
f3714ef4 NB |
6943 | delalloc_end = delalloc_start + delalloc_len; |
6944 | if (delalloc_end < delalloc_start) | |
6945 | delalloc_end = (u64)-1; | |
ec29ed5b CM |
6946 | |
6947 | /* | |
f3714ef4 NB |
6948 | * We didn't find anything useful, return the original results from |
6949 | * get_extent() | |
ec29ed5b | 6950 | */ |
f3714ef4 | 6951 | if (delalloc_start > end || delalloc_end <= start) { |
ec29ed5b CM |
6952 | em = hole_em; |
6953 | hole_em = NULL; | |
6954 | goto out; | |
6955 | } | |
6956 | ||
f3714ef4 NB |
6957 | /* |
6958 | * Adjust the delalloc_start to make sure it doesn't go backwards from | |
6959 | * the start they passed in | |
ec29ed5b | 6960 | */ |
f3714ef4 NB |
6961 | delalloc_start = max(start, delalloc_start); |
6962 | delalloc_len = delalloc_end - delalloc_start; | |
ec29ed5b | 6963 | |
f3714ef4 NB |
6964 | if (delalloc_len > 0) { |
6965 | u64 hole_start; | |
02950af4 | 6966 | u64 hole_len; |
f3714ef4 | 6967 | const u64 hole_end = extent_map_end(hole_em); |
ec29ed5b | 6968 | |
172ddd60 | 6969 | em = alloc_extent_map(); |
ec29ed5b CM |
6970 | if (!em) { |
6971 | err = -ENOMEM; | |
6972 | goto out; | |
6973 | } | |
f3714ef4 NB |
6974 | |
6975 | ASSERT(hole_em); | |
ec29ed5b | 6976 | /* |
f3714ef4 NB |
6977 | * When btrfs_get_extent can't find anything it returns one |
6978 | * huge hole | |
ec29ed5b | 6979 | * |
f3714ef4 NB |
6980 | * Make sure what it found really fits our range, and adjust to |
6981 | * make sure it is based on the start from the caller | |
ec29ed5b | 6982 | */ |
f3714ef4 NB |
6983 | if (hole_end <= start || hole_em->start > end) { |
6984 | free_extent_map(hole_em); | |
6985 | hole_em = NULL; | |
6986 | } else { | |
6987 | hole_start = max(hole_em->start, start); | |
6988 | hole_len = hole_end - hole_start; | |
ec29ed5b | 6989 | } |
f3714ef4 NB |
6990 | |
6991 | if (hole_em && delalloc_start > hole_start) { | |
6992 | /* | |
6993 | * Our hole starts before our delalloc, so we have to | |
6994 | * return just the parts of the hole that go until the | |
6995 | * delalloc starts | |
ec29ed5b | 6996 | */ |
f3714ef4 | 6997 | em->len = min(hole_len, delalloc_start - hole_start); |
ec29ed5b CM |
6998 | em->start = hole_start; |
6999 | em->orig_start = hole_start; | |
7000 | /* | |
f3714ef4 NB |
7001 | * Don't adjust block start at all, it is fixed at |
7002 | * EXTENT_MAP_HOLE | |
ec29ed5b CM |
7003 | */ |
7004 | em->block_start = hole_em->block_start; | |
7005 | em->block_len = hole_len; | |
f9e4fb53 LB |
7006 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
7007 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b | 7008 | } else { |
f3714ef4 NB |
7009 | /* |
7010 | * Hole is out of passed range or it starts after | |
7011 | * delalloc range | |
7012 | */ | |
7013 | em->start = delalloc_start; | |
7014 | em->len = delalloc_len; | |
7015 | em->orig_start = delalloc_start; | |
ec29ed5b | 7016 | em->block_start = EXTENT_MAP_DELALLOC; |
f3714ef4 | 7017 | em->block_len = delalloc_len; |
ec29ed5b | 7018 | } |
bf8d32b9 | 7019 | } else { |
ec29ed5b CM |
7020 | return hole_em; |
7021 | } | |
7022 | out: | |
7023 | ||
7024 | free_extent_map(hole_em); | |
7025 | if (err) { | |
7026 | free_extent_map(em); | |
7027 | return ERR_PTR(err); | |
7028 | } | |
7029 | return em; | |
7030 | } | |
7031 | ||
64f54188 | 7032 | static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode, |
5f9a8a51 FM |
7033 | const u64 start, |
7034 | const u64 len, | |
7035 | const u64 orig_start, | |
7036 | const u64 block_start, | |
7037 | const u64 block_len, | |
7038 | const u64 orig_block_len, | |
7039 | const u64 ram_bytes, | |
7040 | const int type) | |
7041 | { | |
7042 | struct extent_map *em = NULL; | |
7043 | int ret; | |
7044 | ||
5f9a8a51 | 7045 | if (type != BTRFS_ORDERED_NOCOW) { |
64f54188 NB |
7046 | em = create_io_em(inode, start, len, orig_start, block_start, |
7047 | block_len, orig_block_len, ram_bytes, | |
6f9994db LB |
7048 | BTRFS_COMPRESS_NONE, /* compress_type */ |
7049 | type); | |
5f9a8a51 FM |
7050 | if (IS_ERR(em)) |
7051 | goto out; | |
7052 | } | |
64f54188 NB |
7053 | ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len, |
7054 | block_len, type); | |
5f9a8a51 FM |
7055 | if (ret) { |
7056 | if (em) { | |
7057 | free_extent_map(em); | |
64f54188 | 7058 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); |
5f9a8a51 FM |
7059 | } |
7060 | em = ERR_PTR(ret); | |
7061 | } | |
7062 | out: | |
5f9a8a51 FM |
7063 | |
7064 | return em; | |
7065 | } | |
7066 | ||
9fc6f911 | 7067 | static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode, |
4b46fce2 JB |
7068 | u64 start, u64 len) |
7069 | { | |
9fc6f911 NB |
7070 | struct btrfs_root *root = inode->root; |
7071 | struct btrfs_fs_info *fs_info = root->fs_info; | |
70c8a91c | 7072 | struct extent_map *em; |
4b46fce2 JB |
7073 | struct btrfs_key ins; |
7074 | u64 alloc_hint; | |
7075 | int ret; | |
4b46fce2 | 7076 | |
9fc6f911 | 7077 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 7078 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 7079 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
7080 | if (ret) |
7081 | return ERR_PTR(ret); | |
4b46fce2 | 7082 | |
9fc6f911 | 7083 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
5f9a8a51 | 7084 | ins.objectid, ins.offset, ins.offset, |
6288d6ea | 7085 | ins.offset, BTRFS_ORDERED_REGULAR); |
0b246afa | 7086 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 7087 | if (IS_ERR(em)) |
9fc6f911 NB |
7088 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, |
7089 | 1); | |
de0ee0ed | 7090 | |
4b46fce2 JB |
7091 | return em; |
7092 | } | |
7093 | ||
46bfbb5c | 7094 | /* |
e4ecaf90 QW |
7095 | * Check if we can do nocow write into the range [@offset, @offset + @len) |
7096 | * | |
7097 | * @offset: File offset | |
7098 | * @len: The length to write, will be updated to the nocow writeable | |
7099 | * range | |
7100 | * @orig_start: (optional) Return the original file offset of the file extent | |
7101 | * @orig_len: (optional) Return the original on-disk length of the file extent | |
7102 | * @ram_bytes: (optional) Return the ram_bytes of the file extent | |
a84d5d42 BB |
7103 | * @strict: if true, omit optimizations that might force us into unnecessary |
7104 | * cow. e.g., don't trust generation number. | |
e4ecaf90 QW |
7105 | * |
7106 | * This function will flush ordered extents in the range to ensure proper | |
7107 | * nocow checks for (nowait == false) case. | |
7108 | * | |
7109 | * Return: | |
7110 | * >0 and update @len if we can do nocow write | |
7111 | * 0 if we can't do nocow write | |
7112 | * <0 if error happened | |
7113 | * | |
7114 | * NOTE: This only checks the file extents, caller is responsible to wait for | |
7115 | * any ordered extents. | |
46bfbb5c | 7116 | */ |
00361589 | 7117 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 | 7118 | u64 *orig_start, u64 *orig_block_len, |
a84d5d42 | 7119 | u64 *ram_bytes, bool strict) |
46bfbb5c | 7120 | { |
2ff7e61e | 7121 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
46bfbb5c CM |
7122 | struct btrfs_path *path; |
7123 | int ret; | |
7124 | struct extent_buffer *leaf; | |
7125 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7126 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7127 | struct btrfs_file_extent_item *fi; |
7128 | struct btrfs_key key; | |
7129 | u64 disk_bytenr; | |
7130 | u64 backref_offset; | |
7131 | u64 extent_end; | |
7132 | u64 num_bytes; | |
7133 | int slot; | |
7134 | int found_type; | |
7ee9e440 | 7135 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7136 | |
46bfbb5c CM |
7137 | path = btrfs_alloc_path(); |
7138 | if (!path) | |
7139 | return -ENOMEM; | |
7140 | ||
f85b7379 DS |
7141 | ret = btrfs_lookup_file_extent(NULL, root, path, |
7142 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
7143 | if (ret < 0) |
7144 | goto out; | |
7145 | ||
7146 | slot = path->slots[0]; | |
7147 | if (ret == 1) { | |
7148 | if (slot == 0) { | |
7149 | /* can't find the item, must cow */ | |
7150 | ret = 0; | |
7151 | goto out; | |
7152 | } | |
7153 | slot--; | |
7154 | } | |
7155 | ret = 0; | |
7156 | leaf = path->nodes[0]; | |
7157 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4a0cc7ca | 7158 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
7159 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7160 | /* not our file or wrong item type, must cow */ | |
7161 | goto out; | |
7162 | } | |
7163 | ||
7164 | if (key.offset > offset) { | |
7165 | /* Wrong offset, must cow */ | |
7166 | goto out; | |
7167 | } | |
7168 | ||
7169 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7170 | found_type = btrfs_file_extent_type(leaf, fi); | |
7171 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7172 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7173 | /* not a regular extent, must cow */ | |
7174 | goto out; | |
7175 | } | |
7ee9e440 JB |
7176 | |
7177 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7178 | goto out; | |
7179 | ||
e77751aa MX |
7180 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7181 | if (extent_end <= offset) | |
7182 | goto out; | |
7183 | ||
46bfbb5c | 7184 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7185 | if (disk_bytenr == 0) |
7186 | goto out; | |
7187 | ||
7188 | if (btrfs_file_extent_compression(leaf, fi) || | |
7189 | btrfs_file_extent_encryption(leaf, fi) || | |
7190 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7191 | goto out; | |
7192 | ||
78d4295b EL |
7193 | /* |
7194 | * Do the same check as in btrfs_cross_ref_exist but without the | |
7195 | * unnecessary search. | |
7196 | */ | |
a84d5d42 BB |
7197 | if (!strict && |
7198 | (btrfs_file_extent_generation(leaf, fi) <= | |
7199 | btrfs_root_last_snapshot(&root->root_item))) | |
78d4295b EL |
7200 | goto out; |
7201 | ||
46bfbb5c CM |
7202 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7203 | ||
7ee9e440 JB |
7204 | if (orig_start) { |
7205 | *orig_start = key.offset - backref_offset; | |
7206 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7207 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7208 | } | |
eb384b55 | 7209 | |
2ff7e61e | 7210 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
46bfbb5c | 7211 | goto out; |
7b2b7085 MX |
7212 | |
7213 | num_bytes = min(offset + *len, extent_end) - offset; | |
7214 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7215 | u64 range_end; | |
7216 | ||
da17066c JM |
7217 | range_end = round_up(offset + num_bytes, |
7218 | root->fs_info->sectorsize) - 1; | |
7b2b7085 MX |
7219 | ret = test_range_bit(io_tree, offset, range_end, |
7220 | EXTENT_DELALLOC, 0, NULL); | |
7221 | if (ret) { | |
7222 | ret = -EAGAIN; | |
7223 | goto out; | |
7224 | } | |
7225 | } | |
7226 | ||
1bda19eb | 7227 | btrfs_release_path(path); |
46bfbb5c CM |
7228 | |
7229 | /* | |
7230 | * look for other files referencing this extent, if we | |
7231 | * find any we must cow | |
7232 | */ | |
00361589 | 7233 | |
e4c3b2dc | 7234 | ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)), |
a84d5d42 BB |
7235 | key.offset - backref_offset, disk_bytenr, |
7236 | strict); | |
00361589 JB |
7237 | if (ret) { |
7238 | ret = 0; | |
7239 | goto out; | |
7240 | } | |
46bfbb5c CM |
7241 | |
7242 | /* | |
7243 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7244 | * in this extent we are about to write. If there | |
7245 | * are any csums in that range we have to cow in order | |
7246 | * to keep the csums correct | |
7247 | */ | |
7248 | disk_bytenr += backref_offset; | |
7249 | disk_bytenr += offset - key.offset; | |
2ff7e61e JM |
7250 | if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes)) |
7251 | goto out; | |
46bfbb5c CM |
7252 | /* |
7253 | * all of the above have passed, it is safe to overwrite this extent | |
7254 | * without cow | |
7255 | */ | |
eb384b55 | 7256 | *len = num_bytes; |
46bfbb5c CM |
7257 | ret = 1; |
7258 | out: | |
7259 | btrfs_free_path(path); | |
7260 | return ret; | |
7261 | } | |
7262 | ||
eb838e73 | 7263 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
f85781fb | 7264 | struct extent_state **cached_state, bool writing) |
eb838e73 JB |
7265 | { |
7266 | struct btrfs_ordered_extent *ordered; | |
7267 | int ret = 0; | |
7268 | ||
7269 | while (1) { | |
7270 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7271 | cached_state); |
eb838e73 JB |
7272 | /* |
7273 | * We're concerned with the entire range that we're going to be | |
01327610 | 7274 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7275 | * extents in this range. |
7276 | */ | |
a776c6fa | 7277 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart, |
eb838e73 JB |
7278 | lockend - lockstart + 1); |
7279 | ||
7280 | /* | |
7281 | * We need to make sure there are no buffered pages in this | |
7282 | * range either, we could have raced between the invalidate in | |
7283 | * generic_file_direct_write and locking the extent. The | |
7284 | * invalidate needs to happen so that reads after a write do not | |
7285 | * get stale data. | |
7286 | */ | |
fc4adbff | 7287 | if (!ordered && |
051c98eb DS |
7288 | (!writing || !filemap_range_has_page(inode->i_mapping, |
7289 | lockstart, lockend))) | |
eb838e73 JB |
7290 | break; |
7291 | ||
7292 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
e43bbe5e | 7293 | cached_state); |
eb838e73 JB |
7294 | |
7295 | if (ordered) { | |
ade77029 FM |
7296 | /* |
7297 | * If we are doing a DIO read and the ordered extent we | |
7298 | * found is for a buffered write, we can not wait for it | |
7299 | * to complete and retry, because if we do so we can | |
7300 | * deadlock with concurrent buffered writes on page | |
7301 | * locks. This happens only if our DIO read covers more | |
7302 | * than one extent map, if at this point has already | |
7303 | * created an ordered extent for a previous extent map | |
7304 | * and locked its range in the inode's io tree, and a | |
7305 | * concurrent write against that previous extent map's | |
7306 | * range and this range started (we unlock the ranges | |
7307 | * in the io tree only when the bios complete and | |
7308 | * buffered writes always lock pages before attempting | |
7309 | * to lock range in the io tree). | |
7310 | */ | |
7311 | if (writing || | |
7312 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
c0a43603 | 7313 | btrfs_start_ordered_extent(ordered, 1); |
ade77029 FM |
7314 | else |
7315 | ret = -ENOTBLK; | |
eb838e73 JB |
7316 | btrfs_put_ordered_extent(ordered); |
7317 | } else { | |
eb838e73 | 7318 | /* |
b850ae14 FM |
7319 | * We could trigger writeback for this range (and wait |
7320 | * for it to complete) and then invalidate the pages for | |
7321 | * this range (through invalidate_inode_pages2_range()), | |
7322 | * but that can lead us to a deadlock with a concurrent | |
ba206a02 | 7323 | * call to readahead (a buffered read or a defrag call |
b850ae14 FM |
7324 | * triggered a readahead) on a page lock due to an |
7325 | * ordered dio extent we created before but did not have | |
7326 | * yet a corresponding bio submitted (whence it can not | |
ba206a02 | 7327 | * complete), which makes readahead wait for that |
b850ae14 FM |
7328 | * ordered extent to complete while holding a lock on |
7329 | * that page. | |
eb838e73 | 7330 | */ |
b850ae14 | 7331 | ret = -ENOTBLK; |
eb838e73 JB |
7332 | } |
7333 | ||
ade77029 FM |
7334 | if (ret) |
7335 | break; | |
7336 | ||
eb838e73 JB |
7337 | cond_resched(); |
7338 | } | |
7339 | ||
7340 | return ret; | |
7341 | } | |
7342 | ||
6f9994db | 7343 | /* The callers of this must take lock_extent() */ |
4b67c11d NB |
7344 | static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start, |
7345 | u64 len, u64 orig_start, u64 block_start, | |
6f9994db LB |
7346 | u64 block_len, u64 orig_block_len, |
7347 | u64 ram_bytes, int compress_type, | |
7348 | int type) | |
69ffb543 JB |
7349 | { |
7350 | struct extent_map_tree *em_tree; | |
7351 | struct extent_map *em; | |
69ffb543 JB |
7352 | int ret; |
7353 | ||
6f9994db LB |
7354 | ASSERT(type == BTRFS_ORDERED_PREALLOC || |
7355 | type == BTRFS_ORDERED_COMPRESSED || | |
7356 | type == BTRFS_ORDERED_NOCOW || | |
1af4a0aa | 7357 | type == BTRFS_ORDERED_REGULAR); |
6f9994db | 7358 | |
4b67c11d | 7359 | em_tree = &inode->extent_tree; |
69ffb543 JB |
7360 | em = alloc_extent_map(); |
7361 | if (!em) | |
7362 | return ERR_PTR(-ENOMEM); | |
7363 | ||
7364 | em->start = start; | |
7365 | em->orig_start = orig_start; | |
7366 | em->len = len; | |
7367 | em->block_len = block_len; | |
7368 | em->block_start = block_start; | |
b4939680 | 7369 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7370 | em->ram_bytes = ram_bytes; |
70c8a91c | 7371 | em->generation = -1; |
69ffb543 | 7372 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
1af4a0aa | 7373 | if (type == BTRFS_ORDERED_PREALLOC) { |
b11e234d | 7374 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
1af4a0aa | 7375 | } else if (type == BTRFS_ORDERED_COMPRESSED) { |
6f9994db LB |
7376 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
7377 | em->compress_type = compress_type; | |
7378 | } | |
69ffb543 JB |
7379 | |
7380 | do { | |
4b67c11d NB |
7381 | btrfs_drop_extent_cache(inode, em->start, |
7382 | em->start + em->len - 1, 0); | |
69ffb543 | 7383 | write_lock(&em_tree->lock); |
09a2a8f9 | 7384 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 | 7385 | write_unlock(&em_tree->lock); |
6f9994db LB |
7386 | /* |
7387 | * The caller has taken lock_extent(), who could race with us | |
7388 | * to add em? | |
7389 | */ | |
69ffb543 JB |
7390 | } while (ret == -EEXIST); |
7391 | ||
7392 | if (ret) { | |
7393 | free_extent_map(em); | |
7394 | return ERR_PTR(ret); | |
7395 | } | |
7396 | ||
6f9994db | 7397 | /* em got 2 refs now, callers needs to do free_extent_map once. */ |
69ffb543 JB |
7398 | return em; |
7399 | } | |
7400 | ||
1c8d0175 | 7401 | |
c5794e51 | 7402 | static int btrfs_get_blocks_direct_write(struct extent_map **map, |
c5794e51 NB |
7403 | struct inode *inode, |
7404 | struct btrfs_dio_data *dio_data, | |
7405 | u64 start, u64 len) | |
7406 | { | |
7407 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
7408 | struct extent_map *em = *map; | |
7409 | int ret = 0; | |
7410 | ||
7411 | /* | |
7412 | * We don't allocate a new extent in the following cases | |
7413 | * | |
7414 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7415 | * existing extent. | |
7416 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7417 | * just use the extent. | |
7418 | * | |
7419 | */ | |
7420 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7421 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7422 | em->block_start != EXTENT_MAP_HOLE)) { | |
7423 | int type; | |
7424 | u64 block_start, orig_start, orig_block_len, ram_bytes; | |
7425 | ||
7426 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7427 | type = BTRFS_ORDERED_PREALLOC; | |
7428 | else | |
7429 | type = BTRFS_ORDERED_NOCOW; | |
7430 | len = min(len, em->len - (start - em->start)); | |
7431 | block_start = em->block_start + (start - em->start); | |
7432 | ||
7433 | if (can_nocow_extent(inode, start, &len, &orig_start, | |
a84d5d42 | 7434 | &orig_block_len, &ram_bytes, false) == 1 && |
c5794e51 NB |
7435 | btrfs_inc_nocow_writers(fs_info, block_start)) { |
7436 | struct extent_map *em2; | |
7437 | ||
64f54188 | 7438 | em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len, |
c5794e51 NB |
7439 | orig_start, block_start, |
7440 | len, orig_block_len, | |
7441 | ram_bytes, type); | |
7442 | btrfs_dec_nocow_writers(fs_info, block_start); | |
7443 | if (type == BTRFS_ORDERED_PREALLOC) { | |
7444 | free_extent_map(em); | |
7445 | *map = em = em2; | |
7446 | } | |
7447 | ||
7448 | if (em2 && IS_ERR(em2)) { | |
7449 | ret = PTR_ERR(em2); | |
7450 | goto out; | |
7451 | } | |
7452 | /* | |
7453 | * For inode marked NODATACOW or extent marked PREALLOC, | |
7454 | * use the existing or preallocated extent, so does not | |
7455 | * need to adjust btrfs_space_info's bytes_may_use. | |
7456 | */ | |
9db5d510 | 7457 | btrfs_free_reserved_data_space_noquota(fs_info, len); |
c5794e51 NB |
7458 | goto skip_cow; |
7459 | } | |
7460 | } | |
7461 | ||
7462 | /* this will cow the extent */ | |
c5794e51 | 7463 | free_extent_map(em); |
9fc6f911 | 7464 | *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len); |
c5794e51 NB |
7465 | if (IS_ERR(em)) { |
7466 | ret = PTR_ERR(em); | |
7467 | goto out; | |
7468 | } | |
7469 | ||
7470 | len = min(len, em->len - (start - em->start)); | |
7471 | ||
7472 | skip_cow: | |
c5794e51 NB |
7473 | /* |
7474 | * Need to update the i_size under the extent lock so buffered | |
7475 | * readers will get the updated i_size when we unlock. | |
7476 | */ | |
f85781fb | 7477 | if (start + len > i_size_read(inode)) |
c5794e51 NB |
7478 | i_size_write(inode, start + len); |
7479 | ||
c5794e51 | 7480 | dio_data->reserve -= len; |
c5794e51 NB |
7481 | out: |
7482 | return ret; | |
7483 | } | |
7484 | ||
f85781fb GR |
7485 | static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start, |
7486 | loff_t length, unsigned int flags, struct iomap *iomap, | |
7487 | struct iomap *srcmap) | |
4b46fce2 | 7488 | { |
0b246afa | 7489 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7490 | struct extent_map *em; |
eb838e73 | 7491 | struct extent_state *cached_state = NULL; |
50745b0a | 7492 | struct btrfs_dio_data *dio_data = NULL; |
eb838e73 | 7493 | u64 lockstart, lockend; |
f85781fb | 7494 | const bool write = !!(flags & IOMAP_WRITE); |
0934856d | 7495 | int ret = 0; |
f85781fb GR |
7496 | u64 len = length; |
7497 | bool unlock_extents = false; | |
eb838e73 | 7498 | |
f85781fb | 7499 | if (!write) |
0b246afa | 7500 | len = min_t(u64, len, fs_info->sectorsize); |
eb838e73 | 7501 | |
c329861d JB |
7502 | lockstart = start; |
7503 | lockend = start + len - 1; | |
7504 | ||
f85781fb GR |
7505 | /* |
7506 | * The generic stuff only does filemap_write_and_wait_range, which | |
7507 | * isn't enough if we've written compressed pages to this area, so we | |
7508 | * need to flush the dirty pages again to make absolutely sure that any | |
7509 | * outstanding dirty pages are on disk. | |
7510 | */ | |
7511 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
7512 | &BTRFS_I(inode)->runtime_flags)) { | |
7513 | ret = filemap_fdatawrite_range(inode->i_mapping, start, | |
7514 | start + length - 1); | |
7515 | if (ret) | |
7516 | return ret; | |
7517 | } | |
7518 | ||
7519 | dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS); | |
7520 | if (!dio_data) | |
7521 | return -ENOMEM; | |
7522 | ||
7523 | dio_data->length = length; | |
7524 | if (write) { | |
7525 | dio_data->reserve = round_up(length, fs_info->sectorsize); | |
7526 | ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), | |
7527 | &dio_data->data_reserved, | |
7528 | start, dio_data->reserve); | |
7529 | if (ret) { | |
7530 | extent_changeset_free(dio_data->data_reserved); | |
7531 | kfree(dio_data); | |
7532 | return ret; | |
7533 | } | |
e1cbbfa5 | 7534 | } |
f85781fb GR |
7535 | iomap->private = dio_data; |
7536 | ||
e1cbbfa5 | 7537 | |
eb838e73 JB |
7538 | /* |
7539 | * If this errors out it's because we couldn't invalidate pagecache for | |
7540 | * this range and we need to fallback to buffered. | |
7541 | */ | |
f85781fb | 7542 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) { |
9c9464cc FM |
7543 | ret = -ENOTBLK; |
7544 | goto err; | |
7545 | } | |
eb838e73 | 7546 | |
39b07b5d | 7547 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len); |
eb838e73 JB |
7548 | if (IS_ERR(em)) { |
7549 | ret = PTR_ERR(em); | |
7550 | goto unlock_err; | |
7551 | } | |
4b46fce2 JB |
7552 | |
7553 | /* | |
7554 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7555 | * io. INLINE is special, and we could probably kludge it in here, but | |
7556 | * it's still buffered so for safety lets just fall back to the generic | |
7557 | * buffered path. | |
7558 | * | |
7559 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7560 | * decompress it, so there will be buffering required no matter what we | |
7561 | * do, so go ahead and fallback to buffered. | |
7562 | * | |
01327610 | 7563 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7564 | * to buffered IO. Don't blame me, this is the price we pay for using |
7565 | * the generic code. | |
7566 | */ | |
7567 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7568 | em->block_start == EXTENT_MAP_INLINE) { | |
7569 | free_extent_map(em); | |
eb838e73 JB |
7570 | ret = -ENOTBLK; |
7571 | goto unlock_err; | |
4b46fce2 JB |
7572 | } |
7573 | ||
f85781fb GR |
7574 | len = min(len, em->len - (start - em->start)); |
7575 | if (write) { | |
7576 | ret = btrfs_get_blocks_direct_write(&em, inode, dio_data, | |
7577 | start, len); | |
c5794e51 NB |
7578 | if (ret < 0) |
7579 | goto unlock_err; | |
f85781fb GR |
7580 | unlock_extents = true; |
7581 | /* Recalc len in case the new em is smaller than requested */ | |
7582 | len = min(len, em->len - (start - em->start)); | |
c5794e51 | 7583 | } else { |
1c8d0175 NB |
7584 | /* |
7585 | * We need to unlock only the end area that we aren't using. | |
7586 | * The rest is going to be unlocked by the endio routine. | |
7587 | */ | |
f85781fb GR |
7588 | lockstart = start + len; |
7589 | if (lockstart < lockend) | |
7590 | unlock_extents = true; | |
7591 | } | |
7592 | ||
7593 | if (unlock_extents) | |
7594 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, | |
7595 | lockstart, lockend, &cached_state); | |
7596 | else | |
7597 | free_extent_state(cached_state); | |
7598 | ||
7599 | /* | |
7600 | * Translate extent map information to iomap. | |
7601 | * We trim the extents (and move the addr) even though iomap code does | |
7602 | * that, since we have locked only the parts we are performing I/O in. | |
7603 | */ | |
7604 | if ((em->block_start == EXTENT_MAP_HOLE) || | |
7605 | (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) { | |
7606 | iomap->addr = IOMAP_NULL_ADDR; | |
7607 | iomap->type = IOMAP_HOLE; | |
7608 | } else { | |
7609 | iomap->addr = em->block_start + (start - em->start); | |
7610 | iomap->type = IOMAP_MAPPED; | |
a43a67a2 | 7611 | } |
f85781fb GR |
7612 | iomap->offset = start; |
7613 | iomap->bdev = fs_info->fs_devices->latest_bdev; | |
7614 | iomap->length = len; | |
a43a67a2 | 7615 | |
4b46fce2 JB |
7616 | free_extent_map(em); |
7617 | ||
7618 | return 0; | |
eb838e73 JB |
7619 | |
7620 | unlock_err: | |
e182163d OS |
7621 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7622 | &cached_state); | |
9c9464cc | 7623 | err: |
f85781fb GR |
7624 | if (dio_data) { |
7625 | btrfs_delalloc_release_space(BTRFS_I(inode), | |
7626 | dio_data->data_reserved, start, | |
7627 | dio_data->reserve, true); | |
7628 | btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->reserve); | |
7629 | extent_changeset_free(dio_data->data_reserved); | |
7630 | kfree(dio_data); | |
7631 | } | |
7632 | return ret; | |
7633 | } | |
7634 | ||
7635 | static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length, | |
7636 | ssize_t written, unsigned int flags, struct iomap *iomap) | |
7637 | { | |
7638 | int ret = 0; | |
7639 | struct btrfs_dio_data *dio_data = iomap->private; | |
7640 | size_t submitted = dio_data->submitted; | |
7641 | const bool write = !!(flags & IOMAP_WRITE); | |
7642 | ||
7643 | if (!write && (iomap->type == IOMAP_HOLE)) { | |
7644 | /* If reading from a hole, unlock and return */ | |
7645 | unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1); | |
7646 | goto out; | |
7647 | } | |
7648 | ||
7649 | if (submitted < length) { | |
7650 | pos += submitted; | |
7651 | length -= submitted; | |
7652 | if (write) | |
7653 | __endio_write_update_ordered(BTRFS_I(inode), pos, | |
7654 | length, false); | |
7655 | else | |
7656 | unlock_extent(&BTRFS_I(inode)->io_tree, pos, | |
7657 | pos + length - 1); | |
7658 | ret = -ENOTBLK; | |
7659 | } | |
7660 | ||
7661 | if (write) { | |
7662 | if (dio_data->reserve) | |
7663 | btrfs_delalloc_release_space(BTRFS_I(inode), | |
7664 | dio_data->data_reserved, pos, | |
7665 | dio_data->reserve, true); | |
7666 | btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->length); | |
7667 | extent_changeset_free(dio_data->data_reserved); | |
7668 | } | |
7669 | out: | |
7670 | kfree(dio_data); | |
7671 | iomap->private = NULL; | |
7672 | ||
8b110e39 MX |
7673 | return ret; |
7674 | } | |
7675 | ||
769b4f24 | 7676 | static void btrfs_dio_private_put(struct btrfs_dio_private *dip) |
8b110e39 | 7677 | { |
769b4f24 OS |
7678 | /* |
7679 | * This implies a barrier so that stores to dio_bio->bi_status before | |
7680 | * this and loads of dio_bio->bi_status after this are fully ordered. | |
7681 | */ | |
7682 | if (!refcount_dec_and_test(&dip->refs)) | |
7683 | return; | |
8b110e39 | 7684 | |
769b4f24 | 7685 | if (bio_op(dip->dio_bio) == REQ_OP_WRITE) { |
b672b5c1 NB |
7686 | __endio_write_update_ordered(BTRFS_I(dip->inode), |
7687 | dip->logical_offset, | |
769b4f24 OS |
7688 | dip->bytes, |
7689 | !dip->dio_bio->bi_status); | |
7690 | } else { | |
7691 | unlock_extent(&BTRFS_I(dip->inode)->io_tree, | |
7692 | dip->logical_offset, | |
7693 | dip->logical_offset + dip->bytes - 1); | |
8b110e39 MX |
7694 | } |
7695 | ||
f85781fb | 7696 | bio_endio(dip->dio_bio); |
769b4f24 | 7697 | kfree(dip); |
8b110e39 MX |
7698 | } |
7699 | ||
77d5d689 OS |
7700 | static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio, |
7701 | int mirror_num, | |
7702 | unsigned long bio_flags) | |
8b110e39 | 7703 | { |
77d5d689 | 7704 | struct btrfs_dio_private *dip = bio->bi_private; |
2ff7e61e | 7705 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
58efbc9f | 7706 | blk_status_t ret; |
8b110e39 | 7707 | |
37226b21 | 7708 | BUG_ON(bio_op(bio) == REQ_OP_WRITE); |
8b110e39 | 7709 | |
5c047a69 | 7710 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
8b110e39 | 7711 | if (ret) |
ea057f6d | 7712 | return ret; |
8b110e39 | 7713 | |
77d5d689 | 7714 | refcount_inc(&dip->refs); |
08635bae | 7715 | ret = btrfs_map_bio(fs_info, bio, mirror_num); |
8b110e39 | 7716 | if (ret) |
fd9d6670 | 7717 | refcount_dec(&dip->refs); |
77d5d689 | 7718 | return ret; |
8b110e39 MX |
7719 | } |
7720 | ||
fd9d6670 OS |
7721 | static blk_status_t btrfs_check_read_dio_bio(struct inode *inode, |
7722 | struct btrfs_io_bio *io_bio, | |
7723 | const bool uptodate) | |
4b46fce2 | 7724 | { |
fd9d6670 OS |
7725 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
7726 | const u32 sectorsize = fs_info->sectorsize; | |
7727 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
7728 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
7729 | const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM); | |
17347cec LB |
7730 | struct bio_vec bvec; |
7731 | struct bvec_iter iter; | |
fd9d6670 | 7732 | u64 start = io_bio->logical; |
7ffd27e3 | 7733 | u32 bio_offset = 0; |
58efbc9f | 7734 | blk_status_t err = BLK_STS_OK; |
4b46fce2 | 7735 | |
fd9d6670 OS |
7736 | __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) { |
7737 | unsigned int i, nr_sectors, pgoff; | |
8b110e39 | 7738 | |
17347cec LB |
7739 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); |
7740 | pgoff = bvec.bv_offset; | |
fd9d6670 | 7741 | for (i = 0; i < nr_sectors; i++) { |
97bf5a55 | 7742 | ASSERT(pgoff < PAGE_SIZE); |
fd9d6670 | 7743 | if (uptodate && |
7ffd27e3 QW |
7744 | (!csum || !check_data_csum(inode, io_bio, |
7745 | bio_offset, bvec.bv_page, pgoff))) { | |
fd9d6670 OS |
7746 | clean_io_failure(fs_info, failure_tree, io_tree, |
7747 | start, bvec.bv_page, | |
7748 | btrfs_ino(BTRFS_I(inode)), | |
7749 | pgoff); | |
7750 | } else { | |
7751 | blk_status_t status; | |
7752 | ||
7ffd27e3 | 7753 | ASSERT((start - io_bio->logical) < UINT_MAX); |
77d5d689 OS |
7754 | status = btrfs_submit_read_repair(inode, |
7755 | &io_bio->bio, | |
7756 | start - io_bio->logical, | |
fd9d6670 OS |
7757 | bvec.bv_page, pgoff, |
7758 | start, | |
7759 | start + sectorsize - 1, | |
77d5d689 OS |
7760 | io_bio->mirror_num, |
7761 | submit_dio_repair_bio); | |
fd9d6670 OS |
7762 | if (status) |
7763 | err = status; | |
7764 | } | |
7765 | start += sectorsize; | |
7ffd27e3 QW |
7766 | ASSERT(bio_offset + sectorsize > bio_offset); |
7767 | bio_offset += sectorsize; | |
2dabb324 | 7768 | pgoff += sectorsize; |
2dabb324 | 7769 | } |
2c30c71b | 7770 | } |
c1dc0896 MX |
7771 | return err; |
7772 | } | |
7773 | ||
b672b5c1 | 7774 | static void __endio_write_update_ordered(struct btrfs_inode *inode, |
52427260 QW |
7775 | const u64 offset, const u64 bytes, |
7776 | const bool uptodate) | |
4b46fce2 | 7777 | { |
b672b5c1 | 7778 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
4b46fce2 | 7779 | struct btrfs_ordered_extent *ordered = NULL; |
52427260 | 7780 | struct btrfs_workqueue *wq; |
14543774 FM |
7781 | u64 ordered_offset = offset; |
7782 | u64 ordered_bytes = bytes; | |
67c003f9 | 7783 | u64 last_offset; |
4b46fce2 | 7784 | |
b672b5c1 | 7785 | if (btrfs_is_free_space_inode(inode)) |
52427260 | 7786 | wq = fs_info->endio_freespace_worker; |
a0cac0ec | 7787 | else |
52427260 | 7788 | wq = fs_info->endio_write_workers; |
52427260 | 7789 | |
b25f0d00 NB |
7790 | while (ordered_offset < offset + bytes) { |
7791 | last_offset = ordered_offset; | |
b672b5c1 | 7792 | if (btrfs_dec_test_first_ordered_pending(inode, &ordered, |
7095821e NB |
7793 | &ordered_offset, |
7794 | ordered_bytes, | |
7795 | uptodate)) { | |
a0cac0ec OS |
7796 | btrfs_init_work(&ordered->work, finish_ordered_fn, NULL, |
7797 | NULL); | |
b25f0d00 NB |
7798 | btrfs_queue_work(wq, &ordered->work); |
7799 | } | |
7800 | /* | |
7801 | * If btrfs_dec_test_ordered_pending does not find any ordered | |
7802 | * extent in the range, we can exit. | |
7803 | */ | |
7804 | if (ordered_offset == last_offset) | |
7805 | return; | |
7806 | /* | |
7807 | * Our bio might span multiple ordered extents. In this case | |
52042d8e | 7808 | * we keep going until we have accounted the whole dio. |
b25f0d00 NB |
7809 | */ |
7810 | if (ordered_offset < offset + bytes) { | |
7811 | ordered_bytes = offset + bytes - ordered_offset; | |
7812 | ordered = NULL; | |
7813 | } | |
163cf09c | 7814 | } |
14543774 FM |
7815 | } |
7816 | ||
8896a08d | 7817 | static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode, |
1941b64b QW |
7818 | struct bio *bio, |
7819 | u64 dio_file_offset) | |
eaf25d93 | 7820 | { |
1941b64b | 7821 | return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, 1); |
eaf25d93 CM |
7822 | } |
7823 | ||
4246a0b6 | 7824 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
7825 | { |
7826 | struct btrfs_dio_private *dip = bio->bi_private; | |
4e4cbee9 | 7827 | blk_status_t err = bio->bi_status; |
e65e1535 | 7828 | |
8b110e39 MX |
7829 | if (err) |
7830 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
6296b960 | 7831 | "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d", |
f85b7379 | 7832 | btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio), |
1201b58b | 7833 | bio->bi_opf, bio->bi_iter.bi_sector, |
8b110e39 MX |
7834 | bio->bi_iter.bi_size, err); |
7835 | ||
769b4f24 OS |
7836 | if (bio_op(bio) == REQ_OP_READ) { |
7837 | err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio), | |
fd9d6670 | 7838 | !err); |
e65e1535 MX |
7839 | } |
7840 | ||
769b4f24 OS |
7841 | if (err) |
7842 | dip->dio_bio->bi_status = err; | |
e65e1535 | 7843 | |
e65e1535 | 7844 | bio_put(bio); |
769b4f24 | 7845 | btrfs_dio_private_put(dip); |
c1dc0896 MX |
7846 | } |
7847 | ||
d0ee3934 DS |
7848 | static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio, |
7849 | struct inode *inode, u64 file_offset, int async_submit) | |
e65e1535 | 7850 | { |
0b246afa | 7851 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
facc8a22 | 7852 | struct btrfs_dio_private *dip = bio->bi_private; |
37226b21 | 7853 | bool write = bio_op(bio) == REQ_OP_WRITE; |
4e4cbee9 | 7854 | blk_status_t ret; |
e65e1535 | 7855 | |
4c274bc6 | 7856 | /* Check btrfs_submit_bio_hook() for rules about async submit. */ |
b812ce28 JB |
7857 | if (async_submit) |
7858 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
7859 | ||
5fd02043 | 7860 | if (!write) { |
0b246afa | 7861 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
5fd02043 JB |
7862 | if (ret) |
7863 | goto err; | |
7864 | } | |
e65e1535 | 7865 | |
e6961cac | 7866 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
1ae39938 JB |
7867 | goto map; |
7868 | ||
7869 | if (write && async_submit) { | |
1941b64b | 7870 | ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset, |
e288c080 | 7871 | btrfs_submit_bio_start_direct_io); |
e65e1535 | 7872 | goto err; |
1ae39938 JB |
7873 | } else if (write) { |
7874 | /* | |
7875 | * If we aren't doing async submit, calculate the csum of the | |
7876 | * bio now. | |
7877 | */ | |
bd242a08 | 7878 | ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1); |
1ae39938 JB |
7879 | if (ret) |
7880 | goto err; | |
23ea8e5a | 7881 | } else { |
85879573 OS |
7882 | u64 csum_offset; |
7883 | ||
7884 | csum_offset = file_offset - dip->logical_offset; | |
265fdfa6 | 7885 | csum_offset >>= fs_info->sectorsize_bits; |
55fc29be | 7886 | csum_offset *= fs_info->csum_size; |
85879573 | 7887 | btrfs_io_bio(bio)->csum = dip->csums + csum_offset; |
c2db1073 | 7888 | } |
1ae39938 | 7889 | map: |
08635bae | 7890 | ret = btrfs_map_bio(fs_info, bio, 0); |
e65e1535 | 7891 | err: |
e65e1535 MX |
7892 | return ret; |
7893 | } | |
7894 | ||
c36cac28 OS |
7895 | /* |
7896 | * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked | |
7897 | * or ordered extents whether or not we submit any bios. | |
7898 | */ | |
7899 | static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio, | |
7900 | struct inode *inode, | |
7901 | loff_t file_offset) | |
e65e1535 | 7902 | { |
c36cac28 | 7903 | const bool write = (bio_op(dio_bio) == REQ_OP_WRITE); |
85879573 OS |
7904 | const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM); |
7905 | size_t dip_size; | |
c36cac28 | 7906 | struct btrfs_dio_private *dip; |
c36cac28 | 7907 | |
85879573 OS |
7908 | dip_size = sizeof(*dip); |
7909 | if (!write && csum) { | |
7910 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
85879573 OS |
7911 | size_t nblocks; |
7912 | ||
265fdfa6 | 7913 | nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits; |
223486c2 | 7914 | dip_size += fs_info->csum_size * nblocks; |
85879573 OS |
7915 | } |
7916 | ||
7917 | dip = kzalloc(dip_size, GFP_NOFS); | |
c36cac28 OS |
7918 | if (!dip) |
7919 | return NULL; | |
7920 | ||
c36cac28 OS |
7921 | dip->inode = inode; |
7922 | dip->logical_offset = file_offset; | |
7923 | dip->bytes = dio_bio->bi_iter.bi_size; | |
1201b58b | 7924 | dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9; |
c36cac28 | 7925 | dip->dio_bio = dio_bio; |
e3b318d1 | 7926 | refcount_set(&dip->refs, 1); |
c36cac28 OS |
7927 | return dip; |
7928 | } | |
7929 | ||
f85781fb GR |
7930 | static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap, |
7931 | struct bio *dio_bio, loff_t file_offset) | |
c36cac28 OS |
7932 | { |
7933 | const bool write = (bio_op(dio_bio) == REQ_OP_WRITE); | |
0b246afa | 7934 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
769b4f24 OS |
7935 | const bool raid56 = (btrfs_data_alloc_profile(fs_info) & |
7936 | BTRFS_BLOCK_GROUP_RAID56_MASK); | |
c36cac28 | 7937 | struct btrfs_dio_private *dip; |
e65e1535 | 7938 | struct bio *bio; |
c36cac28 | 7939 | u64 start_sector; |
1ae39938 | 7940 | int async_submit = 0; |
725130ba LB |
7941 | u64 submit_len; |
7942 | int clone_offset = 0; | |
7943 | int clone_len; | |
5f4dc8fc | 7944 | int ret; |
58efbc9f | 7945 | blk_status_t status; |
89b798ad | 7946 | struct btrfs_io_geometry geom; |
f85781fb | 7947 | struct btrfs_dio_data *dio_data = iomap->private; |
e65e1535 | 7948 | |
c36cac28 OS |
7949 | dip = btrfs_create_dio_private(dio_bio, inode, file_offset); |
7950 | if (!dip) { | |
7951 | if (!write) { | |
7952 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, | |
7953 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
7954 | } | |
7955 | dio_bio->bi_status = BLK_STS_RESOURCE; | |
f85781fb GR |
7956 | bio_endio(dio_bio); |
7957 | return BLK_QC_T_NONE; | |
c36cac28 | 7958 | } |
facc8a22 | 7959 | |
334c16d8 | 7960 | if (!write) { |
85879573 OS |
7961 | /* |
7962 | * Load the csums up front to reduce csum tree searches and | |
7963 | * contention when submitting bios. | |
334c16d8 JB |
7964 | * |
7965 | * If we have csums disabled this will do nothing. | |
85879573 | 7966 | */ |
6275193e | 7967 | status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums); |
85879573 OS |
7968 | if (status != BLK_STS_OK) |
7969 | goto out_err; | |
02f57c7a JB |
7970 | } |
7971 | ||
769b4f24 OS |
7972 | start_sector = dio_bio->bi_iter.bi_sector; |
7973 | submit_len = dio_bio->bi_iter.bi_size; | |
53b381b3 | 7974 | |
3c91ee69 | 7975 | do { |
769b4f24 OS |
7976 | ret = btrfs_get_io_geometry(fs_info, btrfs_op(dio_bio), |
7977 | start_sector << 9, submit_len, | |
7978 | &geom); | |
7979 | if (ret) { | |
7980 | status = errno_to_blk_status(ret); | |
7981 | goto out_err; | |
7982 | } | |
7983 | ASSERT(geom.len <= INT_MAX); | |
7984 | ||
89b798ad | 7985 | clone_len = min_t(int, submit_len, geom.len); |
02f57c7a | 7986 | |
725130ba LB |
7987 | /* |
7988 | * This will never fail as it's passing GPF_NOFS and | |
7989 | * the allocation is backed by btrfs_bioset. | |
7990 | */ | |
769b4f24 | 7991 | bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len); |
725130ba LB |
7992 | bio->bi_private = dip; |
7993 | bio->bi_end_io = btrfs_end_dio_bio; | |
7994 | btrfs_io_bio(bio)->logical = file_offset; | |
7995 | ||
7996 | ASSERT(submit_len >= clone_len); | |
7997 | submit_len -= clone_len; | |
e65e1535 | 7998 | |
725130ba LB |
7999 | /* |
8000 | * Increase the count before we submit the bio so we know | |
8001 | * the end IO handler won't happen before we increase the | |
8002 | * count. Otherwise, the dip might get freed before we're | |
8003 | * done setting it up. | |
769b4f24 OS |
8004 | * |
8005 | * We transfer the initial reference to the last bio, so we | |
8006 | * don't need to increment the reference count for the last one. | |
725130ba | 8007 | */ |
769b4f24 OS |
8008 | if (submit_len > 0) { |
8009 | refcount_inc(&dip->refs); | |
8010 | /* | |
8011 | * If we are submitting more than one bio, submit them | |
8012 | * all asynchronously. The exception is RAID 5 or 6, as | |
8013 | * asynchronous checksums make it difficult to collect | |
8014 | * full stripe writes. | |
8015 | */ | |
8016 | if (!raid56) | |
8017 | async_submit = 1; | |
8018 | } | |
e65e1535 | 8019 | |
d0ee3934 | 8020 | status = btrfs_submit_dio_bio(bio, inode, file_offset, |
58efbc9f OS |
8021 | async_submit); |
8022 | if (status) { | |
725130ba | 8023 | bio_put(bio); |
769b4f24 OS |
8024 | if (submit_len > 0) |
8025 | refcount_dec(&dip->refs); | |
725130ba LB |
8026 | goto out_err; |
8027 | } | |
e65e1535 | 8028 | |
f85781fb | 8029 | dio_data->submitted += clone_len; |
725130ba LB |
8030 | clone_offset += clone_len; |
8031 | start_sector += clone_len >> 9; | |
8032 | file_offset += clone_len; | |
3c91ee69 | 8033 | } while (submit_len > 0); |
f85781fb | 8034 | return BLK_QC_T_NONE; |
e65e1535 | 8035 | |
e65e1535 | 8036 | out_err: |
769b4f24 OS |
8037 | dip->dio_bio->bi_status = status; |
8038 | btrfs_dio_private_put(dip); | |
f85781fb | 8039 | return BLK_QC_T_NONE; |
4b46fce2 JB |
8040 | } |
8041 | ||
4e4cabec | 8042 | const struct iomap_ops btrfs_dio_iomap_ops = { |
f85781fb GR |
8043 | .iomap_begin = btrfs_dio_iomap_begin, |
8044 | .iomap_end = btrfs_dio_iomap_end, | |
8045 | }; | |
8046 | ||
4e4cabec | 8047 | const struct iomap_dio_ops btrfs_dio_ops = { |
f85781fb GR |
8048 | .submit_io = btrfs_submit_direct, |
8049 | }; | |
8050 | ||
1506fcc8 | 8051 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
bab16e21 | 8052 | u64 start, u64 len) |
1506fcc8 | 8053 | { |
05dadc09 TI |
8054 | int ret; |
8055 | ||
45dd052e | 8056 | ret = fiemap_prep(inode, fieinfo, start, &len, 0); |
05dadc09 TI |
8057 | if (ret) |
8058 | return ret; | |
8059 | ||
facee0a0 | 8060 | return extent_fiemap(BTRFS_I(inode), fieinfo, start, len); |
1506fcc8 YS |
8061 | } |
8062 | ||
a52d9a80 | 8063 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8064 | { |
0f208812 NB |
8065 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
8066 | u64 start = page_offset(page); | |
8067 | u64 end = start + PAGE_SIZE - 1; | |
c1be9c1a | 8068 | unsigned long bio_flags = 0; |
0f208812 | 8069 | struct bio *bio = NULL; |
c1be9c1a NB |
8070 | int ret; |
8071 | ||
0f208812 NB |
8072 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); |
8073 | ||
8074 | ret = btrfs_do_readpage(page, NULL, &bio, &bio_flags, 0, NULL); | |
c1be9c1a NB |
8075 | if (bio) |
8076 | ret = submit_one_bio(bio, 0, bio_flags); | |
8077 | return ret; | |
9ebefb18 | 8078 | } |
1832a6d5 | 8079 | |
a52d9a80 | 8080 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8081 | { |
be7bd730 JB |
8082 | struct inode *inode = page->mapping->host; |
8083 | int ret; | |
b888db2b CM |
8084 | |
8085 | if (current->flags & PF_MEMALLOC) { | |
8086 | redirty_page_for_writepage(wbc, page); | |
8087 | unlock_page(page); | |
8088 | return 0; | |
8089 | } | |
be7bd730 JB |
8090 | |
8091 | /* | |
8092 | * If we are under memory pressure we will call this directly from the | |
8093 | * VM, we need to make sure we have the inode referenced for the ordered | |
8094 | * extent. If not just return like we didn't do anything. | |
8095 | */ | |
8096 | if (!igrab(inode)) { | |
8097 | redirty_page_for_writepage(wbc, page); | |
8098 | return AOP_WRITEPAGE_ACTIVATE; | |
8099 | } | |
0a9b0e53 | 8100 | ret = extent_write_full_page(page, wbc); |
be7bd730 JB |
8101 | btrfs_add_delayed_iput(inode); |
8102 | return ret; | |
9ebefb18 CM |
8103 | } |
8104 | ||
48a3b636 ES |
8105 | static int btrfs_writepages(struct address_space *mapping, |
8106 | struct writeback_control *wbc) | |
b293f02e | 8107 | { |
8ae225a8 | 8108 | return extent_writepages(mapping, wbc); |
b293f02e CM |
8109 | } |
8110 | ||
ba206a02 | 8111 | static void btrfs_readahead(struct readahead_control *rac) |
3ab2fb5a | 8112 | { |
ba206a02 | 8113 | extent_readahead(rac); |
3ab2fb5a | 8114 | } |
2a3ff0ad | 8115 | |
e6dcd2dc | 8116 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8117 | { |
477a30ba | 8118 | int ret = try_release_extent_mapping(page, gfp_flags); |
d1b89bc0 GJ |
8119 | if (ret == 1) |
8120 | detach_page_private(page); | |
a52d9a80 | 8121 | return ret; |
39279cc3 CM |
8122 | } |
8123 | ||
e6dcd2dc CM |
8124 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8125 | { | |
98509cfc CM |
8126 | if (PageWriteback(page) || PageDirty(page)) |
8127 | return 0; | |
3ba7ab22 | 8128 | return __btrfs_releasepage(page, gfp_flags); |
e6dcd2dc CM |
8129 | } |
8130 | ||
f8e66081 RG |
8131 | #ifdef CONFIG_MIGRATION |
8132 | static int btrfs_migratepage(struct address_space *mapping, | |
8133 | struct page *newpage, struct page *page, | |
8134 | enum migrate_mode mode) | |
8135 | { | |
8136 | int ret; | |
8137 | ||
8138 | ret = migrate_page_move_mapping(mapping, newpage, page, 0); | |
8139 | if (ret != MIGRATEPAGE_SUCCESS) | |
8140 | return ret; | |
8141 | ||
d1b89bc0 GJ |
8142 | if (page_has_private(page)) |
8143 | attach_page_private(newpage, detach_page_private(page)); | |
f8e66081 RG |
8144 | |
8145 | if (PagePrivate2(page)) { | |
8146 | ClearPagePrivate2(page); | |
8147 | SetPagePrivate2(newpage); | |
8148 | } | |
8149 | ||
8150 | if (mode != MIGRATE_SYNC_NO_COPY) | |
8151 | migrate_page_copy(newpage, page); | |
8152 | else | |
8153 | migrate_page_states(newpage, page); | |
8154 | return MIGRATEPAGE_SUCCESS; | |
8155 | } | |
8156 | #endif | |
8157 | ||
d47992f8 LC |
8158 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8159 | unsigned int length) | |
39279cc3 | 8160 | { |
53ac7ead NB |
8161 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
8162 | struct extent_io_tree *tree = &inode->io_tree; | |
e6dcd2dc | 8163 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8164 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 8165 | u64 page_start = page_offset(page); |
09cbfeaf | 8166 | u64 page_end = page_start + PAGE_SIZE - 1; |
dbfdb6d1 CR |
8167 | u64 start; |
8168 | u64 end; | |
53ac7ead | 8169 | int inode_evicting = inode->vfs_inode.i_state & I_FREEING; |
2766ff61 FM |
8170 | bool found_ordered = false; |
8171 | bool completed_ordered = false; | |
39279cc3 | 8172 | |
8b62b72b CM |
8173 | /* |
8174 | * we have the page locked, so new writeback can't start, | |
8175 | * and the dirty bit won't be cleared while we are here. | |
8176 | * | |
8177 | * Wait for IO on this page so that we can safely clear | |
8178 | * the PagePrivate2 bit and do ordered accounting | |
8179 | */ | |
e6dcd2dc | 8180 | wait_on_page_writeback(page); |
8b62b72b | 8181 | |
e6dcd2dc CM |
8182 | if (offset) { |
8183 | btrfs_releasepage(page, GFP_NOFS); | |
8184 | return; | |
8185 | } | |
131e404a FDBM |
8186 | |
8187 | if (!inode_evicting) | |
ff13db41 | 8188 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
28b325cf | 8189 | |
dbfdb6d1 | 8190 | start = page_start; |
28b325cf | 8191 | again: |
53ac7ead | 8192 | ordered = btrfs_lookup_ordered_range(inode, start, page_end - start + 1); |
e6dcd2dc | 8193 | if (ordered) { |
2766ff61 | 8194 | found_ordered = true; |
bffe633e OS |
8195 | end = min(page_end, |
8196 | ordered->file_offset + ordered->num_bytes - 1); | |
eb84ae03 | 8197 | /* |
2766ff61 FM |
8198 | * IO on this page will never be started, so we need to account |
8199 | * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW | |
8200 | * here, must leave that up for the ordered extent completion. | |
eb84ae03 | 8201 | */ |
131e404a | 8202 | if (!inode_evicting) |
dbfdb6d1 | 8203 | clear_extent_bit(tree, start, end, |
2766ff61 | 8204 | EXTENT_DELALLOC | |
131e404a | 8205 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
ae0f1625 | 8206 | EXTENT_DEFRAG, 1, 0, &cached_state); |
8b62b72b CM |
8207 | /* |
8208 | * whoever cleared the private bit is responsible | |
8209 | * for the finish_ordered_io | |
8210 | */ | |
77cef2ec JB |
8211 | if (TestClearPagePrivate2(page)) { |
8212 | struct btrfs_ordered_inode_tree *tree; | |
8213 | u64 new_len; | |
8214 | ||
53ac7ead | 8215 | tree = &inode->ordered_tree; |
77cef2ec JB |
8216 | |
8217 | spin_lock_irq(&tree->lock); | |
8218 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
dbfdb6d1 | 8219 | new_len = start - ordered->file_offset; |
77cef2ec JB |
8220 | if (new_len < ordered->truncated_len) |
8221 | ordered->truncated_len = new_len; | |
8222 | spin_unlock_irq(&tree->lock); | |
8223 | ||
53ac7ead NB |
8224 | if (btrfs_dec_test_ordered_pending(inode, &ordered, |
8225 | start, | |
2766ff61 | 8226 | end - start + 1, 1)) { |
77cef2ec | 8227 | btrfs_finish_ordered_io(ordered); |
2766ff61 FM |
8228 | completed_ordered = true; |
8229 | } | |
8b62b72b | 8230 | } |
e6dcd2dc | 8231 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
8232 | if (!inode_evicting) { |
8233 | cached_state = NULL; | |
dbfdb6d1 | 8234 | lock_extent_bits(tree, start, end, |
131e404a FDBM |
8235 | &cached_state); |
8236 | } | |
dbfdb6d1 CR |
8237 | |
8238 | start = end + 1; | |
8239 | if (start < page_end) | |
8240 | goto again; | |
131e404a FDBM |
8241 | } |
8242 | ||
b9d0b389 QW |
8243 | /* |
8244 | * Qgroup reserved space handler | |
8245 | * Page here will be either | |
fa91e4aa QW |
8246 | * 1) Already written to disk or ordered extent already submitted |
8247 | * Then its QGROUP_RESERVED bit in io_tree is already cleaned. | |
8248 | * Qgroup will be handled by its qgroup_record then. | |
8249 | * btrfs_qgroup_free_data() call will do nothing here. | |
8250 | * | |
8251 | * 2) Not written to disk yet | |
8252 | * Then btrfs_qgroup_free_data() call will clear the QGROUP_RESERVED | |
8253 | * bit of its io_tree, and free the qgroup reserved data space. | |
8254 | * Since the IO will never happen for this page. | |
b9d0b389 | 8255 | */ |
53ac7ead | 8256 | btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE); |
131e404a | 8257 | if (!inode_evicting) { |
2766ff61 FM |
8258 | bool delete = true; |
8259 | ||
8260 | /* | |
8261 | * If there's an ordered extent for this range and we have not | |
8262 | * finished it ourselves, we must leave EXTENT_DELALLOC_NEW set | |
8263 | * in the range for the ordered extent completion. We must also | |
8264 | * not delete the range, otherwise we would lose that bit (and | |
8265 | * any other bits set in the range). Make sure EXTENT_UPTODATE | |
8266 | * is cleared if we don't delete, otherwise it can lead to | |
8267 | * corruptions if the i_size is extented later. | |
8268 | */ | |
8269 | if (found_ordered && !completed_ordered) | |
8270 | delete = false; | |
e182163d | 8271 | clear_extent_bit(tree, page_start, page_end, EXTENT_LOCKED | |
2766ff61 FM |
8272 | EXTENT_DELALLOC | EXTENT_UPTODATE | |
8273 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, | |
8274 | delete, &cached_state); | |
131e404a FDBM |
8275 | |
8276 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 8277 | } |
e6dcd2dc | 8278 | |
4a096752 | 8279 | ClearPageChecked(page); |
d1b89bc0 | 8280 | detach_page_private(page); |
39279cc3 CM |
8281 | } |
8282 | ||
9ebefb18 CM |
8283 | /* |
8284 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8285 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8286 | * be careful to check for EOF conditions here. We set the page up correctly | |
8287 | * for a written page which means we get ENOSPC checking when writing into | |
8288 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8289 | * support these features. | |
8290 | * | |
8291 | * We are not allowed to take the i_mutex here so we have to play games to | |
8292 | * protect against truncate races as the page could now be beyond EOF. Because | |
d1342aad OS |
8293 | * truncate_setsize() writes the inode size before removing pages, once we have |
8294 | * the page lock we can determine safely if the page is beyond EOF. If it is not | |
9ebefb18 CM |
8295 | * beyond EOF, then the page is guaranteed safe against truncation until we |
8296 | * unlock the page. | |
8297 | */ | |
a528a241 | 8298 | vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf) |
9ebefb18 | 8299 | { |
c2ec175c | 8300 | struct page *page = vmf->page; |
11bac800 | 8301 | struct inode *inode = file_inode(vmf->vma->vm_file); |
0b246afa | 8302 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
8303 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8304 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8305 | struct extent_state *cached_state = NULL; |
364ecf36 | 8306 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc CM |
8307 | char *kaddr; |
8308 | unsigned long zero_start; | |
9ebefb18 | 8309 | loff_t size; |
a528a241 SJ |
8310 | vm_fault_t ret; |
8311 | int ret2; | |
9998eb70 | 8312 | int reserved = 0; |
d0b7da88 | 8313 | u64 reserved_space; |
a52d9a80 | 8314 | u64 page_start; |
e6dcd2dc | 8315 | u64 page_end; |
d0b7da88 CR |
8316 | u64 end; |
8317 | ||
09cbfeaf | 8318 | reserved_space = PAGE_SIZE; |
9ebefb18 | 8319 | |
b2b5ef5c | 8320 | sb_start_pagefault(inode->i_sb); |
df480633 | 8321 | page_start = page_offset(page); |
09cbfeaf | 8322 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 8323 | end = page_end; |
df480633 | 8324 | |
d0b7da88 CR |
8325 | /* |
8326 | * Reserving delalloc space after obtaining the page lock can lead to | |
8327 | * deadlock. For example, if a dirty page is locked by this function | |
8328 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
8329 | * dirty page write out, then the btrfs_writepage() function could | |
8330 | * end up waiting indefinitely to get a lock on the page currently | |
8331 | * being processed by btrfs_page_mkwrite() function. | |
8332 | */ | |
e5b7231e NB |
8333 | ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved, |
8334 | page_start, reserved_space); | |
a528a241 SJ |
8335 | if (!ret2) { |
8336 | ret2 = file_update_time(vmf->vma->vm_file); | |
9998eb70 CM |
8337 | reserved = 1; |
8338 | } | |
a528a241 SJ |
8339 | if (ret2) { |
8340 | ret = vmf_error(ret2); | |
9998eb70 CM |
8341 | if (reserved) |
8342 | goto out; | |
8343 | goto out_noreserve; | |
56a76f82 | 8344 | } |
1832a6d5 | 8345 | |
56a76f82 | 8346 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8347 | again: |
9ebefb18 | 8348 | lock_page(page); |
9ebefb18 | 8349 | size = i_size_read(inode); |
a52d9a80 | 8350 | |
9ebefb18 | 8351 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8352 | (page_start >= size)) { |
9ebefb18 CM |
8353 | /* page got truncated out from underneath us */ |
8354 | goto out_unlock; | |
8355 | } | |
e6dcd2dc CM |
8356 | wait_on_page_writeback(page); |
8357 | ||
ff13db41 | 8358 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc CM |
8359 | set_page_extent_mapped(page); |
8360 | ||
eb84ae03 CM |
8361 | /* |
8362 | * we can't set the delalloc bits if there are pending ordered | |
8363 | * extents. Drop our locks and wait for them to finish | |
8364 | */ | |
a776c6fa NB |
8365 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
8366 | PAGE_SIZE); | |
e6dcd2dc | 8367 | if (ordered) { |
2ac55d41 | 8368 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 8369 | &cached_state); |
e6dcd2dc | 8370 | unlock_page(page); |
c0a43603 | 8371 | btrfs_start_ordered_extent(ordered, 1); |
e6dcd2dc CM |
8372 | btrfs_put_ordered_extent(ordered); |
8373 | goto again; | |
8374 | } | |
8375 | ||
09cbfeaf | 8376 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 8377 | reserved_space = round_up(size - page_start, |
0b246afa | 8378 | fs_info->sectorsize); |
09cbfeaf | 8379 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 | 8380 | end = page_start + reserved_space - 1; |
86d52921 NB |
8381 | btrfs_delalloc_release_space(BTRFS_I(inode), |
8382 | data_reserved, page_start, | |
8383 | PAGE_SIZE - reserved_space, true); | |
d0b7da88 CR |
8384 | } |
8385 | } | |
8386 | ||
fbf19087 | 8387 | /* |
5416034f LB |
8388 | * page_mkwrite gets called when the page is firstly dirtied after it's |
8389 | * faulted in, but write(2) could also dirty a page and set delalloc | |
8390 | * bits, thus in this case for space account reason, we still need to | |
8391 | * clear any delalloc bits within this page range since we have to | |
8392 | * reserve data&meta space before lock_page() (see above comments). | |
fbf19087 | 8393 | */ |
d0b7da88 | 8394 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
e182163d OS |
8395 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | |
8396 | EXTENT_DEFRAG, 0, 0, &cached_state); | |
fbf19087 | 8397 | |
c2566f22 | 8398 | ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0, |
330a5827 | 8399 | &cached_state); |
a528a241 | 8400 | if (ret2) { |
2ac55d41 | 8401 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 8402 | &cached_state); |
9ed74f2d JB |
8403 | ret = VM_FAULT_SIGBUS; |
8404 | goto out_unlock; | |
8405 | } | |
9ebefb18 CM |
8406 | |
8407 | /* page is wholly or partially inside EOF */ | |
09cbfeaf | 8408 | if (page_start + PAGE_SIZE > size) |
7073017a | 8409 | zero_start = offset_in_page(size); |
9ebefb18 | 8410 | else |
09cbfeaf | 8411 | zero_start = PAGE_SIZE; |
9ebefb18 | 8412 | |
09cbfeaf | 8413 | if (zero_start != PAGE_SIZE) { |
e6dcd2dc | 8414 | kaddr = kmap(page); |
09cbfeaf | 8415 | memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start); |
e6dcd2dc CM |
8416 | flush_dcache_page(page); |
8417 | kunmap(page); | |
8418 | } | |
247e743c | 8419 | ClearPageChecked(page); |
e6dcd2dc | 8420 | set_page_dirty(page); |
50a9b214 | 8421 | SetPageUptodate(page); |
5a3f23d5 | 8422 | |
0b246afa | 8423 | BTRFS_I(inode)->last_trans = fs_info->generation; |
257c62e1 | 8424 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; |
46d8bc34 | 8425 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 8426 | |
e43bbe5e | 8427 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state); |
9ebefb18 | 8428 | |
76de60ed YY |
8429 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
8430 | sb_end_pagefault(inode->i_sb); | |
8431 | extent_changeset_free(data_reserved); | |
8432 | return VM_FAULT_LOCKED; | |
717beb96 CM |
8433 | |
8434 | out_unlock: | |
9ebefb18 | 8435 | unlock_page(page); |
1832a6d5 | 8436 | out: |
8702ba93 | 8437 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
86d52921 | 8438 | btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start, |
43b18595 | 8439 | reserved_space, (ret != 0)); |
9998eb70 | 8440 | out_noreserve: |
b2b5ef5c | 8441 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 8442 | extent_changeset_free(data_reserved); |
9ebefb18 CM |
8443 | return ret; |
8444 | } | |
8445 | ||
213e8c55 | 8446 | static int btrfs_truncate(struct inode *inode, bool skip_writeback) |
39279cc3 | 8447 | { |
0b246afa | 8448 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 8449 | struct btrfs_root *root = BTRFS_I(inode)->root; |
fcb80c2a | 8450 | struct btrfs_block_rsv *rsv; |
ad7e1a74 | 8451 | int ret; |
39279cc3 | 8452 | struct btrfs_trans_handle *trans; |
0b246afa | 8453 | u64 mask = fs_info->sectorsize - 1; |
2bd36e7b | 8454 | u64 min_size = btrfs_calc_metadata_size(fs_info, 1); |
39279cc3 | 8455 | |
213e8c55 FM |
8456 | if (!skip_writeback) { |
8457 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), | |
8458 | (u64)-1); | |
8459 | if (ret) | |
8460 | return ret; | |
8461 | } | |
39279cc3 | 8462 | |
fcb80c2a | 8463 | /* |
f7e9e8fc OS |
8464 | * Yes ladies and gentlemen, this is indeed ugly. We have a couple of |
8465 | * things going on here: | |
fcb80c2a | 8466 | * |
f7e9e8fc | 8467 | * 1) We need to reserve space to update our inode. |
fcb80c2a | 8468 | * |
f7e9e8fc | 8469 | * 2) We need to have something to cache all the space that is going to |
fcb80c2a JB |
8470 | * be free'd up by the truncate operation, but also have some slack |
8471 | * space reserved in case it uses space during the truncate (thank you | |
8472 | * very much snapshotting). | |
8473 | * | |
f7e9e8fc | 8474 | * And we need these to be separate. The fact is we can use a lot of |
fcb80c2a | 8475 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 8476 | * we will use, so we need the truncate reservation to be separate so it |
f7e9e8fc OS |
8477 | * doesn't end up using space reserved for updating the inode. We also |
8478 | * need to be able to stop the transaction and start a new one, which | |
8479 | * means we need to be able to update the inode several times, and we | |
8480 | * have no idea of knowing how many times that will be, so we can't just | |
8481 | * reserve 1 item for the entirety of the operation, so that has to be | |
8482 | * done separately as well. | |
fcb80c2a JB |
8483 | * |
8484 | * So that leaves us with | |
8485 | * | |
f7e9e8fc | 8486 | * 1) rsv - for the truncate reservation, which we will steal from the |
fcb80c2a | 8487 | * transaction reservation. |
f7e9e8fc | 8488 | * 2) fs_info->trans_block_rsv - this will have 1 items worth left for |
fcb80c2a JB |
8489 | * updating the inode. |
8490 | */ | |
2ff7e61e | 8491 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
8492 | if (!rsv) |
8493 | return -ENOMEM; | |
4a338542 | 8494 | rsv->size = min_size; |
ca7e70f5 | 8495 | rsv->failfast = 1; |
f0cd846e | 8496 | |
907cbceb | 8497 | /* |
07127184 | 8498 | * 1 for the truncate slack space |
907cbceb JB |
8499 | * 1 for updating the inode. |
8500 | */ | |
f3fe820c | 8501 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a | 8502 | if (IS_ERR(trans)) { |
ad7e1a74 | 8503 | ret = PTR_ERR(trans); |
fcb80c2a JB |
8504 | goto out; |
8505 | } | |
f0cd846e | 8506 | |
907cbceb | 8507 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 8508 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
3a584174 | 8509 | min_size, false); |
fcb80c2a | 8510 | BUG_ON(ret); |
f0cd846e | 8511 | |
5dc562c5 JB |
8512 | /* |
8513 | * So if we truncate and then write and fsync we normally would just | |
8514 | * write the extents that changed, which is a problem if we need to | |
8515 | * first truncate that entire inode. So set this flag so we write out | |
8516 | * all of the extents in the inode to the sync log so we're completely | |
8517 | * safe. | |
8518 | */ | |
8519 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 8520 | trans->block_rsv = rsv; |
907cbceb | 8521 | |
8082510e | 8522 | while (1) { |
50743398 | 8523 | ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode), |
8082510e YZ |
8524 | inode->i_size, |
8525 | BTRFS_EXTENT_DATA_KEY); | |
ddfae63c | 8526 | trans->block_rsv = &fs_info->trans_block_rsv; |
ad7e1a74 | 8527 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 8528 | break; |
39279cc3 | 8529 | |
9a56fcd1 | 8530 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
ad7e1a74 | 8531 | if (ret) |
3893e33b | 8532 | break; |
ca7e70f5 | 8533 | |
3a45bb20 | 8534 | btrfs_end_transaction(trans); |
2ff7e61e | 8535 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
8536 | |
8537 | trans = btrfs_start_transaction(root, 2); | |
8538 | if (IS_ERR(trans)) { | |
ad7e1a74 | 8539 | ret = PTR_ERR(trans); |
ca7e70f5 JB |
8540 | trans = NULL; |
8541 | break; | |
8542 | } | |
8543 | ||
63f018be | 8544 | btrfs_block_rsv_release(fs_info, rsv, -1, NULL); |
0b246afa | 8545 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
3a584174 | 8546 | rsv, min_size, false); |
ca7e70f5 JB |
8547 | BUG_ON(ret); /* shouldn't happen */ |
8548 | trans->block_rsv = rsv; | |
8082510e YZ |
8549 | } |
8550 | ||
ddfae63c JB |
8551 | /* |
8552 | * We can't call btrfs_truncate_block inside a trans handle as we could | |
8553 | * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know | |
8554 | * we've truncated everything except the last little bit, and can do | |
8555 | * btrfs_truncate_block and then update the disk_i_size. | |
8556 | */ | |
8557 | if (ret == NEED_TRUNCATE_BLOCK) { | |
8558 | btrfs_end_transaction(trans); | |
8559 | btrfs_btree_balance_dirty(fs_info); | |
8560 | ||
217f42eb | 8561 | ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0); |
ddfae63c JB |
8562 | if (ret) |
8563 | goto out; | |
8564 | trans = btrfs_start_transaction(root, 1); | |
8565 | if (IS_ERR(trans)) { | |
8566 | ret = PTR_ERR(trans); | |
8567 | goto out; | |
8568 | } | |
76aea537 | 8569 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
ddfae63c JB |
8570 | } |
8571 | ||
917c16b2 | 8572 | if (trans) { |
ad7e1a74 OS |
8573 | int ret2; |
8574 | ||
0b246afa | 8575 | trans->block_rsv = &fs_info->trans_block_rsv; |
9a56fcd1 | 8576 | ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
ad7e1a74 OS |
8577 | if (ret2 && !ret) |
8578 | ret = ret2; | |
7b128766 | 8579 | |
ad7e1a74 OS |
8580 | ret2 = btrfs_end_transaction(trans); |
8581 | if (ret2 && !ret) | |
8582 | ret = ret2; | |
2ff7e61e | 8583 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 8584 | } |
fcb80c2a | 8585 | out: |
2ff7e61e | 8586 | btrfs_free_block_rsv(fs_info, rsv); |
fcb80c2a | 8587 | |
ad7e1a74 | 8588 | return ret; |
39279cc3 CM |
8589 | } |
8590 | ||
d352ac68 CM |
8591 | /* |
8592 | * create a new subvolume directory/inode (helper for the ioctl). | |
8593 | */ | |
d2fb3437 | 8594 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
8595 | struct btrfs_root *new_root, |
8596 | struct btrfs_root *parent_root, | |
8597 | u64 new_dirid) | |
39279cc3 | 8598 | { |
39279cc3 | 8599 | struct inode *inode; |
76dda93c | 8600 | int err; |
00e4e6b3 | 8601 | u64 index = 0; |
39279cc3 | 8602 | |
12fc9d09 FA |
8603 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
8604 | new_dirid, new_dirid, | |
8605 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
8606 | &index); | |
54aa1f4d | 8607 | if (IS_ERR(inode)) |
f46b5a66 | 8608 | return PTR_ERR(inode); |
39279cc3 CM |
8609 | inode->i_op = &btrfs_dir_inode_operations; |
8610 | inode->i_fop = &btrfs_dir_file_operations; | |
8611 | ||
bfe86848 | 8612 | set_nlink(inode, 1); |
6ef06d27 | 8613 | btrfs_i_size_write(BTRFS_I(inode), 0); |
b0d5d10f | 8614 | unlock_new_inode(inode); |
3b96362c | 8615 | |
63541927 FDBM |
8616 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
8617 | if (err) | |
8618 | btrfs_err(new_root->fs_info, | |
351fd353 | 8619 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
8620 | new_root->root_key.objectid, err); |
8621 | ||
9a56fcd1 | 8622 | err = btrfs_update_inode(trans, new_root, BTRFS_I(inode)); |
cb8e7090 | 8623 | |
76dda93c | 8624 | iput(inode); |
ce598979 | 8625 | return err; |
39279cc3 CM |
8626 | } |
8627 | ||
39279cc3 CM |
8628 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
8629 | { | |
69fe2d75 | 8630 | struct btrfs_fs_info *fs_info = btrfs_sb(sb); |
39279cc3 | 8631 | struct btrfs_inode *ei; |
2ead6ae7 | 8632 | struct inode *inode; |
39279cc3 | 8633 | |
712e36c5 | 8634 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL); |
39279cc3 CM |
8635 | if (!ei) |
8636 | return NULL; | |
2ead6ae7 YZ |
8637 | |
8638 | ei->root = NULL; | |
2ead6ae7 | 8639 | ei->generation = 0; |
15ee9bc7 | 8640 | ei->last_trans = 0; |
257c62e1 | 8641 | ei->last_sub_trans = 0; |
e02119d5 | 8642 | ei->logged_trans = 0; |
2ead6ae7 | 8643 | ei->delalloc_bytes = 0; |
a7e3b975 | 8644 | ei->new_delalloc_bytes = 0; |
47059d93 | 8645 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
8646 | ei->disk_i_size = 0; |
8647 | ei->flags = 0; | |
7709cde3 | 8648 | ei->csum_bytes = 0; |
2ead6ae7 | 8649 | ei->index_cnt = (u64)-1; |
67de1176 | 8650 | ei->dir_index = 0; |
2ead6ae7 | 8651 | ei->last_unlink_trans = 0; |
3ebac17c | 8652 | ei->last_reflink_trans = 0; |
46d8bc34 | 8653 | ei->last_log_commit = 0; |
2ead6ae7 | 8654 | |
9e0baf60 JB |
8655 | spin_lock_init(&ei->lock); |
8656 | ei->outstanding_extents = 0; | |
69fe2d75 JB |
8657 | if (sb->s_magic != BTRFS_TEST_MAGIC) |
8658 | btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv, | |
8659 | BTRFS_BLOCK_RSV_DELALLOC); | |
72ac3c0d | 8660 | ei->runtime_flags = 0; |
b52aa8c9 | 8661 | ei->prop_compress = BTRFS_COMPRESS_NONE; |
eec63c65 | 8662 | ei->defrag_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 8663 | |
16cdcec7 MX |
8664 | ei->delayed_node = NULL; |
8665 | ||
9cc97d64 | 8666 | ei->i_otime.tv_sec = 0; |
8667 | ei->i_otime.tv_nsec = 0; | |
8668 | ||
2ead6ae7 | 8669 | inode = &ei->vfs_inode; |
a8067e02 | 8670 | extent_map_tree_init(&ei->extent_tree); |
43eb5f29 QW |
8671 | extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode); |
8672 | extent_io_tree_init(fs_info, &ei->io_failure_tree, | |
8673 | IO_TREE_INODE_IO_FAILURE, inode); | |
41a2ee75 JB |
8674 | extent_io_tree_init(fs_info, &ei->file_extent_tree, |
8675 | IO_TREE_INODE_FILE_EXTENT, inode); | |
7b439738 DS |
8676 | ei->io_tree.track_uptodate = true; |
8677 | ei->io_failure_tree.track_uptodate = true; | |
b812ce28 | 8678 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 8679 | mutex_init(&ei->log_mutex); |
e6dcd2dc | 8680 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 8681 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 8682 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 YZ |
8683 | RB_CLEAR_NODE(&ei->rb_node); |
8684 | ||
8685 | return inode; | |
39279cc3 CM |
8686 | } |
8687 | ||
aaedb55b JB |
8688 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
8689 | void btrfs_test_destroy_inode(struct inode *inode) | |
8690 | { | |
dcdbc059 | 8691 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
aaedb55b JB |
8692 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8693 | } | |
8694 | #endif | |
8695 | ||
26602cab | 8696 | void btrfs_free_inode(struct inode *inode) |
fa0d7e3d | 8697 | { |
fa0d7e3d NP |
8698 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8699 | } | |
8700 | ||
633cc816 | 8701 | void btrfs_destroy_inode(struct inode *vfs_inode) |
39279cc3 | 8702 | { |
e6dcd2dc | 8703 | struct btrfs_ordered_extent *ordered; |
633cc816 NB |
8704 | struct btrfs_inode *inode = BTRFS_I(vfs_inode); |
8705 | struct btrfs_root *root = inode->root; | |
5a3f23d5 | 8706 | |
633cc816 NB |
8707 | WARN_ON(!hlist_empty(&vfs_inode->i_dentry)); |
8708 | WARN_ON(vfs_inode->i_data.nrpages); | |
8709 | WARN_ON(inode->block_rsv.reserved); | |
8710 | WARN_ON(inode->block_rsv.size); | |
8711 | WARN_ON(inode->outstanding_extents); | |
8712 | WARN_ON(inode->delalloc_bytes); | |
8713 | WARN_ON(inode->new_delalloc_bytes); | |
8714 | WARN_ON(inode->csum_bytes); | |
8715 | WARN_ON(inode->defrag_bytes); | |
39279cc3 | 8716 | |
a6dbd429 JB |
8717 | /* |
8718 | * This can happen where we create an inode, but somebody else also | |
8719 | * created the same inode and we need to destroy the one we already | |
8720 | * created. | |
8721 | */ | |
8722 | if (!root) | |
26602cab | 8723 | return; |
a6dbd429 | 8724 | |
d397712b | 8725 | while (1) { |
633cc816 | 8726 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
e6dcd2dc CM |
8727 | if (!ordered) |
8728 | break; | |
8729 | else { | |
633cc816 | 8730 | btrfs_err(root->fs_info, |
5d163e0e | 8731 | "found ordered extent %llu %llu on inode cleanup", |
bffe633e | 8732 | ordered->file_offset, ordered->num_bytes); |
71fe0a55 | 8733 | btrfs_remove_ordered_extent(inode, ordered); |
e6dcd2dc CM |
8734 | btrfs_put_ordered_extent(ordered); |
8735 | btrfs_put_ordered_extent(ordered); | |
8736 | } | |
8737 | } | |
633cc816 NB |
8738 | btrfs_qgroup_check_reserved_leak(inode); |
8739 | inode_tree_del(inode); | |
8740 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); | |
8741 | btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1); | |
8742 | btrfs_put_root(inode->root); | |
39279cc3 CM |
8743 | } |
8744 | ||
45321ac5 | 8745 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
8746 | { |
8747 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 8748 | |
6379ef9f NA |
8749 | if (root == NULL) |
8750 | return 1; | |
8751 | ||
fa6ac876 | 8752 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 8753 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 8754 | return 1; |
76dda93c | 8755 | else |
45321ac5 | 8756 | return generic_drop_inode(inode); |
76dda93c YZ |
8757 | } |
8758 | ||
0ee0fda0 | 8759 | static void init_once(void *foo) |
39279cc3 CM |
8760 | { |
8761 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
8762 | ||
8763 | inode_init_once(&ei->vfs_inode); | |
8764 | } | |
8765 | ||
e67c718b | 8766 | void __cold btrfs_destroy_cachep(void) |
39279cc3 | 8767 | { |
8c0a8537 KS |
8768 | /* |
8769 | * Make sure all delayed rcu free inodes are flushed before we | |
8770 | * destroy cache. | |
8771 | */ | |
8772 | rcu_barrier(); | |
5598e900 KM |
8773 | kmem_cache_destroy(btrfs_inode_cachep); |
8774 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5598e900 KM |
8775 | kmem_cache_destroy(btrfs_path_cachep); |
8776 | kmem_cache_destroy(btrfs_free_space_cachep); | |
3acd4850 | 8777 | kmem_cache_destroy(btrfs_free_space_bitmap_cachep); |
39279cc3 CM |
8778 | } |
8779 | ||
f5c29bd9 | 8780 | int __init btrfs_init_cachep(void) |
39279cc3 | 8781 | { |
837e1972 | 8782 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 8783 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
8784 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
8785 | init_once); | |
39279cc3 CM |
8786 | if (!btrfs_inode_cachep) |
8787 | goto fail; | |
9601e3f6 | 8788 | |
837e1972 | 8789 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 | 8790 | sizeof(struct btrfs_trans_handle), 0, |
fba4b697 | 8791 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
8792 | if (!btrfs_trans_handle_cachep) |
8793 | goto fail; | |
9601e3f6 | 8794 | |
837e1972 | 8795 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 | 8796 | sizeof(struct btrfs_path), 0, |
fba4b697 | 8797 | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
8798 | if (!btrfs_path_cachep) |
8799 | goto fail; | |
9601e3f6 | 8800 | |
837e1972 | 8801 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 | 8802 | sizeof(struct btrfs_free_space), 0, |
fba4b697 | 8803 | SLAB_MEM_SPREAD, NULL); |
dc89e982 JB |
8804 | if (!btrfs_free_space_cachep) |
8805 | goto fail; | |
8806 | ||
3acd4850 CL |
8807 | btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap", |
8808 | PAGE_SIZE, PAGE_SIZE, | |
8809 | SLAB_RED_ZONE, NULL); | |
8810 | if (!btrfs_free_space_bitmap_cachep) | |
8811 | goto fail; | |
8812 | ||
39279cc3 CM |
8813 | return 0; |
8814 | fail: | |
8815 | btrfs_destroy_cachep(); | |
8816 | return -ENOMEM; | |
8817 | } | |
8818 | ||
a528d35e DH |
8819 | static int btrfs_getattr(const struct path *path, struct kstat *stat, |
8820 | u32 request_mask, unsigned int flags) | |
39279cc3 | 8821 | { |
df0af1a5 | 8822 | u64 delalloc_bytes; |
2766ff61 | 8823 | u64 inode_bytes; |
a528d35e | 8824 | struct inode *inode = d_inode(path->dentry); |
fadc0d8b | 8825 | u32 blocksize = inode->i_sb->s_blocksize; |
04a87e34 YS |
8826 | u32 bi_flags = BTRFS_I(inode)->flags; |
8827 | ||
8828 | stat->result_mask |= STATX_BTIME; | |
8829 | stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec; | |
8830 | stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec; | |
8831 | if (bi_flags & BTRFS_INODE_APPEND) | |
8832 | stat->attributes |= STATX_ATTR_APPEND; | |
8833 | if (bi_flags & BTRFS_INODE_COMPRESS) | |
8834 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
8835 | if (bi_flags & BTRFS_INODE_IMMUTABLE) | |
8836 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
8837 | if (bi_flags & BTRFS_INODE_NODUMP) | |
8838 | stat->attributes |= STATX_ATTR_NODUMP; | |
8839 | ||
8840 | stat->attributes_mask |= (STATX_ATTR_APPEND | | |
8841 | STATX_ATTR_COMPRESSED | | |
8842 | STATX_ATTR_IMMUTABLE | | |
8843 | STATX_ATTR_NODUMP); | |
fadc0d8b | 8844 | |
39279cc3 | 8845 | generic_fillattr(inode, stat); |
0ee5dc67 | 8846 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
8847 | |
8848 | spin_lock(&BTRFS_I(inode)->lock); | |
a7e3b975 | 8849 | delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes; |
2766ff61 | 8850 | inode_bytes = inode_get_bytes(inode); |
df0af1a5 | 8851 | spin_unlock(&BTRFS_I(inode)->lock); |
2766ff61 | 8852 | stat->blocks = (ALIGN(inode_bytes, blocksize) + |
df0af1a5 | 8853 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
8854 | return 0; |
8855 | } | |
8856 | ||
cdd1fedf DF |
8857 | static int btrfs_rename_exchange(struct inode *old_dir, |
8858 | struct dentry *old_dentry, | |
8859 | struct inode *new_dir, | |
8860 | struct dentry *new_dentry) | |
8861 | { | |
0b246afa | 8862 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf DF |
8863 | struct btrfs_trans_handle *trans; |
8864 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
8865 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
8866 | struct inode *new_inode = new_dentry->d_inode; | |
8867 | struct inode *old_inode = old_dentry->d_inode; | |
95582b00 | 8868 | struct timespec64 ctime = current_time(old_inode); |
4a0cc7ca NB |
8869 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
8870 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
8871 | u64 old_idx = 0; |
8872 | u64 new_idx = 0; | |
cdd1fedf | 8873 | int ret; |
75b463d2 | 8874 | int ret2; |
86e8aa0e FM |
8875 | bool root_log_pinned = false; |
8876 | bool dest_log_pinned = false; | |
cdd1fedf DF |
8877 | |
8878 | /* we only allow rename subvolume link between subvolumes */ | |
8879 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
8880 | return -EXDEV; | |
8881 | ||
8882 | /* close the race window with snapshot create/destroy ioctl */ | |
943eb3bf JB |
8883 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID || |
8884 | new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 8885 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
8886 | |
8887 | /* | |
8888 | * We want to reserve the absolute worst case amount of items. So if | |
8889 | * both inodes are subvols and we need to unlink them then that would | |
8890 | * require 4 item modifications, but if they are both normal inodes it | |
8891 | * would require 5 item modifications, so we'll assume their normal | |
8892 | * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items | |
8893 | * should cover the worst case number of items we'll modify. | |
8894 | */ | |
8895 | trans = btrfs_start_transaction(root, 12); | |
8896 | if (IS_ERR(trans)) { | |
8897 | ret = PTR_ERR(trans); | |
8898 | goto out_notrans; | |
8899 | } | |
8900 | ||
3e174099 JB |
8901 | if (dest != root) |
8902 | btrfs_record_root_in_trans(trans, dest); | |
8903 | ||
cdd1fedf DF |
8904 | /* |
8905 | * We need to find a free sequence number both in the source and | |
8906 | * in the destination directory for the exchange. | |
8907 | */ | |
877574e2 | 8908 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx); |
cdd1fedf DF |
8909 | if (ret) |
8910 | goto out_fail; | |
877574e2 | 8911 | ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx); |
cdd1fedf DF |
8912 | if (ret) |
8913 | goto out_fail; | |
8914 | ||
8915 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
8916 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
8917 | ||
8918 | /* Reference for the source. */ | |
8919 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
8920 | /* force full log commit if subvolume involved. */ | |
90787766 | 8921 | btrfs_set_log_full_commit(trans); |
cdd1fedf | 8922 | } else { |
376e5a57 FM |
8923 | btrfs_pin_log_trans(root); |
8924 | root_log_pinned = true; | |
cdd1fedf DF |
8925 | ret = btrfs_insert_inode_ref(trans, dest, |
8926 | new_dentry->d_name.name, | |
8927 | new_dentry->d_name.len, | |
8928 | old_ino, | |
f85b7379 DS |
8929 | btrfs_ino(BTRFS_I(new_dir)), |
8930 | old_idx); | |
cdd1fedf DF |
8931 | if (ret) |
8932 | goto out_fail; | |
cdd1fedf DF |
8933 | } |
8934 | ||
8935 | /* And now for the dest. */ | |
8936 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
8937 | /* force full log commit if subvolume involved. */ | |
90787766 | 8938 | btrfs_set_log_full_commit(trans); |
cdd1fedf | 8939 | } else { |
376e5a57 FM |
8940 | btrfs_pin_log_trans(dest); |
8941 | dest_log_pinned = true; | |
cdd1fedf DF |
8942 | ret = btrfs_insert_inode_ref(trans, root, |
8943 | old_dentry->d_name.name, | |
8944 | old_dentry->d_name.len, | |
8945 | new_ino, | |
f85b7379 DS |
8946 | btrfs_ino(BTRFS_I(old_dir)), |
8947 | new_idx); | |
cdd1fedf DF |
8948 | if (ret) |
8949 | goto out_fail; | |
cdd1fedf DF |
8950 | } |
8951 | ||
8952 | /* Update inode version and ctime/mtime. */ | |
8953 | inode_inc_iversion(old_dir); | |
8954 | inode_inc_iversion(new_dir); | |
8955 | inode_inc_iversion(old_inode); | |
8956 | inode_inc_iversion(new_inode); | |
8957 | old_dir->i_ctime = old_dir->i_mtime = ctime; | |
8958 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
8959 | old_inode->i_ctime = ctime; | |
8960 | new_inode->i_ctime = ctime; | |
8961 | ||
8962 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 DS |
8963 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
8964 | BTRFS_I(old_inode), 1); | |
8965 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), | |
8966 | BTRFS_I(new_inode), 1); | |
cdd1fedf DF |
8967 | } |
8968 | ||
8969 | /* src is a subvolume */ | |
8970 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
045d3967 | 8971 | ret = btrfs_unlink_subvol(trans, old_dir, old_dentry); |
cdd1fedf | 8972 | } else { /* src is an inode */ |
4ec5934e NB |
8973 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
8974 | BTRFS_I(old_dentry->d_inode), | |
cdd1fedf DF |
8975 | old_dentry->d_name.name, |
8976 | old_dentry->d_name.len); | |
8977 | if (!ret) | |
9a56fcd1 | 8978 | ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode)); |
cdd1fedf DF |
8979 | } |
8980 | if (ret) { | |
66642832 | 8981 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
8982 | goto out_fail; |
8983 | } | |
8984 | ||
8985 | /* dest is a subvolume */ | |
8986 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
045d3967 | 8987 | ret = btrfs_unlink_subvol(trans, new_dir, new_dentry); |
cdd1fedf | 8988 | } else { /* dest is an inode */ |
4ec5934e NB |
8989 | ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
8990 | BTRFS_I(new_dentry->d_inode), | |
cdd1fedf DF |
8991 | new_dentry->d_name.name, |
8992 | new_dentry->d_name.len); | |
8993 | if (!ret) | |
9a56fcd1 | 8994 | ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode)); |
cdd1fedf DF |
8995 | } |
8996 | if (ret) { | |
66642832 | 8997 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
8998 | goto out_fail; |
8999 | } | |
9000 | ||
db0a669f | 9001 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
cdd1fedf DF |
9002 | new_dentry->d_name.name, |
9003 | new_dentry->d_name.len, 0, old_idx); | |
9004 | if (ret) { | |
66642832 | 9005 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9006 | goto out_fail; |
9007 | } | |
9008 | ||
db0a669f | 9009 | ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode), |
cdd1fedf DF |
9010 | old_dentry->d_name.name, |
9011 | old_dentry->d_name.len, 0, new_idx); | |
9012 | if (ret) { | |
66642832 | 9013 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9014 | goto out_fail; |
9015 | } | |
9016 | ||
9017 | if (old_inode->i_nlink == 1) | |
9018 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9019 | if (new_inode->i_nlink == 1) | |
9020 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9021 | ||
86e8aa0e | 9022 | if (root_log_pinned) { |
75b463d2 FM |
9023 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
9024 | new_dentry->d_parent); | |
cdd1fedf | 9025 | btrfs_end_log_trans(root); |
86e8aa0e | 9026 | root_log_pinned = false; |
cdd1fedf | 9027 | } |
86e8aa0e | 9028 | if (dest_log_pinned) { |
75b463d2 FM |
9029 | btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir), |
9030 | old_dentry->d_parent); | |
cdd1fedf | 9031 | btrfs_end_log_trans(dest); |
86e8aa0e | 9032 | dest_log_pinned = false; |
cdd1fedf DF |
9033 | } |
9034 | out_fail: | |
86e8aa0e FM |
9035 | /* |
9036 | * If we have pinned a log and an error happened, we unpin tasks | |
9037 | * trying to sync the log and force them to fallback to a transaction | |
9038 | * commit if the log currently contains any of the inodes involved in | |
9039 | * this rename operation (to ensure we do not persist a log with an | |
9040 | * inconsistent state for any of these inodes or leading to any | |
9041 | * inconsistencies when replayed). If the transaction was aborted, the | |
9042 | * abortion reason is propagated to userspace when attempting to commit | |
9043 | * the transaction. If the log does not contain any of these inodes, we | |
9044 | * allow the tasks to sync it. | |
9045 | */ | |
9046 | if (ret && (root_log_pinned || dest_log_pinned)) { | |
0f8939b8 NB |
9047 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9048 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9049 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
86e8aa0e | 9050 | (new_inode && |
0f8939b8 | 9051 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
90787766 | 9052 | btrfs_set_log_full_commit(trans); |
86e8aa0e FM |
9053 | |
9054 | if (root_log_pinned) { | |
9055 | btrfs_end_log_trans(root); | |
9056 | root_log_pinned = false; | |
9057 | } | |
9058 | if (dest_log_pinned) { | |
9059 | btrfs_end_log_trans(dest); | |
9060 | dest_log_pinned = false; | |
9061 | } | |
9062 | } | |
75b463d2 FM |
9063 | ret2 = btrfs_end_transaction(trans); |
9064 | ret = ret ? ret : ret2; | |
cdd1fedf | 9065 | out_notrans: |
943eb3bf JB |
9066 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID || |
9067 | old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9068 | up_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9069 | |
9070 | return ret; | |
9071 | } | |
9072 | ||
9073 | static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans, | |
9074 | struct btrfs_root *root, | |
9075 | struct inode *dir, | |
9076 | struct dentry *dentry) | |
9077 | { | |
9078 | int ret; | |
9079 | struct inode *inode; | |
9080 | u64 objectid; | |
9081 | u64 index; | |
9082 | ||
abadc1fc | 9083 | ret = btrfs_find_free_objectid(root, &objectid); |
cdd1fedf DF |
9084 | if (ret) |
9085 | return ret; | |
9086 | ||
9087 | inode = btrfs_new_inode(trans, root, dir, | |
9088 | dentry->d_name.name, | |
9089 | dentry->d_name.len, | |
4a0cc7ca | 9090 | btrfs_ino(BTRFS_I(dir)), |
cdd1fedf DF |
9091 | objectid, |
9092 | S_IFCHR | WHITEOUT_MODE, | |
9093 | &index); | |
9094 | ||
9095 | if (IS_ERR(inode)) { | |
9096 | ret = PTR_ERR(inode); | |
9097 | return ret; | |
9098 | } | |
9099 | ||
9100 | inode->i_op = &btrfs_special_inode_operations; | |
9101 | init_special_inode(inode, inode->i_mode, | |
9102 | WHITEOUT_DEV); | |
9103 | ||
9104 | ret = btrfs_init_inode_security(trans, inode, dir, | |
9105 | &dentry->d_name); | |
9106 | if (ret) | |
c9901618 | 9107 | goto out; |
cdd1fedf | 9108 | |
cef415af NB |
9109 | ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
9110 | BTRFS_I(inode), 0, index); | |
cdd1fedf | 9111 | if (ret) |
c9901618 | 9112 | goto out; |
cdd1fedf | 9113 | |
9a56fcd1 | 9114 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
c9901618 | 9115 | out: |
cdd1fedf | 9116 | unlock_new_inode(inode); |
c9901618 FM |
9117 | if (ret) |
9118 | inode_dec_link_count(inode); | |
cdd1fedf DF |
9119 | iput(inode); |
9120 | ||
c9901618 | 9121 | return ret; |
cdd1fedf DF |
9122 | } |
9123 | ||
d397712b | 9124 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
cdd1fedf DF |
9125 | struct inode *new_dir, struct dentry *new_dentry, |
9126 | unsigned int flags) | |
39279cc3 | 9127 | { |
0b246afa | 9128 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
39279cc3 | 9129 | struct btrfs_trans_handle *trans; |
5062af35 | 9130 | unsigned int trans_num_items; |
39279cc3 | 9131 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9132 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9133 | struct inode *new_inode = d_inode(new_dentry); |
9134 | struct inode *old_inode = d_inode(old_dentry); | |
00e4e6b3 | 9135 | u64 index = 0; |
39279cc3 | 9136 | int ret; |
75b463d2 | 9137 | int ret2; |
4a0cc7ca | 9138 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
3dc9e8f7 | 9139 | bool log_pinned = false; |
39279cc3 | 9140 | |
4a0cc7ca | 9141 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9142 | return -EPERM; |
9143 | ||
4df27c4d | 9144 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9145 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9146 | return -EXDEV; |
9147 | ||
33345d01 | 9148 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 9149 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 9150 | return -ENOTEMPTY; |
5f39d397 | 9151 | |
4df27c4d YZ |
9152 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9153 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9154 | return -ENOTEMPTY; | |
9c52057c CM |
9155 | |
9156 | ||
9157 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9158 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9159 | new_dentry->d_name.name, |
9160 | new_dentry->d_name.len); | |
9161 | ||
9162 | if (ret) { | |
9163 | if (ret == -EEXIST) { | |
9164 | /* we shouldn't get | |
9165 | * eexist without a new_inode */ | |
fae7f21c | 9166 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9167 | return ret; |
9168 | } | |
9169 | } else { | |
9170 | /* maybe -EOVERFLOW */ | |
9171 | return ret; | |
9172 | } | |
9173 | } | |
9174 | ret = 0; | |
9175 | ||
5a3f23d5 | 9176 | /* |
8d875f95 CM |
9177 | * we're using rename to replace one file with another. Start IO on it |
9178 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9179 | */ |
8d875f95 | 9180 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9181 | filemap_flush(old_inode->i_mapping); |
9182 | ||
76dda93c | 9183 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 9184 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9185 | down_read(&fs_info->subvol_sem); |
a22285a6 YZ |
9186 | /* |
9187 | * We want to reserve the absolute worst case amount of items. So if | |
9188 | * both inodes are subvols and we need to unlink them then that would | |
9189 | * require 4 item modifications, but if they are both normal inodes it | |
cdd1fedf | 9190 | * would require 5 item modifications, so we'll assume they are normal |
a22285a6 YZ |
9191 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items |
9192 | * should cover the worst case number of items we'll modify. | |
5062af35 FM |
9193 | * If our rename has the whiteout flag, we need more 5 units for the |
9194 | * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item | |
9195 | * when selinux is enabled). | |
a22285a6 | 9196 | */ |
5062af35 FM |
9197 | trans_num_items = 11; |
9198 | if (flags & RENAME_WHITEOUT) | |
9199 | trans_num_items += 5; | |
9200 | trans = btrfs_start_transaction(root, trans_num_items); | |
b44c59a8 | 9201 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9202 | ret = PTR_ERR(trans); |
9203 | goto out_notrans; | |
9204 | } | |
76dda93c | 9205 | |
4df27c4d YZ |
9206 | if (dest != root) |
9207 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 9208 | |
877574e2 | 9209 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index); |
a5719521 YZ |
9210 | if (ret) |
9211 | goto out_fail; | |
5a3f23d5 | 9212 | |
67de1176 | 9213 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9214 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9215 | /* force full log commit if subvolume involved. */ |
90787766 | 9216 | btrfs_set_log_full_commit(trans); |
4df27c4d | 9217 | } else { |
c4aba954 FM |
9218 | btrfs_pin_log_trans(root); |
9219 | log_pinned = true; | |
a5719521 YZ |
9220 | ret = btrfs_insert_inode_ref(trans, dest, |
9221 | new_dentry->d_name.name, | |
9222 | new_dentry->d_name.len, | |
33345d01 | 9223 | old_ino, |
4a0cc7ca | 9224 | btrfs_ino(BTRFS_I(new_dir)), index); |
a5719521 YZ |
9225 | if (ret) |
9226 | goto out_fail; | |
4df27c4d | 9227 | } |
5a3f23d5 | 9228 | |
0c4d2d95 JB |
9229 | inode_inc_iversion(old_dir); |
9230 | inode_inc_iversion(new_dir); | |
9231 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
9232 | old_dir->i_ctime = old_dir->i_mtime = |
9233 | new_dir->i_ctime = new_dir->i_mtime = | |
c2050a45 | 9234 | old_inode->i_ctime = current_time(old_dir); |
5f39d397 | 9235 | |
12fcfd22 | 9236 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 DS |
9237 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9238 | BTRFS_I(old_inode), 1); | |
12fcfd22 | 9239 | |
33345d01 | 9240 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
045d3967 | 9241 | ret = btrfs_unlink_subvol(trans, old_dir, old_dentry); |
4df27c4d | 9242 | } else { |
4ec5934e NB |
9243 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9244 | BTRFS_I(d_inode(old_dentry)), | |
92986796 AV |
9245 | old_dentry->d_name.name, |
9246 | old_dentry->d_name.len); | |
9247 | if (!ret) | |
9a56fcd1 | 9248 | ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode)); |
4df27c4d | 9249 | } |
79787eaa | 9250 | if (ret) { |
66642832 | 9251 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9252 | goto out_fail; |
9253 | } | |
39279cc3 CM |
9254 | |
9255 | if (new_inode) { | |
0c4d2d95 | 9256 | inode_inc_iversion(new_inode); |
c2050a45 | 9257 | new_inode->i_ctime = current_time(new_inode); |
4a0cc7ca | 9258 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d | 9259 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
045d3967 | 9260 | ret = btrfs_unlink_subvol(trans, new_dir, new_dentry); |
4df27c4d YZ |
9261 | BUG_ON(new_inode->i_nlink == 0); |
9262 | } else { | |
4ec5934e NB |
9263 | ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9264 | BTRFS_I(d_inode(new_dentry)), | |
4df27c4d YZ |
9265 | new_dentry->d_name.name, |
9266 | new_dentry->d_name.len); | |
9267 | } | |
4ef31a45 | 9268 | if (!ret && new_inode->i_nlink == 0) |
73f2e545 NB |
9269 | ret = btrfs_orphan_add(trans, |
9270 | BTRFS_I(d_inode(new_dentry))); | |
79787eaa | 9271 | if (ret) { |
66642832 | 9272 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9273 | goto out_fail; |
9274 | } | |
39279cc3 | 9275 | } |
aec7477b | 9276 | |
db0a669f | 9277 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
4df27c4d | 9278 | new_dentry->d_name.name, |
a5719521 | 9279 | new_dentry->d_name.len, 0, index); |
79787eaa | 9280 | if (ret) { |
66642832 | 9281 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9282 | goto out_fail; |
9283 | } | |
39279cc3 | 9284 | |
67de1176 MX |
9285 | if (old_inode->i_nlink == 1) |
9286 | BTRFS_I(old_inode)->dir_index = index; | |
9287 | ||
3dc9e8f7 | 9288 | if (log_pinned) { |
75b463d2 FM |
9289 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
9290 | new_dentry->d_parent); | |
4df27c4d | 9291 | btrfs_end_log_trans(root); |
3dc9e8f7 | 9292 | log_pinned = false; |
4df27c4d | 9293 | } |
cdd1fedf DF |
9294 | |
9295 | if (flags & RENAME_WHITEOUT) { | |
9296 | ret = btrfs_whiteout_for_rename(trans, root, old_dir, | |
9297 | old_dentry); | |
9298 | ||
9299 | if (ret) { | |
66642832 | 9300 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9301 | goto out_fail; |
9302 | } | |
4df27c4d | 9303 | } |
39279cc3 | 9304 | out_fail: |
3dc9e8f7 FM |
9305 | /* |
9306 | * If we have pinned the log and an error happened, we unpin tasks | |
9307 | * trying to sync the log and force them to fallback to a transaction | |
9308 | * commit if the log currently contains any of the inodes involved in | |
9309 | * this rename operation (to ensure we do not persist a log with an | |
9310 | * inconsistent state for any of these inodes or leading to any | |
9311 | * inconsistencies when replayed). If the transaction was aborted, the | |
9312 | * abortion reason is propagated to userspace when attempting to commit | |
9313 | * the transaction. If the log does not contain any of these inodes, we | |
9314 | * allow the tasks to sync it. | |
9315 | */ | |
9316 | if (ret && log_pinned) { | |
0f8939b8 NB |
9317 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9318 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9319 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
3dc9e8f7 | 9320 | (new_inode && |
0f8939b8 | 9321 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
90787766 | 9322 | btrfs_set_log_full_commit(trans); |
3dc9e8f7 FM |
9323 | |
9324 | btrfs_end_log_trans(root); | |
9325 | log_pinned = false; | |
9326 | } | |
75b463d2 FM |
9327 | ret2 = btrfs_end_transaction(trans); |
9328 | ret = ret ? ret : ret2; | |
b44c59a8 | 9329 | out_notrans: |
33345d01 | 9330 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9331 | up_read(&fs_info->subvol_sem); |
9ed74f2d | 9332 | |
39279cc3 CM |
9333 | return ret; |
9334 | } | |
9335 | ||
80ace85c MS |
9336 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
9337 | struct inode *new_dir, struct dentry *new_dentry, | |
9338 | unsigned int flags) | |
9339 | { | |
cdd1fedf | 9340 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
9341 | return -EINVAL; |
9342 | ||
cdd1fedf DF |
9343 | if (flags & RENAME_EXCHANGE) |
9344 | return btrfs_rename_exchange(old_dir, old_dentry, new_dir, | |
9345 | new_dentry); | |
9346 | ||
9347 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags); | |
80ace85c MS |
9348 | } |
9349 | ||
3a2f8c07 NB |
9350 | struct btrfs_delalloc_work { |
9351 | struct inode *inode; | |
9352 | struct completion completion; | |
9353 | struct list_head list; | |
9354 | struct btrfs_work work; | |
9355 | }; | |
9356 | ||
8ccf6f19 MX |
9357 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
9358 | { | |
9359 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 9360 | struct inode *inode; |
8ccf6f19 MX |
9361 | |
9362 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
9363 | work); | |
9f23e289 | 9364 | inode = delalloc_work->inode; |
30424601 DS |
9365 | filemap_flush(inode->i_mapping); |
9366 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9367 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 9368 | filemap_flush(inode->i_mapping); |
8ccf6f19 | 9369 | |
076da91c | 9370 | iput(inode); |
8ccf6f19 MX |
9371 | complete(&delalloc_work->completion); |
9372 | } | |
9373 | ||
3a2f8c07 | 9374 | static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode) |
8ccf6f19 MX |
9375 | { |
9376 | struct btrfs_delalloc_work *work; | |
9377 | ||
100d5702 | 9378 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
9379 | if (!work) |
9380 | return NULL; | |
9381 | ||
9382 | init_completion(&work->completion); | |
9383 | INIT_LIST_HEAD(&work->list); | |
9384 | work->inode = inode; | |
a0cac0ec | 9385 | btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL); |
8ccf6f19 MX |
9386 | |
9387 | return work; | |
9388 | } | |
9389 | ||
d352ac68 CM |
9390 | /* |
9391 | * some fairly slow code that needs optimization. This walks the list | |
9392 | * of all the inodes with pending delalloc and forces them to disk. | |
9393 | */ | |
e076ab2a JB |
9394 | static int start_delalloc_inodes(struct btrfs_root *root, |
9395 | struct writeback_control *wbc, bool snapshot, | |
3d45f221 | 9396 | bool in_reclaim_context) |
ea8c2819 | 9397 | { |
ea8c2819 | 9398 | struct btrfs_inode *binode; |
5b21f2ed | 9399 | struct inode *inode; |
8ccf6f19 MX |
9400 | struct btrfs_delalloc_work *work, *next; |
9401 | struct list_head works; | |
1eafa6c7 | 9402 | struct list_head splice; |
8ccf6f19 | 9403 | int ret = 0; |
e076ab2a | 9404 | bool full_flush = wbc->nr_to_write == LONG_MAX; |
ea8c2819 | 9405 | |
8ccf6f19 | 9406 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 9407 | INIT_LIST_HEAD(&splice); |
63607cc8 | 9408 | |
573bfb72 | 9409 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
9410 | spin_lock(&root->delalloc_lock); |
9411 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
9412 | while (!list_empty(&splice)) { |
9413 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 9414 | delalloc_inodes); |
1eafa6c7 | 9415 | |
eb73c1b7 MX |
9416 | list_move_tail(&binode->delalloc_inodes, |
9417 | &root->delalloc_inodes); | |
3d45f221 FM |
9418 | |
9419 | if (in_reclaim_context && | |
9420 | test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags)) | |
9421 | continue; | |
9422 | ||
5b21f2ed | 9423 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 9424 | if (!inode) { |
eb73c1b7 | 9425 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 9426 | continue; |
df0af1a5 | 9427 | } |
eb73c1b7 | 9428 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 9429 | |
3cd24c69 EL |
9430 | if (snapshot) |
9431 | set_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
9432 | &binode->runtime_flags); | |
e076ab2a JB |
9433 | if (full_flush) { |
9434 | work = btrfs_alloc_delalloc_work(inode); | |
9435 | if (!work) { | |
9436 | iput(inode); | |
9437 | ret = -ENOMEM; | |
9438 | goto out; | |
9439 | } | |
9440 | list_add_tail(&work->list, &works); | |
9441 | btrfs_queue_work(root->fs_info->flush_workers, | |
9442 | &work->work); | |
9443 | } else { | |
9444 | ret = sync_inode(inode, wbc); | |
9445 | if (!ret && | |
9446 | test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9447 | &BTRFS_I(inode)->runtime_flags)) | |
9448 | ret = sync_inode(inode, wbc); | |
9449 | btrfs_add_delayed_iput(inode); | |
9450 | if (ret || wbc->nr_to_write <= 0) | |
b4912139 JB |
9451 | goto out; |
9452 | } | |
5b21f2ed | 9453 | cond_resched(); |
eb73c1b7 | 9454 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 9455 | } |
eb73c1b7 | 9456 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 9457 | |
a1ecaabb | 9458 | out: |
eb73c1b7 MX |
9459 | list_for_each_entry_safe(work, next, &works, list) { |
9460 | list_del_init(&work->list); | |
40012f96 NB |
9461 | wait_for_completion(&work->completion); |
9462 | kfree(work); | |
eb73c1b7 MX |
9463 | } |
9464 | ||
81f1d390 | 9465 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
9466 | spin_lock(&root->delalloc_lock); |
9467 | list_splice_tail(&splice, &root->delalloc_inodes); | |
9468 | spin_unlock(&root->delalloc_lock); | |
9469 | } | |
573bfb72 | 9470 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
9471 | return ret; |
9472 | } | |
1eafa6c7 | 9473 | |
3cd24c69 | 9474 | int btrfs_start_delalloc_snapshot(struct btrfs_root *root) |
eb73c1b7 | 9475 | { |
e076ab2a JB |
9476 | struct writeback_control wbc = { |
9477 | .nr_to_write = LONG_MAX, | |
9478 | .sync_mode = WB_SYNC_NONE, | |
9479 | .range_start = 0, | |
9480 | .range_end = LLONG_MAX, | |
9481 | }; | |
0b246afa | 9482 | struct btrfs_fs_info *fs_info = root->fs_info; |
1eafa6c7 | 9483 | |
0b246afa | 9484 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
9485 | return -EROFS; |
9486 | ||
e076ab2a | 9487 | return start_delalloc_inodes(root, &wbc, true, false); |
eb73c1b7 MX |
9488 | } |
9489 | ||
3d45f221 FM |
9490 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, u64 nr, |
9491 | bool in_reclaim_context) | |
eb73c1b7 | 9492 | { |
e076ab2a JB |
9493 | struct writeback_control wbc = { |
9494 | .nr_to_write = (nr == U64_MAX) ? LONG_MAX : (unsigned long)nr, | |
9495 | .sync_mode = WB_SYNC_NONE, | |
9496 | .range_start = 0, | |
9497 | .range_end = LLONG_MAX, | |
9498 | }; | |
eb73c1b7 MX |
9499 | struct btrfs_root *root; |
9500 | struct list_head splice; | |
9501 | int ret; | |
9502 | ||
2c21b4d7 | 9503 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
9504 | return -EROFS; |
9505 | ||
9506 | INIT_LIST_HEAD(&splice); | |
9507 | ||
573bfb72 | 9508 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
9509 | spin_lock(&fs_info->delalloc_root_lock); |
9510 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 9511 | while (!list_empty(&splice) && nr) { |
e076ab2a JB |
9512 | /* |
9513 | * Reset nr_to_write here so we know that we're doing a full | |
9514 | * flush. | |
9515 | */ | |
9516 | if (nr == U64_MAX) | |
9517 | wbc.nr_to_write = LONG_MAX; | |
9518 | ||
eb73c1b7 MX |
9519 | root = list_first_entry(&splice, struct btrfs_root, |
9520 | delalloc_root); | |
00246528 | 9521 | root = btrfs_grab_root(root); |
eb73c1b7 MX |
9522 | BUG_ON(!root); |
9523 | list_move_tail(&root->delalloc_root, | |
9524 | &fs_info->delalloc_roots); | |
9525 | spin_unlock(&fs_info->delalloc_root_lock); | |
9526 | ||
e076ab2a | 9527 | ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context); |
00246528 | 9528 | btrfs_put_root(root); |
e076ab2a | 9529 | if (ret < 0 || wbc.nr_to_write <= 0) |
eb73c1b7 | 9530 | goto out; |
eb73c1b7 | 9531 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 9532 | } |
eb73c1b7 | 9533 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 9534 | |
6c255e67 | 9535 | ret = 0; |
eb73c1b7 | 9536 | out: |
81f1d390 | 9537 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
9538 | spin_lock(&fs_info->delalloc_root_lock); |
9539 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
9540 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 9541 | } |
573bfb72 | 9542 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 9543 | return ret; |
ea8c2819 CM |
9544 | } |
9545 | ||
39279cc3 CM |
9546 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
9547 | const char *symname) | |
9548 | { | |
0b246afa | 9549 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
9550 | struct btrfs_trans_handle *trans; |
9551 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9552 | struct btrfs_path *path; | |
9553 | struct btrfs_key key; | |
1832a6d5 | 9554 | struct inode *inode = NULL; |
39279cc3 | 9555 | int err; |
39279cc3 | 9556 | u64 objectid; |
67871254 | 9557 | u64 index = 0; |
39279cc3 CM |
9558 | int name_len; |
9559 | int datasize; | |
5f39d397 | 9560 | unsigned long ptr; |
39279cc3 | 9561 | struct btrfs_file_extent_item *ei; |
5f39d397 | 9562 | struct extent_buffer *leaf; |
39279cc3 | 9563 | |
f06becc4 | 9564 | name_len = strlen(symname); |
0b246afa | 9565 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 9566 | return -ENAMETOOLONG; |
1832a6d5 | 9567 | |
9ed74f2d JB |
9568 | /* |
9569 | * 2 items for inode item and ref | |
9570 | * 2 items for dir items | |
9269d12b FM |
9571 | * 1 item for updating parent inode item |
9572 | * 1 item for the inline extent item | |
9ed74f2d JB |
9573 | * 1 item for xattr if selinux is on |
9574 | */ | |
9269d12b | 9575 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
9576 | if (IS_ERR(trans)) |
9577 | return PTR_ERR(trans); | |
1832a6d5 | 9578 | |
abadc1fc | 9579 | err = btrfs_find_free_objectid(root, &objectid); |
581bb050 LZ |
9580 | if (err) |
9581 | goto out_unlock; | |
9582 | ||
aec7477b | 9583 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
9584 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), |
9585 | objectid, S_IFLNK|S_IRWXUGO, &index); | |
7cf96da3 TI |
9586 | if (IS_ERR(inode)) { |
9587 | err = PTR_ERR(inode); | |
32955c54 | 9588 | inode = NULL; |
39279cc3 | 9589 | goto out_unlock; |
7cf96da3 | 9590 | } |
39279cc3 | 9591 | |
ad19db71 CS |
9592 | /* |
9593 | * If the active LSM wants to access the inode during | |
9594 | * d_instantiate it needs these. Smack checks to see | |
9595 | * if the filesystem supports xattrs by looking at the | |
9596 | * ops vector. | |
9597 | */ | |
9598 | inode->i_fop = &btrfs_file_operations; | |
9599 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 9600 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
9601 | |
9602 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
9603 | if (err) | |
32955c54 | 9604 | goto out_unlock; |
ad19db71 | 9605 | |
39279cc3 | 9606 | path = btrfs_alloc_path(); |
d8926bb3 MF |
9607 | if (!path) { |
9608 | err = -ENOMEM; | |
32955c54 | 9609 | goto out_unlock; |
d8926bb3 | 9610 | } |
4a0cc7ca | 9611 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 9612 | key.offset = 0; |
962a298f | 9613 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
9614 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
9615 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
9616 | datasize); | |
54aa1f4d | 9617 | if (err) { |
b0839166 | 9618 | btrfs_free_path(path); |
32955c54 | 9619 | goto out_unlock; |
54aa1f4d | 9620 | } |
5f39d397 CM |
9621 | leaf = path->nodes[0]; |
9622 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
9623 | struct btrfs_file_extent_item); | |
9624 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
9625 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 9626 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
9627 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
9628 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
9629 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
9630 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
9631 | ||
39279cc3 | 9632 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
9633 | write_extent_buffer(leaf, symname, ptr, name_len); |
9634 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 9635 | btrfs_free_path(path); |
5f39d397 | 9636 | |
39279cc3 | 9637 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 9638 | inode_nohighmem(inode); |
d899e052 | 9639 | inode_set_bytes(inode, name_len); |
6ef06d27 | 9640 | btrfs_i_size_write(BTRFS_I(inode), name_len); |
9a56fcd1 | 9641 | err = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
d50866d0 FM |
9642 | /* |
9643 | * Last step, add directory indexes for our symlink inode. This is the | |
9644 | * last step to avoid extra cleanup of these indexes if an error happens | |
9645 | * elsewhere above. | |
9646 | */ | |
9647 | if (!err) | |
cef415af NB |
9648 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
9649 | BTRFS_I(inode), 0, index); | |
32955c54 AV |
9650 | if (err) |
9651 | goto out_unlock; | |
b0d5d10f | 9652 | |
1e2e547a | 9653 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
9654 | |
9655 | out_unlock: | |
3a45bb20 | 9656 | btrfs_end_transaction(trans); |
32955c54 | 9657 | if (err && inode) { |
39279cc3 | 9658 | inode_dec_link_count(inode); |
32955c54 | 9659 | discard_new_inode(inode); |
39279cc3 | 9660 | } |
2ff7e61e | 9661 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
9662 | return err; |
9663 | } | |
16432985 | 9664 | |
8fccebfa FM |
9665 | static struct btrfs_trans_handle *insert_prealloc_file_extent( |
9666 | struct btrfs_trans_handle *trans_in, | |
90dffd0c NB |
9667 | struct btrfs_inode *inode, |
9668 | struct btrfs_key *ins, | |
203f44c5 QW |
9669 | u64 file_offset) |
9670 | { | |
9671 | struct btrfs_file_extent_item stack_fi; | |
bf385648 | 9672 | struct btrfs_replace_extent_info extent_info; |
8fccebfa FM |
9673 | struct btrfs_trans_handle *trans = trans_in; |
9674 | struct btrfs_path *path; | |
203f44c5 QW |
9675 | u64 start = ins->objectid; |
9676 | u64 len = ins->offset; | |
9729f10a | 9677 | int ret; |
203f44c5 QW |
9678 | |
9679 | memset(&stack_fi, 0, sizeof(stack_fi)); | |
9680 | ||
9681 | btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC); | |
9682 | btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start); | |
9683 | btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len); | |
9684 | btrfs_set_stack_file_extent_num_bytes(&stack_fi, len); | |
9685 | btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len); | |
9686 | btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE); | |
9687 | /* Encryption and other encoding is reserved and all 0 */ | |
9688 | ||
90dffd0c | 9689 | ret = btrfs_qgroup_release_data(inode, file_offset, len); |
9729f10a | 9690 | if (ret < 0) |
8fccebfa FM |
9691 | return ERR_PTR(ret); |
9692 | ||
9693 | if (trans) { | |
90dffd0c | 9694 | ret = insert_reserved_file_extent(trans, inode, |
2766ff61 FM |
9695 | file_offset, &stack_fi, |
9696 | true, ret); | |
8fccebfa FM |
9697 | if (ret) |
9698 | return ERR_PTR(ret); | |
9699 | return trans; | |
9700 | } | |
9701 | ||
9702 | extent_info.disk_offset = start; | |
9703 | extent_info.disk_len = len; | |
9704 | extent_info.data_offset = 0; | |
9705 | extent_info.data_len = len; | |
9706 | extent_info.file_offset = file_offset; | |
9707 | extent_info.extent_buf = (char *)&stack_fi; | |
8fccebfa FM |
9708 | extent_info.is_new_extent = true; |
9709 | extent_info.qgroup_reserved = ret; | |
9710 | extent_info.insertions = 0; | |
9711 | ||
9712 | path = btrfs_alloc_path(); | |
9713 | if (!path) | |
9714 | return ERR_PTR(-ENOMEM); | |
9715 | ||
90dffd0c | 9716 | ret = btrfs_replace_file_extents(&inode->vfs_inode, path, file_offset, |
8fccebfa FM |
9717 | file_offset + len - 1, &extent_info, |
9718 | &trans); | |
9719 | btrfs_free_path(path); | |
9720 | if (ret) | |
9721 | return ERR_PTR(ret); | |
9722 | ||
9723 | return trans; | |
203f44c5 | 9724 | } |
8fccebfa | 9725 | |
0af3d00b JB |
9726 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
9727 | u64 start, u64 num_bytes, u64 min_size, | |
9728 | loff_t actual_len, u64 *alloc_hint, | |
9729 | struct btrfs_trans_handle *trans) | |
d899e052 | 9730 | { |
0b246afa | 9731 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 JB |
9732 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
9733 | struct extent_map *em; | |
d899e052 YZ |
9734 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9735 | struct btrfs_key ins; | |
d899e052 | 9736 | u64 cur_offset = start; |
b778cf96 | 9737 | u64 clear_offset = start; |
55a61d1d | 9738 | u64 i_size; |
154ea289 | 9739 | u64 cur_bytes; |
0b670dc4 | 9740 | u64 last_alloc = (u64)-1; |
d899e052 | 9741 | int ret = 0; |
0af3d00b | 9742 | bool own_trans = true; |
18513091 | 9743 | u64 end = start + num_bytes - 1; |
d899e052 | 9744 | |
0af3d00b JB |
9745 | if (trans) |
9746 | own_trans = false; | |
d899e052 | 9747 | while (num_bytes > 0) { |
ee22184b | 9748 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 9749 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
9750 | /* |
9751 | * If we are severely fragmented we could end up with really | |
9752 | * small allocations, so if the allocator is returning small | |
9753 | * chunks lets make its job easier by only searching for those | |
9754 | * sized chunks. | |
9755 | */ | |
9756 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
9757 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
9758 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
8fccebfa | 9759 | if (ret) |
a22285a6 | 9760 | break; |
b778cf96 JB |
9761 | |
9762 | /* | |
9763 | * We've reserved this space, and thus converted it from | |
9764 | * ->bytes_may_use to ->bytes_reserved. Any error that happens | |
9765 | * from here on out we will only need to clear our reservation | |
9766 | * for the remaining unreserved area, so advance our | |
9767 | * clear_offset by our extent size. | |
9768 | */ | |
9769 | clear_offset += ins.offset; | |
5a303d5d | 9770 | |
0b670dc4 | 9771 | last_alloc = ins.offset; |
90dffd0c NB |
9772 | trans = insert_prealloc_file_extent(trans, BTRFS_I(inode), |
9773 | &ins, cur_offset); | |
1afc708d FM |
9774 | /* |
9775 | * Now that we inserted the prealloc extent we can finally | |
9776 | * decrement the number of reservations in the block group. | |
9777 | * If we did it before, we could race with relocation and have | |
9778 | * relocation miss the reserved extent, making it fail later. | |
9779 | */ | |
9780 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); | |
8fccebfa FM |
9781 | if (IS_ERR(trans)) { |
9782 | ret = PTR_ERR(trans); | |
2ff7e61e | 9783 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 9784 | ins.offset, 0); |
79787eaa JM |
9785 | break; |
9786 | } | |
31193213 | 9787 | |
dcdbc059 | 9788 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
a1ed835e | 9789 | cur_offset + ins.offset -1, 0); |
5a303d5d | 9790 | |
5dc562c5 JB |
9791 | em = alloc_extent_map(); |
9792 | if (!em) { | |
9793 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
9794 | &BTRFS_I(inode)->runtime_flags); | |
9795 | goto next; | |
9796 | } | |
9797 | ||
9798 | em->start = cur_offset; | |
9799 | em->orig_start = cur_offset; | |
9800 | em->len = ins.offset; | |
9801 | em->block_start = ins.objectid; | |
9802 | em->block_len = ins.offset; | |
b4939680 | 9803 | em->orig_block_len = ins.offset; |
cc95bef6 | 9804 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
9805 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
9806 | em->generation = trans->transid; | |
9807 | ||
9808 | while (1) { | |
9809 | write_lock(&em_tree->lock); | |
09a2a8f9 | 9810 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
9811 | write_unlock(&em_tree->lock); |
9812 | if (ret != -EEXIST) | |
9813 | break; | |
dcdbc059 | 9814 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
9815 | cur_offset + ins.offset - 1, |
9816 | 0); | |
9817 | } | |
9818 | free_extent_map(em); | |
9819 | next: | |
d899e052 YZ |
9820 | num_bytes -= ins.offset; |
9821 | cur_offset += ins.offset; | |
efa56464 | 9822 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 9823 | |
0c4d2d95 | 9824 | inode_inc_iversion(inode); |
c2050a45 | 9825 | inode->i_ctime = current_time(inode); |
6cbff00f | 9826 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 9827 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
9828 | (actual_len > inode->i_size) && |
9829 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 9830 | if (cur_offset > actual_len) |
55a61d1d | 9831 | i_size = actual_len; |
d1ea6a61 | 9832 | else |
55a61d1d JB |
9833 | i_size = cur_offset; |
9834 | i_size_write(inode, i_size); | |
76aea537 | 9835 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
5a303d5d YZ |
9836 | } |
9837 | ||
9a56fcd1 | 9838 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
79787eaa JM |
9839 | |
9840 | if (ret) { | |
66642832 | 9841 | btrfs_abort_transaction(trans, ret); |
79787eaa | 9842 | if (own_trans) |
3a45bb20 | 9843 | btrfs_end_transaction(trans); |
79787eaa JM |
9844 | break; |
9845 | } | |
d899e052 | 9846 | |
8fccebfa | 9847 | if (own_trans) { |
3a45bb20 | 9848 | btrfs_end_transaction(trans); |
8fccebfa FM |
9849 | trans = NULL; |
9850 | } | |
5a303d5d | 9851 | } |
b778cf96 | 9852 | if (clear_offset < end) |
25ce28ca | 9853 | btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset, |
b778cf96 | 9854 | end - clear_offset + 1); |
d899e052 YZ |
9855 | return ret; |
9856 | } | |
9857 | ||
0af3d00b JB |
9858 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
9859 | u64 start, u64 num_bytes, u64 min_size, | |
9860 | loff_t actual_len, u64 *alloc_hint) | |
9861 | { | |
9862 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9863 | min_size, actual_len, alloc_hint, | |
9864 | NULL); | |
9865 | } | |
9866 | ||
9867 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
9868 | struct btrfs_trans_handle *trans, int mode, | |
9869 | u64 start, u64 num_bytes, u64 min_size, | |
9870 | loff_t actual_len, u64 *alloc_hint) | |
9871 | { | |
9872 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9873 | min_size, actual_len, alloc_hint, trans); | |
9874 | } | |
9875 | ||
e6dcd2dc CM |
9876 | static int btrfs_set_page_dirty(struct page *page) |
9877 | { | |
e6dcd2dc CM |
9878 | return __set_page_dirty_nobuffers(page); |
9879 | } | |
9880 | ||
10556cb2 | 9881 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 9882 | { |
b83cc969 | 9883 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 9884 | umode_t mode = inode->i_mode; |
b83cc969 | 9885 | |
cb6db4e5 JM |
9886 | if (mask & MAY_WRITE && |
9887 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
9888 | if (btrfs_root_readonly(root)) | |
9889 | return -EROFS; | |
9890 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
9891 | return -EACCES; | |
9892 | } | |
2830ba7f | 9893 | return generic_permission(inode, mask); |
fdebe2bd | 9894 | } |
39279cc3 | 9895 | |
ef3b9af5 FM |
9896 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
9897 | { | |
2ff7e61e | 9898 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
9899 | struct btrfs_trans_handle *trans; |
9900 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9901 | struct inode *inode = NULL; | |
9902 | u64 objectid; | |
9903 | u64 index; | |
9904 | int ret = 0; | |
9905 | ||
9906 | /* | |
9907 | * 5 units required for adding orphan entry | |
9908 | */ | |
9909 | trans = btrfs_start_transaction(root, 5); | |
9910 | if (IS_ERR(trans)) | |
9911 | return PTR_ERR(trans); | |
9912 | ||
abadc1fc | 9913 | ret = btrfs_find_free_objectid(root, &objectid); |
ef3b9af5 FM |
9914 | if (ret) |
9915 | goto out; | |
9916 | ||
9917 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
f85b7379 | 9918 | btrfs_ino(BTRFS_I(dir)), objectid, mode, &index); |
ef3b9af5 FM |
9919 | if (IS_ERR(inode)) { |
9920 | ret = PTR_ERR(inode); | |
9921 | inode = NULL; | |
9922 | goto out; | |
9923 | } | |
9924 | ||
ef3b9af5 FM |
9925 | inode->i_fop = &btrfs_file_operations; |
9926 | inode->i_op = &btrfs_file_inode_operations; | |
9927 | ||
9928 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 | 9929 | |
b0d5d10f CM |
9930 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
9931 | if (ret) | |
32955c54 | 9932 | goto out; |
b0d5d10f | 9933 | |
9a56fcd1 | 9934 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
b0d5d10f | 9935 | if (ret) |
32955c54 | 9936 | goto out; |
73f2e545 | 9937 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
ef3b9af5 | 9938 | if (ret) |
32955c54 | 9939 | goto out; |
ef3b9af5 | 9940 | |
5762b5c9 FM |
9941 | /* |
9942 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
9943 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
9944 | * through: | |
9945 | * | |
9946 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
9947 | */ | |
9948 | set_nlink(inode, 1); | |
ef3b9af5 | 9949 | d_tmpfile(dentry, inode); |
32955c54 | 9950 | unlock_new_inode(inode); |
ef3b9af5 | 9951 | mark_inode_dirty(inode); |
ef3b9af5 | 9952 | out: |
3a45bb20 | 9953 | btrfs_end_transaction(trans); |
32955c54 AV |
9954 | if (ret && inode) |
9955 | discard_new_inode(inode); | |
2ff7e61e | 9956 | btrfs_btree_balance_dirty(fs_info); |
ef3b9af5 FM |
9957 | return ret; |
9958 | } | |
9959 | ||
5cdc84bf | 9960 | void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
c6100a4b | 9961 | { |
5cdc84bf | 9962 | struct inode *inode = tree->private_data; |
c6100a4b JB |
9963 | unsigned long index = start >> PAGE_SHIFT; |
9964 | unsigned long end_index = end >> PAGE_SHIFT; | |
9965 | struct page *page; | |
9966 | ||
9967 | while (index <= end_index) { | |
9968 | page = find_get_page(inode->i_mapping, index); | |
9969 | ASSERT(page); /* Pages should be in the extent_io_tree */ | |
9970 | set_page_writeback(page); | |
9971 | put_page(page); | |
9972 | index++; | |
9973 | } | |
9974 | } | |
9975 | ||
ed46ff3d OS |
9976 | #ifdef CONFIG_SWAP |
9977 | /* | |
9978 | * Add an entry indicating a block group or device which is pinned by a | |
9979 | * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a | |
9980 | * negative errno on failure. | |
9981 | */ | |
9982 | static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr, | |
9983 | bool is_block_group) | |
9984 | { | |
9985 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
9986 | struct btrfs_swapfile_pin *sp, *entry; | |
9987 | struct rb_node **p; | |
9988 | struct rb_node *parent = NULL; | |
9989 | ||
9990 | sp = kmalloc(sizeof(*sp), GFP_NOFS); | |
9991 | if (!sp) | |
9992 | return -ENOMEM; | |
9993 | sp->ptr = ptr; | |
9994 | sp->inode = inode; | |
9995 | sp->is_block_group = is_block_group; | |
9996 | ||
9997 | spin_lock(&fs_info->swapfile_pins_lock); | |
9998 | p = &fs_info->swapfile_pins.rb_node; | |
9999 | while (*p) { | |
10000 | parent = *p; | |
10001 | entry = rb_entry(parent, struct btrfs_swapfile_pin, node); | |
10002 | if (sp->ptr < entry->ptr || | |
10003 | (sp->ptr == entry->ptr && sp->inode < entry->inode)) { | |
10004 | p = &(*p)->rb_left; | |
10005 | } else if (sp->ptr > entry->ptr || | |
10006 | (sp->ptr == entry->ptr && sp->inode > entry->inode)) { | |
10007 | p = &(*p)->rb_right; | |
10008 | } else { | |
10009 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10010 | kfree(sp); | |
10011 | return 1; | |
10012 | } | |
10013 | } | |
10014 | rb_link_node(&sp->node, parent, p); | |
10015 | rb_insert_color(&sp->node, &fs_info->swapfile_pins); | |
10016 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10017 | return 0; | |
10018 | } | |
10019 | ||
10020 | /* Free all of the entries pinned by this swapfile. */ | |
10021 | static void btrfs_free_swapfile_pins(struct inode *inode) | |
10022 | { | |
10023 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10024 | struct btrfs_swapfile_pin *sp; | |
10025 | struct rb_node *node, *next; | |
10026 | ||
10027 | spin_lock(&fs_info->swapfile_pins_lock); | |
10028 | node = rb_first(&fs_info->swapfile_pins); | |
10029 | while (node) { | |
10030 | next = rb_next(node); | |
10031 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
10032 | if (sp->inode == inode) { | |
10033 | rb_erase(&sp->node, &fs_info->swapfile_pins); | |
10034 | if (sp->is_block_group) | |
10035 | btrfs_put_block_group(sp->ptr); | |
10036 | kfree(sp); | |
10037 | } | |
10038 | node = next; | |
10039 | } | |
10040 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10041 | } | |
10042 | ||
10043 | struct btrfs_swap_info { | |
10044 | u64 start; | |
10045 | u64 block_start; | |
10046 | u64 block_len; | |
10047 | u64 lowest_ppage; | |
10048 | u64 highest_ppage; | |
10049 | unsigned long nr_pages; | |
10050 | int nr_extents; | |
10051 | }; | |
10052 | ||
10053 | static int btrfs_add_swap_extent(struct swap_info_struct *sis, | |
10054 | struct btrfs_swap_info *bsi) | |
10055 | { | |
10056 | unsigned long nr_pages; | |
10057 | u64 first_ppage, first_ppage_reported, next_ppage; | |
10058 | int ret; | |
10059 | ||
10060 | first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT; | |
10061 | next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len, | |
10062 | PAGE_SIZE) >> PAGE_SHIFT; | |
10063 | ||
10064 | if (first_ppage >= next_ppage) | |
10065 | return 0; | |
10066 | nr_pages = next_ppage - first_ppage; | |
10067 | ||
10068 | first_ppage_reported = first_ppage; | |
10069 | if (bsi->start == 0) | |
10070 | first_ppage_reported++; | |
10071 | if (bsi->lowest_ppage > first_ppage_reported) | |
10072 | bsi->lowest_ppage = first_ppage_reported; | |
10073 | if (bsi->highest_ppage < (next_ppage - 1)) | |
10074 | bsi->highest_ppage = next_ppage - 1; | |
10075 | ||
10076 | ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage); | |
10077 | if (ret < 0) | |
10078 | return ret; | |
10079 | bsi->nr_extents += ret; | |
10080 | bsi->nr_pages += nr_pages; | |
10081 | return 0; | |
10082 | } | |
10083 | ||
10084 | static void btrfs_swap_deactivate(struct file *file) | |
10085 | { | |
10086 | struct inode *inode = file_inode(file); | |
10087 | ||
10088 | btrfs_free_swapfile_pins(inode); | |
10089 | atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles); | |
10090 | } | |
10091 | ||
10092 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10093 | sector_t *span) | |
10094 | { | |
10095 | struct inode *inode = file_inode(file); | |
10096 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10097 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
10098 | struct extent_state *cached_state = NULL; | |
10099 | struct extent_map *em = NULL; | |
10100 | struct btrfs_device *device = NULL; | |
10101 | struct btrfs_swap_info bsi = { | |
10102 | .lowest_ppage = (sector_t)-1ULL, | |
10103 | }; | |
10104 | int ret = 0; | |
10105 | u64 isize; | |
10106 | u64 start; | |
10107 | ||
10108 | /* | |
10109 | * If the swap file was just created, make sure delalloc is done. If the | |
10110 | * file changes again after this, the user is doing something stupid and | |
10111 | * we don't really care. | |
10112 | */ | |
10113 | ret = btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
10114 | if (ret) | |
10115 | return ret; | |
10116 | ||
10117 | /* | |
10118 | * The inode is locked, so these flags won't change after we check them. | |
10119 | */ | |
10120 | if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) { | |
10121 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10122 | return -EINVAL; | |
10123 | } | |
10124 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) { | |
10125 | btrfs_warn(fs_info, "swapfile must not be copy-on-write"); | |
10126 | return -EINVAL; | |
10127 | } | |
10128 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
10129 | btrfs_warn(fs_info, "swapfile must not be checksummed"); | |
10130 | return -EINVAL; | |
10131 | } | |
10132 | ||
10133 | /* | |
10134 | * Balance or device remove/replace/resize can move stuff around from | |
c3e1f96c GR |
10135 | * under us. The exclop protection makes sure they aren't running/won't |
10136 | * run concurrently while we are mapping the swap extents, and | |
10137 | * fs_info->swapfile_pins prevents them from running while the swap | |
10138 | * file is active and moving the extents. Note that this also prevents | |
10139 | * a concurrent device add which isn't actually necessary, but it's not | |
ed46ff3d OS |
10140 | * really worth the trouble to allow it. |
10141 | */ | |
c3e1f96c | 10142 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) { |
ed46ff3d OS |
10143 | btrfs_warn(fs_info, |
10144 | "cannot activate swapfile while exclusive operation is running"); | |
10145 | return -EBUSY; | |
10146 | } | |
10147 | /* | |
10148 | * Snapshots can create extents which require COW even if NODATACOW is | |
10149 | * set. We use this counter to prevent snapshots. We must increment it | |
10150 | * before walking the extents because we don't want a concurrent | |
10151 | * snapshot to run after we've already checked the extents. | |
10152 | */ | |
10153 | atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles); | |
10154 | ||
10155 | isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize); | |
10156 | ||
10157 | lock_extent_bits(io_tree, 0, isize - 1, &cached_state); | |
10158 | start = 0; | |
10159 | while (start < isize) { | |
10160 | u64 logical_block_start, physical_block_start; | |
32da5386 | 10161 | struct btrfs_block_group *bg; |
ed46ff3d OS |
10162 | u64 len = isize - start; |
10163 | ||
39b07b5d | 10164 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len); |
ed46ff3d OS |
10165 | if (IS_ERR(em)) { |
10166 | ret = PTR_ERR(em); | |
10167 | goto out; | |
10168 | } | |
10169 | ||
10170 | if (em->block_start == EXTENT_MAP_HOLE) { | |
10171 | btrfs_warn(fs_info, "swapfile must not have holes"); | |
10172 | ret = -EINVAL; | |
10173 | goto out; | |
10174 | } | |
10175 | if (em->block_start == EXTENT_MAP_INLINE) { | |
10176 | /* | |
10177 | * It's unlikely we'll ever actually find ourselves | |
10178 | * here, as a file small enough to fit inline won't be | |
10179 | * big enough to store more than the swap header, but in | |
10180 | * case something changes in the future, let's catch it | |
10181 | * here rather than later. | |
10182 | */ | |
10183 | btrfs_warn(fs_info, "swapfile must not be inline"); | |
10184 | ret = -EINVAL; | |
10185 | goto out; | |
10186 | } | |
10187 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
10188 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10189 | ret = -EINVAL; | |
10190 | goto out; | |
10191 | } | |
10192 | ||
10193 | logical_block_start = em->block_start + (start - em->start); | |
10194 | len = min(len, em->len - (start - em->start)); | |
10195 | free_extent_map(em); | |
10196 | em = NULL; | |
10197 | ||
a84d5d42 | 10198 | ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true); |
ed46ff3d OS |
10199 | if (ret < 0) { |
10200 | goto out; | |
10201 | } else if (ret) { | |
10202 | ret = 0; | |
10203 | } else { | |
10204 | btrfs_warn(fs_info, | |
10205 | "swapfile must not be copy-on-write"); | |
10206 | ret = -EINVAL; | |
10207 | goto out; | |
10208 | } | |
10209 | ||
10210 | em = btrfs_get_chunk_map(fs_info, logical_block_start, len); | |
10211 | if (IS_ERR(em)) { | |
10212 | ret = PTR_ERR(em); | |
10213 | goto out; | |
10214 | } | |
10215 | ||
10216 | if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
10217 | btrfs_warn(fs_info, | |
10218 | "swapfile must have single data profile"); | |
10219 | ret = -EINVAL; | |
10220 | goto out; | |
10221 | } | |
10222 | ||
10223 | if (device == NULL) { | |
10224 | device = em->map_lookup->stripes[0].dev; | |
10225 | ret = btrfs_add_swapfile_pin(inode, device, false); | |
10226 | if (ret == 1) | |
10227 | ret = 0; | |
10228 | else if (ret) | |
10229 | goto out; | |
10230 | } else if (device != em->map_lookup->stripes[0].dev) { | |
10231 | btrfs_warn(fs_info, "swapfile must be on one device"); | |
10232 | ret = -EINVAL; | |
10233 | goto out; | |
10234 | } | |
10235 | ||
10236 | physical_block_start = (em->map_lookup->stripes[0].physical + | |
10237 | (logical_block_start - em->start)); | |
10238 | len = min(len, em->len - (logical_block_start - em->start)); | |
10239 | free_extent_map(em); | |
10240 | em = NULL; | |
10241 | ||
10242 | bg = btrfs_lookup_block_group(fs_info, logical_block_start); | |
10243 | if (!bg) { | |
10244 | btrfs_warn(fs_info, | |
10245 | "could not find block group containing swapfile"); | |
10246 | ret = -EINVAL; | |
10247 | goto out; | |
10248 | } | |
10249 | ||
10250 | ret = btrfs_add_swapfile_pin(inode, bg, true); | |
10251 | if (ret) { | |
10252 | btrfs_put_block_group(bg); | |
10253 | if (ret == 1) | |
10254 | ret = 0; | |
10255 | else | |
10256 | goto out; | |
10257 | } | |
10258 | ||
10259 | if (bsi.block_len && | |
10260 | bsi.block_start + bsi.block_len == physical_block_start) { | |
10261 | bsi.block_len += len; | |
10262 | } else { | |
10263 | if (bsi.block_len) { | |
10264 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10265 | if (ret) | |
10266 | goto out; | |
10267 | } | |
10268 | bsi.start = start; | |
10269 | bsi.block_start = physical_block_start; | |
10270 | bsi.block_len = len; | |
10271 | } | |
10272 | ||
10273 | start += len; | |
10274 | } | |
10275 | ||
10276 | if (bsi.block_len) | |
10277 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10278 | ||
10279 | out: | |
10280 | if (!IS_ERR_OR_NULL(em)) | |
10281 | free_extent_map(em); | |
10282 | ||
10283 | unlock_extent_cached(io_tree, 0, isize - 1, &cached_state); | |
10284 | ||
10285 | if (ret) | |
10286 | btrfs_swap_deactivate(file); | |
10287 | ||
c3e1f96c | 10288 | btrfs_exclop_finish(fs_info); |
ed46ff3d OS |
10289 | |
10290 | if (ret) | |
10291 | return ret; | |
10292 | ||
10293 | if (device) | |
10294 | sis->bdev = device->bdev; | |
10295 | *span = bsi.highest_ppage - bsi.lowest_ppage + 1; | |
10296 | sis->max = bsi.nr_pages; | |
10297 | sis->pages = bsi.nr_pages - 1; | |
10298 | sis->highest_bit = bsi.nr_pages - 1; | |
10299 | return bsi.nr_extents; | |
10300 | } | |
10301 | #else | |
10302 | static void btrfs_swap_deactivate(struct file *file) | |
10303 | { | |
10304 | } | |
10305 | ||
10306 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10307 | sector_t *span) | |
10308 | { | |
10309 | return -EOPNOTSUPP; | |
10310 | } | |
10311 | #endif | |
10312 | ||
2766ff61 FM |
10313 | /* |
10314 | * Update the number of bytes used in the VFS' inode. When we replace extents in | |
10315 | * a range (clone, dedupe, fallocate's zero range), we must update the number of | |
10316 | * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls | |
10317 | * always get a correct value. | |
10318 | */ | |
10319 | void btrfs_update_inode_bytes(struct btrfs_inode *inode, | |
10320 | const u64 add_bytes, | |
10321 | const u64 del_bytes) | |
10322 | { | |
10323 | if (add_bytes == del_bytes) | |
10324 | return; | |
10325 | ||
10326 | spin_lock(&inode->lock); | |
10327 | if (del_bytes > 0) | |
10328 | inode_sub_bytes(&inode->vfs_inode, del_bytes); | |
10329 | if (add_bytes > 0) | |
10330 | inode_add_bytes(&inode->vfs_inode, add_bytes); | |
10331 | spin_unlock(&inode->lock); | |
10332 | } | |
10333 | ||
6e1d5dcc | 10334 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10335 | .getattr = btrfs_getattr, |
39279cc3 CM |
10336 | .lookup = btrfs_lookup, |
10337 | .create = btrfs_create, | |
10338 | .unlink = btrfs_unlink, | |
10339 | .link = btrfs_link, | |
10340 | .mkdir = btrfs_mkdir, | |
10341 | .rmdir = btrfs_rmdir, | |
2773bf00 | 10342 | .rename = btrfs_rename2, |
39279cc3 CM |
10343 | .symlink = btrfs_symlink, |
10344 | .setattr = btrfs_setattr, | |
618e21d5 | 10345 | .mknod = btrfs_mknod, |
5103e947 | 10346 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10347 | .permission = btrfs_permission, |
4e34e719 | 10348 | .get_acl = btrfs_get_acl, |
996a710d | 10349 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10350 | .update_time = btrfs_update_time, |
ef3b9af5 | 10351 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 10352 | }; |
76dda93c | 10353 | |
828c0950 | 10354 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10355 | .llseek = generic_file_llseek, |
10356 | .read = generic_read_dir, | |
02dbfc99 | 10357 | .iterate_shared = btrfs_real_readdir, |
23b5ec74 | 10358 | .open = btrfs_opendir, |
34287aa3 | 10359 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10360 | #ifdef CONFIG_COMPAT |
4c63c245 | 10361 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 10362 | #endif |
6bf13c0c | 10363 | .release = btrfs_release_file, |
e02119d5 | 10364 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10365 | }; |
10366 | ||
35054394 CM |
10367 | /* |
10368 | * btrfs doesn't support the bmap operation because swapfiles | |
10369 | * use bmap to make a mapping of extents in the file. They assume | |
10370 | * these extents won't change over the life of the file and they | |
10371 | * use the bmap result to do IO directly to the drive. | |
10372 | * | |
10373 | * the btrfs bmap call would return logical addresses that aren't | |
10374 | * suitable for IO and they also will change frequently as COW | |
10375 | * operations happen. So, swapfile + btrfs == corruption. | |
10376 | * | |
10377 | * For now we're avoiding this by dropping bmap. | |
10378 | */ | |
7f09410b | 10379 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
10380 | .readpage = btrfs_readpage, |
10381 | .writepage = btrfs_writepage, | |
b293f02e | 10382 | .writepages = btrfs_writepages, |
ba206a02 | 10383 | .readahead = btrfs_readahead, |
f85781fb | 10384 | .direct_IO = noop_direct_IO, |
a52d9a80 CM |
10385 | .invalidatepage = btrfs_invalidatepage, |
10386 | .releasepage = btrfs_releasepage, | |
f8e66081 RG |
10387 | #ifdef CONFIG_MIGRATION |
10388 | .migratepage = btrfs_migratepage, | |
10389 | #endif | |
e6dcd2dc | 10390 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 10391 | .error_remove_page = generic_error_remove_page, |
ed46ff3d OS |
10392 | .swap_activate = btrfs_swap_activate, |
10393 | .swap_deactivate = btrfs_swap_deactivate, | |
39279cc3 CM |
10394 | }; |
10395 | ||
6e1d5dcc | 10396 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10397 | .getattr = btrfs_getattr, |
10398 | .setattr = btrfs_setattr, | |
5103e947 | 10399 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10400 | .permission = btrfs_permission, |
1506fcc8 | 10401 | .fiemap = btrfs_fiemap, |
4e34e719 | 10402 | .get_acl = btrfs_get_acl, |
996a710d | 10403 | .set_acl = btrfs_set_acl, |
e41f941a | 10404 | .update_time = btrfs_update_time, |
39279cc3 | 10405 | }; |
6e1d5dcc | 10406 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10407 | .getattr = btrfs_getattr, |
10408 | .setattr = btrfs_setattr, | |
fdebe2bd | 10409 | .permission = btrfs_permission, |
33268eaf | 10410 | .listxattr = btrfs_listxattr, |
4e34e719 | 10411 | .get_acl = btrfs_get_acl, |
996a710d | 10412 | .set_acl = btrfs_set_acl, |
e41f941a | 10413 | .update_time = btrfs_update_time, |
618e21d5 | 10414 | }; |
6e1d5dcc | 10415 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 10416 | .get_link = page_get_link, |
f209561a | 10417 | .getattr = btrfs_getattr, |
22c44fe6 | 10418 | .setattr = btrfs_setattr, |
fdebe2bd | 10419 | .permission = btrfs_permission, |
0279b4cd | 10420 | .listxattr = btrfs_listxattr, |
e41f941a | 10421 | .update_time = btrfs_update_time, |
39279cc3 | 10422 | }; |
76dda93c | 10423 | |
82d339d9 | 10424 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
10425 | .d_delete = btrfs_dentry_delete, |
10426 | }; |