1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6 #include <crypto/hash.h>
7 #include <linux/kernel.h>
8 #include <linux/bio.h>
9 #include <linux/blk-cgroup.h>
10 #include <linux/file.h>
11 #include <linux/fs.h>
12 #include <linux/pagemap.h>
13 #include <linux/highmem.h>
14 #include <linux/time.h>
15 #include <linux/init.h>
16 #include <linux/string.h>
17 #include <linux/backing-dev.h>
18 #include <linux/writeback.h>
19 #include <linux/compat.h>
20 #include <linux/xattr.h>
21 #include <linux/posix_acl.h>
22 #include <linux/falloc.h>
23 #include <linux/slab.h>
24 #include <linux/ratelimit.h>
25 #include <linux/btrfs.h>
26 #include <linux/blkdev.h>
27 #include <linux/posix_acl_xattr.h>
28 #include <linux/uio.h>
29 #include <linux/magic.h>
30 #include <linux/iversion.h>
31 #include <linux/swap.h>
32 #include <linux/migrate.h>
33 #include <linux/sched/mm.h>
34 #include <linux/iomap.h>
35 #include <asm/unaligned.h>
36 #include <linux/fsverity.h>
37 #include "misc.h"
38 #include "ctree.h"
39 #include "disk-io.h"
40 #include "transaction.h"
41 #include "btrfs_inode.h"
42 #include "ordered-data.h"
43 #include "xattr.h"
44 #include "tree-log.h"
45 #include "bio.h"
46 #include "compression.h"
47 #include "locking.h"
48 #include "props.h"
49 #include "qgroup.h"
50 #include "delalloc-space.h"
51 #include "block-group.h"
52 #include "space-info.h"
53 #include "zoned.h"
54 #include "subpage.h"
55 #include "inode-item.h"
56 #include "fs.h"
57 #include "accessors.h"
58 #include "extent-tree.h"
59 #include "root-tree.h"
60 #include "defrag.h"
61 #include "dir-item.h"
62 #include "file-item.h"
63 #include "uuid-tree.h"
64 #include "ioctl.h"
65 #include "file.h"
66 #include "acl.h"
67 #include "relocation.h"
68 #include "verity.h"
69 #include "super.h"
70 #include "orphan.h"
71 #include "backref.h"
72 #include "raid-stripe-tree.h"
73
74 struct btrfs_iget_args {
75 u64 ino;
76 struct btrfs_root *root;
77 };
78
79 struct btrfs_dio_data {
80 ssize_t submitted;
81 struct extent_changeset *data_reserved;
82 struct btrfs_ordered_extent *ordered;
83 bool data_space_reserved;
84 bool nocow_done;
85 };
86
87 struct btrfs_dio_private {
88 /* Range of I/O */
89 u64 file_offset;
90 u32 bytes;
91
92 /* This must be last */
93 struct btrfs_bio bbio;
94 };
95
96 static struct bio_set btrfs_dio_bioset;
97
98 struct btrfs_rename_ctx {
99 /* Output field. Stores the index number of the old directory entry. */
100 u64 index;
101 };
102
103 /*
104 * Used by data_reloc_print_warning_inode() to pass needed info for filename
105 * resolution and output of error message.
106 */
107 struct data_reloc_warn {
108 struct btrfs_path path;
109 struct btrfs_fs_info *fs_info;
110 u64 extent_item_size;
111 u64 logical;
112 int mirror_num;
113 };
114
115 /*
116 * For the file_extent_tree, we want to hold the inode lock when we lookup and
117 * update the disk_i_size, but lockdep will complain because our io_tree we hold
118 * the tree lock and get the inode lock when setting delalloc. These two things
119 * are unrelated, so make a class for the file_extent_tree so we don't get the
120 * two locking patterns mixed up.
121 */
122 static struct lock_class_key file_extent_tree_class;
123
124 static const struct inode_operations btrfs_dir_inode_operations;
125 static const struct inode_operations btrfs_symlink_inode_operations;
126 static const struct inode_operations btrfs_special_inode_operations;
127 static const struct inode_operations btrfs_file_inode_operations;
128 static const struct address_space_operations btrfs_aops;
129 static const struct file_operations btrfs_dir_file_operations;
130
131 static struct kmem_cache *btrfs_inode_cachep;
132
133 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
134 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback);
135
136 static noinline int run_delalloc_cow(struct btrfs_inode *inode,
137 struct page *locked_page, u64 start,
138 u64 end, struct writeback_control *wbc,
139 bool pages_dirty);
140 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
141 u64 len, u64 orig_start, u64 block_start,
142 u64 block_len, u64 orig_block_len,
143 u64 ram_bytes, int compress_type,
144 int type);
145
data_reloc_print_warning_inode(u64 inum,u64 offset,u64 num_bytes,u64 root,void * warn_ctx)146 static int data_reloc_print_warning_inode(u64 inum, u64 offset, u64 num_bytes,
147 u64 root, void *warn_ctx)
148 {
149 struct data_reloc_warn *warn = warn_ctx;
150 struct btrfs_fs_info *fs_info = warn->fs_info;
151 struct extent_buffer *eb;
152 struct btrfs_inode_item *inode_item;
153 struct inode_fs_paths *ipath = NULL;
154 struct btrfs_root *local_root;
155 struct btrfs_key key;
156 unsigned int nofs_flag;
157 u32 nlink;
158 int ret;
159
160 local_root = btrfs_get_fs_root(fs_info, root, true);
161 if (IS_ERR(local_root)) {
162 ret = PTR_ERR(local_root);
163 goto err;
164 }
165
166 /* This makes the path point to (inum INODE_ITEM ioff). */
167 key.objectid = inum;
168 key.type = BTRFS_INODE_ITEM_KEY;
169 key.offset = 0;
170
171 ret = btrfs_search_slot(NULL, local_root, &key, &warn->path, 0, 0);
172 if (ret) {
173 btrfs_put_root(local_root);
174 btrfs_release_path(&warn->path);
175 goto err;
176 }
177
178 eb = warn->path.nodes[0];
179 inode_item = btrfs_item_ptr(eb, warn->path.slots[0], struct btrfs_inode_item);
180 nlink = btrfs_inode_nlink(eb, inode_item);
181 btrfs_release_path(&warn->path);
182
183 nofs_flag = memalloc_nofs_save();
184 ipath = init_ipath(4096, local_root, &warn->path);
185 memalloc_nofs_restore(nofs_flag);
186 if (IS_ERR(ipath)) {
187 btrfs_put_root(local_root);
188 ret = PTR_ERR(ipath);
189 ipath = NULL;
190 /*
191 * -ENOMEM, not a critical error, just output an generic error
192 * without filename.
193 */
194 btrfs_warn(fs_info,
195 "checksum error at logical %llu mirror %u root %llu, inode %llu offset %llu",
196 warn->logical, warn->mirror_num, root, inum, offset);
197 return ret;
198 }
199 ret = paths_from_inode(inum, ipath);
200 if (ret < 0)
201 goto err;
202
203 /*
204 * We deliberately ignore the bit ipath might have been too small to
205 * hold all of the paths here
206 */
207 for (int i = 0; i < ipath->fspath->elem_cnt; i++) {
208 btrfs_warn(fs_info,
209 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu length %u links %u (path: %s)",
210 warn->logical, warn->mirror_num, root, inum, offset,
211 fs_info->sectorsize, nlink,
212 (char *)(unsigned long)ipath->fspath->val[i]);
213 }
214
215 btrfs_put_root(local_root);
216 free_ipath(ipath);
217 return 0;
218
219 err:
220 btrfs_warn(fs_info,
221 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu, path resolving failed with ret=%d",
222 warn->logical, warn->mirror_num, root, inum, offset, ret);
223
224 free_ipath(ipath);
225 return ret;
226 }
227
228 /*
229 * Do extra user-friendly error output (e.g. lookup all the affected files).
230 *
231 * Return true if we succeeded doing the backref lookup.
232 * Return false if such lookup failed, and has to fallback to the old error message.
233 */
print_data_reloc_error(const struct btrfs_inode * inode,u64 file_off,const u8 * csum,const u8 * csum_expected,int mirror_num)234 static void print_data_reloc_error(const struct btrfs_inode *inode, u64 file_off,
235 const u8 *csum, const u8 *csum_expected,
236 int mirror_num)
237 {
238 struct btrfs_fs_info *fs_info = inode->root->fs_info;
239 struct btrfs_path path = { 0 };
240 struct btrfs_key found_key = { 0 };
241 struct extent_buffer *eb;
242 struct btrfs_extent_item *ei;
243 const u32 csum_size = fs_info->csum_size;
244 u64 logical;
245 u64 flags;
246 u32 item_size;
247 int ret;
248
249 mutex_lock(&fs_info->reloc_mutex);
250 logical = btrfs_get_reloc_bg_bytenr(fs_info);
251 mutex_unlock(&fs_info->reloc_mutex);
252
253 if (logical == U64_MAX) {
254 btrfs_warn_rl(fs_info, "has data reloc tree but no running relocation");
255 btrfs_warn_rl(fs_info,
256 "csum failed root %lld ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
257 btrfs_root_id(inode->root), btrfs_ino(inode), file_off,
258 CSUM_FMT_VALUE(csum_size, csum),
259 CSUM_FMT_VALUE(csum_size, csum_expected),
260 mirror_num);
261 return;
262 }
263
264 logical += file_off;
265 btrfs_warn_rl(fs_info,
266 "csum failed root %lld ino %llu off %llu logical %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
267 btrfs_root_id(inode->root),
268 btrfs_ino(inode), file_off, logical,
269 CSUM_FMT_VALUE(csum_size, csum),
270 CSUM_FMT_VALUE(csum_size, csum_expected),
271 mirror_num);
272
273 ret = extent_from_logical(fs_info, logical, &path, &found_key, &flags);
274 if (ret < 0) {
275 btrfs_err_rl(fs_info, "failed to lookup extent item for logical %llu: %d",
276 logical, ret);
277 return;
278 }
279 eb = path.nodes[0];
280 ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item);
281 item_size = btrfs_item_size(eb, path.slots[0]);
282 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
283 unsigned long ptr = 0;
284 u64 ref_root;
285 u8 ref_level;
286
287 while (true) {
288 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
289 item_size, &ref_root,
290 &ref_level);
291 if (ret < 0) {
292 btrfs_warn_rl(fs_info,
293 "failed to resolve tree backref for logical %llu: %d",
294 logical, ret);
295 break;
296 }
297 if (ret > 0)
298 break;
299
300 btrfs_warn_rl(fs_info,
301 "csum error at logical %llu mirror %u: metadata %s (level %d) in tree %llu",
302 logical, mirror_num,
303 (ref_level ? "node" : "leaf"),
304 ref_level, ref_root);
305 }
306 btrfs_release_path(&path);
307 } else {
308 struct btrfs_backref_walk_ctx ctx = { 0 };
309 struct data_reloc_warn reloc_warn = { 0 };
310
311 btrfs_release_path(&path);
312
313 ctx.bytenr = found_key.objectid;
314 ctx.extent_item_pos = logical - found_key.objectid;
315 ctx.fs_info = fs_info;
316
317 reloc_warn.logical = logical;
318 reloc_warn.extent_item_size = found_key.offset;
319 reloc_warn.mirror_num = mirror_num;
320 reloc_warn.fs_info = fs_info;
321
322 iterate_extent_inodes(&ctx, true,
323 data_reloc_print_warning_inode, &reloc_warn);
324 }
325 }
326
btrfs_print_data_csum_error(struct btrfs_inode * inode,u64 logical_start,u8 * csum,u8 * csum_expected,int mirror_num)327 static void __cold btrfs_print_data_csum_error(struct btrfs_inode *inode,
328 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
329 {
330 struct btrfs_root *root = inode->root;
331 const u32 csum_size = root->fs_info->csum_size;
332
333 /* For data reloc tree, it's better to do a backref lookup instead. */
334 if (btrfs_root_id(root) == BTRFS_DATA_RELOC_TREE_OBJECTID)
335 return print_data_reloc_error(inode, logical_start, csum,
336 csum_expected, mirror_num);
337
338 /* Output without objectid, which is more meaningful */
339 if (btrfs_root_id(root) >= BTRFS_LAST_FREE_OBJECTID) {
340 btrfs_warn_rl(root->fs_info,
341 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
342 btrfs_root_id(root), btrfs_ino(inode),
343 logical_start,
344 CSUM_FMT_VALUE(csum_size, csum),
345 CSUM_FMT_VALUE(csum_size, csum_expected),
346 mirror_num);
347 } else {
348 btrfs_warn_rl(root->fs_info,
349 "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
350 btrfs_root_id(root), btrfs_ino(inode),
351 logical_start,
352 CSUM_FMT_VALUE(csum_size, csum),
353 CSUM_FMT_VALUE(csum_size, csum_expected),
354 mirror_num);
355 }
356 }
357
358 /*
359 * Lock inode i_rwsem based on arguments passed.
360 *
361 * ilock_flags can have the following bit set:
362 *
363 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
364 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
365 * return -EAGAIN
366 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
367 */
btrfs_inode_lock(struct btrfs_inode * inode,unsigned int ilock_flags)368 int btrfs_inode_lock(struct btrfs_inode *inode, unsigned int ilock_flags)
369 {
370 if (ilock_flags & BTRFS_ILOCK_SHARED) {
371 if (ilock_flags & BTRFS_ILOCK_TRY) {
372 if (!inode_trylock_shared(&inode->vfs_inode))
373 return -EAGAIN;
374 else
375 return 0;
376 }
377 inode_lock_shared(&inode->vfs_inode);
378 } else {
379 if (ilock_flags & BTRFS_ILOCK_TRY) {
380 if (!inode_trylock(&inode->vfs_inode))
381 return -EAGAIN;
382 else
383 return 0;
384 }
385 inode_lock(&inode->vfs_inode);
386 }
387 if (ilock_flags & BTRFS_ILOCK_MMAP)
388 down_write(&inode->i_mmap_lock);
389 return 0;
390 }
391
392 /*
393 * Unock inode i_rwsem.
394 *
395 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
396 * to decide whether the lock acquired is shared or exclusive.
397 */
btrfs_inode_unlock(struct btrfs_inode * inode,unsigned int ilock_flags)398 void btrfs_inode_unlock(struct btrfs_inode *inode, unsigned int ilock_flags)
399 {
400 if (ilock_flags & BTRFS_ILOCK_MMAP)
401 up_write(&inode->i_mmap_lock);
402 if (ilock_flags & BTRFS_ILOCK_SHARED)
403 inode_unlock_shared(&inode->vfs_inode);
404 else
405 inode_unlock(&inode->vfs_inode);
406 }
407
408 /*
409 * Cleanup all submitted ordered extents in specified range to handle errors
410 * from the btrfs_run_delalloc_range() callback.
411 *
412 * NOTE: caller must ensure that when an error happens, it can not call
413 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
414 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
415 * to be released, which we want to happen only when finishing the ordered
416 * extent (btrfs_finish_ordered_io()).
417 */
btrfs_cleanup_ordered_extents(struct btrfs_inode * inode,struct page * locked_page,u64 offset,u64 bytes)418 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
419 struct page *locked_page,
420 u64 offset, u64 bytes)
421 {
422 unsigned long index = offset >> PAGE_SHIFT;
423 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
424 u64 page_start = 0, page_end = 0;
425 struct page *page;
426
427 if (locked_page) {
428 page_start = page_offset(locked_page);
429 page_end = page_start + PAGE_SIZE - 1;
430 }
431
432 while (index <= end_index) {
433 /*
434 * For locked page, we will call btrfs_mark_ordered_io_finished
435 * through btrfs_mark_ordered_io_finished() on it
436 * in run_delalloc_range() for the error handling, which will
437 * clear page Ordered and run the ordered extent accounting.
438 *
439 * Here we can't just clear the Ordered bit, or
440 * btrfs_mark_ordered_io_finished() would skip the accounting
441 * for the page range, and the ordered extent will never finish.
442 */
443 if (locked_page && index == (page_start >> PAGE_SHIFT)) {
444 index++;
445 continue;
446 }
447 page = find_get_page(inode->vfs_inode.i_mapping, index);
448 index++;
449 if (!page)
450 continue;
451
452 /*
453 * Here we just clear all Ordered bits for every page in the
454 * range, then btrfs_mark_ordered_io_finished() will handle
455 * the ordered extent accounting for the range.
456 */
457 btrfs_folio_clamp_clear_ordered(inode->root->fs_info,
458 page_folio(page), offset, bytes);
459 put_page(page);
460 }
461
462 if (locked_page) {
463 /* The locked page covers the full range, nothing needs to be done */
464 if (bytes + offset <= page_start + PAGE_SIZE)
465 return;
466 /*
467 * In case this page belongs to the delalloc range being
468 * instantiated then skip it, since the first page of a range is
469 * going to be properly cleaned up by the caller of
470 * run_delalloc_range
471 */
472 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
473 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
474 offset = page_offset(locked_page) + PAGE_SIZE;
475 }
476 }
477
478 return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false);
479 }
480
481 static int btrfs_dirty_inode(struct btrfs_inode *inode);
482
btrfs_init_inode_security(struct btrfs_trans_handle * trans,struct btrfs_new_inode_args * args)483 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
484 struct btrfs_new_inode_args *args)
485 {
486 int err;
487
488 if (args->default_acl) {
489 err = __btrfs_set_acl(trans, args->inode, args->default_acl,
490 ACL_TYPE_DEFAULT);
491 if (err)
492 return err;
493 }
494 if (args->acl) {
495 err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
496 if (err)
497 return err;
498 }
499 if (!args->default_acl && !args->acl)
500 cache_no_acl(args->inode);
501 return btrfs_xattr_security_init(trans, args->inode, args->dir,
502 &args->dentry->d_name);
503 }
504
505 /*
506 * this does all the hard work for inserting an inline extent into
507 * the btree. The caller should have done a btrfs_drop_extents so that
508 * no overlapping inline items exist in the btree
509 */
insert_inline_extent(struct btrfs_trans_handle * trans,struct btrfs_path * path,struct btrfs_inode * inode,bool extent_inserted,size_t size,size_t compressed_size,int compress_type,struct folio * compressed_folio,bool update_i_size)510 static int insert_inline_extent(struct btrfs_trans_handle *trans,
511 struct btrfs_path *path,
512 struct btrfs_inode *inode, bool extent_inserted,
513 size_t size, size_t compressed_size,
514 int compress_type,
515 struct folio *compressed_folio,
516 bool update_i_size)
517 {
518 struct btrfs_root *root = inode->root;
519 struct extent_buffer *leaf;
520 struct page *page = NULL;
521 const u32 sectorsize = trans->fs_info->sectorsize;
522 char *kaddr;
523 unsigned long ptr;
524 struct btrfs_file_extent_item *ei;
525 int ret;
526 size_t cur_size = size;
527 u64 i_size;
528
529 /*
530 * The decompressed size must still be no larger than a sector. Under
531 * heavy race, we can have size == 0 passed in, but that shouldn't be a
532 * big deal and we can continue the insertion.
533 */
534 ASSERT(size <= sectorsize);
535
536 /*
537 * The compressed size also needs to be no larger than a sector.
538 * That's also why we only need one page as the parameter.
539 */
540 if (compressed_folio)
541 ASSERT(compressed_size <= sectorsize);
542 else
543 ASSERT(compressed_size == 0);
544
545 if (compressed_size && compressed_folio)
546 cur_size = compressed_size;
547
548 if (!extent_inserted) {
549 struct btrfs_key key;
550 size_t datasize;
551
552 key.objectid = btrfs_ino(inode);
553 key.offset = 0;
554 key.type = BTRFS_EXTENT_DATA_KEY;
555
556 datasize = btrfs_file_extent_calc_inline_size(cur_size);
557 ret = btrfs_insert_empty_item(trans, root, path, &key,
558 datasize);
559 if (ret)
560 goto fail;
561 }
562 leaf = path->nodes[0];
563 ei = btrfs_item_ptr(leaf, path->slots[0],
564 struct btrfs_file_extent_item);
565 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
566 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
567 btrfs_set_file_extent_encryption(leaf, ei, 0);
568 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
569 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
570 ptr = btrfs_file_extent_inline_start(ei);
571
572 if (compress_type != BTRFS_COMPRESS_NONE) {
573 kaddr = kmap_local_folio(compressed_folio, 0);
574 write_extent_buffer(leaf, kaddr, ptr, compressed_size);
575 kunmap_local(kaddr);
576
577 btrfs_set_file_extent_compression(leaf, ei,
578 compress_type);
579 } else {
580 page = find_get_page(inode->vfs_inode.i_mapping, 0);
581 btrfs_set_file_extent_compression(leaf, ei, 0);
582 kaddr = kmap_local_page(page);
583 write_extent_buffer(leaf, kaddr, ptr, size);
584 kunmap_local(kaddr);
585 put_page(page);
586 }
587 btrfs_mark_buffer_dirty(trans, leaf);
588 btrfs_release_path(path);
589
590 /*
591 * We align size to sectorsize for inline extents just for simplicity
592 * sake.
593 */
594 ret = btrfs_inode_set_file_extent_range(inode, 0,
595 ALIGN(size, root->fs_info->sectorsize));
596 if (ret)
597 goto fail;
598
599 /*
600 * We're an inline extent, so nobody can extend the file past i_size
601 * without locking a page we already have locked.
602 *
603 * We must do any i_size and inode updates before we unlock the pages.
604 * Otherwise we could end up racing with unlink.
605 */
606 i_size = i_size_read(&inode->vfs_inode);
607 if (update_i_size && size > i_size) {
608 i_size_write(&inode->vfs_inode, size);
609 i_size = size;
610 }
611 inode->disk_i_size = i_size;
612
613 fail:
614 return ret;
615 }
616
can_cow_file_range_inline(struct btrfs_inode * inode,u64 offset,u64 size,size_t compressed_size)617 static bool can_cow_file_range_inline(struct btrfs_inode *inode,
618 u64 offset, u64 size,
619 size_t compressed_size)
620 {
621 struct btrfs_fs_info *fs_info = inode->root->fs_info;
622 u64 data_len = (compressed_size ?: size);
623
624 /* Inline extents must start at offset 0. */
625 if (offset != 0)
626 return false;
627
628 /*
629 * Due to the page size limit, for subpage we can only trigger the
630 * writeback for the dirty sectors of page, that means data writeback
631 * is doing more writeback than what we want.
632 *
633 * This is especially unexpected for some call sites like fallocate,
634 * where we only increase i_size after everything is done.
635 * This means we can trigger inline extent even if we didn't want to.
636 * So here we skip inline extent creation completely.
637 */
638 if (fs_info->sectorsize != PAGE_SIZE)
639 return false;
640
641 /* Inline extents are limited to sectorsize. */
642 if (size > fs_info->sectorsize)
643 return false;
644
645 /* We cannot exceed the maximum inline data size. */
646 if (data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
647 return false;
648
649 /* We cannot exceed the user specified max_inline size. */
650 if (data_len > fs_info->max_inline)
651 return false;
652
653 /* Inline extents must be the entirety of the file. */
654 if (size < i_size_read(&inode->vfs_inode))
655 return false;
656
657 return true;
658 }
659
660 /*
661 * conditionally insert an inline extent into the file. This
662 * does the checks required to make sure the data is small enough
663 * to fit as an inline extent.
664 *
665 * If being used directly, you must have already checked we're allowed to cow
666 * the range by getting true from can_cow_file_range_inline().
667 */
__cow_file_range_inline(struct btrfs_inode * inode,u64 offset,u64 size,size_t compressed_size,int compress_type,struct folio * compressed_folio,bool update_i_size)668 static noinline int __cow_file_range_inline(struct btrfs_inode *inode, u64 offset,
669 u64 size, size_t compressed_size,
670 int compress_type,
671 struct folio *compressed_folio,
672 bool update_i_size)
673 {
674 struct btrfs_drop_extents_args drop_args = { 0 };
675 struct btrfs_root *root = inode->root;
676 struct btrfs_fs_info *fs_info = root->fs_info;
677 struct btrfs_trans_handle *trans;
678 u64 data_len = (compressed_size ?: size);
679 int ret;
680 struct btrfs_path *path;
681
682 path = btrfs_alloc_path();
683 if (!path)
684 return -ENOMEM;
685
686 trans = btrfs_join_transaction(root);
687 if (IS_ERR(trans)) {
688 btrfs_free_path(path);
689 return PTR_ERR(trans);
690 }
691 trans->block_rsv = &inode->block_rsv;
692
693 drop_args.path = path;
694 drop_args.start = 0;
695 drop_args.end = fs_info->sectorsize;
696 drop_args.drop_cache = true;
697 drop_args.replace_extent = true;
698 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
699 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
700 if (ret) {
701 btrfs_abort_transaction(trans, ret);
702 goto out;
703 }
704
705 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
706 size, compressed_size, compress_type,
707 compressed_folio, update_i_size);
708 if (ret && ret != -ENOSPC) {
709 btrfs_abort_transaction(trans, ret);
710 goto out;
711 } else if (ret == -ENOSPC) {
712 ret = 1;
713 goto out;
714 }
715
716 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
717 ret = btrfs_update_inode(trans, inode);
718 if (ret && ret != -ENOSPC) {
719 btrfs_abort_transaction(trans, ret);
720 goto out;
721 } else if (ret == -ENOSPC) {
722 ret = 1;
723 goto out;
724 }
725
726 btrfs_set_inode_full_sync(inode);
727 out:
728 /*
729 * Don't forget to free the reserved space, as for inlined extent
730 * it won't count as data extent, free them directly here.
731 * And at reserve time, it's always aligned to page size, so
732 * just free one page here.
733 */
734 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE, NULL);
735 btrfs_free_path(path);
736 btrfs_end_transaction(trans);
737 return ret;
738 }
739
cow_file_range_inline(struct btrfs_inode * inode,u64 offset,u64 end,size_t compressed_size,int compress_type,struct folio * compressed_folio,bool update_i_size)740 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 offset,
741 u64 end,
742 size_t compressed_size,
743 int compress_type,
744 struct folio *compressed_folio,
745 bool update_i_size)
746 {
747 struct extent_state *cached = NULL;
748 unsigned long clear_flags = EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
749 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING | EXTENT_LOCKED;
750 u64 size = min_t(u64, i_size_read(&inode->vfs_inode), end + 1);
751 int ret;
752
753 if (!can_cow_file_range_inline(inode, offset, size, compressed_size))
754 return 1;
755
756 lock_extent(&inode->io_tree, offset, end, &cached);
757 ret = __cow_file_range_inline(inode, offset, size, compressed_size,
758 compress_type, compressed_folio,
759 update_i_size);
760 if (ret > 0) {
761 unlock_extent(&inode->io_tree, offset, end, &cached);
762 return ret;
763 }
764
765 extent_clear_unlock_delalloc(inode, offset, end, NULL, &cached,
766 clear_flags,
767 PAGE_UNLOCK | PAGE_START_WRITEBACK |
768 PAGE_END_WRITEBACK);
769 return ret;
770 }
771
772 struct async_extent {
773 u64 start;
774 u64 ram_size;
775 u64 compressed_size;
776 struct folio **folios;
777 unsigned long nr_folios;
778 int compress_type;
779 struct list_head list;
780 };
781
782 struct async_chunk {
783 struct btrfs_inode *inode;
784 struct page *locked_page;
785 u64 start;
786 u64 end;
787 blk_opf_t write_flags;
788 struct list_head extents;
789 struct cgroup_subsys_state *blkcg_css;
790 struct btrfs_work work;
791 struct async_cow *async_cow;
792 };
793
794 struct async_cow {
795 atomic_t num_chunks;
796 struct async_chunk chunks[];
797 };
798
add_async_extent(struct async_chunk * cow,u64 start,u64 ram_size,u64 compressed_size,struct folio ** folios,unsigned long nr_folios,int compress_type)799 static noinline int add_async_extent(struct async_chunk *cow,
800 u64 start, u64 ram_size,
801 u64 compressed_size,
802 struct folio **folios,
803 unsigned long nr_folios,
804 int compress_type)
805 {
806 struct async_extent *async_extent;
807
808 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
809 if (!async_extent)
810 return -ENOMEM;
811 async_extent->start = start;
812 async_extent->ram_size = ram_size;
813 async_extent->compressed_size = compressed_size;
814 async_extent->folios = folios;
815 async_extent->nr_folios = nr_folios;
816 async_extent->compress_type = compress_type;
817 list_add_tail(&async_extent->list, &cow->extents);
818 return 0;
819 }
820
821 /*
822 * Check if the inode needs to be submitted to compression, based on mount
823 * options, defragmentation, properties or heuristics.
824 */
inode_need_compress(struct btrfs_inode * inode,u64 start,u64 end)825 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
826 u64 end)
827 {
828 struct btrfs_fs_info *fs_info = inode->root->fs_info;
829
830 if (!btrfs_inode_can_compress(inode)) {
831 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
832 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
833 btrfs_ino(inode));
834 return 0;
835 }
836 /*
837 * Special check for subpage.
838 *
839 * We lock the full page then run each delalloc range in the page, thus
840 * for the following case, we will hit some subpage specific corner case:
841 *
842 * 0 32K 64K
843 * | |///////| |///////|
844 * \- A \- B
845 *
846 * In above case, both range A and range B will try to unlock the full
847 * page [0, 64K), causing the one finished later will have page
848 * unlocked already, triggering various page lock requirement BUG_ON()s.
849 *
850 * So here we add an artificial limit that subpage compression can only
851 * if the range is fully page aligned.
852 *
853 * In theory we only need to ensure the first page is fully covered, but
854 * the tailing partial page will be locked until the full compression
855 * finishes, delaying the write of other range.
856 *
857 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
858 * first to prevent any submitted async extent to unlock the full page.
859 * By this, we can ensure for subpage case that only the last async_cow
860 * will unlock the full page.
861 */
862 if (fs_info->sectorsize < PAGE_SIZE) {
863 if (!PAGE_ALIGNED(start) ||
864 !PAGE_ALIGNED(end + 1))
865 return 0;
866 }
867
868 /* force compress */
869 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
870 return 1;
871 /* defrag ioctl */
872 if (inode->defrag_compress)
873 return 1;
874 /* bad compression ratios */
875 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
876 return 0;
877 if (btrfs_test_opt(fs_info, COMPRESS) ||
878 inode->flags & BTRFS_INODE_COMPRESS ||
879 inode->prop_compress)
880 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
881 return 0;
882 }
883
inode_should_defrag(struct btrfs_inode * inode,u64 start,u64 end,u64 num_bytes,u32 small_write)884 static inline void inode_should_defrag(struct btrfs_inode *inode,
885 u64 start, u64 end, u64 num_bytes, u32 small_write)
886 {
887 /* If this is a small write inside eof, kick off a defrag */
888 if (num_bytes < small_write &&
889 (start > 0 || end + 1 < inode->disk_i_size))
890 btrfs_add_inode_defrag(NULL, inode, small_write);
891 }
892
893 /*
894 * Work queue call back to started compression on a file and pages.
895 *
896 * This is done inside an ordered work queue, and the compression is spread
897 * across many cpus. The actual IO submission is step two, and the ordered work
898 * queue takes care of making sure that happens in the same order things were
899 * put onto the queue by writepages and friends.
900 *
901 * If this code finds it can't get good compression, it puts an entry onto the
902 * work queue to write the uncompressed bytes. This makes sure that both
903 * compressed inodes and uncompressed inodes are written in the same order that
904 * the flusher thread sent them down.
905 */
compress_file_range(struct btrfs_work * work)906 static void compress_file_range(struct btrfs_work *work)
907 {
908 struct async_chunk *async_chunk =
909 container_of(work, struct async_chunk, work);
910 struct btrfs_inode *inode = async_chunk->inode;
911 struct btrfs_fs_info *fs_info = inode->root->fs_info;
912 struct address_space *mapping = inode->vfs_inode.i_mapping;
913 u64 blocksize = fs_info->sectorsize;
914 u64 start = async_chunk->start;
915 u64 end = async_chunk->end;
916 u64 actual_end;
917 u64 i_size;
918 int ret = 0;
919 struct folio **folios;
920 unsigned long nr_folios;
921 unsigned long total_compressed = 0;
922 unsigned long total_in = 0;
923 unsigned int poff;
924 int i;
925 int compress_type = fs_info->compress_type;
926
927 inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
928
929 /*
930 * We need to call clear_page_dirty_for_io on each page in the range.
931 * Otherwise applications with the file mmap'd can wander in and change
932 * the page contents while we are compressing them.
933 */
934 extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end);
935
936 /*
937 * We need to save i_size before now because it could change in between
938 * us evaluating the size and assigning it. This is because we lock and
939 * unlock the page in truncate and fallocate, and then modify the i_size
940 * later on.
941 *
942 * The barriers are to emulate READ_ONCE, remove that once i_size_read
943 * does that for us.
944 */
945 barrier();
946 i_size = i_size_read(&inode->vfs_inode);
947 barrier();
948 actual_end = min_t(u64, i_size, end + 1);
949 again:
950 folios = NULL;
951 nr_folios = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
952 nr_folios = min_t(unsigned long, nr_folios, BTRFS_MAX_COMPRESSED_PAGES);
953
954 /*
955 * we don't want to send crud past the end of i_size through
956 * compression, that's just a waste of CPU time. So, if the
957 * end of the file is before the start of our current
958 * requested range of bytes, we bail out to the uncompressed
959 * cleanup code that can deal with all of this.
960 *
961 * It isn't really the fastest way to fix things, but this is a
962 * very uncommon corner.
963 */
964 if (actual_end <= start)
965 goto cleanup_and_bail_uncompressed;
966
967 total_compressed = actual_end - start;
968
969 /*
970 * Skip compression for a small file range(<=blocksize) that
971 * isn't an inline extent, since it doesn't save disk space at all.
972 */
973 if (total_compressed <= blocksize &&
974 (start > 0 || end + 1 < inode->disk_i_size))
975 goto cleanup_and_bail_uncompressed;
976
977 /*
978 * For subpage case, we require full page alignment for the sector
979 * aligned range.
980 * Thus we must also check against @actual_end, not just @end.
981 */
982 if (blocksize < PAGE_SIZE) {
983 if (!PAGE_ALIGNED(start) ||
984 !PAGE_ALIGNED(round_up(actual_end, blocksize)))
985 goto cleanup_and_bail_uncompressed;
986 }
987
988 total_compressed = min_t(unsigned long, total_compressed,
989 BTRFS_MAX_UNCOMPRESSED);
990 total_in = 0;
991 ret = 0;
992
993 /*
994 * We do compression for mount -o compress and when the inode has not
995 * been flagged as NOCOMPRESS. This flag can change at any time if we
996 * discover bad compression ratios.
997 */
998 if (!inode_need_compress(inode, start, end))
999 goto cleanup_and_bail_uncompressed;
1000
1001 folios = kcalloc(nr_folios, sizeof(struct folio *), GFP_NOFS);
1002 if (!folios) {
1003 /*
1004 * Memory allocation failure is not a fatal error, we can fall
1005 * back to uncompressed code.
1006 */
1007 goto cleanup_and_bail_uncompressed;
1008 }
1009
1010 if (inode->defrag_compress)
1011 compress_type = inode->defrag_compress;
1012 else if (inode->prop_compress)
1013 compress_type = inode->prop_compress;
1014
1015 /* Compression level is applied here. */
1016 ret = btrfs_compress_folios(compress_type | (fs_info->compress_level << 4),
1017 mapping, start, folios, &nr_folios, &total_in,
1018 &total_compressed);
1019 if (ret)
1020 goto mark_incompressible;
1021
1022 /*
1023 * Zero the tail end of the last page, as we might be sending it down
1024 * to disk.
1025 */
1026 poff = offset_in_page(total_compressed);
1027 if (poff)
1028 folio_zero_range(folios[nr_folios - 1], poff, PAGE_SIZE - poff);
1029
1030 /*
1031 * Try to create an inline extent.
1032 *
1033 * If we didn't compress the entire range, try to create an uncompressed
1034 * inline extent, else a compressed one.
1035 *
1036 * Check cow_file_range() for why we don't even try to create inline
1037 * extent for the subpage case.
1038 */
1039 if (total_in < actual_end)
1040 ret = cow_file_range_inline(inode, start, end, 0,
1041 BTRFS_COMPRESS_NONE, NULL, false);
1042 else
1043 ret = cow_file_range_inline(inode, start, end, total_compressed,
1044 compress_type, folios[0], false);
1045 if (ret <= 0) {
1046 if (ret < 0)
1047 mapping_set_error(mapping, -EIO);
1048 goto free_pages;
1049 }
1050
1051 /*
1052 * We aren't doing an inline extent. Round the compressed size up to a
1053 * block size boundary so the allocator does sane things.
1054 */
1055 total_compressed = ALIGN(total_compressed, blocksize);
1056
1057 /*
1058 * One last check to make sure the compression is really a win, compare
1059 * the page count read with the blocks on disk, compression must free at
1060 * least one sector.
1061 */
1062 total_in = round_up(total_in, fs_info->sectorsize);
1063 if (total_compressed + blocksize > total_in)
1064 goto mark_incompressible;
1065
1066 /*
1067 * The async work queues will take care of doing actual allocation on
1068 * disk for these compressed pages, and will submit the bios.
1069 */
1070 ret = add_async_extent(async_chunk, start, total_in, total_compressed, folios,
1071 nr_folios, compress_type);
1072 BUG_ON(ret);
1073 if (start + total_in < end) {
1074 start += total_in;
1075 cond_resched();
1076 goto again;
1077 }
1078 return;
1079
1080 mark_incompressible:
1081 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && !inode->prop_compress)
1082 inode->flags |= BTRFS_INODE_NOCOMPRESS;
1083 cleanup_and_bail_uncompressed:
1084 ret = add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
1085 BTRFS_COMPRESS_NONE);
1086 BUG_ON(ret);
1087 free_pages:
1088 if (folios) {
1089 for (i = 0; i < nr_folios; i++) {
1090 WARN_ON(folios[i]->mapping);
1091 btrfs_free_compr_folio(folios[i]);
1092 }
1093 kfree(folios);
1094 }
1095 }
1096
free_async_extent_pages(struct async_extent * async_extent)1097 static void free_async_extent_pages(struct async_extent *async_extent)
1098 {
1099 int i;
1100
1101 if (!async_extent->folios)
1102 return;
1103
1104 for (i = 0; i < async_extent->nr_folios; i++) {
1105 WARN_ON(async_extent->folios[i]->mapping);
1106 btrfs_free_compr_folio(async_extent->folios[i]);
1107 }
1108 kfree(async_extent->folios);
1109 async_extent->nr_folios = 0;
1110 async_extent->folios = NULL;
1111 }
1112
submit_uncompressed_range(struct btrfs_inode * inode,struct async_extent * async_extent,struct page * locked_page)1113 static void submit_uncompressed_range(struct btrfs_inode *inode,
1114 struct async_extent *async_extent,
1115 struct page *locked_page)
1116 {
1117 u64 start = async_extent->start;
1118 u64 end = async_extent->start + async_extent->ram_size - 1;
1119 int ret;
1120 struct writeback_control wbc = {
1121 .sync_mode = WB_SYNC_ALL,
1122 .range_start = start,
1123 .range_end = end,
1124 .no_cgroup_owner = 1,
1125 };
1126
1127 wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode);
1128 ret = run_delalloc_cow(inode, locked_page, start, end, &wbc, false);
1129 wbc_detach_inode(&wbc);
1130 if (ret < 0) {
1131 btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
1132 if (locked_page) {
1133 const u64 page_start = page_offset(locked_page);
1134
1135 set_page_writeback(locked_page);
1136 end_page_writeback(locked_page);
1137 btrfs_mark_ordered_io_finished(inode, locked_page,
1138 page_start, PAGE_SIZE,
1139 !ret);
1140 mapping_set_error(locked_page->mapping, ret);
1141 unlock_page(locked_page);
1142 }
1143 }
1144 }
1145
submit_one_async_extent(struct async_chunk * async_chunk,struct async_extent * async_extent,u64 * alloc_hint)1146 static void submit_one_async_extent(struct async_chunk *async_chunk,
1147 struct async_extent *async_extent,
1148 u64 *alloc_hint)
1149 {
1150 struct btrfs_inode *inode = async_chunk->inode;
1151 struct extent_io_tree *io_tree = &inode->io_tree;
1152 struct btrfs_root *root = inode->root;
1153 struct btrfs_fs_info *fs_info = root->fs_info;
1154 struct btrfs_ordered_extent *ordered;
1155 struct btrfs_key ins;
1156 struct page *locked_page = NULL;
1157 struct extent_state *cached = NULL;
1158 struct extent_map *em;
1159 int ret = 0;
1160 u64 start = async_extent->start;
1161 u64 end = async_extent->start + async_extent->ram_size - 1;
1162
1163 if (async_chunk->blkcg_css)
1164 kthread_associate_blkcg(async_chunk->blkcg_css);
1165
1166 /*
1167 * If async_chunk->locked_page is in the async_extent range, we need to
1168 * handle it.
1169 */
1170 if (async_chunk->locked_page) {
1171 u64 locked_page_start = page_offset(async_chunk->locked_page);
1172 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
1173
1174 if (!(start >= locked_page_end || end <= locked_page_start))
1175 locked_page = async_chunk->locked_page;
1176 }
1177
1178 if (async_extent->compress_type == BTRFS_COMPRESS_NONE) {
1179 submit_uncompressed_range(inode, async_extent, locked_page);
1180 goto done;
1181 }
1182
1183 ret = btrfs_reserve_extent(root, async_extent->ram_size,
1184 async_extent->compressed_size,
1185 async_extent->compressed_size,
1186 0, *alloc_hint, &ins, 1, 1);
1187 if (ret) {
1188 /*
1189 * We can't reserve contiguous space for the compressed size.
1190 * Unlikely, but it's possible that we could have enough
1191 * non-contiguous space for the uncompressed size instead. So
1192 * fall back to uncompressed.
1193 */
1194 submit_uncompressed_range(inode, async_extent, locked_page);
1195 goto done;
1196 }
1197
1198 lock_extent(io_tree, start, end, &cached);
1199
1200 /* Here we're doing allocation and writeback of the compressed pages */
1201 em = create_io_em(inode, start,
1202 async_extent->ram_size, /* len */
1203 start, /* orig_start */
1204 ins.objectid, /* block_start */
1205 ins.offset, /* block_len */
1206 ins.offset, /* orig_block_len */
1207 async_extent->ram_size, /* ram_bytes */
1208 async_extent->compress_type,
1209 BTRFS_ORDERED_COMPRESSED);
1210 if (IS_ERR(em)) {
1211 ret = PTR_ERR(em);
1212 goto out_free_reserve;
1213 }
1214 free_extent_map(em);
1215
1216 ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */
1217 async_extent->ram_size, /* num_bytes */
1218 async_extent->ram_size, /* ram_bytes */
1219 ins.objectid, /* disk_bytenr */
1220 ins.offset, /* disk_num_bytes */
1221 0, /* offset */
1222 1 << BTRFS_ORDERED_COMPRESSED,
1223 async_extent->compress_type);
1224 if (IS_ERR(ordered)) {
1225 btrfs_drop_extent_map_range(inode, start, end, false);
1226 ret = PTR_ERR(ordered);
1227 goto out_free_reserve;
1228 }
1229 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1230
1231 /* Clear dirty, set writeback and unlock the pages. */
1232 extent_clear_unlock_delalloc(inode, start, end,
1233 NULL, &cached, EXTENT_LOCKED | EXTENT_DELALLOC,
1234 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1235 btrfs_submit_compressed_write(ordered,
1236 async_extent->folios, /* compressed_folios */
1237 async_extent->nr_folios,
1238 async_chunk->write_flags, true);
1239 *alloc_hint = ins.objectid + ins.offset;
1240 done:
1241 if (async_chunk->blkcg_css)
1242 kthread_associate_blkcg(NULL);
1243 kfree(async_extent);
1244 return;
1245
1246 out_free_reserve:
1247 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1248 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1249 mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1250 extent_clear_unlock_delalloc(inode, start, end,
1251 NULL, &cached,
1252 EXTENT_LOCKED | EXTENT_DELALLOC |
1253 EXTENT_DELALLOC_NEW |
1254 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1255 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1256 PAGE_END_WRITEBACK);
1257 free_async_extent_pages(async_extent);
1258 if (async_chunk->blkcg_css)
1259 kthread_associate_blkcg(NULL);
1260 btrfs_debug(fs_info,
1261 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1262 btrfs_root_id(root), btrfs_ino(inode), start,
1263 async_extent->ram_size, ret);
1264 kfree(async_extent);
1265 }
1266
get_extent_allocation_hint(struct btrfs_inode * inode,u64 start,u64 num_bytes)1267 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1268 u64 num_bytes)
1269 {
1270 struct extent_map_tree *em_tree = &inode->extent_tree;
1271 struct extent_map *em;
1272 u64 alloc_hint = 0;
1273
1274 read_lock(&em_tree->lock);
1275 em = search_extent_mapping(em_tree, start, num_bytes);
1276 if (em) {
1277 /*
1278 * if block start isn't an actual block number then find the
1279 * first block in this inode and use that as a hint. If that
1280 * block is also bogus then just don't worry about it.
1281 */
1282 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1283 free_extent_map(em);
1284 em = search_extent_mapping(em_tree, 0, 0);
1285 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1286 alloc_hint = em->block_start;
1287 if (em)
1288 free_extent_map(em);
1289 } else {
1290 alloc_hint = em->block_start;
1291 free_extent_map(em);
1292 }
1293 }
1294 read_unlock(&em_tree->lock);
1295
1296 return alloc_hint;
1297 }
1298
1299 /*
1300 * when extent_io.c finds a delayed allocation range in the file,
1301 * the call backs end up in this code. The basic idea is to
1302 * allocate extents on disk for the range, and create ordered data structs
1303 * in ram to track those extents.
1304 *
1305 * locked_page is the page that writepage had locked already. We use
1306 * it to make sure we don't do extra locks or unlocks.
1307 *
1308 * When this function fails, it unlocks all pages except @locked_page.
1309 *
1310 * When this function successfully creates an inline extent, it returns 1 and
1311 * unlocks all pages including locked_page and starts I/O on them.
1312 * (In reality inline extents are limited to a single page, so locked_page is
1313 * the only page handled anyway).
1314 *
1315 * When this function succeed and creates a normal extent, the page locking
1316 * status depends on the passed in flags:
1317 *
1318 * - If @keep_locked is set, all pages are kept locked.
1319 * - Else all pages except for @locked_page are unlocked.
1320 *
1321 * When a failure happens in the second or later iteration of the
1322 * while-loop, the ordered extents created in previous iterations are kept
1323 * intact. So, the caller must clean them up by calling
1324 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1325 * example.
1326 */
cow_file_range(struct btrfs_inode * inode,struct page * locked_page,u64 start,u64 end,u64 * done_offset,bool keep_locked,bool no_inline)1327 static noinline int cow_file_range(struct btrfs_inode *inode,
1328 struct page *locked_page, u64 start, u64 end,
1329 u64 *done_offset,
1330 bool keep_locked, bool no_inline)
1331 {
1332 struct btrfs_root *root = inode->root;
1333 struct btrfs_fs_info *fs_info = root->fs_info;
1334 struct extent_state *cached = NULL;
1335 u64 alloc_hint = 0;
1336 u64 orig_start = start;
1337 u64 num_bytes;
1338 unsigned long ram_size;
1339 u64 cur_alloc_size = 0;
1340 u64 min_alloc_size;
1341 u64 blocksize = fs_info->sectorsize;
1342 struct btrfs_key ins;
1343 struct extent_map *em;
1344 unsigned clear_bits;
1345 unsigned long page_ops;
1346 bool extent_reserved = false;
1347 int ret = 0;
1348
1349 if (btrfs_is_free_space_inode(inode)) {
1350 ret = -EINVAL;
1351 goto out_unlock;
1352 }
1353
1354 num_bytes = ALIGN(end - start + 1, blocksize);
1355 num_bytes = max(blocksize, num_bytes);
1356 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1357
1358 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1359
1360 if (!no_inline) {
1361 /* lets try to make an inline extent */
1362 ret = cow_file_range_inline(inode, start, end, 0,
1363 BTRFS_COMPRESS_NONE, NULL, false);
1364 if (ret <= 0) {
1365 /*
1366 * We succeeded, return 1 so the caller knows we're done
1367 * with this page and already handled the IO.
1368 *
1369 * If there was an error then cow_file_range_inline() has
1370 * already done the cleanup.
1371 */
1372 if (ret == 0)
1373 ret = 1;
1374 goto done;
1375 }
1376 }
1377
1378 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1379
1380 /*
1381 * Relocation relies on the relocated extents to have exactly the same
1382 * size as the original extents. Normally writeback for relocation data
1383 * extents follows a NOCOW path because relocation preallocates the
1384 * extents. However, due to an operation such as scrub turning a block
1385 * group to RO mode, it may fallback to COW mode, so we must make sure
1386 * an extent allocated during COW has exactly the requested size and can
1387 * not be split into smaller extents, otherwise relocation breaks and
1388 * fails during the stage where it updates the bytenr of file extent
1389 * items.
1390 */
1391 if (btrfs_is_data_reloc_root(root))
1392 min_alloc_size = num_bytes;
1393 else
1394 min_alloc_size = fs_info->sectorsize;
1395
1396 while (num_bytes > 0) {
1397 struct btrfs_ordered_extent *ordered;
1398
1399 cur_alloc_size = num_bytes;
1400 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1401 min_alloc_size, 0, alloc_hint,
1402 &ins, 1, 1);
1403 if (ret == -EAGAIN) {
1404 /*
1405 * btrfs_reserve_extent only returns -EAGAIN for zoned
1406 * file systems, which is an indication that there are
1407 * no active zones to allocate from at the moment.
1408 *
1409 * If this is the first loop iteration, wait for at
1410 * least one zone to finish before retrying the
1411 * allocation. Otherwise ask the caller to write out
1412 * the already allocated blocks before coming back to
1413 * us, or return -ENOSPC if it can't handle retries.
1414 */
1415 ASSERT(btrfs_is_zoned(fs_info));
1416 if (start == orig_start) {
1417 wait_on_bit_io(&inode->root->fs_info->flags,
1418 BTRFS_FS_NEED_ZONE_FINISH,
1419 TASK_UNINTERRUPTIBLE);
1420 continue;
1421 }
1422 if (done_offset) {
1423 *done_offset = start - 1;
1424 return 0;
1425 }
1426 ret = -ENOSPC;
1427 }
1428 if (ret < 0)
1429 goto out_unlock;
1430 cur_alloc_size = ins.offset;
1431 extent_reserved = true;
1432
1433 ram_size = ins.offset;
1434
1435 lock_extent(&inode->io_tree, start, start + ram_size - 1,
1436 &cached);
1437
1438 em = create_io_em(inode, start, ins.offset, /* len */
1439 start, /* orig_start */
1440 ins.objectid, /* block_start */
1441 ins.offset, /* block_len */
1442 ins.offset, /* orig_block_len */
1443 ram_size, /* ram_bytes */
1444 BTRFS_COMPRESS_NONE, /* compress_type */
1445 BTRFS_ORDERED_REGULAR /* type */);
1446 if (IS_ERR(em)) {
1447 unlock_extent(&inode->io_tree, start,
1448 start + ram_size - 1, &cached);
1449 ret = PTR_ERR(em);
1450 goto out_reserve;
1451 }
1452 free_extent_map(em);
1453
1454 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1455 ram_size, ins.objectid, cur_alloc_size,
1456 0, 1 << BTRFS_ORDERED_REGULAR,
1457 BTRFS_COMPRESS_NONE);
1458 if (IS_ERR(ordered)) {
1459 unlock_extent(&inode->io_tree, start,
1460 start + ram_size - 1, &cached);
1461 ret = PTR_ERR(ordered);
1462 goto out_drop_extent_cache;
1463 }
1464
1465 if (btrfs_is_data_reloc_root(root)) {
1466 ret = btrfs_reloc_clone_csums(ordered);
1467
1468 /*
1469 * Only drop cache here, and process as normal.
1470 *
1471 * We must not allow extent_clear_unlock_delalloc()
1472 * at out_unlock label to free meta of this ordered
1473 * extent, as its meta should be freed by
1474 * btrfs_finish_ordered_io().
1475 *
1476 * So we must continue until @start is increased to
1477 * skip current ordered extent.
1478 */
1479 if (ret)
1480 btrfs_drop_extent_map_range(inode, start,
1481 start + ram_size - 1,
1482 false);
1483 }
1484 btrfs_put_ordered_extent(ordered);
1485
1486 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1487
1488 /*
1489 * We're not doing compressed IO, don't unlock the first page
1490 * (which the caller expects to stay locked), don't clear any
1491 * dirty bits and don't set any writeback bits
1492 *
1493 * Do set the Ordered (Private2) bit so we know this page was
1494 * properly setup for writepage.
1495 */
1496 page_ops = (keep_locked ? 0 : PAGE_UNLOCK);
1497 page_ops |= PAGE_SET_ORDERED;
1498
1499 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1500 locked_page, &cached,
1501 EXTENT_LOCKED | EXTENT_DELALLOC,
1502 page_ops);
1503 if (num_bytes < cur_alloc_size)
1504 num_bytes = 0;
1505 else
1506 num_bytes -= cur_alloc_size;
1507 alloc_hint = ins.objectid + ins.offset;
1508 start += cur_alloc_size;
1509 extent_reserved = false;
1510
1511 /*
1512 * btrfs_reloc_clone_csums() error, since start is increased
1513 * extent_clear_unlock_delalloc() at out_unlock label won't
1514 * free metadata of current ordered extent, we're OK to exit.
1515 */
1516 if (ret)
1517 goto out_unlock;
1518 }
1519 done:
1520 if (done_offset)
1521 *done_offset = end;
1522 return ret;
1523
1524 out_drop_extent_cache:
1525 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1526 out_reserve:
1527 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1528 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1529 out_unlock:
1530 /*
1531 * Now, we have three regions to clean up:
1532 *
1533 * |-------(1)----|---(2)---|-------------(3)----------|
1534 * `- orig_start `- start `- start + cur_alloc_size `- end
1535 *
1536 * We process each region below.
1537 */
1538
1539 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1540 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1541 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1542
1543 /*
1544 * For the range (1). We have already instantiated the ordered extents
1545 * for this region. They are cleaned up by
1546 * btrfs_cleanup_ordered_extents() in e.g,
1547 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1548 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1549 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1550 * function.
1551 *
1552 * However, in case of @keep_locked, we still need to unlock the pages
1553 * (except @locked_page) to ensure all the pages are unlocked.
1554 */
1555 if (keep_locked && orig_start < start) {
1556 if (!locked_page)
1557 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1558 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1559 locked_page, NULL, 0, page_ops);
1560 }
1561
1562 /*
1563 * At this point we're unlocked, we want to make sure we're only
1564 * clearing these flags under the extent lock, so lock the rest of the
1565 * range and clear everything up.
1566 */
1567 lock_extent(&inode->io_tree, start, end, NULL);
1568
1569 /*
1570 * For the range (2). If we reserved an extent for our delalloc range
1571 * (or a subrange) and failed to create the respective ordered extent,
1572 * then it means that when we reserved the extent we decremented the
1573 * extent's size from the data space_info's bytes_may_use counter and
1574 * incremented the space_info's bytes_reserved counter by the same
1575 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1576 * to decrement again the data space_info's bytes_may_use counter,
1577 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1578 */
1579 if (extent_reserved) {
1580 extent_clear_unlock_delalloc(inode, start,
1581 start + cur_alloc_size - 1,
1582 locked_page, &cached,
1583 clear_bits,
1584 page_ops);
1585 start += cur_alloc_size;
1586 }
1587
1588 /*
1589 * For the range (3). We never touched the region. In addition to the
1590 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1591 * space_info's bytes_may_use counter, reserved in
1592 * btrfs_check_data_free_space().
1593 */
1594 if (start < end) {
1595 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1596 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1597 &cached, clear_bits, page_ops);
1598 }
1599 return ret;
1600 }
1601
1602 /*
1603 * Phase two of compressed writeback. This is the ordered portion of the code,
1604 * which only gets called in the order the work was queued. We walk all the
1605 * async extents created by compress_file_range and send them down to the disk.
1606 *
1607 * If called with @do_free == true then it'll try to finish the work and free
1608 * the work struct eventually.
1609 */
submit_compressed_extents(struct btrfs_work * work,bool do_free)1610 static noinline void submit_compressed_extents(struct btrfs_work *work, bool do_free)
1611 {
1612 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1613 work);
1614 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1615 struct async_extent *async_extent;
1616 unsigned long nr_pages;
1617 u64 alloc_hint = 0;
1618
1619 if (do_free) {
1620 struct async_chunk *async_chunk;
1621 struct async_cow *async_cow;
1622
1623 async_chunk = container_of(work, struct async_chunk, work);
1624 btrfs_add_delayed_iput(async_chunk->inode);
1625 if (async_chunk->blkcg_css)
1626 css_put(async_chunk->blkcg_css);
1627
1628 async_cow = async_chunk->async_cow;
1629 if (atomic_dec_and_test(&async_cow->num_chunks))
1630 kvfree(async_cow);
1631 return;
1632 }
1633
1634 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1635 PAGE_SHIFT;
1636
1637 while (!list_empty(&async_chunk->extents)) {
1638 async_extent = list_entry(async_chunk->extents.next,
1639 struct async_extent, list);
1640 list_del(&async_extent->list);
1641 submit_one_async_extent(async_chunk, async_extent, &alloc_hint);
1642 }
1643
1644 /* atomic_sub_return implies a barrier */
1645 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1646 5 * SZ_1M)
1647 cond_wake_up_nomb(&fs_info->async_submit_wait);
1648 }
1649
run_delalloc_compressed(struct btrfs_inode * inode,struct page * locked_page,u64 start,u64 end,struct writeback_control * wbc)1650 static bool run_delalloc_compressed(struct btrfs_inode *inode,
1651 struct page *locked_page, u64 start,
1652 u64 end, struct writeback_control *wbc)
1653 {
1654 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1655 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1656 struct async_cow *ctx;
1657 struct async_chunk *async_chunk;
1658 unsigned long nr_pages;
1659 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1660 int i;
1661 unsigned nofs_flag;
1662 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1663
1664 nofs_flag = memalloc_nofs_save();
1665 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1666 memalloc_nofs_restore(nofs_flag);
1667 if (!ctx)
1668 return false;
1669
1670 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1671
1672 async_chunk = ctx->chunks;
1673 atomic_set(&ctx->num_chunks, num_chunks);
1674
1675 for (i = 0; i < num_chunks; i++) {
1676 u64 cur_end = min(end, start + SZ_512K - 1);
1677
1678 /*
1679 * igrab is called higher up in the call chain, take only the
1680 * lightweight reference for the callback lifetime
1681 */
1682 ihold(&inode->vfs_inode);
1683 async_chunk[i].async_cow = ctx;
1684 async_chunk[i].inode = inode;
1685 async_chunk[i].start = start;
1686 async_chunk[i].end = cur_end;
1687 async_chunk[i].write_flags = write_flags;
1688 INIT_LIST_HEAD(&async_chunk[i].extents);
1689
1690 /*
1691 * The locked_page comes all the way from writepage and its
1692 * the original page we were actually given. As we spread
1693 * this large delalloc region across multiple async_chunk
1694 * structs, only the first struct needs a pointer to locked_page
1695 *
1696 * This way we don't need racey decisions about who is supposed
1697 * to unlock it.
1698 */
1699 if (locked_page) {
1700 /*
1701 * Depending on the compressibility, the pages might or
1702 * might not go through async. We want all of them to
1703 * be accounted against wbc once. Let's do it here
1704 * before the paths diverge. wbc accounting is used
1705 * only for foreign writeback detection and doesn't
1706 * need full accuracy. Just account the whole thing
1707 * against the first page.
1708 */
1709 wbc_account_cgroup_owner(wbc, locked_page,
1710 cur_end - start);
1711 async_chunk[i].locked_page = locked_page;
1712 locked_page = NULL;
1713 } else {
1714 async_chunk[i].locked_page = NULL;
1715 }
1716
1717 if (blkcg_css != blkcg_root_css) {
1718 css_get(blkcg_css);
1719 async_chunk[i].blkcg_css = blkcg_css;
1720 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1721 } else {
1722 async_chunk[i].blkcg_css = NULL;
1723 }
1724
1725 btrfs_init_work(&async_chunk[i].work, compress_file_range,
1726 submit_compressed_extents);
1727
1728 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1729 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1730
1731 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1732
1733 start = cur_end + 1;
1734 }
1735 return true;
1736 }
1737
1738 /*
1739 * Run the delalloc range from start to end, and write back any dirty pages
1740 * covered by the range.
1741 */
run_delalloc_cow(struct btrfs_inode * inode,struct page * locked_page,u64 start,u64 end,struct writeback_control * wbc,bool pages_dirty)1742 static noinline int run_delalloc_cow(struct btrfs_inode *inode,
1743 struct page *locked_page, u64 start,
1744 u64 end, struct writeback_control *wbc,
1745 bool pages_dirty)
1746 {
1747 u64 done_offset = end;
1748 int ret;
1749
1750 while (start <= end) {
1751 ret = cow_file_range(inode, locked_page, start, end, &done_offset,
1752 true, false);
1753 if (ret)
1754 return ret;
1755 extent_write_locked_range(&inode->vfs_inode, locked_page, start,
1756 done_offset, wbc, pages_dirty);
1757 start = done_offset + 1;
1758 }
1759
1760 return 1;
1761 }
1762
fallback_to_cow(struct btrfs_inode * inode,struct page * locked_page,const u64 start,const u64 end)1763 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1764 const u64 start, const u64 end)
1765 {
1766 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1767 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1768 const u64 range_bytes = end + 1 - start;
1769 struct extent_io_tree *io_tree = &inode->io_tree;
1770 struct extent_state *cached_state = NULL;
1771 u64 range_start = start;
1772 u64 count;
1773 int ret;
1774
1775 /*
1776 * If EXTENT_NORESERVE is set it means that when the buffered write was
1777 * made we had not enough available data space and therefore we did not
1778 * reserve data space for it, since we though we could do NOCOW for the
1779 * respective file range (either there is prealloc extent or the inode
1780 * has the NOCOW bit set).
1781 *
1782 * However when we need to fallback to COW mode (because for example the
1783 * block group for the corresponding extent was turned to RO mode by a
1784 * scrub or relocation) we need to do the following:
1785 *
1786 * 1) We increment the bytes_may_use counter of the data space info.
1787 * If COW succeeds, it allocates a new data extent and after doing
1788 * that it decrements the space info's bytes_may_use counter and
1789 * increments its bytes_reserved counter by the same amount (we do
1790 * this at btrfs_add_reserved_bytes()). So we need to increment the
1791 * bytes_may_use counter to compensate (when space is reserved at
1792 * buffered write time, the bytes_may_use counter is incremented);
1793 *
1794 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1795 * that if the COW path fails for any reason, it decrements (through
1796 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1797 * data space info, which we incremented in the step above.
1798 *
1799 * If we need to fallback to cow and the inode corresponds to a free
1800 * space cache inode or an inode of the data relocation tree, we must
1801 * also increment bytes_may_use of the data space_info for the same
1802 * reason. Space caches and relocated data extents always get a prealloc
1803 * extent for them, however scrub or balance may have set the block
1804 * group that contains that extent to RO mode and therefore force COW
1805 * when starting writeback.
1806 */
1807 lock_extent(io_tree, start, end, &cached_state);
1808 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1809 EXTENT_NORESERVE, 0, NULL);
1810 if (count > 0 || is_space_ino || is_reloc_ino) {
1811 u64 bytes = count;
1812 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1813 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1814
1815 if (is_space_ino || is_reloc_ino)
1816 bytes = range_bytes;
1817
1818 spin_lock(&sinfo->lock);
1819 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1820 spin_unlock(&sinfo->lock);
1821
1822 if (count > 0)
1823 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1824 NULL);
1825 }
1826 unlock_extent(io_tree, start, end, &cached_state);
1827
1828 /*
1829 * Don't try to create inline extents, as a mix of inline extent that
1830 * is written out and unlocked directly and a normal NOCOW extent
1831 * doesn't work.
1832 */
1833 ret = cow_file_range(inode, locked_page, start, end, NULL, false, true);
1834 ASSERT(ret != 1);
1835 return ret;
1836 }
1837
1838 struct can_nocow_file_extent_args {
1839 /* Input fields. */
1840
1841 /* Start file offset of the range we want to NOCOW. */
1842 u64 start;
1843 /* End file offset (inclusive) of the range we want to NOCOW. */
1844 u64 end;
1845 bool writeback_path;
1846 bool strict;
1847 /*
1848 * Free the path passed to can_nocow_file_extent() once it's not needed
1849 * anymore.
1850 */
1851 bool free_path;
1852
1853 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1854
1855 u64 disk_bytenr;
1856 u64 disk_num_bytes;
1857 u64 extent_offset;
1858 /* Number of bytes that can be written to in NOCOW mode. */
1859 u64 num_bytes;
1860 };
1861
1862 /*
1863 * Check if we can NOCOW the file extent that the path points to.
1864 * This function may return with the path released, so the caller should check
1865 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1866 *
1867 * Returns: < 0 on error
1868 * 0 if we can not NOCOW
1869 * 1 if we can NOCOW
1870 */
can_nocow_file_extent(struct btrfs_path * path,struct btrfs_key * key,struct btrfs_inode * inode,struct can_nocow_file_extent_args * args)1871 static int can_nocow_file_extent(struct btrfs_path *path,
1872 struct btrfs_key *key,
1873 struct btrfs_inode *inode,
1874 struct can_nocow_file_extent_args *args)
1875 {
1876 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1877 struct extent_buffer *leaf = path->nodes[0];
1878 struct btrfs_root *root = inode->root;
1879 struct btrfs_file_extent_item *fi;
1880 struct btrfs_root *csum_root;
1881 u64 extent_end;
1882 u8 extent_type;
1883 int can_nocow = 0;
1884 int ret = 0;
1885 bool nowait = path->nowait;
1886
1887 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1888 extent_type = btrfs_file_extent_type(leaf, fi);
1889
1890 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1891 goto out;
1892
1893 /* Can't access these fields unless we know it's not an inline extent. */
1894 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1895 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1896 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1897
1898 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1899 extent_type == BTRFS_FILE_EXTENT_REG)
1900 goto out;
1901
1902 /*
1903 * If the extent was created before the generation where the last snapshot
1904 * for its subvolume was created, then this implies the extent is shared,
1905 * hence we must COW.
1906 */
1907 if (!args->strict &&
1908 btrfs_file_extent_generation(leaf, fi) <=
1909 btrfs_root_last_snapshot(&root->root_item))
1910 goto out;
1911
1912 /* An explicit hole, must COW. */
1913 if (args->disk_bytenr == 0)
1914 goto out;
1915
1916 /* Compressed/encrypted/encoded extents must be COWed. */
1917 if (btrfs_file_extent_compression(leaf, fi) ||
1918 btrfs_file_extent_encryption(leaf, fi) ||
1919 btrfs_file_extent_other_encoding(leaf, fi))
1920 goto out;
1921
1922 extent_end = btrfs_file_extent_end(path);
1923
1924 /*
1925 * The following checks can be expensive, as they need to take other
1926 * locks and do btree or rbtree searches, so release the path to avoid
1927 * blocking other tasks for too long.
1928 */
1929 btrfs_release_path(path);
1930
1931 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1932 key->offset - args->extent_offset,
1933 args->disk_bytenr, args->strict, path);
1934 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1935 if (ret != 0)
1936 goto out;
1937
1938 if (args->free_path) {
1939 /*
1940 * We don't need the path anymore, plus through the
1941 * btrfs_lookup_csums_list() call below we will end up allocating
1942 * another path. So free the path to avoid unnecessary extra
1943 * memory usage.
1944 */
1945 btrfs_free_path(path);
1946 path = NULL;
1947 }
1948
1949 /* If there are pending snapshots for this root, we must COW. */
1950 if (args->writeback_path && !is_freespace_inode &&
1951 atomic_read(&root->snapshot_force_cow))
1952 goto out;
1953
1954 args->disk_bytenr += args->extent_offset;
1955 args->disk_bytenr += args->start - key->offset;
1956 args->num_bytes = min(args->end + 1, extent_end) - args->start;
1957
1958 /*
1959 * Force COW if csums exist in the range. This ensures that csums for a
1960 * given extent are either valid or do not exist.
1961 */
1962
1963 csum_root = btrfs_csum_root(root->fs_info, args->disk_bytenr);
1964 ret = btrfs_lookup_csums_list(csum_root, args->disk_bytenr,
1965 args->disk_bytenr + args->num_bytes - 1,
1966 NULL, nowait);
1967 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1968 if (ret != 0)
1969 goto out;
1970
1971 can_nocow = 1;
1972 out:
1973 if (args->free_path && path)
1974 btrfs_free_path(path);
1975
1976 return ret < 0 ? ret : can_nocow;
1977 }
1978
1979 /*
1980 * when nowcow writeback call back. This checks for snapshots or COW copies
1981 * of the extents that exist in the file, and COWs the file as required.
1982 *
1983 * If no cow copies or snapshots exist, we write directly to the existing
1984 * blocks on disk
1985 */
run_delalloc_nocow(struct btrfs_inode * inode,struct page * locked_page,const u64 start,const u64 end)1986 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1987 struct page *locked_page,
1988 const u64 start, const u64 end)
1989 {
1990 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1991 struct btrfs_root *root = inode->root;
1992 struct btrfs_path *path;
1993 u64 cow_start = (u64)-1;
1994 u64 cur_offset = start;
1995 int ret;
1996 bool check_prev = true;
1997 u64 ino = btrfs_ino(inode);
1998 struct can_nocow_file_extent_args nocow_args = { 0 };
1999
2000 /*
2001 * Normally on a zoned device we're only doing COW writes, but in case
2002 * of relocation on a zoned filesystem serializes I/O so that we're only
2003 * writing sequentially and can end up here as well.
2004 */
2005 ASSERT(!btrfs_is_zoned(fs_info) || btrfs_is_data_reloc_root(root));
2006
2007 path = btrfs_alloc_path();
2008 if (!path) {
2009 ret = -ENOMEM;
2010 goto error;
2011 }
2012
2013 nocow_args.end = end;
2014 nocow_args.writeback_path = true;
2015
2016 while (cur_offset <= end) {
2017 struct btrfs_block_group *nocow_bg = NULL;
2018 struct btrfs_ordered_extent *ordered;
2019 struct btrfs_key found_key;
2020 struct btrfs_file_extent_item *fi;
2021 struct extent_buffer *leaf;
2022 struct extent_state *cached_state = NULL;
2023 u64 extent_end;
2024 u64 ram_bytes;
2025 u64 nocow_end;
2026 int extent_type;
2027 bool is_prealloc;
2028
2029 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2030 cur_offset, 0);
2031 if (ret < 0)
2032 goto error;
2033
2034 /*
2035 * If there is no extent for our range when doing the initial
2036 * search, then go back to the previous slot as it will be the
2037 * one containing the search offset
2038 */
2039 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2040 leaf = path->nodes[0];
2041 btrfs_item_key_to_cpu(leaf, &found_key,
2042 path->slots[0] - 1);
2043 if (found_key.objectid == ino &&
2044 found_key.type == BTRFS_EXTENT_DATA_KEY)
2045 path->slots[0]--;
2046 }
2047 check_prev = false;
2048 next_slot:
2049 /* Go to next leaf if we have exhausted the current one */
2050 leaf = path->nodes[0];
2051 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2052 ret = btrfs_next_leaf(root, path);
2053 if (ret < 0)
2054 goto error;
2055 if (ret > 0)
2056 break;
2057 leaf = path->nodes[0];
2058 }
2059
2060 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2061
2062 /* Didn't find anything for our INO */
2063 if (found_key.objectid > ino)
2064 break;
2065 /*
2066 * Keep searching until we find an EXTENT_ITEM or there are no
2067 * more extents for this inode
2068 */
2069 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2070 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2071 path->slots[0]++;
2072 goto next_slot;
2073 }
2074
2075 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2076 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2077 found_key.offset > end)
2078 break;
2079
2080 /*
2081 * If the found extent starts after requested offset, then
2082 * adjust extent_end to be right before this extent begins
2083 */
2084 if (found_key.offset > cur_offset) {
2085 extent_end = found_key.offset;
2086 extent_type = 0;
2087 goto must_cow;
2088 }
2089
2090 /*
2091 * Found extent which begins before our range and potentially
2092 * intersect it
2093 */
2094 fi = btrfs_item_ptr(leaf, path->slots[0],
2095 struct btrfs_file_extent_item);
2096 extent_type = btrfs_file_extent_type(leaf, fi);
2097 /* If this is triggered then we have a memory corruption. */
2098 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2099 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2100 ret = -EUCLEAN;
2101 goto error;
2102 }
2103 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2104 extent_end = btrfs_file_extent_end(path);
2105
2106 /*
2107 * If the extent we got ends before our current offset, skip to
2108 * the next extent.
2109 */
2110 if (extent_end <= cur_offset) {
2111 path->slots[0]++;
2112 goto next_slot;
2113 }
2114
2115 nocow_args.start = cur_offset;
2116 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2117 if (ret < 0)
2118 goto error;
2119 if (ret == 0)
2120 goto must_cow;
2121
2122 ret = 0;
2123 nocow_bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2124 if (!nocow_bg) {
2125 must_cow:
2126 /*
2127 * If we can't perform NOCOW writeback for the range,
2128 * then record the beginning of the range that needs to
2129 * be COWed. It will be written out before the next
2130 * NOCOW range if we find one, or when exiting this
2131 * loop.
2132 */
2133 if (cow_start == (u64)-1)
2134 cow_start = cur_offset;
2135 cur_offset = extent_end;
2136 if (cur_offset > end)
2137 break;
2138 if (!path->nodes[0])
2139 continue;
2140 path->slots[0]++;
2141 goto next_slot;
2142 }
2143
2144 /*
2145 * COW range from cow_start to found_key.offset - 1. As the key
2146 * will contain the beginning of the first extent that can be
2147 * NOCOW, following one which needs to be COW'ed
2148 */
2149 if (cow_start != (u64)-1) {
2150 ret = fallback_to_cow(inode, locked_page,
2151 cow_start, found_key.offset - 1);
2152 cow_start = (u64)-1;
2153 if (ret) {
2154 btrfs_dec_nocow_writers(nocow_bg);
2155 goto error;
2156 }
2157 }
2158
2159 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2160 lock_extent(&inode->io_tree, cur_offset, nocow_end, &cached_state);
2161
2162 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2163 if (is_prealloc) {
2164 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2165 struct extent_map *em;
2166
2167 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2168 orig_start,
2169 nocow_args.disk_bytenr, /* block_start */
2170 nocow_args.num_bytes, /* block_len */
2171 nocow_args.disk_num_bytes, /* orig_block_len */
2172 ram_bytes, BTRFS_COMPRESS_NONE,
2173 BTRFS_ORDERED_PREALLOC);
2174 if (IS_ERR(em)) {
2175 unlock_extent(&inode->io_tree, cur_offset,
2176 nocow_end, &cached_state);
2177 btrfs_dec_nocow_writers(nocow_bg);
2178 ret = PTR_ERR(em);
2179 goto error;
2180 }
2181 free_extent_map(em);
2182 }
2183
2184 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2185 nocow_args.num_bytes, nocow_args.num_bytes,
2186 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2187 is_prealloc
2188 ? (1 << BTRFS_ORDERED_PREALLOC)
2189 : (1 << BTRFS_ORDERED_NOCOW),
2190 BTRFS_COMPRESS_NONE);
2191 btrfs_dec_nocow_writers(nocow_bg);
2192 if (IS_ERR(ordered)) {
2193 if (is_prealloc) {
2194 btrfs_drop_extent_map_range(inode, cur_offset,
2195 nocow_end, false);
2196 }
2197 unlock_extent(&inode->io_tree, cur_offset,
2198 nocow_end, &cached_state);
2199 ret = PTR_ERR(ordered);
2200 goto error;
2201 }
2202
2203 if (btrfs_is_data_reloc_root(root))
2204 /*
2205 * Error handled later, as we must prevent
2206 * extent_clear_unlock_delalloc() in error handler
2207 * from freeing metadata of created ordered extent.
2208 */
2209 ret = btrfs_reloc_clone_csums(ordered);
2210 btrfs_put_ordered_extent(ordered);
2211
2212 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2213 locked_page, &cached_state,
2214 EXTENT_LOCKED | EXTENT_DELALLOC |
2215 EXTENT_CLEAR_DATA_RESV,
2216 PAGE_UNLOCK | PAGE_SET_ORDERED);
2217
2218 cur_offset = extent_end;
2219
2220 /*
2221 * btrfs_reloc_clone_csums() error, now we're OK to call error
2222 * handler, as metadata for created ordered extent will only
2223 * be freed by btrfs_finish_ordered_io().
2224 */
2225 if (ret)
2226 goto error;
2227 }
2228 btrfs_release_path(path);
2229
2230 if (cur_offset <= end && cow_start == (u64)-1)
2231 cow_start = cur_offset;
2232
2233 if (cow_start != (u64)-1) {
2234 cur_offset = end;
2235 ret = fallback_to_cow(inode, locked_page, cow_start, end);
2236 cow_start = (u64)-1;
2237 if (ret)
2238 goto error;
2239 }
2240
2241 btrfs_free_path(path);
2242 return 0;
2243
2244 error:
2245 /*
2246 * If an error happened while a COW region is outstanding, cur_offset
2247 * needs to be reset to cow_start to ensure the COW region is unlocked
2248 * as well.
2249 */
2250 if (cow_start != (u64)-1)
2251 cur_offset = cow_start;
2252
2253 /*
2254 * We need to lock the extent here because we're clearing DELALLOC and
2255 * we're not locked at this point.
2256 */
2257 if (cur_offset < end) {
2258 struct extent_state *cached = NULL;
2259
2260 lock_extent(&inode->io_tree, cur_offset, end, &cached);
2261 extent_clear_unlock_delalloc(inode, cur_offset, end,
2262 locked_page, &cached,
2263 EXTENT_LOCKED | EXTENT_DELALLOC |
2264 EXTENT_DEFRAG |
2265 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2266 PAGE_START_WRITEBACK |
2267 PAGE_END_WRITEBACK);
2268 }
2269 btrfs_free_path(path);
2270 return ret;
2271 }
2272
should_nocow(struct btrfs_inode * inode,u64 start,u64 end)2273 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2274 {
2275 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2276 if (inode->defrag_bytes &&
2277 test_range_bit_exists(&inode->io_tree, start, end, EXTENT_DEFRAG))
2278 return false;
2279 return true;
2280 }
2281 return false;
2282 }
2283
2284 /*
2285 * Function to process delayed allocation (create CoW) for ranges which are
2286 * being touched for the first time.
2287 */
btrfs_run_delalloc_range(struct btrfs_inode * inode,struct page * locked_page,u64 start,u64 end,struct writeback_control * wbc)2288 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2289 u64 start, u64 end, struct writeback_control *wbc)
2290 {
2291 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2292 int ret;
2293
2294 /*
2295 * The range must cover part of the @locked_page, or a return of 1
2296 * can confuse the caller.
2297 */
2298 ASSERT(!(end <= page_offset(locked_page) ||
2299 start >= page_offset(locked_page) + PAGE_SIZE));
2300
2301 if (should_nocow(inode, start, end)) {
2302 ret = run_delalloc_nocow(inode, locked_page, start, end);
2303 goto out;
2304 }
2305
2306 if (btrfs_inode_can_compress(inode) &&
2307 inode_need_compress(inode, start, end) &&
2308 run_delalloc_compressed(inode, locked_page, start, end, wbc))
2309 return 1;
2310
2311 if (zoned)
2312 ret = run_delalloc_cow(inode, locked_page, start, end, wbc,
2313 true);
2314 else
2315 ret = cow_file_range(inode, locked_page, start, end, NULL,
2316 false, false);
2317
2318 out:
2319 if (ret < 0)
2320 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2321 end - start + 1);
2322 return ret;
2323 }
2324
btrfs_split_delalloc_extent(struct btrfs_inode * inode,struct extent_state * orig,u64 split)2325 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2326 struct extent_state *orig, u64 split)
2327 {
2328 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2329 u64 size;
2330
2331 lockdep_assert_held(&inode->io_tree.lock);
2332
2333 /* not delalloc, ignore it */
2334 if (!(orig->state & EXTENT_DELALLOC))
2335 return;
2336
2337 size = orig->end - orig->start + 1;
2338 if (size > fs_info->max_extent_size) {
2339 u32 num_extents;
2340 u64 new_size;
2341
2342 /*
2343 * See the explanation in btrfs_merge_delalloc_extent, the same
2344 * applies here, just in reverse.
2345 */
2346 new_size = orig->end - split + 1;
2347 num_extents = count_max_extents(fs_info, new_size);
2348 new_size = split - orig->start;
2349 num_extents += count_max_extents(fs_info, new_size);
2350 if (count_max_extents(fs_info, size) >= num_extents)
2351 return;
2352 }
2353
2354 spin_lock(&inode->lock);
2355 btrfs_mod_outstanding_extents(inode, 1);
2356 spin_unlock(&inode->lock);
2357 }
2358
2359 /*
2360 * Handle merged delayed allocation extents so we can keep track of new extents
2361 * that are just merged onto old extents, such as when we are doing sequential
2362 * writes, so we can properly account for the metadata space we'll need.
2363 */
btrfs_merge_delalloc_extent(struct btrfs_inode * inode,struct extent_state * new,struct extent_state * other)2364 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2365 struct extent_state *other)
2366 {
2367 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2368 u64 new_size, old_size;
2369 u32 num_extents;
2370
2371 lockdep_assert_held(&inode->io_tree.lock);
2372
2373 /* not delalloc, ignore it */
2374 if (!(other->state & EXTENT_DELALLOC))
2375 return;
2376
2377 if (new->start > other->start)
2378 new_size = new->end - other->start + 1;
2379 else
2380 new_size = other->end - new->start + 1;
2381
2382 /* we're not bigger than the max, unreserve the space and go */
2383 if (new_size <= fs_info->max_extent_size) {
2384 spin_lock(&inode->lock);
2385 btrfs_mod_outstanding_extents(inode, -1);
2386 spin_unlock(&inode->lock);
2387 return;
2388 }
2389
2390 /*
2391 * We have to add up either side to figure out how many extents were
2392 * accounted for before we merged into one big extent. If the number of
2393 * extents we accounted for is <= the amount we need for the new range
2394 * then we can return, otherwise drop. Think of it like this
2395 *
2396 * [ 4k][MAX_SIZE]
2397 *
2398 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2399 * need 2 outstanding extents, on one side we have 1 and the other side
2400 * we have 1 so they are == and we can return. But in this case
2401 *
2402 * [MAX_SIZE+4k][MAX_SIZE+4k]
2403 *
2404 * Each range on their own accounts for 2 extents, but merged together
2405 * they are only 3 extents worth of accounting, so we need to drop in
2406 * this case.
2407 */
2408 old_size = other->end - other->start + 1;
2409 num_extents = count_max_extents(fs_info, old_size);
2410 old_size = new->end - new->start + 1;
2411 num_extents += count_max_extents(fs_info, old_size);
2412 if (count_max_extents(fs_info, new_size) >= num_extents)
2413 return;
2414
2415 spin_lock(&inode->lock);
2416 btrfs_mod_outstanding_extents(inode, -1);
2417 spin_unlock(&inode->lock);
2418 }
2419
btrfs_add_delalloc_inode(struct btrfs_inode * inode)2420 static void btrfs_add_delalloc_inode(struct btrfs_inode *inode)
2421 {
2422 struct btrfs_root *root = inode->root;
2423 struct btrfs_fs_info *fs_info = root->fs_info;
2424
2425 spin_lock(&root->delalloc_lock);
2426 ASSERT(list_empty(&inode->delalloc_inodes));
2427 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2428 root->nr_delalloc_inodes++;
2429 if (root->nr_delalloc_inodes == 1) {
2430 spin_lock(&fs_info->delalloc_root_lock);
2431 ASSERT(list_empty(&root->delalloc_root));
2432 list_add_tail(&root->delalloc_root, &fs_info->delalloc_roots);
2433 spin_unlock(&fs_info->delalloc_root_lock);
2434 }
2435 spin_unlock(&root->delalloc_lock);
2436 }
2437
btrfs_del_delalloc_inode(struct btrfs_inode * inode)2438 void btrfs_del_delalloc_inode(struct btrfs_inode *inode)
2439 {
2440 struct btrfs_root *root = inode->root;
2441 struct btrfs_fs_info *fs_info = root->fs_info;
2442
2443 lockdep_assert_held(&root->delalloc_lock);
2444
2445 /*
2446 * We may be called after the inode was already deleted from the list,
2447 * namely in the transaction abort path btrfs_destroy_delalloc_inodes(),
2448 * and then later through btrfs_clear_delalloc_extent() while the inode
2449 * still has ->delalloc_bytes > 0.
2450 */
2451 if (!list_empty(&inode->delalloc_inodes)) {
2452 list_del_init(&inode->delalloc_inodes);
2453 root->nr_delalloc_inodes--;
2454 if (!root->nr_delalloc_inodes) {
2455 ASSERT(list_empty(&root->delalloc_inodes));
2456 spin_lock(&fs_info->delalloc_root_lock);
2457 ASSERT(!list_empty(&root->delalloc_root));
2458 list_del_init(&root->delalloc_root);
2459 spin_unlock(&fs_info->delalloc_root_lock);
2460 }
2461 }
2462 }
2463
2464 /*
2465 * Properly track delayed allocation bytes in the inode and to maintain the
2466 * list of inodes that have pending delalloc work to be done.
2467 */
btrfs_set_delalloc_extent(struct btrfs_inode * inode,struct extent_state * state,u32 bits)2468 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2469 u32 bits)
2470 {
2471 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2472
2473 lockdep_assert_held(&inode->io_tree.lock);
2474
2475 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2476 WARN_ON(1);
2477 /*
2478 * set_bit and clear bit hooks normally require _irqsave/restore
2479 * but in this case, we are only testing for the DELALLOC
2480 * bit, which is only set or cleared with irqs on
2481 */
2482 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2483 u64 len = state->end + 1 - state->start;
2484 u64 prev_delalloc_bytes;
2485 u32 num_extents = count_max_extents(fs_info, len);
2486
2487 spin_lock(&inode->lock);
2488 btrfs_mod_outstanding_extents(inode, num_extents);
2489 spin_unlock(&inode->lock);
2490
2491 /* For sanity tests */
2492 if (btrfs_is_testing(fs_info))
2493 return;
2494
2495 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2496 fs_info->delalloc_batch);
2497 spin_lock(&inode->lock);
2498 prev_delalloc_bytes = inode->delalloc_bytes;
2499 inode->delalloc_bytes += len;
2500 if (bits & EXTENT_DEFRAG)
2501 inode->defrag_bytes += len;
2502 spin_unlock(&inode->lock);
2503
2504 /*
2505 * We don't need to be under the protection of the inode's lock,
2506 * because we are called while holding the inode's io_tree lock
2507 * and are therefore protected against concurrent calls of this
2508 * function and btrfs_clear_delalloc_extent().
2509 */
2510 if (!btrfs_is_free_space_inode(inode) && prev_delalloc_bytes == 0)
2511 btrfs_add_delalloc_inode(inode);
2512 }
2513
2514 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2515 (bits & EXTENT_DELALLOC_NEW)) {
2516 spin_lock(&inode->lock);
2517 inode->new_delalloc_bytes += state->end + 1 - state->start;
2518 spin_unlock(&inode->lock);
2519 }
2520 }
2521
2522 /*
2523 * Once a range is no longer delalloc this function ensures that proper
2524 * accounting happens.
2525 */
btrfs_clear_delalloc_extent(struct btrfs_inode * inode,struct extent_state * state,u32 bits)2526 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2527 struct extent_state *state, u32 bits)
2528 {
2529 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2530 u64 len = state->end + 1 - state->start;
2531 u32 num_extents = count_max_extents(fs_info, len);
2532
2533 lockdep_assert_held(&inode->io_tree.lock);
2534
2535 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2536 spin_lock(&inode->lock);
2537 inode->defrag_bytes -= len;
2538 spin_unlock(&inode->lock);
2539 }
2540
2541 /*
2542 * set_bit and clear bit hooks normally require _irqsave/restore
2543 * but in this case, we are only testing for the DELALLOC
2544 * bit, which is only set or cleared with irqs on
2545 */
2546 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2547 struct btrfs_root *root = inode->root;
2548 u64 new_delalloc_bytes;
2549
2550 spin_lock(&inode->lock);
2551 btrfs_mod_outstanding_extents(inode, -num_extents);
2552 spin_unlock(&inode->lock);
2553
2554 /*
2555 * We don't reserve metadata space for space cache inodes so we
2556 * don't need to call delalloc_release_metadata if there is an
2557 * error.
2558 */
2559 if (bits & EXTENT_CLEAR_META_RESV &&
2560 root != fs_info->tree_root)
2561 btrfs_delalloc_release_metadata(inode, len, true);
2562
2563 /* For sanity tests. */
2564 if (btrfs_is_testing(fs_info))
2565 return;
2566
2567 if (!btrfs_is_data_reloc_root(root) &&
2568 !btrfs_is_free_space_inode(inode) &&
2569 !(state->state & EXTENT_NORESERVE) &&
2570 (bits & EXTENT_CLEAR_DATA_RESV))
2571 btrfs_free_reserved_data_space_noquota(fs_info, len);
2572
2573 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2574 fs_info->delalloc_batch);
2575 spin_lock(&inode->lock);
2576 inode->delalloc_bytes -= len;
2577 new_delalloc_bytes = inode->delalloc_bytes;
2578 spin_unlock(&inode->lock);
2579
2580 /*
2581 * We don't need to be under the protection of the inode's lock,
2582 * because we are called while holding the inode's io_tree lock
2583 * and are therefore protected against concurrent calls of this
2584 * function and btrfs_set_delalloc_extent().
2585 */
2586 if (!btrfs_is_free_space_inode(inode) && new_delalloc_bytes == 0) {
2587 spin_lock(&root->delalloc_lock);
2588 btrfs_del_delalloc_inode(inode);
2589 spin_unlock(&root->delalloc_lock);
2590 }
2591 }
2592
2593 if ((state->state & EXTENT_DELALLOC_NEW) &&
2594 (bits & EXTENT_DELALLOC_NEW)) {
2595 spin_lock(&inode->lock);
2596 ASSERT(inode->new_delalloc_bytes >= len);
2597 inode->new_delalloc_bytes -= len;
2598 if (bits & EXTENT_ADD_INODE_BYTES)
2599 inode_add_bytes(&inode->vfs_inode, len);
2600 spin_unlock(&inode->lock);
2601 }
2602 }
2603
btrfs_extract_ordered_extent(struct btrfs_bio * bbio,struct btrfs_ordered_extent * ordered)2604 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2605 struct btrfs_ordered_extent *ordered)
2606 {
2607 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2608 u64 len = bbio->bio.bi_iter.bi_size;
2609 struct btrfs_ordered_extent *new;
2610 int ret;
2611
2612 /* Must always be called for the beginning of an ordered extent. */
2613 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2614 return -EINVAL;
2615
2616 /* No need to split if the ordered extent covers the entire bio. */
2617 if (ordered->disk_num_bytes == len) {
2618 refcount_inc(&ordered->refs);
2619 bbio->ordered = ordered;
2620 return 0;
2621 }
2622
2623 /*
2624 * Don't split the extent_map for NOCOW extents, as we're writing into
2625 * a pre-existing one.
2626 */
2627 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2628 ret = split_extent_map(bbio->inode, bbio->file_offset,
2629 ordered->num_bytes, len,
2630 ordered->disk_bytenr);
2631 if (ret)
2632 return ret;
2633 }
2634
2635 new = btrfs_split_ordered_extent(ordered, len);
2636 if (IS_ERR(new))
2637 return PTR_ERR(new);
2638 bbio->ordered = new;
2639 return 0;
2640 }
2641
2642 /*
2643 * given a list of ordered sums record them in the inode. This happens
2644 * at IO completion time based on sums calculated at bio submission time.
2645 */
add_pending_csums(struct btrfs_trans_handle * trans,struct list_head * list)2646 static int add_pending_csums(struct btrfs_trans_handle *trans,
2647 struct list_head *list)
2648 {
2649 struct btrfs_ordered_sum *sum;
2650 struct btrfs_root *csum_root = NULL;
2651 int ret;
2652
2653 list_for_each_entry(sum, list, list) {
2654 trans->adding_csums = true;
2655 if (!csum_root)
2656 csum_root = btrfs_csum_root(trans->fs_info,
2657 sum->logical);
2658 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2659 trans->adding_csums = false;
2660 if (ret)
2661 return ret;
2662 }
2663 return 0;
2664 }
2665
btrfs_find_new_delalloc_bytes(struct btrfs_inode * inode,const u64 start,const u64 len,struct extent_state ** cached_state)2666 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2667 const u64 start,
2668 const u64 len,
2669 struct extent_state **cached_state)
2670 {
2671 u64 search_start = start;
2672 const u64 end = start + len - 1;
2673
2674 while (search_start < end) {
2675 const u64 search_len = end - search_start + 1;
2676 struct extent_map *em;
2677 u64 em_len;
2678 int ret = 0;
2679
2680 em = btrfs_get_extent(inode, NULL, search_start, search_len);
2681 if (IS_ERR(em))
2682 return PTR_ERR(em);
2683
2684 if (em->block_start != EXTENT_MAP_HOLE)
2685 goto next;
2686
2687 em_len = em->len;
2688 if (em->start < search_start)
2689 em_len -= search_start - em->start;
2690 if (em_len > search_len)
2691 em_len = search_len;
2692
2693 ret = set_extent_bit(&inode->io_tree, search_start,
2694 search_start + em_len - 1,
2695 EXTENT_DELALLOC_NEW, cached_state);
2696 next:
2697 search_start = extent_map_end(em);
2698 free_extent_map(em);
2699 if (ret)
2700 return ret;
2701 }
2702 return 0;
2703 }
2704
btrfs_set_extent_delalloc(struct btrfs_inode * inode,u64 start,u64 end,unsigned int extra_bits,struct extent_state ** cached_state)2705 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2706 unsigned int extra_bits,
2707 struct extent_state **cached_state)
2708 {
2709 WARN_ON(PAGE_ALIGNED(end));
2710
2711 if (start >= i_size_read(&inode->vfs_inode) &&
2712 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2713 /*
2714 * There can't be any extents following eof in this case so just
2715 * set the delalloc new bit for the range directly.
2716 */
2717 extra_bits |= EXTENT_DELALLOC_NEW;
2718 } else {
2719 int ret;
2720
2721 ret = btrfs_find_new_delalloc_bytes(inode, start,
2722 end + 1 - start,
2723 cached_state);
2724 if (ret)
2725 return ret;
2726 }
2727
2728 return set_extent_bit(&inode->io_tree, start, end,
2729 EXTENT_DELALLOC | extra_bits, cached_state);
2730 }
2731
2732 /* see btrfs_writepage_start_hook for details on why this is required */
2733 struct btrfs_writepage_fixup {
2734 struct page *page;
2735 struct btrfs_inode *inode;
2736 struct btrfs_work work;
2737 };
2738
btrfs_writepage_fixup_worker(struct btrfs_work * work)2739 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2740 {
2741 struct btrfs_writepage_fixup *fixup =
2742 container_of(work, struct btrfs_writepage_fixup, work);
2743 struct btrfs_ordered_extent *ordered;
2744 struct extent_state *cached_state = NULL;
2745 struct extent_changeset *data_reserved = NULL;
2746 struct page *page = fixup->page;
2747 struct btrfs_inode *inode = fixup->inode;
2748 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2749 u64 page_start = page_offset(page);
2750 u64 page_end = page_offset(page) + PAGE_SIZE - 1;
2751 int ret = 0;
2752 bool free_delalloc_space = true;
2753
2754 /*
2755 * This is similar to page_mkwrite, we need to reserve the space before
2756 * we take the page lock.
2757 */
2758 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2759 PAGE_SIZE);
2760 again:
2761 lock_page(page);
2762
2763 /*
2764 * Before we queued this fixup, we took a reference on the page.
2765 * page->mapping may go NULL, but it shouldn't be moved to a different
2766 * address space.
2767 */
2768 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2769 /*
2770 * Unfortunately this is a little tricky, either
2771 *
2772 * 1) We got here and our page had already been dealt with and
2773 * we reserved our space, thus ret == 0, so we need to just
2774 * drop our space reservation and bail. This can happen the
2775 * first time we come into the fixup worker, or could happen
2776 * while waiting for the ordered extent.
2777 * 2) Our page was already dealt with, but we happened to get an
2778 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2779 * this case we obviously don't have anything to release, but
2780 * because the page was already dealt with we don't want to
2781 * mark the page with an error, so make sure we're resetting
2782 * ret to 0. This is why we have this check _before_ the ret
2783 * check, because we do not want to have a surprise ENOSPC
2784 * when the page was already properly dealt with.
2785 */
2786 if (!ret) {
2787 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2788 btrfs_delalloc_release_space(inode, data_reserved,
2789 page_start, PAGE_SIZE,
2790 true);
2791 }
2792 ret = 0;
2793 goto out_page;
2794 }
2795
2796 /*
2797 * We can't mess with the page state unless it is locked, so now that
2798 * it is locked bail if we failed to make our space reservation.
2799 */
2800 if (ret)
2801 goto out_page;
2802
2803 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2804
2805 /* already ordered? We're done */
2806 if (PageOrdered(page))
2807 goto out_reserved;
2808
2809 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2810 if (ordered) {
2811 unlock_extent(&inode->io_tree, page_start, page_end,
2812 &cached_state);
2813 unlock_page(page);
2814 btrfs_start_ordered_extent(ordered);
2815 btrfs_put_ordered_extent(ordered);
2816 goto again;
2817 }
2818
2819 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2820 &cached_state);
2821 if (ret)
2822 goto out_reserved;
2823
2824 /*
2825 * Everything went as planned, we're now the owner of a dirty page with
2826 * delayed allocation bits set and space reserved for our COW
2827 * destination.
2828 *
2829 * The page was dirty when we started, nothing should have cleaned it.
2830 */
2831 BUG_ON(!PageDirty(page));
2832 free_delalloc_space = false;
2833 out_reserved:
2834 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2835 if (free_delalloc_space)
2836 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2837 PAGE_SIZE, true);
2838 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2839 out_page:
2840 if (ret) {
2841 /*
2842 * We hit ENOSPC or other errors. Update the mapping and page
2843 * to reflect the errors and clean the page.
2844 */
2845 mapping_set_error(page->mapping, ret);
2846 btrfs_mark_ordered_io_finished(inode, page, page_start,
2847 PAGE_SIZE, !ret);
2848 clear_page_dirty_for_io(page);
2849 }
2850 btrfs_folio_clear_checked(fs_info, page_folio(page), page_start, PAGE_SIZE);
2851 unlock_page(page);
2852 put_page(page);
2853 kfree(fixup);
2854 extent_changeset_free(data_reserved);
2855 /*
2856 * As a precaution, do a delayed iput in case it would be the last iput
2857 * that could need flushing space. Recursing back to fixup worker would
2858 * deadlock.
2859 */
2860 btrfs_add_delayed_iput(inode);
2861 }
2862
2863 /*
2864 * There are a few paths in the higher layers of the kernel that directly
2865 * set the page dirty bit without asking the filesystem if it is a
2866 * good idea. This causes problems because we want to make sure COW
2867 * properly happens and the data=ordered rules are followed.
2868 *
2869 * In our case any range that doesn't have the ORDERED bit set
2870 * hasn't been properly setup for IO. We kick off an async process
2871 * to fix it up. The async helper will wait for ordered extents, set
2872 * the delalloc bit and make it safe to write the page.
2873 */
btrfs_writepage_cow_fixup(struct page * page)2874 int btrfs_writepage_cow_fixup(struct page *page)
2875 {
2876 struct inode *inode = page->mapping->host;
2877 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2878 struct btrfs_writepage_fixup *fixup;
2879
2880 /* This page has ordered extent covering it already */
2881 if (PageOrdered(page))
2882 return 0;
2883
2884 /*
2885 * PageChecked is set below when we create a fixup worker for this page,
2886 * don't try to create another one if we're already PageChecked()
2887 *
2888 * The extent_io writepage code will redirty the page if we send back
2889 * EAGAIN.
2890 */
2891 if (PageChecked(page))
2892 return -EAGAIN;
2893
2894 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2895 if (!fixup)
2896 return -EAGAIN;
2897
2898 /*
2899 * We are already holding a reference to this inode from
2900 * write_cache_pages. We need to hold it because the space reservation
2901 * takes place outside of the page lock, and we can't trust
2902 * page->mapping outside of the page lock.
2903 */
2904 ihold(inode);
2905 btrfs_folio_set_checked(fs_info, page_folio(page), page_offset(page), PAGE_SIZE);
2906 get_page(page);
2907 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL);
2908 fixup->page = page;
2909 fixup->inode = BTRFS_I(inode);
2910 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2911
2912 return -EAGAIN;
2913 }
2914
insert_reserved_file_extent(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,u64 file_pos,struct btrfs_file_extent_item * stack_fi,const bool update_inode_bytes,u64 qgroup_reserved)2915 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2916 struct btrfs_inode *inode, u64 file_pos,
2917 struct btrfs_file_extent_item *stack_fi,
2918 const bool update_inode_bytes,
2919 u64 qgroup_reserved)
2920 {
2921 struct btrfs_root *root = inode->root;
2922 const u64 sectorsize = root->fs_info->sectorsize;
2923 struct btrfs_path *path;
2924 struct extent_buffer *leaf;
2925 struct btrfs_key ins;
2926 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2927 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2928 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2929 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2930 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2931 struct btrfs_drop_extents_args drop_args = { 0 };
2932 int ret;
2933
2934 path = btrfs_alloc_path();
2935 if (!path)
2936 return -ENOMEM;
2937
2938 /*
2939 * we may be replacing one extent in the tree with another.
2940 * The new extent is pinned in the extent map, and we don't want
2941 * to drop it from the cache until it is completely in the btree.
2942 *
2943 * So, tell btrfs_drop_extents to leave this extent in the cache.
2944 * the caller is expected to unpin it and allow it to be merged
2945 * with the others.
2946 */
2947 drop_args.path = path;
2948 drop_args.start = file_pos;
2949 drop_args.end = file_pos + num_bytes;
2950 drop_args.replace_extent = true;
2951 drop_args.extent_item_size = sizeof(*stack_fi);
2952 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2953 if (ret)
2954 goto out;
2955
2956 if (!drop_args.extent_inserted) {
2957 ins.objectid = btrfs_ino(inode);
2958 ins.offset = file_pos;
2959 ins.type = BTRFS_EXTENT_DATA_KEY;
2960
2961 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2962 sizeof(*stack_fi));
2963 if (ret)
2964 goto out;
2965 }
2966 leaf = path->nodes[0];
2967 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2968 write_extent_buffer(leaf, stack_fi,
2969 btrfs_item_ptr_offset(leaf, path->slots[0]),
2970 sizeof(struct btrfs_file_extent_item));
2971
2972 btrfs_mark_buffer_dirty(trans, leaf);
2973 btrfs_release_path(path);
2974
2975 /*
2976 * If we dropped an inline extent here, we know the range where it is
2977 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2978 * number of bytes only for that range containing the inline extent.
2979 * The remaining of the range will be processed when clearning the
2980 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2981 */
2982 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2983 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2984
2985 inline_size = drop_args.bytes_found - inline_size;
2986 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2987 drop_args.bytes_found -= inline_size;
2988 num_bytes -= sectorsize;
2989 }
2990
2991 if (update_inode_bytes)
2992 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2993
2994 ins.objectid = disk_bytenr;
2995 ins.offset = disk_num_bytes;
2996 ins.type = BTRFS_EXTENT_ITEM_KEY;
2997
2998 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2999 if (ret)
3000 goto out;
3001
3002 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
3003 file_pos - offset,
3004 qgroup_reserved, &ins);
3005 out:
3006 btrfs_free_path(path);
3007
3008 return ret;
3009 }
3010
btrfs_release_delalloc_bytes(struct btrfs_fs_info * fs_info,u64 start,u64 len)3011 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
3012 u64 start, u64 len)
3013 {
3014 struct btrfs_block_group *cache;
3015
3016 cache = btrfs_lookup_block_group(fs_info, start);
3017 ASSERT(cache);
3018
3019 spin_lock(&cache->lock);
3020 cache->delalloc_bytes -= len;
3021 spin_unlock(&cache->lock);
3022
3023 btrfs_put_block_group(cache);
3024 }
3025
insert_ordered_extent_file_extent(struct btrfs_trans_handle * trans,struct btrfs_ordered_extent * oe)3026 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3027 struct btrfs_ordered_extent *oe)
3028 {
3029 struct btrfs_file_extent_item stack_fi;
3030 bool update_inode_bytes;
3031 u64 num_bytes = oe->num_bytes;
3032 u64 ram_bytes = oe->ram_bytes;
3033
3034 memset(&stack_fi, 0, sizeof(stack_fi));
3035 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3036 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3037 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3038 oe->disk_num_bytes);
3039 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3040 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3041 num_bytes = oe->truncated_len;
3042 ram_bytes = num_bytes;
3043 }
3044 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3045 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3046 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3047 /* Encryption and other encoding is reserved and all 0 */
3048
3049 /*
3050 * For delalloc, when completing an ordered extent we update the inode's
3051 * bytes when clearing the range in the inode's io tree, so pass false
3052 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3053 * except if the ordered extent was truncated.
3054 */
3055 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3056 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3057 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3058
3059 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3060 oe->file_offset, &stack_fi,
3061 update_inode_bytes, oe->qgroup_rsv);
3062 }
3063
3064 /*
3065 * As ordered data IO finishes, this gets called so we can finish
3066 * an ordered extent if the range of bytes in the file it covers are
3067 * fully written.
3068 */
btrfs_finish_one_ordered(struct btrfs_ordered_extent * ordered_extent)3069 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3070 {
3071 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3072 struct btrfs_root *root = inode->root;
3073 struct btrfs_fs_info *fs_info = root->fs_info;
3074 struct btrfs_trans_handle *trans = NULL;
3075 struct extent_io_tree *io_tree = &inode->io_tree;
3076 struct extent_state *cached_state = NULL;
3077 u64 start, end;
3078 int compress_type = 0;
3079 int ret = 0;
3080 u64 logical_len = ordered_extent->num_bytes;
3081 bool freespace_inode;
3082 bool truncated = false;
3083 bool clear_reserved_extent = true;
3084 unsigned int clear_bits = EXTENT_DEFRAG;
3085
3086 start = ordered_extent->file_offset;
3087 end = start + ordered_extent->num_bytes - 1;
3088
3089 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3090 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3091 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3092 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3093 clear_bits |= EXTENT_DELALLOC_NEW;
3094
3095 freespace_inode = btrfs_is_free_space_inode(inode);
3096 if (!freespace_inode)
3097 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3098
3099 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3100 ret = -EIO;
3101 goto out;
3102 }
3103
3104 if (btrfs_is_zoned(fs_info))
3105 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3106 ordered_extent->disk_num_bytes);
3107
3108 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3109 truncated = true;
3110 logical_len = ordered_extent->truncated_len;
3111 /* Truncated the entire extent, don't bother adding */
3112 if (!logical_len)
3113 goto out;
3114 }
3115
3116 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3117 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3118
3119 btrfs_inode_safe_disk_i_size_write(inode, 0);
3120 if (freespace_inode)
3121 trans = btrfs_join_transaction_spacecache(root);
3122 else
3123 trans = btrfs_join_transaction(root);
3124 if (IS_ERR(trans)) {
3125 ret = PTR_ERR(trans);
3126 trans = NULL;
3127 goto out;
3128 }
3129 trans->block_rsv = &inode->block_rsv;
3130 ret = btrfs_update_inode_fallback(trans, inode);
3131 if (ret) /* -ENOMEM or corruption */
3132 btrfs_abort_transaction(trans, ret);
3133 goto out;
3134 }
3135
3136 clear_bits |= EXTENT_LOCKED;
3137 lock_extent(io_tree, start, end, &cached_state);
3138
3139 if (freespace_inode)
3140 trans = btrfs_join_transaction_spacecache(root);
3141 else
3142 trans = btrfs_join_transaction(root);
3143 if (IS_ERR(trans)) {
3144 ret = PTR_ERR(trans);
3145 trans = NULL;
3146 goto out;
3147 }
3148
3149 trans->block_rsv = &inode->block_rsv;
3150
3151 ret = btrfs_insert_raid_extent(trans, ordered_extent);
3152 if (ret)
3153 goto out;
3154
3155 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3156 compress_type = ordered_extent->compress_type;
3157 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3158 BUG_ON(compress_type);
3159 ret = btrfs_mark_extent_written(trans, inode,
3160 ordered_extent->file_offset,
3161 ordered_extent->file_offset +
3162 logical_len);
3163 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3164 ordered_extent->disk_num_bytes);
3165 } else {
3166 BUG_ON(root == fs_info->tree_root);
3167 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3168 if (!ret) {
3169 clear_reserved_extent = false;
3170 btrfs_release_delalloc_bytes(fs_info,
3171 ordered_extent->disk_bytenr,
3172 ordered_extent->disk_num_bytes);
3173 }
3174 }
3175 if (ret < 0) {
3176 btrfs_abort_transaction(trans, ret);
3177 goto out;
3178 }
3179
3180 ret = unpin_extent_cache(inode, ordered_extent->file_offset,
3181 ordered_extent->num_bytes, trans->transid);
3182 if (ret < 0) {
3183 btrfs_abort_transaction(trans, ret);
3184 goto out;
3185 }
3186
3187 ret = add_pending_csums(trans, &ordered_extent->list);
3188 if (ret) {
3189 btrfs_abort_transaction(trans, ret);
3190 goto out;
3191 }
3192
3193 /*
3194 * If this is a new delalloc range, clear its new delalloc flag to
3195 * update the inode's number of bytes. This needs to be done first
3196 * before updating the inode item.
3197 */
3198 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3199 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3200 clear_extent_bit(&inode->io_tree, start, end,
3201 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3202 &cached_state);
3203
3204 btrfs_inode_safe_disk_i_size_write(inode, 0);
3205 ret = btrfs_update_inode_fallback(trans, inode);
3206 if (ret) { /* -ENOMEM or corruption */
3207 btrfs_abort_transaction(trans, ret);
3208 goto out;
3209 }
3210 out:
3211 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3212 &cached_state);
3213
3214 if (trans)
3215 btrfs_end_transaction(trans);
3216
3217 if (ret || truncated) {
3218 u64 unwritten_start = start;
3219
3220 /*
3221 * If we failed to finish this ordered extent for any reason we
3222 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3223 * extent, and mark the inode with the error if it wasn't
3224 * already set. Any error during writeback would have already
3225 * set the mapping error, so we need to set it if we're the ones
3226 * marking this ordered extent as failed.
3227 */
3228 if (ret)
3229 btrfs_mark_ordered_extent_error(ordered_extent);
3230
3231 if (truncated)
3232 unwritten_start += logical_len;
3233 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3234
3235 /*
3236 * Drop extent maps for the part of the extent we didn't write.
3237 *
3238 * We have an exception here for the free_space_inode, this is
3239 * because when we do btrfs_get_extent() on the free space inode
3240 * we will search the commit root. If this is a new block group
3241 * we won't find anything, and we will trip over the assert in
3242 * writepage where we do ASSERT(em->block_start !=
3243 * EXTENT_MAP_HOLE).
3244 *
3245 * Theoretically we could also skip this for any NOCOW extent as
3246 * we don't mess with the extent map tree in the NOCOW case, but
3247 * for now simply skip this if we are the free space inode.
3248 */
3249 if (!btrfs_is_free_space_inode(inode))
3250 btrfs_drop_extent_map_range(inode, unwritten_start,
3251 end, false);
3252
3253 /*
3254 * If the ordered extent had an IOERR or something else went
3255 * wrong we need to return the space for this ordered extent
3256 * back to the allocator. We only free the extent in the
3257 * truncated case if we didn't write out the extent at all.
3258 *
3259 * If we made it past insert_reserved_file_extent before we
3260 * errored out then we don't need to do this as the accounting
3261 * has already been done.
3262 */
3263 if ((ret || !logical_len) &&
3264 clear_reserved_extent &&
3265 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3266 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3267 /*
3268 * Discard the range before returning it back to the
3269 * free space pool
3270 */
3271 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3272 btrfs_discard_extent(fs_info,
3273 ordered_extent->disk_bytenr,
3274 ordered_extent->disk_num_bytes,
3275 NULL);
3276 btrfs_free_reserved_extent(fs_info,
3277 ordered_extent->disk_bytenr,
3278 ordered_extent->disk_num_bytes, 1);
3279 /*
3280 * Actually free the qgroup rsv which was released when
3281 * the ordered extent was created.
3282 */
3283 btrfs_qgroup_free_refroot(fs_info, btrfs_root_id(inode->root),
3284 ordered_extent->qgroup_rsv,
3285 BTRFS_QGROUP_RSV_DATA);
3286 }
3287 }
3288
3289 /*
3290 * This needs to be done to make sure anybody waiting knows we are done
3291 * updating everything for this ordered extent.
3292 */
3293 btrfs_remove_ordered_extent(inode, ordered_extent);
3294
3295 /* once for us */
3296 btrfs_put_ordered_extent(ordered_extent);
3297 /* once for the tree */
3298 btrfs_put_ordered_extent(ordered_extent);
3299
3300 return ret;
3301 }
3302
btrfs_finish_ordered_io(struct btrfs_ordered_extent * ordered)3303 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3304 {
3305 if (btrfs_is_zoned(inode_to_fs_info(ordered->inode)) &&
3306 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3307 list_empty(&ordered->bioc_list))
3308 btrfs_finish_ordered_zoned(ordered);
3309 return btrfs_finish_one_ordered(ordered);
3310 }
3311
3312 /*
3313 * Verify the checksum for a single sector without any extra action that depend
3314 * on the type of I/O.
3315 */
btrfs_check_sector_csum(struct btrfs_fs_info * fs_info,struct page * page,u32 pgoff,u8 * csum,const u8 * const csum_expected)3316 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3317 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3318 {
3319 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3320 char *kaddr;
3321
3322 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3323
3324 shash->tfm = fs_info->csum_shash;
3325
3326 kaddr = kmap_local_page(page) + pgoff;
3327 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3328 kunmap_local(kaddr);
3329
3330 if (memcmp(csum, csum_expected, fs_info->csum_size))
3331 return -EIO;
3332 return 0;
3333 }
3334
3335 /*
3336 * Verify the checksum of a single data sector.
3337 *
3338 * @bbio: btrfs_io_bio which contains the csum
3339 * @dev: device the sector is on
3340 * @bio_offset: offset to the beginning of the bio (in bytes)
3341 * @bv: bio_vec to check
3342 *
3343 * Check if the checksum on a data block is valid. When a checksum mismatch is
3344 * detected, report the error and fill the corrupted range with zero.
3345 *
3346 * Return %true if the sector is ok or had no checksum to start with, else %false.
3347 */
btrfs_data_csum_ok(struct btrfs_bio * bbio,struct btrfs_device * dev,u32 bio_offset,struct bio_vec * bv)3348 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3349 u32 bio_offset, struct bio_vec *bv)
3350 {
3351 struct btrfs_inode *inode = bbio->inode;
3352 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3353 u64 file_offset = bbio->file_offset + bio_offset;
3354 u64 end = file_offset + bv->bv_len - 1;
3355 u8 *csum_expected;
3356 u8 csum[BTRFS_CSUM_SIZE];
3357
3358 ASSERT(bv->bv_len == fs_info->sectorsize);
3359
3360 if (!bbio->csum)
3361 return true;
3362
3363 if (btrfs_is_data_reloc_root(inode->root) &&
3364 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3365 NULL)) {
3366 /* Skip the range without csum for data reloc inode */
3367 clear_extent_bits(&inode->io_tree, file_offset, end,
3368 EXTENT_NODATASUM);
3369 return true;
3370 }
3371
3372 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3373 fs_info->csum_size;
3374 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3375 csum_expected))
3376 goto zeroit;
3377 return true;
3378
3379 zeroit:
3380 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3381 bbio->mirror_num);
3382 if (dev)
3383 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3384 memzero_bvec(bv);
3385 return false;
3386 }
3387
3388 /*
3389 * Perform a delayed iput on @inode.
3390 *
3391 * @inode: The inode we want to perform iput on
3392 *
3393 * This function uses the generic vfs_inode::i_count to track whether we should
3394 * just decrement it (in case it's > 1) or if this is the last iput then link
3395 * the inode to the delayed iput machinery. Delayed iputs are processed at
3396 * transaction commit time/superblock commit/cleaner kthread.
3397 */
btrfs_add_delayed_iput(struct btrfs_inode * inode)3398 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3399 {
3400 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3401 unsigned long flags;
3402
3403 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3404 return;
3405
3406 atomic_inc(&fs_info->nr_delayed_iputs);
3407 /*
3408 * Need to be irq safe here because we can be called from either an irq
3409 * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3410 * context.
3411 */
3412 spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3413 ASSERT(list_empty(&inode->delayed_iput));
3414 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3415 spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3416 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3417 wake_up_process(fs_info->cleaner_kthread);
3418 }
3419
run_delayed_iput_locked(struct btrfs_fs_info * fs_info,struct btrfs_inode * inode)3420 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3421 struct btrfs_inode *inode)
3422 {
3423 list_del_init(&inode->delayed_iput);
3424 spin_unlock_irq(&fs_info->delayed_iput_lock);
3425 iput(&inode->vfs_inode);
3426 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3427 wake_up(&fs_info->delayed_iputs_wait);
3428 spin_lock_irq(&fs_info->delayed_iput_lock);
3429 }
3430
btrfs_run_delayed_iput(struct btrfs_fs_info * fs_info,struct btrfs_inode * inode)3431 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3432 struct btrfs_inode *inode)
3433 {
3434 if (!list_empty(&inode->delayed_iput)) {
3435 spin_lock_irq(&fs_info->delayed_iput_lock);
3436 if (!list_empty(&inode->delayed_iput))
3437 run_delayed_iput_locked(fs_info, inode);
3438 spin_unlock_irq(&fs_info->delayed_iput_lock);
3439 }
3440 }
3441
btrfs_run_delayed_iputs(struct btrfs_fs_info * fs_info)3442 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3443 {
3444 /*
3445 * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3446 * calls btrfs_add_delayed_iput() and that needs to lock
3447 * fs_info->delayed_iput_lock. So we need to disable irqs here to
3448 * prevent a deadlock.
3449 */
3450 spin_lock_irq(&fs_info->delayed_iput_lock);
3451 while (!list_empty(&fs_info->delayed_iputs)) {
3452 struct btrfs_inode *inode;
3453
3454 inode = list_first_entry(&fs_info->delayed_iputs,
3455 struct btrfs_inode, delayed_iput);
3456 run_delayed_iput_locked(fs_info, inode);
3457 if (need_resched()) {
3458 spin_unlock_irq(&fs_info->delayed_iput_lock);
3459 cond_resched();
3460 spin_lock_irq(&fs_info->delayed_iput_lock);
3461 }
3462 }
3463 spin_unlock_irq(&fs_info->delayed_iput_lock);
3464 }
3465
3466 /*
3467 * Wait for flushing all delayed iputs
3468 *
3469 * @fs_info: the filesystem
3470 *
3471 * This will wait on any delayed iputs that are currently running with KILLABLE
3472 * set. Once they are all done running we will return, unless we are killed in
3473 * which case we return EINTR. This helps in user operations like fallocate etc
3474 * that might get blocked on the iputs.
3475 *
3476 * Return EINTR if we were killed, 0 if nothing's pending
3477 */
btrfs_wait_on_delayed_iputs(struct btrfs_fs_info * fs_info)3478 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3479 {
3480 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3481 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3482 if (ret)
3483 return -EINTR;
3484 return 0;
3485 }
3486
3487 /*
3488 * This creates an orphan entry for the given inode in case something goes wrong
3489 * in the middle of an unlink.
3490 */
btrfs_orphan_add(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)3491 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3492 struct btrfs_inode *inode)
3493 {
3494 int ret;
3495
3496 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3497 if (ret && ret != -EEXIST) {
3498 btrfs_abort_transaction(trans, ret);
3499 return ret;
3500 }
3501
3502 return 0;
3503 }
3504
3505 /*
3506 * We have done the delete so we can go ahead and remove the orphan item for
3507 * this particular inode.
3508 */
btrfs_orphan_del(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)3509 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3510 struct btrfs_inode *inode)
3511 {
3512 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3513 }
3514
3515 /*
3516 * this cleans up any orphans that may be left on the list from the last use
3517 * of this root.
3518 */
btrfs_orphan_cleanup(struct btrfs_root * root)3519 int btrfs_orphan_cleanup(struct btrfs_root *root)
3520 {
3521 struct btrfs_fs_info *fs_info = root->fs_info;
3522 struct btrfs_path *path;
3523 struct extent_buffer *leaf;
3524 struct btrfs_key key, found_key;
3525 struct btrfs_trans_handle *trans;
3526 struct inode *inode;
3527 u64 last_objectid = 0;
3528 int ret = 0, nr_unlink = 0;
3529
3530 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3531 return 0;
3532
3533 path = btrfs_alloc_path();
3534 if (!path) {
3535 ret = -ENOMEM;
3536 goto out;
3537 }
3538 path->reada = READA_BACK;
3539
3540 key.objectid = BTRFS_ORPHAN_OBJECTID;
3541 key.type = BTRFS_ORPHAN_ITEM_KEY;
3542 key.offset = (u64)-1;
3543
3544 while (1) {
3545 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3546 if (ret < 0)
3547 goto out;
3548
3549 /*
3550 * if ret == 0 means we found what we were searching for, which
3551 * is weird, but possible, so only screw with path if we didn't
3552 * find the key and see if we have stuff that matches
3553 */
3554 if (ret > 0) {
3555 ret = 0;
3556 if (path->slots[0] == 0)
3557 break;
3558 path->slots[0]--;
3559 }
3560
3561 /* pull out the item */
3562 leaf = path->nodes[0];
3563 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3564
3565 /* make sure the item matches what we want */
3566 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3567 break;
3568 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3569 break;
3570
3571 /* release the path since we're done with it */
3572 btrfs_release_path(path);
3573
3574 /*
3575 * this is where we are basically btrfs_lookup, without the
3576 * crossing root thing. we store the inode number in the
3577 * offset of the orphan item.
3578 */
3579
3580 if (found_key.offset == last_objectid) {
3581 /*
3582 * We found the same inode as before. This means we were
3583 * not able to remove its items via eviction triggered
3584 * by an iput(). A transaction abort may have happened,
3585 * due to -ENOSPC for example, so try to grab the error
3586 * that lead to a transaction abort, if any.
3587 */
3588 btrfs_err(fs_info,
3589 "Error removing orphan entry, stopping orphan cleanup");
3590 ret = BTRFS_FS_ERROR(fs_info) ?: -EINVAL;
3591 goto out;
3592 }
3593
3594 last_objectid = found_key.offset;
3595
3596 found_key.objectid = found_key.offset;
3597 found_key.type = BTRFS_INODE_ITEM_KEY;
3598 found_key.offset = 0;
3599 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3600 if (IS_ERR(inode)) {
3601 ret = PTR_ERR(inode);
3602 inode = NULL;
3603 if (ret != -ENOENT)
3604 goto out;
3605 }
3606
3607 if (!inode && root == fs_info->tree_root) {
3608 struct btrfs_root *dead_root;
3609 int is_dead_root = 0;
3610
3611 /*
3612 * This is an orphan in the tree root. Currently these
3613 * could come from 2 sources:
3614 * a) a root (snapshot/subvolume) deletion in progress
3615 * b) a free space cache inode
3616 * We need to distinguish those two, as the orphan item
3617 * for a root must not get deleted before the deletion
3618 * of the snapshot/subvolume's tree completes.
3619 *
3620 * btrfs_find_orphan_roots() ran before us, which has
3621 * found all deleted roots and loaded them into
3622 * fs_info->fs_roots_radix. So here we can find if an
3623 * orphan item corresponds to a deleted root by looking
3624 * up the root from that radix tree.
3625 */
3626
3627 spin_lock(&fs_info->fs_roots_radix_lock);
3628 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3629 (unsigned long)found_key.objectid);
3630 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3631 is_dead_root = 1;
3632 spin_unlock(&fs_info->fs_roots_radix_lock);
3633
3634 if (is_dead_root) {
3635 /* prevent this orphan from being found again */
3636 key.offset = found_key.objectid - 1;
3637 continue;
3638 }
3639
3640 }
3641
3642 /*
3643 * If we have an inode with links, there are a couple of
3644 * possibilities:
3645 *
3646 * 1. We were halfway through creating fsverity metadata for the
3647 * file. In that case, the orphan item represents incomplete
3648 * fsverity metadata which must be cleaned up with
3649 * btrfs_drop_verity_items and deleting the orphan item.
3650
3651 * 2. Old kernels (before v3.12) used to create an
3652 * orphan item for truncate indicating that there were possibly
3653 * extent items past i_size that needed to be deleted. In v3.12,
3654 * truncate was changed to update i_size in sync with the extent
3655 * items, but the (useless) orphan item was still created. Since
3656 * v4.18, we don't create the orphan item for truncate at all.
3657 *
3658 * So, this item could mean that we need to do a truncate, but
3659 * only if this filesystem was last used on a pre-v3.12 kernel
3660 * and was not cleanly unmounted. The odds of that are quite
3661 * slim, and it's a pain to do the truncate now, so just delete
3662 * the orphan item.
3663 *
3664 * It's also possible that this orphan item was supposed to be
3665 * deleted but wasn't. The inode number may have been reused,
3666 * but either way, we can delete the orphan item.
3667 */
3668 if (!inode || inode->i_nlink) {
3669 if (inode) {
3670 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3671 iput(inode);
3672 inode = NULL;
3673 if (ret)
3674 goto out;
3675 }
3676 trans = btrfs_start_transaction(root, 1);
3677 if (IS_ERR(trans)) {
3678 ret = PTR_ERR(trans);
3679 goto out;
3680 }
3681 btrfs_debug(fs_info, "auto deleting %Lu",
3682 found_key.objectid);
3683 ret = btrfs_del_orphan_item(trans, root,
3684 found_key.objectid);
3685 btrfs_end_transaction(trans);
3686 if (ret)
3687 goto out;
3688 continue;
3689 }
3690
3691 nr_unlink++;
3692
3693 /* this will do delete_inode and everything for us */
3694 iput(inode);
3695 }
3696 /* release the path since we're done with it */
3697 btrfs_release_path(path);
3698
3699 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3700 trans = btrfs_join_transaction(root);
3701 if (!IS_ERR(trans))
3702 btrfs_end_transaction(trans);
3703 }
3704
3705 if (nr_unlink)
3706 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3707
3708 out:
3709 if (ret)
3710 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3711 btrfs_free_path(path);
3712 return ret;
3713 }
3714
3715 /*
3716 * very simple check to peek ahead in the leaf looking for xattrs. If we
3717 * don't find any xattrs, we know there can't be any acls.
3718 *
3719 * slot is the slot the inode is in, objectid is the objectid of the inode
3720 */
acls_after_inode_item(struct extent_buffer * leaf,int slot,u64 objectid,int * first_xattr_slot)3721 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3722 int slot, u64 objectid,
3723 int *first_xattr_slot)
3724 {
3725 u32 nritems = btrfs_header_nritems(leaf);
3726 struct btrfs_key found_key;
3727 static u64 xattr_access = 0;
3728 static u64 xattr_default = 0;
3729 int scanned = 0;
3730
3731 if (!xattr_access) {
3732 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3733 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3734 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3735 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3736 }
3737
3738 slot++;
3739 *first_xattr_slot = -1;
3740 while (slot < nritems) {
3741 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3742
3743 /* we found a different objectid, there must not be acls */
3744 if (found_key.objectid != objectid)
3745 return 0;
3746
3747 /* we found an xattr, assume we've got an acl */
3748 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3749 if (*first_xattr_slot == -1)
3750 *first_xattr_slot = slot;
3751 if (found_key.offset == xattr_access ||
3752 found_key.offset == xattr_default)
3753 return 1;
3754 }
3755
3756 /*
3757 * we found a key greater than an xattr key, there can't
3758 * be any acls later on
3759 */
3760 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3761 return 0;
3762
3763 slot++;
3764 scanned++;
3765
3766 /*
3767 * it goes inode, inode backrefs, xattrs, extents,
3768 * so if there are a ton of hard links to an inode there can
3769 * be a lot of backrefs. Don't waste time searching too hard,
3770 * this is just an optimization
3771 */
3772 if (scanned >= 8)
3773 break;
3774 }
3775 /* we hit the end of the leaf before we found an xattr or
3776 * something larger than an xattr. We have to assume the inode
3777 * has acls
3778 */
3779 if (*first_xattr_slot == -1)
3780 *first_xattr_slot = slot;
3781 return 1;
3782 }
3783
3784 /*
3785 * read an inode from the btree into the in-memory inode
3786 */
btrfs_read_locked_inode(struct inode * inode,struct btrfs_path * in_path)3787 static int btrfs_read_locked_inode(struct inode *inode,
3788 struct btrfs_path *in_path)
3789 {
3790 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
3791 struct btrfs_path *path = in_path;
3792 struct extent_buffer *leaf;
3793 struct btrfs_inode_item *inode_item;
3794 struct btrfs_root *root = BTRFS_I(inode)->root;
3795 struct btrfs_key location;
3796 unsigned long ptr;
3797 int maybe_acls;
3798 u32 rdev;
3799 int ret;
3800 bool filled = false;
3801 int first_xattr_slot;
3802
3803 ret = btrfs_fill_inode(inode, &rdev);
3804 if (!ret)
3805 filled = true;
3806
3807 if (!path) {
3808 path = btrfs_alloc_path();
3809 if (!path)
3810 return -ENOMEM;
3811 }
3812
3813 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3814
3815 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3816 if (ret) {
3817 if (path != in_path)
3818 btrfs_free_path(path);
3819 return ret;
3820 }
3821
3822 leaf = path->nodes[0];
3823
3824 if (filled)
3825 goto cache_index;
3826
3827 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3828 struct btrfs_inode_item);
3829 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3830 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3831 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3832 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3833 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3834 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3835 round_up(i_size_read(inode), fs_info->sectorsize));
3836
3837 inode_set_atime(inode, btrfs_timespec_sec(leaf, &inode_item->atime),
3838 btrfs_timespec_nsec(leaf, &inode_item->atime));
3839
3840 inode_set_mtime(inode, btrfs_timespec_sec(leaf, &inode_item->mtime),
3841 btrfs_timespec_nsec(leaf, &inode_item->mtime));
3842
3843 inode_set_ctime(inode, btrfs_timespec_sec(leaf, &inode_item->ctime),
3844 btrfs_timespec_nsec(leaf, &inode_item->ctime));
3845
3846 BTRFS_I(inode)->i_otime_sec = btrfs_timespec_sec(leaf, &inode_item->otime);
3847 BTRFS_I(inode)->i_otime_nsec = btrfs_timespec_nsec(leaf, &inode_item->otime);
3848
3849 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3850 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3851 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3852
3853 inode_set_iversion_queried(inode,
3854 btrfs_inode_sequence(leaf, inode_item));
3855 inode->i_generation = BTRFS_I(inode)->generation;
3856 inode->i_rdev = 0;
3857 rdev = btrfs_inode_rdev(leaf, inode_item);
3858
3859 BTRFS_I(inode)->index_cnt = (u64)-1;
3860 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3861 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3862
3863 cache_index:
3864 /*
3865 * If we were modified in the current generation and evicted from memory
3866 * and then re-read we need to do a full sync since we don't have any
3867 * idea about which extents were modified before we were evicted from
3868 * cache.
3869 *
3870 * This is required for both inode re-read from disk and delayed inode
3871 * in the delayed_nodes xarray.
3872 */
3873 if (BTRFS_I(inode)->last_trans == btrfs_get_fs_generation(fs_info))
3874 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3875 &BTRFS_I(inode)->runtime_flags);
3876
3877 /*
3878 * We don't persist the id of the transaction where an unlink operation
3879 * against the inode was last made. So here we assume the inode might
3880 * have been evicted, and therefore the exact value of last_unlink_trans
3881 * lost, and set it to last_trans to avoid metadata inconsistencies
3882 * between the inode and its parent if the inode is fsync'ed and the log
3883 * replayed. For example, in the scenario:
3884 *
3885 * touch mydir/foo
3886 * ln mydir/foo mydir/bar
3887 * sync
3888 * unlink mydir/bar
3889 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3890 * xfs_io -c fsync mydir/foo
3891 * <power failure>
3892 * mount fs, triggers fsync log replay
3893 *
3894 * We must make sure that when we fsync our inode foo we also log its
3895 * parent inode, otherwise after log replay the parent still has the
3896 * dentry with the "bar" name but our inode foo has a link count of 1
3897 * and doesn't have an inode ref with the name "bar" anymore.
3898 *
3899 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3900 * but it guarantees correctness at the expense of occasional full
3901 * transaction commits on fsync if our inode is a directory, or if our
3902 * inode is not a directory, logging its parent unnecessarily.
3903 */
3904 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3905
3906 /*
3907 * Same logic as for last_unlink_trans. We don't persist the generation
3908 * of the last transaction where this inode was used for a reflink
3909 * operation, so after eviction and reloading the inode we must be
3910 * pessimistic and assume the last transaction that modified the inode.
3911 */
3912 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3913
3914 path->slots[0]++;
3915 if (inode->i_nlink != 1 ||
3916 path->slots[0] >= btrfs_header_nritems(leaf))
3917 goto cache_acl;
3918
3919 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3920 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3921 goto cache_acl;
3922
3923 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3924 if (location.type == BTRFS_INODE_REF_KEY) {
3925 struct btrfs_inode_ref *ref;
3926
3927 ref = (struct btrfs_inode_ref *)ptr;
3928 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3929 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3930 struct btrfs_inode_extref *extref;
3931
3932 extref = (struct btrfs_inode_extref *)ptr;
3933 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3934 extref);
3935 }
3936 cache_acl:
3937 /*
3938 * try to precache a NULL acl entry for files that don't have
3939 * any xattrs or acls
3940 */
3941 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3942 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3943 if (first_xattr_slot != -1) {
3944 path->slots[0] = first_xattr_slot;
3945 ret = btrfs_load_inode_props(inode, path);
3946 if (ret)
3947 btrfs_err(fs_info,
3948 "error loading props for ino %llu (root %llu): %d",
3949 btrfs_ino(BTRFS_I(inode)),
3950 btrfs_root_id(root), ret);
3951 }
3952 if (path != in_path)
3953 btrfs_free_path(path);
3954
3955 if (!maybe_acls)
3956 cache_no_acl(inode);
3957
3958 switch (inode->i_mode & S_IFMT) {
3959 case S_IFREG:
3960 inode->i_mapping->a_ops = &btrfs_aops;
3961 inode->i_fop = &btrfs_file_operations;
3962 inode->i_op = &btrfs_file_inode_operations;
3963 break;
3964 case S_IFDIR:
3965 inode->i_fop = &btrfs_dir_file_operations;
3966 inode->i_op = &btrfs_dir_inode_operations;
3967 break;
3968 case S_IFLNK:
3969 inode->i_op = &btrfs_symlink_inode_operations;
3970 inode_nohighmem(inode);
3971 inode->i_mapping->a_ops = &btrfs_aops;
3972 break;
3973 default:
3974 inode->i_op = &btrfs_special_inode_operations;
3975 init_special_inode(inode, inode->i_mode, rdev);
3976 break;
3977 }
3978
3979 btrfs_sync_inode_flags_to_i_flags(inode);
3980 return 0;
3981 }
3982
3983 /*
3984 * given a leaf and an inode, copy the inode fields into the leaf
3985 */
fill_inode_item(struct btrfs_trans_handle * trans,struct extent_buffer * leaf,struct btrfs_inode_item * item,struct inode * inode)3986 static void fill_inode_item(struct btrfs_trans_handle *trans,
3987 struct extent_buffer *leaf,
3988 struct btrfs_inode_item *item,
3989 struct inode *inode)
3990 {
3991 struct btrfs_map_token token;
3992 u64 flags;
3993
3994 btrfs_init_map_token(&token, leaf);
3995
3996 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3997 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3998 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3999 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4000 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4001
4002 btrfs_set_token_timespec_sec(&token, &item->atime,
4003 inode_get_atime_sec(inode));
4004 btrfs_set_token_timespec_nsec(&token, &item->atime,
4005 inode_get_atime_nsec(inode));
4006
4007 btrfs_set_token_timespec_sec(&token, &item->mtime,
4008 inode_get_mtime_sec(inode));
4009 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4010 inode_get_mtime_nsec(inode));
4011
4012 btrfs_set_token_timespec_sec(&token, &item->ctime,
4013 inode_get_ctime_sec(inode));
4014 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4015 inode_get_ctime_nsec(inode));
4016
4017 btrfs_set_token_timespec_sec(&token, &item->otime, BTRFS_I(inode)->i_otime_sec);
4018 btrfs_set_token_timespec_nsec(&token, &item->otime, BTRFS_I(inode)->i_otime_nsec);
4019
4020 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
4021 btrfs_set_token_inode_generation(&token, item,
4022 BTRFS_I(inode)->generation);
4023 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4024 btrfs_set_token_inode_transid(&token, item, trans->transid);
4025 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4026 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4027 BTRFS_I(inode)->ro_flags);
4028 btrfs_set_token_inode_flags(&token, item, flags);
4029 btrfs_set_token_inode_block_group(&token, item, 0);
4030 }
4031
4032 /*
4033 * copy everything in the in-memory inode into the btree.
4034 */
btrfs_update_inode_item(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)4035 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4036 struct btrfs_inode *inode)
4037 {
4038 struct btrfs_inode_item *inode_item;
4039 struct btrfs_path *path;
4040 struct extent_buffer *leaf;
4041 int ret;
4042
4043 path = btrfs_alloc_path();
4044 if (!path)
4045 return -ENOMEM;
4046
4047 ret = btrfs_lookup_inode(trans, inode->root, path, &inode->location, 1);
4048 if (ret) {
4049 if (ret > 0)
4050 ret = -ENOENT;
4051 goto failed;
4052 }
4053
4054 leaf = path->nodes[0];
4055 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4056 struct btrfs_inode_item);
4057
4058 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4059 btrfs_mark_buffer_dirty(trans, leaf);
4060 btrfs_set_inode_last_trans(trans, inode);
4061 ret = 0;
4062 failed:
4063 btrfs_free_path(path);
4064 return ret;
4065 }
4066
4067 /*
4068 * copy everything in the in-memory inode into the btree.
4069 */
btrfs_update_inode(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)4070 int btrfs_update_inode(struct btrfs_trans_handle *trans,
4071 struct btrfs_inode *inode)
4072 {
4073 struct btrfs_root *root = inode->root;
4074 struct btrfs_fs_info *fs_info = root->fs_info;
4075 int ret;
4076
4077 /*
4078 * If the inode is a free space inode, we can deadlock during commit
4079 * if we put it into the delayed code.
4080 *
4081 * The data relocation inode should also be directly updated
4082 * without delay
4083 */
4084 if (!btrfs_is_free_space_inode(inode)
4085 && !btrfs_is_data_reloc_root(root)
4086 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4087 btrfs_update_root_times(trans, root);
4088
4089 ret = btrfs_delayed_update_inode(trans, inode);
4090 if (!ret)
4091 btrfs_set_inode_last_trans(trans, inode);
4092 return ret;
4093 }
4094
4095 return btrfs_update_inode_item(trans, inode);
4096 }
4097
btrfs_update_inode_fallback(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)4098 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4099 struct btrfs_inode *inode)
4100 {
4101 int ret;
4102
4103 ret = btrfs_update_inode(trans, inode);
4104 if (ret == -ENOSPC)
4105 return btrfs_update_inode_item(trans, inode);
4106 return ret;
4107 }
4108
4109 /*
4110 * unlink helper that gets used here in inode.c and in the tree logging
4111 * recovery code. It remove a link in a directory with a given name, and
4112 * also drops the back refs in the inode to the directory
4113 */
__btrfs_unlink_inode(struct btrfs_trans_handle * trans,struct btrfs_inode * dir,struct btrfs_inode * inode,const struct fscrypt_str * name,struct btrfs_rename_ctx * rename_ctx)4114 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4115 struct btrfs_inode *dir,
4116 struct btrfs_inode *inode,
4117 const struct fscrypt_str *name,
4118 struct btrfs_rename_ctx *rename_ctx)
4119 {
4120 struct btrfs_root *root = dir->root;
4121 struct btrfs_fs_info *fs_info = root->fs_info;
4122 struct btrfs_path *path;
4123 int ret = 0;
4124 struct btrfs_dir_item *di;
4125 u64 index;
4126 u64 ino = btrfs_ino(inode);
4127 u64 dir_ino = btrfs_ino(dir);
4128
4129 path = btrfs_alloc_path();
4130 if (!path) {
4131 ret = -ENOMEM;
4132 goto out;
4133 }
4134
4135 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4136 if (IS_ERR_OR_NULL(di)) {
4137 ret = di ? PTR_ERR(di) : -ENOENT;
4138 goto err;
4139 }
4140 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4141 if (ret)
4142 goto err;
4143 btrfs_release_path(path);
4144
4145 /*
4146 * If we don't have dir index, we have to get it by looking up
4147 * the inode ref, since we get the inode ref, remove it directly,
4148 * it is unnecessary to do delayed deletion.
4149 *
4150 * But if we have dir index, needn't search inode ref to get it.
4151 * Since the inode ref is close to the inode item, it is better
4152 * that we delay to delete it, and just do this deletion when
4153 * we update the inode item.
4154 */
4155 if (inode->dir_index) {
4156 ret = btrfs_delayed_delete_inode_ref(inode);
4157 if (!ret) {
4158 index = inode->dir_index;
4159 goto skip_backref;
4160 }
4161 }
4162
4163 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4164 if (ret) {
4165 btrfs_info(fs_info,
4166 "failed to delete reference to %.*s, inode %llu parent %llu",
4167 name->len, name->name, ino, dir_ino);
4168 btrfs_abort_transaction(trans, ret);
4169 goto err;
4170 }
4171 skip_backref:
4172 if (rename_ctx)
4173 rename_ctx->index = index;
4174
4175 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4176 if (ret) {
4177 btrfs_abort_transaction(trans, ret);
4178 goto err;
4179 }
4180
4181 /*
4182 * If we are in a rename context, we don't need to update anything in the
4183 * log. That will be done later during the rename by btrfs_log_new_name().
4184 * Besides that, doing it here would only cause extra unnecessary btree
4185 * operations on the log tree, increasing latency for applications.
4186 */
4187 if (!rename_ctx) {
4188 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4189 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4190 }
4191
4192 /*
4193 * If we have a pending delayed iput we could end up with the final iput
4194 * being run in btrfs-cleaner context. If we have enough of these built
4195 * up we can end up burning a lot of time in btrfs-cleaner without any
4196 * way to throttle the unlinks. Since we're currently holding a ref on
4197 * the inode we can run the delayed iput here without any issues as the
4198 * final iput won't be done until after we drop the ref we're currently
4199 * holding.
4200 */
4201 btrfs_run_delayed_iput(fs_info, inode);
4202 err:
4203 btrfs_free_path(path);
4204 if (ret)
4205 goto out;
4206
4207 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4208 inode_inc_iversion(&inode->vfs_inode);
4209 inode_inc_iversion(&dir->vfs_inode);
4210 inode_set_mtime_to_ts(&dir->vfs_inode, inode_set_ctime_current(&dir->vfs_inode));
4211 ret = btrfs_update_inode(trans, dir);
4212 out:
4213 return ret;
4214 }
4215
btrfs_unlink_inode(struct btrfs_trans_handle * trans,struct btrfs_inode * dir,struct btrfs_inode * inode,const struct fscrypt_str * name)4216 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4217 struct btrfs_inode *dir, struct btrfs_inode *inode,
4218 const struct fscrypt_str *name)
4219 {
4220 int ret;
4221
4222 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4223 if (!ret) {
4224 drop_nlink(&inode->vfs_inode);
4225 ret = btrfs_update_inode(trans, inode);
4226 }
4227 return ret;
4228 }
4229
4230 /*
4231 * helper to start transaction for unlink and rmdir.
4232 *
4233 * unlink and rmdir are special in btrfs, they do not always free space, so
4234 * if we cannot make our reservations the normal way try and see if there is
4235 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4236 * allow the unlink to occur.
4237 */
__unlink_start_trans(struct btrfs_inode * dir)4238 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4239 {
4240 struct btrfs_root *root = dir->root;
4241
4242 return btrfs_start_transaction_fallback_global_rsv(root,
4243 BTRFS_UNLINK_METADATA_UNITS);
4244 }
4245
btrfs_unlink(struct inode * dir,struct dentry * dentry)4246 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4247 {
4248 struct btrfs_trans_handle *trans;
4249 struct inode *inode = d_inode(dentry);
4250 int ret;
4251 struct fscrypt_name fname;
4252
4253 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4254 if (ret)
4255 return ret;
4256
4257 /* This needs to handle no-key deletions later on */
4258
4259 trans = __unlink_start_trans(BTRFS_I(dir));
4260 if (IS_ERR(trans)) {
4261 ret = PTR_ERR(trans);
4262 goto fscrypt_free;
4263 }
4264
4265 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4266 false);
4267
4268 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4269 &fname.disk_name);
4270 if (ret)
4271 goto end_trans;
4272
4273 if (inode->i_nlink == 0) {
4274 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4275 if (ret)
4276 goto end_trans;
4277 }
4278
4279 end_trans:
4280 btrfs_end_transaction(trans);
4281 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4282 fscrypt_free:
4283 fscrypt_free_filename(&fname);
4284 return ret;
4285 }
4286
btrfs_unlink_subvol(struct btrfs_trans_handle * trans,struct btrfs_inode * dir,struct dentry * dentry)4287 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4288 struct btrfs_inode *dir, struct dentry *dentry)
4289 {
4290 struct btrfs_root *root = dir->root;
4291 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4292 struct btrfs_path *path;
4293 struct extent_buffer *leaf;
4294 struct btrfs_dir_item *di;
4295 struct btrfs_key key;
4296 u64 index;
4297 int ret;
4298 u64 objectid;
4299 u64 dir_ino = btrfs_ino(dir);
4300 struct fscrypt_name fname;
4301
4302 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4303 if (ret)
4304 return ret;
4305
4306 /* This needs to handle no-key deletions later on */
4307
4308 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4309 objectid = btrfs_root_id(inode->root);
4310 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4311 objectid = inode->location.objectid;
4312 } else {
4313 WARN_ON(1);
4314 fscrypt_free_filename(&fname);
4315 return -EINVAL;
4316 }
4317
4318 path = btrfs_alloc_path();
4319 if (!path) {
4320 ret = -ENOMEM;
4321 goto out;
4322 }
4323
4324 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4325 &fname.disk_name, -1);
4326 if (IS_ERR_OR_NULL(di)) {
4327 ret = di ? PTR_ERR(di) : -ENOENT;
4328 goto out;
4329 }
4330
4331 leaf = path->nodes[0];
4332 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4333 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4334 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4335 if (ret) {
4336 btrfs_abort_transaction(trans, ret);
4337 goto out;
4338 }
4339 btrfs_release_path(path);
4340
4341 /*
4342 * This is a placeholder inode for a subvolume we didn't have a
4343 * reference to at the time of the snapshot creation. In the meantime
4344 * we could have renamed the real subvol link into our snapshot, so
4345 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4346 * Instead simply lookup the dir_index_item for this entry so we can
4347 * remove it. Otherwise we know we have a ref to the root and we can
4348 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4349 */
4350 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4351 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4352 if (IS_ERR_OR_NULL(di)) {
4353 if (!di)
4354 ret = -ENOENT;
4355 else
4356 ret = PTR_ERR(di);
4357 btrfs_abort_transaction(trans, ret);
4358 goto out;
4359 }
4360
4361 leaf = path->nodes[0];
4362 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4363 index = key.offset;
4364 btrfs_release_path(path);
4365 } else {
4366 ret = btrfs_del_root_ref(trans, objectid,
4367 btrfs_root_id(root), dir_ino,
4368 &index, &fname.disk_name);
4369 if (ret) {
4370 btrfs_abort_transaction(trans, ret);
4371 goto out;
4372 }
4373 }
4374
4375 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4376 if (ret) {
4377 btrfs_abort_transaction(trans, ret);
4378 goto out;
4379 }
4380
4381 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4382 inode_inc_iversion(&dir->vfs_inode);
4383 inode_set_mtime_to_ts(&dir->vfs_inode, inode_set_ctime_current(&dir->vfs_inode));
4384 ret = btrfs_update_inode_fallback(trans, dir);
4385 if (ret)
4386 btrfs_abort_transaction(trans, ret);
4387 out:
4388 btrfs_free_path(path);
4389 fscrypt_free_filename(&fname);
4390 return ret;
4391 }
4392
4393 /*
4394 * Helper to check if the subvolume references other subvolumes or if it's
4395 * default.
4396 */
may_destroy_subvol(struct btrfs_root * root)4397 static noinline int may_destroy_subvol(struct btrfs_root *root)
4398 {
4399 struct btrfs_fs_info *fs_info = root->fs_info;
4400 struct btrfs_path *path;
4401 struct btrfs_dir_item *di;
4402 struct btrfs_key key;
4403 struct fscrypt_str name = FSTR_INIT("default", 7);
4404 u64 dir_id;
4405 int ret;
4406
4407 path = btrfs_alloc_path();
4408 if (!path)
4409 return -ENOMEM;
4410
4411 /* Make sure this root isn't set as the default subvol */
4412 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4413 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4414 dir_id, &name, 0);
4415 if (di && !IS_ERR(di)) {
4416 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4417 if (key.objectid == btrfs_root_id(root)) {
4418 ret = -EPERM;
4419 btrfs_err(fs_info,
4420 "deleting default subvolume %llu is not allowed",
4421 key.objectid);
4422 goto out;
4423 }
4424 btrfs_release_path(path);
4425 }
4426
4427 key.objectid = btrfs_root_id(root);
4428 key.type = BTRFS_ROOT_REF_KEY;
4429 key.offset = (u64)-1;
4430
4431 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4432 if (ret < 0)
4433 goto out;
4434 if (ret == 0) {
4435 /*
4436 * Key with offset -1 found, there would have to exist a root
4437 * with such id, but this is out of valid range.
4438 */
4439 ret = -EUCLEAN;
4440 goto out;
4441 }
4442
4443 ret = 0;
4444 if (path->slots[0] > 0) {
4445 path->slots[0]--;
4446 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4447 if (key.objectid == btrfs_root_id(root) && key.type == BTRFS_ROOT_REF_KEY)
4448 ret = -ENOTEMPTY;
4449 }
4450 out:
4451 btrfs_free_path(path);
4452 return ret;
4453 }
4454
4455 /* Delete all dentries for inodes belonging to the root */
btrfs_prune_dentries(struct btrfs_root * root)4456 static void btrfs_prune_dentries(struct btrfs_root *root)
4457 {
4458 struct btrfs_fs_info *fs_info = root->fs_info;
4459 struct btrfs_inode *inode;
4460 u64 min_ino = 0;
4461
4462 if (!BTRFS_FS_ERROR(fs_info))
4463 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4464
4465 inode = btrfs_find_first_inode(root, min_ino);
4466 while (inode) {
4467 if (atomic_read(&inode->vfs_inode.i_count) > 1)
4468 d_prune_aliases(&inode->vfs_inode);
4469
4470 min_ino = btrfs_ino(inode) + 1;
4471 /*
4472 * btrfs_drop_inode() will have it removed from the inode
4473 * cache when its usage count hits zero.
4474 */
4475 iput(&inode->vfs_inode);
4476 cond_resched();
4477 inode = btrfs_find_first_inode(root, min_ino);
4478 }
4479 }
4480
btrfs_delete_subvolume(struct btrfs_inode * dir,struct dentry * dentry)4481 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4482 {
4483 struct btrfs_root *root = dir->root;
4484 struct btrfs_fs_info *fs_info = root->fs_info;
4485 struct inode *inode = d_inode(dentry);
4486 struct btrfs_root *dest = BTRFS_I(inode)->root;
4487 struct btrfs_trans_handle *trans;
4488 struct btrfs_block_rsv block_rsv;
4489 u64 root_flags;
4490 u64 qgroup_reserved = 0;
4491 int ret;
4492
4493 down_write(&fs_info->subvol_sem);
4494
4495 /*
4496 * Don't allow to delete a subvolume with send in progress. This is
4497 * inside the inode lock so the error handling that has to drop the bit
4498 * again is not run concurrently.
4499 */
4500 spin_lock(&dest->root_item_lock);
4501 if (dest->send_in_progress) {
4502 spin_unlock(&dest->root_item_lock);
4503 btrfs_warn(fs_info,
4504 "attempt to delete subvolume %llu during send",
4505 btrfs_root_id(dest));
4506 ret = -EPERM;
4507 goto out_up_write;
4508 }
4509 if (atomic_read(&dest->nr_swapfiles)) {
4510 spin_unlock(&dest->root_item_lock);
4511 btrfs_warn(fs_info,
4512 "attempt to delete subvolume %llu with active swapfile",
4513 btrfs_root_id(root));
4514 ret = -EPERM;
4515 goto out_up_write;
4516 }
4517 root_flags = btrfs_root_flags(&dest->root_item);
4518 btrfs_set_root_flags(&dest->root_item,
4519 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4520 spin_unlock(&dest->root_item_lock);
4521
4522 ret = may_destroy_subvol(dest);
4523 if (ret)
4524 goto out_undead;
4525
4526 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4527 /*
4528 * One for dir inode,
4529 * two for dir entries,
4530 * two for root ref/backref.
4531 */
4532 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4533 if (ret)
4534 goto out_undead;
4535 qgroup_reserved = block_rsv.qgroup_rsv_reserved;
4536
4537 trans = btrfs_start_transaction(root, 0);
4538 if (IS_ERR(trans)) {
4539 ret = PTR_ERR(trans);
4540 goto out_release;
4541 }
4542 ret = btrfs_record_root_in_trans(trans, root);
4543 if (ret) {
4544 btrfs_abort_transaction(trans, ret);
4545 goto out_end_trans;
4546 }
4547 btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
4548 qgroup_reserved = 0;
4549 trans->block_rsv = &block_rsv;
4550 trans->bytes_reserved = block_rsv.size;
4551
4552 btrfs_record_snapshot_destroy(trans, dir);
4553
4554 ret = btrfs_unlink_subvol(trans, dir, dentry);
4555 if (ret) {
4556 btrfs_abort_transaction(trans, ret);
4557 goto out_end_trans;
4558 }
4559
4560 ret = btrfs_record_root_in_trans(trans, dest);
4561 if (ret) {
4562 btrfs_abort_transaction(trans, ret);
4563 goto out_end_trans;
4564 }
4565
4566 memset(&dest->root_item.drop_progress, 0,
4567 sizeof(dest->root_item.drop_progress));
4568 btrfs_set_root_drop_level(&dest->root_item, 0);
4569 btrfs_set_root_refs(&dest->root_item, 0);
4570
4571 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4572 ret = btrfs_insert_orphan_item(trans,
4573 fs_info->tree_root,
4574 btrfs_root_id(dest));
4575 if (ret) {
4576 btrfs_abort_transaction(trans, ret);
4577 goto out_end_trans;
4578 }
4579 }
4580
4581 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4582 BTRFS_UUID_KEY_SUBVOL, btrfs_root_id(dest));
4583 if (ret && ret != -ENOENT) {
4584 btrfs_abort_transaction(trans, ret);
4585 goto out_end_trans;
4586 }
4587 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4588 ret = btrfs_uuid_tree_remove(trans,
4589 dest->root_item.received_uuid,
4590 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4591 btrfs_root_id(dest));
4592 if (ret && ret != -ENOENT) {
4593 btrfs_abort_transaction(trans, ret);
4594 goto out_end_trans;
4595 }
4596 }
4597
4598 free_anon_bdev(dest->anon_dev);
4599 dest->anon_dev = 0;
4600 out_end_trans:
4601 trans->block_rsv = NULL;
4602 trans->bytes_reserved = 0;
4603 ret = btrfs_end_transaction(trans);
4604 inode->i_flags |= S_DEAD;
4605 out_release:
4606 btrfs_block_rsv_release(fs_info, &block_rsv, (u64)-1, NULL);
4607 if (qgroup_reserved)
4608 btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved);
4609 out_undead:
4610 if (ret) {
4611 spin_lock(&dest->root_item_lock);
4612 root_flags = btrfs_root_flags(&dest->root_item);
4613 btrfs_set_root_flags(&dest->root_item,
4614 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4615 spin_unlock(&dest->root_item_lock);
4616 }
4617 out_up_write:
4618 up_write(&fs_info->subvol_sem);
4619 if (!ret) {
4620 d_invalidate(dentry);
4621 btrfs_prune_dentries(dest);
4622 ASSERT(dest->send_in_progress == 0);
4623 }
4624
4625 return ret;
4626 }
4627
btrfs_rmdir(struct inode * dir,struct dentry * dentry)4628 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4629 {
4630 struct inode *inode = d_inode(dentry);
4631 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4632 int ret = 0;
4633 struct btrfs_trans_handle *trans;
4634 u64 last_unlink_trans;
4635 struct fscrypt_name fname;
4636
4637 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4638 return -ENOTEMPTY;
4639 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4640 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4641 btrfs_err(fs_info,
4642 "extent tree v2 doesn't support snapshot deletion yet");
4643 return -EOPNOTSUPP;
4644 }
4645 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4646 }
4647
4648 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4649 if (ret)
4650 return ret;
4651
4652 /* This needs to handle no-key deletions later on */
4653
4654 trans = __unlink_start_trans(BTRFS_I(dir));
4655 if (IS_ERR(trans)) {
4656 ret = PTR_ERR(trans);
4657 goto out_notrans;
4658 }
4659
4660 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4661 ret = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4662 goto out;
4663 }
4664
4665 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4666 if (ret)
4667 goto out;
4668
4669 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4670
4671 /* now the directory is empty */
4672 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4673 &fname.disk_name);
4674 if (!ret) {
4675 btrfs_i_size_write(BTRFS_I(inode), 0);
4676 /*
4677 * Propagate the last_unlink_trans value of the deleted dir to
4678 * its parent directory. This is to prevent an unrecoverable
4679 * log tree in the case we do something like this:
4680 * 1) create dir foo
4681 * 2) create snapshot under dir foo
4682 * 3) delete the snapshot
4683 * 4) rmdir foo
4684 * 5) mkdir foo
4685 * 6) fsync foo or some file inside foo
4686 */
4687 if (last_unlink_trans >= trans->transid)
4688 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4689 }
4690 out:
4691 btrfs_end_transaction(trans);
4692 out_notrans:
4693 btrfs_btree_balance_dirty(fs_info);
4694 fscrypt_free_filename(&fname);
4695
4696 return ret;
4697 }
4698
4699 /*
4700 * Read, zero a chunk and write a block.
4701 *
4702 * @inode - inode that we're zeroing
4703 * @from - the offset to start zeroing
4704 * @len - the length to zero, 0 to zero the entire range respective to the
4705 * offset
4706 * @front - zero up to the offset instead of from the offset on
4707 *
4708 * This will find the block for the "from" offset and cow the block and zero the
4709 * part we want to zero. This is used with truncate and hole punching.
4710 */
btrfs_truncate_block(struct btrfs_inode * inode,loff_t from,loff_t len,int front)4711 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4712 int front)
4713 {
4714 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4715 struct address_space *mapping = inode->vfs_inode.i_mapping;
4716 struct extent_io_tree *io_tree = &inode->io_tree;
4717 struct btrfs_ordered_extent *ordered;
4718 struct extent_state *cached_state = NULL;
4719 struct extent_changeset *data_reserved = NULL;
4720 bool only_release_metadata = false;
4721 u32 blocksize = fs_info->sectorsize;
4722 pgoff_t index = from >> PAGE_SHIFT;
4723 unsigned offset = from & (blocksize - 1);
4724 struct folio *folio;
4725 gfp_t mask = btrfs_alloc_write_mask(mapping);
4726 size_t write_bytes = blocksize;
4727 int ret = 0;
4728 u64 block_start;
4729 u64 block_end;
4730
4731 if (IS_ALIGNED(offset, blocksize) &&
4732 (!len || IS_ALIGNED(len, blocksize)))
4733 goto out;
4734
4735 block_start = round_down(from, blocksize);
4736 block_end = block_start + blocksize - 1;
4737
4738 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4739 blocksize, false);
4740 if (ret < 0) {
4741 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4742 /* For nocow case, no need to reserve data space */
4743 only_release_metadata = true;
4744 } else {
4745 goto out;
4746 }
4747 }
4748 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4749 if (ret < 0) {
4750 if (!only_release_metadata)
4751 btrfs_free_reserved_data_space(inode, data_reserved,
4752 block_start, blocksize);
4753 goto out;
4754 }
4755 again:
4756 folio = __filemap_get_folio(mapping, index,
4757 FGP_LOCK | FGP_ACCESSED | FGP_CREAT, mask);
4758 if (IS_ERR(folio)) {
4759 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4760 blocksize, true);
4761 btrfs_delalloc_release_extents(inode, blocksize);
4762 ret = -ENOMEM;
4763 goto out;
4764 }
4765
4766 if (!folio_test_uptodate(folio)) {
4767 ret = btrfs_read_folio(NULL, folio);
4768 folio_lock(folio);
4769 if (folio->mapping != mapping) {
4770 folio_unlock(folio);
4771 folio_put(folio);
4772 goto again;
4773 }
4774 if (!folio_test_uptodate(folio)) {
4775 ret = -EIO;
4776 goto out_unlock;
4777 }
4778 }
4779
4780 /*
4781 * We unlock the page after the io is completed and then re-lock it
4782 * above. release_folio() could have come in between that and cleared
4783 * folio private, but left the page in the mapping. Set the page mapped
4784 * here to make sure it's properly set for the subpage stuff.
4785 */
4786 ret = set_folio_extent_mapped(folio);
4787 if (ret < 0)
4788 goto out_unlock;
4789
4790 folio_wait_writeback(folio);
4791
4792 lock_extent(io_tree, block_start, block_end, &cached_state);
4793
4794 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4795 if (ordered) {
4796 unlock_extent(io_tree, block_start, block_end, &cached_state);
4797 folio_unlock(folio);
4798 folio_put(folio);
4799 btrfs_start_ordered_extent(ordered);
4800 btrfs_put_ordered_extent(ordered);
4801 goto again;
4802 }
4803
4804 clear_extent_bit(&inode->io_tree, block_start, block_end,
4805 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4806 &cached_state);
4807
4808 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4809 &cached_state);
4810 if (ret) {
4811 unlock_extent(io_tree, block_start, block_end, &cached_state);
4812 goto out_unlock;
4813 }
4814
4815 if (offset != blocksize) {
4816 if (!len)
4817 len = blocksize - offset;
4818 if (front)
4819 folio_zero_range(folio, block_start - folio_pos(folio),
4820 offset);
4821 else
4822 folio_zero_range(folio,
4823 (block_start - folio_pos(folio)) + offset,
4824 len);
4825 }
4826 btrfs_folio_clear_checked(fs_info, folio, block_start,
4827 block_end + 1 - block_start);
4828 btrfs_folio_set_dirty(fs_info, folio, block_start,
4829 block_end + 1 - block_start);
4830 unlock_extent(io_tree, block_start, block_end, &cached_state);
4831
4832 if (only_release_metadata)
4833 set_extent_bit(&inode->io_tree, block_start, block_end,
4834 EXTENT_NORESERVE, NULL);
4835
4836 out_unlock:
4837 if (ret) {
4838 if (only_release_metadata)
4839 btrfs_delalloc_release_metadata(inode, blocksize, true);
4840 else
4841 btrfs_delalloc_release_space(inode, data_reserved,
4842 block_start, blocksize, true);
4843 }
4844 btrfs_delalloc_release_extents(inode, blocksize);
4845 folio_unlock(folio);
4846 folio_put(folio);
4847 out:
4848 if (only_release_metadata)
4849 btrfs_check_nocow_unlock(inode);
4850 extent_changeset_free(data_reserved);
4851 return ret;
4852 }
4853
maybe_insert_hole(struct btrfs_inode * inode,u64 offset,u64 len)4854 static int maybe_insert_hole(struct btrfs_inode *inode, u64 offset, u64 len)
4855 {
4856 struct btrfs_root *root = inode->root;
4857 struct btrfs_fs_info *fs_info = root->fs_info;
4858 struct btrfs_trans_handle *trans;
4859 struct btrfs_drop_extents_args drop_args = { 0 };
4860 int ret;
4861
4862 /*
4863 * If NO_HOLES is enabled, we don't need to do anything.
4864 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4865 * or btrfs_update_inode() will be called, which guarantee that the next
4866 * fsync will know this inode was changed and needs to be logged.
4867 */
4868 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4869 return 0;
4870
4871 /*
4872 * 1 - for the one we're dropping
4873 * 1 - for the one we're adding
4874 * 1 - for updating the inode.
4875 */
4876 trans = btrfs_start_transaction(root, 3);
4877 if (IS_ERR(trans))
4878 return PTR_ERR(trans);
4879
4880 drop_args.start = offset;
4881 drop_args.end = offset + len;
4882 drop_args.drop_cache = true;
4883
4884 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4885 if (ret) {
4886 btrfs_abort_transaction(trans, ret);
4887 btrfs_end_transaction(trans);
4888 return ret;
4889 }
4890
4891 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4892 if (ret) {
4893 btrfs_abort_transaction(trans, ret);
4894 } else {
4895 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4896 btrfs_update_inode(trans, inode);
4897 }
4898 btrfs_end_transaction(trans);
4899 return ret;
4900 }
4901
4902 /*
4903 * This function puts in dummy file extents for the area we're creating a hole
4904 * for. So if we are truncating this file to a larger size we need to insert
4905 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4906 * the range between oldsize and size
4907 */
btrfs_cont_expand(struct btrfs_inode * inode,loff_t oldsize,loff_t size)4908 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4909 {
4910 struct btrfs_root *root = inode->root;
4911 struct btrfs_fs_info *fs_info = root->fs_info;
4912 struct extent_io_tree *io_tree = &inode->io_tree;
4913 struct extent_map *em = NULL;
4914 struct extent_state *cached_state = NULL;
4915 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4916 u64 block_end = ALIGN(size, fs_info->sectorsize);
4917 u64 last_byte;
4918 u64 cur_offset;
4919 u64 hole_size;
4920 int ret = 0;
4921
4922 /*
4923 * If our size started in the middle of a block we need to zero out the
4924 * rest of the block before we expand the i_size, otherwise we could
4925 * expose stale data.
4926 */
4927 ret = btrfs_truncate_block(inode, oldsize, 0, 0);
4928 if (ret)
4929 return ret;
4930
4931 if (size <= hole_start)
4932 return 0;
4933
4934 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4935 &cached_state);
4936 cur_offset = hole_start;
4937 while (1) {
4938 em = btrfs_get_extent(inode, NULL, cur_offset, block_end - cur_offset);
4939 if (IS_ERR(em)) {
4940 ret = PTR_ERR(em);
4941 em = NULL;
4942 break;
4943 }
4944 last_byte = min(extent_map_end(em), block_end);
4945 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4946 hole_size = last_byte - cur_offset;
4947
4948 if (!(em->flags & EXTENT_FLAG_PREALLOC)) {
4949 struct extent_map *hole_em;
4950
4951 ret = maybe_insert_hole(inode, cur_offset, hole_size);
4952 if (ret)
4953 break;
4954
4955 ret = btrfs_inode_set_file_extent_range(inode,
4956 cur_offset, hole_size);
4957 if (ret)
4958 break;
4959
4960 hole_em = alloc_extent_map();
4961 if (!hole_em) {
4962 btrfs_drop_extent_map_range(inode, cur_offset,
4963 cur_offset + hole_size - 1,
4964 false);
4965 btrfs_set_inode_full_sync(inode);
4966 goto next;
4967 }
4968 hole_em->start = cur_offset;
4969 hole_em->len = hole_size;
4970 hole_em->orig_start = cur_offset;
4971
4972 hole_em->block_start = EXTENT_MAP_HOLE;
4973 hole_em->block_len = 0;
4974 hole_em->orig_block_len = 0;
4975 hole_em->ram_bytes = hole_size;
4976 hole_em->generation = btrfs_get_fs_generation(fs_info);
4977
4978 ret = btrfs_replace_extent_map_range(inode, hole_em, true);
4979 free_extent_map(hole_em);
4980 } else {
4981 ret = btrfs_inode_set_file_extent_range(inode,
4982 cur_offset, hole_size);
4983 if (ret)
4984 break;
4985 }
4986 next:
4987 free_extent_map(em);
4988 em = NULL;
4989 cur_offset = last_byte;
4990 if (cur_offset >= block_end)
4991 break;
4992 }
4993 free_extent_map(em);
4994 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
4995 return ret;
4996 }
4997
btrfs_setsize(struct inode * inode,struct iattr * attr)4998 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4999 {
5000 struct btrfs_root *root = BTRFS_I(inode)->root;
5001 struct btrfs_trans_handle *trans;
5002 loff_t oldsize = i_size_read(inode);
5003 loff_t newsize = attr->ia_size;
5004 int mask = attr->ia_valid;
5005 int ret;
5006
5007 /*
5008 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5009 * special case where we need to update the times despite not having
5010 * these flags set. For all other operations the VFS set these flags
5011 * explicitly if it wants a timestamp update.
5012 */
5013 if (newsize != oldsize) {
5014 inode_inc_iversion(inode);
5015 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
5016 inode_set_mtime_to_ts(inode,
5017 inode_set_ctime_current(inode));
5018 }
5019 }
5020
5021 if (newsize > oldsize) {
5022 /*
5023 * Don't do an expanding truncate while snapshotting is ongoing.
5024 * This is to ensure the snapshot captures a fully consistent
5025 * state of this file - if the snapshot captures this expanding
5026 * truncation, it must capture all writes that happened before
5027 * this truncation.
5028 */
5029 btrfs_drew_write_lock(&root->snapshot_lock);
5030 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5031 if (ret) {
5032 btrfs_drew_write_unlock(&root->snapshot_lock);
5033 return ret;
5034 }
5035
5036 trans = btrfs_start_transaction(root, 1);
5037 if (IS_ERR(trans)) {
5038 btrfs_drew_write_unlock(&root->snapshot_lock);
5039 return PTR_ERR(trans);
5040 }
5041
5042 i_size_write(inode, newsize);
5043 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5044 pagecache_isize_extended(inode, oldsize, newsize);
5045 ret = btrfs_update_inode(trans, BTRFS_I(inode));
5046 btrfs_drew_write_unlock(&root->snapshot_lock);
5047 btrfs_end_transaction(trans);
5048 } else {
5049 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
5050
5051 if (btrfs_is_zoned(fs_info)) {
5052 ret = btrfs_wait_ordered_range(inode,
5053 ALIGN(newsize, fs_info->sectorsize),
5054 (u64)-1);
5055 if (ret)
5056 return ret;
5057 }
5058
5059 /*
5060 * We're truncating a file that used to have good data down to
5061 * zero. Make sure any new writes to the file get on disk
5062 * on close.
5063 */
5064 if (newsize == 0)
5065 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5066 &BTRFS_I(inode)->runtime_flags);
5067
5068 truncate_setsize(inode, newsize);
5069
5070 inode_dio_wait(inode);
5071
5072 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5073 if (ret && inode->i_nlink) {
5074 int err;
5075
5076 /*
5077 * Truncate failed, so fix up the in-memory size. We
5078 * adjusted disk_i_size down as we removed extents, so
5079 * wait for disk_i_size to be stable and then update the
5080 * in-memory size to match.
5081 */
5082 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5083 if (err)
5084 return err;
5085 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5086 }
5087 }
5088
5089 return ret;
5090 }
5091
btrfs_setattr(struct mnt_idmap * idmap,struct dentry * dentry,struct iattr * attr)5092 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5093 struct iattr *attr)
5094 {
5095 struct inode *inode = d_inode(dentry);
5096 struct btrfs_root *root = BTRFS_I(inode)->root;
5097 int err;
5098
5099 if (btrfs_root_readonly(root))
5100 return -EROFS;
5101
5102 err = setattr_prepare(idmap, dentry, attr);
5103 if (err)
5104 return err;
5105
5106 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5107 err = btrfs_setsize(inode, attr);
5108 if (err)
5109 return err;
5110 }
5111
5112 if (attr->ia_valid) {
5113 setattr_copy(idmap, inode, attr);
5114 inode_inc_iversion(inode);
5115 err = btrfs_dirty_inode(BTRFS_I(inode));
5116
5117 if (!err && attr->ia_valid & ATTR_MODE)
5118 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5119 }
5120
5121 return err;
5122 }
5123
5124 /*
5125 * While truncating the inode pages during eviction, we get the VFS
5126 * calling btrfs_invalidate_folio() against each folio of the inode. This
5127 * is slow because the calls to btrfs_invalidate_folio() result in a
5128 * huge amount of calls to lock_extent() and clear_extent_bit(),
5129 * which keep merging and splitting extent_state structures over and over,
5130 * wasting lots of time.
5131 *
5132 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5133 * skip all those expensive operations on a per folio basis and do only
5134 * the ordered io finishing, while we release here the extent_map and
5135 * extent_state structures, without the excessive merging and splitting.
5136 */
evict_inode_truncate_pages(struct inode * inode)5137 static void evict_inode_truncate_pages(struct inode *inode)
5138 {
5139 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5140 struct rb_node *node;
5141
5142 ASSERT(inode->i_state & I_FREEING);
5143 truncate_inode_pages_final(&inode->i_data);
5144
5145 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5146
5147 /*
5148 * Keep looping until we have no more ranges in the io tree.
5149 * We can have ongoing bios started by readahead that have
5150 * their endio callback (extent_io.c:end_bio_extent_readpage)
5151 * still in progress (unlocked the pages in the bio but did not yet
5152 * unlocked the ranges in the io tree). Therefore this means some
5153 * ranges can still be locked and eviction started because before
5154 * submitting those bios, which are executed by a separate task (work
5155 * queue kthread), inode references (inode->i_count) were not taken
5156 * (which would be dropped in the end io callback of each bio).
5157 * Therefore here we effectively end up waiting for those bios and
5158 * anyone else holding locked ranges without having bumped the inode's
5159 * reference count - if we don't do it, when they access the inode's
5160 * io_tree to unlock a range it may be too late, leading to an
5161 * use-after-free issue.
5162 */
5163 spin_lock(&io_tree->lock);
5164 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5165 struct extent_state *state;
5166 struct extent_state *cached_state = NULL;
5167 u64 start;
5168 u64 end;
5169 unsigned state_flags;
5170
5171 node = rb_first(&io_tree->state);
5172 state = rb_entry(node, struct extent_state, rb_node);
5173 start = state->start;
5174 end = state->end;
5175 state_flags = state->state;
5176 spin_unlock(&io_tree->lock);
5177
5178 lock_extent(io_tree, start, end, &cached_state);
5179
5180 /*
5181 * If still has DELALLOC flag, the extent didn't reach disk,
5182 * and its reserved space won't be freed by delayed_ref.
5183 * So we need to free its reserved space here.
5184 * (Refer to comment in btrfs_invalidate_folio, case 2)
5185 *
5186 * Note, end is the bytenr of last byte, so we need + 1 here.
5187 */
5188 if (state_flags & EXTENT_DELALLOC)
5189 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5190 end - start + 1, NULL);
5191
5192 clear_extent_bit(io_tree, start, end,
5193 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5194 &cached_state);
5195
5196 cond_resched();
5197 spin_lock(&io_tree->lock);
5198 }
5199 spin_unlock(&io_tree->lock);
5200 }
5201
evict_refill_and_join(struct btrfs_root * root,struct btrfs_block_rsv * rsv)5202 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5203 struct btrfs_block_rsv *rsv)
5204 {
5205 struct btrfs_fs_info *fs_info = root->fs_info;
5206 struct btrfs_trans_handle *trans;
5207 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5208 int ret;
5209
5210 /*
5211 * Eviction should be taking place at some place safe because of our
5212 * delayed iputs. However the normal flushing code will run delayed
5213 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5214 *
5215 * We reserve the delayed_refs_extra here again because we can't use
5216 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5217 * above. We reserve our extra bit here because we generate a ton of
5218 * delayed refs activity by truncating.
5219 *
5220 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5221 * if we fail to make this reservation we can re-try without the
5222 * delayed_refs_extra so we can make some forward progress.
5223 */
5224 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5225 BTRFS_RESERVE_FLUSH_EVICT);
5226 if (ret) {
5227 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5228 BTRFS_RESERVE_FLUSH_EVICT);
5229 if (ret) {
5230 btrfs_warn(fs_info,
5231 "could not allocate space for delete; will truncate on mount");
5232 return ERR_PTR(-ENOSPC);
5233 }
5234 delayed_refs_extra = 0;
5235 }
5236
5237 trans = btrfs_join_transaction(root);
5238 if (IS_ERR(trans))
5239 return trans;
5240
5241 if (delayed_refs_extra) {
5242 trans->block_rsv = &fs_info->trans_block_rsv;
5243 trans->bytes_reserved = delayed_refs_extra;
5244 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5245 delayed_refs_extra, true);
5246 }
5247 return trans;
5248 }
5249
btrfs_evict_inode(struct inode * inode)5250 void btrfs_evict_inode(struct inode *inode)
5251 {
5252 struct btrfs_fs_info *fs_info;
5253 struct btrfs_trans_handle *trans;
5254 struct btrfs_root *root = BTRFS_I(inode)->root;
5255 struct btrfs_block_rsv *rsv = NULL;
5256 int ret;
5257
5258 trace_btrfs_inode_evict(inode);
5259
5260 if (!root) {
5261 fsverity_cleanup_inode(inode);
5262 clear_inode(inode);
5263 return;
5264 }
5265
5266 fs_info = inode_to_fs_info(inode);
5267 evict_inode_truncate_pages(inode);
5268
5269 if (inode->i_nlink &&
5270 ((btrfs_root_refs(&root->root_item) != 0 &&
5271 btrfs_root_id(root) != BTRFS_ROOT_TREE_OBJECTID) ||
5272 btrfs_is_free_space_inode(BTRFS_I(inode))))
5273 goto out;
5274
5275 if (is_bad_inode(inode))
5276 goto out;
5277
5278 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5279 goto out;
5280
5281 if (inode->i_nlink > 0) {
5282 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5283 btrfs_root_id(root) != BTRFS_ROOT_TREE_OBJECTID);
5284 goto out;
5285 }
5286
5287 /*
5288 * This makes sure the inode item in tree is uptodate and the space for
5289 * the inode update is released.
5290 */
5291 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5292 if (ret)
5293 goto out;
5294
5295 /*
5296 * This drops any pending insert or delete operations we have for this
5297 * inode. We could have a delayed dir index deletion queued up, but
5298 * we're removing the inode completely so that'll be taken care of in
5299 * the truncate.
5300 */
5301 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5302
5303 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5304 if (!rsv)
5305 goto out;
5306 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5307 rsv->failfast = true;
5308
5309 btrfs_i_size_write(BTRFS_I(inode), 0);
5310
5311 while (1) {
5312 struct btrfs_truncate_control control = {
5313 .inode = BTRFS_I(inode),
5314 .ino = btrfs_ino(BTRFS_I(inode)),
5315 .new_size = 0,
5316 .min_type = 0,
5317 };
5318
5319 trans = evict_refill_and_join(root, rsv);
5320 if (IS_ERR(trans))
5321 goto out;
5322
5323 trans->block_rsv = rsv;
5324
5325 ret = btrfs_truncate_inode_items(trans, root, &control);
5326 trans->block_rsv = &fs_info->trans_block_rsv;
5327 btrfs_end_transaction(trans);
5328 /*
5329 * We have not added new delayed items for our inode after we
5330 * have flushed its delayed items, so no need to throttle on
5331 * delayed items. However we have modified extent buffers.
5332 */
5333 btrfs_btree_balance_dirty_nodelay(fs_info);
5334 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5335 goto out;
5336 else if (!ret)
5337 break;
5338 }
5339
5340 /*
5341 * Errors here aren't a big deal, it just means we leave orphan items in
5342 * the tree. They will be cleaned up on the next mount. If the inode
5343 * number gets reused, cleanup deletes the orphan item without doing
5344 * anything, and unlink reuses the existing orphan item.
5345 *
5346 * If it turns out that we are dropping too many of these, we might want
5347 * to add a mechanism for retrying these after a commit.
5348 */
5349 trans = evict_refill_and_join(root, rsv);
5350 if (!IS_ERR(trans)) {
5351 trans->block_rsv = rsv;
5352 btrfs_orphan_del(trans, BTRFS_I(inode));
5353 trans->block_rsv = &fs_info->trans_block_rsv;
5354 btrfs_end_transaction(trans);
5355 }
5356
5357 out:
5358 btrfs_free_block_rsv(fs_info, rsv);
5359 /*
5360 * If we didn't successfully delete, the orphan item will still be in
5361 * the tree and we'll retry on the next mount. Again, we might also want
5362 * to retry these periodically in the future.
5363 */
5364 btrfs_remove_delayed_node(BTRFS_I(inode));
5365 fsverity_cleanup_inode(inode);
5366 clear_inode(inode);
5367 }
5368
5369 /*
5370 * Return the key found in the dir entry in the location pointer, fill @type
5371 * with BTRFS_FT_*, and return 0.
5372 *
5373 * If no dir entries were found, returns -ENOENT.
5374 * If found a corrupted location in dir entry, returns -EUCLEAN.
5375 */
btrfs_inode_by_name(struct btrfs_inode * dir,struct dentry * dentry,struct btrfs_key * location,u8 * type)5376 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5377 struct btrfs_key *location, u8 *type)
5378 {
5379 struct btrfs_dir_item *di;
5380 struct btrfs_path *path;
5381 struct btrfs_root *root = dir->root;
5382 int ret = 0;
5383 struct fscrypt_name fname;
5384
5385 path = btrfs_alloc_path();
5386 if (!path)
5387 return -ENOMEM;
5388
5389 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5390 if (ret < 0)
5391 goto out;
5392 /*
5393 * fscrypt_setup_filename() should never return a positive value, but
5394 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5395 */
5396 ASSERT(ret == 0);
5397
5398 /* This needs to handle no-key deletions later on */
5399
5400 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5401 &fname.disk_name, 0);
5402 if (IS_ERR_OR_NULL(di)) {
5403 ret = di ? PTR_ERR(di) : -ENOENT;
5404 goto out;
5405 }
5406
5407 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5408 if (location->type != BTRFS_INODE_ITEM_KEY &&
5409 location->type != BTRFS_ROOT_ITEM_KEY) {
5410 ret = -EUCLEAN;
5411 btrfs_warn(root->fs_info,
5412 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5413 __func__, fname.disk_name.name, btrfs_ino(dir),
5414 location->objectid, location->type, location->offset);
5415 }
5416 if (!ret)
5417 *type = btrfs_dir_ftype(path->nodes[0], di);
5418 out:
5419 fscrypt_free_filename(&fname);
5420 btrfs_free_path(path);
5421 return ret;
5422 }
5423
5424 /*
5425 * when we hit a tree root in a directory, the btrfs part of the inode
5426 * needs to be changed to reflect the root directory of the tree root. This
5427 * is kind of like crossing a mount point.
5428 */
fixup_tree_root_location(struct btrfs_fs_info * fs_info,struct btrfs_inode * dir,struct dentry * dentry,struct btrfs_key * location,struct btrfs_root ** sub_root)5429 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5430 struct btrfs_inode *dir,
5431 struct dentry *dentry,
5432 struct btrfs_key *location,
5433 struct btrfs_root **sub_root)
5434 {
5435 struct btrfs_path *path;
5436 struct btrfs_root *new_root;
5437 struct btrfs_root_ref *ref;
5438 struct extent_buffer *leaf;
5439 struct btrfs_key key;
5440 int ret;
5441 int err = 0;
5442 struct fscrypt_name fname;
5443
5444 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5445 if (ret)
5446 return ret;
5447
5448 path = btrfs_alloc_path();
5449 if (!path) {
5450 err = -ENOMEM;
5451 goto out;
5452 }
5453
5454 err = -ENOENT;
5455 key.objectid = btrfs_root_id(dir->root);
5456 key.type = BTRFS_ROOT_REF_KEY;
5457 key.offset = location->objectid;
5458
5459 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5460 if (ret) {
5461 if (ret < 0)
5462 err = ret;
5463 goto out;
5464 }
5465
5466 leaf = path->nodes[0];
5467 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5468 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5469 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5470 goto out;
5471
5472 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5473 (unsigned long)(ref + 1), fname.disk_name.len);
5474 if (ret)
5475 goto out;
5476
5477 btrfs_release_path(path);
5478
5479 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5480 if (IS_ERR(new_root)) {
5481 err = PTR_ERR(new_root);
5482 goto out;
5483 }
5484
5485 *sub_root = new_root;
5486 location->objectid = btrfs_root_dirid(&new_root->root_item);
5487 location->type = BTRFS_INODE_ITEM_KEY;
5488 location->offset = 0;
5489 err = 0;
5490 out:
5491 btrfs_free_path(path);
5492 fscrypt_free_filename(&fname);
5493 return err;
5494 }
5495
inode_tree_add(struct btrfs_inode * inode)5496 static void inode_tree_add(struct btrfs_inode *inode)
5497 {
5498 struct btrfs_root *root = inode->root;
5499 struct btrfs_inode *entry;
5500 struct rb_node **p;
5501 struct rb_node *parent;
5502 struct rb_node *new = &inode->rb_node;
5503 u64 ino = btrfs_ino(inode);
5504
5505 if (inode_unhashed(&inode->vfs_inode))
5506 return;
5507 parent = NULL;
5508 spin_lock(&root->inode_lock);
5509 p = &root->inode_tree.rb_node;
5510 while (*p) {
5511 parent = *p;
5512 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5513
5514 if (ino < btrfs_ino(entry))
5515 p = &parent->rb_left;
5516 else if (ino > btrfs_ino(entry))
5517 p = &parent->rb_right;
5518 else {
5519 WARN_ON(!(entry->vfs_inode.i_state &
5520 (I_WILL_FREE | I_FREEING)));
5521 rb_replace_node(parent, new, &root->inode_tree);
5522 RB_CLEAR_NODE(parent);
5523 spin_unlock(&root->inode_lock);
5524 return;
5525 }
5526 }
5527 rb_link_node(new, parent, p);
5528 rb_insert_color(new, &root->inode_tree);
5529 spin_unlock(&root->inode_lock);
5530 }
5531
inode_tree_del(struct btrfs_inode * inode)5532 static void inode_tree_del(struct btrfs_inode *inode)
5533 {
5534 struct btrfs_root *root = inode->root;
5535 int empty = 0;
5536
5537 spin_lock(&root->inode_lock);
5538 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5539 rb_erase(&inode->rb_node, &root->inode_tree);
5540 RB_CLEAR_NODE(&inode->rb_node);
5541 empty = RB_EMPTY_ROOT(&root->inode_tree);
5542 }
5543 spin_unlock(&root->inode_lock);
5544
5545 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5546 spin_lock(&root->inode_lock);
5547 empty = RB_EMPTY_ROOT(&root->inode_tree);
5548 spin_unlock(&root->inode_lock);
5549 if (empty)
5550 btrfs_add_dead_root(root);
5551 }
5552 }
5553
5554
btrfs_init_locked_inode(struct inode * inode,void * p)5555 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5556 {
5557 struct btrfs_iget_args *args = p;
5558
5559 inode->i_ino = args->ino;
5560 BTRFS_I(inode)->location.objectid = args->ino;
5561 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5562 BTRFS_I(inode)->location.offset = 0;
5563 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5564
5565 if (args->root && args->root == args->root->fs_info->tree_root &&
5566 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5567 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5568 &BTRFS_I(inode)->runtime_flags);
5569 return 0;
5570 }
5571
btrfs_find_actor(struct inode * inode,void * opaque)5572 static int btrfs_find_actor(struct inode *inode, void *opaque)
5573 {
5574 struct btrfs_iget_args *args = opaque;
5575
5576 return args->ino == BTRFS_I(inode)->location.objectid &&
5577 args->root == BTRFS_I(inode)->root;
5578 }
5579
btrfs_iget_locked(struct super_block * s,u64 ino,struct btrfs_root * root)5580 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5581 struct btrfs_root *root)
5582 {
5583 struct inode *inode;
5584 struct btrfs_iget_args args;
5585 unsigned long hashval = btrfs_inode_hash(ino, root);
5586
5587 args.ino = ino;
5588 args.root = root;
5589
5590 inode = iget5_locked(s, hashval, btrfs_find_actor,
5591 btrfs_init_locked_inode,
5592 (void *)&args);
5593 return inode;
5594 }
5595
5596 /*
5597 * Get an inode object given its inode number and corresponding root.
5598 * Path can be preallocated to prevent recursing back to iget through
5599 * allocator. NULL is also valid but may require an additional allocation
5600 * later.
5601 */
btrfs_iget_path(struct super_block * s,u64 ino,struct btrfs_root * root,struct btrfs_path * path)5602 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5603 struct btrfs_root *root, struct btrfs_path *path)
5604 {
5605 struct inode *inode;
5606
5607 inode = btrfs_iget_locked(s, ino, root);
5608 if (!inode)
5609 return ERR_PTR(-ENOMEM);
5610
5611 if (inode->i_state & I_NEW) {
5612 int ret;
5613
5614 ret = btrfs_read_locked_inode(inode, path);
5615 if (!ret) {
5616 inode_tree_add(BTRFS_I(inode));
5617 unlock_new_inode(inode);
5618 } else {
5619 iget_failed(inode);
5620 /*
5621 * ret > 0 can come from btrfs_search_slot called by
5622 * btrfs_read_locked_inode, this means the inode item
5623 * was not found.
5624 */
5625 if (ret > 0)
5626 ret = -ENOENT;
5627 inode = ERR_PTR(ret);
5628 }
5629 }
5630
5631 return inode;
5632 }
5633
btrfs_iget(struct super_block * s,u64 ino,struct btrfs_root * root)5634 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5635 {
5636 return btrfs_iget_path(s, ino, root, NULL);
5637 }
5638
new_simple_dir(struct inode * dir,struct btrfs_key * key,struct btrfs_root * root)5639 static struct inode *new_simple_dir(struct inode *dir,
5640 struct btrfs_key *key,
5641 struct btrfs_root *root)
5642 {
5643 struct timespec64 ts;
5644 struct inode *inode = new_inode(dir->i_sb);
5645
5646 if (!inode)
5647 return ERR_PTR(-ENOMEM);
5648
5649 BTRFS_I(inode)->root = btrfs_grab_root(root);
5650 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5651 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5652
5653 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5654 /*
5655 * We only need lookup, the rest is read-only and there's no inode
5656 * associated with the dentry
5657 */
5658 inode->i_op = &simple_dir_inode_operations;
5659 inode->i_opflags &= ~IOP_XATTR;
5660 inode->i_fop = &simple_dir_operations;
5661 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5662
5663 ts = inode_set_ctime_current(inode);
5664 inode_set_mtime_to_ts(inode, ts);
5665 inode_set_atime_to_ts(inode, inode_get_atime(dir));
5666 BTRFS_I(inode)->i_otime_sec = ts.tv_sec;
5667 BTRFS_I(inode)->i_otime_nsec = ts.tv_nsec;
5668
5669 inode->i_uid = dir->i_uid;
5670 inode->i_gid = dir->i_gid;
5671
5672 return inode;
5673 }
5674
5675 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5676 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5677 static_assert(BTRFS_FT_DIR == FT_DIR);
5678 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5679 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5680 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5681 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5682 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5683
btrfs_inode_type(struct inode * inode)5684 static inline u8 btrfs_inode_type(struct inode *inode)
5685 {
5686 return fs_umode_to_ftype(inode->i_mode);
5687 }
5688
btrfs_lookup_dentry(struct inode * dir,struct dentry * dentry)5689 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5690 {
5691 struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
5692 struct inode *inode;
5693 struct btrfs_root *root = BTRFS_I(dir)->root;
5694 struct btrfs_root *sub_root = root;
5695 struct btrfs_key location;
5696 u8 di_type = 0;
5697 int ret = 0;
5698
5699 if (dentry->d_name.len > BTRFS_NAME_LEN)
5700 return ERR_PTR(-ENAMETOOLONG);
5701
5702 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5703 if (ret < 0)
5704 return ERR_PTR(ret);
5705
5706 if (location.type == BTRFS_INODE_ITEM_KEY) {
5707 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5708 if (IS_ERR(inode))
5709 return inode;
5710
5711 /* Do extra check against inode mode with di_type */
5712 if (btrfs_inode_type(inode) != di_type) {
5713 btrfs_crit(fs_info,
5714 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5715 inode->i_mode, btrfs_inode_type(inode),
5716 di_type);
5717 iput(inode);
5718 return ERR_PTR(-EUCLEAN);
5719 }
5720 return inode;
5721 }
5722
5723 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5724 &location, &sub_root);
5725 if (ret < 0) {
5726 if (ret != -ENOENT)
5727 inode = ERR_PTR(ret);
5728 else
5729 inode = new_simple_dir(dir, &location, root);
5730 } else {
5731 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5732 btrfs_put_root(sub_root);
5733
5734 if (IS_ERR(inode))
5735 return inode;
5736
5737 down_read(&fs_info->cleanup_work_sem);
5738 if (!sb_rdonly(inode->i_sb))
5739 ret = btrfs_orphan_cleanup(sub_root);
5740 up_read(&fs_info->cleanup_work_sem);
5741 if (ret) {
5742 iput(inode);
5743 inode = ERR_PTR(ret);
5744 }
5745 }
5746
5747 return inode;
5748 }
5749
btrfs_dentry_delete(const struct dentry * dentry)5750 static int btrfs_dentry_delete(const struct dentry *dentry)
5751 {
5752 struct btrfs_root *root;
5753 struct inode *inode = d_inode(dentry);
5754
5755 if (!inode && !IS_ROOT(dentry))
5756 inode = d_inode(dentry->d_parent);
5757
5758 if (inode) {
5759 root = BTRFS_I(inode)->root;
5760 if (btrfs_root_refs(&root->root_item) == 0)
5761 return 1;
5762
5763 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5764 return 1;
5765 }
5766 return 0;
5767 }
5768
btrfs_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)5769 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5770 unsigned int flags)
5771 {
5772 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5773
5774 if (inode == ERR_PTR(-ENOENT))
5775 inode = NULL;
5776 return d_splice_alias(inode, dentry);
5777 }
5778
5779 /*
5780 * Find the highest existing sequence number in a directory and then set the
5781 * in-memory index_cnt variable to the first free sequence number.
5782 */
btrfs_set_inode_index_count(struct btrfs_inode * inode)5783 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5784 {
5785 struct btrfs_root *root = inode->root;
5786 struct btrfs_key key, found_key;
5787 struct btrfs_path *path;
5788 struct extent_buffer *leaf;
5789 int ret;
5790
5791 key.objectid = btrfs_ino(inode);
5792 key.type = BTRFS_DIR_INDEX_KEY;
5793 key.offset = (u64)-1;
5794
5795 path = btrfs_alloc_path();
5796 if (!path)
5797 return -ENOMEM;
5798
5799 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5800 if (ret < 0)
5801 goto out;
5802 /* FIXME: we should be able to handle this */
5803 if (ret == 0)
5804 goto out;
5805 ret = 0;
5806
5807 if (path->slots[0] == 0) {
5808 inode->index_cnt = BTRFS_DIR_START_INDEX;
5809 goto out;
5810 }
5811
5812 path->slots[0]--;
5813
5814 leaf = path->nodes[0];
5815 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5816
5817 if (found_key.objectid != btrfs_ino(inode) ||
5818 found_key.type != BTRFS_DIR_INDEX_KEY) {
5819 inode->index_cnt = BTRFS_DIR_START_INDEX;
5820 goto out;
5821 }
5822
5823 inode->index_cnt = found_key.offset + 1;
5824 out:
5825 btrfs_free_path(path);
5826 return ret;
5827 }
5828
btrfs_get_dir_last_index(struct btrfs_inode * dir,u64 * index)5829 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5830 {
5831 int ret = 0;
5832
5833 btrfs_inode_lock(dir, 0);
5834 if (dir->index_cnt == (u64)-1) {
5835 ret = btrfs_inode_delayed_dir_index_count(dir);
5836 if (ret) {
5837 ret = btrfs_set_inode_index_count(dir);
5838 if (ret)
5839 goto out;
5840 }
5841 }
5842
5843 /* index_cnt is the index number of next new entry, so decrement it. */
5844 *index = dir->index_cnt - 1;
5845 out:
5846 btrfs_inode_unlock(dir, 0);
5847
5848 return ret;
5849 }
5850
5851 /*
5852 * All this infrastructure exists because dir_emit can fault, and we are holding
5853 * the tree lock when doing readdir. For now just allocate a buffer and copy
5854 * our information into that, and then dir_emit from the buffer. This is
5855 * similar to what NFS does, only we don't keep the buffer around in pagecache
5856 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5857 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5858 * tree lock.
5859 */
btrfs_opendir(struct inode * inode,struct file * file)5860 static int btrfs_opendir(struct inode *inode, struct file *file)
5861 {
5862 struct btrfs_file_private *private;
5863 u64 last_index;
5864 int ret;
5865
5866 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5867 if (ret)
5868 return ret;
5869
5870 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5871 if (!private)
5872 return -ENOMEM;
5873 private->last_index = last_index;
5874 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5875 if (!private->filldir_buf) {
5876 kfree(private);
5877 return -ENOMEM;
5878 }
5879 file->private_data = private;
5880 return 0;
5881 }
5882
btrfs_dir_llseek(struct file * file,loff_t offset,int whence)5883 static loff_t btrfs_dir_llseek(struct file *file, loff_t offset, int whence)
5884 {
5885 struct btrfs_file_private *private = file->private_data;
5886 int ret;
5887
5888 ret = btrfs_get_dir_last_index(BTRFS_I(file_inode(file)),
5889 &private->last_index);
5890 if (ret)
5891 return ret;
5892
5893 return generic_file_llseek(file, offset, whence);
5894 }
5895
5896 struct dir_entry {
5897 u64 ino;
5898 u64 offset;
5899 unsigned type;
5900 int name_len;
5901 };
5902
btrfs_filldir(void * addr,int entries,struct dir_context * ctx)5903 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5904 {
5905 while (entries--) {
5906 struct dir_entry *entry = addr;
5907 char *name = (char *)(entry + 1);
5908
5909 ctx->pos = get_unaligned(&entry->offset);
5910 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5911 get_unaligned(&entry->ino),
5912 get_unaligned(&entry->type)))
5913 return 1;
5914 addr += sizeof(struct dir_entry) +
5915 get_unaligned(&entry->name_len);
5916 ctx->pos++;
5917 }
5918 return 0;
5919 }
5920
btrfs_real_readdir(struct file * file,struct dir_context * ctx)5921 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5922 {
5923 struct inode *inode = file_inode(file);
5924 struct btrfs_root *root = BTRFS_I(inode)->root;
5925 struct btrfs_file_private *private = file->private_data;
5926 struct btrfs_dir_item *di;
5927 struct btrfs_key key;
5928 struct btrfs_key found_key;
5929 struct btrfs_path *path;
5930 void *addr;
5931 LIST_HEAD(ins_list);
5932 LIST_HEAD(del_list);
5933 int ret;
5934 char *name_ptr;
5935 int name_len;
5936 int entries = 0;
5937 int total_len = 0;
5938 bool put = false;
5939 struct btrfs_key location;
5940
5941 if (!dir_emit_dots(file, ctx))
5942 return 0;
5943
5944 path = btrfs_alloc_path();
5945 if (!path)
5946 return -ENOMEM;
5947
5948 addr = private->filldir_buf;
5949 path->reada = READA_FORWARD;
5950
5951 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
5952 &ins_list, &del_list);
5953
5954 again:
5955 key.type = BTRFS_DIR_INDEX_KEY;
5956 key.offset = ctx->pos;
5957 key.objectid = btrfs_ino(BTRFS_I(inode));
5958
5959 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5960 struct dir_entry *entry;
5961 struct extent_buffer *leaf = path->nodes[0];
5962 u8 ftype;
5963
5964 if (found_key.objectid != key.objectid)
5965 break;
5966 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5967 break;
5968 if (found_key.offset < ctx->pos)
5969 continue;
5970 if (found_key.offset > private->last_index)
5971 break;
5972 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5973 continue;
5974 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5975 name_len = btrfs_dir_name_len(leaf, di);
5976 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5977 PAGE_SIZE) {
5978 btrfs_release_path(path);
5979 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5980 if (ret)
5981 goto nopos;
5982 addr = private->filldir_buf;
5983 entries = 0;
5984 total_len = 0;
5985 goto again;
5986 }
5987
5988 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
5989 entry = addr;
5990 name_ptr = (char *)(entry + 1);
5991 read_extent_buffer(leaf, name_ptr,
5992 (unsigned long)(di + 1), name_len);
5993 put_unaligned(name_len, &entry->name_len);
5994 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
5995 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5996 put_unaligned(location.objectid, &entry->ino);
5997 put_unaligned(found_key.offset, &entry->offset);
5998 entries++;
5999 addr += sizeof(struct dir_entry) + name_len;
6000 total_len += sizeof(struct dir_entry) + name_len;
6001 }
6002 /* Catch error encountered during iteration */
6003 if (ret < 0)
6004 goto err;
6005
6006 btrfs_release_path(path);
6007
6008 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6009 if (ret)
6010 goto nopos;
6011
6012 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6013 if (ret)
6014 goto nopos;
6015
6016 /*
6017 * Stop new entries from being returned after we return the last
6018 * entry.
6019 *
6020 * New directory entries are assigned a strictly increasing
6021 * offset. This means that new entries created during readdir
6022 * are *guaranteed* to be seen in the future by that readdir.
6023 * This has broken buggy programs which operate on names as
6024 * they're returned by readdir. Until we re-use freed offsets
6025 * we have this hack to stop new entries from being returned
6026 * under the assumption that they'll never reach this huge
6027 * offset.
6028 *
6029 * This is being careful not to overflow 32bit loff_t unless the
6030 * last entry requires it because doing so has broken 32bit apps
6031 * in the past.
6032 */
6033 if (ctx->pos >= INT_MAX)
6034 ctx->pos = LLONG_MAX;
6035 else
6036 ctx->pos = INT_MAX;
6037 nopos:
6038 ret = 0;
6039 err:
6040 if (put)
6041 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6042 btrfs_free_path(path);
6043 return ret;
6044 }
6045
6046 /*
6047 * This is somewhat expensive, updating the tree every time the
6048 * inode changes. But, it is most likely to find the inode in cache.
6049 * FIXME, needs more benchmarking...there are no reasons other than performance
6050 * to keep or drop this code.
6051 */
btrfs_dirty_inode(struct btrfs_inode * inode)6052 static int btrfs_dirty_inode(struct btrfs_inode *inode)
6053 {
6054 struct btrfs_root *root = inode->root;
6055 struct btrfs_fs_info *fs_info = root->fs_info;
6056 struct btrfs_trans_handle *trans;
6057 int ret;
6058
6059 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6060 return 0;
6061
6062 trans = btrfs_join_transaction(root);
6063 if (IS_ERR(trans))
6064 return PTR_ERR(trans);
6065
6066 ret = btrfs_update_inode(trans, inode);
6067 if (ret == -ENOSPC || ret == -EDQUOT) {
6068 /* whoops, lets try again with the full transaction */
6069 btrfs_end_transaction(trans);
6070 trans = btrfs_start_transaction(root, 1);
6071 if (IS_ERR(trans))
6072 return PTR_ERR(trans);
6073
6074 ret = btrfs_update_inode(trans, inode);
6075 }
6076 btrfs_end_transaction(trans);
6077 if (inode->delayed_node)
6078 btrfs_balance_delayed_items(fs_info);
6079
6080 return ret;
6081 }
6082
6083 /*
6084 * This is a copy of file_update_time. We need this so we can return error on
6085 * ENOSPC for updating the inode in the case of file write and mmap writes.
6086 */
btrfs_update_time(struct inode * inode,int flags)6087 static int btrfs_update_time(struct inode *inode, int flags)
6088 {
6089 struct btrfs_root *root = BTRFS_I(inode)->root;
6090 bool dirty;
6091
6092 if (btrfs_root_readonly(root))
6093 return -EROFS;
6094
6095 dirty = inode_update_timestamps(inode, flags);
6096 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6097 }
6098
6099 /*
6100 * helper to find a free sequence number in a given directory. This current
6101 * code is very simple, later versions will do smarter things in the btree
6102 */
btrfs_set_inode_index(struct btrfs_inode * dir,u64 * index)6103 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6104 {
6105 int ret = 0;
6106
6107 if (dir->index_cnt == (u64)-1) {
6108 ret = btrfs_inode_delayed_dir_index_count(dir);
6109 if (ret) {
6110 ret = btrfs_set_inode_index_count(dir);
6111 if (ret)
6112 return ret;
6113 }
6114 }
6115
6116 *index = dir->index_cnt;
6117 dir->index_cnt++;
6118
6119 return ret;
6120 }
6121
btrfs_insert_inode_locked(struct inode * inode)6122 static int btrfs_insert_inode_locked(struct inode *inode)
6123 {
6124 struct btrfs_iget_args args;
6125
6126 args.ino = BTRFS_I(inode)->location.objectid;
6127 args.root = BTRFS_I(inode)->root;
6128
6129 return insert_inode_locked4(inode,
6130 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6131 btrfs_find_actor, &args);
6132 }
6133
btrfs_new_inode_prepare(struct btrfs_new_inode_args * args,unsigned int * trans_num_items)6134 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6135 unsigned int *trans_num_items)
6136 {
6137 struct inode *dir = args->dir;
6138 struct inode *inode = args->inode;
6139 int ret;
6140
6141 if (!args->orphan) {
6142 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6143 &args->fname);
6144 if (ret)
6145 return ret;
6146 }
6147
6148 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6149 if (ret) {
6150 fscrypt_free_filename(&args->fname);
6151 return ret;
6152 }
6153
6154 /* 1 to add inode item */
6155 *trans_num_items = 1;
6156 /* 1 to add compression property */
6157 if (BTRFS_I(dir)->prop_compress)
6158 (*trans_num_items)++;
6159 /* 1 to add default ACL xattr */
6160 if (args->default_acl)
6161 (*trans_num_items)++;
6162 /* 1 to add access ACL xattr */
6163 if (args->acl)
6164 (*trans_num_items)++;
6165 #ifdef CONFIG_SECURITY
6166 /* 1 to add LSM xattr */
6167 if (dir->i_security)
6168 (*trans_num_items)++;
6169 #endif
6170 if (args->orphan) {
6171 /* 1 to add orphan item */
6172 (*trans_num_items)++;
6173 } else {
6174 /*
6175 * 1 to add dir item
6176 * 1 to add dir index
6177 * 1 to update parent inode item
6178 *
6179 * No need for 1 unit for the inode ref item because it is
6180 * inserted in a batch together with the inode item at
6181 * btrfs_create_new_inode().
6182 */
6183 *trans_num_items += 3;
6184 }
6185 return 0;
6186 }
6187
btrfs_new_inode_args_destroy(struct btrfs_new_inode_args * args)6188 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6189 {
6190 posix_acl_release(args->acl);
6191 posix_acl_release(args->default_acl);
6192 fscrypt_free_filename(&args->fname);
6193 }
6194
6195 /*
6196 * Inherit flags from the parent inode.
6197 *
6198 * Currently only the compression flags and the cow flags are inherited.
6199 */
btrfs_inherit_iflags(struct btrfs_inode * inode,struct btrfs_inode * dir)6200 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6201 {
6202 unsigned int flags;
6203
6204 flags = dir->flags;
6205
6206 if (flags & BTRFS_INODE_NOCOMPRESS) {
6207 inode->flags &= ~BTRFS_INODE_COMPRESS;
6208 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6209 } else if (flags & BTRFS_INODE_COMPRESS) {
6210 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6211 inode->flags |= BTRFS_INODE_COMPRESS;
6212 }
6213
6214 if (flags & BTRFS_INODE_NODATACOW) {
6215 inode->flags |= BTRFS_INODE_NODATACOW;
6216 if (S_ISREG(inode->vfs_inode.i_mode))
6217 inode->flags |= BTRFS_INODE_NODATASUM;
6218 }
6219
6220 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6221 }
6222
btrfs_create_new_inode(struct btrfs_trans_handle * trans,struct btrfs_new_inode_args * args)6223 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6224 struct btrfs_new_inode_args *args)
6225 {
6226 struct timespec64 ts;
6227 struct inode *dir = args->dir;
6228 struct inode *inode = args->inode;
6229 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6230 struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
6231 struct btrfs_root *root;
6232 struct btrfs_inode_item *inode_item;
6233 struct btrfs_key *location;
6234 struct btrfs_path *path;
6235 u64 objectid;
6236 struct btrfs_inode_ref *ref;
6237 struct btrfs_key key[2];
6238 u32 sizes[2];
6239 struct btrfs_item_batch batch;
6240 unsigned long ptr;
6241 int ret;
6242
6243 path = btrfs_alloc_path();
6244 if (!path)
6245 return -ENOMEM;
6246
6247 if (!args->subvol)
6248 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6249 root = BTRFS_I(inode)->root;
6250
6251 ret = btrfs_get_free_objectid(root, &objectid);
6252 if (ret)
6253 goto out;
6254 inode->i_ino = objectid;
6255
6256 if (args->orphan) {
6257 /*
6258 * O_TMPFILE, set link count to 0, so that after this point, we
6259 * fill in an inode item with the correct link count.
6260 */
6261 set_nlink(inode, 0);
6262 } else {
6263 trace_btrfs_inode_request(dir);
6264
6265 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6266 if (ret)
6267 goto out;
6268 }
6269 /* index_cnt is ignored for everything but a dir. */
6270 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6271 BTRFS_I(inode)->generation = trans->transid;
6272 inode->i_generation = BTRFS_I(inode)->generation;
6273
6274 /*
6275 * We don't have any capability xattrs set here yet, shortcut any
6276 * queries for the xattrs here. If we add them later via the inode
6277 * security init path or any other path this flag will be cleared.
6278 */
6279 set_bit(BTRFS_INODE_NO_CAP_XATTR, &BTRFS_I(inode)->runtime_flags);
6280
6281 /*
6282 * Subvolumes don't inherit flags from their parent directory.
6283 * Originally this was probably by accident, but we probably can't
6284 * change it now without compatibility issues.
6285 */
6286 if (!args->subvol)
6287 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6288
6289 if (S_ISREG(inode->i_mode)) {
6290 if (btrfs_test_opt(fs_info, NODATASUM))
6291 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6292 if (btrfs_test_opt(fs_info, NODATACOW))
6293 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6294 BTRFS_INODE_NODATASUM;
6295 }
6296
6297 location = &BTRFS_I(inode)->location;
6298 location->objectid = objectid;
6299 location->offset = 0;
6300 location->type = BTRFS_INODE_ITEM_KEY;
6301
6302 ret = btrfs_insert_inode_locked(inode);
6303 if (ret < 0) {
6304 if (!args->orphan)
6305 BTRFS_I(dir)->index_cnt--;
6306 goto out;
6307 }
6308
6309 /*
6310 * We could have gotten an inode number from somebody who was fsynced
6311 * and then removed in this same transaction, so let's just set full
6312 * sync since it will be a full sync anyway and this will blow away the
6313 * old info in the log.
6314 */
6315 btrfs_set_inode_full_sync(BTRFS_I(inode));
6316
6317 key[0].objectid = objectid;
6318 key[0].type = BTRFS_INODE_ITEM_KEY;
6319 key[0].offset = 0;
6320
6321 sizes[0] = sizeof(struct btrfs_inode_item);
6322
6323 if (!args->orphan) {
6324 /*
6325 * Start new inodes with an inode_ref. This is slightly more
6326 * efficient for small numbers of hard links since they will
6327 * be packed into one item. Extended refs will kick in if we
6328 * add more hard links than can fit in the ref item.
6329 */
6330 key[1].objectid = objectid;
6331 key[1].type = BTRFS_INODE_REF_KEY;
6332 if (args->subvol) {
6333 key[1].offset = objectid;
6334 sizes[1] = 2 + sizeof(*ref);
6335 } else {
6336 key[1].offset = btrfs_ino(BTRFS_I(dir));
6337 sizes[1] = name->len + sizeof(*ref);
6338 }
6339 }
6340
6341 batch.keys = &key[0];
6342 batch.data_sizes = &sizes[0];
6343 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6344 batch.nr = args->orphan ? 1 : 2;
6345 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6346 if (ret != 0) {
6347 btrfs_abort_transaction(trans, ret);
6348 goto discard;
6349 }
6350
6351 ts = simple_inode_init_ts(inode);
6352 BTRFS_I(inode)->i_otime_sec = ts.tv_sec;
6353 BTRFS_I(inode)->i_otime_nsec = ts.tv_nsec;
6354
6355 /*
6356 * We're going to fill the inode item now, so at this point the inode
6357 * must be fully initialized.
6358 */
6359
6360 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6361 struct btrfs_inode_item);
6362 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6363 sizeof(*inode_item));
6364 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6365
6366 if (!args->orphan) {
6367 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6368 struct btrfs_inode_ref);
6369 ptr = (unsigned long)(ref + 1);
6370 if (args->subvol) {
6371 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6372 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6373 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6374 } else {
6375 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6376 name->len);
6377 btrfs_set_inode_ref_index(path->nodes[0], ref,
6378 BTRFS_I(inode)->dir_index);
6379 write_extent_buffer(path->nodes[0], name->name, ptr,
6380 name->len);
6381 }
6382 }
6383
6384 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
6385 /*
6386 * We don't need the path anymore, plus inheriting properties, adding
6387 * ACLs, security xattrs, orphan item or adding the link, will result in
6388 * allocating yet another path. So just free our path.
6389 */
6390 btrfs_free_path(path);
6391 path = NULL;
6392
6393 if (args->subvol) {
6394 struct inode *parent;
6395
6396 /*
6397 * Subvolumes inherit properties from their parent subvolume,
6398 * not the directory they were created in.
6399 */
6400 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6401 BTRFS_I(dir)->root);
6402 if (IS_ERR(parent)) {
6403 ret = PTR_ERR(parent);
6404 } else {
6405 ret = btrfs_inode_inherit_props(trans, inode, parent);
6406 iput(parent);
6407 }
6408 } else {
6409 ret = btrfs_inode_inherit_props(trans, inode, dir);
6410 }
6411 if (ret) {
6412 btrfs_err(fs_info,
6413 "error inheriting props for ino %llu (root %llu): %d",
6414 btrfs_ino(BTRFS_I(inode)), btrfs_root_id(root), ret);
6415 }
6416
6417 /*
6418 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6419 * probably a bug.
6420 */
6421 if (!args->subvol) {
6422 ret = btrfs_init_inode_security(trans, args);
6423 if (ret) {
6424 btrfs_abort_transaction(trans, ret);
6425 goto discard;
6426 }
6427 }
6428
6429 inode_tree_add(BTRFS_I(inode));
6430
6431 trace_btrfs_inode_new(inode);
6432 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6433
6434 btrfs_update_root_times(trans, root);
6435
6436 if (args->orphan) {
6437 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6438 } else {
6439 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6440 0, BTRFS_I(inode)->dir_index);
6441 }
6442 if (ret) {
6443 btrfs_abort_transaction(trans, ret);
6444 goto discard;
6445 }
6446
6447 return 0;
6448
6449 discard:
6450 /*
6451 * discard_new_inode() calls iput(), but the caller owns the reference
6452 * to the inode.
6453 */
6454 ihold(inode);
6455 discard_new_inode(inode);
6456 out:
6457 btrfs_free_path(path);
6458 return ret;
6459 }
6460
6461 /*
6462 * utility function to add 'inode' into 'parent_inode' with
6463 * a give name and a given sequence number.
6464 * if 'add_backref' is true, also insert a backref from the
6465 * inode to the parent directory.
6466 */
btrfs_add_link(struct btrfs_trans_handle * trans,struct btrfs_inode * parent_inode,struct btrfs_inode * inode,const struct fscrypt_str * name,int add_backref,u64 index)6467 int btrfs_add_link(struct btrfs_trans_handle *trans,
6468 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6469 const struct fscrypt_str *name, int add_backref, u64 index)
6470 {
6471 int ret = 0;
6472 struct btrfs_key key;
6473 struct btrfs_root *root = parent_inode->root;
6474 u64 ino = btrfs_ino(inode);
6475 u64 parent_ino = btrfs_ino(parent_inode);
6476
6477 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6478 memcpy(&key, &inode->root->root_key, sizeof(key));
6479 } else {
6480 key.objectid = ino;
6481 key.type = BTRFS_INODE_ITEM_KEY;
6482 key.offset = 0;
6483 }
6484
6485 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6486 ret = btrfs_add_root_ref(trans, key.objectid,
6487 btrfs_root_id(root), parent_ino,
6488 index, name);
6489 } else if (add_backref) {
6490 ret = btrfs_insert_inode_ref(trans, root, name,
6491 ino, parent_ino, index);
6492 }
6493
6494 /* Nothing to clean up yet */
6495 if (ret)
6496 return ret;
6497
6498 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6499 btrfs_inode_type(&inode->vfs_inode), index);
6500 if (ret == -EEXIST || ret == -EOVERFLOW)
6501 goto fail_dir_item;
6502 else if (ret) {
6503 btrfs_abort_transaction(trans, ret);
6504 return ret;
6505 }
6506
6507 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6508 name->len * 2);
6509 inode_inc_iversion(&parent_inode->vfs_inode);
6510 /*
6511 * If we are replaying a log tree, we do not want to update the mtime
6512 * and ctime of the parent directory with the current time, since the
6513 * log replay procedure is responsible for setting them to their correct
6514 * values (the ones it had when the fsync was done).
6515 */
6516 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags))
6517 inode_set_mtime_to_ts(&parent_inode->vfs_inode,
6518 inode_set_ctime_current(&parent_inode->vfs_inode));
6519
6520 ret = btrfs_update_inode(trans, parent_inode);
6521 if (ret)
6522 btrfs_abort_transaction(trans, ret);
6523 return ret;
6524
6525 fail_dir_item:
6526 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6527 u64 local_index;
6528 int err;
6529 err = btrfs_del_root_ref(trans, key.objectid,
6530 btrfs_root_id(root), parent_ino,
6531 &local_index, name);
6532 if (err)
6533 btrfs_abort_transaction(trans, err);
6534 } else if (add_backref) {
6535 u64 local_index;
6536 int err;
6537
6538 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6539 &local_index);
6540 if (err)
6541 btrfs_abort_transaction(trans, err);
6542 }
6543
6544 /* Return the original error code */
6545 return ret;
6546 }
6547
btrfs_create_common(struct inode * dir,struct dentry * dentry,struct inode * inode)6548 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6549 struct inode *inode)
6550 {
6551 struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
6552 struct btrfs_root *root = BTRFS_I(dir)->root;
6553 struct btrfs_new_inode_args new_inode_args = {
6554 .dir = dir,
6555 .dentry = dentry,
6556 .inode = inode,
6557 };
6558 unsigned int trans_num_items;
6559 struct btrfs_trans_handle *trans;
6560 int err;
6561
6562 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6563 if (err)
6564 goto out_inode;
6565
6566 trans = btrfs_start_transaction(root, trans_num_items);
6567 if (IS_ERR(trans)) {
6568 err = PTR_ERR(trans);
6569 goto out_new_inode_args;
6570 }
6571
6572 err = btrfs_create_new_inode(trans, &new_inode_args);
6573 if (!err)
6574 d_instantiate_new(dentry, inode);
6575
6576 btrfs_end_transaction(trans);
6577 btrfs_btree_balance_dirty(fs_info);
6578 out_new_inode_args:
6579 btrfs_new_inode_args_destroy(&new_inode_args);
6580 out_inode:
6581 if (err)
6582 iput(inode);
6583 return err;
6584 }
6585
btrfs_mknod(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode,dev_t rdev)6586 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6587 struct dentry *dentry, umode_t mode, dev_t rdev)
6588 {
6589 struct inode *inode;
6590
6591 inode = new_inode(dir->i_sb);
6592 if (!inode)
6593 return -ENOMEM;
6594 inode_init_owner(idmap, inode, dir, mode);
6595 inode->i_op = &btrfs_special_inode_operations;
6596 init_special_inode(inode, inode->i_mode, rdev);
6597 return btrfs_create_common(dir, dentry, inode);
6598 }
6599
btrfs_create(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode,bool excl)6600 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6601 struct dentry *dentry, umode_t mode, bool excl)
6602 {
6603 struct inode *inode;
6604
6605 inode = new_inode(dir->i_sb);
6606 if (!inode)
6607 return -ENOMEM;
6608 inode_init_owner(idmap, inode, dir, mode);
6609 inode->i_fop = &btrfs_file_operations;
6610 inode->i_op = &btrfs_file_inode_operations;
6611 inode->i_mapping->a_ops = &btrfs_aops;
6612 return btrfs_create_common(dir, dentry, inode);
6613 }
6614
btrfs_link(struct dentry * old_dentry,struct inode * dir,struct dentry * dentry)6615 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6616 struct dentry *dentry)
6617 {
6618 struct btrfs_trans_handle *trans = NULL;
6619 struct btrfs_root *root = BTRFS_I(dir)->root;
6620 struct inode *inode = d_inode(old_dentry);
6621 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
6622 struct fscrypt_name fname;
6623 u64 index;
6624 int err;
6625 int drop_inode = 0;
6626
6627 /* do not allow sys_link's with other subvols of the same device */
6628 if (btrfs_root_id(root) != btrfs_root_id(BTRFS_I(inode)->root))
6629 return -EXDEV;
6630
6631 if (inode->i_nlink >= BTRFS_LINK_MAX)
6632 return -EMLINK;
6633
6634 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6635 if (err)
6636 goto fail;
6637
6638 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6639 if (err)
6640 goto fail;
6641
6642 /*
6643 * 2 items for inode and inode ref
6644 * 2 items for dir items
6645 * 1 item for parent inode
6646 * 1 item for orphan item deletion if O_TMPFILE
6647 */
6648 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6649 if (IS_ERR(trans)) {
6650 err = PTR_ERR(trans);
6651 trans = NULL;
6652 goto fail;
6653 }
6654
6655 /* There are several dir indexes for this inode, clear the cache. */
6656 BTRFS_I(inode)->dir_index = 0ULL;
6657 inc_nlink(inode);
6658 inode_inc_iversion(inode);
6659 inode_set_ctime_current(inode);
6660 ihold(inode);
6661 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6662
6663 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6664 &fname.disk_name, 1, index);
6665
6666 if (err) {
6667 drop_inode = 1;
6668 } else {
6669 struct dentry *parent = dentry->d_parent;
6670
6671 err = btrfs_update_inode(trans, BTRFS_I(inode));
6672 if (err)
6673 goto fail;
6674 if (inode->i_nlink == 1) {
6675 /*
6676 * If new hard link count is 1, it's a file created
6677 * with open(2) O_TMPFILE flag.
6678 */
6679 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6680 if (err)
6681 goto fail;
6682 }
6683 d_instantiate(dentry, inode);
6684 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6685 }
6686
6687 fail:
6688 fscrypt_free_filename(&fname);
6689 if (trans)
6690 btrfs_end_transaction(trans);
6691 if (drop_inode) {
6692 inode_dec_link_count(inode);
6693 iput(inode);
6694 }
6695 btrfs_btree_balance_dirty(fs_info);
6696 return err;
6697 }
6698
btrfs_mkdir(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode)6699 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6700 struct dentry *dentry, umode_t mode)
6701 {
6702 struct inode *inode;
6703
6704 inode = new_inode(dir->i_sb);
6705 if (!inode)
6706 return -ENOMEM;
6707 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6708 inode->i_op = &btrfs_dir_inode_operations;
6709 inode->i_fop = &btrfs_dir_file_operations;
6710 return btrfs_create_common(dir, dentry, inode);
6711 }
6712
uncompress_inline(struct btrfs_path * path,struct page * page,struct btrfs_file_extent_item * item)6713 static noinline int uncompress_inline(struct btrfs_path *path,
6714 struct page *page,
6715 struct btrfs_file_extent_item *item)
6716 {
6717 int ret;
6718 struct extent_buffer *leaf = path->nodes[0];
6719 char *tmp;
6720 size_t max_size;
6721 unsigned long inline_size;
6722 unsigned long ptr;
6723 int compress_type;
6724
6725 compress_type = btrfs_file_extent_compression(leaf, item);
6726 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6727 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6728 tmp = kmalloc(inline_size, GFP_NOFS);
6729 if (!tmp)
6730 return -ENOMEM;
6731 ptr = btrfs_file_extent_inline_start(item);
6732
6733 read_extent_buffer(leaf, tmp, ptr, inline_size);
6734
6735 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6736 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6737
6738 /*
6739 * decompression code contains a memset to fill in any space between the end
6740 * of the uncompressed data and the end of max_size in case the decompressed
6741 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6742 * the end of an inline extent and the beginning of the next block, so we
6743 * cover that region here.
6744 */
6745
6746 if (max_size < PAGE_SIZE)
6747 memzero_page(page, max_size, PAGE_SIZE - max_size);
6748 kfree(tmp);
6749 return ret;
6750 }
6751
read_inline_extent(struct btrfs_inode * inode,struct btrfs_path * path,struct page * page)6752 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6753 struct page *page)
6754 {
6755 struct btrfs_file_extent_item *fi;
6756 void *kaddr;
6757 size_t copy_size;
6758
6759 if (!page || PageUptodate(page))
6760 return 0;
6761
6762 ASSERT(page_offset(page) == 0);
6763
6764 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6765 struct btrfs_file_extent_item);
6766 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6767 return uncompress_inline(path, page, fi);
6768
6769 copy_size = min_t(u64, PAGE_SIZE,
6770 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6771 kaddr = kmap_local_page(page);
6772 read_extent_buffer(path->nodes[0], kaddr,
6773 btrfs_file_extent_inline_start(fi), copy_size);
6774 kunmap_local(kaddr);
6775 if (copy_size < PAGE_SIZE)
6776 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6777 return 0;
6778 }
6779
6780 /*
6781 * Lookup the first extent overlapping a range in a file.
6782 *
6783 * @inode: file to search in
6784 * @page: page to read extent data into if the extent is inline
6785 * @start: file offset
6786 * @len: length of range starting at @start
6787 *
6788 * Return the first &struct extent_map which overlaps the given range, reading
6789 * it from the B-tree and caching it if necessary. Note that there may be more
6790 * extents which overlap the given range after the returned extent_map.
6791 *
6792 * If @page is not NULL and the extent is inline, this also reads the extent
6793 * data directly into the page and marks the extent up to date in the io_tree.
6794 *
6795 * Return: ERR_PTR on error, non-NULL extent_map on success.
6796 */
btrfs_get_extent(struct btrfs_inode * inode,struct page * page,u64 start,u64 len)6797 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6798 struct page *page, u64 start, u64 len)
6799 {
6800 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6801 int ret = 0;
6802 u64 extent_start = 0;
6803 u64 extent_end = 0;
6804 u64 objectid = btrfs_ino(inode);
6805 int extent_type = -1;
6806 struct btrfs_path *path = NULL;
6807 struct btrfs_root *root = inode->root;
6808 struct btrfs_file_extent_item *item;
6809 struct extent_buffer *leaf;
6810 struct btrfs_key found_key;
6811 struct extent_map *em = NULL;
6812 struct extent_map_tree *em_tree = &inode->extent_tree;
6813
6814 read_lock(&em_tree->lock);
6815 em = lookup_extent_mapping(em_tree, start, len);
6816 read_unlock(&em_tree->lock);
6817
6818 if (em) {
6819 if (em->start > start || em->start + em->len <= start)
6820 free_extent_map(em);
6821 else if (em->block_start == EXTENT_MAP_INLINE && page)
6822 free_extent_map(em);
6823 else
6824 goto out;
6825 }
6826 em = alloc_extent_map();
6827 if (!em) {
6828 ret = -ENOMEM;
6829 goto out;
6830 }
6831 em->start = EXTENT_MAP_HOLE;
6832 em->orig_start = EXTENT_MAP_HOLE;
6833 em->len = (u64)-1;
6834 em->block_len = (u64)-1;
6835
6836 path = btrfs_alloc_path();
6837 if (!path) {
6838 ret = -ENOMEM;
6839 goto out;
6840 }
6841
6842 /* Chances are we'll be called again, so go ahead and do readahead */
6843 path->reada = READA_FORWARD;
6844
6845 /*
6846 * The same explanation in load_free_space_cache applies here as well,
6847 * we only read when we're loading the free space cache, and at that
6848 * point the commit_root has everything we need.
6849 */
6850 if (btrfs_is_free_space_inode(inode)) {
6851 path->search_commit_root = 1;
6852 path->skip_locking = 1;
6853 }
6854
6855 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6856 if (ret < 0) {
6857 goto out;
6858 } else if (ret > 0) {
6859 if (path->slots[0] == 0)
6860 goto not_found;
6861 path->slots[0]--;
6862 ret = 0;
6863 }
6864
6865 leaf = path->nodes[0];
6866 item = btrfs_item_ptr(leaf, path->slots[0],
6867 struct btrfs_file_extent_item);
6868 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6869 if (found_key.objectid != objectid ||
6870 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6871 /*
6872 * If we backup past the first extent we want to move forward
6873 * and see if there is an extent in front of us, otherwise we'll
6874 * say there is a hole for our whole search range which can
6875 * cause problems.
6876 */
6877 extent_end = start;
6878 goto next;
6879 }
6880
6881 extent_type = btrfs_file_extent_type(leaf, item);
6882 extent_start = found_key.offset;
6883 extent_end = btrfs_file_extent_end(path);
6884 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6885 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6886 /* Only regular file could have regular/prealloc extent */
6887 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6888 ret = -EUCLEAN;
6889 btrfs_crit(fs_info,
6890 "regular/prealloc extent found for non-regular inode %llu",
6891 btrfs_ino(inode));
6892 goto out;
6893 }
6894 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6895 extent_start);
6896 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6897 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6898 path->slots[0],
6899 extent_start);
6900 }
6901 next:
6902 if (start >= extent_end) {
6903 path->slots[0]++;
6904 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6905 ret = btrfs_next_leaf(root, path);
6906 if (ret < 0)
6907 goto out;
6908 else if (ret > 0)
6909 goto not_found;
6910
6911 leaf = path->nodes[0];
6912 }
6913 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6914 if (found_key.objectid != objectid ||
6915 found_key.type != BTRFS_EXTENT_DATA_KEY)
6916 goto not_found;
6917 if (start + len <= found_key.offset)
6918 goto not_found;
6919 if (start > found_key.offset)
6920 goto next;
6921
6922 /* New extent overlaps with existing one */
6923 em->start = start;
6924 em->orig_start = start;
6925 em->len = found_key.offset - start;
6926 em->block_start = EXTENT_MAP_HOLE;
6927 goto insert;
6928 }
6929
6930 btrfs_extent_item_to_extent_map(inode, path, item, em);
6931
6932 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6933 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6934 goto insert;
6935 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6936 /*
6937 * Inline extent can only exist at file offset 0. This is
6938 * ensured by tree-checker and inline extent creation path.
6939 * Thus all members representing file offsets should be zero.
6940 */
6941 ASSERT(extent_start == 0);
6942 ASSERT(em->start == 0);
6943
6944 /*
6945 * btrfs_extent_item_to_extent_map() should have properly
6946 * initialized em members already.
6947 *
6948 * Other members are not utilized for inline extents.
6949 */
6950 ASSERT(em->block_start == EXTENT_MAP_INLINE);
6951 ASSERT(em->len == fs_info->sectorsize);
6952
6953 ret = read_inline_extent(inode, path, page);
6954 if (ret < 0)
6955 goto out;
6956 goto insert;
6957 }
6958 not_found:
6959 em->start = start;
6960 em->orig_start = start;
6961 em->len = len;
6962 em->block_start = EXTENT_MAP_HOLE;
6963 insert:
6964 ret = 0;
6965 btrfs_release_path(path);
6966 if (em->start > start || extent_map_end(em) <= start) {
6967 btrfs_err(fs_info,
6968 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6969 em->start, em->len, start, len);
6970 ret = -EIO;
6971 goto out;
6972 }
6973
6974 write_lock(&em_tree->lock);
6975 ret = btrfs_add_extent_mapping(inode, &em, start, len);
6976 write_unlock(&em_tree->lock);
6977 out:
6978 btrfs_free_path(path);
6979
6980 trace_btrfs_get_extent(root, inode, em);
6981
6982 if (ret) {
6983 free_extent_map(em);
6984 return ERR_PTR(ret);
6985 }
6986 return em;
6987 }
6988
btrfs_create_dio_extent(struct btrfs_inode * inode,struct btrfs_dio_data * dio_data,const u64 start,const u64 len,const u64 orig_start,const u64 block_start,const u64 block_len,const u64 orig_block_len,const u64 ram_bytes,const int type)6989 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
6990 struct btrfs_dio_data *dio_data,
6991 const u64 start,
6992 const u64 len,
6993 const u64 orig_start,
6994 const u64 block_start,
6995 const u64 block_len,
6996 const u64 orig_block_len,
6997 const u64 ram_bytes,
6998 const int type)
6999 {
7000 struct extent_map *em = NULL;
7001 struct btrfs_ordered_extent *ordered;
7002
7003 if (type != BTRFS_ORDERED_NOCOW) {
7004 em = create_io_em(inode, start, len, orig_start, block_start,
7005 block_len, orig_block_len, ram_bytes,
7006 BTRFS_COMPRESS_NONE, /* compress_type */
7007 type);
7008 if (IS_ERR(em))
7009 goto out;
7010 }
7011 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
7012 block_start, block_len, 0,
7013 (1 << type) |
7014 (1 << BTRFS_ORDERED_DIRECT),
7015 BTRFS_COMPRESS_NONE);
7016 if (IS_ERR(ordered)) {
7017 if (em) {
7018 free_extent_map(em);
7019 btrfs_drop_extent_map_range(inode, start,
7020 start + len - 1, false);
7021 }
7022 em = ERR_CAST(ordered);
7023 } else {
7024 ASSERT(!dio_data->ordered);
7025 dio_data->ordered = ordered;
7026 }
7027 out:
7028
7029 return em;
7030 }
7031
btrfs_new_extent_direct(struct btrfs_inode * inode,struct btrfs_dio_data * dio_data,u64 start,u64 len)7032 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7033 struct btrfs_dio_data *dio_data,
7034 u64 start, u64 len)
7035 {
7036 struct btrfs_root *root = inode->root;
7037 struct btrfs_fs_info *fs_info = root->fs_info;
7038 struct extent_map *em;
7039 struct btrfs_key ins;
7040 u64 alloc_hint;
7041 int ret;
7042
7043 alloc_hint = get_extent_allocation_hint(inode, start, len);
7044 again:
7045 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7046 0, alloc_hint, &ins, 1, 1);
7047 if (ret == -EAGAIN) {
7048 ASSERT(btrfs_is_zoned(fs_info));
7049 wait_on_bit_io(&inode->root->fs_info->flags, BTRFS_FS_NEED_ZONE_FINISH,
7050 TASK_UNINTERRUPTIBLE);
7051 goto again;
7052 }
7053 if (ret)
7054 return ERR_PTR(ret);
7055
7056 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
7057 ins.objectid, ins.offset, ins.offset,
7058 ins.offset, BTRFS_ORDERED_REGULAR);
7059 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7060 if (IS_ERR(em))
7061 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7062 1);
7063
7064 return em;
7065 }
7066
btrfs_extent_readonly(struct btrfs_fs_info * fs_info,u64 bytenr)7067 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7068 {
7069 struct btrfs_block_group *block_group;
7070 bool readonly = false;
7071
7072 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7073 if (!block_group || block_group->ro)
7074 readonly = true;
7075 if (block_group)
7076 btrfs_put_block_group(block_group);
7077 return readonly;
7078 }
7079
7080 /*
7081 * Check if we can do nocow write into the range [@offset, @offset + @len)
7082 *
7083 * @offset: File offset
7084 * @len: The length to write, will be updated to the nocow writeable
7085 * range
7086 * @orig_start: (optional) Return the original file offset of the file extent
7087 * @orig_len: (optional) Return the original on-disk length of the file extent
7088 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7089 * @strict: if true, omit optimizations that might force us into unnecessary
7090 * cow. e.g., don't trust generation number.
7091 *
7092 * Return:
7093 * >0 and update @len if we can do nocow write
7094 * 0 if we can't do nocow write
7095 * <0 if error happened
7096 *
7097 * NOTE: This only checks the file extents, caller is responsible to wait for
7098 * any ordered extents.
7099 */
can_nocow_extent(struct inode * inode,u64 offset,u64 * len,u64 * orig_start,u64 * orig_block_len,u64 * ram_bytes,bool nowait,bool strict)7100 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7101 u64 *orig_start, u64 *orig_block_len,
7102 u64 *ram_bytes, bool nowait, bool strict)
7103 {
7104 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
7105 struct can_nocow_file_extent_args nocow_args = { 0 };
7106 struct btrfs_path *path;
7107 int ret;
7108 struct extent_buffer *leaf;
7109 struct btrfs_root *root = BTRFS_I(inode)->root;
7110 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7111 struct btrfs_file_extent_item *fi;
7112 struct btrfs_key key;
7113 int found_type;
7114
7115 path = btrfs_alloc_path();
7116 if (!path)
7117 return -ENOMEM;
7118 path->nowait = nowait;
7119
7120 ret = btrfs_lookup_file_extent(NULL, root, path,
7121 btrfs_ino(BTRFS_I(inode)), offset, 0);
7122 if (ret < 0)
7123 goto out;
7124
7125 if (ret == 1) {
7126 if (path->slots[0] == 0) {
7127 /* can't find the item, must cow */
7128 ret = 0;
7129 goto out;
7130 }
7131 path->slots[0]--;
7132 }
7133 ret = 0;
7134 leaf = path->nodes[0];
7135 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7136 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7137 key.type != BTRFS_EXTENT_DATA_KEY) {
7138 /* not our file or wrong item type, must cow */
7139 goto out;
7140 }
7141
7142 if (key.offset > offset) {
7143 /* Wrong offset, must cow */
7144 goto out;
7145 }
7146
7147 if (btrfs_file_extent_end(path) <= offset)
7148 goto out;
7149
7150 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7151 found_type = btrfs_file_extent_type(leaf, fi);
7152 if (ram_bytes)
7153 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7154
7155 nocow_args.start = offset;
7156 nocow_args.end = offset + *len - 1;
7157 nocow_args.strict = strict;
7158 nocow_args.free_path = true;
7159
7160 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7161 /* can_nocow_file_extent() has freed the path. */
7162 path = NULL;
7163
7164 if (ret != 1) {
7165 /* Treat errors as not being able to NOCOW. */
7166 ret = 0;
7167 goto out;
7168 }
7169
7170 ret = 0;
7171 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7172 goto out;
7173
7174 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7175 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7176 u64 range_end;
7177
7178 range_end = round_up(offset + nocow_args.num_bytes,
7179 root->fs_info->sectorsize) - 1;
7180 ret = test_range_bit_exists(io_tree, offset, range_end, EXTENT_DELALLOC);
7181 if (ret) {
7182 ret = -EAGAIN;
7183 goto out;
7184 }
7185 }
7186
7187 if (orig_start)
7188 *orig_start = key.offset - nocow_args.extent_offset;
7189 if (orig_block_len)
7190 *orig_block_len = nocow_args.disk_num_bytes;
7191
7192 *len = nocow_args.num_bytes;
7193 ret = 1;
7194 out:
7195 btrfs_free_path(path);
7196 return ret;
7197 }
7198
lock_extent_direct(struct inode * inode,u64 lockstart,u64 lockend,struct extent_state ** cached_state,unsigned int iomap_flags)7199 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7200 struct extent_state **cached_state,
7201 unsigned int iomap_flags)
7202 {
7203 const bool writing = (iomap_flags & IOMAP_WRITE);
7204 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7205 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7206 struct btrfs_ordered_extent *ordered;
7207 int ret = 0;
7208
7209 while (1) {
7210 if (nowait) {
7211 if (!try_lock_extent(io_tree, lockstart, lockend,
7212 cached_state))
7213 return -EAGAIN;
7214 } else {
7215 lock_extent(io_tree, lockstart, lockend, cached_state);
7216 }
7217 /*
7218 * We're concerned with the entire range that we're going to be
7219 * doing DIO to, so we need to make sure there's no ordered
7220 * extents in this range.
7221 */
7222 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7223 lockend - lockstart + 1);
7224
7225 /*
7226 * We need to make sure there are no buffered pages in this
7227 * range either, we could have raced between the invalidate in
7228 * generic_file_direct_write and locking the extent. The
7229 * invalidate needs to happen so that reads after a write do not
7230 * get stale data.
7231 */
7232 if (!ordered &&
7233 (!writing || !filemap_range_has_page(inode->i_mapping,
7234 lockstart, lockend)))
7235 break;
7236
7237 unlock_extent(io_tree, lockstart, lockend, cached_state);
7238
7239 if (ordered) {
7240 if (nowait) {
7241 btrfs_put_ordered_extent(ordered);
7242 ret = -EAGAIN;
7243 break;
7244 }
7245 /*
7246 * If we are doing a DIO read and the ordered extent we
7247 * found is for a buffered write, we can not wait for it
7248 * to complete and retry, because if we do so we can
7249 * deadlock with concurrent buffered writes on page
7250 * locks. This happens only if our DIO read covers more
7251 * than one extent map, if at this point has already
7252 * created an ordered extent for a previous extent map
7253 * and locked its range in the inode's io tree, and a
7254 * concurrent write against that previous extent map's
7255 * range and this range started (we unlock the ranges
7256 * in the io tree only when the bios complete and
7257 * buffered writes always lock pages before attempting
7258 * to lock range in the io tree).
7259 */
7260 if (writing ||
7261 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7262 btrfs_start_ordered_extent(ordered);
7263 else
7264 ret = nowait ? -EAGAIN : -ENOTBLK;
7265 btrfs_put_ordered_extent(ordered);
7266 } else {
7267 /*
7268 * We could trigger writeback for this range (and wait
7269 * for it to complete) and then invalidate the pages for
7270 * this range (through invalidate_inode_pages2_range()),
7271 * but that can lead us to a deadlock with a concurrent
7272 * call to readahead (a buffered read or a defrag call
7273 * triggered a readahead) on a page lock due to an
7274 * ordered dio extent we created before but did not have
7275 * yet a corresponding bio submitted (whence it can not
7276 * complete), which makes readahead wait for that
7277 * ordered extent to complete while holding a lock on
7278 * that page.
7279 */
7280 ret = nowait ? -EAGAIN : -ENOTBLK;
7281 }
7282
7283 if (ret)
7284 break;
7285
7286 cond_resched();
7287 }
7288
7289 return ret;
7290 }
7291
7292 /* The callers of this must take lock_extent() */
create_io_em(struct btrfs_inode * inode,u64 start,u64 len,u64 orig_start,u64 block_start,u64 block_len,u64 orig_block_len,u64 ram_bytes,int compress_type,int type)7293 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7294 u64 len, u64 orig_start, u64 block_start,
7295 u64 block_len, u64 orig_block_len,
7296 u64 ram_bytes, int compress_type,
7297 int type)
7298 {
7299 struct extent_map *em;
7300 int ret;
7301
7302 /*
7303 * Note the missing NOCOW type.
7304 *
7305 * For pure NOCOW writes, we should not create an io extent map, but
7306 * just reusing the existing one.
7307 * Only PREALLOC writes (NOCOW write into preallocated range) can
7308 * create an io extent map.
7309 */
7310 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7311 type == BTRFS_ORDERED_COMPRESSED ||
7312 type == BTRFS_ORDERED_REGULAR);
7313
7314 switch (type) {
7315 case BTRFS_ORDERED_PREALLOC:
7316 /* Uncompressed extents. */
7317 ASSERT(block_len == len);
7318
7319 /* We're only referring part of a larger preallocated extent. */
7320 ASSERT(block_len <= ram_bytes);
7321 break;
7322 case BTRFS_ORDERED_REGULAR:
7323 /* Uncompressed extents. */
7324 ASSERT(block_len == len);
7325
7326 /* COW results a new extent matching our file extent size. */
7327 ASSERT(orig_block_len == len);
7328 ASSERT(ram_bytes == len);
7329
7330 /* Since it's a new extent, we should not have any offset. */
7331 ASSERT(orig_start == start);
7332 break;
7333 case BTRFS_ORDERED_COMPRESSED:
7334 /* Must be compressed. */
7335 ASSERT(compress_type != BTRFS_COMPRESS_NONE);
7336
7337 /*
7338 * Encoded write can make us to refer to part of the
7339 * uncompressed extent.
7340 */
7341 ASSERT(len <= ram_bytes);
7342 break;
7343 }
7344
7345 em = alloc_extent_map();
7346 if (!em)
7347 return ERR_PTR(-ENOMEM);
7348
7349 em->start = start;
7350 em->orig_start = orig_start;
7351 em->len = len;
7352 em->block_len = block_len;
7353 em->block_start = block_start;
7354 em->orig_block_len = orig_block_len;
7355 em->ram_bytes = ram_bytes;
7356 em->generation = -1;
7357 em->flags |= EXTENT_FLAG_PINNED;
7358 if (type == BTRFS_ORDERED_COMPRESSED)
7359 extent_map_set_compression(em, compress_type);
7360
7361 ret = btrfs_replace_extent_map_range(inode, em, true);
7362 if (ret) {
7363 free_extent_map(em);
7364 return ERR_PTR(ret);
7365 }
7366
7367 /* em got 2 refs now, callers needs to do free_extent_map once. */
7368 return em;
7369 }
7370
7371
btrfs_get_blocks_direct_write(struct extent_map ** map,struct inode * inode,struct btrfs_dio_data * dio_data,u64 start,u64 * lenp,unsigned int iomap_flags)7372 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7373 struct inode *inode,
7374 struct btrfs_dio_data *dio_data,
7375 u64 start, u64 *lenp,
7376 unsigned int iomap_flags)
7377 {
7378 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7379 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
7380 struct extent_map *em = *map;
7381 int type;
7382 u64 block_start, orig_start, orig_block_len, ram_bytes;
7383 struct btrfs_block_group *bg;
7384 bool can_nocow = false;
7385 bool space_reserved = false;
7386 u64 len = *lenp;
7387 u64 prev_len;
7388 int ret = 0;
7389
7390 /*
7391 * We don't allocate a new extent in the following cases
7392 *
7393 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7394 * existing extent.
7395 * 2) The extent is marked as PREALLOC. We're good to go here and can
7396 * just use the extent.
7397 *
7398 */
7399 if ((em->flags & EXTENT_FLAG_PREALLOC) ||
7400 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7401 em->block_start != EXTENT_MAP_HOLE)) {
7402 if (em->flags & EXTENT_FLAG_PREALLOC)
7403 type = BTRFS_ORDERED_PREALLOC;
7404 else
7405 type = BTRFS_ORDERED_NOCOW;
7406 len = min(len, em->len - (start - em->start));
7407 block_start = em->block_start + (start - em->start);
7408
7409 if (can_nocow_extent(inode, start, &len, &orig_start,
7410 &orig_block_len, &ram_bytes, false, false) == 1) {
7411 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7412 if (bg)
7413 can_nocow = true;
7414 }
7415 }
7416
7417 prev_len = len;
7418 if (can_nocow) {
7419 struct extent_map *em2;
7420
7421 /* We can NOCOW, so only need to reserve metadata space. */
7422 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7423 nowait);
7424 if (ret < 0) {
7425 /* Our caller expects us to free the input extent map. */
7426 free_extent_map(em);
7427 *map = NULL;
7428 btrfs_dec_nocow_writers(bg);
7429 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7430 ret = -EAGAIN;
7431 goto out;
7432 }
7433 space_reserved = true;
7434
7435 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7436 orig_start, block_start,
7437 len, orig_block_len,
7438 ram_bytes, type);
7439 btrfs_dec_nocow_writers(bg);
7440 if (type == BTRFS_ORDERED_PREALLOC) {
7441 free_extent_map(em);
7442 *map = em2;
7443 em = em2;
7444 }
7445
7446 if (IS_ERR(em2)) {
7447 ret = PTR_ERR(em2);
7448 goto out;
7449 }
7450
7451 dio_data->nocow_done = true;
7452 } else {
7453 /* Our caller expects us to free the input extent map. */
7454 free_extent_map(em);
7455 *map = NULL;
7456
7457 if (nowait) {
7458 ret = -EAGAIN;
7459 goto out;
7460 }
7461
7462 /*
7463 * If we could not allocate data space before locking the file
7464 * range and we can't do a NOCOW write, then we have to fail.
7465 */
7466 if (!dio_data->data_space_reserved) {
7467 ret = -ENOSPC;
7468 goto out;
7469 }
7470
7471 /*
7472 * We have to COW and we have already reserved data space before,
7473 * so now we reserve only metadata.
7474 */
7475 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7476 false);
7477 if (ret < 0)
7478 goto out;
7479 space_reserved = true;
7480
7481 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7482 if (IS_ERR(em)) {
7483 ret = PTR_ERR(em);
7484 goto out;
7485 }
7486 *map = em;
7487 len = min(len, em->len - (start - em->start));
7488 if (len < prev_len)
7489 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7490 prev_len - len, true);
7491 }
7492
7493 /*
7494 * We have created our ordered extent, so we can now release our reservation
7495 * for an outstanding extent.
7496 */
7497 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7498
7499 /*
7500 * Need to update the i_size under the extent lock so buffered
7501 * readers will get the updated i_size when we unlock.
7502 */
7503 if (start + len > i_size_read(inode))
7504 i_size_write(inode, start + len);
7505 out:
7506 if (ret && space_reserved) {
7507 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7508 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7509 }
7510 *lenp = len;
7511 return ret;
7512 }
7513
btrfs_dio_iomap_begin(struct inode * inode,loff_t start,loff_t length,unsigned int flags,struct iomap * iomap,struct iomap * srcmap)7514 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7515 loff_t length, unsigned int flags, struct iomap *iomap,
7516 struct iomap *srcmap)
7517 {
7518 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7519 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
7520 struct extent_map *em;
7521 struct extent_state *cached_state = NULL;
7522 struct btrfs_dio_data *dio_data = iter->private;
7523 u64 lockstart, lockend;
7524 const bool write = !!(flags & IOMAP_WRITE);
7525 int ret = 0;
7526 u64 len = length;
7527 const u64 data_alloc_len = length;
7528 bool unlock_extents = false;
7529
7530 /*
7531 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7532 * we're NOWAIT we may submit a bio for a partial range and return
7533 * EIOCBQUEUED, which would result in an errant short read.
7534 *
7535 * The best way to handle this would be to allow for partial completions
7536 * of iocb's, so we could submit the partial bio, return and fault in
7537 * the rest of the pages, and then submit the io for the rest of the
7538 * range. However we don't have that currently, so simply return
7539 * -EAGAIN at this point so that the normal path is used.
7540 */
7541 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7542 return -EAGAIN;
7543
7544 /*
7545 * Cap the size of reads to that usually seen in buffered I/O as we need
7546 * to allocate a contiguous array for the checksums.
7547 */
7548 if (!write)
7549 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7550
7551 lockstart = start;
7552 lockend = start + len - 1;
7553
7554 /*
7555 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7556 * enough if we've written compressed pages to this area, so we need to
7557 * flush the dirty pages again to make absolutely sure that any
7558 * outstanding dirty pages are on disk - the first flush only starts
7559 * compression on the data, while keeping the pages locked, so by the
7560 * time the second flush returns we know bios for the compressed pages
7561 * were submitted and finished, and the pages no longer under writeback.
7562 *
7563 * If we have a NOWAIT request and we have any pages in the range that
7564 * are locked, likely due to compression still in progress, we don't want
7565 * to block on page locks. We also don't want to block on pages marked as
7566 * dirty or under writeback (same as for the non-compression case).
7567 * iomap_dio_rw() did the same check, but after that and before we got
7568 * here, mmap'ed writes may have happened or buffered reads started
7569 * (readpage() and readahead(), which lock pages), as we haven't locked
7570 * the file range yet.
7571 */
7572 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7573 &BTRFS_I(inode)->runtime_flags)) {
7574 if (flags & IOMAP_NOWAIT) {
7575 if (filemap_range_needs_writeback(inode->i_mapping,
7576 lockstart, lockend))
7577 return -EAGAIN;
7578 } else {
7579 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7580 start + length - 1);
7581 if (ret)
7582 return ret;
7583 }
7584 }
7585
7586 memset(dio_data, 0, sizeof(*dio_data));
7587
7588 /*
7589 * We always try to allocate data space and must do it before locking
7590 * the file range, to avoid deadlocks with concurrent writes to the same
7591 * range if the range has several extents and the writes don't expand the
7592 * current i_size (the inode lock is taken in shared mode). If we fail to
7593 * allocate data space here we continue and later, after locking the
7594 * file range, we fail with ENOSPC only if we figure out we can not do a
7595 * NOCOW write.
7596 */
7597 if (write && !(flags & IOMAP_NOWAIT)) {
7598 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7599 &dio_data->data_reserved,
7600 start, data_alloc_len, false);
7601 if (!ret)
7602 dio_data->data_space_reserved = true;
7603 else if (ret && !(BTRFS_I(inode)->flags &
7604 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7605 goto err;
7606 }
7607
7608 /*
7609 * If this errors out it's because we couldn't invalidate pagecache for
7610 * this range and we need to fallback to buffered IO, or we are doing a
7611 * NOWAIT read/write and we need to block.
7612 */
7613 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7614 if (ret < 0)
7615 goto err;
7616
7617 em = btrfs_get_extent(BTRFS_I(inode), NULL, start, len);
7618 if (IS_ERR(em)) {
7619 ret = PTR_ERR(em);
7620 goto unlock_err;
7621 }
7622
7623 /*
7624 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7625 * io. INLINE is special, and we could probably kludge it in here, but
7626 * it's still buffered so for safety lets just fall back to the generic
7627 * buffered path.
7628 *
7629 * For COMPRESSED we _have_ to read the entire extent in so we can
7630 * decompress it, so there will be buffering required no matter what we
7631 * do, so go ahead and fallback to buffered.
7632 *
7633 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7634 * to buffered IO. Don't blame me, this is the price we pay for using
7635 * the generic code.
7636 */
7637 if (extent_map_is_compressed(em) ||
7638 em->block_start == EXTENT_MAP_INLINE) {
7639 free_extent_map(em);
7640 /*
7641 * If we are in a NOWAIT context, return -EAGAIN in order to
7642 * fallback to buffered IO. This is not only because we can
7643 * block with buffered IO (no support for NOWAIT semantics at
7644 * the moment) but also to avoid returning short reads to user
7645 * space - this happens if we were able to read some data from
7646 * previous non-compressed extents and then when we fallback to
7647 * buffered IO, at btrfs_file_read_iter() by calling
7648 * filemap_read(), we fail to fault in pages for the read buffer,
7649 * in which case filemap_read() returns a short read (the number
7650 * of bytes previously read is > 0, so it does not return -EFAULT).
7651 */
7652 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7653 goto unlock_err;
7654 }
7655
7656 len = min(len, em->len - (start - em->start));
7657
7658 /*
7659 * If we have a NOWAIT request and the range contains multiple extents
7660 * (or a mix of extents and holes), then we return -EAGAIN to make the
7661 * caller fallback to a context where it can do a blocking (without
7662 * NOWAIT) request. This way we avoid doing partial IO and returning
7663 * success to the caller, which is not optimal for writes and for reads
7664 * it can result in unexpected behaviour for an application.
7665 *
7666 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7667 * iomap_dio_rw(), we can end up returning less data then what the caller
7668 * asked for, resulting in an unexpected, and incorrect, short read.
7669 * That is, the caller asked to read N bytes and we return less than that,
7670 * which is wrong unless we are crossing EOF. This happens if we get a
7671 * page fault error when trying to fault in pages for the buffer that is
7672 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7673 * have previously submitted bios for other extents in the range, in
7674 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7675 * those bios have completed by the time we get the page fault error,
7676 * which we return back to our caller - we should only return EIOCBQUEUED
7677 * after we have submitted bios for all the extents in the range.
7678 */
7679 if ((flags & IOMAP_NOWAIT) && len < length) {
7680 free_extent_map(em);
7681 ret = -EAGAIN;
7682 goto unlock_err;
7683 }
7684
7685 if (write) {
7686 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7687 start, &len, flags);
7688 if (ret < 0)
7689 goto unlock_err;
7690 unlock_extents = true;
7691 /* Recalc len in case the new em is smaller than requested */
7692 len = min(len, em->len - (start - em->start));
7693 if (dio_data->data_space_reserved) {
7694 u64 release_offset;
7695 u64 release_len = 0;
7696
7697 if (dio_data->nocow_done) {
7698 release_offset = start;
7699 release_len = data_alloc_len;
7700 } else if (len < data_alloc_len) {
7701 release_offset = start + len;
7702 release_len = data_alloc_len - len;
7703 }
7704
7705 if (release_len > 0)
7706 btrfs_free_reserved_data_space(BTRFS_I(inode),
7707 dio_data->data_reserved,
7708 release_offset,
7709 release_len);
7710 }
7711 } else {
7712 /*
7713 * We need to unlock only the end area that we aren't using.
7714 * The rest is going to be unlocked by the endio routine.
7715 */
7716 lockstart = start + len;
7717 if (lockstart < lockend)
7718 unlock_extents = true;
7719 }
7720
7721 if (unlock_extents)
7722 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7723 &cached_state);
7724 else
7725 free_extent_state(cached_state);
7726
7727 /*
7728 * Translate extent map information to iomap.
7729 * We trim the extents (and move the addr) even though iomap code does
7730 * that, since we have locked only the parts we are performing I/O in.
7731 */
7732 if ((em->block_start == EXTENT_MAP_HOLE) ||
7733 ((em->flags & EXTENT_FLAG_PREALLOC) && !write)) {
7734 iomap->addr = IOMAP_NULL_ADDR;
7735 iomap->type = IOMAP_HOLE;
7736 } else {
7737 iomap->addr = em->block_start + (start - em->start);
7738 iomap->type = IOMAP_MAPPED;
7739 }
7740 iomap->offset = start;
7741 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7742 iomap->length = len;
7743 free_extent_map(em);
7744
7745 return 0;
7746
7747 unlock_err:
7748 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7749 &cached_state);
7750 err:
7751 if (dio_data->data_space_reserved) {
7752 btrfs_free_reserved_data_space(BTRFS_I(inode),
7753 dio_data->data_reserved,
7754 start, data_alloc_len);
7755 extent_changeset_free(dio_data->data_reserved);
7756 }
7757
7758 return ret;
7759 }
7760
btrfs_dio_iomap_end(struct inode * inode,loff_t pos,loff_t length,ssize_t written,unsigned int flags,struct iomap * iomap)7761 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7762 ssize_t written, unsigned int flags, struct iomap *iomap)
7763 {
7764 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7765 struct btrfs_dio_data *dio_data = iter->private;
7766 size_t submitted = dio_data->submitted;
7767 const bool write = !!(flags & IOMAP_WRITE);
7768 int ret = 0;
7769
7770 if (!write && (iomap->type == IOMAP_HOLE)) {
7771 /* If reading from a hole, unlock and return */
7772 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7773 NULL);
7774 return 0;
7775 }
7776
7777 if (submitted < length) {
7778 pos += submitted;
7779 length -= submitted;
7780 if (write)
7781 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7782 pos, length, false);
7783 else
7784 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7785 pos + length - 1, NULL);
7786 ret = -ENOTBLK;
7787 }
7788 if (write) {
7789 btrfs_put_ordered_extent(dio_data->ordered);
7790 dio_data->ordered = NULL;
7791 }
7792
7793 if (write)
7794 extent_changeset_free(dio_data->data_reserved);
7795 return ret;
7796 }
7797
btrfs_dio_end_io(struct btrfs_bio * bbio)7798 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7799 {
7800 struct btrfs_dio_private *dip =
7801 container_of(bbio, struct btrfs_dio_private, bbio);
7802 struct btrfs_inode *inode = bbio->inode;
7803 struct bio *bio = &bbio->bio;
7804
7805 if (bio->bi_status) {
7806 btrfs_warn(inode->root->fs_info,
7807 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7808 btrfs_ino(inode), bio->bi_opf,
7809 dip->file_offset, dip->bytes, bio->bi_status);
7810 }
7811
7812 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7813 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7814 dip->file_offset, dip->bytes,
7815 !bio->bi_status);
7816 } else {
7817 unlock_extent(&inode->io_tree, dip->file_offset,
7818 dip->file_offset + dip->bytes - 1, NULL);
7819 }
7820
7821 bbio->bio.bi_private = bbio->private;
7822 iomap_dio_bio_end_io(bio);
7823 }
7824
btrfs_dio_submit_io(const struct iomap_iter * iter,struct bio * bio,loff_t file_offset)7825 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7826 loff_t file_offset)
7827 {
7828 struct btrfs_bio *bbio = btrfs_bio(bio);
7829 struct btrfs_dio_private *dip =
7830 container_of(bbio, struct btrfs_dio_private, bbio);
7831 struct btrfs_dio_data *dio_data = iter->private;
7832
7833 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7834 btrfs_dio_end_io, bio->bi_private);
7835 bbio->inode = BTRFS_I(iter->inode);
7836 bbio->file_offset = file_offset;
7837
7838 dip->file_offset = file_offset;
7839 dip->bytes = bio->bi_iter.bi_size;
7840
7841 dio_data->submitted += bio->bi_iter.bi_size;
7842
7843 /*
7844 * Check if we are doing a partial write. If we are, we need to split
7845 * the ordered extent to match the submitted bio. Hang on to the
7846 * remaining unfinishable ordered_extent in dio_data so that it can be
7847 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7848 * remaining pages is blocked on the outstanding ordered extent.
7849 */
7850 if (iter->flags & IOMAP_WRITE) {
7851 int ret;
7852
7853 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7854 if (ret) {
7855 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7856 file_offset, dip->bytes,
7857 !ret);
7858 bio->bi_status = errno_to_blk_status(ret);
7859 iomap_dio_bio_end_io(bio);
7860 return;
7861 }
7862 }
7863
7864 btrfs_submit_bio(bbio, 0);
7865 }
7866
7867 static const struct iomap_ops btrfs_dio_iomap_ops = {
7868 .iomap_begin = btrfs_dio_iomap_begin,
7869 .iomap_end = btrfs_dio_iomap_end,
7870 };
7871
7872 static const struct iomap_dio_ops btrfs_dio_ops = {
7873 .submit_io = btrfs_dio_submit_io,
7874 .bio_set = &btrfs_dio_bioset,
7875 };
7876
btrfs_dio_read(struct kiocb * iocb,struct iov_iter * iter,size_t done_before)7877 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7878 {
7879 struct btrfs_dio_data data = { 0 };
7880
7881 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7882 IOMAP_DIO_PARTIAL, &data, done_before);
7883 }
7884
btrfs_dio_write(struct kiocb * iocb,struct iov_iter * iter,size_t done_before)7885 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7886 size_t done_before)
7887 {
7888 struct btrfs_dio_data data = { 0 };
7889
7890 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7891 IOMAP_DIO_PARTIAL, &data, done_before);
7892 }
7893
btrfs_fiemap(struct inode * inode,struct fiemap_extent_info * fieinfo,u64 start,u64 len)7894 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7895 u64 start, u64 len)
7896 {
7897 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
7898 int ret;
7899
7900 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7901 if (ret)
7902 return ret;
7903
7904 /*
7905 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7906 * file range (0 to LLONG_MAX), but that is not enough if we have
7907 * compression enabled. The first filemap_fdatawrite_range() only kicks
7908 * in the compression of data (in an async thread) and will return
7909 * before the compression is done and writeback is started. A second
7910 * filemap_fdatawrite_range() is needed to wait for the compression to
7911 * complete and writeback to start. We also need to wait for ordered
7912 * extents to complete, because our fiemap implementation uses mainly
7913 * file extent items to list the extents, searching for extent maps
7914 * only for file ranges with holes or prealloc extents to figure out
7915 * if we have delalloc in those ranges.
7916 */
7917 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7918 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7919 if (ret)
7920 return ret;
7921 }
7922
7923 btrfs_inode_lock(btrfs_inode, BTRFS_ILOCK_SHARED);
7924
7925 /*
7926 * We did an initial flush to avoid holding the inode's lock while
7927 * triggering writeback and waiting for the completion of IO and ordered
7928 * extents. Now after we locked the inode we do it again, because it's
7929 * possible a new write may have happened in between those two steps.
7930 */
7931 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7932 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7933 if (ret) {
7934 btrfs_inode_unlock(btrfs_inode, BTRFS_ILOCK_SHARED);
7935 return ret;
7936 }
7937 }
7938
7939 ret = extent_fiemap(btrfs_inode, fieinfo, start, len);
7940 btrfs_inode_unlock(btrfs_inode, BTRFS_ILOCK_SHARED);
7941
7942 return ret;
7943 }
7944
7945 /*
7946 * For release_folio() and invalidate_folio() we have a race window where
7947 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7948 * If we continue to release/invalidate the page, we could cause use-after-free
7949 * for subpage spinlock. So this function is to spin and wait for subpage
7950 * spinlock.
7951 */
wait_subpage_spinlock(struct page * page)7952 static void wait_subpage_spinlock(struct page *page)
7953 {
7954 struct btrfs_fs_info *fs_info = page_to_fs_info(page);
7955 struct folio *folio = page_folio(page);
7956 struct btrfs_subpage *subpage;
7957
7958 if (!btrfs_is_subpage(fs_info, page->mapping))
7959 return;
7960
7961 ASSERT(folio_test_private(folio) && folio_get_private(folio));
7962 subpage = folio_get_private(folio);
7963
7964 /*
7965 * This may look insane as we just acquire the spinlock and release it,
7966 * without doing anything. But we just want to make sure no one is
7967 * still holding the subpage spinlock.
7968 * And since the page is not dirty nor writeback, and we have page
7969 * locked, the only possible way to hold a spinlock is from the endio
7970 * function to clear page writeback.
7971 *
7972 * Here we just acquire the spinlock so that all existing callers
7973 * should exit and we're safe to release/invalidate the page.
7974 */
7975 spin_lock_irq(&subpage->lock);
7976 spin_unlock_irq(&subpage->lock);
7977 }
7978
__btrfs_release_folio(struct folio * folio,gfp_t gfp_flags)7979 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7980 {
7981 if (try_release_extent_mapping(&folio->page, gfp_flags)) {
7982 wait_subpage_spinlock(&folio->page);
7983 clear_page_extent_mapped(&folio->page);
7984 return true;
7985 }
7986 return false;
7987 }
7988
btrfs_release_folio(struct folio * folio,gfp_t gfp_flags)7989 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7990 {
7991 if (folio_test_writeback(folio) || folio_test_dirty(folio))
7992 return false;
7993 return __btrfs_release_folio(folio, gfp_flags);
7994 }
7995
7996 #ifdef CONFIG_MIGRATION
btrfs_migrate_folio(struct address_space * mapping,struct folio * dst,struct folio * src,enum migrate_mode mode)7997 static int btrfs_migrate_folio(struct address_space *mapping,
7998 struct folio *dst, struct folio *src,
7999 enum migrate_mode mode)
8000 {
8001 int ret = filemap_migrate_folio(mapping, dst, src, mode);
8002
8003 if (ret != MIGRATEPAGE_SUCCESS)
8004 return ret;
8005
8006 if (folio_test_ordered(src)) {
8007 folio_clear_ordered(src);
8008 folio_set_ordered(dst);
8009 }
8010
8011 return MIGRATEPAGE_SUCCESS;
8012 }
8013 #else
8014 #define btrfs_migrate_folio NULL
8015 #endif
8016
btrfs_invalidate_folio(struct folio * folio,size_t offset,size_t length)8017 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8018 size_t length)
8019 {
8020 struct btrfs_inode *inode = folio_to_inode(folio);
8021 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8022 struct extent_io_tree *tree = &inode->io_tree;
8023 struct extent_state *cached_state = NULL;
8024 u64 page_start = folio_pos(folio);
8025 u64 page_end = page_start + folio_size(folio) - 1;
8026 u64 cur;
8027 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8028
8029 /*
8030 * We have folio locked so no new ordered extent can be created on this
8031 * page, nor bio can be submitted for this folio.
8032 *
8033 * But already submitted bio can still be finished on this folio.
8034 * Furthermore, endio function won't skip folio which has Ordered
8035 * (Private2) already cleared, so it's possible for endio and
8036 * invalidate_folio to do the same ordered extent accounting twice
8037 * on one folio.
8038 *
8039 * So here we wait for any submitted bios to finish, so that we won't
8040 * do double ordered extent accounting on the same folio.
8041 */
8042 folio_wait_writeback(folio);
8043 wait_subpage_spinlock(&folio->page);
8044
8045 /*
8046 * For subpage case, we have call sites like
8047 * btrfs_punch_hole_lock_range() which passes range not aligned to
8048 * sectorsize.
8049 * If the range doesn't cover the full folio, we don't need to and
8050 * shouldn't clear page extent mapped, as folio->private can still
8051 * record subpage dirty bits for other part of the range.
8052 *
8053 * For cases that invalidate the full folio even the range doesn't
8054 * cover the full folio, like invalidating the last folio, we're
8055 * still safe to wait for ordered extent to finish.
8056 */
8057 if (!(offset == 0 && length == folio_size(folio))) {
8058 btrfs_release_folio(folio, GFP_NOFS);
8059 return;
8060 }
8061
8062 if (!inode_evicting)
8063 lock_extent(tree, page_start, page_end, &cached_state);
8064
8065 cur = page_start;
8066 while (cur < page_end) {
8067 struct btrfs_ordered_extent *ordered;
8068 u64 range_end;
8069 u32 range_len;
8070 u32 extra_flags = 0;
8071
8072 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8073 page_end + 1 - cur);
8074 if (!ordered) {
8075 range_end = page_end;
8076 /*
8077 * No ordered extent covering this range, we are safe
8078 * to delete all extent states in the range.
8079 */
8080 extra_flags = EXTENT_CLEAR_ALL_BITS;
8081 goto next;
8082 }
8083 if (ordered->file_offset > cur) {
8084 /*
8085 * There is a range between [cur, oe->file_offset) not
8086 * covered by any ordered extent.
8087 * We are safe to delete all extent states, and handle
8088 * the ordered extent in the next iteration.
8089 */
8090 range_end = ordered->file_offset - 1;
8091 extra_flags = EXTENT_CLEAR_ALL_BITS;
8092 goto next;
8093 }
8094
8095 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8096 page_end);
8097 ASSERT(range_end + 1 - cur < U32_MAX);
8098 range_len = range_end + 1 - cur;
8099 if (!btrfs_folio_test_ordered(fs_info, folio, cur, range_len)) {
8100 /*
8101 * If Ordered (Private2) is cleared, it means endio has
8102 * already been executed for the range.
8103 * We can't delete the extent states as
8104 * btrfs_finish_ordered_io() may still use some of them.
8105 */
8106 goto next;
8107 }
8108 btrfs_folio_clear_ordered(fs_info, folio, cur, range_len);
8109
8110 /*
8111 * IO on this page will never be started, so we need to account
8112 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8113 * here, must leave that up for the ordered extent completion.
8114 *
8115 * This will also unlock the range for incoming
8116 * btrfs_finish_ordered_io().
8117 */
8118 if (!inode_evicting)
8119 clear_extent_bit(tree, cur, range_end,
8120 EXTENT_DELALLOC |
8121 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8122 EXTENT_DEFRAG, &cached_state);
8123
8124 spin_lock_irq(&inode->ordered_tree_lock);
8125 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8126 ordered->truncated_len = min(ordered->truncated_len,
8127 cur - ordered->file_offset);
8128 spin_unlock_irq(&inode->ordered_tree_lock);
8129
8130 /*
8131 * If the ordered extent has finished, we're safe to delete all
8132 * the extent states of the range, otherwise
8133 * btrfs_finish_ordered_io() will get executed by endio for
8134 * other pages, so we can't delete extent states.
8135 */
8136 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8137 cur, range_end + 1 - cur)) {
8138 btrfs_finish_ordered_io(ordered);
8139 /*
8140 * The ordered extent has finished, now we're again
8141 * safe to delete all extent states of the range.
8142 */
8143 extra_flags = EXTENT_CLEAR_ALL_BITS;
8144 }
8145 next:
8146 if (ordered)
8147 btrfs_put_ordered_extent(ordered);
8148 /*
8149 * Qgroup reserved space handler
8150 * Sector(s) here will be either:
8151 *
8152 * 1) Already written to disk or bio already finished
8153 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8154 * Qgroup will be handled by its qgroup_record then.
8155 * btrfs_qgroup_free_data() call will do nothing here.
8156 *
8157 * 2) Not written to disk yet
8158 * Then btrfs_qgroup_free_data() call will clear the
8159 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8160 * reserved data space.
8161 * Since the IO will never happen for this page.
8162 */
8163 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur, NULL);
8164 if (!inode_evicting) {
8165 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8166 EXTENT_DELALLOC | EXTENT_UPTODATE |
8167 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8168 extra_flags, &cached_state);
8169 }
8170 cur = range_end + 1;
8171 }
8172 /*
8173 * We have iterated through all ordered extents of the page, the page
8174 * should not have Ordered (Private2) anymore, or the above iteration
8175 * did something wrong.
8176 */
8177 ASSERT(!folio_test_ordered(folio));
8178 btrfs_folio_clear_checked(fs_info, folio, folio_pos(folio), folio_size(folio));
8179 if (!inode_evicting)
8180 __btrfs_release_folio(folio, GFP_NOFS);
8181 clear_page_extent_mapped(&folio->page);
8182 }
8183
btrfs_truncate(struct btrfs_inode * inode,bool skip_writeback)8184 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8185 {
8186 struct btrfs_truncate_control control = {
8187 .inode = inode,
8188 .ino = btrfs_ino(inode),
8189 .min_type = BTRFS_EXTENT_DATA_KEY,
8190 .clear_extent_range = true,
8191 };
8192 struct btrfs_root *root = inode->root;
8193 struct btrfs_fs_info *fs_info = root->fs_info;
8194 struct btrfs_block_rsv *rsv;
8195 int ret;
8196 struct btrfs_trans_handle *trans;
8197 u64 mask = fs_info->sectorsize - 1;
8198 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8199
8200 if (!skip_writeback) {
8201 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8202 inode->vfs_inode.i_size & (~mask),
8203 (u64)-1);
8204 if (ret)
8205 return ret;
8206 }
8207
8208 /*
8209 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8210 * things going on here:
8211 *
8212 * 1) We need to reserve space to update our inode.
8213 *
8214 * 2) We need to have something to cache all the space that is going to
8215 * be free'd up by the truncate operation, but also have some slack
8216 * space reserved in case it uses space during the truncate (thank you
8217 * very much snapshotting).
8218 *
8219 * And we need these to be separate. The fact is we can use a lot of
8220 * space doing the truncate, and we have no earthly idea how much space
8221 * we will use, so we need the truncate reservation to be separate so it
8222 * doesn't end up using space reserved for updating the inode. We also
8223 * need to be able to stop the transaction and start a new one, which
8224 * means we need to be able to update the inode several times, and we
8225 * have no idea of knowing how many times that will be, so we can't just
8226 * reserve 1 item for the entirety of the operation, so that has to be
8227 * done separately as well.
8228 *
8229 * So that leaves us with
8230 *
8231 * 1) rsv - for the truncate reservation, which we will steal from the
8232 * transaction reservation.
8233 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8234 * updating the inode.
8235 */
8236 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8237 if (!rsv)
8238 return -ENOMEM;
8239 rsv->size = min_size;
8240 rsv->failfast = true;
8241
8242 /*
8243 * 1 for the truncate slack space
8244 * 1 for updating the inode.
8245 */
8246 trans = btrfs_start_transaction(root, 2);
8247 if (IS_ERR(trans)) {
8248 ret = PTR_ERR(trans);
8249 goto out;
8250 }
8251
8252 /* Migrate the slack space for the truncate to our reserve */
8253 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8254 min_size, false);
8255 /*
8256 * We have reserved 2 metadata units when we started the transaction and
8257 * min_size matches 1 unit, so this should never fail, but if it does,
8258 * it's not critical we just fail truncation.
8259 */
8260 if (WARN_ON(ret)) {
8261 btrfs_end_transaction(trans);
8262 goto out;
8263 }
8264
8265 trans->block_rsv = rsv;
8266
8267 while (1) {
8268 struct extent_state *cached_state = NULL;
8269 const u64 new_size = inode->vfs_inode.i_size;
8270 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8271
8272 control.new_size = new_size;
8273 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8274 /*
8275 * We want to drop from the next block forward in case this new
8276 * size is not block aligned since we will be keeping the last
8277 * block of the extent just the way it is.
8278 */
8279 btrfs_drop_extent_map_range(inode,
8280 ALIGN(new_size, fs_info->sectorsize),
8281 (u64)-1, false);
8282
8283 ret = btrfs_truncate_inode_items(trans, root, &control);
8284
8285 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8286 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8287
8288 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8289
8290 trans->block_rsv = &fs_info->trans_block_rsv;
8291 if (ret != -ENOSPC && ret != -EAGAIN)
8292 break;
8293
8294 ret = btrfs_update_inode(trans, inode);
8295 if (ret)
8296 break;
8297
8298 btrfs_end_transaction(trans);
8299 btrfs_btree_balance_dirty(fs_info);
8300
8301 trans = btrfs_start_transaction(root, 2);
8302 if (IS_ERR(trans)) {
8303 ret = PTR_ERR(trans);
8304 trans = NULL;
8305 break;
8306 }
8307
8308 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8309 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8310 rsv, min_size, false);
8311 /*
8312 * We have reserved 2 metadata units when we started the
8313 * transaction and min_size matches 1 unit, so this should never
8314 * fail, but if it does, it's not critical we just fail truncation.
8315 */
8316 if (WARN_ON(ret))
8317 break;
8318
8319 trans->block_rsv = rsv;
8320 }
8321
8322 /*
8323 * We can't call btrfs_truncate_block inside a trans handle as we could
8324 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8325 * know we've truncated everything except the last little bit, and can
8326 * do btrfs_truncate_block and then update the disk_i_size.
8327 */
8328 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8329 btrfs_end_transaction(trans);
8330 btrfs_btree_balance_dirty(fs_info);
8331
8332 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8333 if (ret)
8334 goto out;
8335 trans = btrfs_start_transaction(root, 1);
8336 if (IS_ERR(trans)) {
8337 ret = PTR_ERR(trans);
8338 goto out;
8339 }
8340 btrfs_inode_safe_disk_i_size_write(inode, 0);
8341 }
8342
8343 if (trans) {
8344 int ret2;
8345
8346 trans->block_rsv = &fs_info->trans_block_rsv;
8347 ret2 = btrfs_update_inode(trans, inode);
8348 if (ret2 && !ret)
8349 ret = ret2;
8350
8351 ret2 = btrfs_end_transaction(trans);
8352 if (ret2 && !ret)
8353 ret = ret2;
8354 btrfs_btree_balance_dirty(fs_info);
8355 }
8356 out:
8357 btrfs_free_block_rsv(fs_info, rsv);
8358 /*
8359 * So if we truncate and then write and fsync we normally would just
8360 * write the extents that changed, which is a problem if we need to
8361 * first truncate that entire inode. So set this flag so we write out
8362 * all of the extents in the inode to the sync log so we're completely
8363 * safe.
8364 *
8365 * If no extents were dropped or trimmed we don't need to force the next
8366 * fsync to truncate all the inode's items from the log and re-log them
8367 * all. This means the truncate operation did not change the file size,
8368 * or changed it to a smaller size but there was only an implicit hole
8369 * between the old i_size and the new i_size, and there were no prealloc
8370 * extents beyond i_size to drop.
8371 */
8372 if (control.extents_found > 0)
8373 btrfs_set_inode_full_sync(inode);
8374
8375 return ret;
8376 }
8377
btrfs_new_subvol_inode(struct mnt_idmap * idmap,struct inode * dir)8378 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8379 struct inode *dir)
8380 {
8381 struct inode *inode;
8382
8383 inode = new_inode(dir->i_sb);
8384 if (inode) {
8385 /*
8386 * Subvolumes don't inherit the sgid bit or the parent's gid if
8387 * the parent's sgid bit is set. This is probably a bug.
8388 */
8389 inode_init_owner(idmap, inode, NULL,
8390 S_IFDIR | (~current_umask() & S_IRWXUGO));
8391 inode->i_op = &btrfs_dir_inode_operations;
8392 inode->i_fop = &btrfs_dir_file_operations;
8393 }
8394 return inode;
8395 }
8396
btrfs_alloc_inode(struct super_block * sb)8397 struct inode *btrfs_alloc_inode(struct super_block *sb)
8398 {
8399 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8400 struct btrfs_inode *ei;
8401 struct inode *inode;
8402 struct extent_io_tree *file_extent_tree = NULL;
8403
8404 /* Self tests may pass a NULL fs_info. */
8405 if (fs_info && !btrfs_fs_incompat(fs_info, NO_HOLES)) {
8406 file_extent_tree = kmalloc(sizeof(struct extent_io_tree), GFP_KERNEL);
8407 if (!file_extent_tree)
8408 return NULL;
8409 }
8410
8411 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8412 if (!ei) {
8413 kfree(file_extent_tree);
8414 return NULL;
8415 }
8416
8417 ei->root = NULL;
8418 ei->generation = 0;
8419 ei->last_trans = 0;
8420 ei->last_sub_trans = 0;
8421 ei->logged_trans = 0;
8422 ei->delalloc_bytes = 0;
8423 ei->new_delalloc_bytes = 0;
8424 ei->defrag_bytes = 0;
8425 ei->disk_i_size = 0;
8426 ei->flags = 0;
8427 ei->ro_flags = 0;
8428 ei->csum_bytes = 0;
8429 ei->index_cnt = (u64)-1;
8430 ei->dir_index = 0;
8431 ei->last_unlink_trans = 0;
8432 ei->last_reflink_trans = 0;
8433 ei->last_log_commit = 0;
8434
8435 spin_lock_init(&ei->lock);
8436 ei->outstanding_extents = 0;
8437 if (sb->s_magic != BTRFS_TEST_MAGIC)
8438 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8439 BTRFS_BLOCK_RSV_DELALLOC);
8440 ei->runtime_flags = 0;
8441 ei->prop_compress = BTRFS_COMPRESS_NONE;
8442 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8443
8444 ei->delayed_node = NULL;
8445
8446 ei->i_otime_sec = 0;
8447 ei->i_otime_nsec = 0;
8448
8449 inode = &ei->vfs_inode;
8450 extent_map_tree_init(&ei->extent_tree);
8451
8452 /* This io tree sets the valid inode. */
8453 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8454 ei->io_tree.inode = ei;
8455
8456 ei->file_extent_tree = file_extent_tree;
8457 if (file_extent_tree) {
8458 extent_io_tree_init(fs_info, ei->file_extent_tree,
8459 IO_TREE_INODE_FILE_EXTENT);
8460 /* Lockdep class is set only for the file extent tree. */
8461 lockdep_set_class(&ei->file_extent_tree->lock, &file_extent_tree_class);
8462 }
8463 mutex_init(&ei->log_mutex);
8464 spin_lock_init(&ei->ordered_tree_lock);
8465 ei->ordered_tree = RB_ROOT;
8466 ei->ordered_tree_last = NULL;
8467 INIT_LIST_HEAD(&ei->delalloc_inodes);
8468 INIT_LIST_HEAD(&ei->delayed_iput);
8469 RB_CLEAR_NODE(&ei->rb_node);
8470 init_rwsem(&ei->i_mmap_lock);
8471
8472 return inode;
8473 }
8474
8475 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
btrfs_test_destroy_inode(struct inode * inode)8476 void btrfs_test_destroy_inode(struct inode *inode)
8477 {
8478 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8479 kfree(BTRFS_I(inode)->file_extent_tree);
8480 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8481 }
8482 #endif
8483
btrfs_free_inode(struct inode * inode)8484 void btrfs_free_inode(struct inode *inode)
8485 {
8486 kfree(BTRFS_I(inode)->file_extent_tree);
8487 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8488 }
8489
btrfs_destroy_inode(struct inode * vfs_inode)8490 void btrfs_destroy_inode(struct inode *vfs_inode)
8491 {
8492 struct btrfs_ordered_extent *ordered;
8493 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8494 struct btrfs_root *root = inode->root;
8495 bool freespace_inode;
8496
8497 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8498 WARN_ON(vfs_inode->i_data.nrpages);
8499 WARN_ON(inode->block_rsv.reserved);
8500 WARN_ON(inode->block_rsv.size);
8501 WARN_ON(inode->outstanding_extents);
8502 if (!S_ISDIR(vfs_inode->i_mode)) {
8503 WARN_ON(inode->delalloc_bytes);
8504 WARN_ON(inode->new_delalloc_bytes);
8505 }
8506 WARN_ON(inode->csum_bytes);
8507 WARN_ON(inode->defrag_bytes);
8508
8509 /*
8510 * This can happen where we create an inode, but somebody else also
8511 * created the same inode and we need to destroy the one we already
8512 * created.
8513 */
8514 if (!root)
8515 return;
8516
8517 /*
8518 * If this is a free space inode do not take the ordered extents lockdep
8519 * map.
8520 */
8521 freespace_inode = btrfs_is_free_space_inode(inode);
8522
8523 while (1) {
8524 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8525 if (!ordered)
8526 break;
8527 else {
8528 btrfs_err(root->fs_info,
8529 "found ordered extent %llu %llu on inode cleanup",
8530 ordered->file_offset, ordered->num_bytes);
8531
8532 if (!freespace_inode)
8533 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8534
8535 btrfs_remove_ordered_extent(inode, ordered);
8536 btrfs_put_ordered_extent(ordered);
8537 btrfs_put_ordered_extent(ordered);
8538 }
8539 }
8540 btrfs_qgroup_check_reserved_leak(inode);
8541 inode_tree_del(inode);
8542 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8543 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8544 btrfs_put_root(inode->root);
8545 }
8546
btrfs_drop_inode(struct inode * inode)8547 int btrfs_drop_inode(struct inode *inode)
8548 {
8549 struct btrfs_root *root = BTRFS_I(inode)->root;
8550
8551 if (root == NULL)
8552 return 1;
8553
8554 /* the snap/subvol tree is on deleting */
8555 if (btrfs_root_refs(&root->root_item) == 0)
8556 return 1;
8557 else
8558 return generic_drop_inode(inode);
8559 }
8560
init_once(void * foo)8561 static void init_once(void *foo)
8562 {
8563 struct btrfs_inode *ei = foo;
8564
8565 inode_init_once(&ei->vfs_inode);
8566 }
8567
btrfs_destroy_cachep(void)8568 void __cold btrfs_destroy_cachep(void)
8569 {
8570 /*
8571 * Make sure all delayed rcu free inodes are flushed before we
8572 * destroy cache.
8573 */
8574 rcu_barrier();
8575 bioset_exit(&btrfs_dio_bioset);
8576 kmem_cache_destroy(btrfs_inode_cachep);
8577 }
8578
btrfs_init_cachep(void)8579 int __init btrfs_init_cachep(void)
8580 {
8581 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8582 sizeof(struct btrfs_inode), 0,
8583 SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT,
8584 init_once);
8585 if (!btrfs_inode_cachep)
8586 goto fail;
8587
8588 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8589 offsetof(struct btrfs_dio_private, bbio.bio),
8590 BIOSET_NEED_BVECS))
8591 goto fail;
8592
8593 return 0;
8594 fail:
8595 btrfs_destroy_cachep();
8596 return -ENOMEM;
8597 }
8598
btrfs_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int flags)8599 static int btrfs_getattr(struct mnt_idmap *idmap,
8600 const struct path *path, struct kstat *stat,
8601 u32 request_mask, unsigned int flags)
8602 {
8603 u64 delalloc_bytes;
8604 u64 inode_bytes;
8605 struct inode *inode = d_inode(path->dentry);
8606 u32 blocksize = btrfs_sb(inode->i_sb)->sectorsize;
8607 u32 bi_flags = BTRFS_I(inode)->flags;
8608 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8609
8610 stat->result_mask |= STATX_BTIME;
8611 stat->btime.tv_sec = BTRFS_I(inode)->i_otime_sec;
8612 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime_nsec;
8613 if (bi_flags & BTRFS_INODE_APPEND)
8614 stat->attributes |= STATX_ATTR_APPEND;
8615 if (bi_flags & BTRFS_INODE_COMPRESS)
8616 stat->attributes |= STATX_ATTR_COMPRESSED;
8617 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8618 stat->attributes |= STATX_ATTR_IMMUTABLE;
8619 if (bi_flags & BTRFS_INODE_NODUMP)
8620 stat->attributes |= STATX_ATTR_NODUMP;
8621 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8622 stat->attributes |= STATX_ATTR_VERITY;
8623
8624 stat->attributes_mask |= (STATX_ATTR_APPEND |
8625 STATX_ATTR_COMPRESSED |
8626 STATX_ATTR_IMMUTABLE |
8627 STATX_ATTR_NODUMP);
8628
8629 generic_fillattr(idmap, request_mask, inode, stat);
8630 stat->dev = BTRFS_I(inode)->root->anon_dev;
8631
8632 stat->subvol = BTRFS_I(inode)->root->root_key.objectid;
8633 stat->result_mask |= STATX_SUBVOL;
8634
8635 spin_lock(&BTRFS_I(inode)->lock);
8636 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8637 inode_bytes = inode_get_bytes(inode);
8638 spin_unlock(&BTRFS_I(inode)->lock);
8639 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8640 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8641 return 0;
8642 }
8643
btrfs_rename_exchange(struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry)8644 static int btrfs_rename_exchange(struct inode *old_dir,
8645 struct dentry *old_dentry,
8646 struct inode *new_dir,
8647 struct dentry *new_dentry)
8648 {
8649 struct btrfs_fs_info *fs_info = inode_to_fs_info(old_dir);
8650 struct btrfs_trans_handle *trans;
8651 unsigned int trans_num_items;
8652 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8653 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8654 struct inode *new_inode = new_dentry->d_inode;
8655 struct inode *old_inode = old_dentry->d_inode;
8656 struct btrfs_rename_ctx old_rename_ctx;
8657 struct btrfs_rename_ctx new_rename_ctx;
8658 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8659 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8660 u64 old_idx = 0;
8661 u64 new_idx = 0;
8662 int ret;
8663 int ret2;
8664 bool need_abort = false;
8665 struct fscrypt_name old_fname, new_fname;
8666 struct fscrypt_str *old_name, *new_name;
8667
8668 /*
8669 * For non-subvolumes allow exchange only within one subvolume, in the
8670 * same inode namespace. Two subvolumes (represented as directory) can
8671 * be exchanged as they're a logical link and have a fixed inode number.
8672 */
8673 if (root != dest &&
8674 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8675 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8676 return -EXDEV;
8677
8678 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8679 if (ret)
8680 return ret;
8681
8682 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8683 if (ret) {
8684 fscrypt_free_filename(&old_fname);
8685 return ret;
8686 }
8687
8688 old_name = &old_fname.disk_name;
8689 new_name = &new_fname.disk_name;
8690
8691 /* close the race window with snapshot create/destroy ioctl */
8692 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8693 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8694 down_read(&fs_info->subvol_sem);
8695
8696 /*
8697 * For each inode:
8698 * 1 to remove old dir item
8699 * 1 to remove old dir index
8700 * 1 to add new dir item
8701 * 1 to add new dir index
8702 * 1 to update parent inode
8703 *
8704 * If the parents are the same, we only need to account for one
8705 */
8706 trans_num_items = (old_dir == new_dir ? 9 : 10);
8707 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8708 /*
8709 * 1 to remove old root ref
8710 * 1 to remove old root backref
8711 * 1 to add new root ref
8712 * 1 to add new root backref
8713 */
8714 trans_num_items += 4;
8715 } else {
8716 /*
8717 * 1 to update inode item
8718 * 1 to remove old inode ref
8719 * 1 to add new inode ref
8720 */
8721 trans_num_items += 3;
8722 }
8723 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8724 trans_num_items += 4;
8725 else
8726 trans_num_items += 3;
8727 trans = btrfs_start_transaction(root, trans_num_items);
8728 if (IS_ERR(trans)) {
8729 ret = PTR_ERR(trans);
8730 goto out_notrans;
8731 }
8732
8733 if (dest != root) {
8734 ret = btrfs_record_root_in_trans(trans, dest);
8735 if (ret)
8736 goto out_fail;
8737 }
8738
8739 /*
8740 * We need to find a free sequence number both in the source and
8741 * in the destination directory for the exchange.
8742 */
8743 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8744 if (ret)
8745 goto out_fail;
8746 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8747 if (ret)
8748 goto out_fail;
8749
8750 BTRFS_I(old_inode)->dir_index = 0ULL;
8751 BTRFS_I(new_inode)->dir_index = 0ULL;
8752
8753 /* Reference for the source. */
8754 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8755 /* force full log commit if subvolume involved. */
8756 btrfs_set_log_full_commit(trans);
8757 } else {
8758 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8759 btrfs_ino(BTRFS_I(new_dir)),
8760 old_idx);
8761 if (ret)
8762 goto out_fail;
8763 need_abort = true;
8764 }
8765
8766 /* And now for the dest. */
8767 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8768 /* force full log commit if subvolume involved. */
8769 btrfs_set_log_full_commit(trans);
8770 } else {
8771 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8772 btrfs_ino(BTRFS_I(old_dir)),
8773 new_idx);
8774 if (ret) {
8775 if (need_abort)
8776 btrfs_abort_transaction(trans, ret);
8777 goto out_fail;
8778 }
8779 }
8780
8781 /* Update inode version and ctime/mtime. */
8782 inode_inc_iversion(old_dir);
8783 inode_inc_iversion(new_dir);
8784 inode_inc_iversion(old_inode);
8785 inode_inc_iversion(new_inode);
8786 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
8787
8788 if (old_dentry->d_parent != new_dentry->d_parent) {
8789 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8790 BTRFS_I(old_inode), true);
8791 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8792 BTRFS_I(new_inode), true);
8793 }
8794
8795 /* src is a subvolume */
8796 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8797 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8798 } else { /* src is an inode */
8799 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8800 BTRFS_I(old_dentry->d_inode),
8801 old_name, &old_rename_ctx);
8802 if (!ret)
8803 ret = btrfs_update_inode(trans, BTRFS_I(old_inode));
8804 }
8805 if (ret) {
8806 btrfs_abort_transaction(trans, ret);
8807 goto out_fail;
8808 }
8809
8810 /* dest is a subvolume */
8811 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8812 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8813 } else { /* dest is an inode */
8814 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8815 BTRFS_I(new_dentry->d_inode),
8816 new_name, &new_rename_ctx);
8817 if (!ret)
8818 ret = btrfs_update_inode(trans, BTRFS_I(new_inode));
8819 }
8820 if (ret) {
8821 btrfs_abort_transaction(trans, ret);
8822 goto out_fail;
8823 }
8824
8825 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8826 new_name, 0, old_idx);
8827 if (ret) {
8828 btrfs_abort_transaction(trans, ret);
8829 goto out_fail;
8830 }
8831
8832 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8833 old_name, 0, new_idx);
8834 if (ret) {
8835 btrfs_abort_transaction(trans, ret);
8836 goto out_fail;
8837 }
8838
8839 if (old_inode->i_nlink == 1)
8840 BTRFS_I(old_inode)->dir_index = old_idx;
8841 if (new_inode->i_nlink == 1)
8842 BTRFS_I(new_inode)->dir_index = new_idx;
8843
8844 /*
8845 * Now pin the logs of the roots. We do it to ensure that no other task
8846 * can sync the logs while we are in progress with the rename, because
8847 * that could result in an inconsistency in case any of the inodes that
8848 * are part of this rename operation were logged before.
8849 */
8850 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8851 btrfs_pin_log_trans(root);
8852 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8853 btrfs_pin_log_trans(dest);
8854
8855 /* Do the log updates for all inodes. */
8856 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8857 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
8858 old_rename_ctx.index, new_dentry->d_parent);
8859 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8860 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
8861 new_rename_ctx.index, old_dentry->d_parent);
8862
8863 /* Now unpin the logs. */
8864 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8865 btrfs_end_log_trans(root);
8866 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8867 btrfs_end_log_trans(dest);
8868 out_fail:
8869 ret2 = btrfs_end_transaction(trans);
8870 ret = ret ? ret : ret2;
8871 out_notrans:
8872 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
8873 old_ino == BTRFS_FIRST_FREE_OBJECTID)
8874 up_read(&fs_info->subvol_sem);
8875
8876 fscrypt_free_filename(&new_fname);
8877 fscrypt_free_filename(&old_fname);
8878 return ret;
8879 }
8880
new_whiteout_inode(struct mnt_idmap * idmap,struct inode * dir)8881 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
8882 struct inode *dir)
8883 {
8884 struct inode *inode;
8885
8886 inode = new_inode(dir->i_sb);
8887 if (inode) {
8888 inode_init_owner(idmap, inode, dir,
8889 S_IFCHR | WHITEOUT_MODE);
8890 inode->i_op = &btrfs_special_inode_operations;
8891 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
8892 }
8893 return inode;
8894 }
8895
btrfs_rename(struct mnt_idmap * idmap,struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry,unsigned int flags)8896 static int btrfs_rename(struct mnt_idmap *idmap,
8897 struct inode *old_dir, struct dentry *old_dentry,
8898 struct inode *new_dir, struct dentry *new_dentry,
8899 unsigned int flags)
8900 {
8901 struct btrfs_fs_info *fs_info = inode_to_fs_info(old_dir);
8902 struct btrfs_new_inode_args whiteout_args = {
8903 .dir = old_dir,
8904 .dentry = old_dentry,
8905 };
8906 struct btrfs_trans_handle *trans;
8907 unsigned int trans_num_items;
8908 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8909 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8910 struct inode *new_inode = d_inode(new_dentry);
8911 struct inode *old_inode = d_inode(old_dentry);
8912 struct btrfs_rename_ctx rename_ctx;
8913 u64 index = 0;
8914 int ret;
8915 int ret2;
8916 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8917 struct fscrypt_name old_fname, new_fname;
8918
8919 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8920 return -EPERM;
8921
8922 /* we only allow rename subvolume link between subvolumes */
8923 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8924 return -EXDEV;
8925
8926 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8927 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
8928 return -ENOTEMPTY;
8929
8930 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8931 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8932 return -ENOTEMPTY;
8933
8934 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8935 if (ret)
8936 return ret;
8937
8938 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8939 if (ret) {
8940 fscrypt_free_filename(&old_fname);
8941 return ret;
8942 }
8943
8944 /* check for collisions, even if the name isn't there */
8945 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
8946 if (ret) {
8947 if (ret == -EEXIST) {
8948 /* we shouldn't get
8949 * eexist without a new_inode */
8950 if (WARN_ON(!new_inode)) {
8951 goto out_fscrypt_names;
8952 }
8953 } else {
8954 /* maybe -EOVERFLOW */
8955 goto out_fscrypt_names;
8956 }
8957 }
8958 ret = 0;
8959
8960 /*
8961 * we're using rename to replace one file with another. Start IO on it
8962 * now so we don't add too much work to the end of the transaction
8963 */
8964 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
8965 filemap_flush(old_inode->i_mapping);
8966
8967 if (flags & RENAME_WHITEOUT) {
8968 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
8969 if (!whiteout_args.inode) {
8970 ret = -ENOMEM;
8971 goto out_fscrypt_names;
8972 }
8973 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
8974 if (ret)
8975 goto out_whiteout_inode;
8976 } else {
8977 /* 1 to update the old parent inode. */
8978 trans_num_items = 1;
8979 }
8980
8981 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8982 /* Close the race window with snapshot create/destroy ioctl */
8983 down_read(&fs_info->subvol_sem);
8984 /*
8985 * 1 to remove old root ref
8986 * 1 to remove old root backref
8987 * 1 to add new root ref
8988 * 1 to add new root backref
8989 */
8990 trans_num_items += 4;
8991 } else {
8992 /*
8993 * 1 to update inode
8994 * 1 to remove old inode ref
8995 * 1 to add new inode ref
8996 */
8997 trans_num_items += 3;
8998 }
8999 /*
9000 * 1 to remove old dir item
9001 * 1 to remove old dir index
9002 * 1 to add new dir item
9003 * 1 to add new dir index
9004 */
9005 trans_num_items += 4;
9006 /* 1 to update new parent inode if it's not the same as the old parent */
9007 if (new_dir != old_dir)
9008 trans_num_items++;
9009 if (new_inode) {
9010 /*
9011 * 1 to update inode
9012 * 1 to remove inode ref
9013 * 1 to remove dir item
9014 * 1 to remove dir index
9015 * 1 to possibly add orphan item
9016 */
9017 trans_num_items += 5;
9018 }
9019 trans = btrfs_start_transaction(root, trans_num_items);
9020 if (IS_ERR(trans)) {
9021 ret = PTR_ERR(trans);
9022 goto out_notrans;
9023 }
9024
9025 if (dest != root) {
9026 ret = btrfs_record_root_in_trans(trans, dest);
9027 if (ret)
9028 goto out_fail;
9029 }
9030
9031 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9032 if (ret)
9033 goto out_fail;
9034
9035 BTRFS_I(old_inode)->dir_index = 0ULL;
9036 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9037 /* force full log commit if subvolume involved. */
9038 btrfs_set_log_full_commit(trans);
9039 } else {
9040 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9041 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9042 index);
9043 if (ret)
9044 goto out_fail;
9045 }
9046
9047 inode_inc_iversion(old_dir);
9048 inode_inc_iversion(new_dir);
9049 inode_inc_iversion(old_inode);
9050 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
9051
9052 if (old_dentry->d_parent != new_dentry->d_parent)
9053 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9054 BTRFS_I(old_inode), true);
9055
9056 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9057 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9058 } else {
9059 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9060 BTRFS_I(d_inode(old_dentry)),
9061 &old_fname.disk_name, &rename_ctx);
9062 if (!ret)
9063 ret = btrfs_update_inode(trans, BTRFS_I(old_inode));
9064 }
9065 if (ret) {
9066 btrfs_abort_transaction(trans, ret);
9067 goto out_fail;
9068 }
9069
9070 if (new_inode) {
9071 inode_inc_iversion(new_inode);
9072 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9073 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9074 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9075 BUG_ON(new_inode->i_nlink == 0);
9076 } else {
9077 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9078 BTRFS_I(d_inode(new_dentry)),
9079 &new_fname.disk_name);
9080 }
9081 if (!ret && new_inode->i_nlink == 0)
9082 ret = btrfs_orphan_add(trans,
9083 BTRFS_I(d_inode(new_dentry)));
9084 if (ret) {
9085 btrfs_abort_transaction(trans, ret);
9086 goto out_fail;
9087 }
9088 }
9089
9090 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9091 &new_fname.disk_name, 0, index);
9092 if (ret) {
9093 btrfs_abort_transaction(trans, ret);
9094 goto out_fail;
9095 }
9096
9097 if (old_inode->i_nlink == 1)
9098 BTRFS_I(old_inode)->dir_index = index;
9099
9100 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9101 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9102 rename_ctx.index, new_dentry->d_parent);
9103
9104 if (flags & RENAME_WHITEOUT) {
9105 ret = btrfs_create_new_inode(trans, &whiteout_args);
9106 if (ret) {
9107 btrfs_abort_transaction(trans, ret);
9108 goto out_fail;
9109 } else {
9110 unlock_new_inode(whiteout_args.inode);
9111 iput(whiteout_args.inode);
9112 whiteout_args.inode = NULL;
9113 }
9114 }
9115 out_fail:
9116 ret2 = btrfs_end_transaction(trans);
9117 ret = ret ? ret : ret2;
9118 out_notrans:
9119 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9120 up_read(&fs_info->subvol_sem);
9121 if (flags & RENAME_WHITEOUT)
9122 btrfs_new_inode_args_destroy(&whiteout_args);
9123 out_whiteout_inode:
9124 if (flags & RENAME_WHITEOUT)
9125 iput(whiteout_args.inode);
9126 out_fscrypt_names:
9127 fscrypt_free_filename(&old_fname);
9128 fscrypt_free_filename(&new_fname);
9129 return ret;
9130 }
9131
btrfs_rename2(struct mnt_idmap * idmap,struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry,unsigned int flags)9132 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9133 struct dentry *old_dentry, struct inode *new_dir,
9134 struct dentry *new_dentry, unsigned int flags)
9135 {
9136 int ret;
9137
9138 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9139 return -EINVAL;
9140
9141 if (flags & RENAME_EXCHANGE)
9142 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9143 new_dentry);
9144 else
9145 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9146 new_dentry, flags);
9147
9148 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9149
9150 return ret;
9151 }
9152
9153 struct btrfs_delalloc_work {
9154 struct inode *inode;
9155 struct completion completion;
9156 struct list_head list;
9157 struct btrfs_work work;
9158 };
9159
btrfs_run_delalloc_work(struct btrfs_work * work)9160 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9161 {
9162 struct btrfs_delalloc_work *delalloc_work;
9163 struct inode *inode;
9164
9165 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9166 work);
9167 inode = delalloc_work->inode;
9168 filemap_flush(inode->i_mapping);
9169 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9170 &BTRFS_I(inode)->runtime_flags))
9171 filemap_flush(inode->i_mapping);
9172
9173 iput(inode);
9174 complete(&delalloc_work->completion);
9175 }
9176
btrfs_alloc_delalloc_work(struct inode * inode)9177 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9178 {
9179 struct btrfs_delalloc_work *work;
9180
9181 work = kmalloc(sizeof(*work), GFP_NOFS);
9182 if (!work)
9183 return NULL;
9184
9185 init_completion(&work->completion);
9186 INIT_LIST_HEAD(&work->list);
9187 work->inode = inode;
9188 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL);
9189
9190 return work;
9191 }
9192
9193 /*
9194 * some fairly slow code that needs optimization. This walks the list
9195 * of all the inodes with pending delalloc and forces them to disk.
9196 */
start_delalloc_inodes(struct btrfs_root * root,struct writeback_control * wbc,bool snapshot,bool in_reclaim_context)9197 static int start_delalloc_inodes(struct btrfs_root *root,
9198 struct writeback_control *wbc, bool snapshot,
9199 bool in_reclaim_context)
9200 {
9201 struct btrfs_inode *binode;
9202 struct inode *inode;
9203 struct btrfs_delalloc_work *work, *next;
9204 LIST_HEAD(works);
9205 LIST_HEAD(splice);
9206 int ret = 0;
9207 bool full_flush = wbc->nr_to_write == LONG_MAX;
9208
9209 mutex_lock(&root->delalloc_mutex);
9210 spin_lock(&root->delalloc_lock);
9211 list_splice_init(&root->delalloc_inodes, &splice);
9212 while (!list_empty(&splice)) {
9213 binode = list_entry(splice.next, struct btrfs_inode,
9214 delalloc_inodes);
9215
9216 list_move_tail(&binode->delalloc_inodes,
9217 &root->delalloc_inodes);
9218
9219 if (in_reclaim_context &&
9220 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9221 continue;
9222
9223 inode = igrab(&binode->vfs_inode);
9224 if (!inode) {
9225 cond_resched_lock(&root->delalloc_lock);
9226 continue;
9227 }
9228 spin_unlock(&root->delalloc_lock);
9229
9230 if (snapshot)
9231 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9232 &binode->runtime_flags);
9233 if (full_flush) {
9234 work = btrfs_alloc_delalloc_work(inode);
9235 if (!work) {
9236 iput(inode);
9237 ret = -ENOMEM;
9238 goto out;
9239 }
9240 list_add_tail(&work->list, &works);
9241 btrfs_queue_work(root->fs_info->flush_workers,
9242 &work->work);
9243 } else {
9244 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9245 btrfs_add_delayed_iput(BTRFS_I(inode));
9246 if (ret || wbc->nr_to_write <= 0)
9247 goto out;
9248 }
9249 cond_resched();
9250 spin_lock(&root->delalloc_lock);
9251 }
9252 spin_unlock(&root->delalloc_lock);
9253
9254 out:
9255 list_for_each_entry_safe(work, next, &works, list) {
9256 list_del_init(&work->list);
9257 wait_for_completion(&work->completion);
9258 kfree(work);
9259 }
9260
9261 if (!list_empty(&splice)) {
9262 spin_lock(&root->delalloc_lock);
9263 list_splice_tail(&splice, &root->delalloc_inodes);
9264 spin_unlock(&root->delalloc_lock);
9265 }
9266 mutex_unlock(&root->delalloc_mutex);
9267 return ret;
9268 }
9269
btrfs_start_delalloc_snapshot(struct btrfs_root * root,bool in_reclaim_context)9270 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9271 {
9272 struct writeback_control wbc = {
9273 .nr_to_write = LONG_MAX,
9274 .sync_mode = WB_SYNC_NONE,
9275 .range_start = 0,
9276 .range_end = LLONG_MAX,
9277 };
9278 struct btrfs_fs_info *fs_info = root->fs_info;
9279
9280 if (BTRFS_FS_ERROR(fs_info))
9281 return -EROFS;
9282
9283 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9284 }
9285
btrfs_start_delalloc_roots(struct btrfs_fs_info * fs_info,long nr,bool in_reclaim_context)9286 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9287 bool in_reclaim_context)
9288 {
9289 struct writeback_control wbc = {
9290 .nr_to_write = nr,
9291 .sync_mode = WB_SYNC_NONE,
9292 .range_start = 0,
9293 .range_end = LLONG_MAX,
9294 };
9295 struct btrfs_root *root;
9296 LIST_HEAD(splice);
9297 int ret;
9298
9299 if (BTRFS_FS_ERROR(fs_info))
9300 return -EROFS;
9301
9302 mutex_lock(&fs_info->delalloc_root_mutex);
9303 spin_lock(&fs_info->delalloc_root_lock);
9304 list_splice_init(&fs_info->delalloc_roots, &splice);
9305 while (!list_empty(&splice)) {
9306 /*
9307 * Reset nr_to_write here so we know that we're doing a full
9308 * flush.
9309 */
9310 if (nr == LONG_MAX)
9311 wbc.nr_to_write = LONG_MAX;
9312
9313 root = list_first_entry(&splice, struct btrfs_root,
9314 delalloc_root);
9315 root = btrfs_grab_root(root);
9316 BUG_ON(!root);
9317 list_move_tail(&root->delalloc_root,
9318 &fs_info->delalloc_roots);
9319 spin_unlock(&fs_info->delalloc_root_lock);
9320
9321 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9322 btrfs_put_root(root);
9323 if (ret < 0 || wbc.nr_to_write <= 0)
9324 goto out;
9325 spin_lock(&fs_info->delalloc_root_lock);
9326 }
9327 spin_unlock(&fs_info->delalloc_root_lock);
9328
9329 ret = 0;
9330 out:
9331 if (!list_empty(&splice)) {
9332 spin_lock(&fs_info->delalloc_root_lock);
9333 list_splice_tail(&splice, &fs_info->delalloc_roots);
9334 spin_unlock(&fs_info->delalloc_root_lock);
9335 }
9336 mutex_unlock(&fs_info->delalloc_root_mutex);
9337 return ret;
9338 }
9339
btrfs_symlink(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,const char * symname)9340 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9341 struct dentry *dentry, const char *symname)
9342 {
9343 struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
9344 struct btrfs_trans_handle *trans;
9345 struct btrfs_root *root = BTRFS_I(dir)->root;
9346 struct btrfs_path *path;
9347 struct btrfs_key key;
9348 struct inode *inode;
9349 struct btrfs_new_inode_args new_inode_args = {
9350 .dir = dir,
9351 .dentry = dentry,
9352 };
9353 unsigned int trans_num_items;
9354 int err;
9355 int name_len;
9356 int datasize;
9357 unsigned long ptr;
9358 struct btrfs_file_extent_item *ei;
9359 struct extent_buffer *leaf;
9360
9361 name_len = strlen(symname);
9362 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9363 return -ENAMETOOLONG;
9364
9365 inode = new_inode(dir->i_sb);
9366 if (!inode)
9367 return -ENOMEM;
9368 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9369 inode->i_op = &btrfs_symlink_inode_operations;
9370 inode_nohighmem(inode);
9371 inode->i_mapping->a_ops = &btrfs_aops;
9372 btrfs_i_size_write(BTRFS_I(inode), name_len);
9373 inode_set_bytes(inode, name_len);
9374
9375 new_inode_args.inode = inode;
9376 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9377 if (err)
9378 goto out_inode;
9379 /* 1 additional item for the inline extent */
9380 trans_num_items++;
9381
9382 trans = btrfs_start_transaction(root, trans_num_items);
9383 if (IS_ERR(trans)) {
9384 err = PTR_ERR(trans);
9385 goto out_new_inode_args;
9386 }
9387
9388 err = btrfs_create_new_inode(trans, &new_inode_args);
9389 if (err)
9390 goto out;
9391
9392 path = btrfs_alloc_path();
9393 if (!path) {
9394 err = -ENOMEM;
9395 btrfs_abort_transaction(trans, err);
9396 discard_new_inode(inode);
9397 inode = NULL;
9398 goto out;
9399 }
9400 key.objectid = btrfs_ino(BTRFS_I(inode));
9401 key.offset = 0;
9402 key.type = BTRFS_EXTENT_DATA_KEY;
9403 datasize = btrfs_file_extent_calc_inline_size(name_len);
9404 err = btrfs_insert_empty_item(trans, root, path, &key,
9405 datasize);
9406 if (err) {
9407 btrfs_abort_transaction(trans, err);
9408 btrfs_free_path(path);
9409 discard_new_inode(inode);
9410 inode = NULL;
9411 goto out;
9412 }
9413 leaf = path->nodes[0];
9414 ei = btrfs_item_ptr(leaf, path->slots[0],
9415 struct btrfs_file_extent_item);
9416 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9417 btrfs_set_file_extent_type(leaf, ei,
9418 BTRFS_FILE_EXTENT_INLINE);
9419 btrfs_set_file_extent_encryption(leaf, ei, 0);
9420 btrfs_set_file_extent_compression(leaf, ei, 0);
9421 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9422 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9423
9424 ptr = btrfs_file_extent_inline_start(ei);
9425 write_extent_buffer(leaf, symname, ptr, name_len);
9426 btrfs_mark_buffer_dirty(trans, leaf);
9427 btrfs_free_path(path);
9428
9429 d_instantiate_new(dentry, inode);
9430 err = 0;
9431 out:
9432 btrfs_end_transaction(trans);
9433 btrfs_btree_balance_dirty(fs_info);
9434 out_new_inode_args:
9435 btrfs_new_inode_args_destroy(&new_inode_args);
9436 out_inode:
9437 if (err)
9438 iput(inode);
9439 return err;
9440 }
9441
insert_prealloc_file_extent(struct btrfs_trans_handle * trans_in,struct btrfs_inode * inode,struct btrfs_key * ins,u64 file_offset)9442 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9443 struct btrfs_trans_handle *trans_in,
9444 struct btrfs_inode *inode,
9445 struct btrfs_key *ins,
9446 u64 file_offset)
9447 {
9448 struct btrfs_file_extent_item stack_fi;
9449 struct btrfs_replace_extent_info extent_info;
9450 struct btrfs_trans_handle *trans = trans_in;
9451 struct btrfs_path *path;
9452 u64 start = ins->objectid;
9453 u64 len = ins->offset;
9454 u64 qgroup_released = 0;
9455 int ret;
9456
9457 memset(&stack_fi, 0, sizeof(stack_fi));
9458
9459 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9460 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9461 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9462 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9463 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9464 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9465 /* Encryption and other encoding is reserved and all 0 */
9466
9467 ret = btrfs_qgroup_release_data(inode, file_offset, len, &qgroup_released);
9468 if (ret < 0)
9469 return ERR_PTR(ret);
9470
9471 if (trans) {
9472 ret = insert_reserved_file_extent(trans, inode,
9473 file_offset, &stack_fi,
9474 true, qgroup_released);
9475 if (ret)
9476 goto free_qgroup;
9477 return trans;
9478 }
9479
9480 extent_info.disk_offset = start;
9481 extent_info.disk_len = len;
9482 extent_info.data_offset = 0;
9483 extent_info.data_len = len;
9484 extent_info.file_offset = file_offset;
9485 extent_info.extent_buf = (char *)&stack_fi;
9486 extent_info.is_new_extent = true;
9487 extent_info.update_times = true;
9488 extent_info.qgroup_reserved = qgroup_released;
9489 extent_info.insertions = 0;
9490
9491 path = btrfs_alloc_path();
9492 if (!path) {
9493 ret = -ENOMEM;
9494 goto free_qgroup;
9495 }
9496
9497 ret = btrfs_replace_file_extents(inode, path, file_offset,
9498 file_offset + len - 1, &extent_info,
9499 &trans);
9500 btrfs_free_path(path);
9501 if (ret)
9502 goto free_qgroup;
9503 return trans;
9504
9505 free_qgroup:
9506 /*
9507 * We have released qgroup data range at the beginning of the function,
9508 * and normally qgroup_released bytes will be freed when committing
9509 * transaction.
9510 * But if we error out early, we have to free what we have released
9511 * or we leak qgroup data reservation.
9512 */
9513 btrfs_qgroup_free_refroot(inode->root->fs_info,
9514 btrfs_root_id(inode->root), qgroup_released,
9515 BTRFS_QGROUP_RSV_DATA);
9516 return ERR_PTR(ret);
9517 }
9518
__btrfs_prealloc_file_range(struct inode * inode,int mode,u64 start,u64 num_bytes,u64 min_size,loff_t actual_len,u64 * alloc_hint,struct btrfs_trans_handle * trans)9519 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9520 u64 start, u64 num_bytes, u64 min_size,
9521 loff_t actual_len, u64 *alloc_hint,
9522 struct btrfs_trans_handle *trans)
9523 {
9524 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
9525 struct extent_map *em;
9526 struct btrfs_root *root = BTRFS_I(inode)->root;
9527 struct btrfs_key ins;
9528 u64 cur_offset = start;
9529 u64 clear_offset = start;
9530 u64 i_size;
9531 u64 cur_bytes;
9532 u64 last_alloc = (u64)-1;
9533 int ret = 0;
9534 bool own_trans = true;
9535 u64 end = start + num_bytes - 1;
9536
9537 if (trans)
9538 own_trans = false;
9539 while (num_bytes > 0) {
9540 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9541 cur_bytes = max(cur_bytes, min_size);
9542 /*
9543 * If we are severely fragmented we could end up with really
9544 * small allocations, so if the allocator is returning small
9545 * chunks lets make its job easier by only searching for those
9546 * sized chunks.
9547 */
9548 cur_bytes = min(cur_bytes, last_alloc);
9549 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9550 min_size, 0, *alloc_hint, &ins, 1, 0);
9551 if (ret)
9552 break;
9553
9554 /*
9555 * We've reserved this space, and thus converted it from
9556 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9557 * from here on out we will only need to clear our reservation
9558 * for the remaining unreserved area, so advance our
9559 * clear_offset by our extent size.
9560 */
9561 clear_offset += ins.offset;
9562
9563 last_alloc = ins.offset;
9564 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9565 &ins, cur_offset);
9566 /*
9567 * Now that we inserted the prealloc extent we can finally
9568 * decrement the number of reservations in the block group.
9569 * If we did it before, we could race with relocation and have
9570 * relocation miss the reserved extent, making it fail later.
9571 */
9572 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9573 if (IS_ERR(trans)) {
9574 ret = PTR_ERR(trans);
9575 btrfs_free_reserved_extent(fs_info, ins.objectid,
9576 ins.offset, 0);
9577 break;
9578 }
9579
9580 em = alloc_extent_map();
9581 if (!em) {
9582 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9583 cur_offset + ins.offset - 1, false);
9584 btrfs_set_inode_full_sync(BTRFS_I(inode));
9585 goto next;
9586 }
9587
9588 em->start = cur_offset;
9589 em->orig_start = cur_offset;
9590 em->len = ins.offset;
9591 em->block_start = ins.objectid;
9592 em->block_len = ins.offset;
9593 em->orig_block_len = ins.offset;
9594 em->ram_bytes = ins.offset;
9595 em->flags |= EXTENT_FLAG_PREALLOC;
9596 em->generation = trans->transid;
9597
9598 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9599 free_extent_map(em);
9600 next:
9601 num_bytes -= ins.offset;
9602 cur_offset += ins.offset;
9603 *alloc_hint = ins.objectid + ins.offset;
9604
9605 inode_inc_iversion(inode);
9606 inode_set_ctime_current(inode);
9607 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9608 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9609 (actual_len > inode->i_size) &&
9610 (cur_offset > inode->i_size)) {
9611 if (cur_offset > actual_len)
9612 i_size = actual_len;
9613 else
9614 i_size = cur_offset;
9615 i_size_write(inode, i_size);
9616 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9617 }
9618
9619 ret = btrfs_update_inode(trans, BTRFS_I(inode));
9620
9621 if (ret) {
9622 btrfs_abort_transaction(trans, ret);
9623 if (own_trans)
9624 btrfs_end_transaction(trans);
9625 break;
9626 }
9627
9628 if (own_trans) {
9629 btrfs_end_transaction(trans);
9630 trans = NULL;
9631 }
9632 }
9633 if (clear_offset < end)
9634 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9635 end - clear_offset + 1);
9636 return ret;
9637 }
9638
btrfs_prealloc_file_range(struct inode * inode,int mode,u64 start,u64 num_bytes,u64 min_size,loff_t actual_len,u64 * alloc_hint)9639 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9640 u64 start, u64 num_bytes, u64 min_size,
9641 loff_t actual_len, u64 *alloc_hint)
9642 {
9643 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9644 min_size, actual_len, alloc_hint,
9645 NULL);
9646 }
9647
btrfs_prealloc_file_range_trans(struct inode * inode,struct btrfs_trans_handle * trans,int mode,u64 start,u64 num_bytes,u64 min_size,loff_t actual_len,u64 * alloc_hint)9648 int btrfs_prealloc_file_range_trans(struct inode *inode,
9649 struct btrfs_trans_handle *trans, int mode,
9650 u64 start, u64 num_bytes, u64 min_size,
9651 loff_t actual_len, u64 *alloc_hint)
9652 {
9653 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9654 min_size, actual_len, alloc_hint, trans);
9655 }
9656
btrfs_permission(struct mnt_idmap * idmap,struct inode * inode,int mask)9657 static int btrfs_permission(struct mnt_idmap *idmap,
9658 struct inode *inode, int mask)
9659 {
9660 struct btrfs_root *root = BTRFS_I(inode)->root;
9661 umode_t mode = inode->i_mode;
9662
9663 if (mask & MAY_WRITE &&
9664 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9665 if (btrfs_root_readonly(root))
9666 return -EROFS;
9667 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9668 return -EACCES;
9669 }
9670 return generic_permission(idmap, inode, mask);
9671 }
9672
btrfs_tmpfile(struct mnt_idmap * idmap,struct inode * dir,struct file * file,umode_t mode)9673 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9674 struct file *file, umode_t mode)
9675 {
9676 struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
9677 struct btrfs_trans_handle *trans;
9678 struct btrfs_root *root = BTRFS_I(dir)->root;
9679 struct inode *inode;
9680 struct btrfs_new_inode_args new_inode_args = {
9681 .dir = dir,
9682 .dentry = file->f_path.dentry,
9683 .orphan = true,
9684 };
9685 unsigned int trans_num_items;
9686 int ret;
9687
9688 inode = new_inode(dir->i_sb);
9689 if (!inode)
9690 return -ENOMEM;
9691 inode_init_owner(idmap, inode, dir, mode);
9692 inode->i_fop = &btrfs_file_operations;
9693 inode->i_op = &btrfs_file_inode_operations;
9694 inode->i_mapping->a_ops = &btrfs_aops;
9695
9696 new_inode_args.inode = inode;
9697 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9698 if (ret)
9699 goto out_inode;
9700
9701 trans = btrfs_start_transaction(root, trans_num_items);
9702 if (IS_ERR(trans)) {
9703 ret = PTR_ERR(trans);
9704 goto out_new_inode_args;
9705 }
9706
9707 ret = btrfs_create_new_inode(trans, &new_inode_args);
9708
9709 /*
9710 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9711 * set it to 1 because d_tmpfile() will issue a warning if the count is
9712 * 0, through:
9713 *
9714 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9715 */
9716 set_nlink(inode, 1);
9717
9718 if (!ret) {
9719 d_tmpfile(file, inode);
9720 unlock_new_inode(inode);
9721 mark_inode_dirty(inode);
9722 }
9723
9724 btrfs_end_transaction(trans);
9725 btrfs_btree_balance_dirty(fs_info);
9726 out_new_inode_args:
9727 btrfs_new_inode_args_destroy(&new_inode_args);
9728 out_inode:
9729 if (ret)
9730 iput(inode);
9731 return finish_open_simple(file, ret);
9732 }
9733
btrfs_set_range_writeback(struct btrfs_inode * inode,u64 start,u64 end)9734 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9735 {
9736 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9737 unsigned long index = start >> PAGE_SHIFT;
9738 unsigned long end_index = end >> PAGE_SHIFT;
9739 struct page *page;
9740 u32 len;
9741
9742 ASSERT(end + 1 - start <= U32_MAX);
9743 len = end + 1 - start;
9744 while (index <= end_index) {
9745 page = find_get_page(inode->vfs_inode.i_mapping, index);
9746 ASSERT(page); /* Pages should be in the extent_io_tree */
9747
9748 /* This is for data, which doesn't yet support larger folio. */
9749 ASSERT(folio_order(page_folio(page)) == 0);
9750 btrfs_folio_set_writeback(fs_info, page_folio(page), start, len);
9751 put_page(page);
9752 index++;
9753 }
9754 }
9755
btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info * fs_info,int compress_type)9756 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9757 int compress_type)
9758 {
9759 switch (compress_type) {
9760 case BTRFS_COMPRESS_NONE:
9761 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9762 case BTRFS_COMPRESS_ZLIB:
9763 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9764 case BTRFS_COMPRESS_LZO:
9765 /*
9766 * The LZO format depends on the sector size. 64K is the maximum
9767 * sector size that we support.
9768 */
9769 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9770 return -EINVAL;
9771 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9772 (fs_info->sectorsize_bits - 12);
9773 case BTRFS_COMPRESS_ZSTD:
9774 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9775 default:
9776 return -EUCLEAN;
9777 }
9778 }
9779
btrfs_encoded_read_inline(struct kiocb * iocb,struct iov_iter * iter,u64 start,u64 lockend,struct extent_state ** cached_state,u64 extent_start,size_t count,struct btrfs_ioctl_encoded_io_args * encoded,bool * unlocked)9780 static ssize_t btrfs_encoded_read_inline(
9781 struct kiocb *iocb,
9782 struct iov_iter *iter, u64 start,
9783 u64 lockend,
9784 struct extent_state **cached_state,
9785 u64 extent_start, size_t count,
9786 struct btrfs_ioctl_encoded_io_args *encoded,
9787 bool *unlocked)
9788 {
9789 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9790 struct btrfs_root *root = inode->root;
9791 struct btrfs_fs_info *fs_info = root->fs_info;
9792 struct extent_io_tree *io_tree = &inode->io_tree;
9793 struct btrfs_path *path;
9794 struct extent_buffer *leaf;
9795 struct btrfs_file_extent_item *item;
9796 u64 ram_bytes;
9797 unsigned long ptr;
9798 void *tmp;
9799 ssize_t ret;
9800
9801 path = btrfs_alloc_path();
9802 if (!path) {
9803 ret = -ENOMEM;
9804 goto out;
9805 }
9806 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9807 extent_start, 0);
9808 if (ret) {
9809 if (ret > 0) {
9810 /* The extent item disappeared? */
9811 ret = -EIO;
9812 }
9813 goto out;
9814 }
9815 leaf = path->nodes[0];
9816 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9817
9818 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9819 ptr = btrfs_file_extent_inline_start(item);
9820
9821 encoded->len = min_t(u64, extent_start + ram_bytes,
9822 inode->vfs_inode.i_size) - iocb->ki_pos;
9823 ret = btrfs_encoded_io_compression_from_extent(fs_info,
9824 btrfs_file_extent_compression(leaf, item));
9825 if (ret < 0)
9826 goto out;
9827 encoded->compression = ret;
9828 if (encoded->compression) {
9829 size_t inline_size;
9830
9831 inline_size = btrfs_file_extent_inline_item_len(leaf,
9832 path->slots[0]);
9833 if (inline_size > count) {
9834 ret = -ENOBUFS;
9835 goto out;
9836 }
9837 count = inline_size;
9838 encoded->unencoded_len = ram_bytes;
9839 encoded->unencoded_offset = iocb->ki_pos - extent_start;
9840 } else {
9841 count = min_t(u64, count, encoded->len);
9842 encoded->len = count;
9843 encoded->unencoded_len = count;
9844 ptr += iocb->ki_pos - extent_start;
9845 }
9846
9847 tmp = kmalloc(count, GFP_NOFS);
9848 if (!tmp) {
9849 ret = -ENOMEM;
9850 goto out;
9851 }
9852 read_extent_buffer(leaf, tmp, ptr, count);
9853 btrfs_release_path(path);
9854 unlock_extent(io_tree, start, lockend, cached_state);
9855 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
9856 *unlocked = true;
9857
9858 ret = copy_to_iter(tmp, count, iter);
9859 if (ret != count)
9860 ret = -EFAULT;
9861 kfree(tmp);
9862 out:
9863 btrfs_free_path(path);
9864 return ret;
9865 }
9866
9867 struct btrfs_encoded_read_private {
9868 wait_queue_head_t wait;
9869 atomic_t pending;
9870 blk_status_t status;
9871 };
9872
btrfs_encoded_read_endio(struct btrfs_bio * bbio)9873 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
9874 {
9875 struct btrfs_encoded_read_private *priv = bbio->private;
9876
9877 if (bbio->bio.bi_status) {
9878 /*
9879 * The memory barrier implied by the atomic_dec_return() here
9880 * pairs with the memory barrier implied by the
9881 * atomic_dec_return() or io_wait_event() in
9882 * btrfs_encoded_read_regular_fill_pages() to ensure that this
9883 * write is observed before the load of status in
9884 * btrfs_encoded_read_regular_fill_pages().
9885 */
9886 WRITE_ONCE(priv->status, bbio->bio.bi_status);
9887 }
9888 if (!atomic_dec_return(&priv->pending))
9889 wake_up(&priv->wait);
9890 bio_put(&bbio->bio);
9891 }
9892
btrfs_encoded_read_regular_fill_pages(struct btrfs_inode * inode,u64 file_offset,u64 disk_bytenr,u64 disk_io_size,struct page ** pages)9893 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
9894 u64 file_offset, u64 disk_bytenr,
9895 u64 disk_io_size, struct page **pages)
9896 {
9897 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9898 struct btrfs_encoded_read_private priv = {
9899 .pending = ATOMIC_INIT(1),
9900 };
9901 unsigned long i = 0;
9902 struct btrfs_bio *bbio;
9903
9904 init_waitqueue_head(&priv.wait);
9905
9906 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9907 btrfs_encoded_read_endio, &priv);
9908 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9909 bbio->inode = inode;
9910
9911 do {
9912 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
9913
9914 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
9915 atomic_inc(&priv.pending);
9916 btrfs_submit_bio(bbio, 0);
9917
9918 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9919 btrfs_encoded_read_endio, &priv);
9920 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9921 bbio->inode = inode;
9922 continue;
9923 }
9924
9925 i++;
9926 disk_bytenr += bytes;
9927 disk_io_size -= bytes;
9928 } while (disk_io_size);
9929
9930 atomic_inc(&priv.pending);
9931 btrfs_submit_bio(bbio, 0);
9932
9933 if (atomic_dec_return(&priv.pending))
9934 io_wait_event(priv.wait, !atomic_read(&priv.pending));
9935 /* See btrfs_encoded_read_endio() for ordering. */
9936 return blk_status_to_errno(READ_ONCE(priv.status));
9937 }
9938
btrfs_encoded_read_regular(struct kiocb * iocb,struct iov_iter * iter,u64 start,u64 lockend,struct extent_state ** cached_state,u64 disk_bytenr,u64 disk_io_size,size_t count,bool compressed,bool * unlocked)9939 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
9940 struct iov_iter *iter,
9941 u64 start, u64 lockend,
9942 struct extent_state **cached_state,
9943 u64 disk_bytenr, u64 disk_io_size,
9944 size_t count, bool compressed,
9945 bool *unlocked)
9946 {
9947 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9948 struct extent_io_tree *io_tree = &inode->io_tree;
9949 struct page **pages;
9950 unsigned long nr_pages, i;
9951 u64 cur;
9952 size_t page_offset;
9953 ssize_t ret;
9954
9955 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
9956 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
9957 if (!pages)
9958 return -ENOMEM;
9959 ret = btrfs_alloc_page_array(nr_pages, pages, 0);
9960 if (ret) {
9961 ret = -ENOMEM;
9962 goto out;
9963 }
9964
9965 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
9966 disk_io_size, pages);
9967 if (ret)
9968 goto out;
9969
9970 unlock_extent(io_tree, start, lockend, cached_state);
9971 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
9972 *unlocked = true;
9973
9974 if (compressed) {
9975 i = 0;
9976 page_offset = 0;
9977 } else {
9978 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
9979 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
9980 }
9981 cur = 0;
9982 while (cur < count) {
9983 size_t bytes = min_t(size_t, count - cur,
9984 PAGE_SIZE - page_offset);
9985
9986 if (copy_page_to_iter(pages[i], page_offset, bytes,
9987 iter) != bytes) {
9988 ret = -EFAULT;
9989 goto out;
9990 }
9991 i++;
9992 cur += bytes;
9993 page_offset = 0;
9994 }
9995 ret = count;
9996 out:
9997 for (i = 0; i < nr_pages; i++) {
9998 if (pages[i])
9999 __free_page(pages[i]);
10000 }
10001 kfree(pages);
10002 return ret;
10003 }
10004
btrfs_encoded_read(struct kiocb * iocb,struct iov_iter * iter,struct btrfs_ioctl_encoded_io_args * encoded)10005 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10006 struct btrfs_ioctl_encoded_io_args *encoded)
10007 {
10008 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10009 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10010 struct extent_io_tree *io_tree = &inode->io_tree;
10011 ssize_t ret;
10012 size_t count = iov_iter_count(iter);
10013 u64 start, lockend, disk_bytenr, disk_io_size;
10014 struct extent_state *cached_state = NULL;
10015 struct extent_map *em;
10016 bool unlocked = false;
10017
10018 file_accessed(iocb->ki_filp);
10019
10020 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10021
10022 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10023 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10024 return 0;
10025 }
10026 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10027 /*
10028 * We don't know how long the extent containing iocb->ki_pos is, but if
10029 * it's compressed we know that it won't be longer than this.
10030 */
10031 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10032
10033 for (;;) {
10034 struct btrfs_ordered_extent *ordered;
10035
10036 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10037 lockend - start + 1);
10038 if (ret)
10039 goto out_unlock_inode;
10040 lock_extent(io_tree, start, lockend, &cached_state);
10041 ordered = btrfs_lookup_ordered_range(inode, start,
10042 lockend - start + 1);
10043 if (!ordered)
10044 break;
10045 btrfs_put_ordered_extent(ordered);
10046 unlock_extent(io_tree, start, lockend, &cached_state);
10047 cond_resched();
10048 }
10049
10050 em = btrfs_get_extent(inode, NULL, start, lockend - start + 1);
10051 if (IS_ERR(em)) {
10052 ret = PTR_ERR(em);
10053 goto out_unlock_extent;
10054 }
10055
10056 if (em->block_start == EXTENT_MAP_INLINE) {
10057 u64 extent_start = em->start;
10058
10059 /*
10060 * For inline extents we get everything we need out of the
10061 * extent item.
10062 */
10063 free_extent_map(em);
10064 em = NULL;
10065 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10066 &cached_state, extent_start,
10067 count, encoded, &unlocked);
10068 goto out;
10069 }
10070
10071 /*
10072 * We only want to return up to EOF even if the extent extends beyond
10073 * that.
10074 */
10075 encoded->len = min_t(u64, extent_map_end(em),
10076 inode->vfs_inode.i_size) - iocb->ki_pos;
10077 if (em->block_start == EXTENT_MAP_HOLE ||
10078 (em->flags & EXTENT_FLAG_PREALLOC)) {
10079 disk_bytenr = EXTENT_MAP_HOLE;
10080 count = min_t(u64, count, encoded->len);
10081 encoded->len = count;
10082 encoded->unencoded_len = count;
10083 } else if (extent_map_is_compressed(em)) {
10084 disk_bytenr = em->block_start;
10085 /*
10086 * Bail if the buffer isn't large enough to return the whole
10087 * compressed extent.
10088 */
10089 if (em->block_len > count) {
10090 ret = -ENOBUFS;
10091 goto out_em;
10092 }
10093 disk_io_size = em->block_len;
10094 count = em->block_len;
10095 encoded->unencoded_len = em->ram_bytes;
10096 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10097 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10098 extent_map_compression(em));
10099 if (ret < 0)
10100 goto out_em;
10101 encoded->compression = ret;
10102 } else {
10103 disk_bytenr = em->block_start + (start - em->start);
10104 if (encoded->len > count)
10105 encoded->len = count;
10106 /*
10107 * Don't read beyond what we locked. This also limits the page
10108 * allocations that we'll do.
10109 */
10110 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10111 count = start + disk_io_size - iocb->ki_pos;
10112 encoded->len = count;
10113 encoded->unencoded_len = count;
10114 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10115 }
10116 free_extent_map(em);
10117 em = NULL;
10118
10119 if (disk_bytenr == EXTENT_MAP_HOLE) {
10120 unlock_extent(io_tree, start, lockend, &cached_state);
10121 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10122 unlocked = true;
10123 ret = iov_iter_zero(count, iter);
10124 if (ret != count)
10125 ret = -EFAULT;
10126 } else {
10127 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10128 &cached_state, disk_bytenr,
10129 disk_io_size, count,
10130 encoded->compression,
10131 &unlocked);
10132 }
10133
10134 out:
10135 if (ret >= 0)
10136 iocb->ki_pos += encoded->len;
10137 out_em:
10138 free_extent_map(em);
10139 out_unlock_extent:
10140 if (!unlocked)
10141 unlock_extent(io_tree, start, lockend, &cached_state);
10142 out_unlock_inode:
10143 if (!unlocked)
10144 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10145 return ret;
10146 }
10147
btrfs_do_encoded_write(struct kiocb * iocb,struct iov_iter * from,const struct btrfs_ioctl_encoded_io_args * encoded)10148 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10149 const struct btrfs_ioctl_encoded_io_args *encoded)
10150 {
10151 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10152 struct btrfs_root *root = inode->root;
10153 struct btrfs_fs_info *fs_info = root->fs_info;
10154 struct extent_io_tree *io_tree = &inode->io_tree;
10155 struct extent_changeset *data_reserved = NULL;
10156 struct extent_state *cached_state = NULL;
10157 struct btrfs_ordered_extent *ordered;
10158 int compression;
10159 size_t orig_count;
10160 u64 start, end;
10161 u64 num_bytes, ram_bytes, disk_num_bytes;
10162 unsigned long nr_folios, i;
10163 struct folio **folios;
10164 struct btrfs_key ins;
10165 bool extent_reserved = false;
10166 struct extent_map *em;
10167 ssize_t ret;
10168
10169 switch (encoded->compression) {
10170 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10171 compression = BTRFS_COMPRESS_ZLIB;
10172 break;
10173 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10174 compression = BTRFS_COMPRESS_ZSTD;
10175 break;
10176 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10177 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10178 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10179 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10180 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10181 /* The sector size must match for LZO. */
10182 if (encoded->compression -
10183 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10184 fs_info->sectorsize_bits)
10185 return -EINVAL;
10186 compression = BTRFS_COMPRESS_LZO;
10187 break;
10188 default:
10189 return -EINVAL;
10190 }
10191 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10192 return -EINVAL;
10193
10194 /*
10195 * Compressed extents should always have checksums, so error out if we
10196 * have a NOCOW file or inode was created while mounted with NODATASUM.
10197 */
10198 if (inode->flags & BTRFS_INODE_NODATASUM)
10199 return -EINVAL;
10200
10201 orig_count = iov_iter_count(from);
10202
10203 /* The extent size must be sane. */
10204 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10205 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10206 return -EINVAL;
10207
10208 /*
10209 * The compressed data must be smaller than the decompressed data.
10210 *
10211 * It's of course possible for data to compress to larger or the same
10212 * size, but the buffered I/O path falls back to no compression for such
10213 * data, and we don't want to break any assumptions by creating these
10214 * extents.
10215 *
10216 * Note that this is less strict than the current check we have that the
10217 * compressed data must be at least one sector smaller than the
10218 * decompressed data. We only want to enforce the weaker requirement
10219 * from old kernels that it is at least one byte smaller.
10220 */
10221 if (orig_count >= encoded->unencoded_len)
10222 return -EINVAL;
10223
10224 /* The extent must start on a sector boundary. */
10225 start = iocb->ki_pos;
10226 if (!IS_ALIGNED(start, fs_info->sectorsize))
10227 return -EINVAL;
10228
10229 /*
10230 * The extent must end on a sector boundary. However, we allow a write
10231 * which ends at or extends i_size to have an unaligned length; we round
10232 * up the extent size and set i_size to the unaligned end.
10233 */
10234 if (start + encoded->len < inode->vfs_inode.i_size &&
10235 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10236 return -EINVAL;
10237
10238 /* Finally, the offset in the unencoded data must be sector-aligned. */
10239 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10240 return -EINVAL;
10241
10242 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10243 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10244 end = start + num_bytes - 1;
10245
10246 /*
10247 * If the extent cannot be inline, the compressed data on disk must be
10248 * sector-aligned. For convenience, we extend it with zeroes if it
10249 * isn't.
10250 */
10251 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10252 nr_folios = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10253 folios = kvcalloc(nr_folios, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10254 if (!folios)
10255 return -ENOMEM;
10256 for (i = 0; i < nr_folios; i++) {
10257 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10258 char *kaddr;
10259
10260 folios[i] = folio_alloc(GFP_KERNEL_ACCOUNT, 0);
10261 if (!folios[i]) {
10262 ret = -ENOMEM;
10263 goto out_folios;
10264 }
10265 kaddr = kmap_local_folio(folios[i], 0);
10266 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10267 kunmap_local(kaddr);
10268 ret = -EFAULT;
10269 goto out_folios;
10270 }
10271 if (bytes < PAGE_SIZE)
10272 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10273 kunmap_local(kaddr);
10274 }
10275
10276 for (;;) {
10277 struct btrfs_ordered_extent *ordered;
10278
10279 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10280 if (ret)
10281 goto out_folios;
10282 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10283 start >> PAGE_SHIFT,
10284 end >> PAGE_SHIFT);
10285 if (ret)
10286 goto out_folios;
10287 lock_extent(io_tree, start, end, &cached_state);
10288 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10289 if (!ordered &&
10290 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10291 break;
10292 if (ordered)
10293 btrfs_put_ordered_extent(ordered);
10294 unlock_extent(io_tree, start, end, &cached_state);
10295 cond_resched();
10296 }
10297
10298 /*
10299 * We don't use the higher-level delalloc space functions because our
10300 * num_bytes and disk_num_bytes are different.
10301 */
10302 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10303 if (ret)
10304 goto out_unlock;
10305 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10306 if (ret)
10307 goto out_free_data_space;
10308 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10309 false);
10310 if (ret)
10311 goto out_qgroup_free_data;
10312
10313 /* Try an inline extent first. */
10314 if (encoded->unencoded_len == encoded->len &&
10315 encoded->unencoded_offset == 0 &&
10316 can_cow_file_range_inline(inode, start, encoded->len, orig_count)) {
10317 ret = __cow_file_range_inline(inode, start, encoded->len,
10318 orig_count, compression, folios[0],
10319 true);
10320 if (ret <= 0) {
10321 if (ret == 0)
10322 ret = orig_count;
10323 goto out_delalloc_release;
10324 }
10325 }
10326
10327 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10328 disk_num_bytes, 0, 0, &ins, 1, 1);
10329 if (ret)
10330 goto out_delalloc_release;
10331 extent_reserved = true;
10332
10333 em = create_io_em(inode, start, num_bytes,
10334 start - encoded->unencoded_offset, ins.objectid,
10335 ins.offset, ins.offset, ram_bytes, compression,
10336 BTRFS_ORDERED_COMPRESSED);
10337 if (IS_ERR(em)) {
10338 ret = PTR_ERR(em);
10339 goto out_free_reserved;
10340 }
10341 free_extent_map(em);
10342
10343 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10344 ins.objectid, ins.offset,
10345 encoded->unencoded_offset,
10346 (1 << BTRFS_ORDERED_ENCODED) |
10347 (1 << BTRFS_ORDERED_COMPRESSED),
10348 compression);
10349 if (IS_ERR(ordered)) {
10350 btrfs_drop_extent_map_range(inode, start, end, false);
10351 ret = PTR_ERR(ordered);
10352 goto out_free_reserved;
10353 }
10354 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10355
10356 if (start + encoded->len > inode->vfs_inode.i_size)
10357 i_size_write(&inode->vfs_inode, start + encoded->len);
10358
10359 unlock_extent(io_tree, start, end, &cached_state);
10360
10361 btrfs_delalloc_release_extents(inode, num_bytes);
10362
10363 btrfs_submit_compressed_write(ordered, folios, nr_folios, 0, false);
10364 ret = orig_count;
10365 goto out;
10366
10367 out_free_reserved:
10368 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10369 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10370 out_delalloc_release:
10371 btrfs_delalloc_release_extents(inode, num_bytes);
10372 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10373 out_qgroup_free_data:
10374 if (ret < 0)
10375 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes, NULL);
10376 out_free_data_space:
10377 /*
10378 * If btrfs_reserve_extent() succeeded, then we already decremented
10379 * bytes_may_use.
10380 */
10381 if (!extent_reserved)
10382 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10383 out_unlock:
10384 unlock_extent(io_tree, start, end, &cached_state);
10385 out_folios:
10386 for (i = 0; i < nr_folios; i++) {
10387 if (folios[i])
10388 __folio_put(folios[i]);
10389 }
10390 kvfree(folios);
10391 out:
10392 if (ret >= 0)
10393 iocb->ki_pos += encoded->len;
10394 return ret;
10395 }
10396
10397 #ifdef CONFIG_SWAP
10398 /*
10399 * Add an entry indicating a block group or device which is pinned by a
10400 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10401 * negative errno on failure.
10402 */
btrfs_add_swapfile_pin(struct inode * inode,void * ptr,bool is_block_group)10403 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10404 bool is_block_group)
10405 {
10406 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10407 struct btrfs_swapfile_pin *sp, *entry;
10408 struct rb_node **p;
10409 struct rb_node *parent = NULL;
10410
10411 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10412 if (!sp)
10413 return -ENOMEM;
10414 sp->ptr = ptr;
10415 sp->inode = inode;
10416 sp->is_block_group = is_block_group;
10417 sp->bg_extent_count = 1;
10418
10419 spin_lock(&fs_info->swapfile_pins_lock);
10420 p = &fs_info->swapfile_pins.rb_node;
10421 while (*p) {
10422 parent = *p;
10423 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10424 if (sp->ptr < entry->ptr ||
10425 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10426 p = &(*p)->rb_left;
10427 } else if (sp->ptr > entry->ptr ||
10428 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10429 p = &(*p)->rb_right;
10430 } else {
10431 if (is_block_group)
10432 entry->bg_extent_count++;
10433 spin_unlock(&fs_info->swapfile_pins_lock);
10434 kfree(sp);
10435 return 1;
10436 }
10437 }
10438 rb_link_node(&sp->node, parent, p);
10439 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10440 spin_unlock(&fs_info->swapfile_pins_lock);
10441 return 0;
10442 }
10443
10444 /* Free all of the entries pinned by this swapfile. */
btrfs_free_swapfile_pins(struct inode * inode)10445 static void btrfs_free_swapfile_pins(struct inode *inode)
10446 {
10447 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10448 struct btrfs_swapfile_pin *sp;
10449 struct rb_node *node, *next;
10450
10451 spin_lock(&fs_info->swapfile_pins_lock);
10452 node = rb_first(&fs_info->swapfile_pins);
10453 while (node) {
10454 next = rb_next(node);
10455 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10456 if (sp->inode == inode) {
10457 rb_erase(&sp->node, &fs_info->swapfile_pins);
10458 if (sp->is_block_group) {
10459 btrfs_dec_block_group_swap_extents(sp->ptr,
10460 sp->bg_extent_count);
10461 btrfs_put_block_group(sp->ptr);
10462 }
10463 kfree(sp);
10464 }
10465 node = next;
10466 }
10467 spin_unlock(&fs_info->swapfile_pins_lock);
10468 }
10469
10470 struct btrfs_swap_info {
10471 u64 start;
10472 u64 block_start;
10473 u64 block_len;
10474 u64 lowest_ppage;
10475 u64 highest_ppage;
10476 unsigned long nr_pages;
10477 int nr_extents;
10478 };
10479
btrfs_add_swap_extent(struct swap_info_struct * sis,struct btrfs_swap_info * bsi)10480 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10481 struct btrfs_swap_info *bsi)
10482 {
10483 unsigned long nr_pages;
10484 unsigned long max_pages;
10485 u64 first_ppage, first_ppage_reported, next_ppage;
10486 int ret;
10487
10488 /*
10489 * Our swapfile may have had its size extended after the swap header was
10490 * written. In that case activating the swapfile should not go beyond
10491 * the max size set in the swap header.
10492 */
10493 if (bsi->nr_pages >= sis->max)
10494 return 0;
10495
10496 max_pages = sis->max - bsi->nr_pages;
10497 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10498 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10499
10500 if (first_ppage >= next_ppage)
10501 return 0;
10502 nr_pages = next_ppage - first_ppage;
10503 nr_pages = min(nr_pages, max_pages);
10504
10505 first_ppage_reported = first_ppage;
10506 if (bsi->start == 0)
10507 first_ppage_reported++;
10508 if (bsi->lowest_ppage > first_ppage_reported)
10509 bsi->lowest_ppage = first_ppage_reported;
10510 if (bsi->highest_ppage < (next_ppage - 1))
10511 bsi->highest_ppage = next_ppage - 1;
10512
10513 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10514 if (ret < 0)
10515 return ret;
10516 bsi->nr_extents += ret;
10517 bsi->nr_pages += nr_pages;
10518 return 0;
10519 }
10520
btrfs_swap_deactivate(struct file * file)10521 static void btrfs_swap_deactivate(struct file *file)
10522 {
10523 struct inode *inode = file_inode(file);
10524
10525 btrfs_free_swapfile_pins(inode);
10526 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10527 }
10528
btrfs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)10529 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10530 sector_t *span)
10531 {
10532 struct inode *inode = file_inode(file);
10533 struct btrfs_root *root = BTRFS_I(inode)->root;
10534 struct btrfs_fs_info *fs_info = root->fs_info;
10535 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10536 struct extent_state *cached_state = NULL;
10537 struct extent_map *em = NULL;
10538 struct btrfs_chunk_map *map = NULL;
10539 struct btrfs_device *device = NULL;
10540 struct btrfs_swap_info bsi = {
10541 .lowest_ppage = (sector_t)-1ULL,
10542 };
10543 int ret = 0;
10544 u64 isize;
10545 u64 start;
10546
10547 /*
10548 * If the swap file was just created, make sure delalloc is done. If the
10549 * file changes again after this, the user is doing something stupid and
10550 * we don't really care.
10551 */
10552 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10553 if (ret)
10554 return ret;
10555
10556 /*
10557 * The inode is locked, so these flags won't change after we check them.
10558 */
10559 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10560 btrfs_warn(fs_info, "swapfile must not be compressed");
10561 return -EINVAL;
10562 }
10563 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10564 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10565 return -EINVAL;
10566 }
10567 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10568 btrfs_warn(fs_info, "swapfile must not be checksummed");
10569 return -EINVAL;
10570 }
10571
10572 /*
10573 * Balance or device remove/replace/resize can move stuff around from
10574 * under us. The exclop protection makes sure they aren't running/won't
10575 * run concurrently while we are mapping the swap extents, and
10576 * fs_info->swapfile_pins prevents them from running while the swap
10577 * file is active and moving the extents. Note that this also prevents
10578 * a concurrent device add which isn't actually necessary, but it's not
10579 * really worth the trouble to allow it.
10580 */
10581 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10582 btrfs_warn(fs_info,
10583 "cannot activate swapfile while exclusive operation is running");
10584 return -EBUSY;
10585 }
10586
10587 /*
10588 * Prevent snapshot creation while we are activating the swap file.
10589 * We do not want to race with snapshot creation. If snapshot creation
10590 * already started before we bumped nr_swapfiles from 0 to 1 and
10591 * completes before the first write into the swap file after it is
10592 * activated, than that write would fallback to COW.
10593 */
10594 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10595 btrfs_exclop_finish(fs_info);
10596 btrfs_warn(fs_info,
10597 "cannot activate swapfile because snapshot creation is in progress");
10598 return -EINVAL;
10599 }
10600 /*
10601 * Snapshots can create extents which require COW even if NODATACOW is
10602 * set. We use this counter to prevent snapshots. We must increment it
10603 * before walking the extents because we don't want a concurrent
10604 * snapshot to run after we've already checked the extents.
10605 *
10606 * It is possible that subvolume is marked for deletion but still not
10607 * removed yet. To prevent this race, we check the root status before
10608 * activating the swapfile.
10609 */
10610 spin_lock(&root->root_item_lock);
10611 if (btrfs_root_dead(root)) {
10612 spin_unlock(&root->root_item_lock);
10613
10614 btrfs_exclop_finish(fs_info);
10615 btrfs_warn(fs_info,
10616 "cannot activate swapfile because subvolume %llu is being deleted",
10617 btrfs_root_id(root));
10618 return -EPERM;
10619 }
10620 atomic_inc(&root->nr_swapfiles);
10621 spin_unlock(&root->root_item_lock);
10622
10623 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10624
10625 lock_extent(io_tree, 0, isize - 1, &cached_state);
10626 start = 0;
10627 while (start < isize) {
10628 u64 logical_block_start, physical_block_start;
10629 struct btrfs_block_group *bg;
10630 u64 len = isize - start;
10631
10632 em = btrfs_get_extent(BTRFS_I(inode), NULL, start, len);
10633 if (IS_ERR(em)) {
10634 ret = PTR_ERR(em);
10635 goto out;
10636 }
10637
10638 if (em->block_start == EXTENT_MAP_HOLE) {
10639 btrfs_warn(fs_info, "swapfile must not have holes");
10640 ret = -EINVAL;
10641 goto out;
10642 }
10643 if (em->block_start == EXTENT_MAP_INLINE) {
10644 /*
10645 * It's unlikely we'll ever actually find ourselves
10646 * here, as a file small enough to fit inline won't be
10647 * big enough to store more than the swap header, but in
10648 * case something changes in the future, let's catch it
10649 * here rather than later.
10650 */
10651 btrfs_warn(fs_info, "swapfile must not be inline");
10652 ret = -EINVAL;
10653 goto out;
10654 }
10655 if (extent_map_is_compressed(em)) {
10656 btrfs_warn(fs_info, "swapfile must not be compressed");
10657 ret = -EINVAL;
10658 goto out;
10659 }
10660
10661 logical_block_start = em->block_start + (start - em->start);
10662 len = min(len, em->len - (start - em->start));
10663 free_extent_map(em);
10664 em = NULL;
10665
10666 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10667 if (ret < 0) {
10668 goto out;
10669 } else if (ret) {
10670 ret = 0;
10671 } else {
10672 btrfs_warn(fs_info,
10673 "swapfile must not be copy-on-write");
10674 ret = -EINVAL;
10675 goto out;
10676 }
10677
10678 map = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10679 if (IS_ERR(map)) {
10680 ret = PTR_ERR(map);
10681 goto out;
10682 }
10683
10684 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10685 btrfs_warn(fs_info,
10686 "swapfile must have single data profile");
10687 ret = -EINVAL;
10688 goto out;
10689 }
10690
10691 if (device == NULL) {
10692 device = map->stripes[0].dev;
10693 ret = btrfs_add_swapfile_pin(inode, device, false);
10694 if (ret == 1)
10695 ret = 0;
10696 else if (ret)
10697 goto out;
10698 } else if (device != map->stripes[0].dev) {
10699 btrfs_warn(fs_info, "swapfile must be on one device");
10700 ret = -EINVAL;
10701 goto out;
10702 }
10703
10704 physical_block_start = (map->stripes[0].physical +
10705 (logical_block_start - map->start));
10706 len = min(len, map->chunk_len - (logical_block_start - map->start));
10707 btrfs_free_chunk_map(map);
10708 map = NULL;
10709
10710 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10711 if (!bg) {
10712 btrfs_warn(fs_info,
10713 "could not find block group containing swapfile");
10714 ret = -EINVAL;
10715 goto out;
10716 }
10717
10718 if (!btrfs_inc_block_group_swap_extents(bg)) {
10719 btrfs_warn(fs_info,
10720 "block group for swapfile at %llu is read-only%s",
10721 bg->start,
10722 atomic_read(&fs_info->scrubs_running) ?
10723 " (scrub running)" : "");
10724 btrfs_put_block_group(bg);
10725 ret = -EINVAL;
10726 goto out;
10727 }
10728
10729 ret = btrfs_add_swapfile_pin(inode, bg, true);
10730 if (ret) {
10731 btrfs_put_block_group(bg);
10732 if (ret == 1)
10733 ret = 0;
10734 else
10735 goto out;
10736 }
10737
10738 if (bsi.block_len &&
10739 bsi.block_start + bsi.block_len == physical_block_start) {
10740 bsi.block_len += len;
10741 } else {
10742 if (bsi.block_len) {
10743 ret = btrfs_add_swap_extent(sis, &bsi);
10744 if (ret)
10745 goto out;
10746 }
10747 bsi.start = start;
10748 bsi.block_start = physical_block_start;
10749 bsi.block_len = len;
10750 }
10751
10752 start += len;
10753 }
10754
10755 if (bsi.block_len)
10756 ret = btrfs_add_swap_extent(sis, &bsi);
10757
10758 out:
10759 if (!IS_ERR_OR_NULL(em))
10760 free_extent_map(em);
10761 if (!IS_ERR_OR_NULL(map))
10762 btrfs_free_chunk_map(map);
10763
10764 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10765
10766 if (ret)
10767 btrfs_swap_deactivate(file);
10768
10769 btrfs_drew_write_unlock(&root->snapshot_lock);
10770
10771 btrfs_exclop_finish(fs_info);
10772
10773 if (ret)
10774 return ret;
10775
10776 if (device)
10777 sis->bdev = device->bdev;
10778 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10779 sis->max = bsi.nr_pages;
10780 sis->pages = bsi.nr_pages - 1;
10781 sis->highest_bit = bsi.nr_pages - 1;
10782 return bsi.nr_extents;
10783 }
10784 #else
btrfs_swap_deactivate(struct file * file)10785 static void btrfs_swap_deactivate(struct file *file)
10786 {
10787 }
10788
btrfs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)10789 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10790 sector_t *span)
10791 {
10792 return -EOPNOTSUPP;
10793 }
10794 #endif
10795
10796 /*
10797 * Update the number of bytes used in the VFS' inode. When we replace extents in
10798 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10799 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10800 * always get a correct value.
10801 */
btrfs_update_inode_bytes(struct btrfs_inode * inode,const u64 add_bytes,const u64 del_bytes)10802 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10803 const u64 add_bytes,
10804 const u64 del_bytes)
10805 {
10806 if (add_bytes == del_bytes)
10807 return;
10808
10809 spin_lock(&inode->lock);
10810 if (del_bytes > 0)
10811 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10812 if (add_bytes > 0)
10813 inode_add_bytes(&inode->vfs_inode, add_bytes);
10814 spin_unlock(&inode->lock);
10815 }
10816
10817 /*
10818 * Verify that there are no ordered extents for a given file range.
10819 *
10820 * @inode: The target inode.
10821 * @start: Start offset of the file range, should be sector size aligned.
10822 * @end: End offset (inclusive) of the file range, its value +1 should be
10823 * sector size aligned.
10824 *
10825 * This should typically be used for cases where we locked an inode's VFS lock in
10826 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10827 * we have flushed all delalloc in the range, we have waited for all ordered
10828 * extents in the range to complete and finally we have locked the file range in
10829 * the inode's io_tree.
10830 */
btrfs_assert_inode_range_clean(struct btrfs_inode * inode,u64 start,u64 end)10831 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10832 {
10833 struct btrfs_root *root = inode->root;
10834 struct btrfs_ordered_extent *ordered;
10835
10836 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10837 return;
10838
10839 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10840 if (ordered) {
10841 btrfs_err(root->fs_info,
10842 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10843 start, end, btrfs_ino(inode), btrfs_root_id(root),
10844 ordered->file_offset,
10845 ordered->file_offset + ordered->num_bytes - 1);
10846 btrfs_put_ordered_extent(ordered);
10847 }
10848
10849 ASSERT(ordered == NULL);
10850 }
10851
10852 /*
10853 * Find the first inode with a minimum number.
10854 *
10855 * @root: The root to search for.
10856 * @min_ino: The minimum inode number.
10857 *
10858 * Find the first inode in the @root with a number >= @min_ino and return it.
10859 * Returns NULL if no such inode found.
10860 */
btrfs_find_first_inode(struct btrfs_root * root,u64 min_ino)10861 struct btrfs_inode *btrfs_find_first_inode(struct btrfs_root *root, u64 min_ino)
10862 {
10863 struct rb_node *node;
10864 struct rb_node *prev;
10865 struct btrfs_inode *inode;
10866
10867 spin_lock(&root->inode_lock);
10868 again:
10869 node = root->inode_tree.rb_node;
10870 prev = NULL;
10871 while (node) {
10872 prev = node;
10873 inode = rb_entry(node, struct btrfs_inode, rb_node);
10874 if (min_ino < btrfs_ino(inode))
10875 node = node->rb_left;
10876 else if (min_ino > btrfs_ino(inode))
10877 node = node->rb_right;
10878 else
10879 break;
10880 }
10881
10882 if (!node) {
10883 while (prev) {
10884 inode = rb_entry(prev, struct btrfs_inode, rb_node);
10885 if (min_ino <= btrfs_ino(inode)) {
10886 node = prev;
10887 break;
10888 }
10889 prev = rb_next(prev);
10890 }
10891 }
10892
10893 while (node) {
10894 inode = rb_entry(prev, struct btrfs_inode, rb_node);
10895 if (igrab(&inode->vfs_inode)) {
10896 spin_unlock(&root->inode_lock);
10897 return inode;
10898 }
10899
10900 min_ino = btrfs_ino(inode) + 1;
10901 if (cond_resched_lock(&root->inode_lock))
10902 goto again;
10903
10904 node = rb_next(node);
10905 }
10906 spin_unlock(&root->inode_lock);
10907
10908 return NULL;
10909 }
10910
10911 static const struct inode_operations btrfs_dir_inode_operations = {
10912 .getattr = btrfs_getattr,
10913 .lookup = btrfs_lookup,
10914 .create = btrfs_create,
10915 .unlink = btrfs_unlink,
10916 .link = btrfs_link,
10917 .mkdir = btrfs_mkdir,
10918 .rmdir = btrfs_rmdir,
10919 .rename = btrfs_rename2,
10920 .symlink = btrfs_symlink,
10921 .setattr = btrfs_setattr,
10922 .mknod = btrfs_mknod,
10923 .listxattr = btrfs_listxattr,
10924 .permission = btrfs_permission,
10925 .get_inode_acl = btrfs_get_acl,
10926 .set_acl = btrfs_set_acl,
10927 .update_time = btrfs_update_time,
10928 .tmpfile = btrfs_tmpfile,
10929 .fileattr_get = btrfs_fileattr_get,
10930 .fileattr_set = btrfs_fileattr_set,
10931 };
10932
10933 static const struct file_operations btrfs_dir_file_operations = {
10934 .llseek = btrfs_dir_llseek,
10935 .read = generic_read_dir,
10936 .iterate_shared = btrfs_real_readdir,
10937 .open = btrfs_opendir,
10938 .unlocked_ioctl = btrfs_ioctl,
10939 #ifdef CONFIG_COMPAT
10940 .compat_ioctl = btrfs_compat_ioctl,
10941 #endif
10942 .release = btrfs_release_file,
10943 .fsync = btrfs_sync_file,
10944 };
10945
10946 /*
10947 * btrfs doesn't support the bmap operation because swapfiles
10948 * use bmap to make a mapping of extents in the file. They assume
10949 * these extents won't change over the life of the file and they
10950 * use the bmap result to do IO directly to the drive.
10951 *
10952 * the btrfs bmap call would return logical addresses that aren't
10953 * suitable for IO and they also will change frequently as COW
10954 * operations happen. So, swapfile + btrfs == corruption.
10955 *
10956 * For now we're avoiding this by dropping bmap.
10957 */
10958 static const struct address_space_operations btrfs_aops = {
10959 .read_folio = btrfs_read_folio,
10960 .writepages = btrfs_writepages,
10961 .readahead = btrfs_readahead,
10962 .invalidate_folio = btrfs_invalidate_folio,
10963 .release_folio = btrfs_release_folio,
10964 .migrate_folio = btrfs_migrate_folio,
10965 .dirty_folio = filemap_dirty_folio,
10966 .error_remove_folio = generic_error_remove_folio,
10967 .swap_activate = btrfs_swap_activate,
10968 .swap_deactivate = btrfs_swap_deactivate,
10969 };
10970
10971 static const struct inode_operations btrfs_file_inode_operations = {
10972 .getattr = btrfs_getattr,
10973 .setattr = btrfs_setattr,
10974 .listxattr = btrfs_listxattr,
10975 .permission = btrfs_permission,
10976 .fiemap = btrfs_fiemap,
10977 .get_inode_acl = btrfs_get_acl,
10978 .set_acl = btrfs_set_acl,
10979 .update_time = btrfs_update_time,
10980 .fileattr_get = btrfs_fileattr_get,
10981 .fileattr_set = btrfs_fileattr_set,
10982 };
10983 static const struct inode_operations btrfs_special_inode_operations = {
10984 .getattr = btrfs_getattr,
10985 .setattr = btrfs_setattr,
10986 .permission = btrfs_permission,
10987 .listxattr = btrfs_listxattr,
10988 .get_inode_acl = btrfs_get_acl,
10989 .set_acl = btrfs_set_acl,
10990 .update_time = btrfs_update_time,
10991 };
10992 static const struct inode_operations btrfs_symlink_inode_operations = {
10993 .get_link = page_get_link,
10994 .getattr = btrfs_getattr,
10995 .setattr = btrfs_setattr,
10996 .permission = btrfs_permission,
10997 .listxattr = btrfs_listxattr,
10998 .update_time = btrfs_update_time,
10999 };
11000
11001 const struct dentry_operations btrfs_dentry_operations = {
11002 .d_delete = btrfs_dentry_delete,
11003 };
11004