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 "print-tree.h"
43 #include "ordered-data.h"
44 #include "xattr.h"
45 #include "tree-log.h"
46 #include "bio.h"
47 #include "compression.h"
48 #include "locking.h"
49 #include "free-space-cache.h"
50 #include "props.h"
51 #include "qgroup.h"
52 #include "delalloc-space.h"
53 #include "block-group.h"
54 #include "space-info.h"
55 #include "zoned.h"
56 #include "subpage.h"
57 #include "inode-item.h"
58 #include "fs.h"
59 #include "accessors.h"
60 #include "extent-tree.h"
61 #include "root-tree.h"
62 #include "defrag.h"
63 #include "dir-item.h"
64 #include "file-item.h"
65 #include "uuid-tree.h"
66 #include "ioctl.h"
67 #include "file.h"
68 #include "acl.h"
69 #include "relocation.h"
70 #include "verity.h"
71 #include "super.h"
72 #include "orphan.h"
73 #include "backref.h"
74
75 struct btrfs_iget_args {
76 u64 ino;
77 struct btrfs_root *root;
78 };
79
80 struct btrfs_dio_data {
81 ssize_t submitted;
82 struct extent_changeset *data_reserved;
83 struct btrfs_ordered_extent *ordered;
84 bool data_space_reserved;
85 bool nocow_done;
86 };
87
88 struct btrfs_dio_private {
89 /* Range of I/O */
90 u64 file_offset;
91 u32 bytes;
92
93 /* This must be last */
94 struct btrfs_bio bbio;
95 };
96
97 static struct bio_set btrfs_dio_bioset;
98
99 struct btrfs_rename_ctx {
100 /* Output field. Stores the index number of the old directory entry. */
101 u64 index;
102 };
103
104 /*
105 * Used by data_reloc_print_warning_inode() to pass needed info for filename
106 * resolution and output of error message.
107 */
108 struct data_reloc_warn {
109 struct btrfs_path path;
110 struct btrfs_fs_info *fs_info;
111 u64 extent_item_size;
112 u64 logical;
113 int mirror_num;
114 };
115
116 static const struct inode_operations btrfs_dir_inode_operations;
117 static const struct inode_operations btrfs_symlink_inode_operations;
118 static const struct inode_operations btrfs_special_inode_operations;
119 static const struct inode_operations btrfs_file_inode_operations;
120 static const struct address_space_operations btrfs_aops;
121 static const struct file_operations btrfs_dir_file_operations;
122
123 static struct kmem_cache *btrfs_inode_cachep;
124
125 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
126 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback);
127
128 static noinline int run_delalloc_cow(struct btrfs_inode *inode,
129 struct page *locked_page, u64 start,
130 u64 end, struct writeback_control *wbc,
131 bool pages_dirty);
132 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
133 u64 len, u64 orig_start, u64 block_start,
134 u64 block_len, u64 orig_block_len,
135 u64 ram_bytes, int compress_type,
136 int type);
137
data_reloc_print_warning_inode(u64 inum,u64 offset,u64 num_bytes,u64 root,void * warn_ctx)138 static int data_reloc_print_warning_inode(u64 inum, u64 offset, u64 num_bytes,
139 u64 root, void *warn_ctx)
140 {
141 struct data_reloc_warn *warn = warn_ctx;
142 struct btrfs_fs_info *fs_info = warn->fs_info;
143 struct extent_buffer *eb;
144 struct btrfs_inode_item *inode_item;
145 struct inode_fs_paths *ipath = NULL;
146 struct btrfs_root *local_root;
147 struct btrfs_key key;
148 unsigned int nofs_flag;
149 u32 nlink;
150 int ret;
151
152 local_root = btrfs_get_fs_root(fs_info, root, true);
153 if (IS_ERR(local_root)) {
154 ret = PTR_ERR(local_root);
155 goto err;
156 }
157
158 /* This makes the path point to (inum INODE_ITEM ioff). */
159 key.objectid = inum;
160 key.type = BTRFS_INODE_ITEM_KEY;
161 key.offset = 0;
162
163 ret = btrfs_search_slot(NULL, local_root, &key, &warn->path, 0, 0);
164 if (ret) {
165 btrfs_put_root(local_root);
166 btrfs_release_path(&warn->path);
167 goto err;
168 }
169
170 eb = warn->path.nodes[0];
171 inode_item = btrfs_item_ptr(eb, warn->path.slots[0], struct btrfs_inode_item);
172 nlink = btrfs_inode_nlink(eb, inode_item);
173 btrfs_release_path(&warn->path);
174
175 nofs_flag = memalloc_nofs_save();
176 ipath = init_ipath(4096, local_root, &warn->path);
177 memalloc_nofs_restore(nofs_flag);
178 if (IS_ERR(ipath)) {
179 btrfs_put_root(local_root);
180 ret = PTR_ERR(ipath);
181 ipath = NULL;
182 /*
183 * -ENOMEM, not a critical error, just output an generic error
184 * without filename.
185 */
186 btrfs_warn(fs_info,
187 "checksum error at logical %llu mirror %u root %llu, inode %llu offset %llu",
188 warn->logical, warn->mirror_num, root, inum, offset);
189 return ret;
190 }
191 ret = paths_from_inode(inum, ipath);
192 if (ret < 0)
193 goto err;
194
195 /*
196 * We deliberately ignore the bit ipath might have been too small to
197 * hold all of the paths here
198 */
199 for (int i = 0; i < ipath->fspath->elem_cnt; i++) {
200 btrfs_warn(fs_info,
201 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu length %u links %u (path: %s)",
202 warn->logical, warn->mirror_num, root, inum, offset,
203 fs_info->sectorsize, nlink,
204 (char *)(unsigned long)ipath->fspath->val[i]);
205 }
206
207 btrfs_put_root(local_root);
208 free_ipath(ipath);
209 return 0;
210
211 err:
212 btrfs_warn(fs_info,
213 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu, path resolving failed with ret=%d",
214 warn->logical, warn->mirror_num, root, inum, offset, ret);
215
216 free_ipath(ipath);
217 return ret;
218 }
219
220 /*
221 * Do extra user-friendly error output (e.g. lookup all the affected files).
222 *
223 * Return true if we succeeded doing the backref lookup.
224 * Return false if such lookup failed, and has to fallback to the old error message.
225 */
print_data_reloc_error(const struct btrfs_inode * inode,u64 file_off,const u8 * csum,const u8 * csum_expected,int mirror_num)226 static void print_data_reloc_error(const struct btrfs_inode *inode, u64 file_off,
227 const u8 *csum, const u8 *csum_expected,
228 int mirror_num)
229 {
230 struct btrfs_fs_info *fs_info = inode->root->fs_info;
231 struct btrfs_path path = { 0 };
232 struct btrfs_key found_key = { 0 };
233 struct extent_buffer *eb;
234 struct btrfs_extent_item *ei;
235 const u32 csum_size = fs_info->csum_size;
236 u64 logical;
237 u64 flags;
238 u32 item_size;
239 int ret;
240
241 mutex_lock(&fs_info->reloc_mutex);
242 logical = btrfs_get_reloc_bg_bytenr(fs_info);
243 mutex_unlock(&fs_info->reloc_mutex);
244
245 if (logical == U64_MAX) {
246 btrfs_warn_rl(fs_info, "has data reloc tree but no running relocation");
247 btrfs_warn_rl(fs_info,
248 "csum failed root %lld ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
249 inode->root->root_key.objectid, btrfs_ino(inode), file_off,
250 CSUM_FMT_VALUE(csum_size, csum),
251 CSUM_FMT_VALUE(csum_size, csum_expected),
252 mirror_num);
253 return;
254 }
255
256 logical += file_off;
257 btrfs_warn_rl(fs_info,
258 "csum failed root %lld ino %llu off %llu logical %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
259 inode->root->root_key.objectid,
260 btrfs_ino(inode), file_off, logical,
261 CSUM_FMT_VALUE(csum_size, csum),
262 CSUM_FMT_VALUE(csum_size, csum_expected),
263 mirror_num);
264
265 ret = extent_from_logical(fs_info, logical, &path, &found_key, &flags);
266 if (ret < 0) {
267 btrfs_err_rl(fs_info, "failed to lookup extent item for logical %llu: %d",
268 logical, ret);
269 return;
270 }
271 eb = path.nodes[0];
272 ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item);
273 item_size = btrfs_item_size(eb, path.slots[0]);
274 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
275 unsigned long ptr = 0;
276 u64 ref_root;
277 u8 ref_level;
278
279 while (true) {
280 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
281 item_size, &ref_root,
282 &ref_level);
283 if (ret < 0) {
284 btrfs_warn_rl(fs_info,
285 "failed to resolve tree backref for logical %llu: %d",
286 logical, ret);
287 break;
288 }
289 if (ret > 0)
290 break;
291
292 btrfs_warn_rl(fs_info,
293 "csum error at logical %llu mirror %u: metadata %s (level %d) in tree %llu",
294 logical, mirror_num,
295 (ref_level ? "node" : "leaf"),
296 ref_level, ref_root);
297 }
298 btrfs_release_path(&path);
299 } else {
300 struct btrfs_backref_walk_ctx ctx = { 0 };
301 struct data_reloc_warn reloc_warn = { 0 };
302
303 btrfs_release_path(&path);
304
305 ctx.bytenr = found_key.objectid;
306 ctx.extent_item_pos = logical - found_key.objectid;
307 ctx.fs_info = fs_info;
308
309 reloc_warn.logical = logical;
310 reloc_warn.extent_item_size = found_key.offset;
311 reloc_warn.mirror_num = mirror_num;
312 reloc_warn.fs_info = fs_info;
313
314 iterate_extent_inodes(&ctx, true,
315 data_reloc_print_warning_inode, &reloc_warn);
316 }
317 }
318
btrfs_print_data_csum_error(struct btrfs_inode * inode,u64 logical_start,u8 * csum,u8 * csum_expected,int mirror_num)319 static void __cold btrfs_print_data_csum_error(struct btrfs_inode *inode,
320 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
321 {
322 struct btrfs_root *root = inode->root;
323 const u32 csum_size = root->fs_info->csum_size;
324
325 /* For data reloc tree, it's better to do a backref lookup instead. */
326 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
327 return print_data_reloc_error(inode, logical_start, csum,
328 csum_expected, mirror_num);
329
330 /* Output without objectid, which is more meaningful */
331 if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID) {
332 btrfs_warn_rl(root->fs_info,
333 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
334 root->root_key.objectid, btrfs_ino(inode),
335 logical_start,
336 CSUM_FMT_VALUE(csum_size, csum),
337 CSUM_FMT_VALUE(csum_size, csum_expected),
338 mirror_num);
339 } else {
340 btrfs_warn_rl(root->fs_info,
341 "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
342 root->root_key.objectid, 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 }
348 }
349
350 /*
351 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
352 *
353 * ilock_flags can have the following bit set:
354 *
355 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
356 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
357 * return -EAGAIN
358 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
359 */
btrfs_inode_lock(struct btrfs_inode * inode,unsigned int ilock_flags)360 int btrfs_inode_lock(struct btrfs_inode *inode, unsigned int ilock_flags)
361 {
362 if (ilock_flags & BTRFS_ILOCK_SHARED) {
363 if (ilock_flags & BTRFS_ILOCK_TRY) {
364 if (!inode_trylock_shared(&inode->vfs_inode))
365 return -EAGAIN;
366 else
367 return 0;
368 }
369 inode_lock_shared(&inode->vfs_inode);
370 } else {
371 if (ilock_flags & BTRFS_ILOCK_TRY) {
372 if (!inode_trylock(&inode->vfs_inode))
373 return -EAGAIN;
374 else
375 return 0;
376 }
377 inode_lock(&inode->vfs_inode);
378 }
379 if (ilock_flags & BTRFS_ILOCK_MMAP)
380 down_write(&inode->i_mmap_lock);
381 return 0;
382 }
383
384 /*
385 * btrfs_inode_unlock - unock inode i_rwsem
386 *
387 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
388 * to decide whether the lock acquired is shared or exclusive.
389 */
btrfs_inode_unlock(struct btrfs_inode * inode,unsigned int ilock_flags)390 void btrfs_inode_unlock(struct btrfs_inode *inode, unsigned int ilock_flags)
391 {
392 if (ilock_flags & BTRFS_ILOCK_MMAP)
393 up_write(&inode->i_mmap_lock);
394 if (ilock_flags & BTRFS_ILOCK_SHARED)
395 inode_unlock_shared(&inode->vfs_inode);
396 else
397 inode_unlock(&inode->vfs_inode);
398 }
399
400 /*
401 * Cleanup all submitted ordered extents in specified range to handle errors
402 * from the btrfs_run_delalloc_range() callback.
403 *
404 * NOTE: caller must ensure that when an error happens, it can not call
405 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
406 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
407 * to be released, which we want to happen only when finishing the ordered
408 * extent (btrfs_finish_ordered_io()).
409 */
btrfs_cleanup_ordered_extents(struct btrfs_inode * inode,struct page * locked_page,u64 offset,u64 bytes)410 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
411 struct page *locked_page,
412 u64 offset, u64 bytes)
413 {
414 unsigned long index = offset >> PAGE_SHIFT;
415 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
416 u64 page_start = 0, page_end = 0;
417 struct page *page;
418
419 if (locked_page) {
420 page_start = page_offset(locked_page);
421 page_end = page_start + PAGE_SIZE - 1;
422 }
423
424 while (index <= end_index) {
425 /*
426 * For locked page, we will call btrfs_mark_ordered_io_finished
427 * through btrfs_mark_ordered_io_finished() on it
428 * in run_delalloc_range() for the error handling, which will
429 * clear page Ordered and run the ordered extent accounting.
430 *
431 * Here we can't just clear the Ordered bit, or
432 * btrfs_mark_ordered_io_finished() would skip the accounting
433 * for the page range, and the ordered extent will never finish.
434 */
435 if (locked_page && index == (page_start >> PAGE_SHIFT)) {
436 index++;
437 continue;
438 }
439 page = find_get_page(inode->vfs_inode.i_mapping, index);
440 index++;
441 if (!page)
442 continue;
443
444 /*
445 * Here we just clear all Ordered bits for every page in the
446 * range, then btrfs_mark_ordered_io_finished() will handle
447 * the ordered extent accounting for the range.
448 */
449 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
450 offset, bytes);
451 put_page(page);
452 }
453
454 if (locked_page) {
455 /* The locked page covers the full range, nothing needs to be done */
456 if (bytes + offset <= page_start + PAGE_SIZE)
457 return;
458 /*
459 * In case this page belongs to the delalloc range being
460 * instantiated then skip it, since the first page of a range is
461 * going to be properly cleaned up by the caller of
462 * run_delalloc_range
463 */
464 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
465 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
466 offset = page_offset(locked_page) + PAGE_SIZE;
467 }
468 }
469
470 return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false);
471 }
472
473 static int btrfs_dirty_inode(struct btrfs_inode *inode);
474
btrfs_init_inode_security(struct btrfs_trans_handle * trans,struct btrfs_new_inode_args * args)475 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
476 struct btrfs_new_inode_args *args)
477 {
478 int err;
479
480 if (args->default_acl) {
481 err = __btrfs_set_acl(trans, args->inode, args->default_acl,
482 ACL_TYPE_DEFAULT);
483 if (err)
484 return err;
485 }
486 if (args->acl) {
487 err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
488 if (err)
489 return err;
490 }
491 if (!args->default_acl && !args->acl)
492 cache_no_acl(args->inode);
493 return btrfs_xattr_security_init(trans, args->inode, args->dir,
494 &args->dentry->d_name);
495 }
496
497 /*
498 * this does all the hard work for inserting an inline extent into
499 * the btree. The caller should have done a btrfs_drop_extents so that
500 * no overlapping inline items exist in the btree
501 */
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 page ** compressed_pages,bool update_i_size)502 static int insert_inline_extent(struct btrfs_trans_handle *trans,
503 struct btrfs_path *path,
504 struct btrfs_inode *inode, bool extent_inserted,
505 size_t size, size_t compressed_size,
506 int compress_type,
507 struct page **compressed_pages,
508 bool update_i_size)
509 {
510 struct btrfs_root *root = inode->root;
511 struct extent_buffer *leaf;
512 struct page *page = NULL;
513 char *kaddr;
514 unsigned long ptr;
515 struct btrfs_file_extent_item *ei;
516 int ret;
517 size_t cur_size = size;
518 u64 i_size;
519
520 ASSERT((compressed_size > 0 && compressed_pages) ||
521 (compressed_size == 0 && !compressed_pages));
522
523 if (compressed_size && compressed_pages)
524 cur_size = compressed_size;
525
526 if (!extent_inserted) {
527 struct btrfs_key key;
528 size_t datasize;
529
530 key.objectid = btrfs_ino(inode);
531 key.offset = 0;
532 key.type = BTRFS_EXTENT_DATA_KEY;
533
534 datasize = btrfs_file_extent_calc_inline_size(cur_size);
535 ret = btrfs_insert_empty_item(trans, root, path, &key,
536 datasize);
537 if (ret)
538 goto fail;
539 }
540 leaf = path->nodes[0];
541 ei = btrfs_item_ptr(leaf, path->slots[0],
542 struct btrfs_file_extent_item);
543 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
544 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
545 btrfs_set_file_extent_encryption(leaf, ei, 0);
546 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
547 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
548 ptr = btrfs_file_extent_inline_start(ei);
549
550 if (compress_type != BTRFS_COMPRESS_NONE) {
551 struct page *cpage;
552 int i = 0;
553 while (compressed_size > 0) {
554 cpage = compressed_pages[i];
555 cur_size = min_t(unsigned long, compressed_size,
556 PAGE_SIZE);
557
558 kaddr = kmap_local_page(cpage);
559 write_extent_buffer(leaf, kaddr, ptr, cur_size);
560 kunmap_local(kaddr);
561
562 i++;
563 ptr += cur_size;
564 compressed_size -= cur_size;
565 }
566 btrfs_set_file_extent_compression(leaf, ei,
567 compress_type);
568 } else {
569 page = find_get_page(inode->vfs_inode.i_mapping, 0);
570 btrfs_set_file_extent_compression(leaf, ei, 0);
571 kaddr = kmap_local_page(page);
572 write_extent_buffer(leaf, kaddr, ptr, size);
573 kunmap_local(kaddr);
574 put_page(page);
575 }
576 btrfs_mark_buffer_dirty(trans, leaf);
577 btrfs_release_path(path);
578
579 /*
580 * We align size to sectorsize for inline extents just for simplicity
581 * sake.
582 */
583 ret = btrfs_inode_set_file_extent_range(inode, 0,
584 ALIGN(size, root->fs_info->sectorsize));
585 if (ret)
586 goto fail;
587
588 /*
589 * We're an inline extent, so nobody can extend the file past i_size
590 * without locking a page we already have locked.
591 *
592 * We must do any i_size and inode updates before we unlock the pages.
593 * Otherwise we could end up racing with unlink.
594 */
595 i_size = i_size_read(&inode->vfs_inode);
596 if (update_i_size && size > i_size) {
597 i_size_write(&inode->vfs_inode, size);
598 i_size = size;
599 }
600 inode->disk_i_size = i_size;
601
602 fail:
603 return ret;
604 }
605
606
607 /*
608 * conditionally insert an inline extent into the file. This
609 * does the checks required to make sure the data is small enough
610 * to fit as an inline extent.
611 */
cow_file_range_inline(struct btrfs_inode * inode,u64 size,size_t compressed_size,int compress_type,struct page ** compressed_pages,bool update_i_size)612 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
613 size_t compressed_size,
614 int compress_type,
615 struct page **compressed_pages,
616 bool update_i_size)
617 {
618 struct btrfs_drop_extents_args drop_args = { 0 };
619 struct btrfs_root *root = inode->root;
620 struct btrfs_fs_info *fs_info = root->fs_info;
621 struct btrfs_trans_handle *trans;
622 u64 data_len = (compressed_size ?: size);
623 int ret;
624 struct btrfs_path *path;
625
626 /*
627 * We can create an inline extent if it ends at or beyond the current
628 * i_size, is no larger than a sector (decompressed), and the (possibly
629 * compressed) data fits in a leaf and the configured maximum inline
630 * size.
631 */
632 if (size < i_size_read(&inode->vfs_inode) ||
633 size > fs_info->sectorsize ||
634 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
635 data_len > fs_info->max_inline)
636 return 1;
637
638 path = btrfs_alloc_path();
639 if (!path)
640 return -ENOMEM;
641
642 trans = btrfs_join_transaction(root);
643 if (IS_ERR(trans)) {
644 btrfs_free_path(path);
645 return PTR_ERR(trans);
646 }
647 trans->block_rsv = &inode->block_rsv;
648
649 drop_args.path = path;
650 drop_args.start = 0;
651 drop_args.end = fs_info->sectorsize;
652 drop_args.drop_cache = true;
653 drop_args.replace_extent = true;
654 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
655 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
656 if (ret) {
657 btrfs_abort_transaction(trans, ret);
658 goto out;
659 }
660
661 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
662 size, compressed_size, compress_type,
663 compressed_pages, update_i_size);
664 if (ret && ret != -ENOSPC) {
665 btrfs_abort_transaction(trans, ret);
666 goto out;
667 } else if (ret == -ENOSPC) {
668 ret = 1;
669 goto out;
670 }
671
672 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
673 ret = btrfs_update_inode(trans, root, inode);
674 if (ret && ret != -ENOSPC) {
675 btrfs_abort_transaction(trans, ret);
676 goto out;
677 } else if (ret == -ENOSPC) {
678 ret = 1;
679 goto out;
680 }
681
682 btrfs_set_inode_full_sync(inode);
683 out:
684 /*
685 * Don't forget to free the reserved space, as for inlined extent
686 * it won't count as data extent, free them directly here.
687 * And at reserve time, it's always aligned to page size, so
688 * just free one page here.
689 */
690 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE, NULL);
691 btrfs_free_path(path);
692 btrfs_end_transaction(trans);
693 return ret;
694 }
695
696 struct async_extent {
697 u64 start;
698 u64 ram_size;
699 u64 compressed_size;
700 struct page **pages;
701 unsigned long nr_pages;
702 int compress_type;
703 struct list_head list;
704 };
705
706 struct async_chunk {
707 struct btrfs_inode *inode;
708 struct page *locked_page;
709 u64 start;
710 u64 end;
711 blk_opf_t write_flags;
712 struct list_head extents;
713 struct cgroup_subsys_state *blkcg_css;
714 struct btrfs_work work;
715 struct async_cow *async_cow;
716 };
717
718 struct async_cow {
719 atomic_t num_chunks;
720 struct async_chunk chunks[];
721 };
722
add_async_extent(struct async_chunk * cow,u64 start,u64 ram_size,u64 compressed_size,struct page ** pages,unsigned long nr_pages,int compress_type)723 static noinline int add_async_extent(struct async_chunk *cow,
724 u64 start, u64 ram_size,
725 u64 compressed_size,
726 struct page **pages,
727 unsigned long nr_pages,
728 int compress_type)
729 {
730 struct async_extent *async_extent;
731
732 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
733 if (!async_extent)
734 return -ENOMEM;
735 async_extent->start = start;
736 async_extent->ram_size = ram_size;
737 async_extent->compressed_size = compressed_size;
738 async_extent->pages = pages;
739 async_extent->nr_pages = nr_pages;
740 async_extent->compress_type = compress_type;
741 list_add_tail(&async_extent->list, &cow->extents);
742 return 0;
743 }
744
745 /*
746 * Check if the inode needs to be submitted to compression, based on mount
747 * options, defragmentation, properties or heuristics.
748 */
inode_need_compress(struct btrfs_inode * inode,u64 start,u64 end)749 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
750 u64 end)
751 {
752 struct btrfs_fs_info *fs_info = inode->root->fs_info;
753
754 if (!btrfs_inode_can_compress(inode)) {
755 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
756 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
757 btrfs_ino(inode));
758 return 0;
759 }
760 /*
761 * Special check for subpage.
762 *
763 * We lock the full page then run each delalloc range in the page, thus
764 * for the following case, we will hit some subpage specific corner case:
765 *
766 * 0 32K 64K
767 * | |///////| |///////|
768 * \- A \- B
769 *
770 * In above case, both range A and range B will try to unlock the full
771 * page [0, 64K), causing the one finished later will have page
772 * unlocked already, triggering various page lock requirement BUG_ON()s.
773 *
774 * So here we add an artificial limit that subpage compression can only
775 * if the range is fully page aligned.
776 *
777 * In theory we only need to ensure the first page is fully covered, but
778 * the tailing partial page will be locked until the full compression
779 * finishes, delaying the write of other range.
780 *
781 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
782 * first to prevent any submitted async extent to unlock the full page.
783 * By this, we can ensure for subpage case that only the last async_cow
784 * will unlock the full page.
785 */
786 if (fs_info->sectorsize < PAGE_SIZE) {
787 if (!PAGE_ALIGNED(start) ||
788 !PAGE_ALIGNED(end + 1))
789 return 0;
790 }
791
792 /* force compress */
793 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
794 return 1;
795 /* defrag ioctl */
796 if (inode->defrag_compress)
797 return 1;
798 /* bad compression ratios */
799 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
800 return 0;
801 if (btrfs_test_opt(fs_info, COMPRESS) ||
802 inode->flags & BTRFS_INODE_COMPRESS ||
803 inode->prop_compress)
804 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
805 return 0;
806 }
807
inode_should_defrag(struct btrfs_inode * inode,u64 start,u64 end,u64 num_bytes,u32 small_write)808 static inline void inode_should_defrag(struct btrfs_inode *inode,
809 u64 start, u64 end, u64 num_bytes, u32 small_write)
810 {
811 /* If this is a small write inside eof, kick off a defrag */
812 if (num_bytes < small_write &&
813 (start > 0 || end + 1 < inode->disk_i_size))
814 btrfs_add_inode_defrag(NULL, inode, small_write);
815 }
816
817 /*
818 * Work queue call back to started compression on a file and pages.
819 *
820 * This is done inside an ordered work queue, and the compression is spread
821 * across many cpus. The actual IO submission is step two, and the ordered work
822 * queue takes care of making sure that happens in the same order things were
823 * put onto the queue by writepages and friends.
824 *
825 * If this code finds it can't get good compression, it puts an entry onto the
826 * work queue to write the uncompressed bytes. This makes sure that both
827 * compressed inodes and uncompressed inodes are written in the same order that
828 * the flusher thread sent them down.
829 */
compress_file_range(struct btrfs_work * work)830 static void compress_file_range(struct btrfs_work *work)
831 {
832 struct async_chunk *async_chunk =
833 container_of(work, struct async_chunk, work);
834 struct btrfs_inode *inode = async_chunk->inode;
835 struct btrfs_fs_info *fs_info = inode->root->fs_info;
836 struct address_space *mapping = inode->vfs_inode.i_mapping;
837 u64 blocksize = fs_info->sectorsize;
838 u64 start = async_chunk->start;
839 u64 end = async_chunk->end;
840 u64 actual_end;
841 u64 i_size;
842 int ret = 0;
843 struct page **pages;
844 unsigned long nr_pages;
845 unsigned long total_compressed = 0;
846 unsigned long total_in = 0;
847 unsigned int poff;
848 int i;
849 int compress_type = fs_info->compress_type;
850
851 inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
852
853 /*
854 * We need to call clear_page_dirty_for_io on each page in the range.
855 * Otherwise applications with the file mmap'd can wander in and change
856 * the page contents while we are compressing them.
857 */
858 extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end);
859
860 /*
861 * We need to save i_size before now because it could change in between
862 * us evaluating the size and assigning it. This is because we lock and
863 * unlock the page in truncate and fallocate, and then modify the i_size
864 * later on.
865 *
866 * The barriers are to emulate READ_ONCE, remove that once i_size_read
867 * does that for us.
868 */
869 barrier();
870 i_size = i_size_read(&inode->vfs_inode);
871 barrier();
872 actual_end = min_t(u64, i_size, end + 1);
873 again:
874 pages = NULL;
875 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
876 nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES);
877
878 /*
879 * we don't want to send crud past the end of i_size through
880 * compression, that's just a waste of CPU time. So, if the
881 * end of the file is before the start of our current
882 * requested range of bytes, we bail out to the uncompressed
883 * cleanup code that can deal with all of this.
884 *
885 * It isn't really the fastest way to fix things, but this is a
886 * very uncommon corner.
887 */
888 if (actual_end <= start)
889 goto cleanup_and_bail_uncompressed;
890
891 total_compressed = actual_end - start;
892
893 /*
894 * Skip compression for a small file range(<=blocksize) that
895 * isn't an inline extent, since it doesn't save disk space at all.
896 */
897 if (total_compressed <= blocksize &&
898 (start > 0 || end + 1 < inode->disk_i_size))
899 goto cleanup_and_bail_uncompressed;
900
901 /*
902 * For subpage case, we require full page alignment for the sector
903 * aligned range.
904 * Thus we must also check against @actual_end, not just @end.
905 */
906 if (blocksize < PAGE_SIZE) {
907 if (!PAGE_ALIGNED(start) ||
908 !PAGE_ALIGNED(round_up(actual_end, blocksize)))
909 goto cleanup_and_bail_uncompressed;
910 }
911
912 total_compressed = min_t(unsigned long, total_compressed,
913 BTRFS_MAX_UNCOMPRESSED);
914 total_in = 0;
915 ret = 0;
916
917 /*
918 * We do compression for mount -o compress and when the inode has not
919 * been flagged as NOCOMPRESS. This flag can change at any time if we
920 * discover bad compression ratios.
921 */
922 if (!inode_need_compress(inode, start, end))
923 goto cleanup_and_bail_uncompressed;
924
925 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
926 if (!pages) {
927 /*
928 * Memory allocation failure is not a fatal error, we can fall
929 * back to uncompressed code.
930 */
931 goto cleanup_and_bail_uncompressed;
932 }
933
934 if (inode->defrag_compress)
935 compress_type = inode->defrag_compress;
936 else if (inode->prop_compress)
937 compress_type = inode->prop_compress;
938
939 /* Compression level is applied here. */
940 ret = btrfs_compress_pages(compress_type | (fs_info->compress_level << 4),
941 mapping, start, pages, &nr_pages, &total_in,
942 &total_compressed);
943 if (ret)
944 goto mark_incompressible;
945
946 /*
947 * Zero the tail end of the last page, as we might be sending it down
948 * to disk.
949 */
950 poff = offset_in_page(total_compressed);
951 if (poff)
952 memzero_page(pages[nr_pages - 1], poff, PAGE_SIZE - poff);
953
954 /*
955 * Try to create an inline extent.
956 *
957 * If we didn't compress the entire range, try to create an uncompressed
958 * inline extent, else a compressed one.
959 *
960 * Check cow_file_range() for why we don't even try to create inline
961 * extent for the subpage case.
962 */
963 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
964 if (total_in < actual_end) {
965 ret = cow_file_range_inline(inode, actual_end, 0,
966 BTRFS_COMPRESS_NONE, NULL,
967 false);
968 } else {
969 ret = cow_file_range_inline(inode, actual_end,
970 total_compressed,
971 compress_type, pages,
972 false);
973 }
974 if (ret <= 0) {
975 unsigned long clear_flags = EXTENT_DELALLOC |
976 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
977 EXTENT_DO_ACCOUNTING;
978
979 if (ret < 0)
980 mapping_set_error(mapping, -EIO);
981
982 /*
983 * inline extent creation worked or returned error,
984 * we don't need to create any more async work items.
985 * Unlock and free up our temp pages.
986 *
987 * We use DO_ACCOUNTING here because we need the
988 * delalloc_release_metadata to be done _after_ we drop
989 * our outstanding extent for clearing delalloc for this
990 * range.
991 */
992 extent_clear_unlock_delalloc(inode, start, end,
993 NULL,
994 clear_flags,
995 PAGE_UNLOCK |
996 PAGE_START_WRITEBACK |
997 PAGE_END_WRITEBACK);
998 goto free_pages;
999 }
1000 }
1001
1002 /*
1003 * We aren't doing an inline extent. Round the compressed size up to a
1004 * block size boundary so the allocator does sane things.
1005 */
1006 total_compressed = ALIGN(total_compressed, blocksize);
1007
1008 /*
1009 * One last check to make sure the compression is really a win, compare
1010 * the page count read with the blocks on disk, compression must free at
1011 * least one sector.
1012 */
1013 total_in = round_up(total_in, fs_info->sectorsize);
1014 if (total_compressed + blocksize > total_in)
1015 goto mark_incompressible;
1016
1017 /*
1018 * The async work queues will take care of doing actual allocation on
1019 * disk for these compressed pages, and will submit the bios.
1020 */
1021 ret = add_async_extent(async_chunk, start, total_in, total_compressed, pages,
1022 nr_pages, compress_type);
1023 BUG_ON(ret);
1024 if (start + total_in < end) {
1025 start += total_in;
1026 cond_resched();
1027 goto again;
1028 }
1029 return;
1030
1031 mark_incompressible:
1032 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && !inode->prop_compress)
1033 inode->flags |= BTRFS_INODE_NOCOMPRESS;
1034 cleanup_and_bail_uncompressed:
1035 ret = add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
1036 BTRFS_COMPRESS_NONE);
1037 BUG_ON(ret);
1038 free_pages:
1039 if (pages) {
1040 for (i = 0; i < nr_pages; i++) {
1041 WARN_ON(pages[i]->mapping);
1042 put_page(pages[i]);
1043 }
1044 kfree(pages);
1045 }
1046 }
1047
free_async_extent_pages(struct async_extent * async_extent)1048 static void free_async_extent_pages(struct async_extent *async_extent)
1049 {
1050 int i;
1051
1052 if (!async_extent->pages)
1053 return;
1054
1055 for (i = 0; i < async_extent->nr_pages; i++) {
1056 WARN_ON(async_extent->pages[i]->mapping);
1057 put_page(async_extent->pages[i]);
1058 }
1059 kfree(async_extent->pages);
1060 async_extent->nr_pages = 0;
1061 async_extent->pages = NULL;
1062 }
1063
submit_uncompressed_range(struct btrfs_inode * inode,struct async_extent * async_extent,struct page * locked_page)1064 static void submit_uncompressed_range(struct btrfs_inode *inode,
1065 struct async_extent *async_extent,
1066 struct page *locked_page)
1067 {
1068 u64 start = async_extent->start;
1069 u64 end = async_extent->start + async_extent->ram_size - 1;
1070 int ret;
1071 struct writeback_control wbc = {
1072 .sync_mode = WB_SYNC_ALL,
1073 .range_start = start,
1074 .range_end = end,
1075 .no_cgroup_owner = 1,
1076 };
1077
1078 wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode);
1079 ret = run_delalloc_cow(inode, locked_page, start, end, &wbc, false);
1080 wbc_detach_inode(&wbc);
1081 if (ret < 0) {
1082 btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
1083 if (locked_page) {
1084 const u64 page_start = page_offset(locked_page);
1085
1086 set_page_writeback(locked_page);
1087 end_page_writeback(locked_page);
1088 btrfs_mark_ordered_io_finished(inode, locked_page,
1089 page_start, PAGE_SIZE,
1090 !ret);
1091 mapping_set_error(locked_page->mapping, ret);
1092 unlock_page(locked_page);
1093 }
1094 }
1095 }
1096
submit_one_async_extent(struct async_chunk * async_chunk,struct async_extent * async_extent,u64 * alloc_hint)1097 static void submit_one_async_extent(struct async_chunk *async_chunk,
1098 struct async_extent *async_extent,
1099 u64 *alloc_hint)
1100 {
1101 struct btrfs_inode *inode = async_chunk->inode;
1102 struct extent_io_tree *io_tree = &inode->io_tree;
1103 struct btrfs_root *root = inode->root;
1104 struct btrfs_fs_info *fs_info = root->fs_info;
1105 struct btrfs_ordered_extent *ordered;
1106 struct btrfs_key ins;
1107 struct page *locked_page = NULL;
1108 struct extent_map *em;
1109 int ret = 0;
1110 u64 start = async_extent->start;
1111 u64 end = async_extent->start + async_extent->ram_size - 1;
1112
1113 if (async_chunk->blkcg_css)
1114 kthread_associate_blkcg(async_chunk->blkcg_css);
1115
1116 /*
1117 * If async_chunk->locked_page is in the async_extent range, we need to
1118 * handle it.
1119 */
1120 if (async_chunk->locked_page) {
1121 u64 locked_page_start = page_offset(async_chunk->locked_page);
1122 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
1123
1124 if (!(start >= locked_page_end || end <= locked_page_start))
1125 locked_page = async_chunk->locked_page;
1126 }
1127 lock_extent(io_tree, start, end, NULL);
1128
1129 if (async_extent->compress_type == BTRFS_COMPRESS_NONE) {
1130 submit_uncompressed_range(inode, async_extent, locked_page);
1131 goto done;
1132 }
1133
1134 ret = btrfs_reserve_extent(root, async_extent->ram_size,
1135 async_extent->compressed_size,
1136 async_extent->compressed_size,
1137 0, *alloc_hint, &ins, 1, 1);
1138 if (ret) {
1139 /*
1140 * We can't reserve contiguous space for the compressed size.
1141 * Unlikely, but it's possible that we could have enough
1142 * non-contiguous space for the uncompressed size instead. So
1143 * fall back to uncompressed.
1144 */
1145 submit_uncompressed_range(inode, async_extent, locked_page);
1146 goto done;
1147 }
1148
1149 /* Here we're doing allocation and writeback of the compressed pages */
1150 em = create_io_em(inode, start,
1151 async_extent->ram_size, /* len */
1152 start, /* orig_start */
1153 ins.objectid, /* block_start */
1154 ins.offset, /* block_len */
1155 ins.offset, /* orig_block_len */
1156 async_extent->ram_size, /* ram_bytes */
1157 async_extent->compress_type,
1158 BTRFS_ORDERED_COMPRESSED);
1159 if (IS_ERR(em)) {
1160 ret = PTR_ERR(em);
1161 goto out_free_reserve;
1162 }
1163 free_extent_map(em);
1164
1165 ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */
1166 async_extent->ram_size, /* num_bytes */
1167 async_extent->ram_size, /* ram_bytes */
1168 ins.objectid, /* disk_bytenr */
1169 ins.offset, /* disk_num_bytes */
1170 0, /* offset */
1171 1 << BTRFS_ORDERED_COMPRESSED,
1172 async_extent->compress_type);
1173 if (IS_ERR(ordered)) {
1174 btrfs_drop_extent_map_range(inode, start, end, false);
1175 ret = PTR_ERR(ordered);
1176 goto out_free_reserve;
1177 }
1178 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1179
1180 /* Clear dirty, set writeback and unlock the pages. */
1181 extent_clear_unlock_delalloc(inode, start, end,
1182 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1183 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1184 btrfs_submit_compressed_write(ordered,
1185 async_extent->pages, /* compressed_pages */
1186 async_extent->nr_pages,
1187 async_chunk->write_flags, true);
1188 *alloc_hint = ins.objectid + ins.offset;
1189 done:
1190 if (async_chunk->blkcg_css)
1191 kthread_associate_blkcg(NULL);
1192 kfree(async_extent);
1193 return;
1194
1195 out_free_reserve:
1196 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1197 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1198 mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1199 extent_clear_unlock_delalloc(inode, start, end,
1200 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1201 EXTENT_DELALLOC_NEW |
1202 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1203 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1204 PAGE_END_WRITEBACK);
1205 free_async_extent_pages(async_extent);
1206 if (async_chunk->blkcg_css)
1207 kthread_associate_blkcg(NULL);
1208 btrfs_debug(fs_info,
1209 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1210 root->root_key.objectid, btrfs_ino(inode), start,
1211 async_extent->ram_size, ret);
1212 kfree(async_extent);
1213 }
1214
get_extent_allocation_hint(struct btrfs_inode * inode,u64 start,u64 num_bytes)1215 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1216 u64 num_bytes)
1217 {
1218 struct extent_map_tree *em_tree = &inode->extent_tree;
1219 struct extent_map *em;
1220 u64 alloc_hint = 0;
1221
1222 read_lock(&em_tree->lock);
1223 em = search_extent_mapping(em_tree, start, num_bytes);
1224 if (em) {
1225 /*
1226 * if block start isn't an actual block number then find the
1227 * first block in this inode and use that as a hint. If that
1228 * block is also bogus then just don't worry about it.
1229 */
1230 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1231 free_extent_map(em);
1232 em = search_extent_mapping(em_tree, 0, 0);
1233 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1234 alloc_hint = em->block_start;
1235 if (em)
1236 free_extent_map(em);
1237 } else {
1238 alloc_hint = em->block_start;
1239 free_extent_map(em);
1240 }
1241 }
1242 read_unlock(&em_tree->lock);
1243
1244 return alloc_hint;
1245 }
1246
1247 /*
1248 * when extent_io.c finds a delayed allocation range in the file,
1249 * the call backs end up in this code. The basic idea is to
1250 * allocate extents on disk for the range, and create ordered data structs
1251 * in ram to track those extents.
1252 *
1253 * locked_page is the page that writepage had locked already. We use
1254 * it to make sure we don't do extra locks or unlocks.
1255 *
1256 * When this function fails, it unlocks all pages except @locked_page.
1257 *
1258 * When this function successfully creates an inline extent, it returns 1 and
1259 * unlocks all pages including locked_page and starts I/O on them.
1260 * (In reality inline extents are limited to a single page, so locked_page is
1261 * the only page handled anyway).
1262 *
1263 * When this function succeed and creates a normal extent, the page locking
1264 * status depends on the passed in flags:
1265 *
1266 * - If @keep_locked is set, all pages are kept locked.
1267 * - Else all pages except for @locked_page are unlocked.
1268 *
1269 * When a failure happens in the second or later iteration of the
1270 * while-loop, the ordered extents created in previous iterations are kept
1271 * intact. So, the caller must clean them up by calling
1272 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1273 * example.
1274 */
cow_file_range(struct btrfs_inode * inode,struct page * locked_page,u64 start,u64 end,u64 * done_offset,bool keep_locked,bool no_inline)1275 static noinline int cow_file_range(struct btrfs_inode *inode,
1276 struct page *locked_page, u64 start, u64 end,
1277 u64 *done_offset,
1278 bool keep_locked, bool no_inline)
1279 {
1280 struct btrfs_root *root = inode->root;
1281 struct btrfs_fs_info *fs_info = root->fs_info;
1282 u64 alloc_hint = 0;
1283 u64 orig_start = start;
1284 u64 num_bytes;
1285 unsigned long ram_size;
1286 u64 cur_alloc_size = 0;
1287 u64 min_alloc_size;
1288 u64 blocksize = fs_info->sectorsize;
1289 struct btrfs_key ins;
1290 struct extent_map *em;
1291 unsigned clear_bits;
1292 unsigned long page_ops;
1293 bool extent_reserved = false;
1294 int ret = 0;
1295
1296 if (btrfs_is_free_space_inode(inode)) {
1297 ret = -EINVAL;
1298 goto out_unlock;
1299 }
1300
1301 num_bytes = ALIGN(end - start + 1, blocksize);
1302 num_bytes = max(blocksize, num_bytes);
1303 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1304
1305 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1306
1307 /*
1308 * Due to the page size limit, for subpage we can only trigger the
1309 * writeback for the dirty sectors of page, that means data writeback
1310 * is doing more writeback than what we want.
1311 *
1312 * This is especially unexpected for some call sites like fallocate,
1313 * where we only increase i_size after everything is done.
1314 * This means we can trigger inline extent even if we didn't want to.
1315 * So here we skip inline extent creation completely.
1316 */
1317 if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) {
1318 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1319 end + 1);
1320
1321 /* lets try to make an inline extent */
1322 ret = cow_file_range_inline(inode, actual_end, 0,
1323 BTRFS_COMPRESS_NONE, NULL, false);
1324 if (ret == 0) {
1325 /*
1326 * We use DO_ACCOUNTING here because we need the
1327 * delalloc_release_metadata to be run _after_ we drop
1328 * our outstanding extent for clearing delalloc for this
1329 * range.
1330 */
1331 extent_clear_unlock_delalloc(inode, start, end,
1332 locked_page,
1333 EXTENT_LOCKED | EXTENT_DELALLOC |
1334 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1335 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1336 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1337 /*
1338 * locked_page is locked by the caller of
1339 * writepage_delalloc(), not locked by
1340 * __process_pages_contig().
1341 *
1342 * We can't let __process_pages_contig() to unlock it,
1343 * as it doesn't have any subpage::writers recorded.
1344 *
1345 * Here we manually unlock the page, since the caller
1346 * can't determine if it's an inline extent or a
1347 * compressed extent.
1348 */
1349 unlock_page(locked_page);
1350 ret = 1;
1351 goto done;
1352 } else if (ret < 0) {
1353 goto out_unlock;
1354 }
1355 }
1356
1357 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1358
1359 /*
1360 * Relocation relies on the relocated extents to have exactly the same
1361 * size as the original extents. Normally writeback for relocation data
1362 * extents follows a NOCOW path because relocation preallocates the
1363 * extents. However, due to an operation such as scrub turning a block
1364 * group to RO mode, it may fallback to COW mode, so we must make sure
1365 * an extent allocated during COW has exactly the requested size and can
1366 * not be split into smaller extents, otherwise relocation breaks and
1367 * fails during the stage where it updates the bytenr of file extent
1368 * items.
1369 */
1370 if (btrfs_is_data_reloc_root(root))
1371 min_alloc_size = num_bytes;
1372 else
1373 min_alloc_size = fs_info->sectorsize;
1374
1375 while (num_bytes > 0) {
1376 struct btrfs_ordered_extent *ordered;
1377
1378 cur_alloc_size = num_bytes;
1379 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1380 min_alloc_size, 0, alloc_hint,
1381 &ins, 1, 1);
1382 if (ret == -EAGAIN) {
1383 /*
1384 * btrfs_reserve_extent only returns -EAGAIN for zoned
1385 * file systems, which is an indication that there are
1386 * no active zones to allocate from at the moment.
1387 *
1388 * If this is the first loop iteration, wait for at
1389 * least one zone to finish before retrying the
1390 * allocation. Otherwise ask the caller to write out
1391 * the already allocated blocks before coming back to
1392 * us, or return -ENOSPC if it can't handle retries.
1393 */
1394 ASSERT(btrfs_is_zoned(fs_info));
1395 if (start == orig_start) {
1396 wait_on_bit_io(&inode->root->fs_info->flags,
1397 BTRFS_FS_NEED_ZONE_FINISH,
1398 TASK_UNINTERRUPTIBLE);
1399 continue;
1400 }
1401 if (done_offset) {
1402 *done_offset = start - 1;
1403 return 0;
1404 }
1405 ret = -ENOSPC;
1406 }
1407 if (ret < 0)
1408 goto out_unlock;
1409 cur_alloc_size = ins.offset;
1410 extent_reserved = true;
1411
1412 ram_size = ins.offset;
1413 em = create_io_em(inode, start, ins.offset, /* len */
1414 start, /* orig_start */
1415 ins.objectid, /* block_start */
1416 ins.offset, /* block_len */
1417 ins.offset, /* orig_block_len */
1418 ram_size, /* ram_bytes */
1419 BTRFS_COMPRESS_NONE, /* compress_type */
1420 BTRFS_ORDERED_REGULAR /* type */);
1421 if (IS_ERR(em)) {
1422 ret = PTR_ERR(em);
1423 goto out_reserve;
1424 }
1425 free_extent_map(em);
1426
1427 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1428 ram_size, ins.objectid, cur_alloc_size,
1429 0, 1 << BTRFS_ORDERED_REGULAR,
1430 BTRFS_COMPRESS_NONE);
1431 if (IS_ERR(ordered)) {
1432 ret = PTR_ERR(ordered);
1433 goto out_drop_extent_cache;
1434 }
1435
1436 if (btrfs_is_data_reloc_root(root)) {
1437 ret = btrfs_reloc_clone_csums(ordered);
1438
1439 /*
1440 * Only drop cache here, and process as normal.
1441 *
1442 * We must not allow extent_clear_unlock_delalloc()
1443 * at out_unlock label to free meta of this ordered
1444 * extent, as its meta should be freed by
1445 * btrfs_finish_ordered_io().
1446 *
1447 * So we must continue until @start is increased to
1448 * skip current ordered extent.
1449 */
1450 if (ret)
1451 btrfs_drop_extent_map_range(inode, start,
1452 start + ram_size - 1,
1453 false);
1454 }
1455 btrfs_put_ordered_extent(ordered);
1456
1457 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1458
1459 /*
1460 * We're not doing compressed IO, don't unlock the first page
1461 * (which the caller expects to stay locked), don't clear any
1462 * dirty bits and don't set any writeback bits
1463 *
1464 * Do set the Ordered (Private2) bit so we know this page was
1465 * properly setup for writepage.
1466 */
1467 page_ops = (keep_locked ? 0 : PAGE_UNLOCK);
1468 page_ops |= PAGE_SET_ORDERED;
1469
1470 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1471 locked_page,
1472 EXTENT_LOCKED | EXTENT_DELALLOC,
1473 page_ops);
1474 if (num_bytes < cur_alloc_size)
1475 num_bytes = 0;
1476 else
1477 num_bytes -= cur_alloc_size;
1478 alloc_hint = ins.objectid + ins.offset;
1479 start += cur_alloc_size;
1480 extent_reserved = false;
1481
1482 /*
1483 * btrfs_reloc_clone_csums() error, since start is increased
1484 * extent_clear_unlock_delalloc() at out_unlock label won't
1485 * free metadata of current ordered extent, we're OK to exit.
1486 */
1487 if (ret)
1488 goto out_unlock;
1489 }
1490 done:
1491 if (done_offset)
1492 *done_offset = end;
1493 return ret;
1494
1495 out_drop_extent_cache:
1496 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1497 out_reserve:
1498 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1499 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1500 out_unlock:
1501 /*
1502 * Now, we have three regions to clean up:
1503 *
1504 * |-------(1)----|---(2)---|-------------(3)----------|
1505 * `- orig_start `- start `- start + cur_alloc_size `- end
1506 *
1507 * We process each region below.
1508 */
1509
1510 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1511 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1512 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1513
1514 /*
1515 * For the range (1). We have already instantiated the ordered extents
1516 * for this region. They are cleaned up by
1517 * btrfs_cleanup_ordered_extents() in e.g,
1518 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1519 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1520 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1521 * function.
1522 *
1523 * However, in case of @keep_locked, we still need to unlock the pages
1524 * (except @locked_page) to ensure all the pages are unlocked.
1525 */
1526 if (keep_locked && orig_start < start) {
1527 if (!locked_page)
1528 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1529 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1530 locked_page, 0, page_ops);
1531 }
1532
1533 /*
1534 * For the range (2). If we reserved an extent for our delalloc range
1535 * (or a subrange) and failed to create the respective ordered extent,
1536 * then it means that when we reserved the extent we decremented the
1537 * extent's size from the data space_info's bytes_may_use counter and
1538 * incremented the space_info's bytes_reserved counter by the same
1539 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1540 * to decrement again the data space_info's bytes_may_use counter,
1541 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1542 */
1543 if (extent_reserved) {
1544 extent_clear_unlock_delalloc(inode, start,
1545 start + cur_alloc_size - 1,
1546 locked_page,
1547 clear_bits,
1548 page_ops);
1549 start += cur_alloc_size;
1550 }
1551
1552 /*
1553 * For the range (3). We never touched the region. In addition to the
1554 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1555 * space_info's bytes_may_use counter, reserved in
1556 * btrfs_check_data_free_space().
1557 */
1558 if (start < end) {
1559 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1560 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1561 clear_bits, page_ops);
1562 }
1563 return ret;
1564 }
1565
1566 /*
1567 * Phase two of compressed writeback. This is the ordered portion of the code,
1568 * which only gets called in the order the work was queued. We walk all the
1569 * async extents created by compress_file_range and send them down to the disk.
1570 */
submit_compressed_extents(struct btrfs_work * work)1571 static noinline void submit_compressed_extents(struct btrfs_work *work)
1572 {
1573 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1574 work);
1575 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1576 struct async_extent *async_extent;
1577 unsigned long nr_pages;
1578 u64 alloc_hint = 0;
1579
1580 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1581 PAGE_SHIFT;
1582
1583 while (!list_empty(&async_chunk->extents)) {
1584 async_extent = list_entry(async_chunk->extents.next,
1585 struct async_extent, list);
1586 list_del(&async_extent->list);
1587 submit_one_async_extent(async_chunk, async_extent, &alloc_hint);
1588 }
1589
1590 /* atomic_sub_return implies a barrier */
1591 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1592 5 * SZ_1M)
1593 cond_wake_up_nomb(&fs_info->async_submit_wait);
1594 }
1595
async_cow_free(struct btrfs_work * work)1596 static noinline void async_cow_free(struct btrfs_work *work)
1597 {
1598 struct async_chunk *async_chunk;
1599 struct async_cow *async_cow;
1600
1601 async_chunk = container_of(work, struct async_chunk, work);
1602 btrfs_add_delayed_iput(async_chunk->inode);
1603 if (async_chunk->blkcg_css)
1604 css_put(async_chunk->blkcg_css);
1605
1606 async_cow = async_chunk->async_cow;
1607 if (atomic_dec_and_test(&async_cow->num_chunks))
1608 kvfree(async_cow);
1609 }
1610
run_delalloc_compressed(struct btrfs_inode * inode,struct page * locked_page,u64 start,u64 end,struct writeback_control * wbc)1611 static bool run_delalloc_compressed(struct btrfs_inode *inode,
1612 struct page *locked_page, u64 start,
1613 u64 end, struct writeback_control *wbc)
1614 {
1615 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1616 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1617 struct async_cow *ctx;
1618 struct async_chunk *async_chunk;
1619 unsigned long nr_pages;
1620 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1621 int i;
1622 unsigned nofs_flag;
1623 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1624
1625 nofs_flag = memalloc_nofs_save();
1626 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1627 memalloc_nofs_restore(nofs_flag);
1628 if (!ctx)
1629 return false;
1630
1631 unlock_extent(&inode->io_tree, start, end, NULL);
1632 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1633
1634 async_chunk = ctx->chunks;
1635 atomic_set(&ctx->num_chunks, num_chunks);
1636
1637 for (i = 0; i < num_chunks; i++) {
1638 u64 cur_end = min(end, start + SZ_512K - 1);
1639
1640 /*
1641 * igrab is called higher up in the call chain, take only the
1642 * lightweight reference for the callback lifetime
1643 */
1644 ihold(&inode->vfs_inode);
1645 async_chunk[i].async_cow = ctx;
1646 async_chunk[i].inode = inode;
1647 async_chunk[i].start = start;
1648 async_chunk[i].end = cur_end;
1649 async_chunk[i].write_flags = write_flags;
1650 INIT_LIST_HEAD(&async_chunk[i].extents);
1651
1652 /*
1653 * The locked_page comes all the way from writepage and its
1654 * the original page we were actually given. As we spread
1655 * this large delalloc region across multiple async_chunk
1656 * structs, only the first struct needs a pointer to locked_page
1657 *
1658 * This way we don't need racey decisions about who is supposed
1659 * to unlock it.
1660 */
1661 if (locked_page) {
1662 /*
1663 * Depending on the compressibility, the pages might or
1664 * might not go through async. We want all of them to
1665 * be accounted against wbc once. Let's do it here
1666 * before the paths diverge. wbc accounting is used
1667 * only for foreign writeback detection and doesn't
1668 * need full accuracy. Just account the whole thing
1669 * against the first page.
1670 */
1671 wbc_account_cgroup_owner(wbc, locked_page,
1672 cur_end - start);
1673 async_chunk[i].locked_page = locked_page;
1674 locked_page = NULL;
1675 } else {
1676 async_chunk[i].locked_page = NULL;
1677 }
1678
1679 if (blkcg_css != blkcg_root_css) {
1680 css_get(blkcg_css);
1681 async_chunk[i].blkcg_css = blkcg_css;
1682 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1683 } else {
1684 async_chunk[i].blkcg_css = NULL;
1685 }
1686
1687 btrfs_init_work(&async_chunk[i].work, compress_file_range,
1688 submit_compressed_extents, async_cow_free);
1689
1690 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1691 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1692
1693 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1694
1695 start = cur_end + 1;
1696 }
1697 return true;
1698 }
1699
1700 /*
1701 * Run the delalloc range from start to end, and write back any dirty pages
1702 * covered by the range.
1703 */
run_delalloc_cow(struct btrfs_inode * inode,struct page * locked_page,u64 start,u64 end,struct writeback_control * wbc,bool pages_dirty)1704 static noinline int run_delalloc_cow(struct btrfs_inode *inode,
1705 struct page *locked_page, u64 start,
1706 u64 end, struct writeback_control *wbc,
1707 bool pages_dirty)
1708 {
1709 u64 done_offset = end;
1710 int ret;
1711
1712 while (start <= end) {
1713 ret = cow_file_range(inode, locked_page, start, end, &done_offset,
1714 true, false);
1715 if (ret)
1716 return ret;
1717 extent_write_locked_range(&inode->vfs_inode, locked_page, start,
1718 done_offset, wbc, pages_dirty);
1719 start = done_offset + 1;
1720 }
1721
1722 return 1;
1723 }
1724
csum_exist_in_range(struct btrfs_fs_info * fs_info,u64 bytenr,u64 num_bytes,bool nowait)1725 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1726 u64 bytenr, u64 num_bytes, bool nowait)
1727 {
1728 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1729 struct btrfs_ordered_sum *sums;
1730 int ret;
1731 LIST_HEAD(list);
1732
1733 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1734 &list, 0, nowait);
1735 if (ret == 0 && list_empty(&list))
1736 return 0;
1737
1738 while (!list_empty(&list)) {
1739 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1740 list_del(&sums->list);
1741 kfree(sums);
1742 }
1743 if (ret < 0)
1744 return ret;
1745 return 1;
1746 }
1747
fallback_to_cow(struct btrfs_inode * inode,struct page * locked_page,const u64 start,const u64 end)1748 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1749 const u64 start, const u64 end)
1750 {
1751 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1752 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1753 const u64 range_bytes = end + 1 - start;
1754 struct extent_io_tree *io_tree = &inode->io_tree;
1755 u64 range_start = start;
1756 u64 count;
1757 int ret;
1758
1759 /*
1760 * If EXTENT_NORESERVE is set it means that when the buffered write was
1761 * made we had not enough available data space and therefore we did not
1762 * reserve data space for it, since we though we could do NOCOW for the
1763 * respective file range (either there is prealloc extent or the inode
1764 * has the NOCOW bit set).
1765 *
1766 * However when we need to fallback to COW mode (because for example the
1767 * block group for the corresponding extent was turned to RO mode by a
1768 * scrub or relocation) we need to do the following:
1769 *
1770 * 1) We increment the bytes_may_use counter of the data space info.
1771 * If COW succeeds, it allocates a new data extent and after doing
1772 * that it decrements the space info's bytes_may_use counter and
1773 * increments its bytes_reserved counter by the same amount (we do
1774 * this at btrfs_add_reserved_bytes()). So we need to increment the
1775 * bytes_may_use counter to compensate (when space is reserved at
1776 * buffered write time, the bytes_may_use counter is incremented);
1777 *
1778 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1779 * that if the COW path fails for any reason, it decrements (through
1780 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1781 * data space info, which we incremented in the step above.
1782 *
1783 * If we need to fallback to cow and the inode corresponds to a free
1784 * space cache inode or an inode of the data relocation tree, we must
1785 * also increment bytes_may_use of the data space_info for the same
1786 * reason. Space caches and relocated data extents always get a prealloc
1787 * extent for them, however scrub or balance may have set the block
1788 * group that contains that extent to RO mode and therefore force COW
1789 * when starting writeback.
1790 */
1791 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1792 EXTENT_NORESERVE, 0, NULL);
1793 if (count > 0 || is_space_ino || is_reloc_ino) {
1794 u64 bytes = count;
1795 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1796 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1797
1798 if (is_space_ino || is_reloc_ino)
1799 bytes = range_bytes;
1800
1801 spin_lock(&sinfo->lock);
1802 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1803 spin_unlock(&sinfo->lock);
1804
1805 if (count > 0)
1806 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1807 NULL);
1808 }
1809
1810 /*
1811 * Don't try to create inline extents, as a mix of inline extent that
1812 * is written out and unlocked directly and a normal NOCOW extent
1813 * doesn't work.
1814 */
1815 ret = cow_file_range(inode, locked_page, start, end, NULL, false, true);
1816 ASSERT(ret != 1);
1817 return ret;
1818 }
1819
1820 struct can_nocow_file_extent_args {
1821 /* Input fields. */
1822
1823 /* Start file offset of the range we want to NOCOW. */
1824 u64 start;
1825 /* End file offset (inclusive) of the range we want to NOCOW. */
1826 u64 end;
1827 bool writeback_path;
1828 bool strict;
1829 /*
1830 * Free the path passed to can_nocow_file_extent() once it's not needed
1831 * anymore.
1832 */
1833 bool free_path;
1834
1835 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1836
1837 u64 disk_bytenr;
1838 u64 disk_num_bytes;
1839 u64 extent_offset;
1840 /* Number of bytes that can be written to in NOCOW mode. */
1841 u64 num_bytes;
1842 };
1843
1844 /*
1845 * Check if we can NOCOW the file extent that the path points to.
1846 * This function may return with the path released, so the caller should check
1847 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1848 *
1849 * Returns: < 0 on error
1850 * 0 if we can not NOCOW
1851 * 1 if we can NOCOW
1852 */
can_nocow_file_extent(struct btrfs_path * path,struct btrfs_key * key,struct btrfs_inode * inode,struct can_nocow_file_extent_args * args)1853 static int can_nocow_file_extent(struct btrfs_path *path,
1854 struct btrfs_key *key,
1855 struct btrfs_inode *inode,
1856 struct can_nocow_file_extent_args *args)
1857 {
1858 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1859 struct extent_buffer *leaf = path->nodes[0];
1860 struct btrfs_root *root = inode->root;
1861 struct btrfs_file_extent_item *fi;
1862 u64 extent_end;
1863 u8 extent_type;
1864 int can_nocow = 0;
1865 int ret = 0;
1866 bool nowait = path->nowait;
1867
1868 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1869 extent_type = btrfs_file_extent_type(leaf, fi);
1870
1871 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1872 goto out;
1873
1874 /* Can't access these fields unless we know it's not an inline extent. */
1875 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1876 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1877 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1878
1879 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1880 extent_type == BTRFS_FILE_EXTENT_REG)
1881 goto out;
1882
1883 /*
1884 * If the extent was created before the generation where the last snapshot
1885 * for its subvolume was created, then this implies the extent is shared,
1886 * hence we must COW.
1887 */
1888 if (!args->strict &&
1889 btrfs_file_extent_generation(leaf, fi) <=
1890 btrfs_root_last_snapshot(&root->root_item))
1891 goto out;
1892
1893 /* An explicit hole, must COW. */
1894 if (args->disk_bytenr == 0)
1895 goto out;
1896
1897 /* Compressed/encrypted/encoded extents must be COWed. */
1898 if (btrfs_file_extent_compression(leaf, fi) ||
1899 btrfs_file_extent_encryption(leaf, fi) ||
1900 btrfs_file_extent_other_encoding(leaf, fi))
1901 goto out;
1902
1903 extent_end = btrfs_file_extent_end(path);
1904
1905 /*
1906 * The following checks can be expensive, as they need to take other
1907 * locks and do btree or rbtree searches, so release the path to avoid
1908 * blocking other tasks for too long.
1909 */
1910 btrfs_release_path(path);
1911
1912 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1913 key->offset - args->extent_offset,
1914 args->disk_bytenr, args->strict, path);
1915 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1916 if (ret != 0)
1917 goto out;
1918
1919 if (args->free_path) {
1920 /*
1921 * We don't need the path anymore, plus through the
1922 * csum_exist_in_range() call below we will end up allocating
1923 * another path. So free the path to avoid unnecessary extra
1924 * memory usage.
1925 */
1926 btrfs_free_path(path);
1927 path = NULL;
1928 }
1929
1930 /* If there are pending snapshots for this root, we must COW. */
1931 if (args->writeback_path && !is_freespace_inode &&
1932 atomic_read(&root->snapshot_force_cow))
1933 goto out;
1934
1935 args->disk_bytenr += args->extent_offset;
1936 args->disk_bytenr += args->start - key->offset;
1937 args->num_bytes = min(args->end + 1, extent_end) - args->start;
1938
1939 /*
1940 * Force COW if csums exist in the range. This ensures that csums for a
1941 * given extent are either valid or do not exist.
1942 */
1943 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
1944 nowait);
1945 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1946 if (ret != 0)
1947 goto out;
1948
1949 can_nocow = 1;
1950 out:
1951 if (args->free_path && path)
1952 btrfs_free_path(path);
1953
1954 return ret < 0 ? ret : can_nocow;
1955 }
1956
1957 /*
1958 * when nowcow writeback call back. This checks for snapshots or COW copies
1959 * of the extents that exist in the file, and COWs the file as required.
1960 *
1961 * If no cow copies or snapshots exist, we write directly to the existing
1962 * blocks on disk
1963 */
run_delalloc_nocow(struct btrfs_inode * inode,struct page * locked_page,const u64 start,const u64 end)1964 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1965 struct page *locked_page,
1966 const u64 start, const u64 end)
1967 {
1968 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1969 struct btrfs_root *root = inode->root;
1970 struct btrfs_path *path;
1971 u64 cow_start = (u64)-1;
1972 u64 cur_offset = start;
1973 int ret;
1974 bool check_prev = true;
1975 u64 ino = btrfs_ino(inode);
1976 struct can_nocow_file_extent_args nocow_args = { 0 };
1977
1978 /*
1979 * Normally on a zoned device we're only doing COW writes, but in case
1980 * of relocation on a zoned filesystem serializes I/O so that we're only
1981 * writing sequentially and can end up here as well.
1982 */
1983 ASSERT(!btrfs_is_zoned(fs_info) || btrfs_is_data_reloc_root(root));
1984
1985 path = btrfs_alloc_path();
1986 if (!path) {
1987 ret = -ENOMEM;
1988 goto error;
1989 }
1990
1991 nocow_args.end = end;
1992 nocow_args.writeback_path = true;
1993
1994 while (1) {
1995 struct btrfs_block_group *nocow_bg = NULL;
1996 struct btrfs_ordered_extent *ordered;
1997 struct btrfs_key found_key;
1998 struct btrfs_file_extent_item *fi;
1999 struct extent_buffer *leaf;
2000 u64 extent_end;
2001 u64 ram_bytes;
2002 u64 nocow_end;
2003 int extent_type;
2004 bool is_prealloc;
2005
2006 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2007 cur_offset, 0);
2008 if (ret < 0)
2009 goto error;
2010
2011 /*
2012 * If there is no extent for our range when doing the initial
2013 * search, then go back to the previous slot as it will be the
2014 * one containing the search offset
2015 */
2016 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2017 leaf = path->nodes[0];
2018 btrfs_item_key_to_cpu(leaf, &found_key,
2019 path->slots[0] - 1);
2020 if (found_key.objectid == ino &&
2021 found_key.type == BTRFS_EXTENT_DATA_KEY)
2022 path->slots[0]--;
2023 }
2024 check_prev = false;
2025 next_slot:
2026 /* Go to next leaf if we have exhausted the current one */
2027 leaf = path->nodes[0];
2028 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2029 ret = btrfs_next_leaf(root, path);
2030 if (ret < 0)
2031 goto error;
2032 if (ret > 0)
2033 break;
2034 leaf = path->nodes[0];
2035 }
2036
2037 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2038
2039 /* Didn't find anything for our INO */
2040 if (found_key.objectid > ino)
2041 break;
2042 /*
2043 * Keep searching until we find an EXTENT_ITEM or there are no
2044 * more extents for this inode
2045 */
2046 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2047 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2048 path->slots[0]++;
2049 goto next_slot;
2050 }
2051
2052 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2053 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2054 found_key.offset > end)
2055 break;
2056
2057 /*
2058 * If the found extent starts after requested offset, then
2059 * adjust extent_end to be right before this extent begins
2060 */
2061 if (found_key.offset > cur_offset) {
2062 extent_end = found_key.offset;
2063 extent_type = 0;
2064 goto must_cow;
2065 }
2066
2067 /*
2068 * Found extent which begins before our range and potentially
2069 * intersect it
2070 */
2071 fi = btrfs_item_ptr(leaf, path->slots[0],
2072 struct btrfs_file_extent_item);
2073 extent_type = btrfs_file_extent_type(leaf, fi);
2074 /* If this is triggered then we have a memory corruption. */
2075 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2076 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2077 ret = -EUCLEAN;
2078 goto error;
2079 }
2080 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2081 extent_end = btrfs_file_extent_end(path);
2082
2083 /*
2084 * If the extent we got ends before our current offset, skip to
2085 * the next extent.
2086 */
2087 if (extent_end <= cur_offset) {
2088 path->slots[0]++;
2089 goto next_slot;
2090 }
2091
2092 nocow_args.start = cur_offset;
2093 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2094 if (ret < 0)
2095 goto error;
2096 if (ret == 0)
2097 goto must_cow;
2098
2099 ret = 0;
2100 nocow_bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2101 if (!nocow_bg) {
2102 must_cow:
2103 /*
2104 * If we can't perform NOCOW writeback for the range,
2105 * then record the beginning of the range that needs to
2106 * be COWed. It will be written out before the next
2107 * NOCOW range if we find one, or when exiting this
2108 * loop.
2109 */
2110 if (cow_start == (u64)-1)
2111 cow_start = cur_offset;
2112 cur_offset = extent_end;
2113 if (cur_offset > end)
2114 break;
2115 if (!path->nodes[0])
2116 continue;
2117 path->slots[0]++;
2118 goto next_slot;
2119 }
2120
2121 /*
2122 * COW range from cow_start to found_key.offset - 1. As the key
2123 * will contain the beginning of the first extent that can be
2124 * NOCOW, following one which needs to be COW'ed
2125 */
2126 if (cow_start != (u64)-1) {
2127 ret = fallback_to_cow(inode, locked_page,
2128 cow_start, found_key.offset - 1);
2129 cow_start = (u64)-1;
2130 if (ret) {
2131 btrfs_dec_nocow_writers(nocow_bg);
2132 goto error;
2133 }
2134 }
2135
2136 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2137 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2138 if (is_prealloc) {
2139 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2140 struct extent_map *em;
2141
2142 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2143 orig_start,
2144 nocow_args.disk_bytenr, /* block_start */
2145 nocow_args.num_bytes, /* block_len */
2146 nocow_args.disk_num_bytes, /* orig_block_len */
2147 ram_bytes, BTRFS_COMPRESS_NONE,
2148 BTRFS_ORDERED_PREALLOC);
2149 if (IS_ERR(em)) {
2150 btrfs_dec_nocow_writers(nocow_bg);
2151 ret = PTR_ERR(em);
2152 goto error;
2153 }
2154 free_extent_map(em);
2155 }
2156
2157 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2158 nocow_args.num_bytes, nocow_args.num_bytes,
2159 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2160 is_prealloc
2161 ? (1 << BTRFS_ORDERED_PREALLOC)
2162 : (1 << BTRFS_ORDERED_NOCOW),
2163 BTRFS_COMPRESS_NONE);
2164 btrfs_dec_nocow_writers(nocow_bg);
2165 if (IS_ERR(ordered)) {
2166 if (is_prealloc) {
2167 btrfs_drop_extent_map_range(inode, cur_offset,
2168 nocow_end, false);
2169 }
2170 ret = PTR_ERR(ordered);
2171 goto error;
2172 }
2173
2174 if (btrfs_is_data_reloc_root(root))
2175 /*
2176 * Error handled later, as we must prevent
2177 * extent_clear_unlock_delalloc() in error handler
2178 * from freeing metadata of created ordered extent.
2179 */
2180 ret = btrfs_reloc_clone_csums(ordered);
2181 btrfs_put_ordered_extent(ordered);
2182
2183 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2184 locked_page, EXTENT_LOCKED |
2185 EXTENT_DELALLOC |
2186 EXTENT_CLEAR_DATA_RESV,
2187 PAGE_UNLOCK | PAGE_SET_ORDERED);
2188
2189 cur_offset = extent_end;
2190
2191 /*
2192 * btrfs_reloc_clone_csums() error, now we're OK to call error
2193 * handler, as metadata for created ordered extent will only
2194 * be freed by btrfs_finish_ordered_io().
2195 */
2196 if (ret)
2197 goto error;
2198 if (cur_offset > end)
2199 break;
2200 }
2201 btrfs_release_path(path);
2202
2203 if (cur_offset <= end && cow_start == (u64)-1)
2204 cow_start = cur_offset;
2205
2206 if (cow_start != (u64)-1) {
2207 cur_offset = end;
2208 ret = fallback_to_cow(inode, locked_page, cow_start, end);
2209 cow_start = (u64)-1;
2210 if (ret)
2211 goto error;
2212 }
2213
2214 btrfs_free_path(path);
2215 return 0;
2216
2217 error:
2218 /*
2219 * If an error happened while a COW region is outstanding, cur_offset
2220 * needs to be reset to cow_start to ensure the COW region is unlocked
2221 * as well.
2222 */
2223 if (cow_start != (u64)-1)
2224 cur_offset = cow_start;
2225 if (cur_offset < end)
2226 extent_clear_unlock_delalloc(inode, cur_offset, end,
2227 locked_page, EXTENT_LOCKED |
2228 EXTENT_DELALLOC | EXTENT_DEFRAG |
2229 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2230 PAGE_START_WRITEBACK |
2231 PAGE_END_WRITEBACK);
2232 btrfs_free_path(path);
2233 return ret;
2234 }
2235
should_nocow(struct btrfs_inode * inode,u64 start,u64 end)2236 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2237 {
2238 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2239 if (inode->defrag_bytes &&
2240 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2241 0, NULL))
2242 return false;
2243 return true;
2244 }
2245 return false;
2246 }
2247
2248 /*
2249 * Function to process delayed allocation (create CoW) for ranges which are
2250 * being touched for the first time.
2251 */
btrfs_run_delalloc_range(struct btrfs_inode * inode,struct page * locked_page,u64 start,u64 end,struct writeback_control * wbc)2252 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2253 u64 start, u64 end, struct writeback_control *wbc)
2254 {
2255 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2256 int ret;
2257
2258 /*
2259 * The range must cover part of the @locked_page, or a return of 1
2260 * can confuse the caller.
2261 */
2262 ASSERT(!(end <= page_offset(locked_page) ||
2263 start >= page_offset(locked_page) + PAGE_SIZE));
2264
2265 if (should_nocow(inode, start, end)) {
2266 ret = run_delalloc_nocow(inode, locked_page, start, end);
2267 goto out;
2268 }
2269
2270 if (btrfs_inode_can_compress(inode) &&
2271 inode_need_compress(inode, start, end) &&
2272 run_delalloc_compressed(inode, locked_page, start, end, wbc))
2273 return 1;
2274
2275 if (zoned)
2276 ret = run_delalloc_cow(inode, locked_page, start, end, wbc,
2277 true);
2278 else
2279 ret = cow_file_range(inode, locked_page, start, end, NULL,
2280 false, false);
2281
2282 out:
2283 if (ret < 0)
2284 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2285 end - start + 1);
2286 return ret;
2287 }
2288
btrfs_split_delalloc_extent(struct btrfs_inode * inode,struct extent_state * orig,u64 split)2289 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2290 struct extent_state *orig, u64 split)
2291 {
2292 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2293 u64 size;
2294
2295 /* not delalloc, ignore it */
2296 if (!(orig->state & EXTENT_DELALLOC))
2297 return;
2298
2299 size = orig->end - orig->start + 1;
2300 if (size > fs_info->max_extent_size) {
2301 u32 num_extents;
2302 u64 new_size;
2303
2304 /*
2305 * See the explanation in btrfs_merge_delalloc_extent, the same
2306 * applies here, just in reverse.
2307 */
2308 new_size = orig->end - split + 1;
2309 num_extents = count_max_extents(fs_info, new_size);
2310 new_size = split - orig->start;
2311 num_extents += count_max_extents(fs_info, new_size);
2312 if (count_max_extents(fs_info, size) >= num_extents)
2313 return;
2314 }
2315
2316 spin_lock(&inode->lock);
2317 btrfs_mod_outstanding_extents(inode, 1);
2318 spin_unlock(&inode->lock);
2319 }
2320
2321 /*
2322 * Handle merged delayed allocation extents so we can keep track of new extents
2323 * that are just merged onto old extents, such as when we are doing sequential
2324 * writes, so we can properly account for the metadata space we'll need.
2325 */
btrfs_merge_delalloc_extent(struct btrfs_inode * inode,struct extent_state * new,struct extent_state * other)2326 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2327 struct extent_state *other)
2328 {
2329 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2330 u64 new_size, old_size;
2331 u32 num_extents;
2332
2333 /* not delalloc, ignore it */
2334 if (!(other->state & EXTENT_DELALLOC))
2335 return;
2336
2337 if (new->start > other->start)
2338 new_size = new->end - other->start + 1;
2339 else
2340 new_size = other->end - new->start + 1;
2341
2342 /* we're not bigger than the max, unreserve the space and go */
2343 if (new_size <= fs_info->max_extent_size) {
2344 spin_lock(&inode->lock);
2345 btrfs_mod_outstanding_extents(inode, -1);
2346 spin_unlock(&inode->lock);
2347 return;
2348 }
2349
2350 /*
2351 * We have to add up either side to figure out how many extents were
2352 * accounted for before we merged into one big extent. If the number of
2353 * extents we accounted for is <= the amount we need for the new range
2354 * then we can return, otherwise drop. Think of it like this
2355 *
2356 * [ 4k][MAX_SIZE]
2357 *
2358 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2359 * need 2 outstanding extents, on one side we have 1 and the other side
2360 * we have 1 so they are == and we can return. But in this case
2361 *
2362 * [MAX_SIZE+4k][MAX_SIZE+4k]
2363 *
2364 * Each range on their own accounts for 2 extents, but merged together
2365 * they are only 3 extents worth of accounting, so we need to drop in
2366 * this case.
2367 */
2368 old_size = other->end - other->start + 1;
2369 num_extents = count_max_extents(fs_info, old_size);
2370 old_size = new->end - new->start + 1;
2371 num_extents += count_max_extents(fs_info, old_size);
2372 if (count_max_extents(fs_info, new_size) >= num_extents)
2373 return;
2374
2375 spin_lock(&inode->lock);
2376 btrfs_mod_outstanding_extents(inode, -1);
2377 spin_unlock(&inode->lock);
2378 }
2379
btrfs_add_delalloc_inodes(struct btrfs_root * root,struct btrfs_inode * inode)2380 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2381 struct btrfs_inode *inode)
2382 {
2383 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2384
2385 spin_lock(&root->delalloc_lock);
2386 if (list_empty(&inode->delalloc_inodes)) {
2387 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2388 set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags);
2389 root->nr_delalloc_inodes++;
2390 if (root->nr_delalloc_inodes == 1) {
2391 spin_lock(&fs_info->delalloc_root_lock);
2392 BUG_ON(!list_empty(&root->delalloc_root));
2393 list_add_tail(&root->delalloc_root,
2394 &fs_info->delalloc_roots);
2395 spin_unlock(&fs_info->delalloc_root_lock);
2396 }
2397 }
2398 spin_unlock(&root->delalloc_lock);
2399 }
2400
__btrfs_del_delalloc_inode(struct btrfs_root * root,struct btrfs_inode * inode)2401 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2402 struct btrfs_inode *inode)
2403 {
2404 struct btrfs_fs_info *fs_info = root->fs_info;
2405
2406 if (!list_empty(&inode->delalloc_inodes)) {
2407 list_del_init(&inode->delalloc_inodes);
2408 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2409 &inode->runtime_flags);
2410 root->nr_delalloc_inodes--;
2411 if (!root->nr_delalloc_inodes) {
2412 ASSERT(list_empty(&root->delalloc_inodes));
2413 spin_lock(&fs_info->delalloc_root_lock);
2414 BUG_ON(list_empty(&root->delalloc_root));
2415 list_del_init(&root->delalloc_root);
2416 spin_unlock(&fs_info->delalloc_root_lock);
2417 }
2418 }
2419 }
2420
btrfs_del_delalloc_inode(struct btrfs_root * root,struct btrfs_inode * inode)2421 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2422 struct btrfs_inode *inode)
2423 {
2424 spin_lock(&root->delalloc_lock);
2425 __btrfs_del_delalloc_inode(root, inode);
2426 spin_unlock(&root->delalloc_lock);
2427 }
2428
2429 /*
2430 * Properly track delayed allocation bytes in the inode and to maintain the
2431 * list of inodes that have pending delalloc work to be done.
2432 */
btrfs_set_delalloc_extent(struct btrfs_inode * inode,struct extent_state * state,u32 bits)2433 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2434 u32 bits)
2435 {
2436 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2437
2438 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2439 WARN_ON(1);
2440 /*
2441 * set_bit and clear bit hooks normally require _irqsave/restore
2442 * but in this case, we are only testing for the DELALLOC
2443 * bit, which is only set or cleared with irqs on
2444 */
2445 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2446 struct btrfs_root *root = inode->root;
2447 u64 len = state->end + 1 - state->start;
2448 u32 num_extents = count_max_extents(fs_info, len);
2449 bool do_list = !btrfs_is_free_space_inode(inode);
2450
2451 spin_lock(&inode->lock);
2452 btrfs_mod_outstanding_extents(inode, num_extents);
2453 spin_unlock(&inode->lock);
2454
2455 /* For sanity tests */
2456 if (btrfs_is_testing(fs_info))
2457 return;
2458
2459 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2460 fs_info->delalloc_batch);
2461 spin_lock(&inode->lock);
2462 inode->delalloc_bytes += len;
2463 if (bits & EXTENT_DEFRAG)
2464 inode->defrag_bytes += len;
2465 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2466 &inode->runtime_flags))
2467 btrfs_add_delalloc_inodes(root, inode);
2468 spin_unlock(&inode->lock);
2469 }
2470
2471 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2472 (bits & EXTENT_DELALLOC_NEW)) {
2473 spin_lock(&inode->lock);
2474 inode->new_delalloc_bytes += state->end + 1 - state->start;
2475 spin_unlock(&inode->lock);
2476 }
2477 }
2478
2479 /*
2480 * Once a range is no longer delalloc this function ensures that proper
2481 * accounting happens.
2482 */
btrfs_clear_delalloc_extent(struct btrfs_inode * inode,struct extent_state * state,u32 bits)2483 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2484 struct extent_state *state, u32 bits)
2485 {
2486 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2487 u64 len = state->end + 1 - state->start;
2488 u32 num_extents = count_max_extents(fs_info, len);
2489
2490 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2491 spin_lock(&inode->lock);
2492 inode->defrag_bytes -= len;
2493 spin_unlock(&inode->lock);
2494 }
2495
2496 /*
2497 * set_bit and clear bit hooks normally require _irqsave/restore
2498 * but in this case, we are only testing for the DELALLOC
2499 * bit, which is only set or cleared with irqs on
2500 */
2501 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2502 struct btrfs_root *root = inode->root;
2503 bool do_list = !btrfs_is_free_space_inode(inode);
2504
2505 spin_lock(&inode->lock);
2506 btrfs_mod_outstanding_extents(inode, -num_extents);
2507 spin_unlock(&inode->lock);
2508
2509 /*
2510 * We don't reserve metadata space for space cache inodes so we
2511 * don't need to call delalloc_release_metadata if there is an
2512 * error.
2513 */
2514 if (bits & EXTENT_CLEAR_META_RESV &&
2515 root != fs_info->tree_root)
2516 btrfs_delalloc_release_metadata(inode, len, true);
2517
2518 /* For sanity tests. */
2519 if (btrfs_is_testing(fs_info))
2520 return;
2521
2522 if (!btrfs_is_data_reloc_root(root) &&
2523 do_list && !(state->state & EXTENT_NORESERVE) &&
2524 (bits & EXTENT_CLEAR_DATA_RESV))
2525 btrfs_free_reserved_data_space_noquota(fs_info, len);
2526
2527 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2528 fs_info->delalloc_batch);
2529 spin_lock(&inode->lock);
2530 inode->delalloc_bytes -= len;
2531 if (do_list && inode->delalloc_bytes == 0 &&
2532 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2533 &inode->runtime_flags))
2534 btrfs_del_delalloc_inode(root, inode);
2535 spin_unlock(&inode->lock);
2536 }
2537
2538 if ((state->state & EXTENT_DELALLOC_NEW) &&
2539 (bits & EXTENT_DELALLOC_NEW)) {
2540 spin_lock(&inode->lock);
2541 ASSERT(inode->new_delalloc_bytes >= len);
2542 inode->new_delalloc_bytes -= len;
2543 if (bits & EXTENT_ADD_INODE_BYTES)
2544 inode_add_bytes(&inode->vfs_inode, len);
2545 spin_unlock(&inode->lock);
2546 }
2547 }
2548
btrfs_extract_ordered_extent(struct btrfs_bio * bbio,struct btrfs_ordered_extent * ordered)2549 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2550 struct btrfs_ordered_extent *ordered)
2551 {
2552 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2553 u64 len = bbio->bio.bi_iter.bi_size;
2554 struct btrfs_ordered_extent *new;
2555 int ret;
2556
2557 /* Must always be called for the beginning of an ordered extent. */
2558 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2559 return -EINVAL;
2560
2561 /* No need to split if the ordered extent covers the entire bio. */
2562 if (ordered->disk_num_bytes == len) {
2563 refcount_inc(&ordered->refs);
2564 bbio->ordered = ordered;
2565 return 0;
2566 }
2567
2568 /*
2569 * Don't split the extent_map for NOCOW extents, as we're writing into
2570 * a pre-existing one.
2571 */
2572 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2573 ret = split_extent_map(bbio->inode, bbio->file_offset,
2574 ordered->num_bytes, len,
2575 ordered->disk_bytenr);
2576 if (ret)
2577 return ret;
2578 }
2579
2580 new = btrfs_split_ordered_extent(ordered, len);
2581 if (IS_ERR(new))
2582 return PTR_ERR(new);
2583 bbio->ordered = new;
2584 return 0;
2585 }
2586
2587 /*
2588 * given a list of ordered sums record them in the inode. This happens
2589 * at IO completion time based on sums calculated at bio submission time.
2590 */
add_pending_csums(struct btrfs_trans_handle * trans,struct list_head * list)2591 static int add_pending_csums(struct btrfs_trans_handle *trans,
2592 struct list_head *list)
2593 {
2594 struct btrfs_ordered_sum *sum;
2595 struct btrfs_root *csum_root = NULL;
2596 int ret;
2597
2598 list_for_each_entry(sum, list, list) {
2599 trans->adding_csums = true;
2600 if (!csum_root)
2601 csum_root = btrfs_csum_root(trans->fs_info,
2602 sum->logical);
2603 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2604 trans->adding_csums = false;
2605 if (ret)
2606 return ret;
2607 }
2608 return 0;
2609 }
2610
btrfs_find_new_delalloc_bytes(struct btrfs_inode * inode,const u64 start,const u64 len,struct extent_state ** cached_state)2611 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2612 const u64 start,
2613 const u64 len,
2614 struct extent_state **cached_state)
2615 {
2616 u64 search_start = start;
2617 const u64 end = start + len - 1;
2618
2619 while (search_start < end) {
2620 const u64 search_len = end - search_start + 1;
2621 struct extent_map *em;
2622 u64 em_len;
2623 int ret = 0;
2624
2625 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2626 if (IS_ERR(em))
2627 return PTR_ERR(em);
2628
2629 if (em->block_start != EXTENT_MAP_HOLE)
2630 goto next;
2631
2632 em_len = em->len;
2633 if (em->start < search_start)
2634 em_len -= search_start - em->start;
2635 if (em_len > search_len)
2636 em_len = search_len;
2637
2638 ret = set_extent_bit(&inode->io_tree, search_start,
2639 search_start + em_len - 1,
2640 EXTENT_DELALLOC_NEW, cached_state);
2641 next:
2642 search_start = extent_map_end(em);
2643 free_extent_map(em);
2644 if (ret)
2645 return ret;
2646 }
2647 return 0;
2648 }
2649
btrfs_set_extent_delalloc(struct btrfs_inode * inode,u64 start,u64 end,unsigned int extra_bits,struct extent_state ** cached_state)2650 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2651 unsigned int extra_bits,
2652 struct extent_state **cached_state)
2653 {
2654 WARN_ON(PAGE_ALIGNED(end));
2655
2656 if (start >= i_size_read(&inode->vfs_inode) &&
2657 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2658 /*
2659 * There can't be any extents following eof in this case so just
2660 * set the delalloc new bit for the range directly.
2661 */
2662 extra_bits |= EXTENT_DELALLOC_NEW;
2663 } else {
2664 int ret;
2665
2666 ret = btrfs_find_new_delalloc_bytes(inode, start,
2667 end + 1 - start,
2668 cached_state);
2669 if (ret)
2670 return ret;
2671 }
2672
2673 return set_extent_bit(&inode->io_tree, start, end,
2674 EXTENT_DELALLOC | extra_bits, cached_state);
2675 }
2676
2677 /* see btrfs_writepage_start_hook for details on why this is required */
2678 struct btrfs_writepage_fixup {
2679 struct page *page;
2680 struct btrfs_inode *inode;
2681 struct btrfs_work work;
2682 };
2683
btrfs_writepage_fixup_worker(struct btrfs_work * work)2684 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2685 {
2686 struct btrfs_writepage_fixup *fixup =
2687 container_of(work, struct btrfs_writepage_fixup, work);
2688 struct btrfs_ordered_extent *ordered;
2689 struct extent_state *cached_state = NULL;
2690 struct extent_changeset *data_reserved = NULL;
2691 struct page *page = fixup->page;
2692 struct btrfs_inode *inode = fixup->inode;
2693 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2694 u64 page_start = page_offset(page);
2695 u64 page_end = page_offset(page) + PAGE_SIZE - 1;
2696 int ret = 0;
2697 bool free_delalloc_space = true;
2698
2699 /*
2700 * This is similar to page_mkwrite, we need to reserve the space before
2701 * we take the page lock.
2702 */
2703 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2704 PAGE_SIZE);
2705 again:
2706 lock_page(page);
2707
2708 /*
2709 * Before we queued this fixup, we took a reference on the page.
2710 * page->mapping may go NULL, but it shouldn't be moved to a different
2711 * address space.
2712 */
2713 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2714 /*
2715 * Unfortunately this is a little tricky, either
2716 *
2717 * 1) We got here and our page had already been dealt with and
2718 * we reserved our space, thus ret == 0, so we need to just
2719 * drop our space reservation and bail. This can happen the
2720 * first time we come into the fixup worker, or could happen
2721 * while waiting for the ordered extent.
2722 * 2) Our page was already dealt with, but we happened to get an
2723 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2724 * this case we obviously don't have anything to release, but
2725 * because the page was already dealt with we don't want to
2726 * mark the page with an error, so make sure we're resetting
2727 * ret to 0. This is why we have this check _before_ the ret
2728 * check, because we do not want to have a surprise ENOSPC
2729 * when the page was already properly dealt with.
2730 */
2731 if (!ret) {
2732 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2733 btrfs_delalloc_release_space(inode, data_reserved,
2734 page_start, PAGE_SIZE,
2735 true);
2736 }
2737 ret = 0;
2738 goto out_page;
2739 }
2740
2741 /*
2742 * We can't mess with the page state unless it is locked, so now that
2743 * it is locked bail if we failed to make our space reservation.
2744 */
2745 if (ret)
2746 goto out_page;
2747
2748 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2749
2750 /* already ordered? We're done */
2751 if (PageOrdered(page))
2752 goto out_reserved;
2753
2754 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2755 if (ordered) {
2756 unlock_extent(&inode->io_tree, page_start, page_end,
2757 &cached_state);
2758 unlock_page(page);
2759 btrfs_start_ordered_extent(ordered);
2760 btrfs_put_ordered_extent(ordered);
2761 goto again;
2762 }
2763
2764 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2765 &cached_state);
2766 if (ret)
2767 goto out_reserved;
2768
2769 /*
2770 * Everything went as planned, we're now the owner of a dirty page with
2771 * delayed allocation bits set and space reserved for our COW
2772 * destination.
2773 *
2774 * The page was dirty when we started, nothing should have cleaned it.
2775 */
2776 BUG_ON(!PageDirty(page));
2777 free_delalloc_space = false;
2778 out_reserved:
2779 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2780 if (free_delalloc_space)
2781 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2782 PAGE_SIZE, true);
2783 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2784 out_page:
2785 if (ret) {
2786 /*
2787 * We hit ENOSPC or other errors. Update the mapping and page
2788 * to reflect the errors and clean the page.
2789 */
2790 mapping_set_error(page->mapping, ret);
2791 btrfs_mark_ordered_io_finished(inode, page, page_start,
2792 PAGE_SIZE, !ret);
2793 clear_page_dirty_for_io(page);
2794 }
2795 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
2796 unlock_page(page);
2797 put_page(page);
2798 kfree(fixup);
2799 extent_changeset_free(data_reserved);
2800 /*
2801 * As a precaution, do a delayed iput in case it would be the last iput
2802 * that could need flushing space. Recursing back to fixup worker would
2803 * deadlock.
2804 */
2805 btrfs_add_delayed_iput(inode);
2806 }
2807
2808 /*
2809 * There are a few paths in the higher layers of the kernel that directly
2810 * set the page dirty bit without asking the filesystem if it is a
2811 * good idea. This causes problems because we want to make sure COW
2812 * properly happens and the data=ordered rules are followed.
2813 *
2814 * In our case any range that doesn't have the ORDERED bit set
2815 * hasn't been properly setup for IO. We kick off an async process
2816 * to fix it up. The async helper will wait for ordered extents, set
2817 * the delalloc bit and make it safe to write the page.
2818 */
btrfs_writepage_cow_fixup(struct page * page)2819 int btrfs_writepage_cow_fixup(struct page *page)
2820 {
2821 struct inode *inode = page->mapping->host;
2822 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2823 struct btrfs_writepage_fixup *fixup;
2824
2825 /* This page has ordered extent covering it already */
2826 if (PageOrdered(page))
2827 return 0;
2828
2829 /*
2830 * PageChecked is set below when we create a fixup worker for this page,
2831 * don't try to create another one if we're already PageChecked()
2832 *
2833 * The extent_io writepage code will redirty the page if we send back
2834 * EAGAIN.
2835 */
2836 if (PageChecked(page))
2837 return -EAGAIN;
2838
2839 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2840 if (!fixup)
2841 return -EAGAIN;
2842
2843 /*
2844 * We are already holding a reference to this inode from
2845 * write_cache_pages. We need to hold it because the space reservation
2846 * takes place outside of the page lock, and we can't trust
2847 * page->mapping outside of the page lock.
2848 */
2849 ihold(inode);
2850 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
2851 get_page(page);
2852 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2853 fixup->page = page;
2854 fixup->inode = BTRFS_I(inode);
2855 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2856
2857 return -EAGAIN;
2858 }
2859
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)2860 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2861 struct btrfs_inode *inode, u64 file_pos,
2862 struct btrfs_file_extent_item *stack_fi,
2863 const bool update_inode_bytes,
2864 u64 qgroup_reserved)
2865 {
2866 struct btrfs_root *root = inode->root;
2867 const u64 sectorsize = root->fs_info->sectorsize;
2868 struct btrfs_path *path;
2869 struct extent_buffer *leaf;
2870 struct btrfs_key ins;
2871 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2872 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2873 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2874 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2875 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2876 struct btrfs_drop_extents_args drop_args = { 0 };
2877 int ret;
2878
2879 path = btrfs_alloc_path();
2880 if (!path)
2881 return -ENOMEM;
2882
2883 /*
2884 * we may be replacing one extent in the tree with another.
2885 * The new extent is pinned in the extent map, and we don't want
2886 * to drop it from the cache until it is completely in the btree.
2887 *
2888 * So, tell btrfs_drop_extents to leave this extent in the cache.
2889 * the caller is expected to unpin it and allow it to be merged
2890 * with the others.
2891 */
2892 drop_args.path = path;
2893 drop_args.start = file_pos;
2894 drop_args.end = file_pos + num_bytes;
2895 drop_args.replace_extent = true;
2896 drop_args.extent_item_size = sizeof(*stack_fi);
2897 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2898 if (ret)
2899 goto out;
2900
2901 if (!drop_args.extent_inserted) {
2902 ins.objectid = btrfs_ino(inode);
2903 ins.offset = file_pos;
2904 ins.type = BTRFS_EXTENT_DATA_KEY;
2905
2906 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2907 sizeof(*stack_fi));
2908 if (ret)
2909 goto out;
2910 }
2911 leaf = path->nodes[0];
2912 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2913 write_extent_buffer(leaf, stack_fi,
2914 btrfs_item_ptr_offset(leaf, path->slots[0]),
2915 sizeof(struct btrfs_file_extent_item));
2916
2917 btrfs_mark_buffer_dirty(trans, leaf);
2918 btrfs_release_path(path);
2919
2920 /*
2921 * If we dropped an inline extent here, we know the range where it is
2922 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2923 * number of bytes only for that range containing the inline extent.
2924 * The remaining of the range will be processed when clearning the
2925 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2926 */
2927 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2928 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2929
2930 inline_size = drop_args.bytes_found - inline_size;
2931 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2932 drop_args.bytes_found -= inline_size;
2933 num_bytes -= sectorsize;
2934 }
2935
2936 if (update_inode_bytes)
2937 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2938
2939 ins.objectid = disk_bytenr;
2940 ins.offset = disk_num_bytes;
2941 ins.type = BTRFS_EXTENT_ITEM_KEY;
2942
2943 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2944 if (ret)
2945 goto out;
2946
2947 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2948 file_pos - offset,
2949 qgroup_reserved, &ins);
2950 out:
2951 btrfs_free_path(path);
2952
2953 return ret;
2954 }
2955
btrfs_release_delalloc_bytes(struct btrfs_fs_info * fs_info,u64 start,u64 len)2956 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2957 u64 start, u64 len)
2958 {
2959 struct btrfs_block_group *cache;
2960
2961 cache = btrfs_lookup_block_group(fs_info, start);
2962 ASSERT(cache);
2963
2964 spin_lock(&cache->lock);
2965 cache->delalloc_bytes -= len;
2966 spin_unlock(&cache->lock);
2967
2968 btrfs_put_block_group(cache);
2969 }
2970
insert_ordered_extent_file_extent(struct btrfs_trans_handle * trans,struct btrfs_ordered_extent * oe)2971 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2972 struct btrfs_ordered_extent *oe)
2973 {
2974 struct btrfs_file_extent_item stack_fi;
2975 bool update_inode_bytes;
2976 u64 num_bytes = oe->num_bytes;
2977 u64 ram_bytes = oe->ram_bytes;
2978
2979 memset(&stack_fi, 0, sizeof(stack_fi));
2980 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2981 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2982 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2983 oe->disk_num_bytes);
2984 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
2985 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
2986 num_bytes = oe->truncated_len;
2987 ram_bytes = num_bytes;
2988 }
2989 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
2990 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
2991 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2992 /* Encryption and other encoding is reserved and all 0 */
2993
2994 /*
2995 * For delalloc, when completing an ordered extent we update the inode's
2996 * bytes when clearing the range in the inode's io tree, so pass false
2997 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2998 * except if the ordered extent was truncated.
2999 */
3000 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3001 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3002 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3003
3004 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3005 oe->file_offset, &stack_fi,
3006 update_inode_bytes, oe->qgroup_rsv);
3007 }
3008
3009 /*
3010 * As ordered data IO finishes, this gets called so we can finish
3011 * an ordered extent if the range of bytes in the file it covers are
3012 * fully written.
3013 */
btrfs_finish_one_ordered(struct btrfs_ordered_extent * ordered_extent)3014 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3015 {
3016 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3017 struct btrfs_root *root = inode->root;
3018 struct btrfs_fs_info *fs_info = root->fs_info;
3019 struct btrfs_trans_handle *trans = NULL;
3020 struct extent_io_tree *io_tree = &inode->io_tree;
3021 struct extent_state *cached_state = NULL;
3022 u64 start, end;
3023 int compress_type = 0;
3024 int ret = 0;
3025 u64 logical_len = ordered_extent->num_bytes;
3026 bool freespace_inode;
3027 bool truncated = false;
3028 bool clear_reserved_extent = true;
3029 unsigned int clear_bits = EXTENT_DEFRAG;
3030
3031 start = ordered_extent->file_offset;
3032 end = start + ordered_extent->num_bytes - 1;
3033
3034 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3035 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3036 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3037 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3038 clear_bits |= EXTENT_DELALLOC_NEW;
3039
3040 freespace_inode = btrfs_is_free_space_inode(inode);
3041 if (!freespace_inode)
3042 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3043
3044 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3045 ret = -EIO;
3046 goto out;
3047 }
3048
3049 if (btrfs_is_zoned(fs_info))
3050 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3051 ordered_extent->disk_num_bytes);
3052
3053 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3054 truncated = true;
3055 logical_len = ordered_extent->truncated_len;
3056 /* Truncated the entire extent, don't bother adding */
3057 if (!logical_len)
3058 goto out;
3059 }
3060
3061 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3062 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3063
3064 btrfs_inode_safe_disk_i_size_write(inode, 0);
3065 if (freespace_inode)
3066 trans = btrfs_join_transaction_spacecache(root);
3067 else
3068 trans = btrfs_join_transaction(root);
3069 if (IS_ERR(trans)) {
3070 ret = PTR_ERR(trans);
3071 trans = NULL;
3072 goto out;
3073 }
3074 trans->block_rsv = &inode->block_rsv;
3075 ret = btrfs_update_inode_fallback(trans, root, inode);
3076 if (ret) /* -ENOMEM or corruption */
3077 btrfs_abort_transaction(trans, ret);
3078 goto out;
3079 }
3080
3081 clear_bits |= EXTENT_LOCKED;
3082 lock_extent(io_tree, start, end, &cached_state);
3083
3084 if (freespace_inode)
3085 trans = btrfs_join_transaction_spacecache(root);
3086 else
3087 trans = btrfs_join_transaction(root);
3088 if (IS_ERR(trans)) {
3089 ret = PTR_ERR(trans);
3090 trans = NULL;
3091 goto out;
3092 }
3093
3094 trans->block_rsv = &inode->block_rsv;
3095
3096 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3097 compress_type = ordered_extent->compress_type;
3098 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3099 BUG_ON(compress_type);
3100 ret = btrfs_mark_extent_written(trans, inode,
3101 ordered_extent->file_offset,
3102 ordered_extent->file_offset +
3103 logical_len);
3104 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3105 ordered_extent->disk_num_bytes);
3106 } else {
3107 BUG_ON(root == fs_info->tree_root);
3108 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3109 if (!ret) {
3110 clear_reserved_extent = false;
3111 btrfs_release_delalloc_bytes(fs_info,
3112 ordered_extent->disk_bytenr,
3113 ordered_extent->disk_num_bytes);
3114 }
3115 }
3116 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3117 ordered_extent->num_bytes, trans->transid);
3118 if (ret < 0) {
3119 btrfs_abort_transaction(trans, ret);
3120 goto out;
3121 }
3122
3123 ret = add_pending_csums(trans, &ordered_extent->list);
3124 if (ret) {
3125 btrfs_abort_transaction(trans, ret);
3126 goto out;
3127 }
3128
3129 /*
3130 * If this is a new delalloc range, clear its new delalloc flag to
3131 * update the inode's number of bytes. This needs to be done first
3132 * before updating the inode item.
3133 */
3134 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3135 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3136 clear_extent_bit(&inode->io_tree, start, end,
3137 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3138 &cached_state);
3139
3140 btrfs_inode_safe_disk_i_size_write(inode, 0);
3141 ret = btrfs_update_inode_fallback(trans, root, inode);
3142 if (ret) { /* -ENOMEM or corruption */
3143 btrfs_abort_transaction(trans, ret);
3144 goto out;
3145 }
3146 ret = 0;
3147 out:
3148 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3149 &cached_state);
3150
3151 if (trans)
3152 btrfs_end_transaction(trans);
3153
3154 if (ret || truncated) {
3155 u64 unwritten_start = start;
3156
3157 /*
3158 * If we failed to finish this ordered extent for any reason we
3159 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3160 * extent, and mark the inode with the error if it wasn't
3161 * already set. Any error during writeback would have already
3162 * set the mapping error, so we need to set it if we're the ones
3163 * marking this ordered extent as failed.
3164 */
3165 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3166 &ordered_extent->flags))
3167 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3168
3169 if (truncated)
3170 unwritten_start += logical_len;
3171 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3172
3173 /*
3174 * Drop extent maps for the part of the extent we didn't write.
3175 *
3176 * We have an exception here for the free_space_inode, this is
3177 * because when we do btrfs_get_extent() on the free space inode
3178 * we will search the commit root. If this is a new block group
3179 * we won't find anything, and we will trip over the assert in
3180 * writepage where we do ASSERT(em->block_start !=
3181 * EXTENT_MAP_HOLE).
3182 *
3183 * Theoretically we could also skip this for any NOCOW extent as
3184 * we don't mess with the extent map tree in the NOCOW case, but
3185 * for now simply skip this if we are the free space inode.
3186 */
3187 if (!btrfs_is_free_space_inode(inode))
3188 btrfs_drop_extent_map_range(inode, unwritten_start,
3189 end, false);
3190
3191 /*
3192 * If the ordered extent had an IOERR or something else went
3193 * wrong we need to return the space for this ordered extent
3194 * back to the allocator. We only free the extent in the
3195 * truncated case if we didn't write out the extent at all.
3196 *
3197 * If we made it past insert_reserved_file_extent before we
3198 * errored out then we don't need to do this as the accounting
3199 * has already been done.
3200 */
3201 if ((ret || !logical_len) &&
3202 clear_reserved_extent &&
3203 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3204 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3205 /*
3206 * Discard the range before returning it back to the
3207 * free space pool
3208 */
3209 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3210 btrfs_discard_extent(fs_info,
3211 ordered_extent->disk_bytenr,
3212 ordered_extent->disk_num_bytes,
3213 NULL);
3214 btrfs_free_reserved_extent(fs_info,
3215 ordered_extent->disk_bytenr,
3216 ordered_extent->disk_num_bytes, 1);
3217 /*
3218 * Actually free the qgroup rsv which was released when
3219 * the ordered extent was created.
3220 */
3221 btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid,
3222 ordered_extent->qgroup_rsv,
3223 BTRFS_QGROUP_RSV_DATA);
3224 }
3225 }
3226
3227 /*
3228 * This needs to be done to make sure anybody waiting knows we are done
3229 * updating everything for this ordered extent.
3230 */
3231 btrfs_remove_ordered_extent(inode, ordered_extent);
3232
3233 /* once for us */
3234 btrfs_put_ordered_extent(ordered_extent);
3235 /* once for the tree */
3236 btrfs_put_ordered_extent(ordered_extent);
3237
3238 return ret;
3239 }
3240
btrfs_finish_ordered_io(struct btrfs_ordered_extent * ordered)3241 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3242 {
3243 if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) &&
3244 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
3245 btrfs_finish_ordered_zoned(ordered);
3246 return btrfs_finish_one_ordered(ordered);
3247 }
3248
3249 /*
3250 * Verify the checksum for a single sector without any extra action that depend
3251 * on the type of I/O.
3252 */
btrfs_check_sector_csum(struct btrfs_fs_info * fs_info,struct page * page,u32 pgoff,u8 * csum,const u8 * const csum_expected)3253 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3254 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3255 {
3256 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3257 char *kaddr;
3258
3259 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3260
3261 shash->tfm = fs_info->csum_shash;
3262
3263 kaddr = kmap_local_page(page) + pgoff;
3264 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3265 kunmap_local(kaddr);
3266
3267 if (memcmp(csum, csum_expected, fs_info->csum_size))
3268 return -EIO;
3269 return 0;
3270 }
3271
3272 /*
3273 * Verify the checksum of a single data sector.
3274 *
3275 * @bbio: btrfs_io_bio which contains the csum
3276 * @dev: device the sector is on
3277 * @bio_offset: offset to the beginning of the bio (in bytes)
3278 * @bv: bio_vec to check
3279 *
3280 * Check if the checksum on a data block is valid. When a checksum mismatch is
3281 * detected, report the error and fill the corrupted range with zero.
3282 *
3283 * Return %true if the sector is ok or had no checksum to start with, else %false.
3284 */
btrfs_data_csum_ok(struct btrfs_bio * bbio,struct btrfs_device * dev,u32 bio_offset,struct bio_vec * bv)3285 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3286 u32 bio_offset, struct bio_vec *bv)
3287 {
3288 struct btrfs_inode *inode = bbio->inode;
3289 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3290 u64 file_offset = bbio->file_offset + bio_offset;
3291 u64 end = file_offset + bv->bv_len - 1;
3292 u8 *csum_expected;
3293 u8 csum[BTRFS_CSUM_SIZE];
3294
3295 ASSERT(bv->bv_len == fs_info->sectorsize);
3296
3297 if (!bbio->csum)
3298 return true;
3299
3300 if (btrfs_is_data_reloc_root(inode->root) &&
3301 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3302 1, NULL)) {
3303 /* Skip the range without csum for data reloc inode */
3304 clear_extent_bits(&inode->io_tree, file_offset, end,
3305 EXTENT_NODATASUM);
3306 return true;
3307 }
3308
3309 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3310 fs_info->csum_size;
3311 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3312 csum_expected))
3313 goto zeroit;
3314 return true;
3315
3316 zeroit:
3317 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3318 bbio->mirror_num);
3319 if (dev)
3320 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3321 memzero_bvec(bv);
3322 return false;
3323 }
3324
3325 /*
3326 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3327 *
3328 * @inode: The inode we want to perform iput on
3329 *
3330 * This function uses the generic vfs_inode::i_count to track whether we should
3331 * just decrement it (in case it's > 1) or if this is the last iput then link
3332 * the inode to the delayed iput machinery. Delayed iputs are processed at
3333 * transaction commit time/superblock commit/cleaner kthread.
3334 */
btrfs_add_delayed_iput(struct btrfs_inode * inode)3335 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3336 {
3337 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3338 unsigned long flags;
3339
3340 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3341 return;
3342
3343 atomic_inc(&fs_info->nr_delayed_iputs);
3344 /*
3345 * Need to be irq safe here because we can be called from either an irq
3346 * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3347 * context.
3348 */
3349 spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3350 ASSERT(list_empty(&inode->delayed_iput));
3351 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3352 spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3353 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3354 wake_up_process(fs_info->cleaner_kthread);
3355 }
3356
run_delayed_iput_locked(struct btrfs_fs_info * fs_info,struct btrfs_inode * inode)3357 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3358 struct btrfs_inode *inode)
3359 {
3360 list_del_init(&inode->delayed_iput);
3361 spin_unlock_irq(&fs_info->delayed_iput_lock);
3362 iput(&inode->vfs_inode);
3363 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3364 wake_up(&fs_info->delayed_iputs_wait);
3365 spin_lock_irq(&fs_info->delayed_iput_lock);
3366 }
3367
btrfs_run_delayed_iput(struct btrfs_fs_info * fs_info,struct btrfs_inode * inode)3368 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3369 struct btrfs_inode *inode)
3370 {
3371 if (!list_empty(&inode->delayed_iput)) {
3372 spin_lock_irq(&fs_info->delayed_iput_lock);
3373 if (!list_empty(&inode->delayed_iput))
3374 run_delayed_iput_locked(fs_info, inode);
3375 spin_unlock_irq(&fs_info->delayed_iput_lock);
3376 }
3377 }
3378
btrfs_run_delayed_iputs(struct btrfs_fs_info * fs_info)3379 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3380 {
3381 /*
3382 * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3383 * calls btrfs_add_delayed_iput() and that needs to lock
3384 * fs_info->delayed_iput_lock. So we need to disable irqs here to
3385 * prevent a deadlock.
3386 */
3387 spin_lock_irq(&fs_info->delayed_iput_lock);
3388 while (!list_empty(&fs_info->delayed_iputs)) {
3389 struct btrfs_inode *inode;
3390
3391 inode = list_first_entry(&fs_info->delayed_iputs,
3392 struct btrfs_inode, delayed_iput);
3393 run_delayed_iput_locked(fs_info, inode);
3394 if (need_resched()) {
3395 spin_unlock_irq(&fs_info->delayed_iput_lock);
3396 cond_resched();
3397 spin_lock_irq(&fs_info->delayed_iput_lock);
3398 }
3399 }
3400 spin_unlock_irq(&fs_info->delayed_iput_lock);
3401 }
3402
3403 /*
3404 * Wait for flushing all delayed iputs
3405 *
3406 * @fs_info: the filesystem
3407 *
3408 * This will wait on any delayed iputs that are currently running with KILLABLE
3409 * set. Once they are all done running we will return, unless we are killed in
3410 * which case we return EINTR. This helps in user operations like fallocate etc
3411 * that might get blocked on the iputs.
3412 *
3413 * Return EINTR if we were killed, 0 if nothing's pending
3414 */
btrfs_wait_on_delayed_iputs(struct btrfs_fs_info * fs_info)3415 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3416 {
3417 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3418 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3419 if (ret)
3420 return -EINTR;
3421 return 0;
3422 }
3423
3424 /*
3425 * This creates an orphan entry for the given inode in case something goes wrong
3426 * in the middle of an unlink.
3427 */
btrfs_orphan_add(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)3428 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3429 struct btrfs_inode *inode)
3430 {
3431 int ret;
3432
3433 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3434 if (ret && ret != -EEXIST) {
3435 btrfs_abort_transaction(trans, ret);
3436 return ret;
3437 }
3438
3439 return 0;
3440 }
3441
3442 /*
3443 * We have done the delete so we can go ahead and remove the orphan item for
3444 * this particular inode.
3445 */
btrfs_orphan_del(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)3446 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3447 struct btrfs_inode *inode)
3448 {
3449 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3450 }
3451
3452 /*
3453 * this cleans up any orphans that may be left on the list from the last use
3454 * of this root.
3455 */
btrfs_orphan_cleanup(struct btrfs_root * root)3456 int btrfs_orphan_cleanup(struct btrfs_root *root)
3457 {
3458 struct btrfs_fs_info *fs_info = root->fs_info;
3459 struct btrfs_path *path;
3460 struct extent_buffer *leaf;
3461 struct btrfs_key key, found_key;
3462 struct btrfs_trans_handle *trans;
3463 struct inode *inode;
3464 u64 last_objectid = 0;
3465 int ret = 0, nr_unlink = 0;
3466
3467 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3468 return 0;
3469
3470 path = btrfs_alloc_path();
3471 if (!path) {
3472 ret = -ENOMEM;
3473 goto out;
3474 }
3475 path->reada = READA_BACK;
3476
3477 key.objectid = BTRFS_ORPHAN_OBJECTID;
3478 key.type = BTRFS_ORPHAN_ITEM_KEY;
3479 key.offset = (u64)-1;
3480
3481 while (1) {
3482 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3483 if (ret < 0)
3484 goto out;
3485
3486 /*
3487 * if ret == 0 means we found what we were searching for, which
3488 * is weird, but possible, so only screw with path if we didn't
3489 * find the key and see if we have stuff that matches
3490 */
3491 if (ret > 0) {
3492 ret = 0;
3493 if (path->slots[0] == 0)
3494 break;
3495 path->slots[0]--;
3496 }
3497
3498 /* pull out the item */
3499 leaf = path->nodes[0];
3500 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3501
3502 /* make sure the item matches what we want */
3503 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3504 break;
3505 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3506 break;
3507
3508 /* release the path since we're done with it */
3509 btrfs_release_path(path);
3510
3511 /*
3512 * this is where we are basically btrfs_lookup, without the
3513 * crossing root thing. we store the inode number in the
3514 * offset of the orphan item.
3515 */
3516
3517 if (found_key.offset == last_objectid) {
3518 /*
3519 * We found the same inode as before. This means we were
3520 * not able to remove its items via eviction triggered
3521 * by an iput(). A transaction abort may have happened,
3522 * due to -ENOSPC for example, so try to grab the error
3523 * that lead to a transaction abort, if any.
3524 */
3525 btrfs_err(fs_info,
3526 "Error removing orphan entry, stopping orphan cleanup");
3527 ret = BTRFS_FS_ERROR(fs_info) ?: -EINVAL;
3528 goto out;
3529 }
3530
3531 last_objectid = found_key.offset;
3532
3533 found_key.objectid = found_key.offset;
3534 found_key.type = BTRFS_INODE_ITEM_KEY;
3535 found_key.offset = 0;
3536 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3537 if (IS_ERR(inode)) {
3538 ret = PTR_ERR(inode);
3539 inode = NULL;
3540 if (ret != -ENOENT)
3541 goto out;
3542 }
3543
3544 if (!inode && root == fs_info->tree_root) {
3545 struct btrfs_root *dead_root;
3546 int is_dead_root = 0;
3547
3548 /*
3549 * This is an orphan in the tree root. Currently these
3550 * could come from 2 sources:
3551 * a) a root (snapshot/subvolume) deletion in progress
3552 * b) a free space cache inode
3553 * We need to distinguish those two, as the orphan item
3554 * for a root must not get deleted before the deletion
3555 * of the snapshot/subvolume's tree completes.
3556 *
3557 * btrfs_find_orphan_roots() ran before us, which has
3558 * found all deleted roots and loaded them into
3559 * fs_info->fs_roots_radix. So here we can find if an
3560 * orphan item corresponds to a deleted root by looking
3561 * up the root from that radix tree.
3562 */
3563
3564 spin_lock(&fs_info->fs_roots_radix_lock);
3565 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3566 (unsigned long)found_key.objectid);
3567 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3568 is_dead_root = 1;
3569 spin_unlock(&fs_info->fs_roots_radix_lock);
3570
3571 if (is_dead_root) {
3572 /* prevent this orphan from being found again */
3573 key.offset = found_key.objectid - 1;
3574 continue;
3575 }
3576
3577 }
3578
3579 /*
3580 * If we have an inode with links, there are a couple of
3581 * possibilities:
3582 *
3583 * 1. We were halfway through creating fsverity metadata for the
3584 * file. In that case, the orphan item represents incomplete
3585 * fsverity metadata which must be cleaned up with
3586 * btrfs_drop_verity_items and deleting the orphan item.
3587
3588 * 2. Old kernels (before v3.12) used to create an
3589 * orphan item for truncate indicating that there were possibly
3590 * extent items past i_size that needed to be deleted. In v3.12,
3591 * truncate was changed to update i_size in sync with the extent
3592 * items, but the (useless) orphan item was still created. Since
3593 * v4.18, we don't create the orphan item for truncate at all.
3594 *
3595 * So, this item could mean that we need to do a truncate, but
3596 * only if this filesystem was last used on a pre-v3.12 kernel
3597 * and was not cleanly unmounted. The odds of that are quite
3598 * slim, and it's a pain to do the truncate now, so just delete
3599 * the orphan item.
3600 *
3601 * It's also possible that this orphan item was supposed to be
3602 * deleted but wasn't. The inode number may have been reused,
3603 * but either way, we can delete the orphan item.
3604 */
3605 if (!inode || inode->i_nlink) {
3606 if (inode) {
3607 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3608 iput(inode);
3609 inode = NULL;
3610 if (ret)
3611 goto out;
3612 }
3613 trans = btrfs_start_transaction(root, 1);
3614 if (IS_ERR(trans)) {
3615 ret = PTR_ERR(trans);
3616 goto out;
3617 }
3618 btrfs_debug(fs_info, "auto deleting %Lu",
3619 found_key.objectid);
3620 ret = btrfs_del_orphan_item(trans, root,
3621 found_key.objectid);
3622 btrfs_end_transaction(trans);
3623 if (ret)
3624 goto out;
3625 continue;
3626 }
3627
3628 nr_unlink++;
3629
3630 /* this will do delete_inode and everything for us */
3631 iput(inode);
3632 }
3633 /* release the path since we're done with it */
3634 btrfs_release_path(path);
3635
3636 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3637 trans = btrfs_join_transaction(root);
3638 if (!IS_ERR(trans))
3639 btrfs_end_transaction(trans);
3640 }
3641
3642 if (nr_unlink)
3643 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3644
3645 out:
3646 if (ret)
3647 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3648 btrfs_free_path(path);
3649 return ret;
3650 }
3651
3652 /*
3653 * very simple check to peek ahead in the leaf looking for xattrs. If we
3654 * don't find any xattrs, we know there can't be any acls.
3655 *
3656 * slot is the slot the inode is in, objectid is the objectid of the inode
3657 */
acls_after_inode_item(struct extent_buffer * leaf,int slot,u64 objectid,int * first_xattr_slot)3658 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3659 int slot, u64 objectid,
3660 int *first_xattr_slot)
3661 {
3662 u32 nritems = btrfs_header_nritems(leaf);
3663 struct btrfs_key found_key;
3664 static u64 xattr_access = 0;
3665 static u64 xattr_default = 0;
3666 int scanned = 0;
3667
3668 if (!xattr_access) {
3669 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3670 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3671 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3672 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3673 }
3674
3675 slot++;
3676 *first_xattr_slot = -1;
3677 while (slot < nritems) {
3678 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3679
3680 /* we found a different objectid, there must not be acls */
3681 if (found_key.objectid != objectid)
3682 return 0;
3683
3684 /* we found an xattr, assume we've got an acl */
3685 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3686 if (*first_xattr_slot == -1)
3687 *first_xattr_slot = slot;
3688 if (found_key.offset == xattr_access ||
3689 found_key.offset == xattr_default)
3690 return 1;
3691 }
3692
3693 /*
3694 * we found a key greater than an xattr key, there can't
3695 * be any acls later on
3696 */
3697 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3698 return 0;
3699
3700 slot++;
3701 scanned++;
3702
3703 /*
3704 * it goes inode, inode backrefs, xattrs, extents,
3705 * so if there are a ton of hard links to an inode there can
3706 * be a lot of backrefs. Don't waste time searching too hard,
3707 * this is just an optimization
3708 */
3709 if (scanned >= 8)
3710 break;
3711 }
3712 /* we hit the end of the leaf before we found an xattr or
3713 * something larger than an xattr. We have to assume the inode
3714 * has acls
3715 */
3716 if (*first_xattr_slot == -1)
3717 *first_xattr_slot = slot;
3718 return 1;
3719 }
3720
3721 /*
3722 * read an inode from the btree into the in-memory inode
3723 */
btrfs_read_locked_inode(struct inode * inode,struct btrfs_path * in_path)3724 static int btrfs_read_locked_inode(struct inode *inode,
3725 struct btrfs_path *in_path)
3726 {
3727 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3728 struct btrfs_path *path = in_path;
3729 struct extent_buffer *leaf;
3730 struct btrfs_inode_item *inode_item;
3731 struct btrfs_root *root = BTRFS_I(inode)->root;
3732 struct btrfs_key location;
3733 unsigned long ptr;
3734 int maybe_acls;
3735 u32 rdev;
3736 int ret;
3737 bool filled = false;
3738 int first_xattr_slot;
3739
3740 ret = btrfs_fill_inode(inode, &rdev);
3741 if (!ret)
3742 filled = true;
3743
3744 if (!path) {
3745 path = btrfs_alloc_path();
3746 if (!path)
3747 return -ENOMEM;
3748 }
3749
3750 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3751
3752 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3753 if (ret) {
3754 if (path != in_path)
3755 btrfs_free_path(path);
3756 return ret;
3757 }
3758
3759 leaf = path->nodes[0];
3760
3761 if (filled)
3762 goto cache_index;
3763
3764 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3765 struct btrfs_inode_item);
3766 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3767 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3768 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3769 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3770 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3771 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3772 round_up(i_size_read(inode), fs_info->sectorsize));
3773
3774 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3775 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3776
3777 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3778 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3779
3780 inode_set_ctime(inode, btrfs_timespec_sec(leaf, &inode_item->ctime),
3781 btrfs_timespec_nsec(leaf, &inode_item->ctime));
3782
3783 BTRFS_I(inode)->i_otime.tv_sec =
3784 btrfs_timespec_sec(leaf, &inode_item->otime);
3785 BTRFS_I(inode)->i_otime.tv_nsec =
3786 btrfs_timespec_nsec(leaf, &inode_item->otime);
3787
3788 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3789 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3790 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3791
3792 inode_set_iversion_queried(inode,
3793 btrfs_inode_sequence(leaf, inode_item));
3794 inode->i_generation = BTRFS_I(inode)->generation;
3795 inode->i_rdev = 0;
3796 rdev = btrfs_inode_rdev(leaf, inode_item);
3797
3798 BTRFS_I(inode)->index_cnt = (u64)-1;
3799 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3800 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3801
3802 cache_index:
3803 /*
3804 * If we were modified in the current generation and evicted from memory
3805 * and then re-read we need to do a full sync since we don't have any
3806 * idea about which extents were modified before we were evicted from
3807 * cache.
3808 *
3809 * This is required for both inode re-read from disk and delayed inode
3810 * in delayed_nodes_tree.
3811 */
3812 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3813 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3814 &BTRFS_I(inode)->runtime_flags);
3815
3816 /*
3817 * We don't persist the id of the transaction where an unlink operation
3818 * against the inode was last made. So here we assume the inode might
3819 * have been evicted, and therefore the exact value of last_unlink_trans
3820 * lost, and set it to last_trans to avoid metadata inconsistencies
3821 * between the inode and its parent if the inode is fsync'ed and the log
3822 * replayed. For example, in the scenario:
3823 *
3824 * touch mydir/foo
3825 * ln mydir/foo mydir/bar
3826 * sync
3827 * unlink mydir/bar
3828 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3829 * xfs_io -c fsync mydir/foo
3830 * <power failure>
3831 * mount fs, triggers fsync log replay
3832 *
3833 * We must make sure that when we fsync our inode foo we also log its
3834 * parent inode, otherwise after log replay the parent still has the
3835 * dentry with the "bar" name but our inode foo has a link count of 1
3836 * and doesn't have an inode ref with the name "bar" anymore.
3837 *
3838 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3839 * but it guarantees correctness at the expense of occasional full
3840 * transaction commits on fsync if our inode is a directory, or if our
3841 * inode is not a directory, logging its parent unnecessarily.
3842 */
3843 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3844
3845 /*
3846 * Same logic as for last_unlink_trans. We don't persist the generation
3847 * of the last transaction where this inode was used for a reflink
3848 * operation, so after eviction and reloading the inode we must be
3849 * pessimistic and assume the last transaction that modified the inode.
3850 */
3851 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3852
3853 path->slots[0]++;
3854 if (inode->i_nlink != 1 ||
3855 path->slots[0] >= btrfs_header_nritems(leaf))
3856 goto cache_acl;
3857
3858 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3859 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3860 goto cache_acl;
3861
3862 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3863 if (location.type == BTRFS_INODE_REF_KEY) {
3864 struct btrfs_inode_ref *ref;
3865
3866 ref = (struct btrfs_inode_ref *)ptr;
3867 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3868 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3869 struct btrfs_inode_extref *extref;
3870
3871 extref = (struct btrfs_inode_extref *)ptr;
3872 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3873 extref);
3874 }
3875 cache_acl:
3876 /*
3877 * try to precache a NULL acl entry for files that don't have
3878 * any xattrs or acls
3879 */
3880 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3881 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3882 if (first_xattr_slot != -1) {
3883 path->slots[0] = first_xattr_slot;
3884 ret = btrfs_load_inode_props(inode, path);
3885 if (ret)
3886 btrfs_err(fs_info,
3887 "error loading props for ino %llu (root %llu): %d",
3888 btrfs_ino(BTRFS_I(inode)),
3889 root->root_key.objectid, ret);
3890 }
3891 if (path != in_path)
3892 btrfs_free_path(path);
3893
3894 if (!maybe_acls)
3895 cache_no_acl(inode);
3896
3897 switch (inode->i_mode & S_IFMT) {
3898 case S_IFREG:
3899 inode->i_mapping->a_ops = &btrfs_aops;
3900 inode->i_fop = &btrfs_file_operations;
3901 inode->i_op = &btrfs_file_inode_operations;
3902 break;
3903 case S_IFDIR:
3904 inode->i_fop = &btrfs_dir_file_operations;
3905 inode->i_op = &btrfs_dir_inode_operations;
3906 break;
3907 case S_IFLNK:
3908 inode->i_op = &btrfs_symlink_inode_operations;
3909 inode_nohighmem(inode);
3910 inode->i_mapping->a_ops = &btrfs_aops;
3911 break;
3912 default:
3913 inode->i_op = &btrfs_special_inode_operations;
3914 init_special_inode(inode, inode->i_mode, rdev);
3915 break;
3916 }
3917
3918 btrfs_sync_inode_flags_to_i_flags(inode);
3919 return 0;
3920 }
3921
3922 /*
3923 * given a leaf and an inode, copy the inode fields into the leaf
3924 */
fill_inode_item(struct btrfs_trans_handle * trans,struct extent_buffer * leaf,struct btrfs_inode_item * item,struct inode * inode)3925 static void fill_inode_item(struct btrfs_trans_handle *trans,
3926 struct extent_buffer *leaf,
3927 struct btrfs_inode_item *item,
3928 struct inode *inode)
3929 {
3930 struct btrfs_map_token token;
3931 u64 flags;
3932
3933 btrfs_init_map_token(&token, leaf);
3934
3935 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3936 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3937 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3938 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3939 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3940
3941 btrfs_set_token_timespec_sec(&token, &item->atime,
3942 inode->i_atime.tv_sec);
3943 btrfs_set_token_timespec_nsec(&token, &item->atime,
3944 inode->i_atime.tv_nsec);
3945
3946 btrfs_set_token_timespec_sec(&token, &item->mtime,
3947 inode->i_mtime.tv_sec);
3948 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3949 inode->i_mtime.tv_nsec);
3950
3951 btrfs_set_token_timespec_sec(&token, &item->ctime,
3952 inode_get_ctime(inode).tv_sec);
3953 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3954 inode_get_ctime(inode).tv_nsec);
3955
3956 btrfs_set_token_timespec_sec(&token, &item->otime,
3957 BTRFS_I(inode)->i_otime.tv_sec);
3958 btrfs_set_token_timespec_nsec(&token, &item->otime,
3959 BTRFS_I(inode)->i_otime.tv_nsec);
3960
3961 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3962 btrfs_set_token_inode_generation(&token, item,
3963 BTRFS_I(inode)->generation);
3964 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3965 btrfs_set_token_inode_transid(&token, item, trans->transid);
3966 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3967 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3968 BTRFS_I(inode)->ro_flags);
3969 btrfs_set_token_inode_flags(&token, item, flags);
3970 btrfs_set_token_inode_block_group(&token, item, 0);
3971 }
3972
3973 /*
3974 * copy everything in the in-memory inode into the btree.
3975 */
btrfs_update_inode_item(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode)3976 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3977 struct btrfs_root *root,
3978 struct btrfs_inode *inode)
3979 {
3980 struct btrfs_inode_item *inode_item;
3981 struct btrfs_path *path;
3982 struct extent_buffer *leaf;
3983 int ret;
3984
3985 path = btrfs_alloc_path();
3986 if (!path)
3987 return -ENOMEM;
3988
3989 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
3990 if (ret) {
3991 if (ret > 0)
3992 ret = -ENOENT;
3993 goto failed;
3994 }
3995
3996 leaf = path->nodes[0];
3997 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3998 struct btrfs_inode_item);
3999
4000 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4001 btrfs_mark_buffer_dirty(trans, leaf);
4002 btrfs_set_inode_last_trans(trans, inode);
4003 ret = 0;
4004 failed:
4005 btrfs_free_path(path);
4006 return ret;
4007 }
4008
4009 /*
4010 * copy everything in the in-memory inode into the btree.
4011 */
btrfs_update_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode)4012 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4013 struct btrfs_root *root,
4014 struct btrfs_inode *inode)
4015 {
4016 struct btrfs_fs_info *fs_info = root->fs_info;
4017 int ret;
4018
4019 /*
4020 * If the inode is a free space inode, we can deadlock during commit
4021 * if we put it into the delayed code.
4022 *
4023 * The data relocation inode should also be directly updated
4024 * without delay
4025 */
4026 if (!btrfs_is_free_space_inode(inode)
4027 && !btrfs_is_data_reloc_root(root)
4028 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4029 btrfs_update_root_times(trans, root);
4030
4031 ret = btrfs_delayed_update_inode(trans, root, inode);
4032 if (!ret)
4033 btrfs_set_inode_last_trans(trans, inode);
4034 return ret;
4035 }
4036
4037 return btrfs_update_inode_item(trans, root, inode);
4038 }
4039
btrfs_update_inode_fallback(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode)4040 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4041 struct btrfs_root *root, struct btrfs_inode *inode)
4042 {
4043 int ret;
4044
4045 ret = btrfs_update_inode(trans, root, inode);
4046 if (ret == -ENOSPC)
4047 return btrfs_update_inode_item(trans, root, inode);
4048 return ret;
4049 }
4050
4051 /*
4052 * unlink helper that gets used here in inode.c and in the tree logging
4053 * recovery code. It remove a link in a directory with a given name, and
4054 * also drops the back refs in the inode to the directory
4055 */
__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)4056 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4057 struct btrfs_inode *dir,
4058 struct btrfs_inode *inode,
4059 const struct fscrypt_str *name,
4060 struct btrfs_rename_ctx *rename_ctx)
4061 {
4062 struct btrfs_root *root = dir->root;
4063 struct btrfs_fs_info *fs_info = root->fs_info;
4064 struct btrfs_path *path;
4065 int ret = 0;
4066 struct btrfs_dir_item *di;
4067 u64 index;
4068 u64 ino = btrfs_ino(inode);
4069 u64 dir_ino = btrfs_ino(dir);
4070
4071 path = btrfs_alloc_path();
4072 if (!path) {
4073 ret = -ENOMEM;
4074 goto out;
4075 }
4076
4077 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4078 if (IS_ERR_OR_NULL(di)) {
4079 ret = di ? PTR_ERR(di) : -ENOENT;
4080 goto err;
4081 }
4082 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4083 if (ret)
4084 goto err;
4085 btrfs_release_path(path);
4086
4087 /*
4088 * If we don't have dir index, we have to get it by looking up
4089 * the inode ref, since we get the inode ref, remove it directly,
4090 * it is unnecessary to do delayed deletion.
4091 *
4092 * But if we have dir index, needn't search inode ref to get it.
4093 * Since the inode ref is close to the inode item, it is better
4094 * that we delay to delete it, and just do this deletion when
4095 * we update the inode item.
4096 */
4097 if (inode->dir_index) {
4098 ret = btrfs_delayed_delete_inode_ref(inode);
4099 if (!ret) {
4100 index = inode->dir_index;
4101 goto skip_backref;
4102 }
4103 }
4104
4105 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4106 if (ret) {
4107 btrfs_info(fs_info,
4108 "failed to delete reference to %.*s, inode %llu parent %llu",
4109 name->len, name->name, ino, dir_ino);
4110 btrfs_abort_transaction(trans, ret);
4111 goto err;
4112 }
4113 skip_backref:
4114 if (rename_ctx)
4115 rename_ctx->index = index;
4116
4117 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4118 if (ret) {
4119 btrfs_abort_transaction(trans, ret);
4120 goto err;
4121 }
4122
4123 /*
4124 * If we are in a rename context, we don't need to update anything in the
4125 * log. That will be done later during the rename by btrfs_log_new_name().
4126 * Besides that, doing it here would only cause extra unnecessary btree
4127 * operations on the log tree, increasing latency for applications.
4128 */
4129 if (!rename_ctx) {
4130 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4131 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4132 }
4133
4134 /*
4135 * If we have a pending delayed iput we could end up with the final iput
4136 * being run in btrfs-cleaner context. If we have enough of these built
4137 * up we can end up burning a lot of time in btrfs-cleaner without any
4138 * way to throttle the unlinks. Since we're currently holding a ref on
4139 * the inode we can run the delayed iput here without any issues as the
4140 * final iput won't be done until after we drop the ref we're currently
4141 * holding.
4142 */
4143 btrfs_run_delayed_iput(fs_info, inode);
4144 err:
4145 btrfs_free_path(path);
4146 if (ret)
4147 goto out;
4148
4149 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4150 inode_inc_iversion(&inode->vfs_inode);
4151 inode_set_ctime_current(&inode->vfs_inode);
4152 inode_inc_iversion(&dir->vfs_inode);
4153 inode_set_ctime_current(&inode->vfs_inode);
4154 dir->vfs_inode.i_mtime = inode_set_ctime_current(&dir->vfs_inode);
4155 ret = btrfs_update_inode(trans, root, dir);
4156 out:
4157 return ret;
4158 }
4159
btrfs_unlink_inode(struct btrfs_trans_handle * trans,struct btrfs_inode * dir,struct btrfs_inode * inode,const struct fscrypt_str * name)4160 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4161 struct btrfs_inode *dir, struct btrfs_inode *inode,
4162 const struct fscrypt_str *name)
4163 {
4164 int ret;
4165
4166 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4167 if (!ret) {
4168 drop_nlink(&inode->vfs_inode);
4169 ret = btrfs_update_inode(trans, inode->root, inode);
4170 }
4171 return ret;
4172 }
4173
4174 /*
4175 * helper to start transaction for unlink and rmdir.
4176 *
4177 * unlink and rmdir are special in btrfs, they do not always free space, so
4178 * if we cannot make our reservations the normal way try and see if there is
4179 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4180 * allow the unlink to occur.
4181 */
__unlink_start_trans(struct btrfs_inode * dir)4182 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4183 {
4184 struct btrfs_root *root = dir->root;
4185
4186 return btrfs_start_transaction_fallback_global_rsv(root,
4187 BTRFS_UNLINK_METADATA_UNITS);
4188 }
4189
btrfs_unlink(struct inode * dir,struct dentry * dentry)4190 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4191 {
4192 struct btrfs_trans_handle *trans;
4193 struct inode *inode = d_inode(dentry);
4194 int ret;
4195 struct fscrypt_name fname;
4196
4197 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4198 if (ret)
4199 return ret;
4200
4201 /* This needs to handle no-key deletions later on */
4202
4203 trans = __unlink_start_trans(BTRFS_I(dir));
4204 if (IS_ERR(trans)) {
4205 ret = PTR_ERR(trans);
4206 goto fscrypt_free;
4207 }
4208
4209 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4210 false);
4211
4212 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4213 &fname.disk_name);
4214 if (ret)
4215 goto end_trans;
4216
4217 if (inode->i_nlink == 0) {
4218 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4219 if (ret)
4220 goto end_trans;
4221 }
4222
4223 end_trans:
4224 btrfs_end_transaction(trans);
4225 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4226 fscrypt_free:
4227 fscrypt_free_filename(&fname);
4228 return ret;
4229 }
4230
btrfs_unlink_subvol(struct btrfs_trans_handle * trans,struct btrfs_inode * dir,struct dentry * dentry)4231 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4232 struct btrfs_inode *dir, struct dentry *dentry)
4233 {
4234 struct btrfs_root *root = dir->root;
4235 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4236 struct btrfs_path *path;
4237 struct extent_buffer *leaf;
4238 struct btrfs_dir_item *di;
4239 struct btrfs_key key;
4240 u64 index;
4241 int ret;
4242 u64 objectid;
4243 u64 dir_ino = btrfs_ino(dir);
4244 struct fscrypt_name fname;
4245
4246 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4247 if (ret)
4248 return ret;
4249
4250 /* This needs to handle no-key deletions later on */
4251
4252 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4253 objectid = inode->root->root_key.objectid;
4254 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4255 objectid = inode->location.objectid;
4256 } else {
4257 WARN_ON(1);
4258 fscrypt_free_filename(&fname);
4259 return -EINVAL;
4260 }
4261
4262 path = btrfs_alloc_path();
4263 if (!path) {
4264 ret = -ENOMEM;
4265 goto out;
4266 }
4267
4268 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4269 &fname.disk_name, -1);
4270 if (IS_ERR_OR_NULL(di)) {
4271 ret = di ? PTR_ERR(di) : -ENOENT;
4272 goto out;
4273 }
4274
4275 leaf = path->nodes[0];
4276 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4277 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4278 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4279 if (ret) {
4280 btrfs_abort_transaction(trans, ret);
4281 goto out;
4282 }
4283 btrfs_release_path(path);
4284
4285 /*
4286 * This is a placeholder inode for a subvolume we didn't have a
4287 * reference to at the time of the snapshot creation. In the meantime
4288 * we could have renamed the real subvol link into our snapshot, so
4289 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4290 * Instead simply lookup the dir_index_item for this entry so we can
4291 * remove it. Otherwise we know we have a ref to the root and we can
4292 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4293 */
4294 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4295 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4296 if (IS_ERR_OR_NULL(di)) {
4297 if (!di)
4298 ret = -ENOENT;
4299 else
4300 ret = PTR_ERR(di);
4301 btrfs_abort_transaction(trans, ret);
4302 goto out;
4303 }
4304
4305 leaf = path->nodes[0];
4306 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4307 index = key.offset;
4308 btrfs_release_path(path);
4309 } else {
4310 ret = btrfs_del_root_ref(trans, objectid,
4311 root->root_key.objectid, dir_ino,
4312 &index, &fname.disk_name);
4313 if (ret) {
4314 btrfs_abort_transaction(trans, ret);
4315 goto out;
4316 }
4317 }
4318
4319 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4320 if (ret) {
4321 btrfs_abort_transaction(trans, ret);
4322 goto out;
4323 }
4324
4325 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4326 inode_inc_iversion(&dir->vfs_inode);
4327 dir->vfs_inode.i_mtime = inode_set_ctime_current(&dir->vfs_inode);
4328 ret = btrfs_update_inode_fallback(trans, root, dir);
4329 if (ret)
4330 btrfs_abort_transaction(trans, ret);
4331 out:
4332 btrfs_free_path(path);
4333 fscrypt_free_filename(&fname);
4334 return ret;
4335 }
4336
4337 /*
4338 * Helper to check if the subvolume references other subvolumes or if it's
4339 * default.
4340 */
may_destroy_subvol(struct btrfs_root * root)4341 static noinline int may_destroy_subvol(struct btrfs_root *root)
4342 {
4343 struct btrfs_fs_info *fs_info = root->fs_info;
4344 struct btrfs_path *path;
4345 struct btrfs_dir_item *di;
4346 struct btrfs_key key;
4347 struct fscrypt_str name = FSTR_INIT("default", 7);
4348 u64 dir_id;
4349 int ret;
4350
4351 path = btrfs_alloc_path();
4352 if (!path)
4353 return -ENOMEM;
4354
4355 /* Make sure this root isn't set as the default subvol */
4356 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4357 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4358 dir_id, &name, 0);
4359 if (di && !IS_ERR(di)) {
4360 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4361 if (key.objectid == root->root_key.objectid) {
4362 ret = -EPERM;
4363 btrfs_err(fs_info,
4364 "deleting default subvolume %llu is not allowed",
4365 key.objectid);
4366 goto out;
4367 }
4368 btrfs_release_path(path);
4369 }
4370
4371 key.objectid = root->root_key.objectid;
4372 key.type = BTRFS_ROOT_REF_KEY;
4373 key.offset = (u64)-1;
4374
4375 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4376 if (ret < 0)
4377 goto out;
4378 if (ret == 0) {
4379 /*
4380 * Key with offset -1 found, there would have to exist a root
4381 * with such id, but this is out of valid range.
4382 */
4383 ret = -EUCLEAN;
4384 goto out;
4385 }
4386
4387 ret = 0;
4388 if (path->slots[0] > 0) {
4389 path->slots[0]--;
4390 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4391 if (key.objectid == root->root_key.objectid &&
4392 key.type == BTRFS_ROOT_REF_KEY)
4393 ret = -ENOTEMPTY;
4394 }
4395 out:
4396 btrfs_free_path(path);
4397 return ret;
4398 }
4399
4400 /* Delete all dentries for inodes belonging to the root */
btrfs_prune_dentries(struct btrfs_root * root)4401 static void btrfs_prune_dentries(struct btrfs_root *root)
4402 {
4403 struct btrfs_fs_info *fs_info = root->fs_info;
4404 struct rb_node *node;
4405 struct rb_node *prev;
4406 struct btrfs_inode *entry;
4407 struct inode *inode;
4408 u64 objectid = 0;
4409
4410 if (!BTRFS_FS_ERROR(fs_info))
4411 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4412
4413 spin_lock(&root->inode_lock);
4414 again:
4415 node = root->inode_tree.rb_node;
4416 prev = NULL;
4417 while (node) {
4418 prev = node;
4419 entry = rb_entry(node, struct btrfs_inode, rb_node);
4420
4421 if (objectid < btrfs_ino(entry))
4422 node = node->rb_left;
4423 else if (objectid > btrfs_ino(entry))
4424 node = node->rb_right;
4425 else
4426 break;
4427 }
4428 if (!node) {
4429 while (prev) {
4430 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4431 if (objectid <= btrfs_ino(entry)) {
4432 node = prev;
4433 break;
4434 }
4435 prev = rb_next(prev);
4436 }
4437 }
4438 while (node) {
4439 entry = rb_entry(node, struct btrfs_inode, rb_node);
4440 objectid = btrfs_ino(entry) + 1;
4441 inode = igrab(&entry->vfs_inode);
4442 if (inode) {
4443 spin_unlock(&root->inode_lock);
4444 if (atomic_read(&inode->i_count) > 1)
4445 d_prune_aliases(inode);
4446 /*
4447 * btrfs_drop_inode will have it removed from the inode
4448 * cache when its usage count hits zero.
4449 */
4450 iput(inode);
4451 cond_resched();
4452 spin_lock(&root->inode_lock);
4453 goto again;
4454 }
4455
4456 if (cond_resched_lock(&root->inode_lock))
4457 goto again;
4458
4459 node = rb_next(node);
4460 }
4461 spin_unlock(&root->inode_lock);
4462 }
4463
btrfs_delete_subvolume(struct btrfs_inode * dir,struct dentry * dentry)4464 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4465 {
4466 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4467 struct btrfs_root *root = dir->root;
4468 struct inode *inode = d_inode(dentry);
4469 struct btrfs_root *dest = BTRFS_I(inode)->root;
4470 struct btrfs_trans_handle *trans;
4471 struct btrfs_block_rsv block_rsv;
4472 u64 root_flags;
4473 u64 qgroup_reserved = 0;
4474 int ret;
4475
4476 down_write(&fs_info->subvol_sem);
4477
4478 /*
4479 * Don't allow to delete a subvolume with send in progress. This is
4480 * inside the inode lock so the error handling that has to drop the bit
4481 * again is not run concurrently.
4482 */
4483 spin_lock(&dest->root_item_lock);
4484 if (dest->send_in_progress) {
4485 spin_unlock(&dest->root_item_lock);
4486 btrfs_warn(fs_info,
4487 "attempt to delete subvolume %llu during send",
4488 dest->root_key.objectid);
4489 ret = -EPERM;
4490 goto out_up_write;
4491 }
4492 if (atomic_read(&dest->nr_swapfiles)) {
4493 spin_unlock(&dest->root_item_lock);
4494 btrfs_warn(fs_info,
4495 "attempt to delete subvolume %llu with active swapfile",
4496 root->root_key.objectid);
4497 ret = -EPERM;
4498 goto out_up_write;
4499 }
4500 root_flags = btrfs_root_flags(&dest->root_item);
4501 btrfs_set_root_flags(&dest->root_item,
4502 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4503 spin_unlock(&dest->root_item_lock);
4504
4505 ret = may_destroy_subvol(dest);
4506 if (ret)
4507 goto out_undead;
4508
4509 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4510 /*
4511 * One for dir inode,
4512 * two for dir entries,
4513 * two for root ref/backref.
4514 */
4515 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4516 if (ret)
4517 goto out_undead;
4518 qgroup_reserved = block_rsv.qgroup_rsv_reserved;
4519
4520 trans = btrfs_start_transaction(root, 0);
4521 if (IS_ERR(trans)) {
4522 ret = PTR_ERR(trans);
4523 goto out_release;
4524 }
4525 ret = btrfs_record_root_in_trans(trans, root);
4526 if (ret) {
4527 btrfs_abort_transaction(trans, ret);
4528 goto out_end_trans;
4529 }
4530 btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
4531 qgroup_reserved = 0;
4532 trans->block_rsv = &block_rsv;
4533 trans->bytes_reserved = block_rsv.size;
4534
4535 btrfs_record_snapshot_destroy(trans, dir);
4536
4537 ret = btrfs_unlink_subvol(trans, dir, dentry);
4538 if (ret) {
4539 btrfs_abort_transaction(trans, ret);
4540 goto out_end_trans;
4541 }
4542
4543 ret = btrfs_record_root_in_trans(trans, dest);
4544 if (ret) {
4545 btrfs_abort_transaction(trans, ret);
4546 goto out_end_trans;
4547 }
4548
4549 memset(&dest->root_item.drop_progress, 0,
4550 sizeof(dest->root_item.drop_progress));
4551 btrfs_set_root_drop_level(&dest->root_item, 0);
4552 btrfs_set_root_refs(&dest->root_item, 0);
4553
4554 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4555 ret = btrfs_insert_orphan_item(trans,
4556 fs_info->tree_root,
4557 dest->root_key.objectid);
4558 if (ret) {
4559 btrfs_abort_transaction(trans, ret);
4560 goto out_end_trans;
4561 }
4562 }
4563
4564 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4565 BTRFS_UUID_KEY_SUBVOL,
4566 dest->root_key.objectid);
4567 if (ret && ret != -ENOENT) {
4568 btrfs_abort_transaction(trans, ret);
4569 goto out_end_trans;
4570 }
4571 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4572 ret = btrfs_uuid_tree_remove(trans,
4573 dest->root_item.received_uuid,
4574 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4575 dest->root_key.objectid);
4576 if (ret && ret != -ENOENT) {
4577 btrfs_abort_transaction(trans, ret);
4578 goto out_end_trans;
4579 }
4580 }
4581
4582 free_anon_bdev(dest->anon_dev);
4583 dest->anon_dev = 0;
4584 out_end_trans:
4585 trans->block_rsv = NULL;
4586 trans->bytes_reserved = 0;
4587 ret = btrfs_end_transaction(trans);
4588 inode->i_flags |= S_DEAD;
4589 out_release:
4590 btrfs_block_rsv_release(fs_info, &block_rsv, (u64)-1, NULL);
4591 if (qgroup_reserved)
4592 btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved);
4593 out_undead:
4594 if (ret) {
4595 spin_lock(&dest->root_item_lock);
4596 root_flags = btrfs_root_flags(&dest->root_item);
4597 btrfs_set_root_flags(&dest->root_item,
4598 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4599 spin_unlock(&dest->root_item_lock);
4600 }
4601 out_up_write:
4602 up_write(&fs_info->subvol_sem);
4603 if (!ret) {
4604 d_invalidate(dentry);
4605 btrfs_prune_dentries(dest);
4606 ASSERT(dest->send_in_progress == 0);
4607 }
4608
4609 return ret;
4610 }
4611
btrfs_rmdir(struct inode * dir,struct dentry * dentry)4612 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4613 {
4614 struct inode *inode = d_inode(dentry);
4615 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4616 int err = 0;
4617 struct btrfs_trans_handle *trans;
4618 u64 last_unlink_trans;
4619 struct fscrypt_name fname;
4620
4621 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4622 return -ENOTEMPTY;
4623 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4624 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4625 btrfs_err(fs_info,
4626 "extent tree v2 doesn't support snapshot deletion yet");
4627 return -EOPNOTSUPP;
4628 }
4629 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4630 }
4631
4632 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4633 if (err)
4634 return err;
4635
4636 /* This needs to handle no-key deletions later on */
4637
4638 trans = __unlink_start_trans(BTRFS_I(dir));
4639 if (IS_ERR(trans)) {
4640 err = PTR_ERR(trans);
4641 goto out_notrans;
4642 }
4643
4644 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4645 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4646 goto out;
4647 }
4648
4649 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4650 if (err)
4651 goto out;
4652
4653 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4654
4655 /* now the directory is empty */
4656 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4657 &fname.disk_name);
4658 if (!err) {
4659 btrfs_i_size_write(BTRFS_I(inode), 0);
4660 /*
4661 * Propagate the last_unlink_trans value of the deleted dir to
4662 * its parent directory. This is to prevent an unrecoverable
4663 * log tree in the case we do something like this:
4664 * 1) create dir foo
4665 * 2) create snapshot under dir foo
4666 * 3) delete the snapshot
4667 * 4) rmdir foo
4668 * 5) mkdir foo
4669 * 6) fsync foo or some file inside foo
4670 */
4671 if (last_unlink_trans >= trans->transid)
4672 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4673 }
4674 out:
4675 btrfs_end_transaction(trans);
4676 out_notrans:
4677 btrfs_btree_balance_dirty(fs_info);
4678 fscrypt_free_filename(&fname);
4679
4680 return err;
4681 }
4682
4683 /*
4684 * btrfs_truncate_block - read, zero a chunk and write a block
4685 * @inode - inode that we're zeroing
4686 * @from - the offset to start zeroing
4687 * @len - the length to zero, 0 to zero the entire range respective to the
4688 * offset
4689 * @front - zero up to the offset instead of from the offset on
4690 *
4691 * This will find the block for the "from" offset and cow the block and zero the
4692 * part we want to zero. This is used with truncate and hole punching.
4693 */
btrfs_truncate_block(struct btrfs_inode * inode,loff_t from,loff_t len,int front)4694 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4695 int front)
4696 {
4697 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4698 struct address_space *mapping = inode->vfs_inode.i_mapping;
4699 struct extent_io_tree *io_tree = &inode->io_tree;
4700 struct btrfs_ordered_extent *ordered;
4701 struct extent_state *cached_state = NULL;
4702 struct extent_changeset *data_reserved = NULL;
4703 bool only_release_metadata = false;
4704 u32 blocksize = fs_info->sectorsize;
4705 pgoff_t index = from >> PAGE_SHIFT;
4706 unsigned offset = from & (blocksize - 1);
4707 struct page *page;
4708 gfp_t mask = btrfs_alloc_write_mask(mapping);
4709 size_t write_bytes = blocksize;
4710 int ret = 0;
4711 u64 block_start;
4712 u64 block_end;
4713
4714 if (IS_ALIGNED(offset, blocksize) &&
4715 (!len || IS_ALIGNED(len, blocksize)))
4716 goto out;
4717
4718 block_start = round_down(from, blocksize);
4719 block_end = block_start + blocksize - 1;
4720
4721 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4722 blocksize, false);
4723 if (ret < 0) {
4724 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4725 /* For nocow case, no need to reserve data space */
4726 only_release_metadata = true;
4727 } else {
4728 goto out;
4729 }
4730 }
4731 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4732 if (ret < 0) {
4733 if (!only_release_metadata)
4734 btrfs_free_reserved_data_space(inode, data_reserved,
4735 block_start, blocksize);
4736 goto out;
4737 }
4738 again:
4739 page = find_or_create_page(mapping, index, mask);
4740 if (!page) {
4741 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4742 blocksize, true);
4743 btrfs_delalloc_release_extents(inode, blocksize);
4744 ret = -ENOMEM;
4745 goto out;
4746 }
4747
4748 if (!PageUptodate(page)) {
4749 ret = btrfs_read_folio(NULL, page_folio(page));
4750 lock_page(page);
4751 if (page->mapping != mapping) {
4752 unlock_page(page);
4753 put_page(page);
4754 goto again;
4755 }
4756 if (!PageUptodate(page)) {
4757 ret = -EIO;
4758 goto out_unlock;
4759 }
4760 }
4761
4762 /*
4763 * We unlock the page after the io is completed and then re-lock it
4764 * above. release_folio() could have come in between that and cleared
4765 * PagePrivate(), but left the page in the mapping. Set the page mapped
4766 * here to make sure it's properly set for the subpage stuff.
4767 */
4768 ret = set_page_extent_mapped(page);
4769 if (ret < 0)
4770 goto out_unlock;
4771
4772 wait_on_page_writeback(page);
4773
4774 lock_extent(io_tree, block_start, block_end, &cached_state);
4775
4776 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4777 if (ordered) {
4778 unlock_extent(io_tree, block_start, block_end, &cached_state);
4779 unlock_page(page);
4780 put_page(page);
4781 btrfs_start_ordered_extent(ordered);
4782 btrfs_put_ordered_extent(ordered);
4783 goto again;
4784 }
4785
4786 clear_extent_bit(&inode->io_tree, block_start, block_end,
4787 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4788 &cached_state);
4789
4790 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4791 &cached_state);
4792 if (ret) {
4793 unlock_extent(io_tree, block_start, block_end, &cached_state);
4794 goto out_unlock;
4795 }
4796
4797 if (offset != blocksize) {
4798 if (!len)
4799 len = blocksize - offset;
4800 if (front)
4801 memzero_page(page, (block_start - page_offset(page)),
4802 offset);
4803 else
4804 memzero_page(page, (block_start - page_offset(page)) + offset,
4805 len);
4806 }
4807 btrfs_page_clear_checked(fs_info, page, block_start,
4808 block_end + 1 - block_start);
4809 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4810 unlock_extent(io_tree, block_start, block_end, &cached_state);
4811
4812 if (only_release_metadata)
4813 set_extent_bit(&inode->io_tree, block_start, block_end,
4814 EXTENT_NORESERVE, NULL);
4815
4816 out_unlock:
4817 if (ret) {
4818 if (only_release_metadata)
4819 btrfs_delalloc_release_metadata(inode, blocksize, true);
4820 else
4821 btrfs_delalloc_release_space(inode, data_reserved,
4822 block_start, blocksize, true);
4823 }
4824 btrfs_delalloc_release_extents(inode, blocksize);
4825 unlock_page(page);
4826 put_page(page);
4827 out:
4828 if (only_release_metadata)
4829 btrfs_check_nocow_unlock(inode);
4830 extent_changeset_free(data_reserved);
4831 return ret;
4832 }
4833
maybe_insert_hole(struct btrfs_root * root,struct btrfs_inode * inode,u64 offset,u64 len)4834 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4835 u64 offset, u64 len)
4836 {
4837 struct btrfs_fs_info *fs_info = root->fs_info;
4838 struct btrfs_trans_handle *trans;
4839 struct btrfs_drop_extents_args drop_args = { 0 };
4840 int ret;
4841
4842 /*
4843 * If NO_HOLES is enabled, we don't need to do anything.
4844 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4845 * or btrfs_update_inode() will be called, which guarantee that the next
4846 * fsync will know this inode was changed and needs to be logged.
4847 */
4848 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4849 return 0;
4850
4851 /*
4852 * 1 - for the one we're dropping
4853 * 1 - for the one we're adding
4854 * 1 - for updating the inode.
4855 */
4856 trans = btrfs_start_transaction(root, 3);
4857 if (IS_ERR(trans))
4858 return PTR_ERR(trans);
4859
4860 drop_args.start = offset;
4861 drop_args.end = offset + len;
4862 drop_args.drop_cache = true;
4863
4864 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4865 if (ret) {
4866 btrfs_abort_transaction(trans, ret);
4867 btrfs_end_transaction(trans);
4868 return ret;
4869 }
4870
4871 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4872 if (ret) {
4873 btrfs_abort_transaction(trans, ret);
4874 } else {
4875 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4876 btrfs_update_inode(trans, root, inode);
4877 }
4878 btrfs_end_transaction(trans);
4879 return ret;
4880 }
4881
4882 /*
4883 * This function puts in dummy file extents for the area we're creating a hole
4884 * for. So if we are truncating this file to a larger size we need to insert
4885 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4886 * the range between oldsize and size
4887 */
btrfs_cont_expand(struct btrfs_inode * inode,loff_t oldsize,loff_t size)4888 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4889 {
4890 struct btrfs_root *root = inode->root;
4891 struct btrfs_fs_info *fs_info = root->fs_info;
4892 struct extent_io_tree *io_tree = &inode->io_tree;
4893 struct extent_map *em = NULL;
4894 struct extent_state *cached_state = NULL;
4895 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4896 u64 block_end = ALIGN(size, fs_info->sectorsize);
4897 u64 last_byte;
4898 u64 cur_offset;
4899 u64 hole_size;
4900 int err = 0;
4901
4902 /*
4903 * If our size started in the middle of a block we need to zero out the
4904 * rest of the block before we expand the i_size, otherwise we could
4905 * expose stale data.
4906 */
4907 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4908 if (err)
4909 return err;
4910
4911 if (size <= hole_start)
4912 return 0;
4913
4914 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4915 &cached_state);
4916 cur_offset = hole_start;
4917 while (1) {
4918 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4919 block_end - cur_offset);
4920 if (IS_ERR(em)) {
4921 err = PTR_ERR(em);
4922 em = NULL;
4923 break;
4924 }
4925 last_byte = min(extent_map_end(em), block_end);
4926 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4927 hole_size = last_byte - cur_offset;
4928
4929 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4930 struct extent_map *hole_em;
4931
4932 err = maybe_insert_hole(root, inode, cur_offset,
4933 hole_size);
4934 if (err)
4935 break;
4936
4937 err = btrfs_inode_set_file_extent_range(inode,
4938 cur_offset, hole_size);
4939 if (err)
4940 break;
4941
4942 hole_em = alloc_extent_map();
4943 if (!hole_em) {
4944 btrfs_drop_extent_map_range(inode, cur_offset,
4945 cur_offset + hole_size - 1,
4946 false);
4947 btrfs_set_inode_full_sync(inode);
4948 goto next;
4949 }
4950 hole_em->start = cur_offset;
4951 hole_em->len = hole_size;
4952 hole_em->orig_start = cur_offset;
4953
4954 hole_em->block_start = EXTENT_MAP_HOLE;
4955 hole_em->block_len = 0;
4956 hole_em->orig_block_len = 0;
4957 hole_em->ram_bytes = hole_size;
4958 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4959 hole_em->generation = fs_info->generation;
4960
4961 err = btrfs_replace_extent_map_range(inode, hole_em, true);
4962 free_extent_map(hole_em);
4963 } else {
4964 err = btrfs_inode_set_file_extent_range(inode,
4965 cur_offset, hole_size);
4966 if (err)
4967 break;
4968 }
4969 next:
4970 free_extent_map(em);
4971 em = NULL;
4972 cur_offset = last_byte;
4973 if (cur_offset >= block_end)
4974 break;
4975 }
4976 free_extent_map(em);
4977 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
4978 return err;
4979 }
4980
btrfs_setsize(struct inode * inode,struct iattr * attr)4981 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4982 {
4983 struct btrfs_root *root = BTRFS_I(inode)->root;
4984 struct btrfs_trans_handle *trans;
4985 loff_t oldsize = i_size_read(inode);
4986 loff_t newsize = attr->ia_size;
4987 int mask = attr->ia_valid;
4988 int ret;
4989
4990 /*
4991 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4992 * special case where we need to update the times despite not having
4993 * these flags set. For all other operations the VFS set these flags
4994 * explicitly if it wants a timestamp update.
4995 */
4996 if (newsize != oldsize) {
4997 inode_inc_iversion(inode);
4998 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
4999 inode->i_mtime = inode_set_ctime_current(inode);
5000 }
5001 }
5002
5003 if (newsize > oldsize) {
5004 /*
5005 * Don't do an expanding truncate while snapshotting is ongoing.
5006 * This is to ensure the snapshot captures a fully consistent
5007 * state of this file - if the snapshot captures this expanding
5008 * truncation, it must capture all writes that happened before
5009 * this truncation.
5010 */
5011 btrfs_drew_write_lock(&root->snapshot_lock);
5012 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5013 if (ret) {
5014 btrfs_drew_write_unlock(&root->snapshot_lock);
5015 return ret;
5016 }
5017
5018 trans = btrfs_start_transaction(root, 1);
5019 if (IS_ERR(trans)) {
5020 btrfs_drew_write_unlock(&root->snapshot_lock);
5021 return PTR_ERR(trans);
5022 }
5023
5024 i_size_write(inode, newsize);
5025 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5026 pagecache_isize_extended(inode, oldsize, newsize);
5027 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5028 btrfs_drew_write_unlock(&root->snapshot_lock);
5029 btrfs_end_transaction(trans);
5030 } else {
5031 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5032
5033 if (btrfs_is_zoned(fs_info)) {
5034 ret = btrfs_wait_ordered_range(inode,
5035 ALIGN(newsize, fs_info->sectorsize),
5036 (u64)-1);
5037 if (ret)
5038 return ret;
5039 }
5040
5041 /*
5042 * We're truncating a file that used to have good data down to
5043 * zero. Make sure any new writes to the file get on disk
5044 * on close.
5045 */
5046 if (newsize == 0)
5047 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5048 &BTRFS_I(inode)->runtime_flags);
5049
5050 truncate_setsize(inode, newsize);
5051
5052 inode_dio_wait(inode);
5053
5054 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5055 if (ret && inode->i_nlink) {
5056 int err;
5057
5058 /*
5059 * Truncate failed, so fix up the in-memory size. We
5060 * adjusted disk_i_size down as we removed extents, so
5061 * wait for disk_i_size to be stable and then update the
5062 * in-memory size to match.
5063 */
5064 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5065 if (err)
5066 return err;
5067 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5068 }
5069 }
5070
5071 return ret;
5072 }
5073
btrfs_setattr(struct mnt_idmap * idmap,struct dentry * dentry,struct iattr * attr)5074 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5075 struct iattr *attr)
5076 {
5077 struct inode *inode = d_inode(dentry);
5078 struct btrfs_root *root = BTRFS_I(inode)->root;
5079 int err;
5080
5081 if (btrfs_root_readonly(root))
5082 return -EROFS;
5083
5084 err = setattr_prepare(idmap, dentry, attr);
5085 if (err)
5086 return err;
5087
5088 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5089 err = btrfs_setsize(inode, attr);
5090 if (err)
5091 return err;
5092 }
5093
5094 if (attr->ia_valid) {
5095 setattr_copy(idmap, inode, attr);
5096 inode_inc_iversion(inode);
5097 err = btrfs_dirty_inode(BTRFS_I(inode));
5098
5099 if (!err && attr->ia_valid & ATTR_MODE)
5100 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5101 }
5102
5103 return err;
5104 }
5105
5106 /*
5107 * While truncating the inode pages during eviction, we get the VFS
5108 * calling btrfs_invalidate_folio() against each folio of the inode. This
5109 * is slow because the calls to btrfs_invalidate_folio() result in a
5110 * huge amount of calls to lock_extent() and clear_extent_bit(),
5111 * which keep merging and splitting extent_state structures over and over,
5112 * wasting lots of time.
5113 *
5114 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5115 * skip all those expensive operations on a per folio basis and do only
5116 * the ordered io finishing, while we release here the extent_map and
5117 * extent_state structures, without the excessive merging and splitting.
5118 */
evict_inode_truncate_pages(struct inode * inode)5119 static void evict_inode_truncate_pages(struct inode *inode)
5120 {
5121 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5122 struct rb_node *node;
5123
5124 ASSERT(inode->i_state & I_FREEING);
5125 truncate_inode_pages_final(&inode->i_data);
5126
5127 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5128
5129 /*
5130 * Keep looping until we have no more ranges in the io tree.
5131 * We can have ongoing bios started by readahead that have
5132 * their endio callback (extent_io.c:end_bio_extent_readpage)
5133 * still in progress (unlocked the pages in the bio but did not yet
5134 * unlocked the ranges in the io tree). Therefore this means some
5135 * ranges can still be locked and eviction started because before
5136 * submitting those bios, which are executed by a separate task (work
5137 * queue kthread), inode references (inode->i_count) were not taken
5138 * (which would be dropped in the end io callback of each bio).
5139 * Therefore here we effectively end up waiting for those bios and
5140 * anyone else holding locked ranges without having bumped the inode's
5141 * reference count - if we don't do it, when they access the inode's
5142 * io_tree to unlock a range it may be too late, leading to an
5143 * use-after-free issue.
5144 */
5145 spin_lock(&io_tree->lock);
5146 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5147 struct extent_state *state;
5148 struct extent_state *cached_state = NULL;
5149 u64 start;
5150 u64 end;
5151 unsigned state_flags;
5152
5153 node = rb_first(&io_tree->state);
5154 state = rb_entry(node, struct extent_state, rb_node);
5155 start = state->start;
5156 end = state->end;
5157 state_flags = state->state;
5158 spin_unlock(&io_tree->lock);
5159
5160 lock_extent(io_tree, start, end, &cached_state);
5161
5162 /*
5163 * If still has DELALLOC flag, the extent didn't reach disk,
5164 * and its reserved space won't be freed by delayed_ref.
5165 * So we need to free its reserved space here.
5166 * (Refer to comment in btrfs_invalidate_folio, case 2)
5167 *
5168 * Note, end is the bytenr of last byte, so we need + 1 here.
5169 */
5170 if (state_flags & EXTENT_DELALLOC)
5171 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5172 end - start + 1, NULL);
5173
5174 clear_extent_bit(io_tree, start, end,
5175 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5176 &cached_state);
5177
5178 cond_resched();
5179 spin_lock(&io_tree->lock);
5180 }
5181 spin_unlock(&io_tree->lock);
5182 }
5183
evict_refill_and_join(struct btrfs_root * root,struct btrfs_block_rsv * rsv)5184 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5185 struct btrfs_block_rsv *rsv)
5186 {
5187 struct btrfs_fs_info *fs_info = root->fs_info;
5188 struct btrfs_trans_handle *trans;
5189 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5190 int ret;
5191
5192 /*
5193 * Eviction should be taking place at some place safe because of our
5194 * delayed iputs. However the normal flushing code will run delayed
5195 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5196 *
5197 * We reserve the delayed_refs_extra here again because we can't use
5198 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5199 * above. We reserve our extra bit here because we generate a ton of
5200 * delayed refs activity by truncating.
5201 *
5202 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5203 * if we fail to make this reservation we can re-try without the
5204 * delayed_refs_extra so we can make some forward progress.
5205 */
5206 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5207 BTRFS_RESERVE_FLUSH_EVICT);
5208 if (ret) {
5209 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5210 BTRFS_RESERVE_FLUSH_EVICT);
5211 if (ret) {
5212 btrfs_warn(fs_info,
5213 "could not allocate space for delete; will truncate on mount");
5214 return ERR_PTR(-ENOSPC);
5215 }
5216 delayed_refs_extra = 0;
5217 }
5218
5219 trans = btrfs_join_transaction(root);
5220 if (IS_ERR(trans))
5221 return trans;
5222
5223 if (delayed_refs_extra) {
5224 trans->block_rsv = &fs_info->trans_block_rsv;
5225 trans->bytes_reserved = delayed_refs_extra;
5226 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5227 delayed_refs_extra, true);
5228 }
5229 return trans;
5230 }
5231
btrfs_evict_inode(struct inode * inode)5232 void btrfs_evict_inode(struct inode *inode)
5233 {
5234 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5235 struct btrfs_trans_handle *trans;
5236 struct btrfs_root *root = BTRFS_I(inode)->root;
5237 struct btrfs_block_rsv *rsv = NULL;
5238 int ret;
5239
5240 trace_btrfs_inode_evict(inode);
5241
5242 if (!root) {
5243 fsverity_cleanup_inode(inode);
5244 clear_inode(inode);
5245 return;
5246 }
5247
5248 evict_inode_truncate_pages(inode);
5249
5250 if (inode->i_nlink &&
5251 ((btrfs_root_refs(&root->root_item) != 0 &&
5252 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5253 btrfs_is_free_space_inode(BTRFS_I(inode))))
5254 goto out;
5255
5256 if (is_bad_inode(inode))
5257 goto out;
5258
5259 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5260 goto out;
5261
5262 if (inode->i_nlink > 0) {
5263 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5264 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5265 goto out;
5266 }
5267
5268 /*
5269 * This makes sure the inode item in tree is uptodate and the space for
5270 * the inode update is released.
5271 */
5272 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5273 if (ret)
5274 goto out;
5275
5276 /*
5277 * This drops any pending insert or delete operations we have for this
5278 * inode. We could have a delayed dir index deletion queued up, but
5279 * we're removing the inode completely so that'll be taken care of in
5280 * the truncate.
5281 */
5282 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5283
5284 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5285 if (!rsv)
5286 goto out;
5287 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5288 rsv->failfast = true;
5289
5290 btrfs_i_size_write(BTRFS_I(inode), 0);
5291
5292 while (1) {
5293 struct btrfs_truncate_control control = {
5294 .inode = BTRFS_I(inode),
5295 .ino = btrfs_ino(BTRFS_I(inode)),
5296 .new_size = 0,
5297 .min_type = 0,
5298 };
5299
5300 trans = evict_refill_and_join(root, rsv);
5301 if (IS_ERR(trans))
5302 goto out;
5303
5304 trans->block_rsv = rsv;
5305
5306 ret = btrfs_truncate_inode_items(trans, root, &control);
5307 trans->block_rsv = &fs_info->trans_block_rsv;
5308 btrfs_end_transaction(trans);
5309 /*
5310 * We have not added new delayed items for our inode after we
5311 * have flushed its delayed items, so no need to throttle on
5312 * delayed items. However we have modified extent buffers.
5313 */
5314 btrfs_btree_balance_dirty_nodelay(fs_info);
5315 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5316 goto out;
5317 else if (!ret)
5318 break;
5319 }
5320
5321 /*
5322 * Errors here aren't a big deal, it just means we leave orphan items in
5323 * the tree. They will be cleaned up on the next mount. If the inode
5324 * number gets reused, cleanup deletes the orphan item without doing
5325 * anything, and unlink reuses the existing orphan item.
5326 *
5327 * If it turns out that we are dropping too many of these, we might want
5328 * to add a mechanism for retrying these after a commit.
5329 */
5330 trans = evict_refill_and_join(root, rsv);
5331 if (!IS_ERR(trans)) {
5332 trans->block_rsv = rsv;
5333 btrfs_orphan_del(trans, BTRFS_I(inode));
5334 trans->block_rsv = &fs_info->trans_block_rsv;
5335 btrfs_end_transaction(trans);
5336 }
5337
5338 out:
5339 btrfs_free_block_rsv(fs_info, rsv);
5340 /*
5341 * If we didn't successfully delete, the orphan item will still be in
5342 * the tree and we'll retry on the next mount. Again, we might also want
5343 * to retry these periodically in the future.
5344 */
5345 btrfs_remove_delayed_node(BTRFS_I(inode));
5346 fsverity_cleanup_inode(inode);
5347 clear_inode(inode);
5348 }
5349
5350 /*
5351 * Return the key found in the dir entry in the location pointer, fill @type
5352 * with BTRFS_FT_*, and return 0.
5353 *
5354 * If no dir entries were found, returns -ENOENT.
5355 * If found a corrupted location in dir entry, returns -EUCLEAN.
5356 */
btrfs_inode_by_name(struct btrfs_inode * dir,struct dentry * dentry,struct btrfs_key * location,u8 * type)5357 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5358 struct btrfs_key *location, u8 *type)
5359 {
5360 struct btrfs_dir_item *di;
5361 struct btrfs_path *path;
5362 struct btrfs_root *root = dir->root;
5363 int ret = 0;
5364 struct fscrypt_name fname;
5365
5366 path = btrfs_alloc_path();
5367 if (!path)
5368 return -ENOMEM;
5369
5370 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5371 if (ret < 0)
5372 goto out;
5373 /*
5374 * fscrypt_setup_filename() should never return a positive value, but
5375 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5376 */
5377 ASSERT(ret == 0);
5378
5379 /* This needs to handle no-key deletions later on */
5380
5381 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5382 &fname.disk_name, 0);
5383 if (IS_ERR_OR_NULL(di)) {
5384 ret = di ? PTR_ERR(di) : -ENOENT;
5385 goto out;
5386 }
5387
5388 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5389 if (location->type != BTRFS_INODE_ITEM_KEY &&
5390 location->type != BTRFS_ROOT_ITEM_KEY) {
5391 ret = -EUCLEAN;
5392 btrfs_warn(root->fs_info,
5393 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5394 __func__, fname.disk_name.name, btrfs_ino(dir),
5395 location->objectid, location->type, location->offset);
5396 }
5397 if (!ret)
5398 *type = btrfs_dir_ftype(path->nodes[0], di);
5399 out:
5400 fscrypt_free_filename(&fname);
5401 btrfs_free_path(path);
5402 return ret;
5403 }
5404
5405 /*
5406 * when we hit a tree root in a directory, the btrfs part of the inode
5407 * needs to be changed to reflect the root directory of the tree root. This
5408 * is kind of like crossing a mount point.
5409 */
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)5410 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5411 struct btrfs_inode *dir,
5412 struct dentry *dentry,
5413 struct btrfs_key *location,
5414 struct btrfs_root **sub_root)
5415 {
5416 struct btrfs_path *path;
5417 struct btrfs_root *new_root;
5418 struct btrfs_root_ref *ref;
5419 struct extent_buffer *leaf;
5420 struct btrfs_key key;
5421 int ret;
5422 int err = 0;
5423 struct fscrypt_name fname;
5424
5425 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5426 if (ret)
5427 return ret;
5428
5429 path = btrfs_alloc_path();
5430 if (!path) {
5431 err = -ENOMEM;
5432 goto out;
5433 }
5434
5435 err = -ENOENT;
5436 key.objectid = dir->root->root_key.objectid;
5437 key.type = BTRFS_ROOT_REF_KEY;
5438 key.offset = location->objectid;
5439
5440 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5441 if (ret) {
5442 if (ret < 0)
5443 err = ret;
5444 goto out;
5445 }
5446
5447 leaf = path->nodes[0];
5448 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5449 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5450 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5451 goto out;
5452
5453 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5454 (unsigned long)(ref + 1), fname.disk_name.len);
5455 if (ret)
5456 goto out;
5457
5458 btrfs_release_path(path);
5459
5460 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5461 if (IS_ERR(new_root)) {
5462 err = PTR_ERR(new_root);
5463 goto out;
5464 }
5465
5466 *sub_root = new_root;
5467 location->objectid = btrfs_root_dirid(&new_root->root_item);
5468 location->type = BTRFS_INODE_ITEM_KEY;
5469 location->offset = 0;
5470 err = 0;
5471 out:
5472 btrfs_free_path(path);
5473 fscrypt_free_filename(&fname);
5474 return err;
5475 }
5476
inode_tree_add(struct btrfs_inode * inode)5477 static void inode_tree_add(struct btrfs_inode *inode)
5478 {
5479 struct btrfs_root *root = inode->root;
5480 struct btrfs_inode *entry;
5481 struct rb_node **p;
5482 struct rb_node *parent;
5483 struct rb_node *new = &inode->rb_node;
5484 u64 ino = btrfs_ino(inode);
5485
5486 if (inode_unhashed(&inode->vfs_inode))
5487 return;
5488 parent = NULL;
5489 spin_lock(&root->inode_lock);
5490 p = &root->inode_tree.rb_node;
5491 while (*p) {
5492 parent = *p;
5493 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5494
5495 if (ino < btrfs_ino(entry))
5496 p = &parent->rb_left;
5497 else if (ino > btrfs_ino(entry))
5498 p = &parent->rb_right;
5499 else {
5500 WARN_ON(!(entry->vfs_inode.i_state &
5501 (I_WILL_FREE | I_FREEING)));
5502 rb_replace_node(parent, new, &root->inode_tree);
5503 RB_CLEAR_NODE(parent);
5504 spin_unlock(&root->inode_lock);
5505 return;
5506 }
5507 }
5508 rb_link_node(new, parent, p);
5509 rb_insert_color(new, &root->inode_tree);
5510 spin_unlock(&root->inode_lock);
5511 }
5512
inode_tree_del(struct btrfs_inode * inode)5513 static void inode_tree_del(struct btrfs_inode *inode)
5514 {
5515 struct btrfs_root *root = inode->root;
5516 int empty = 0;
5517
5518 spin_lock(&root->inode_lock);
5519 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5520 rb_erase(&inode->rb_node, &root->inode_tree);
5521 RB_CLEAR_NODE(&inode->rb_node);
5522 empty = RB_EMPTY_ROOT(&root->inode_tree);
5523 }
5524 spin_unlock(&root->inode_lock);
5525
5526 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5527 spin_lock(&root->inode_lock);
5528 empty = RB_EMPTY_ROOT(&root->inode_tree);
5529 spin_unlock(&root->inode_lock);
5530 if (empty)
5531 btrfs_add_dead_root(root);
5532 }
5533 }
5534
5535
btrfs_init_locked_inode(struct inode * inode,void * p)5536 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5537 {
5538 struct btrfs_iget_args *args = p;
5539
5540 inode->i_ino = args->ino;
5541 BTRFS_I(inode)->location.objectid = args->ino;
5542 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5543 BTRFS_I(inode)->location.offset = 0;
5544 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5545 BUG_ON(args->root && !BTRFS_I(inode)->root);
5546
5547 if (args->root && args->root == args->root->fs_info->tree_root &&
5548 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5549 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5550 &BTRFS_I(inode)->runtime_flags);
5551 return 0;
5552 }
5553
btrfs_find_actor(struct inode * inode,void * opaque)5554 static int btrfs_find_actor(struct inode *inode, void *opaque)
5555 {
5556 struct btrfs_iget_args *args = opaque;
5557
5558 return args->ino == BTRFS_I(inode)->location.objectid &&
5559 args->root == BTRFS_I(inode)->root;
5560 }
5561
btrfs_iget_locked(struct super_block * s,u64 ino,struct btrfs_root * root)5562 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5563 struct btrfs_root *root)
5564 {
5565 struct inode *inode;
5566 struct btrfs_iget_args args;
5567 unsigned long hashval = btrfs_inode_hash(ino, root);
5568
5569 args.ino = ino;
5570 args.root = root;
5571
5572 inode = iget5_locked(s, hashval, btrfs_find_actor,
5573 btrfs_init_locked_inode,
5574 (void *)&args);
5575 return inode;
5576 }
5577
5578 /*
5579 * Get an inode object given its inode number and corresponding root.
5580 * Path can be preallocated to prevent recursing back to iget through
5581 * allocator. NULL is also valid but may require an additional allocation
5582 * later.
5583 */
btrfs_iget_path(struct super_block * s,u64 ino,struct btrfs_root * root,struct btrfs_path * path)5584 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5585 struct btrfs_root *root, struct btrfs_path *path)
5586 {
5587 struct inode *inode;
5588
5589 inode = btrfs_iget_locked(s, ino, root);
5590 if (!inode)
5591 return ERR_PTR(-ENOMEM);
5592
5593 if (inode->i_state & I_NEW) {
5594 int ret;
5595
5596 ret = btrfs_read_locked_inode(inode, path);
5597 if (!ret) {
5598 inode_tree_add(BTRFS_I(inode));
5599 unlock_new_inode(inode);
5600 } else {
5601 iget_failed(inode);
5602 /*
5603 * ret > 0 can come from btrfs_search_slot called by
5604 * btrfs_read_locked_inode, this means the inode item
5605 * was not found.
5606 */
5607 if (ret > 0)
5608 ret = -ENOENT;
5609 inode = ERR_PTR(ret);
5610 }
5611 }
5612
5613 return inode;
5614 }
5615
btrfs_iget(struct super_block * s,u64 ino,struct btrfs_root * root)5616 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5617 {
5618 return btrfs_iget_path(s, ino, root, NULL);
5619 }
5620
new_simple_dir(struct inode * dir,struct btrfs_key * key,struct btrfs_root * root)5621 static struct inode *new_simple_dir(struct inode *dir,
5622 struct btrfs_key *key,
5623 struct btrfs_root *root)
5624 {
5625 struct inode *inode = new_inode(dir->i_sb);
5626
5627 if (!inode)
5628 return ERR_PTR(-ENOMEM);
5629
5630 BTRFS_I(inode)->root = btrfs_grab_root(root);
5631 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5632 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5633
5634 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5635 /*
5636 * We only need lookup, the rest is read-only and there's no inode
5637 * associated with the dentry
5638 */
5639 inode->i_op = &simple_dir_inode_operations;
5640 inode->i_opflags &= ~IOP_XATTR;
5641 inode->i_fop = &simple_dir_operations;
5642 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5643 inode->i_mtime = inode_set_ctime_current(inode);
5644 inode->i_atime = dir->i_atime;
5645 BTRFS_I(inode)->i_otime = inode->i_mtime;
5646 inode->i_uid = dir->i_uid;
5647 inode->i_gid = dir->i_gid;
5648
5649 return inode;
5650 }
5651
5652 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5653 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5654 static_assert(BTRFS_FT_DIR == FT_DIR);
5655 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5656 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5657 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5658 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5659 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5660
btrfs_inode_type(struct inode * inode)5661 static inline u8 btrfs_inode_type(struct inode *inode)
5662 {
5663 return fs_umode_to_ftype(inode->i_mode);
5664 }
5665
btrfs_lookup_dentry(struct inode * dir,struct dentry * dentry)5666 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5667 {
5668 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5669 struct inode *inode;
5670 struct btrfs_root *root = BTRFS_I(dir)->root;
5671 struct btrfs_root *sub_root = root;
5672 struct btrfs_key location = { 0 };
5673 u8 di_type = 0;
5674 int ret = 0;
5675
5676 if (dentry->d_name.len > BTRFS_NAME_LEN)
5677 return ERR_PTR(-ENAMETOOLONG);
5678
5679 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5680 if (ret < 0)
5681 return ERR_PTR(ret);
5682
5683 if (location.type == BTRFS_INODE_ITEM_KEY) {
5684 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5685 if (IS_ERR(inode))
5686 return inode;
5687
5688 /* Do extra check against inode mode with di_type */
5689 if (btrfs_inode_type(inode) != di_type) {
5690 btrfs_crit(fs_info,
5691 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5692 inode->i_mode, btrfs_inode_type(inode),
5693 di_type);
5694 iput(inode);
5695 return ERR_PTR(-EUCLEAN);
5696 }
5697 return inode;
5698 }
5699
5700 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5701 &location, &sub_root);
5702 if (ret < 0) {
5703 if (ret != -ENOENT)
5704 inode = ERR_PTR(ret);
5705 else
5706 inode = new_simple_dir(dir, &location, root);
5707 } else {
5708 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5709 btrfs_put_root(sub_root);
5710
5711 if (IS_ERR(inode))
5712 return inode;
5713
5714 down_read(&fs_info->cleanup_work_sem);
5715 if (!sb_rdonly(inode->i_sb))
5716 ret = btrfs_orphan_cleanup(sub_root);
5717 up_read(&fs_info->cleanup_work_sem);
5718 if (ret) {
5719 iput(inode);
5720 inode = ERR_PTR(ret);
5721 }
5722 }
5723
5724 return inode;
5725 }
5726
btrfs_dentry_delete(const struct dentry * dentry)5727 static int btrfs_dentry_delete(const struct dentry *dentry)
5728 {
5729 struct btrfs_root *root;
5730 struct inode *inode = d_inode(dentry);
5731
5732 if (!inode && !IS_ROOT(dentry))
5733 inode = d_inode(dentry->d_parent);
5734
5735 if (inode) {
5736 root = BTRFS_I(inode)->root;
5737 if (btrfs_root_refs(&root->root_item) == 0)
5738 return 1;
5739
5740 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5741 return 1;
5742 }
5743 return 0;
5744 }
5745
btrfs_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)5746 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5747 unsigned int flags)
5748 {
5749 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5750
5751 if (inode == ERR_PTR(-ENOENT))
5752 inode = NULL;
5753 return d_splice_alias(inode, dentry);
5754 }
5755
5756 /*
5757 * Find the highest existing sequence number in a directory and then set the
5758 * in-memory index_cnt variable to the first free sequence number.
5759 */
btrfs_set_inode_index_count(struct btrfs_inode * inode)5760 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5761 {
5762 struct btrfs_root *root = inode->root;
5763 struct btrfs_key key, found_key;
5764 struct btrfs_path *path;
5765 struct extent_buffer *leaf;
5766 int ret;
5767
5768 key.objectid = btrfs_ino(inode);
5769 key.type = BTRFS_DIR_INDEX_KEY;
5770 key.offset = (u64)-1;
5771
5772 path = btrfs_alloc_path();
5773 if (!path)
5774 return -ENOMEM;
5775
5776 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5777 if (ret < 0)
5778 goto out;
5779 /* FIXME: we should be able to handle this */
5780 if (ret == 0)
5781 goto out;
5782 ret = 0;
5783
5784 if (path->slots[0] == 0) {
5785 inode->index_cnt = BTRFS_DIR_START_INDEX;
5786 goto out;
5787 }
5788
5789 path->slots[0]--;
5790
5791 leaf = path->nodes[0];
5792 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5793
5794 if (found_key.objectid != btrfs_ino(inode) ||
5795 found_key.type != BTRFS_DIR_INDEX_KEY) {
5796 inode->index_cnt = BTRFS_DIR_START_INDEX;
5797 goto out;
5798 }
5799
5800 inode->index_cnt = found_key.offset + 1;
5801 out:
5802 btrfs_free_path(path);
5803 return ret;
5804 }
5805
btrfs_get_dir_last_index(struct btrfs_inode * dir,u64 * index)5806 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5807 {
5808 int ret = 0;
5809
5810 btrfs_inode_lock(dir, 0);
5811 if (dir->index_cnt == (u64)-1) {
5812 ret = btrfs_inode_delayed_dir_index_count(dir);
5813 if (ret) {
5814 ret = btrfs_set_inode_index_count(dir);
5815 if (ret)
5816 goto out;
5817 }
5818 }
5819
5820 /* index_cnt is the index number of next new entry, so decrement it. */
5821 *index = dir->index_cnt - 1;
5822 out:
5823 btrfs_inode_unlock(dir, 0);
5824
5825 return ret;
5826 }
5827
5828 /*
5829 * All this infrastructure exists because dir_emit can fault, and we are holding
5830 * the tree lock when doing readdir. For now just allocate a buffer and copy
5831 * our information into that, and then dir_emit from the buffer. This is
5832 * similar to what NFS does, only we don't keep the buffer around in pagecache
5833 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5834 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5835 * tree lock.
5836 */
btrfs_opendir(struct inode * inode,struct file * file)5837 static int btrfs_opendir(struct inode *inode, struct file *file)
5838 {
5839 struct btrfs_file_private *private;
5840 u64 last_index;
5841 int ret;
5842
5843 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5844 if (ret)
5845 return ret;
5846
5847 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5848 if (!private)
5849 return -ENOMEM;
5850 private->last_index = last_index;
5851 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5852 if (!private->filldir_buf) {
5853 kfree(private);
5854 return -ENOMEM;
5855 }
5856 file->private_data = private;
5857 return 0;
5858 }
5859
btrfs_dir_llseek(struct file * file,loff_t offset,int whence)5860 static loff_t btrfs_dir_llseek(struct file *file, loff_t offset, int whence)
5861 {
5862 struct btrfs_file_private *private = file->private_data;
5863 int ret;
5864
5865 ret = btrfs_get_dir_last_index(BTRFS_I(file_inode(file)),
5866 &private->last_index);
5867 if (ret)
5868 return ret;
5869
5870 return generic_file_llseek(file, offset, whence);
5871 }
5872
5873 struct dir_entry {
5874 u64 ino;
5875 u64 offset;
5876 unsigned type;
5877 int name_len;
5878 };
5879
btrfs_filldir(void * addr,int entries,struct dir_context * ctx)5880 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5881 {
5882 while (entries--) {
5883 struct dir_entry *entry = addr;
5884 char *name = (char *)(entry + 1);
5885
5886 ctx->pos = get_unaligned(&entry->offset);
5887 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5888 get_unaligned(&entry->ino),
5889 get_unaligned(&entry->type)))
5890 return 1;
5891 addr += sizeof(struct dir_entry) +
5892 get_unaligned(&entry->name_len);
5893 ctx->pos++;
5894 }
5895 return 0;
5896 }
5897
btrfs_real_readdir(struct file * file,struct dir_context * ctx)5898 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5899 {
5900 struct inode *inode = file_inode(file);
5901 struct btrfs_root *root = BTRFS_I(inode)->root;
5902 struct btrfs_file_private *private = file->private_data;
5903 struct btrfs_dir_item *di;
5904 struct btrfs_key key;
5905 struct btrfs_key found_key;
5906 struct btrfs_path *path;
5907 void *addr;
5908 LIST_HEAD(ins_list);
5909 LIST_HEAD(del_list);
5910 int ret;
5911 char *name_ptr;
5912 int name_len;
5913 int entries = 0;
5914 int total_len = 0;
5915 bool put = false;
5916 struct btrfs_key location;
5917
5918 if (!dir_emit_dots(file, ctx))
5919 return 0;
5920
5921 path = btrfs_alloc_path();
5922 if (!path)
5923 return -ENOMEM;
5924
5925 addr = private->filldir_buf;
5926 path->reada = READA_FORWARD;
5927
5928 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
5929 &ins_list, &del_list);
5930
5931 again:
5932 key.type = BTRFS_DIR_INDEX_KEY;
5933 key.offset = ctx->pos;
5934 key.objectid = btrfs_ino(BTRFS_I(inode));
5935
5936 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5937 struct dir_entry *entry;
5938 struct extent_buffer *leaf = path->nodes[0];
5939 u8 ftype;
5940
5941 if (found_key.objectid != key.objectid)
5942 break;
5943 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5944 break;
5945 if (found_key.offset < ctx->pos)
5946 continue;
5947 if (found_key.offset > private->last_index)
5948 break;
5949 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5950 continue;
5951 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5952 name_len = btrfs_dir_name_len(leaf, di);
5953 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5954 PAGE_SIZE) {
5955 btrfs_release_path(path);
5956 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5957 if (ret)
5958 goto nopos;
5959 addr = private->filldir_buf;
5960 entries = 0;
5961 total_len = 0;
5962 goto again;
5963 }
5964
5965 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
5966 entry = addr;
5967 name_ptr = (char *)(entry + 1);
5968 read_extent_buffer(leaf, name_ptr,
5969 (unsigned long)(di + 1), name_len);
5970 put_unaligned(name_len, &entry->name_len);
5971 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
5972 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5973 put_unaligned(location.objectid, &entry->ino);
5974 put_unaligned(found_key.offset, &entry->offset);
5975 entries++;
5976 addr += sizeof(struct dir_entry) + name_len;
5977 total_len += sizeof(struct dir_entry) + name_len;
5978 }
5979 /* Catch error encountered during iteration */
5980 if (ret < 0)
5981 goto err;
5982
5983 btrfs_release_path(path);
5984
5985 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5986 if (ret)
5987 goto nopos;
5988
5989 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5990 if (ret)
5991 goto nopos;
5992
5993 /*
5994 * Stop new entries from being returned after we return the last
5995 * entry.
5996 *
5997 * New directory entries are assigned a strictly increasing
5998 * offset. This means that new entries created during readdir
5999 * are *guaranteed* to be seen in the future by that readdir.
6000 * This has broken buggy programs which operate on names as
6001 * they're returned by readdir. Until we re-use freed offsets
6002 * we have this hack to stop new entries from being returned
6003 * under the assumption that they'll never reach this huge
6004 * offset.
6005 *
6006 * This is being careful not to overflow 32bit loff_t unless the
6007 * last entry requires it because doing so has broken 32bit apps
6008 * in the past.
6009 */
6010 if (ctx->pos >= INT_MAX)
6011 ctx->pos = LLONG_MAX;
6012 else
6013 ctx->pos = INT_MAX;
6014 nopos:
6015 ret = 0;
6016 err:
6017 if (put)
6018 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6019 btrfs_free_path(path);
6020 return ret;
6021 }
6022
6023 /*
6024 * This is somewhat expensive, updating the tree every time the
6025 * inode changes. But, it is most likely to find the inode in cache.
6026 * FIXME, needs more benchmarking...there are no reasons other than performance
6027 * to keep or drop this code.
6028 */
btrfs_dirty_inode(struct btrfs_inode * inode)6029 static int btrfs_dirty_inode(struct btrfs_inode *inode)
6030 {
6031 struct btrfs_root *root = inode->root;
6032 struct btrfs_fs_info *fs_info = root->fs_info;
6033 struct btrfs_trans_handle *trans;
6034 int ret;
6035
6036 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6037 return 0;
6038
6039 trans = btrfs_join_transaction(root);
6040 if (IS_ERR(trans))
6041 return PTR_ERR(trans);
6042
6043 ret = btrfs_update_inode(trans, root, inode);
6044 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6045 /* whoops, lets try again with the full transaction */
6046 btrfs_end_transaction(trans);
6047 trans = btrfs_start_transaction(root, 1);
6048 if (IS_ERR(trans))
6049 return PTR_ERR(trans);
6050
6051 ret = btrfs_update_inode(trans, root, inode);
6052 }
6053 btrfs_end_transaction(trans);
6054 if (inode->delayed_node)
6055 btrfs_balance_delayed_items(fs_info);
6056
6057 return ret;
6058 }
6059
6060 /*
6061 * This is a copy of file_update_time. We need this so we can return error on
6062 * ENOSPC for updating the inode in the case of file write and mmap writes.
6063 */
btrfs_update_time(struct inode * inode,int flags)6064 static int btrfs_update_time(struct inode *inode, int flags)
6065 {
6066 struct btrfs_root *root = BTRFS_I(inode)->root;
6067 bool dirty = flags & ~S_VERSION;
6068
6069 if (btrfs_root_readonly(root))
6070 return -EROFS;
6071
6072 dirty = inode_update_timestamps(inode, flags);
6073 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6074 }
6075
6076 /*
6077 * helper to find a free sequence number in a given directory. This current
6078 * code is very simple, later versions will do smarter things in the btree
6079 */
btrfs_set_inode_index(struct btrfs_inode * dir,u64 * index)6080 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6081 {
6082 int ret = 0;
6083
6084 if (dir->index_cnt == (u64)-1) {
6085 ret = btrfs_inode_delayed_dir_index_count(dir);
6086 if (ret) {
6087 ret = btrfs_set_inode_index_count(dir);
6088 if (ret)
6089 return ret;
6090 }
6091 }
6092
6093 *index = dir->index_cnt;
6094 dir->index_cnt++;
6095
6096 return ret;
6097 }
6098
btrfs_insert_inode_locked(struct inode * inode)6099 static int btrfs_insert_inode_locked(struct inode *inode)
6100 {
6101 struct btrfs_iget_args args;
6102
6103 args.ino = BTRFS_I(inode)->location.objectid;
6104 args.root = BTRFS_I(inode)->root;
6105
6106 return insert_inode_locked4(inode,
6107 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6108 btrfs_find_actor, &args);
6109 }
6110
btrfs_new_inode_prepare(struct btrfs_new_inode_args * args,unsigned int * trans_num_items)6111 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6112 unsigned int *trans_num_items)
6113 {
6114 struct inode *dir = args->dir;
6115 struct inode *inode = args->inode;
6116 int ret;
6117
6118 if (!args->orphan) {
6119 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6120 &args->fname);
6121 if (ret)
6122 return ret;
6123 }
6124
6125 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6126 if (ret) {
6127 fscrypt_free_filename(&args->fname);
6128 return ret;
6129 }
6130
6131 /* 1 to add inode item */
6132 *trans_num_items = 1;
6133 /* 1 to add compression property */
6134 if (BTRFS_I(dir)->prop_compress)
6135 (*trans_num_items)++;
6136 /* 1 to add default ACL xattr */
6137 if (args->default_acl)
6138 (*trans_num_items)++;
6139 /* 1 to add access ACL xattr */
6140 if (args->acl)
6141 (*trans_num_items)++;
6142 #ifdef CONFIG_SECURITY
6143 /* 1 to add LSM xattr */
6144 if (dir->i_security)
6145 (*trans_num_items)++;
6146 #endif
6147 if (args->orphan) {
6148 /* 1 to add orphan item */
6149 (*trans_num_items)++;
6150 } else {
6151 /*
6152 * 1 to add dir item
6153 * 1 to add dir index
6154 * 1 to update parent inode item
6155 *
6156 * No need for 1 unit for the inode ref item because it is
6157 * inserted in a batch together with the inode item at
6158 * btrfs_create_new_inode().
6159 */
6160 *trans_num_items += 3;
6161 }
6162 return 0;
6163 }
6164
btrfs_new_inode_args_destroy(struct btrfs_new_inode_args * args)6165 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6166 {
6167 posix_acl_release(args->acl);
6168 posix_acl_release(args->default_acl);
6169 fscrypt_free_filename(&args->fname);
6170 }
6171
6172 /*
6173 * Inherit flags from the parent inode.
6174 *
6175 * Currently only the compression flags and the cow flags are inherited.
6176 */
btrfs_inherit_iflags(struct btrfs_inode * inode,struct btrfs_inode * dir)6177 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6178 {
6179 unsigned int flags;
6180
6181 flags = dir->flags;
6182
6183 if (flags & BTRFS_INODE_NOCOMPRESS) {
6184 inode->flags &= ~BTRFS_INODE_COMPRESS;
6185 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6186 } else if (flags & BTRFS_INODE_COMPRESS) {
6187 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6188 inode->flags |= BTRFS_INODE_COMPRESS;
6189 }
6190
6191 if (flags & BTRFS_INODE_NODATACOW) {
6192 inode->flags |= BTRFS_INODE_NODATACOW;
6193 if (S_ISREG(inode->vfs_inode.i_mode))
6194 inode->flags |= BTRFS_INODE_NODATASUM;
6195 }
6196
6197 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6198 }
6199
btrfs_create_new_inode(struct btrfs_trans_handle * trans,struct btrfs_new_inode_args * args)6200 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6201 struct btrfs_new_inode_args *args)
6202 {
6203 struct inode *dir = args->dir;
6204 struct inode *inode = args->inode;
6205 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6206 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6207 struct btrfs_root *root;
6208 struct btrfs_inode_item *inode_item;
6209 struct btrfs_key *location;
6210 struct btrfs_path *path;
6211 u64 objectid;
6212 struct btrfs_inode_ref *ref;
6213 struct btrfs_key key[2];
6214 u32 sizes[2];
6215 struct btrfs_item_batch batch;
6216 unsigned long ptr;
6217 int ret;
6218
6219 path = btrfs_alloc_path();
6220 if (!path)
6221 return -ENOMEM;
6222
6223 if (!args->subvol)
6224 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6225 root = BTRFS_I(inode)->root;
6226
6227 ret = btrfs_get_free_objectid(root, &objectid);
6228 if (ret)
6229 goto out;
6230 inode->i_ino = objectid;
6231
6232 if (args->orphan) {
6233 /*
6234 * O_TMPFILE, set link count to 0, so that after this point, we
6235 * fill in an inode item with the correct link count.
6236 */
6237 set_nlink(inode, 0);
6238 } else {
6239 trace_btrfs_inode_request(dir);
6240
6241 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6242 if (ret)
6243 goto out;
6244 }
6245 /* index_cnt is ignored for everything but a dir. */
6246 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6247 BTRFS_I(inode)->generation = trans->transid;
6248 inode->i_generation = BTRFS_I(inode)->generation;
6249
6250 /*
6251 * Subvolumes don't inherit flags from their parent directory.
6252 * Originally this was probably by accident, but we probably can't
6253 * change it now without compatibility issues.
6254 */
6255 if (!args->subvol)
6256 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6257
6258 if (S_ISREG(inode->i_mode)) {
6259 if (btrfs_test_opt(fs_info, NODATASUM))
6260 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6261 if (btrfs_test_opt(fs_info, NODATACOW))
6262 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6263 BTRFS_INODE_NODATASUM;
6264 }
6265
6266 location = &BTRFS_I(inode)->location;
6267 location->objectid = objectid;
6268 location->offset = 0;
6269 location->type = BTRFS_INODE_ITEM_KEY;
6270
6271 ret = btrfs_insert_inode_locked(inode);
6272 if (ret < 0) {
6273 if (!args->orphan)
6274 BTRFS_I(dir)->index_cnt--;
6275 goto out;
6276 }
6277
6278 /*
6279 * We could have gotten an inode number from somebody who was fsynced
6280 * and then removed in this same transaction, so let's just set full
6281 * sync since it will be a full sync anyway and this will blow away the
6282 * old info in the log.
6283 */
6284 btrfs_set_inode_full_sync(BTRFS_I(inode));
6285
6286 key[0].objectid = objectid;
6287 key[0].type = BTRFS_INODE_ITEM_KEY;
6288 key[0].offset = 0;
6289
6290 sizes[0] = sizeof(struct btrfs_inode_item);
6291
6292 if (!args->orphan) {
6293 /*
6294 * Start new inodes with an inode_ref. This is slightly more
6295 * efficient for small numbers of hard links since they will
6296 * be packed into one item. Extended refs will kick in if we
6297 * add more hard links than can fit in the ref item.
6298 */
6299 key[1].objectid = objectid;
6300 key[1].type = BTRFS_INODE_REF_KEY;
6301 if (args->subvol) {
6302 key[1].offset = objectid;
6303 sizes[1] = 2 + sizeof(*ref);
6304 } else {
6305 key[1].offset = btrfs_ino(BTRFS_I(dir));
6306 sizes[1] = name->len + sizeof(*ref);
6307 }
6308 }
6309
6310 batch.keys = &key[0];
6311 batch.data_sizes = &sizes[0];
6312 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6313 batch.nr = args->orphan ? 1 : 2;
6314 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6315 if (ret != 0) {
6316 btrfs_abort_transaction(trans, ret);
6317 goto discard;
6318 }
6319
6320 inode->i_mtime = inode_set_ctime_current(inode);
6321 inode->i_atime = inode->i_mtime;
6322 BTRFS_I(inode)->i_otime = inode->i_mtime;
6323
6324 /*
6325 * We're going to fill the inode item now, so at this point the inode
6326 * must be fully initialized.
6327 */
6328
6329 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6330 struct btrfs_inode_item);
6331 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6332 sizeof(*inode_item));
6333 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6334
6335 if (!args->orphan) {
6336 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6337 struct btrfs_inode_ref);
6338 ptr = (unsigned long)(ref + 1);
6339 if (args->subvol) {
6340 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6341 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6342 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6343 } else {
6344 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6345 name->len);
6346 btrfs_set_inode_ref_index(path->nodes[0], ref,
6347 BTRFS_I(inode)->dir_index);
6348 write_extent_buffer(path->nodes[0], name->name, ptr,
6349 name->len);
6350 }
6351 }
6352
6353 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
6354 /*
6355 * We don't need the path anymore, plus inheriting properties, adding
6356 * ACLs, security xattrs, orphan item or adding the link, will result in
6357 * allocating yet another path. So just free our path.
6358 */
6359 btrfs_free_path(path);
6360 path = NULL;
6361
6362 if (args->subvol) {
6363 struct inode *parent;
6364
6365 /*
6366 * Subvolumes inherit properties from their parent subvolume,
6367 * not the directory they were created in.
6368 */
6369 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6370 BTRFS_I(dir)->root);
6371 if (IS_ERR(parent)) {
6372 ret = PTR_ERR(parent);
6373 } else {
6374 ret = btrfs_inode_inherit_props(trans, inode, parent);
6375 iput(parent);
6376 }
6377 } else {
6378 ret = btrfs_inode_inherit_props(trans, inode, dir);
6379 }
6380 if (ret) {
6381 btrfs_err(fs_info,
6382 "error inheriting props for ino %llu (root %llu): %d",
6383 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6384 ret);
6385 }
6386
6387 /*
6388 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6389 * probably a bug.
6390 */
6391 if (!args->subvol) {
6392 ret = btrfs_init_inode_security(trans, args);
6393 if (ret) {
6394 btrfs_abort_transaction(trans, ret);
6395 goto discard;
6396 }
6397 }
6398
6399 inode_tree_add(BTRFS_I(inode));
6400
6401 trace_btrfs_inode_new(inode);
6402 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6403
6404 btrfs_update_root_times(trans, root);
6405
6406 if (args->orphan) {
6407 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6408 } else {
6409 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6410 0, BTRFS_I(inode)->dir_index);
6411 }
6412 if (ret) {
6413 btrfs_abort_transaction(trans, ret);
6414 goto discard;
6415 }
6416
6417 return 0;
6418
6419 discard:
6420 /*
6421 * discard_new_inode() calls iput(), but the caller owns the reference
6422 * to the inode.
6423 */
6424 ihold(inode);
6425 discard_new_inode(inode);
6426 out:
6427 btrfs_free_path(path);
6428 return ret;
6429 }
6430
6431 /*
6432 * utility function to add 'inode' into 'parent_inode' with
6433 * a give name and a given sequence number.
6434 * if 'add_backref' is true, also insert a backref from the
6435 * inode to the parent directory.
6436 */
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)6437 int btrfs_add_link(struct btrfs_trans_handle *trans,
6438 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6439 const struct fscrypt_str *name, int add_backref, u64 index)
6440 {
6441 int ret = 0;
6442 struct btrfs_key key;
6443 struct btrfs_root *root = parent_inode->root;
6444 u64 ino = btrfs_ino(inode);
6445 u64 parent_ino = btrfs_ino(parent_inode);
6446
6447 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6448 memcpy(&key, &inode->root->root_key, sizeof(key));
6449 } else {
6450 key.objectid = ino;
6451 key.type = BTRFS_INODE_ITEM_KEY;
6452 key.offset = 0;
6453 }
6454
6455 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6456 ret = btrfs_add_root_ref(trans, key.objectid,
6457 root->root_key.objectid, parent_ino,
6458 index, name);
6459 } else if (add_backref) {
6460 ret = btrfs_insert_inode_ref(trans, root, name,
6461 ino, parent_ino, index);
6462 }
6463
6464 /* Nothing to clean up yet */
6465 if (ret)
6466 return ret;
6467
6468 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6469 btrfs_inode_type(&inode->vfs_inode), index);
6470 if (ret == -EEXIST || ret == -EOVERFLOW)
6471 goto fail_dir_item;
6472 else if (ret) {
6473 btrfs_abort_transaction(trans, ret);
6474 return ret;
6475 }
6476
6477 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6478 name->len * 2);
6479 inode_inc_iversion(&parent_inode->vfs_inode);
6480 /*
6481 * If we are replaying a log tree, we do not want to update the mtime
6482 * and ctime of the parent directory with the current time, since the
6483 * log replay procedure is responsible for setting them to their correct
6484 * values (the ones it had when the fsync was done).
6485 */
6486 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags))
6487 parent_inode->vfs_inode.i_mtime =
6488 inode_set_ctime_current(&parent_inode->vfs_inode);
6489
6490 ret = btrfs_update_inode(trans, root, parent_inode);
6491 if (ret)
6492 btrfs_abort_transaction(trans, ret);
6493 return ret;
6494
6495 fail_dir_item:
6496 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6497 u64 local_index;
6498 int err;
6499 err = btrfs_del_root_ref(trans, key.objectid,
6500 root->root_key.objectid, parent_ino,
6501 &local_index, name);
6502 if (err)
6503 btrfs_abort_transaction(trans, err);
6504 } else if (add_backref) {
6505 u64 local_index;
6506 int err;
6507
6508 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6509 &local_index);
6510 if (err)
6511 btrfs_abort_transaction(trans, err);
6512 }
6513
6514 /* Return the original error code */
6515 return ret;
6516 }
6517
btrfs_create_common(struct inode * dir,struct dentry * dentry,struct inode * inode)6518 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6519 struct inode *inode)
6520 {
6521 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6522 struct btrfs_root *root = BTRFS_I(dir)->root;
6523 struct btrfs_new_inode_args new_inode_args = {
6524 .dir = dir,
6525 .dentry = dentry,
6526 .inode = inode,
6527 };
6528 unsigned int trans_num_items;
6529 struct btrfs_trans_handle *trans;
6530 int err;
6531
6532 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6533 if (err)
6534 goto out_inode;
6535
6536 trans = btrfs_start_transaction(root, trans_num_items);
6537 if (IS_ERR(trans)) {
6538 err = PTR_ERR(trans);
6539 goto out_new_inode_args;
6540 }
6541
6542 err = btrfs_create_new_inode(trans, &new_inode_args);
6543 if (!err)
6544 d_instantiate_new(dentry, inode);
6545
6546 btrfs_end_transaction(trans);
6547 btrfs_btree_balance_dirty(fs_info);
6548 out_new_inode_args:
6549 btrfs_new_inode_args_destroy(&new_inode_args);
6550 out_inode:
6551 if (err)
6552 iput(inode);
6553 return err;
6554 }
6555
btrfs_mknod(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode,dev_t rdev)6556 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6557 struct dentry *dentry, umode_t mode, dev_t rdev)
6558 {
6559 struct inode *inode;
6560
6561 inode = new_inode(dir->i_sb);
6562 if (!inode)
6563 return -ENOMEM;
6564 inode_init_owner(idmap, inode, dir, mode);
6565 inode->i_op = &btrfs_special_inode_operations;
6566 init_special_inode(inode, inode->i_mode, rdev);
6567 return btrfs_create_common(dir, dentry, inode);
6568 }
6569
btrfs_create(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode,bool excl)6570 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6571 struct dentry *dentry, umode_t mode, bool excl)
6572 {
6573 struct inode *inode;
6574
6575 inode = new_inode(dir->i_sb);
6576 if (!inode)
6577 return -ENOMEM;
6578 inode_init_owner(idmap, inode, dir, mode);
6579 inode->i_fop = &btrfs_file_operations;
6580 inode->i_op = &btrfs_file_inode_operations;
6581 inode->i_mapping->a_ops = &btrfs_aops;
6582 return btrfs_create_common(dir, dentry, inode);
6583 }
6584
btrfs_link(struct dentry * old_dentry,struct inode * dir,struct dentry * dentry)6585 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6586 struct dentry *dentry)
6587 {
6588 struct btrfs_trans_handle *trans = NULL;
6589 struct btrfs_root *root = BTRFS_I(dir)->root;
6590 struct inode *inode = d_inode(old_dentry);
6591 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6592 struct fscrypt_name fname;
6593 u64 index;
6594 int err;
6595 int drop_inode = 0;
6596
6597 /* do not allow sys_link's with other subvols of the same device */
6598 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6599 return -EXDEV;
6600
6601 if (inode->i_nlink >= BTRFS_LINK_MAX)
6602 return -EMLINK;
6603
6604 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6605 if (err)
6606 goto fail;
6607
6608 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6609 if (err)
6610 goto fail;
6611
6612 /*
6613 * 2 items for inode and inode ref
6614 * 2 items for dir items
6615 * 1 item for parent inode
6616 * 1 item for orphan item deletion if O_TMPFILE
6617 */
6618 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6619 if (IS_ERR(trans)) {
6620 err = PTR_ERR(trans);
6621 trans = NULL;
6622 goto fail;
6623 }
6624
6625 /* There are several dir indexes for this inode, clear the cache. */
6626 BTRFS_I(inode)->dir_index = 0ULL;
6627 inc_nlink(inode);
6628 inode_inc_iversion(inode);
6629 inode_set_ctime_current(inode);
6630 ihold(inode);
6631 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6632
6633 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6634 &fname.disk_name, 1, index);
6635
6636 if (err) {
6637 drop_inode = 1;
6638 } else {
6639 struct dentry *parent = dentry->d_parent;
6640
6641 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6642 if (err)
6643 goto fail;
6644 if (inode->i_nlink == 1) {
6645 /*
6646 * If new hard link count is 1, it's a file created
6647 * with open(2) O_TMPFILE flag.
6648 */
6649 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6650 if (err)
6651 goto fail;
6652 }
6653 d_instantiate(dentry, inode);
6654 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6655 }
6656
6657 fail:
6658 fscrypt_free_filename(&fname);
6659 if (trans)
6660 btrfs_end_transaction(trans);
6661 if (drop_inode) {
6662 inode_dec_link_count(inode);
6663 iput(inode);
6664 }
6665 btrfs_btree_balance_dirty(fs_info);
6666 return err;
6667 }
6668
btrfs_mkdir(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode)6669 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6670 struct dentry *dentry, umode_t mode)
6671 {
6672 struct inode *inode;
6673
6674 inode = new_inode(dir->i_sb);
6675 if (!inode)
6676 return -ENOMEM;
6677 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6678 inode->i_op = &btrfs_dir_inode_operations;
6679 inode->i_fop = &btrfs_dir_file_operations;
6680 return btrfs_create_common(dir, dentry, inode);
6681 }
6682
uncompress_inline(struct btrfs_path * path,struct page * page,struct btrfs_file_extent_item * item)6683 static noinline int uncompress_inline(struct btrfs_path *path,
6684 struct page *page,
6685 struct btrfs_file_extent_item *item)
6686 {
6687 int ret;
6688 struct extent_buffer *leaf = path->nodes[0];
6689 char *tmp;
6690 size_t max_size;
6691 unsigned long inline_size;
6692 unsigned long ptr;
6693 int compress_type;
6694
6695 compress_type = btrfs_file_extent_compression(leaf, item);
6696 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6697 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6698 tmp = kmalloc(inline_size, GFP_NOFS);
6699 if (!tmp)
6700 return -ENOMEM;
6701 ptr = btrfs_file_extent_inline_start(item);
6702
6703 read_extent_buffer(leaf, tmp, ptr, inline_size);
6704
6705 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6706 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6707
6708 /*
6709 * decompression code contains a memset to fill in any space between the end
6710 * of the uncompressed data and the end of max_size in case the decompressed
6711 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6712 * the end of an inline extent and the beginning of the next block, so we
6713 * cover that region here.
6714 */
6715
6716 if (max_size < PAGE_SIZE)
6717 memzero_page(page, max_size, PAGE_SIZE - max_size);
6718 kfree(tmp);
6719 return ret;
6720 }
6721
read_inline_extent(struct btrfs_inode * inode,struct btrfs_path * path,struct page * page)6722 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6723 struct page *page)
6724 {
6725 struct btrfs_file_extent_item *fi;
6726 void *kaddr;
6727 size_t copy_size;
6728
6729 if (!page || PageUptodate(page))
6730 return 0;
6731
6732 ASSERT(page_offset(page) == 0);
6733
6734 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6735 struct btrfs_file_extent_item);
6736 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6737 return uncompress_inline(path, page, fi);
6738
6739 copy_size = min_t(u64, PAGE_SIZE,
6740 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6741 kaddr = kmap_local_page(page);
6742 read_extent_buffer(path->nodes[0], kaddr,
6743 btrfs_file_extent_inline_start(fi), copy_size);
6744 kunmap_local(kaddr);
6745 if (copy_size < PAGE_SIZE)
6746 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6747 return 0;
6748 }
6749
6750 /*
6751 * Lookup the first extent overlapping a range in a file.
6752 *
6753 * @inode: file to search in
6754 * @page: page to read extent data into if the extent is inline
6755 * @pg_offset: offset into @page to copy to
6756 * @start: file offset
6757 * @len: length of range starting at @start
6758 *
6759 * Return the first &struct extent_map which overlaps the given range, reading
6760 * it from the B-tree and caching it if necessary. Note that there may be more
6761 * extents which overlap the given range after the returned extent_map.
6762 *
6763 * If @page is not NULL and the extent is inline, this also reads the extent
6764 * data directly into the page and marks the extent up to date in the io_tree.
6765 *
6766 * Return: ERR_PTR on error, non-NULL extent_map on success.
6767 */
btrfs_get_extent(struct btrfs_inode * inode,struct page * page,size_t pg_offset,u64 start,u64 len)6768 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6769 struct page *page, size_t pg_offset,
6770 u64 start, u64 len)
6771 {
6772 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6773 int ret = 0;
6774 u64 extent_start = 0;
6775 u64 extent_end = 0;
6776 u64 objectid = btrfs_ino(inode);
6777 int extent_type = -1;
6778 struct btrfs_path *path = NULL;
6779 struct btrfs_root *root = inode->root;
6780 struct btrfs_file_extent_item *item;
6781 struct extent_buffer *leaf;
6782 struct btrfs_key found_key;
6783 struct extent_map *em = NULL;
6784 struct extent_map_tree *em_tree = &inode->extent_tree;
6785
6786 read_lock(&em_tree->lock);
6787 em = lookup_extent_mapping(em_tree, start, len);
6788 read_unlock(&em_tree->lock);
6789
6790 if (em) {
6791 if (em->start > start || em->start + em->len <= start)
6792 free_extent_map(em);
6793 else if (em->block_start == EXTENT_MAP_INLINE && page)
6794 free_extent_map(em);
6795 else
6796 goto out;
6797 }
6798 em = alloc_extent_map();
6799 if (!em) {
6800 ret = -ENOMEM;
6801 goto out;
6802 }
6803 em->start = EXTENT_MAP_HOLE;
6804 em->orig_start = EXTENT_MAP_HOLE;
6805 em->len = (u64)-1;
6806 em->block_len = (u64)-1;
6807
6808 path = btrfs_alloc_path();
6809 if (!path) {
6810 ret = -ENOMEM;
6811 goto out;
6812 }
6813
6814 /* Chances are we'll be called again, so go ahead and do readahead */
6815 path->reada = READA_FORWARD;
6816
6817 /*
6818 * The same explanation in load_free_space_cache applies here as well,
6819 * we only read when we're loading the free space cache, and at that
6820 * point the commit_root has everything we need.
6821 */
6822 if (btrfs_is_free_space_inode(inode)) {
6823 path->search_commit_root = 1;
6824 path->skip_locking = 1;
6825 }
6826
6827 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6828 if (ret < 0) {
6829 goto out;
6830 } else if (ret > 0) {
6831 if (path->slots[0] == 0)
6832 goto not_found;
6833 path->slots[0]--;
6834 ret = 0;
6835 }
6836
6837 leaf = path->nodes[0];
6838 item = btrfs_item_ptr(leaf, path->slots[0],
6839 struct btrfs_file_extent_item);
6840 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6841 if (found_key.objectid != objectid ||
6842 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6843 /*
6844 * If we backup past the first extent we want to move forward
6845 * and see if there is an extent in front of us, otherwise we'll
6846 * say there is a hole for our whole search range which can
6847 * cause problems.
6848 */
6849 extent_end = start;
6850 goto next;
6851 }
6852
6853 extent_type = btrfs_file_extent_type(leaf, item);
6854 extent_start = found_key.offset;
6855 extent_end = btrfs_file_extent_end(path);
6856 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6857 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6858 /* Only regular file could have regular/prealloc extent */
6859 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6860 ret = -EUCLEAN;
6861 btrfs_crit(fs_info,
6862 "regular/prealloc extent found for non-regular inode %llu",
6863 btrfs_ino(inode));
6864 goto out;
6865 }
6866 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6867 extent_start);
6868 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6869 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6870 path->slots[0],
6871 extent_start);
6872 }
6873 next:
6874 if (start >= extent_end) {
6875 path->slots[0]++;
6876 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6877 ret = btrfs_next_leaf(root, path);
6878 if (ret < 0)
6879 goto out;
6880 else if (ret > 0)
6881 goto not_found;
6882
6883 leaf = path->nodes[0];
6884 }
6885 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6886 if (found_key.objectid != objectid ||
6887 found_key.type != BTRFS_EXTENT_DATA_KEY)
6888 goto not_found;
6889 if (start + len <= found_key.offset)
6890 goto not_found;
6891 if (start > found_key.offset)
6892 goto next;
6893
6894 /* New extent overlaps with existing one */
6895 em->start = start;
6896 em->orig_start = start;
6897 em->len = found_key.offset - start;
6898 em->block_start = EXTENT_MAP_HOLE;
6899 goto insert;
6900 }
6901
6902 btrfs_extent_item_to_extent_map(inode, path, item, em);
6903
6904 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6905 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6906 goto insert;
6907 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6908 /*
6909 * Inline extent can only exist at file offset 0. This is
6910 * ensured by tree-checker and inline extent creation path.
6911 * Thus all members representing file offsets should be zero.
6912 */
6913 ASSERT(pg_offset == 0);
6914 ASSERT(extent_start == 0);
6915 ASSERT(em->start == 0);
6916
6917 /*
6918 * btrfs_extent_item_to_extent_map() should have properly
6919 * initialized em members already.
6920 *
6921 * Other members are not utilized for inline extents.
6922 */
6923 ASSERT(em->block_start == EXTENT_MAP_INLINE);
6924 ASSERT(em->len == fs_info->sectorsize);
6925
6926 ret = read_inline_extent(inode, path, page);
6927 if (ret < 0)
6928 goto out;
6929 goto insert;
6930 }
6931 not_found:
6932 em->start = start;
6933 em->orig_start = start;
6934 em->len = len;
6935 em->block_start = EXTENT_MAP_HOLE;
6936 insert:
6937 ret = 0;
6938 btrfs_release_path(path);
6939 if (em->start > start || extent_map_end(em) <= start) {
6940 btrfs_err(fs_info,
6941 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6942 em->start, em->len, start, len);
6943 ret = -EIO;
6944 goto out;
6945 }
6946
6947 write_lock(&em_tree->lock);
6948 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6949 write_unlock(&em_tree->lock);
6950 out:
6951 btrfs_free_path(path);
6952
6953 trace_btrfs_get_extent(root, inode, em);
6954
6955 if (ret) {
6956 free_extent_map(em);
6957 return ERR_PTR(ret);
6958 }
6959 return em;
6960 }
6961
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)6962 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
6963 struct btrfs_dio_data *dio_data,
6964 const u64 start,
6965 const u64 len,
6966 const u64 orig_start,
6967 const u64 block_start,
6968 const u64 block_len,
6969 const u64 orig_block_len,
6970 const u64 ram_bytes,
6971 const int type)
6972 {
6973 struct extent_map *em = NULL;
6974 struct btrfs_ordered_extent *ordered;
6975
6976 if (type != BTRFS_ORDERED_NOCOW) {
6977 em = create_io_em(inode, start, len, orig_start, block_start,
6978 block_len, orig_block_len, ram_bytes,
6979 BTRFS_COMPRESS_NONE, /* compress_type */
6980 type);
6981 if (IS_ERR(em))
6982 goto out;
6983 }
6984 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
6985 block_start, block_len, 0,
6986 (1 << type) |
6987 (1 << BTRFS_ORDERED_DIRECT),
6988 BTRFS_COMPRESS_NONE);
6989 if (IS_ERR(ordered)) {
6990 if (em) {
6991 free_extent_map(em);
6992 btrfs_drop_extent_map_range(inode, start,
6993 start + len - 1, false);
6994 }
6995 em = ERR_CAST(ordered);
6996 } else {
6997 ASSERT(!dio_data->ordered);
6998 dio_data->ordered = ordered;
6999 }
7000 out:
7001
7002 return em;
7003 }
7004
btrfs_new_extent_direct(struct btrfs_inode * inode,struct btrfs_dio_data * dio_data,u64 start,u64 len)7005 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7006 struct btrfs_dio_data *dio_data,
7007 u64 start, u64 len)
7008 {
7009 struct btrfs_root *root = inode->root;
7010 struct btrfs_fs_info *fs_info = root->fs_info;
7011 struct extent_map *em;
7012 struct btrfs_key ins;
7013 u64 alloc_hint;
7014 int ret;
7015
7016 alloc_hint = get_extent_allocation_hint(inode, start, len);
7017 again:
7018 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7019 0, alloc_hint, &ins, 1, 1);
7020 if (ret == -EAGAIN) {
7021 ASSERT(btrfs_is_zoned(fs_info));
7022 wait_on_bit_io(&inode->root->fs_info->flags, BTRFS_FS_NEED_ZONE_FINISH,
7023 TASK_UNINTERRUPTIBLE);
7024 goto again;
7025 }
7026 if (ret)
7027 return ERR_PTR(ret);
7028
7029 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
7030 ins.objectid, ins.offset, ins.offset,
7031 ins.offset, BTRFS_ORDERED_REGULAR);
7032 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7033 if (IS_ERR(em))
7034 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7035 1);
7036
7037 return em;
7038 }
7039
btrfs_extent_readonly(struct btrfs_fs_info * fs_info,u64 bytenr)7040 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7041 {
7042 struct btrfs_block_group *block_group;
7043 bool readonly = false;
7044
7045 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7046 if (!block_group || block_group->ro)
7047 readonly = true;
7048 if (block_group)
7049 btrfs_put_block_group(block_group);
7050 return readonly;
7051 }
7052
7053 /*
7054 * Check if we can do nocow write into the range [@offset, @offset + @len)
7055 *
7056 * @offset: File offset
7057 * @len: The length to write, will be updated to the nocow writeable
7058 * range
7059 * @orig_start: (optional) Return the original file offset of the file extent
7060 * @orig_len: (optional) Return the original on-disk length of the file extent
7061 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7062 * @strict: if true, omit optimizations that might force us into unnecessary
7063 * cow. e.g., don't trust generation number.
7064 *
7065 * Return:
7066 * >0 and update @len if we can do nocow write
7067 * 0 if we can't do nocow write
7068 * <0 if error happened
7069 *
7070 * NOTE: This only checks the file extents, caller is responsible to wait for
7071 * any ordered extents.
7072 */
can_nocow_extent(struct inode * inode,u64 offset,u64 * len,u64 * orig_start,u64 * orig_block_len,u64 * ram_bytes,bool nowait,bool strict)7073 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7074 u64 *orig_start, u64 *orig_block_len,
7075 u64 *ram_bytes, bool nowait, bool strict)
7076 {
7077 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7078 struct can_nocow_file_extent_args nocow_args = { 0 };
7079 struct btrfs_path *path;
7080 int ret;
7081 struct extent_buffer *leaf;
7082 struct btrfs_root *root = BTRFS_I(inode)->root;
7083 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7084 struct btrfs_file_extent_item *fi;
7085 struct btrfs_key key;
7086 int found_type;
7087
7088 path = btrfs_alloc_path();
7089 if (!path)
7090 return -ENOMEM;
7091 path->nowait = nowait;
7092
7093 ret = btrfs_lookup_file_extent(NULL, root, path,
7094 btrfs_ino(BTRFS_I(inode)), offset, 0);
7095 if (ret < 0)
7096 goto out;
7097
7098 if (ret == 1) {
7099 if (path->slots[0] == 0) {
7100 /* can't find the item, must cow */
7101 ret = 0;
7102 goto out;
7103 }
7104 path->slots[0]--;
7105 }
7106 ret = 0;
7107 leaf = path->nodes[0];
7108 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7109 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7110 key.type != BTRFS_EXTENT_DATA_KEY) {
7111 /* not our file or wrong item type, must cow */
7112 goto out;
7113 }
7114
7115 if (key.offset > offset) {
7116 /* Wrong offset, must cow */
7117 goto out;
7118 }
7119
7120 if (btrfs_file_extent_end(path) <= offset)
7121 goto out;
7122
7123 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7124 found_type = btrfs_file_extent_type(leaf, fi);
7125 if (ram_bytes)
7126 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7127
7128 nocow_args.start = offset;
7129 nocow_args.end = offset + *len - 1;
7130 nocow_args.strict = strict;
7131 nocow_args.free_path = true;
7132
7133 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7134 /* can_nocow_file_extent() has freed the path. */
7135 path = NULL;
7136
7137 if (ret != 1) {
7138 /* Treat errors as not being able to NOCOW. */
7139 ret = 0;
7140 goto out;
7141 }
7142
7143 ret = 0;
7144 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7145 goto out;
7146
7147 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7148 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7149 u64 range_end;
7150
7151 range_end = round_up(offset + nocow_args.num_bytes,
7152 root->fs_info->sectorsize) - 1;
7153 ret = test_range_bit(io_tree, offset, range_end,
7154 EXTENT_DELALLOC, 0, NULL);
7155 if (ret) {
7156 ret = -EAGAIN;
7157 goto out;
7158 }
7159 }
7160
7161 if (orig_start)
7162 *orig_start = key.offset - nocow_args.extent_offset;
7163 if (orig_block_len)
7164 *orig_block_len = nocow_args.disk_num_bytes;
7165
7166 *len = nocow_args.num_bytes;
7167 ret = 1;
7168 out:
7169 btrfs_free_path(path);
7170 return ret;
7171 }
7172
lock_extent_direct(struct inode * inode,u64 lockstart,u64 lockend,struct extent_state ** cached_state,unsigned int iomap_flags)7173 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7174 struct extent_state **cached_state,
7175 unsigned int iomap_flags)
7176 {
7177 const bool writing = (iomap_flags & IOMAP_WRITE);
7178 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7179 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7180 struct btrfs_ordered_extent *ordered;
7181 int ret = 0;
7182
7183 while (1) {
7184 if (nowait) {
7185 if (!try_lock_extent(io_tree, lockstart, lockend,
7186 cached_state))
7187 return -EAGAIN;
7188 } else {
7189 lock_extent(io_tree, lockstart, lockend, cached_state);
7190 }
7191 /*
7192 * We're concerned with the entire range that we're going to be
7193 * doing DIO to, so we need to make sure there's no ordered
7194 * extents in this range.
7195 */
7196 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7197 lockend - lockstart + 1);
7198
7199 /*
7200 * We need to make sure there are no buffered pages in this
7201 * range either, we could have raced between the invalidate in
7202 * generic_file_direct_write and locking the extent. The
7203 * invalidate needs to happen so that reads after a write do not
7204 * get stale data.
7205 */
7206 if (!ordered &&
7207 (!writing || !filemap_range_has_page(inode->i_mapping,
7208 lockstart, lockend)))
7209 break;
7210
7211 unlock_extent(io_tree, lockstart, lockend, cached_state);
7212
7213 if (ordered) {
7214 if (nowait) {
7215 btrfs_put_ordered_extent(ordered);
7216 ret = -EAGAIN;
7217 break;
7218 }
7219 /*
7220 * If we are doing a DIO read and the ordered extent we
7221 * found is for a buffered write, we can not wait for it
7222 * to complete and retry, because if we do so we can
7223 * deadlock with concurrent buffered writes on page
7224 * locks. This happens only if our DIO read covers more
7225 * than one extent map, if at this point has already
7226 * created an ordered extent for a previous extent map
7227 * and locked its range in the inode's io tree, and a
7228 * concurrent write against that previous extent map's
7229 * range and this range started (we unlock the ranges
7230 * in the io tree only when the bios complete and
7231 * buffered writes always lock pages before attempting
7232 * to lock range in the io tree).
7233 */
7234 if (writing ||
7235 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7236 btrfs_start_ordered_extent(ordered);
7237 else
7238 ret = nowait ? -EAGAIN : -ENOTBLK;
7239 btrfs_put_ordered_extent(ordered);
7240 } else {
7241 /*
7242 * We could trigger writeback for this range (and wait
7243 * for it to complete) and then invalidate the pages for
7244 * this range (through invalidate_inode_pages2_range()),
7245 * but that can lead us to a deadlock with a concurrent
7246 * call to readahead (a buffered read or a defrag call
7247 * triggered a readahead) on a page lock due to an
7248 * ordered dio extent we created before but did not have
7249 * yet a corresponding bio submitted (whence it can not
7250 * complete), which makes readahead wait for that
7251 * ordered extent to complete while holding a lock on
7252 * that page.
7253 */
7254 ret = nowait ? -EAGAIN : -ENOTBLK;
7255 }
7256
7257 if (ret)
7258 break;
7259
7260 cond_resched();
7261 }
7262
7263 return ret;
7264 }
7265
7266 /* 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)7267 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7268 u64 len, u64 orig_start, u64 block_start,
7269 u64 block_len, u64 orig_block_len,
7270 u64 ram_bytes, int compress_type,
7271 int type)
7272 {
7273 struct extent_map *em;
7274 int ret;
7275
7276 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7277 type == BTRFS_ORDERED_COMPRESSED ||
7278 type == BTRFS_ORDERED_NOCOW ||
7279 type == BTRFS_ORDERED_REGULAR);
7280
7281 em = alloc_extent_map();
7282 if (!em)
7283 return ERR_PTR(-ENOMEM);
7284
7285 em->start = start;
7286 em->orig_start = orig_start;
7287 em->len = len;
7288 em->block_len = block_len;
7289 em->block_start = block_start;
7290 em->orig_block_len = orig_block_len;
7291 em->ram_bytes = ram_bytes;
7292 em->generation = -1;
7293 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7294 if (type == BTRFS_ORDERED_PREALLOC) {
7295 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7296 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7297 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7298 em->compress_type = compress_type;
7299 }
7300
7301 ret = btrfs_replace_extent_map_range(inode, em, true);
7302 if (ret) {
7303 free_extent_map(em);
7304 return ERR_PTR(ret);
7305 }
7306
7307 /* em got 2 refs now, callers needs to do free_extent_map once. */
7308 return em;
7309 }
7310
7311
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)7312 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7313 struct inode *inode,
7314 struct btrfs_dio_data *dio_data,
7315 u64 start, u64 *lenp,
7316 unsigned int iomap_flags)
7317 {
7318 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7319 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7320 struct extent_map *em = *map;
7321 int type;
7322 u64 block_start, orig_start, orig_block_len, ram_bytes;
7323 struct btrfs_block_group *bg;
7324 bool can_nocow = false;
7325 bool space_reserved = false;
7326 u64 len = *lenp;
7327 u64 prev_len;
7328 int ret = 0;
7329
7330 /*
7331 * We don't allocate a new extent in the following cases
7332 *
7333 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7334 * existing extent.
7335 * 2) The extent is marked as PREALLOC. We're good to go here and can
7336 * just use the extent.
7337 *
7338 */
7339 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7340 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7341 em->block_start != EXTENT_MAP_HOLE)) {
7342 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7343 type = BTRFS_ORDERED_PREALLOC;
7344 else
7345 type = BTRFS_ORDERED_NOCOW;
7346 len = min(len, em->len - (start - em->start));
7347 block_start = em->block_start + (start - em->start);
7348
7349 if (can_nocow_extent(inode, start, &len, &orig_start,
7350 &orig_block_len, &ram_bytes, false, false) == 1) {
7351 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7352 if (bg)
7353 can_nocow = true;
7354 }
7355 }
7356
7357 prev_len = len;
7358 if (can_nocow) {
7359 struct extent_map *em2;
7360
7361 /* We can NOCOW, so only need to reserve metadata space. */
7362 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7363 nowait);
7364 if (ret < 0) {
7365 /* Our caller expects us to free the input extent map. */
7366 free_extent_map(em);
7367 *map = NULL;
7368 btrfs_dec_nocow_writers(bg);
7369 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7370 ret = -EAGAIN;
7371 goto out;
7372 }
7373 space_reserved = true;
7374
7375 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7376 orig_start, block_start,
7377 len, orig_block_len,
7378 ram_bytes, type);
7379 btrfs_dec_nocow_writers(bg);
7380 if (type == BTRFS_ORDERED_PREALLOC) {
7381 free_extent_map(em);
7382 *map = em2;
7383 em = em2;
7384 }
7385
7386 if (IS_ERR(em2)) {
7387 ret = PTR_ERR(em2);
7388 goto out;
7389 }
7390
7391 dio_data->nocow_done = true;
7392 } else {
7393 /* Our caller expects us to free the input extent map. */
7394 free_extent_map(em);
7395 *map = NULL;
7396
7397 if (nowait) {
7398 ret = -EAGAIN;
7399 goto out;
7400 }
7401
7402 /*
7403 * If we could not allocate data space before locking the file
7404 * range and we can't do a NOCOW write, then we have to fail.
7405 */
7406 if (!dio_data->data_space_reserved) {
7407 ret = -ENOSPC;
7408 goto out;
7409 }
7410
7411 /*
7412 * We have to COW and we have already reserved data space before,
7413 * so now we reserve only metadata.
7414 */
7415 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7416 false);
7417 if (ret < 0)
7418 goto out;
7419 space_reserved = true;
7420
7421 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7422 if (IS_ERR(em)) {
7423 ret = PTR_ERR(em);
7424 goto out;
7425 }
7426 *map = em;
7427 len = min(len, em->len - (start - em->start));
7428 if (len < prev_len)
7429 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7430 prev_len - len, true);
7431 }
7432
7433 /*
7434 * We have created our ordered extent, so we can now release our reservation
7435 * for an outstanding extent.
7436 */
7437 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7438
7439 /*
7440 * Need to update the i_size under the extent lock so buffered
7441 * readers will get the updated i_size when we unlock.
7442 */
7443 if (start + len > i_size_read(inode))
7444 i_size_write(inode, start + len);
7445 out:
7446 if (ret && space_reserved) {
7447 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7448 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7449 }
7450 *lenp = len;
7451 return ret;
7452 }
7453
btrfs_dio_iomap_begin(struct inode * inode,loff_t start,loff_t length,unsigned int flags,struct iomap * iomap,struct iomap * srcmap)7454 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7455 loff_t length, unsigned int flags, struct iomap *iomap,
7456 struct iomap *srcmap)
7457 {
7458 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7459 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7460 struct extent_map *em;
7461 struct extent_state *cached_state = NULL;
7462 struct btrfs_dio_data *dio_data = iter->private;
7463 u64 lockstart, lockend;
7464 const bool write = !!(flags & IOMAP_WRITE);
7465 int ret = 0;
7466 u64 len = length;
7467 const u64 data_alloc_len = length;
7468 bool unlock_extents = false;
7469
7470 /*
7471 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7472 * we're NOWAIT we may submit a bio for a partial range and return
7473 * EIOCBQUEUED, which would result in an errant short read.
7474 *
7475 * The best way to handle this would be to allow for partial completions
7476 * of iocb's, so we could submit the partial bio, return and fault in
7477 * the rest of the pages, and then submit the io for the rest of the
7478 * range. However we don't have that currently, so simply return
7479 * -EAGAIN at this point so that the normal path is used.
7480 */
7481 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7482 return -EAGAIN;
7483
7484 /*
7485 * Cap the size of reads to that usually seen in buffered I/O as we need
7486 * to allocate a contiguous array for the checksums.
7487 */
7488 if (!write)
7489 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7490
7491 lockstart = start;
7492 lockend = start + len - 1;
7493
7494 /*
7495 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7496 * enough if we've written compressed pages to this area, so we need to
7497 * flush the dirty pages again to make absolutely sure that any
7498 * outstanding dirty pages are on disk - the first flush only starts
7499 * compression on the data, while keeping the pages locked, so by the
7500 * time the second flush returns we know bios for the compressed pages
7501 * were submitted and finished, and the pages no longer under writeback.
7502 *
7503 * If we have a NOWAIT request and we have any pages in the range that
7504 * are locked, likely due to compression still in progress, we don't want
7505 * to block on page locks. We also don't want to block on pages marked as
7506 * dirty or under writeback (same as for the non-compression case).
7507 * iomap_dio_rw() did the same check, but after that and before we got
7508 * here, mmap'ed writes may have happened or buffered reads started
7509 * (readpage() and readahead(), which lock pages), as we haven't locked
7510 * the file range yet.
7511 */
7512 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7513 &BTRFS_I(inode)->runtime_flags)) {
7514 if (flags & IOMAP_NOWAIT) {
7515 if (filemap_range_needs_writeback(inode->i_mapping,
7516 lockstart, lockend))
7517 return -EAGAIN;
7518 } else {
7519 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7520 start + length - 1);
7521 if (ret)
7522 return ret;
7523 }
7524 }
7525
7526 memset(dio_data, 0, sizeof(*dio_data));
7527
7528 /*
7529 * We always try to allocate data space and must do it before locking
7530 * the file range, to avoid deadlocks with concurrent writes to the same
7531 * range if the range has several extents and the writes don't expand the
7532 * current i_size (the inode lock is taken in shared mode). If we fail to
7533 * allocate data space here we continue and later, after locking the
7534 * file range, we fail with ENOSPC only if we figure out we can not do a
7535 * NOCOW write.
7536 */
7537 if (write && !(flags & IOMAP_NOWAIT)) {
7538 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7539 &dio_data->data_reserved,
7540 start, data_alloc_len, false);
7541 if (!ret)
7542 dio_data->data_space_reserved = true;
7543 else if (ret && !(BTRFS_I(inode)->flags &
7544 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7545 goto err;
7546 }
7547
7548 /*
7549 * If this errors out it's because we couldn't invalidate pagecache for
7550 * this range and we need to fallback to buffered IO, or we are doing a
7551 * NOWAIT read/write and we need to block.
7552 */
7553 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7554 if (ret < 0)
7555 goto err;
7556
7557 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7558 if (IS_ERR(em)) {
7559 ret = PTR_ERR(em);
7560 goto unlock_err;
7561 }
7562
7563 /*
7564 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7565 * io. INLINE is special, and we could probably kludge it in here, but
7566 * it's still buffered so for safety lets just fall back to the generic
7567 * buffered path.
7568 *
7569 * For COMPRESSED we _have_ to read the entire extent in so we can
7570 * decompress it, so there will be buffering required no matter what we
7571 * do, so go ahead and fallback to buffered.
7572 *
7573 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7574 * to buffered IO. Don't blame me, this is the price we pay for using
7575 * the generic code.
7576 */
7577 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7578 em->block_start == EXTENT_MAP_INLINE) {
7579 free_extent_map(em);
7580 /*
7581 * If we are in a NOWAIT context, return -EAGAIN in order to
7582 * fallback to buffered IO. This is not only because we can
7583 * block with buffered IO (no support for NOWAIT semantics at
7584 * the moment) but also to avoid returning short reads to user
7585 * space - this happens if we were able to read some data from
7586 * previous non-compressed extents and then when we fallback to
7587 * buffered IO, at btrfs_file_read_iter() by calling
7588 * filemap_read(), we fail to fault in pages for the read buffer,
7589 * in which case filemap_read() returns a short read (the number
7590 * of bytes previously read is > 0, so it does not return -EFAULT).
7591 */
7592 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7593 goto unlock_err;
7594 }
7595
7596 len = min(len, em->len - (start - em->start));
7597
7598 /*
7599 * If we have a NOWAIT request and the range contains multiple extents
7600 * (or a mix of extents and holes), then we return -EAGAIN to make the
7601 * caller fallback to a context where it can do a blocking (without
7602 * NOWAIT) request. This way we avoid doing partial IO and returning
7603 * success to the caller, which is not optimal for writes and for reads
7604 * it can result in unexpected behaviour for an application.
7605 *
7606 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7607 * iomap_dio_rw(), we can end up returning less data then what the caller
7608 * asked for, resulting in an unexpected, and incorrect, short read.
7609 * That is, the caller asked to read N bytes and we return less than that,
7610 * which is wrong unless we are crossing EOF. This happens if we get a
7611 * page fault error when trying to fault in pages for the buffer that is
7612 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7613 * have previously submitted bios for other extents in the range, in
7614 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7615 * those bios have completed by the time we get the page fault error,
7616 * which we return back to our caller - we should only return EIOCBQUEUED
7617 * after we have submitted bios for all the extents in the range.
7618 */
7619 if ((flags & IOMAP_NOWAIT) && len < length) {
7620 free_extent_map(em);
7621 ret = -EAGAIN;
7622 goto unlock_err;
7623 }
7624
7625 if (write) {
7626 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7627 start, &len, flags);
7628 if (ret < 0)
7629 goto unlock_err;
7630 unlock_extents = true;
7631 /* Recalc len in case the new em is smaller than requested */
7632 len = min(len, em->len - (start - em->start));
7633 if (dio_data->data_space_reserved) {
7634 u64 release_offset;
7635 u64 release_len = 0;
7636
7637 if (dio_data->nocow_done) {
7638 release_offset = start;
7639 release_len = data_alloc_len;
7640 } else if (len < data_alloc_len) {
7641 release_offset = start + len;
7642 release_len = data_alloc_len - len;
7643 }
7644
7645 if (release_len > 0)
7646 btrfs_free_reserved_data_space(BTRFS_I(inode),
7647 dio_data->data_reserved,
7648 release_offset,
7649 release_len);
7650 }
7651 } else {
7652 /*
7653 * We need to unlock only the end area that we aren't using.
7654 * The rest is going to be unlocked by the endio routine.
7655 */
7656 lockstart = start + len;
7657 if (lockstart < lockend)
7658 unlock_extents = true;
7659 }
7660
7661 if (unlock_extents)
7662 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7663 &cached_state);
7664 else
7665 free_extent_state(cached_state);
7666
7667 /*
7668 * Translate extent map information to iomap.
7669 * We trim the extents (and move the addr) even though iomap code does
7670 * that, since we have locked only the parts we are performing I/O in.
7671 */
7672 if ((em->block_start == EXTENT_MAP_HOLE) ||
7673 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7674 iomap->addr = IOMAP_NULL_ADDR;
7675 iomap->type = IOMAP_HOLE;
7676 } else {
7677 iomap->addr = em->block_start + (start - em->start);
7678 iomap->type = IOMAP_MAPPED;
7679 }
7680 iomap->offset = start;
7681 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7682 iomap->length = len;
7683 free_extent_map(em);
7684
7685 return 0;
7686
7687 unlock_err:
7688 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7689 &cached_state);
7690 err:
7691 if (dio_data->data_space_reserved) {
7692 btrfs_free_reserved_data_space(BTRFS_I(inode),
7693 dio_data->data_reserved,
7694 start, data_alloc_len);
7695 extent_changeset_free(dio_data->data_reserved);
7696 }
7697
7698 return ret;
7699 }
7700
btrfs_dio_iomap_end(struct inode * inode,loff_t pos,loff_t length,ssize_t written,unsigned int flags,struct iomap * iomap)7701 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7702 ssize_t written, unsigned int flags, struct iomap *iomap)
7703 {
7704 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7705 struct btrfs_dio_data *dio_data = iter->private;
7706 size_t submitted = dio_data->submitted;
7707 const bool write = !!(flags & IOMAP_WRITE);
7708 int ret = 0;
7709
7710 if (!write && (iomap->type == IOMAP_HOLE)) {
7711 /* If reading from a hole, unlock and return */
7712 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7713 NULL);
7714 return 0;
7715 }
7716
7717 if (submitted < length) {
7718 pos += submitted;
7719 length -= submitted;
7720 if (write)
7721 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7722 pos, length, false);
7723 else
7724 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7725 pos + length - 1, NULL);
7726 ret = -ENOTBLK;
7727 }
7728 if (write) {
7729 btrfs_put_ordered_extent(dio_data->ordered);
7730 dio_data->ordered = NULL;
7731 }
7732
7733 if (write)
7734 extent_changeset_free(dio_data->data_reserved);
7735 return ret;
7736 }
7737
btrfs_dio_end_io(struct btrfs_bio * bbio)7738 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7739 {
7740 struct btrfs_dio_private *dip =
7741 container_of(bbio, struct btrfs_dio_private, bbio);
7742 struct btrfs_inode *inode = bbio->inode;
7743 struct bio *bio = &bbio->bio;
7744
7745 if (bio->bi_status) {
7746 btrfs_warn(inode->root->fs_info,
7747 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7748 btrfs_ino(inode), bio->bi_opf,
7749 dip->file_offset, dip->bytes, bio->bi_status);
7750 }
7751
7752 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7753 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7754 dip->file_offset, dip->bytes,
7755 !bio->bi_status);
7756 } else {
7757 unlock_extent(&inode->io_tree, dip->file_offset,
7758 dip->file_offset + dip->bytes - 1, NULL);
7759 }
7760
7761 bbio->bio.bi_private = bbio->private;
7762 iomap_dio_bio_end_io(bio);
7763 }
7764
btrfs_dio_submit_io(const struct iomap_iter * iter,struct bio * bio,loff_t file_offset)7765 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7766 loff_t file_offset)
7767 {
7768 struct btrfs_bio *bbio = btrfs_bio(bio);
7769 struct btrfs_dio_private *dip =
7770 container_of(bbio, struct btrfs_dio_private, bbio);
7771 struct btrfs_dio_data *dio_data = iter->private;
7772
7773 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7774 btrfs_dio_end_io, bio->bi_private);
7775 bbio->inode = BTRFS_I(iter->inode);
7776 bbio->file_offset = file_offset;
7777
7778 dip->file_offset = file_offset;
7779 dip->bytes = bio->bi_iter.bi_size;
7780
7781 dio_data->submitted += bio->bi_iter.bi_size;
7782
7783 /*
7784 * Check if we are doing a partial write. If we are, we need to split
7785 * the ordered extent to match the submitted bio. Hang on to the
7786 * remaining unfinishable ordered_extent in dio_data so that it can be
7787 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7788 * remaining pages is blocked on the outstanding ordered extent.
7789 */
7790 if (iter->flags & IOMAP_WRITE) {
7791 int ret;
7792
7793 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7794 if (ret) {
7795 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7796 file_offset, dip->bytes,
7797 !ret);
7798 bio->bi_status = errno_to_blk_status(ret);
7799 iomap_dio_bio_end_io(bio);
7800 return;
7801 }
7802 }
7803
7804 btrfs_submit_bio(bbio, 0);
7805 }
7806
7807 static const struct iomap_ops btrfs_dio_iomap_ops = {
7808 .iomap_begin = btrfs_dio_iomap_begin,
7809 .iomap_end = btrfs_dio_iomap_end,
7810 };
7811
7812 static const struct iomap_dio_ops btrfs_dio_ops = {
7813 .submit_io = btrfs_dio_submit_io,
7814 .bio_set = &btrfs_dio_bioset,
7815 };
7816
btrfs_dio_read(struct kiocb * iocb,struct iov_iter * iter,size_t done_before)7817 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7818 {
7819 struct btrfs_dio_data data = { 0 };
7820
7821 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7822 IOMAP_DIO_PARTIAL, &data, done_before);
7823 }
7824
btrfs_dio_write(struct kiocb * iocb,struct iov_iter * iter,size_t done_before)7825 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7826 size_t done_before)
7827 {
7828 struct btrfs_dio_data data = { 0 };
7829
7830 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7831 IOMAP_DIO_PARTIAL, &data, done_before);
7832 }
7833
btrfs_fiemap(struct inode * inode,struct fiemap_extent_info * fieinfo,u64 start,u64 len)7834 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7835 u64 start, u64 len)
7836 {
7837 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
7838 int ret;
7839
7840 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7841 if (ret)
7842 return ret;
7843
7844 /*
7845 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7846 * file range (0 to LLONG_MAX), but that is not enough if we have
7847 * compression enabled. The first filemap_fdatawrite_range() only kicks
7848 * in the compression of data (in an async thread) and will return
7849 * before the compression is done and writeback is started. A second
7850 * filemap_fdatawrite_range() is needed to wait for the compression to
7851 * complete and writeback to start. We also need to wait for ordered
7852 * extents to complete, because our fiemap implementation uses mainly
7853 * file extent items to list the extents, searching for extent maps
7854 * only for file ranges with holes or prealloc extents to figure out
7855 * if we have delalloc in those ranges.
7856 */
7857 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7858 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7859 if (ret)
7860 return ret;
7861 }
7862
7863 btrfs_inode_lock(btrfs_inode, BTRFS_ILOCK_SHARED);
7864
7865 /*
7866 * We did an initial flush to avoid holding the inode's lock while
7867 * triggering writeback and waiting for the completion of IO and ordered
7868 * extents. Now after we locked the inode we do it again, because it's
7869 * possible a new write may have happened in between those two steps.
7870 */
7871 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7872 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7873 if (ret) {
7874 btrfs_inode_unlock(btrfs_inode, BTRFS_ILOCK_SHARED);
7875 return ret;
7876 }
7877 }
7878
7879 ret = extent_fiemap(btrfs_inode, fieinfo, start, len);
7880 btrfs_inode_unlock(btrfs_inode, BTRFS_ILOCK_SHARED);
7881
7882 return ret;
7883 }
7884
btrfs_writepages(struct address_space * mapping,struct writeback_control * wbc)7885 static int btrfs_writepages(struct address_space *mapping,
7886 struct writeback_control *wbc)
7887 {
7888 return extent_writepages(mapping, wbc);
7889 }
7890
btrfs_readahead(struct readahead_control * rac)7891 static void btrfs_readahead(struct readahead_control *rac)
7892 {
7893 extent_readahead(rac);
7894 }
7895
7896 /*
7897 * For release_folio() and invalidate_folio() we have a race window where
7898 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7899 * If we continue to release/invalidate the page, we could cause use-after-free
7900 * for subpage spinlock. So this function is to spin and wait for subpage
7901 * spinlock.
7902 */
wait_subpage_spinlock(struct page * page)7903 static void wait_subpage_spinlock(struct page *page)
7904 {
7905 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7906 struct btrfs_subpage *subpage;
7907
7908 if (!btrfs_is_subpage(fs_info, page))
7909 return;
7910
7911 ASSERT(PagePrivate(page) && page->private);
7912 subpage = (struct btrfs_subpage *)page->private;
7913
7914 /*
7915 * This may look insane as we just acquire the spinlock and release it,
7916 * without doing anything. But we just want to make sure no one is
7917 * still holding the subpage spinlock.
7918 * And since the page is not dirty nor writeback, and we have page
7919 * locked, the only possible way to hold a spinlock is from the endio
7920 * function to clear page writeback.
7921 *
7922 * Here we just acquire the spinlock so that all existing callers
7923 * should exit and we're safe to release/invalidate the page.
7924 */
7925 spin_lock_irq(&subpage->lock);
7926 spin_unlock_irq(&subpage->lock);
7927 }
7928
__btrfs_release_folio(struct folio * folio,gfp_t gfp_flags)7929 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7930 {
7931 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
7932
7933 if (ret == 1) {
7934 wait_subpage_spinlock(&folio->page);
7935 clear_page_extent_mapped(&folio->page);
7936 }
7937 return ret;
7938 }
7939
btrfs_release_folio(struct folio * folio,gfp_t gfp_flags)7940 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7941 {
7942 if (folio_test_writeback(folio) || folio_test_dirty(folio))
7943 return false;
7944 return __btrfs_release_folio(folio, gfp_flags);
7945 }
7946
7947 #ifdef CONFIG_MIGRATION
btrfs_migrate_folio(struct address_space * mapping,struct folio * dst,struct folio * src,enum migrate_mode mode)7948 static int btrfs_migrate_folio(struct address_space *mapping,
7949 struct folio *dst, struct folio *src,
7950 enum migrate_mode mode)
7951 {
7952 int ret = filemap_migrate_folio(mapping, dst, src, mode);
7953
7954 if (ret != MIGRATEPAGE_SUCCESS)
7955 return ret;
7956
7957 if (folio_test_ordered(src)) {
7958 folio_clear_ordered(src);
7959 folio_set_ordered(dst);
7960 }
7961
7962 return MIGRATEPAGE_SUCCESS;
7963 }
7964 #else
7965 #define btrfs_migrate_folio NULL
7966 #endif
7967
btrfs_invalidate_folio(struct folio * folio,size_t offset,size_t length)7968 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
7969 size_t length)
7970 {
7971 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
7972 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7973 struct extent_io_tree *tree = &inode->io_tree;
7974 struct extent_state *cached_state = NULL;
7975 u64 page_start = folio_pos(folio);
7976 u64 page_end = page_start + folio_size(folio) - 1;
7977 u64 cur;
7978 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
7979
7980 /*
7981 * We have folio locked so no new ordered extent can be created on this
7982 * page, nor bio can be submitted for this folio.
7983 *
7984 * But already submitted bio can still be finished on this folio.
7985 * Furthermore, endio function won't skip folio which has Ordered
7986 * (Private2) already cleared, so it's possible for endio and
7987 * invalidate_folio to do the same ordered extent accounting twice
7988 * on one folio.
7989 *
7990 * So here we wait for any submitted bios to finish, so that we won't
7991 * do double ordered extent accounting on the same folio.
7992 */
7993 folio_wait_writeback(folio);
7994 wait_subpage_spinlock(&folio->page);
7995
7996 /*
7997 * For subpage case, we have call sites like
7998 * btrfs_punch_hole_lock_range() which passes range not aligned to
7999 * sectorsize.
8000 * If the range doesn't cover the full folio, we don't need to and
8001 * shouldn't clear page extent mapped, as folio->private can still
8002 * record subpage dirty bits for other part of the range.
8003 *
8004 * For cases that invalidate the full folio even the range doesn't
8005 * cover the full folio, like invalidating the last folio, we're
8006 * still safe to wait for ordered extent to finish.
8007 */
8008 if (!(offset == 0 && length == folio_size(folio))) {
8009 btrfs_release_folio(folio, GFP_NOFS);
8010 return;
8011 }
8012
8013 if (!inode_evicting)
8014 lock_extent(tree, page_start, page_end, &cached_state);
8015
8016 cur = page_start;
8017 while (cur < page_end) {
8018 struct btrfs_ordered_extent *ordered;
8019 u64 range_end;
8020 u32 range_len;
8021 u32 extra_flags = 0;
8022
8023 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8024 page_end + 1 - cur);
8025 if (!ordered) {
8026 range_end = page_end;
8027 /*
8028 * No ordered extent covering this range, we are safe
8029 * to delete all extent states in the range.
8030 */
8031 extra_flags = EXTENT_CLEAR_ALL_BITS;
8032 goto next;
8033 }
8034 if (ordered->file_offset > cur) {
8035 /*
8036 * There is a range between [cur, oe->file_offset) not
8037 * covered by any ordered extent.
8038 * We are safe to delete all extent states, and handle
8039 * the ordered extent in the next iteration.
8040 */
8041 range_end = ordered->file_offset - 1;
8042 extra_flags = EXTENT_CLEAR_ALL_BITS;
8043 goto next;
8044 }
8045
8046 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8047 page_end);
8048 ASSERT(range_end + 1 - cur < U32_MAX);
8049 range_len = range_end + 1 - cur;
8050 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8051 /*
8052 * If Ordered (Private2) is cleared, it means endio has
8053 * already been executed for the range.
8054 * We can't delete the extent states as
8055 * btrfs_finish_ordered_io() may still use some of them.
8056 */
8057 goto next;
8058 }
8059 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8060
8061 /*
8062 * IO on this page will never be started, so we need to account
8063 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8064 * here, must leave that up for the ordered extent completion.
8065 *
8066 * This will also unlock the range for incoming
8067 * btrfs_finish_ordered_io().
8068 */
8069 if (!inode_evicting)
8070 clear_extent_bit(tree, cur, range_end,
8071 EXTENT_DELALLOC |
8072 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8073 EXTENT_DEFRAG, &cached_state);
8074
8075 spin_lock_irq(&inode->ordered_tree.lock);
8076 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8077 ordered->truncated_len = min(ordered->truncated_len,
8078 cur - ordered->file_offset);
8079 spin_unlock_irq(&inode->ordered_tree.lock);
8080
8081 /*
8082 * If the ordered extent has finished, we're safe to delete all
8083 * the extent states of the range, otherwise
8084 * btrfs_finish_ordered_io() will get executed by endio for
8085 * other pages, so we can't delete extent states.
8086 */
8087 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8088 cur, range_end + 1 - cur)) {
8089 btrfs_finish_ordered_io(ordered);
8090 /*
8091 * The ordered extent has finished, now we're again
8092 * safe to delete all extent states of the range.
8093 */
8094 extra_flags = EXTENT_CLEAR_ALL_BITS;
8095 }
8096 next:
8097 if (ordered)
8098 btrfs_put_ordered_extent(ordered);
8099 /*
8100 * Qgroup reserved space handler
8101 * Sector(s) here will be either:
8102 *
8103 * 1) Already written to disk or bio already finished
8104 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8105 * Qgroup will be handled by its qgroup_record then.
8106 * btrfs_qgroup_free_data() call will do nothing here.
8107 *
8108 * 2) Not written to disk yet
8109 * Then btrfs_qgroup_free_data() call will clear the
8110 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8111 * reserved data space.
8112 * Since the IO will never happen for this page.
8113 */
8114 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur, NULL);
8115 if (!inode_evicting) {
8116 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8117 EXTENT_DELALLOC | EXTENT_UPTODATE |
8118 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8119 extra_flags, &cached_state);
8120 }
8121 cur = range_end + 1;
8122 }
8123 /*
8124 * We have iterated through all ordered extents of the page, the page
8125 * should not have Ordered (Private2) anymore, or the above iteration
8126 * did something wrong.
8127 */
8128 ASSERT(!folio_test_ordered(folio));
8129 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8130 if (!inode_evicting)
8131 __btrfs_release_folio(folio, GFP_NOFS);
8132 clear_page_extent_mapped(&folio->page);
8133 }
8134
8135 /*
8136 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8137 * called from a page fault handler when a page is first dirtied. Hence we must
8138 * be careful to check for EOF conditions here. We set the page up correctly
8139 * for a written page which means we get ENOSPC checking when writing into
8140 * holes and correct delalloc and unwritten extent mapping on filesystems that
8141 * support these features.
8142 *
8143 * We are not allowed to take the i_mutex here so we have to play games to
8144 * protect against truncate races as the page could now be beyond EOF. Because
8145 * truncate_setsize() writes the inode size before removing pages, once we have
8146 * the page lock we can determine safely if the page is beyond EOF. If it is not
8147 * beyond EOF, then the page is guaranteed safe against truncation until we
8148 * unlock the page.
8149 */
btrfs_page_mkwrite(struct vm_fault * vmf)8150 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8151 {
8152 struct page *page = vmf->page;
8153 struct inode *inode = file_inode(vmf->vma->vm_file);
8154 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8155 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8156 struct btrfs_ordered_extent *ordered;
8157 struct extent_state *cached_state = NULL;
8158 struct extent_changeset *data_reserved = NULL;
8159 unsigned long zero_start;
8160 loff_t size;
8161 vm_fault_t ret;
8162 int ret2;
8163 int reserved = 0;
8164 u64 reserved_space;
8165 u64 page_start;
8166 u64 page_end;
8167 u64 end;
8168
8169 reserved_space = PAGE_SIZE;
8170
8171 sb_start_pagefault(inode->i_sb);
8172 page_start = page_offset(page);
8173 page_end = page_start + PAGE_SIZE - 1;
8174 end = page_end;
8175
8176 /*
8177 * Reserving delalloc space after obtaining the page lock can lead to
8178 * deadlock. For example, if a dirty page is locked by this function
8179 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8180 * dirty page write out, then the btrfs_writepages() function could
8181 * end up waiting indefinitely to get a lock on the page currently
8182 * being processed by btrfs_page_mkwrite() function.
8183 */
8184 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8185 page_start, reserved_space);
8186 if (!ret2) {
8187 ret2 = file_update_time(vmf->vma->vm_file);
8188 reserved = 1;
8189 }
8190 if (ret2) {
8191 ret = vmf_error(ret2);
8192 if (reserved)
8193 goto out;
8194 goto out_noreserve;
8195 }
8196
8197 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8198 again:
8199 down_read(&BTRFS_I(inode)->i_mmap_lock);
8200 lock_page(page);
8201 size = i_size_read(inode);
8202
8203 if ((page->mapping != inode->i_mapping) ||
8204 (page_start >= size)) {
8205 /* page got truncated out from underneath us */
8206 goto out_unlock;
8207 }
8208 wait_on_page_writeback(page);
8209
8210 lock_extent(io_tree, page_start, page_end, &cached_state);
8211 ret2 = set_page_extent_mapped(page);
8212 if (ret2 < 0) {
8213 ret = vmf_error(ret2);
8214 unlock_extent(io_tree, page_start, page_end, &cached_state);
8215 goto out_unlock;
8216 }
8217
8218 /*
8219 * we can't set the delalloc bits if there are pending ordered
8220 * extents. Drop our locks and wait for them to finish
8221 */
8222 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8223 PAGE_SIZE);
8224 if (ordered) {
8225 unlock_extent(io_tree, page_start, page_end, &cached_state);
8226 unlock_page(page);
8227 up_read(&BTRFS_I(inode)->i_mmap_lock);
8228 btrfs_start_ordered_extent(ordered);
8229 btrfs_put_ordered_extent(ordered);
8230 goto again;
8231 }
8232
8233 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8234 reserved_space = round_up(size - page_start,
8235 fs_info->sectorsize);
8236 if (reserved_space < PAGE_SIZE) {
8237 end = page_start + reserved_space - 1;
8238 btrfs_delalloc_release_space(BTRFS_I(inode),
8239 data_reserved, page_start,
8240 PAGE_SIZE - reserved_space, true);
8241 }
8242 }
8243
8244 /*
8245 * page_mkwrite gets called when the page is firstly dirtied after it's
8246 * faulted in, but write(2) could also dirty a page and set delalloc
8247 * bits, thus in this case for space account reason, we still need to
8248 * clear any delalloc bits within this page range since we have to
8249 * reserve data&meta space before lock_page() (see above comments).
8250 */
8251 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8252 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8253 EXTENT_DEFRAG, &cached_state);
8254
8255 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8256 &cached_state);
8257 if (ret2) {
8258 unlock_extent(io_tree, page_start, page_end, &cached_state);
8259 ret = VM_FAULT_SIGBUS;
8260 goto out_unlock;
8261 }
8262
8263 /* page is wholly or partially inside EOF */
8264 if (page_start + PAGE_SIZE > size)
8265 zero_start = offset_in_page(size);
8266 else
8267 zero_start = PAGE_SIZE;
8268
8269 if (zero_start != PAGE_SIZE)
8270 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8271
8272 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8273 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8274 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8275
8276 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8277
8278 unlock_extent(io_tree, page_start, page_end, &cached_state);
8279 up_read(&BTRFS_I(inode)->i_mmap_lock);
8280
8281 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8282 sb_end_pagefault(inode->i_sb);
8283 extent_changeset_free(data_reserved);
8284 return VM_FAULT_LOCKED;
8285
8286 out_unlock:
8287 unlock_page(page);
8288 up_read(&BTRFS_I(inode)->i_mmap_lock);
8289 out:
8290 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8291 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8292 reserved_space, (ret != 0));
8293 out_noreserve:
8294 sb_end_pagefault(inode->i_sb);
8295 extent_changeset_free(data_reserved);
8296 return ret;
8297 }
8298
btrfs_truncate(struct btrfs_inode * inode,bool skip_writeback)8299 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8300 {
8301 struct btrfs_truncate_control control = {
8302 .inode = inode,
8303 .ino = btrfs_ino(inode),
8304 .min_type = BTRFS_EXTENT_DATA_KEY,
8305 .clear_extent_range = true,
8306 };
8307 struct btrfs_root *root = inode->root;
8308 struct btrfs_fs_info *fs_info = root->fs_info;
8309 struct btrfs_block_rsv *rsv;
8310 int ret;
8311 struct btrfs_trans_handle *trans;
8312 u64 mask = fs_info->sectorsize - 1;
8313 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8314
8315 if (!skip_writeback) {
8316 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8317 inode->vfs_inode.i_size & (~mask),
8318 (u64)-1);
8319 if (ret)
8320 return ret;
8321 }
8322
8323 /*
8324 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8325 * things going on here:
8326 *
8327 * 1) We need to reserve space to update our inode.
8328 *
8329 * 2) We need to have something to cache all the space that is going to
8330 * be free'd up by the truncate operation, but also have some slack
8331 * space reserved in case it uses space during the truncate (thank you
8332 * very much snapshotting).
8333 *
8334 * And we need these to be separate. The fact is we can use a lot of
8335 * space doing the truncate, and we have no earthly idea how much space
8336 * we will use, so we need the truncate reservation to be separate so it
8337 * doesn't end up using space reserved for updating the inode. We also
8338 * need to be able to stop the transaction and start a new one, which
8339 * means we need to be able to update the inode several times, and we
8340 * have no idea of knowing how many times that will be, so we can't just
8341 * reserve 1 item for the entirety of the operation, so that has to be
8342 * done separately as well.
8343 *
8344 * So that leaves us with
8345 *
8346 * 1) rsv - for the truncate reservation, which we will steal from the
8347 * transaction reservation.
8348 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8349 * updating the inode.
8350 */
8351 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8352 if (!rsv)
8353 return -ENOMEM;
8354 rsv->size = min_size;
8355 rsv->failfast = true;
8356
8357 /*
8358 * 1 for the truncate slack space
8359 * 1 for updating the inode.
8360 */
8361 trans = btrfs_start_transaction(root, 2);
8362 if (IS_ERR(trans)) {
8363 ret = PTR_ERR(trans);
8364 goto out;
8365 }
8366
8367 /* Migrate the slack space for the truncate to our reserve */
8368 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8369 min_size, false);
8370 /*
8371 * We have reserved 2 metadata units when we started the transaction and
8372 * min_size matches 1 unit, so this should never fail, but if it does,
8373 * it's not critical we just fail truncation.
8374 */
8375 if (WARN_ON(ret)) {
8376 btrfs_end_transaction(trans);
8377 goto out;
8378 }
8379
8380 trans->block_rsv = rsv;
8381
8382 while (1) {
8383 struct extent_state *cached_state = NULL;
8384 const u64 new_size = inode->vfs_inode.i_size;
8385 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8386
8387 control.new_size = new_size;
8388 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8389 /*
8390 * We want to drop from the next block forward in case this new
8391 * size is not block aligned since we will be keeping the last
8392 * block of the extent just the way it is.
8393 */
8394 btrfs_drop_extent_map_range(inode,
8395 ALIGN(new_size, fs_info->sectorsize),
8396 (u64)-1, false);
8397
8398 ret = btrfs_truncate_inode_items(trans, root, &control);
8399
8400 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8401 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8402
8403 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8404
8405 trans->block_rsv = &fs_info->trans_block_rsv;
8406 if (ret != -ENOSPC && ret != -EAGAIN)
8407 break;
8408
8409 ret = btrfs_update_inode(trans, root, inode);
8410 if (ret)
8411 break;
8412
8413 btrfs_end_transaction(trans);
8414 btrfs_btree_balance_dirty(fs_info);
8415
8416 trans = btrfs_start_transaction(root, 2);
8417 if (IS_ERR(trans)) {
8418 ret = PTR_ERR(trans);
8419 trans = NULL;
8420 break;
8421 }
8422
8423 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8424 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8425 rsv, min_size, false);
8426 /*
8427 * We have reserved 2 metadata units when we started the
8428 * transaction and min_size matches 1 unit, so this should never
8429 * fail, but if it does, it's not critical we just fail truncation.
8430 */
8431 if (WARN_ON(ret))
8432 break;
8433
8434 trans->block_rsv = rsv;
8435 }
8436
8437 /*
8438 * We can't call btrfs_truncate_block inside a trans handle as we could
8439 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8440 * know we've truncated everything except the last little bit, and can
8441 * do btrfs_truncate_block and then update the disk_i_size.
8442 */
8443 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8444 btrfs_end_transaction(trans);
8445 btrfs_btree_balance_dirty(fs_info);
8446
8447 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8448 if (ret)
8449 goto out;
8450 trans = btrfs_start_transaction(root, 1);
8451 if (IS_ERR(trans)) {
8452 ret = PTR_ERR(trans);
8453 goto out;
8454 }
8455 btrfs_inode_safe_disk_i_size_write(inode, 0);
8456 }
8457
8458 if (trans) {
8459 int ret2;
8460
8461 trans->block_rsv = &fs_info->trans_block_rsv;
8462 ret2 = btrfs_update_inode(trans, root, inode);
8463 if (ret2 && !ret)
8464 ret = ret2;
8465
8466 ret2 = btrfs_end_transaction(trans);
8467 if (ret2 && !ret)
8468 ret = ret2;
8469 btrfs_btree_balance_dirty(fs_info);
8470 }
8471 out:
8472 btrfs_free_block_rsv(fs_info, rsv);
8473 /*
8474 * So if we truncate and then write and fsync we normally would just
8475 * write the extents that changed, which is a problem if we need to
8476 * first truncate that entire inode. So set this flag so we write out
8477 * all of the extents in the inode to the sync log so we're completely
8478 * safe.
8479 *
8480 * If no extents were dropped or trimmed we don't need to force the next
8481 * fsync to truncate all the inode's items from the log and re-log them
8482 * all. This means the truncate operation did not change the file size,
8483 * or changed it to a smaller size but there was only an implicit hole
8484 * between the old i_size and the new i_size, and there were no prealloc
8485 * extents beyond i_size to drop.
8486 */
8487 if (control.extents_found > 0)
8488 btrfs_set_inode_full_sync(inode);
8489
8490 return ret;
8491 }
8492
btrfs_new_subvol_inode(struct mnt_idmap * idmap,struct inode * dir)8493 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8494 struct inode *dir)
8495 {
8496 struct inode *inode;
8497
8498 inode = new_inode(dir->i_sb);
8499 if (inode) {
8500 /*
8501 * Subvolumes don't inherit the sgid bit or the parent's gid if
8502 * the parent's sgid bit is set. This is probably a bug.
8503 */
8504 inode_init_owner(idmap, inode, NULL,
8505 S_IFDIR | (~current_umask() & S_IRWXUGO));
8506 inode->i_op = &btrfs_dir_inode_operations;
8507 inode->i_fop = &btrfs_dir_file_operations;
8508 }
8509 return inode;
8510 }
8511
btrfs_alloc_inode(struct super_block * sb)8512 struct inode *btrfs_alloc_inode(struct super_block *sb)
8513 {
8514 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8515 struct btrfs_inode *ei;
8516 struct inode *inode;
8517
8518 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8519 if (!ei)
8520 return NULL;
8521
8522 ei->root = NULL;
8523 ei->generation = 0;
8524 ei->last_trans = 0;
8525 ei->last_sub_trans = 0;
8526 ei->logged_trans = 0;
8527 ei->delalloc_bytes = 0;
8528 ei->new_delalloc_bytes = 0;
8529 ei->defrag_bytes = 0;
8530 ei->disk_i_size = 0;
8531 ei->flags = 0;
8532 ei->ro_flags = 0;
8533 ei->csum_bytes = 0;
8534 ei->index_cnt = (u64)-1;
8535 ei->dir_index = 0;
8536 ei->last_unlink_trans = 0;
8537 ei->last_reflink_trans = 0;
8538 ei->last_log_commit = 0;
8539
8540 spin_lock_init(&ei->lock);
8541 ei->outstanding_extents = 0;
8542 if (sb->s_magic != BTRFS_TEST_MAGIC)
8543 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8544 BTRFS_BLOCK_RSV_DELALLOC);
8545 ei->runtime_flags = 0;
8546 ei->prop_compress = BTRFS_COMPRESS_NONE;
8547 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8548
8549 ei->delayed_node = NULL;
8550
8551 ei->i_otime.tv_sec = 0;
8552 ei->i_otime.tv_nsec = 0;
8553
8554 inode = &ei->vfs_inode;
8555 extent_map_tree_init(&ei->extent_tree);
8556 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8557 ei->io_tree.inode = ei;
8558 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8559 IO_TREE_INODE_FILE_EXTENT);
8560 mutex_init(&ei->log_mutex);
8561 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8562 INIT_LIST_HEAD(&ei->delalloc_inodes);
8563 INIT_LIST_HEAD(&ei->delayed_iput);
8564 RB_CLEAR_NODE(&ei->rb_node);
8565 init_rwsem(&ei->i_mmap_lock);
8566
8567 return inode;
8568 }
8569
8570 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
btrfs_test_destroy_inode(struct inode * inode)8571 void btrfs_test_destroy_inode(struct inode *inode)
8572 {
8573 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8574 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8575 }
8576 #endif
8577
btrfs_free_inode(struct inode * inode)8578 void btrfs_free_inode(struct inode *inode)
8579 {
8580 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8581 }
8582
btrfs_destroy_inode(struct inode * vfs_inode)8583 void btrfs_destroy_inode(struct inode *vfs_inode)
8584 {
8585 struct btrfs_ordered_extent *ordered;
8586 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8587 struct btrfs_root *root = inode->root;
8588 bool freespace_inode;
8589
8590 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8591 WARN_ON(vfs_inode->i_data.nrpages);
8592 WARN_ON(inode->block_rsv.reserved);
8593 WARN_ON(inode->block_rsv.size);
8594 WARN_ON(inode->outstanding_extents);
8595 if (!S_ISDIR(vfs_inode->i_mode)) {
8596 WARN_ON(inode->delalloc_bytes);
8597 WARN_ON(inode->new_delalloc_bytes);
8598 }
8599 WARN_ON(inode->csum_bytes);
8600 WARN_ON(inode->defrag_bytes);
8601
8602 /*
8603 * This can happen where we create an inode, but somebody else also
8604 * created the same inode and we need to destroy the one we already
8605 * created.
8606 */
8607 if (!root)
8608 return;
8609
8610 /*
8611 * If this is a free space inode do not take the ordered extents lockdep
8612 * map.
8613 */
8614 freespace_inode = btrfs_is_free_space_inode(inode);
8615
8616 while (1) {
8617 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8618 if (!ordered)
8619 break;
8620 else {
8621 btrfs_err(root->fs_info,
8622 "found ordered extent %llu %llu on inode cleanup",
8623 ordered->file_offset, ordered->num_bytes);
8624
8625 if (!freespace_inode)
8626 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8627
8628 btrfs_remove_ordered_extent(inode, ordered);
8629 btrfs_put_ordered_extent(ordered);
8630 btrfs_put_ordered_extent(ordered);
8631 }
8632 }
8633 btrfs_qgroup_check_reserved_leak(inode);
8634 inode_tree_del(inode);
8635 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8636 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8637 btrfs_put_root(inode->root);
8638 }
8639
btrfs_drop_inode(struct inode * inode)8640 int btrfs_drop_inode(struct inode *inode)
8641 {
8642 struct btrfs_root *root = BTRFS_I(inode)->root;
8643
8644 if (root == NULL)
8645 return 1;
8646
8647 /* the snap/subvol tree is on deleting */
8648 if (btrfs_root_refs(&root->root_item) == 0)
8649 return 1;
8650 else
8651 return generic_drop_inode(inode);
8652 }
8653
init_once(void * foo)8654 static void init_once(void *foo)
8655 {
8656 struct btrfs_inode *ei = foo;
8657
8658 inode_init_once(&ei->vfs_inode);
8659 }
8660
btrfs_destroy_cachep(void)8661 void __cold btrfs_destroy_cachep(void)
8662 {
8663 /*
8664 * Make sure all delayed rcu free inodes are flushed before we
8665 * destroy cache.
8666 */
8667 rcu_barrier();
8668 bioset_exit(&btrfs_dio_bioset);
8669 kmem_cache_destroy(btrfs_inode_cachep);
8670 }
8671
btrfs_init_cachep(void)8672 int __init btrfs_init_cachep(void)
8673 {
8674 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8675 sizeof(struct btrfs_inode), 0,
8676 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8677 init_once);
8678 if (!btrfs_inode_cachep)
8679 goto fail;
8680
8681 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8682 offsetof(struct btrfs_dio_private, bbio.bio),
8683 BIOSET_NEED_BVECS))
8684 goto fail;
8685
8686 return 0;
8687 fail:
8688 btrfs_destroy_cachep();
8689 return -ENOMEM;
8690 }
8691
btrfs_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int flags)8692 static int btrfs_getattr(struct mnt_idmap *idmap,
8693 const struct path *path, struct kstat *stat,
8694 u32 request_mask, unsigned int flags)
8695 {
8696 u64 delalloc_bytes;
8697 u64 inode_bytes;
8698 struct inode *inode = d_inode(path->dentry);
8699 u32 blocksize = btrfs_sb(inode->i_sb)->sectorsize;
8700 u32 bi_flags = BTRFS_I(inode)->flags;
8701 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8702
8703 stat->result_mask |= STATX_BTIME;
8704 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8705 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8706 if (bi_flags & BTRFS_INODE_APPEND)
8707 stat->attributes |= STATX_ATTR_APPEND;
8708 if (bi_flags & BTRFS_INODE_COMPRESS)
8709 stat->attributes |= STATX_ATTR_COMPRESSED;
8710 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8711 stat->attributes |= STATX_ATTR_IMMUTABLE;
8712 if (bi_flags & BTRFS_INODE_NODUMP)
8713 stat->attributes |= STATX_ATTR_NODUMP;
8714 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8715 stat->attributes |= STATX_ATTR_VERITY;
8716
8717 stat->attributes_mask |= (STATX_ATTR_APPEND |
8718 STATX_ATTR_COMPRESSED |
8719 STATX_ATTR_IMMUTABLE |
8720 STATX_ATTR_NODUMP);
8721
8722 generic_fillattr(idmap, request_mask, inode, stat);
8723 stat->dev = BTRFS_I(inode)->root->anon_dev;
8724
8725 spin_lock(&BTRFS_I(inode)->lock);
8726 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8727 inode_bytes = inode_get_bytes(inode);
8728 spin_unlock(&BTRFS_I(inode)->lock);
8729 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8730 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8731 return 0;
8732 }
8733
btrfs_rename_exchange(struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry)8734 static int btrfs_rename_exchange(struct inode *old_dir,
8735 struct dentry *old_dentry,
8736 struct inode *new_dir,
8737 struct dentry *new_dentry)
8738 {
8739 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8740 struct btrfs_trans_handle *trans;
8741 unsigned int trans_num_items;
8742 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8743 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8744 struct inode *new_inode = new_dentry->d_inode;
8745 struct inode *old_inode = old_dentry->d_inode;
8746 struct btrfs_rename_ctx old_rename_ctx;
8747 struct btrfs_rename_ctx new_rename_ctx;
8748 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8749 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8750 u64 old_idx = 0;
8751 u64 new_idx = 0;
8752 int ret;
8753 int ret2;
8754 bool need_abort = false;
8755 struct fscrypt_name old_fname, new_fname;
8756 struct fscrypt_str *old_name, *new_name;
8757
8758 /*
8759 * For non-subvolumes allow exchange only within one subvolume, in the
8760 * same inode namespace. Two subvolumes (represented as directory) can
8761 * be exchanged as they're a logical link and have a fixed inode number.
8762 */
8763 if (root != dest &&
8764 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8765 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8766 return -EXDEV;
8767
8768 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8769 if (ret)
8770 return ret;
8771
8772 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8773 if (ret) {
8774 fscrypt_free_filename(&old_fname);
8775 return ret;
8776 }
8777
8778 old_name = &old_fname.disk_name;
8779 new_name = &new_fname.disk_name;
8780
8781 /* close the race window with snapshot create/destroy ioctl */
8782 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8783 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8784 down_read(&fs_info->subvol_sem);
8785
8786 /*
8787 * For each inode:
8788 * 1 to remove old dir item
8789 * 1 to remove old dir index
8790 * 1 to add new dir item
8791 * 1 to add new dir index
8792 * 1 to update parent inode
8793 *
8794 * If the parents are the same, we only need to account for one
8795 */
8796 trans_num_items = (old_dir == new_dir ? 9 : 10);
8797 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8798 /*
8799 * 1 to remove old root ref
8800 * 1 to remove old root backref
8801 * 1 to add new root ref
8802 * 1 to add new root backref
8803 */
8804 trans_num_items += 4;
8805 } else {
8806 /*
8807 * 1 to update inode item
8808 * 1 to remove old inode ref
8809 * 1 to add new inode ref
8810 */
8811 trans_num_items += 3;
8812 }
8813 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8814 trans_num_items += 4;
8815 else
8816 trans_num_items += 3;
8817 trans = btrfs_start_transaction(root, trans_num_items);
8818 if (IS_ERR(trans)) {
8819 ret = PTR_ERR(trans);
8820 goto out_notrans;
8821 }
8822
8823 if (dest != root) {
8824 ret = btrfs_record_root_in_trans(trans, dest);
8825 if (ret)
8826 goto out_fail;
8827 }
8828
8829 /*
8830 * We need to find a free sequence number both in the source and
8831 * in the destination directory for the exchange.
8832 */
8833 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8834 if (ret)
8835 goto out_fail;
8836 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8837 if (ret)
8838 goto out_fail;
8839
8840 BTRFS_I(old_inode)->dir_index = 0ULL;
8841 BTRFS_I(new_inode)->dir_index = 0ULL;
8842
8843 /* Reference for the source. */
8844 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8845 /* force full log commit if subvolume involved. */
8846 btrfs_set_log_full_commit(trans);
8847 } else {
8848 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8849 btrfs_ino(BTRFS_I(new_dir)),
8850 old_idx);
8851 if (ret)
8852 goto out_fail;
8853 need_abort = true;
8854 }
8855
8856 /* And now for the dest. */
8857 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8858 /* force full log commit if subvolume involved. */
8859 btrfs_set_log_full_commit(trans);
8860 } else {
8861 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8862 btrfs_ino(BTRFS_I(old_dir)),
8863 new_idx);
8864 if (ret) {
8865 if (need_abort)
8866 btrfs_abort_transaction(trans, ret);
8867 goto out_fail;
8868 }
8869 }
8870
8871 /* Update inode version and ctime/mtime. */
8872 inode_inc_iversion(old_dir);
8873 inode_inc_iversion(new_dir);
8874 inode_inc_iversion(old_inode);
8875 inode_inc_iversion(new_inode);
8876 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
8877
8878 if (old_dentry->d_parent != new_dentry->d_parent) {
8879 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8880 BTRFS_I(old_inode), true);
8881 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8882 BTRFS_I(new_inode), true);
8883 }
8884
8885 /* src is a subvolume */
8886 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8887 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8888 } else { /* src is an inode */
8889 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8890 BTRFS_I(old_dentry->d_inode),
8891 old_name, &old_rename_ctx);
8892 if (!ret)
8893 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
8894 }
8895 if (ret) {
8896 btrfs_abort_transaction(trans, ret);
8897 goto out_fail;
8898 }
8899
8900 /* dest is a subvolume */
8901 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8902 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8903 } else { /* dest is an inode */
8904 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8905 BTRFS_I(new_dentry->d_inode),
8906 new_name, &new_rename_ctx);
8907 if (!ret)
8908 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
8909 }
8910 if (ret) {
8911 btrfs_abort_transaction(trans, ret);
8912 goto out_fail;
8913 }
8914
8915 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8916 new_name, 0, old_idx);
8917 if (ret) {
8918 btrfs_abort_transaction(trans, ret);
8919 goto out_fail;
8920 }
8921
8922 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8923 old_name, 0, new_idx);
8924 if (ret) {
8925 btrfs_abort_transaction(trans, ret);
8926 goto out_fail;
8927 }
8928
8929 if (old_inode->i_nlink == 1)
8930 BTRFS_I(old_inode)->dir_index = old_idx;
8931 if (new_inode->i_nlink == 1)
8932 BTRFS_I(new_inode)->dir_index = new_idx;
8933
8934 /*
8935 * Now pin the logs of the roots. We do it to ensure that no other task
8936 * can sync the logs while we are in progress with the rename, because
8937 * that could result in an inconsistency in case any of the inodes that
8938 * are part of this rename operation were logged before.
8939 */
8940 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8941 btrfs_pin_log_trans(root);
8942 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8943 btrfs_pin_log_trans(dest);
8944
8945 /* Do the log updates for all inodes. */
8946 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8947 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
8948 old_rename_ctx.index, new_dentry->d_parent);
8949 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8950 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
8951 new_rename_ctx.index, old_dentry->d_parent);
8952
8953 /* Now unpin the logs. */
8954 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8955 btrfs_end_log_trans(root);
8956 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8957 btrfs_end_log_trans(dest);
8958 out_fail:
8959 ret2 = btrfs_end_transaction(trans);
8960 ret = ret ? ret : ret2;
8961 out_notrans:
8962 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
8963 old_ino == BTRFS_FIRST_FREE_OBJECTID)
8964 up_read(&fs_info->subvol_sem);
8965
8966 fscrypt_free_filename(&new_fname);
8967 fscrypt_free_filename(&old_fname);
8968 return ret;
8969 }
8970
new_whiteout_inode(struct mnt_idmap * idmap,struct inode * dir)8971 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
8972 struct inode *dir)
8973 {
8974 struct inode *inode;
8975
8976 inode = new_inode(dir->i_sb);
8977 if (inode) {
8978 inode_init_owner(idmap, inode, dir,
8979 S_IFCHR | WHITEOUT_MODE);
8980 inode->i_op = &btrfs_special_inode_operations;
8981 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
8982 }
8983 return inode;
8984 }
8985
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)8986 static int btrfs_rename(struct mnt_idmap *idmap,
8987 struct inode *old_dir, struct dentry *old_dentry,
8988 struct inode *new_dir, struct dentry *new_dentry,
8989 unsigned int flags)
8990 {
8991 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8992 struct btrfs_new_inode_args whiteout_args = {
8993 .dir = old_dir,
8994 .dentry = old_dentry,
8995 };
8996 struct btrfs_trans_handle *trans;
8997 unsigned int trans_num_items;
8998 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8999 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9000 struct inode *new_inode = d_inode(new_dentry);
9001 struct inode *old_inode = d_inode(old_dentry);
9002 struct btrfs_rename_ctx rename_ctx;
9003 u64 index = 0;
9004 int ret;
9005 int ret2;
9006 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9007 struct fscrypt_name old_fname, new_fname;
9008
9009 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9010 return -EPERM;
9011
9012 /* we only allow rename subvolume link between subvolumes */
9013 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9014 return -EXDEV;
9015
9016 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9017 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9018 return -ENOTEMPTY;
9019
9020 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9021 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9022 return -ENOTEMPTY;
9023
9024 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
9025 if (ret)
9026 return ret;
9027
9028 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
9029 if (ret) {
9030 fscrypt_free_filename(&old_fname);
9031 return ret;
9032 }
9033
9034 /* check for collisions, even if the name isn't there */
9035 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
9036 if (ret) {
9037 if (ret == -EEXIST) {
9038 /* we shouldn't get
9039 * eexist without a new_inode */
9040 if (WARN_ON(!new_inode)) {
9041 goto out_fscrypt_names;
9042 }
9043 } else {
9044 /* maybe -EOVERFLOW */
9045 goto out_fscrypt_names;
9046 }
9047 }
9048 ret = 0;
9049
9050 /*
9051 * we're using rename to replace one file with another. Start IO on it
9052 * now so we don't add too much work to the end of the transaction
9053 */
9054 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9055 filemap_flush(old_inode->i_mapping);
9056
9057 if (flags & RENAME_WHITEOUT) {
9058 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9059 if (!whiteout_args.inode) {
9060 ret = -ENOMEM;
9061 goto out_fscrypt_names;
9062 }
9063 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9064 if (ret)
9065 goto out_whiteout_inode;
9066 } else {
9067 /* 1 to update the old parent inode. */
9068 trans_num_items = 1;
9069 }
9070
9071 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9072 /* Close the race window with snapshot create/destroy ioctl */
9073 down_read(&fs_info->subvol_sem);
9074 /*
9075 * 1 to remove old root ref
9076 * 1 to remove old root backref
9077 * 1 to add new root ref
9078 * 1 to add new root backref
9079 */
9080 trans_num_items += 4;
9081 } else {
9082 /*
9083 * 1 to update inode
9084 * 1 to remove old inode ref
9085 * 1 to add new inode ref
9086 */
9087 trans_num_items += 3;
9088 }
9089 /*
9090 * 1 to remove old dir item
9091 * 1 to remove old dir index
9092 * 1 to add new dir item
9093 * 1 to add new dir index
9094 */
9095 trans_num_items += 4;
9096 /* 1 to update new parent inode if it's not the same as the old parent */
9097 if (new_dir != old_dir)
9098 trans_num_items++;
9099 if (new_inode) {
9100 /*
9101 * 1 to update inode
9102 * 1 to remove inode ref
9103 * 1 to remove dir item
9104 * 1 to remove dir index
9105 * 1 to possibly add orphan item
9106 */
9107 trans_num_items += 5;
9108 }
9109 trans = btrfs_start_transaction(root, trans_num_items);
9110 if (IS_ERR(trans)) {
9111 ret = PTR_ERR(trans);
9112 goto out_notrans;
9113 }
9114
9115 if (dest != root) {
9116 ret = btrfs_record_root_in_trans(trans, dest);
9117 if (ret)
9118 goto out_fail;
9119 }
9120
9121 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9122 if (ret)
9123 goto out_fail;
9124
9125 BTRFS_I(old_inode)->dir_index = 0ULL;
9126 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9127 /* force full log commit if subvolume involved. */
9128 btrfs_set_log_full_commit(trans);
9129 } else {
9130 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9131 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9132 index);
9133 if (ret)
9134 goto out_fail;
9135 }
9136
9137 inode_inc_iversion(old_dir);
9138 inode_inc_iversion(new_dir);
9139 inode_inc_iversion(old_inode);
9140 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
9141
9142 if (old_dentry->d_parent != new_dentry->d_parent)
9143 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9144 BTRFS_I(old_inode), true);
9145
9146 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9147 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9148 } else {
9149 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9150 BTRFS_I(d_inode(old_dentry)),
9151 &old_fname.disk_name, &rename_ctx);
9152 if (!ret)
9153 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9154 }
9155 if (ret) {
9156 btrfs_abort_transaction(trans, ret);
9157 goto out_fail;
9158 }
9159
9160 if (new_inode) {
9161 inode_inc_iversion(new_inode);
9162 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9163 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9164 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9165 BUG_ON(new_inode->i_nlink == 0);
9166 } else {
9167 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9168 BTRFS_I(d_inode(new_dentry)),
9169 &new_fname.disk_name);
9170 }
9171 if (!ret && new_inode->i_nlink == 0)
9172 ret = btrfs_orphan_add(trans,
9173 BTRFS_I(d_inode(new_dentry)));
9174 if (ret) {
9175 btrfs_abort_transaction(trans, ret);
9176 goto out_fail;
9177 }
9178 }
9179
9180 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9181 &new_fname.disk_name, 0, index);
9182 if (ret) {
9183 btrfs_abort_transaction(trans, ret);
9184 goto out_fail;
9185 }
9186
9187 if (old_inode->i_nlink == 1)
9188 BTRFS_I(old_inode)->dir_index = index;
9189
9190 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9191 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9192 rename_ctx.index, new_dentry->d_parent);
9193
9194 if (flags & RENAME_WHITEOUT) {
9195 ret = btrfs_create_new_inode(trans, &whiteout_args);
9196 if (ret) {
9197 btrfs_abort_transaction(trans, ret);
9198 goto out_fail;
9199 } else {
9200 unlock_new_inode(whiteout_args.inode);
9201 iput(whiteout_args.inode);
9202 whiteout_args.inode = NULL;
9203 }
9204 }
9205 out_fail:
9206 ret2 = btrfs_end_transaction(trans);
9207 ret = ret ? ret : ret2;
9208 out_notrans:
9209 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9210 up_read(&fs_info->subvol_sem);
9211 if (flags & RENAME_WHITEOUT)
9212 btrfs_new_inode_args_destroy(&whiteout_args);
9213 out_whiteout_inode:
9214 if (flags & RENAME_WHITEOUT)
9215 iput(whiteout_args.inode);
9216 out_fscrypt_names:
9217 fscrypt_free_filename(&old_fname);
9218 fscrypt_free_filename(&new_fname);
9219 return ret;
9220 }
9221
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)9222 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9223 struct dentry *old_dentry, struct inode *new_dir,
9224 struct dentry *new_dentry, unsigned int flags)
9225 {
9226 int ret;
9227
9228 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9229 return -EINVAL;
9230
9231 if (flags & RENAME_EXCHANGE)
9232 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9233 new_dentry);
9234 else
9235 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9236 new_dentry, flags);
9237
9238 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9239
9240 return ret;
9241 }
9242
9243 struct btrfs_delalloc_work {
9244 struct inode *inode;
9245 struct completion completion;
9246 struct list_head list;
9247 struct btrfs_work work;
9248 };
9249
btrfs_run_delalloc_work(struct btrfs_work * work)9250 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9251 {
9252 struct btrfs_delalloc_work *delalloc_work;
9253 struct inode *inode;
9254
9255 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9256 work);
9257 inode = delalloc_work->inode;
9258 filemap_flush(inode->i_mapping);
9259 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9260 &BTRFS_I(inode)->runtime_flags))
9261 filemap_flush(inode->i_mapping);
9262
9263 iput(inode);
9264 complete(&delalloc_work->completion);
9265 }
9266
btrfs_alloc_delalloc_work(struct inode * inode)9267 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9268 {
9269 struct btrfs_delalloc_work *work;
9270
9271 work = kmalloc(sizeof(*work), GFP_NOFS);
9272 if (!work)
9273 return NULL;
9274
9275 init_completion(&work->completion);
9276 INIT_LIST_HEAD(&work->list);
9277 work->inode = inode;
9278 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9279
9280 return work;
9281 }
9282
9283 /*
9284 * some fairly slow code that needs optimization. This walks the list
9285 * of all the inodes with pending delalloc and forces them to disk.
9286 */
start_delalloc_inodes(struct btrfs_root * root,struct writeback_control * wbc,bool snapshot,bool in_reclaim_context)9287 static int start_delalloc_inodes(struct btrfs_root *root,
9288 struct writeback_control *wbc, bool snapshot,
9289 bool in_reclaim_context)
9290 {
9291 struct btrfs_inode *binode;
9292 struct inode *inode;
9293 struct btrfs_delalloc_work *work, *next;
9294 LIST_HEAD(works);
9295 LIST_HEAD(splice);
9296 int ret = 0;
9297 bool full_flush = wbc->nr_to_write == LONG_MAX;
9298
9299 mutex_lock(&root->delalloc_mutex);
9300 spin_lock(&root->delalloc_lock);
9301 list_splice_init(&root->delalloc_inodes, &splice);
9302 while (!list_empty(&splice)) {
9303 binode = list_entry(splice.next, struct btrfs_inode,
9304 delalloc_inodes);
9305
9306 list_move_tail(&binode->delalloc_inodes,
9307 &root->delalloc_inodes);
9308
9309 if (in_reclaim_context &&
9310 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9311 continue;
9312
9313 inode = igrab(&binode->vfs_inode);
9314 if (!inode) {
9315 cond_resched_lock(&root->delalloc_lock);
9316 continue;
9317 }
9318 spin_unlock(&root->delalloc_lock);
9319
9320 if (snapshot)
9321 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9322 &binode->runtime_flags);
9323 if (full_flush) {
9324 work = btrfs_alloc_delalloc_work(inode);
9325 if (!work) {
9326 iput(inode);
9327 ret = -ENOMEM;
9328 goto out;
9329 }
9330 list_add_tail(&work->list, &works);
9331 btrfs_queue_work(root->fs_info->flush_workers,
9332 &work->work);
9333 } else {
9334 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9335 btrfs_add_delayed_iput(BTRFS_I(inode));
9336 if (ret || wbc->nr_to_write <= 0)
9337 goto out;
9338 }
9339 cond_resched();
9340 spin_lock(&root->delalloc_lock);
9341 }
9342 spin_unlock(&root->delalloc_lock);
9343
9344 out:
9345 list_for_each_entry_safe(work, next, &works, list) {
9346 list_del_init(&work->list);
9347 wait_for_completion(&work->completion);
9348 kfree(work);
9349 }
9350
9351 if (!list_empty(&splice)) {
9352 spin_lock(&root->delalloc_lock);
9353 list_splice_tail(&splice, &root->delalloc_inodes);
9354 spin_unlock(&root->delalloc_lock);
9355 }
9356 mutex_unlock(&root->delalloc_mutex);
9357 return ret;
9358 }
9359
btrfs_start_delalloc_snapshot(struct btrfs_root * root,bool in_reclaim_context)9360 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9361 {
9362 struct writeback_control wbc = {
9363 .nr_to_write = LONG_MAX,
9364 .sync_mode = WB_SYNC_NONE,
9365 .range_start = 0,
9366 .range_end = LLONG_MAX,
9367 };
9368 struct btrfs_fs_info *fs_info = root->fs_info;
9369
9370 if (BTRFS_FS_ERROR(fs_info))
9371 return -EROFS;
9372
9373 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9374 }
9375
btrfs_start_delalloc_roots(struct btrfs_fs_info * fs_info,long nr,bool in_reclaim_context)9376 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9377 bool in_reclaim_context)
9378 {
9379 struct writeback_control wbc = {
9380 .nr_to_write = nr,
9381 .sync_mode = WB_SYNC_NONE,
9382 .range_start = 0,
9383 .range_end = LLONG_MAX,
9384 };
9385 struct btrfs_root *root;
9386 LIST_HEAD(splice);
9387 int ret;
9388
9389 if (BTRFS_FS_ERROR(fs_info))
9390 return -EROFS;
9391
9392 mutex_lock(&fs_info->delalloc_root_mutex);
9393 spin_lock(&fs_info->delalloc_root_lock);
9394 list_splice_init(&fs_info->delalloc_roots, &splice);
9395 while (!list_empty(&splice)) {
9396 /*
9397 * Reset nr_to_write here so we know that we're doing a full
9398 * flush.
9399 */
9400 if (nr == LONG_MAX)
9401 wbc.nr_to_write = LONG_MAX;
9402
9403 root = list_first_entry(&splice, struct btrfs_root,
9404 delalloc_root);
9405 root = btrfs_grab_root(root);
9406 BUG_ON(!root);
9407 list_move_tail(&root->delalloc_root,
9408 &fs_info->delalloc_roots);
9409 spin_unlock(&fs_info->delalloc_root_lock);
9410
9411 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9412 btrfs_put_root(root);
9413 if (ret < 0 || wbc.nr_to_write <= 0)
9414 goto out;
9415 spin_lock(&fs_info->delalloc_root_lock);
9416 }
9417 spin_unlock(&fs_info->delalloc_root_lock);
9418
9419 ret = 0;
9420 out:
9421 if (!list_empty(&splice)) {
9422 spin_lock(&fs_info->delalloc_root_lock);
9423 list_splice_tail(&splice, &fs_info->delalloc_roots);
9424 spin_unlock(&fs_info->delalloc_root_lock);
9425 }
9426 mutex_unlock(&fs_info->delalloc_root_mutex);
9427 return ret;
9428 }
9429
btrfs_symlink(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,const char * symname)9430 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9431 struct dentry *dentry, const char *symname)
9432 {
9433 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9434 struct btrfs_trans_handle *trans;
9435 struct btrfs_root *root = BTRFS_I(dir)->root;
9436 struct btrfs_path *path;
9437 struct btrfs_key key;
9438 struct inode *inode;
9439 struct btrfs_new_inode_args new_inode_args = {
9440 .dir = dir,
9441 .dentry = dentry,
9442 };
9443 unsigned int trans_num_items;
9444 int err;
9445 int name_len;
9446 int datasize;
9447 unsigned long ptr;
9448 struct btrfs_file_extent_item *ei;
9449 struct extent_buffer *leaf;
9450
9451 name_len = strlen(symname);
9452 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9453 return -ENAMETOOLONG;
9454
9455 inode = new_inode(dir->i_sb);
9456 if (!inode)
9457 return -ENOMEM;
9458 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9459 inode->i_op = &btrfs_symlink_inode_operations;
9460 inode_nohighmem(inode);
9461 inode->i_mapping->a_ops = &btrfs_aops;
9462 btrfs_i_size_write(BTRFS_I(inode), name_len);
9463 inode_set_bytes(inode, name_len);
9464
9465 new_inode_args.inode = inode;
9466 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9467 if (err)
9468 goto out_inode;
9469 /* 1 additional item for the inline extent */
9470 trans_num_items++;
9471
9472 trans = btrfs_start_transaction(root, trans_num_items);
9473 if (IS_ERR(trans)) {
9474 err = PTR_ERR(trans);
9475 goto out_new_inode_args;
9476 }
9477
9478 err = btrfs_create_new_inode(trans, &new_inode_args);
9479 if (err)
9480 goto out;
9481
9482 path = btrfs_alloc_path();
9483 if (!path) {
9484 err = -ENOMEM;
9485 btrfs_abort_transaction(trans, err);
9486 discard_new_inode(inode);
9487 inode = NULL;
9488 goto out;
9489 }
9490 key.objectid = btrfs_ino(BTRFS_I(inode));
9491 key.offset = 0;
9492 key.type = BTRFS_EXTENT_DATA_KEY;
9493 datasize = btrfs_file_extent_calc_inline_size(name_len);
9494 err = btrfs_insert_empty_item(trans, root, path, &key,
9495 datasize);
9496 if (err) {
9497 btrfs_abort_transaction(trans, err);
9498 btrfs_free_path(path);
9499 discard_new_inode(inode);
9500 inode = NULL;
9501 goto out;
9502 }
9503 leaf = path->nodes[0];
9504 ei = btrfs_item_ptr(leaf, path->slots[0],
9505 struct btrfs_file_extent_item);
9506 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9507 btrfs_set_file_extent_type(leaf, ei,
9508 BTRFS_FILE_EXTENT_INLINE);
9509 btrfs_set_file_extent_encryption(leaf, ei, 0);
9510 btrfs_set_file_extent_compression(leaf, ei, 0);
9511 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9512 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9513
9514 ptr = btrfs_file_extent_inline_start(ei);
9515 write_extent_buffer(leaf, symname, ptr, name_len);
9516 btrfs_mark_buffer_dirty(trans, leaf);
9517 btrfs_free_path(path);
9518
9519 d_instantiate_new(dentry, inode);
9520 err = 0;
9521 out:
9522 btrfs_end_transaction(trans);
9523 btrfs_btree_balance_dirty(fs_info);
9524 out_new_inode_args:
9525 btrfs_new_inode_args_destroy(&new_inode_args);
9526 out_inode:
9527 if (err)
9528 iput(inode);
9529 return err;
9530 }
9531
insert_prealloc_file_extent(struct btrfs_trans_handle * trans_in,struct btrfs_inode * inode,struct btrfs_key * ins,u64 file_offset)9532 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9533 struct btrfs_trans_handle *trans_in,
9534 struct btrfs_inode *inode,
9535 struct btrfs_key *ins,
9536 u64 file_offset)
9537 {
9538 struct btrfs_file_extent_item stack_fi;
9539 struct btrfs_replace_extent_info extent_info;
9540 struct btrfs_trans_handle *trans = trans_in;
9541 struct btrfs_path *path;
9542 u64 start = ins->objectid;
9543 u64 len = ins->offset;
9544 u64 qgroup_released = 0;
9545 int ret;
9546
9547 memset(&stack_fi, 0, sizeof(stack_fi));
9548
9549 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9550 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9551 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9552 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9553 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9554 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9555 /* Encryption and other encoding is reserved and all 0 */
9556
9557 ret = btrfs_qgroup_release_data(inode, file_offset, len, &qgroup_released);
9558 if (ret < 0)
9559 return ERR_PTR(ret);
9560
9561 if (trans) {
9562 ret = insert_reserved_file_extent(trans, inode,
9563 file_offset, &stack_fi,
9564 true, qgroup_released);
9565 if (ret)
9566 goto free_qgroup;
9567 return trans;
9568 }
9569
9570 extent_info.disk_offset = start;
9571 extent_info.disk_len = len;
9572 extent_info.data_offset = 0;
9573 extent_info.data_len = len;
9574 extent_info.file_offset = file_offset;
9575 extent_info.extent_buf = (char *)&stack_fi;
9576 extent_info.is_new_extent = true;
9577 extent_info.update_times = true;
9578 extent_info.qgroup_reserved = qgroup_released;
9579 extent_info.insertions = 0;
9580
9581 path = btrfs_alloc_path();
9582 if (!path) {
9583 ret = -ENOMEM;
9584 goto free_qgroup;
9585 }
9586
9587 ret = btrfs_replace_file_extents(inode, path, file_offset,
9588 file_offset + len - 1, &extent_info,
9589 &trans);
9590 btrfs_free_path(path);
9591 if (ret)
9592 goto free_qgroup;
9593 return trans;
9594
9595 free_qgroup:
9596 /*
9597 * We have released qgroup data range at the beginning of the function,
9598 * and normally qgroup_released bytes will be freed when committing
9599 * transaction.
9600 * But if we error out early, we have to free what we have released
9601 * or we leak qgroup data reservation.
9602 */
9603 btrfs_qgroup_free_refroot(inode->root->fs_info,
9604 inode->root->root_key.objectid, qgroup_released,
9605 BTRFS_QGROUP_RSV_DATA);
9606 return ERR_PTR(ret);
9607 }
9608
__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)9609 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9610 u64 start, u64 num_bytes, u64 min_size,
9611 loff_t actual_len, u64 *alloc_hint,
9612 struct btrfs_trans_handle *trans)
9613 {
9614 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9615 struct extent_map *em;
9616 struct btrfs_root *root = BTRFS_I(inode)->root;
9617 struct btrfs_key ins;
9618 u64 cur_offset = start;
9619 u64 clear_offset = start;
9620 u64 i_size;
9621 u64 cur_bytes;
9622 u64 last_alloc = (u64)-1;
9623 int ret = 0;
9624 bool own_trans = true;
9625 u64 end = start + num_bytes - 1;
9626
9627 if (trans)
9628 own_trans = false;
9629 while (num_bytes > 0) {
9630 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9631 cur_bytes = max(cur_bytes, min_size);
9632 /*
9633 * If we are severely fragmented we could end up with really
9634 * small allocations, so if the allocator is returning small
9635 * chunks lets make its job easier by only searching for those
9636 * sized chunks.
9637 */
9638 cur_bytes = min(cur_bytes, last_alloc);
9639 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9640 min_size, 0, *alloc_hint, &ins, 1, 0);
9641 if (ret)
9642 break;
9643
9644 /*
9645 * We've reserved this space, and thus converted it from
9646 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9647 * from here on out we will only need to clear our reservation
9648 * for the remaining unreserved area, so advance our
9649 * clear_offset by our extent size.
9650 */
9651 clear_offset += ins.offset;
9652
9653 last_alloc = ins.offset;
9654 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9655 &ins, cur_offset);
9656 /*
9657 * Now that we inserted the prealloc extent we can finally
9658 * decrement the number of reservations in the block group.
9659 * If we did it before, we could race with relocation and have
9660 * relocation miss the reserved extent, making it fail later.
9661 */
9662 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9663 if (IS_ERR(trans)) {
9664 ret = PTR_ERR(trans);
9665 btrfs_free_reserved_extent(fs_info, ins.objectid,
9666 ins.offset, 0);
9667 break;
9668 }
9669
9670 em = alloc_extent_map();
9671 if (!em) {
9672 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9673 cur_offset + ins.offset - 1, false);
9674 btrfs_set_inode_full_sync(BTRFS_I(inode));
9675 goto next;
9676 }
9677
9678 em->start = cur_offset;
9679 em->orig_start = cur_offset;
9680 em->len = ins.offset;
9681 em->block_start = ins.objectid;
9682 em->block_len = ins.offset;
9683 em->orig_block_len = ins.offset;
9684 em->ram_bytes = ins.offset;
9685 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9686 em->generation = trans->transid;
9687
9688 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9689 free_extent_map(em);
9690 next:
9691 num_bytes -= ins.offset;
9692 cur_offset += ins.offset;
9693 *alloc_hint = ins.objectid + ins.offset;
9694
9695 inode_inc_iversion(inode);
9696 inode_set_ctime_current(inode);
9697 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9698 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9699 (actual_len > inode->i_size) &&
9700 (cur_offset > inode->i_size)) {
9701 if (cur_offset > actual_len)
9702 i_size = actual_len;
9703 else
9704 i_size = cur_offset;
9705 i_size_write(inode, i_size);
9706 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9707 }
9708
9709 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9710
9711 if (ret) {
9712 btrfs_abort_transaction(trans, ret);
9713 if (own_trans)
9714 btrfs_end_transaction(trans);
9715 break;
9716 }
9717
9718 if (own_trans) {
9719 btrfs_end_transaction(trans);
9720 trans = NULL;
9721 }
9722 }
9723 if (clear_offset < end)
9724 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9725 end - clear_offset + 1);
9726 return ret;
9727 }
9728
btrfs_prealloc_file_range(struct inode * inode,int mode,u64 start,u64 num_bytes,u64 min_size,loff_t actual_len,u64 * alloc_hint)9729 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9730 u64 start, u64 num_bytes, u64 min_size,
9731 loff_t actual_len, u64 *alloc_hint)
9732 {
9733 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9734 min_size, actual_len, alloc_hint,
9735 NULL);
9736 }
9737
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)9738 int btrfs_prealloc_file_range_trans(struct inode *inode,
9739 struct btrfs_trans_handle *trans, int mode,
9740 u64 start, u64 num_bytes, u64 min_size,
9741 loff_t actual_len, u64 *alloc_hint)
9742 {
9743 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9744 min_size, actual_len, alloc_hint, trans);
9745 }
9746
btrfs_permission(struct mnt_idmap * idmap,struct inode * inode,int mask)9747 static int btrfs_permission(struct mnt_idmap *idmap,
9748 struct inode *inode, int mask)
9749 {
9750 struct btrfs_root *root = BTRFS_I(inode)->root;
9751 umode_t mode = inode->i_mode;
9752
9753 if (mask & MAY_WRITE &&
9754 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9755 if (btrfs_root_readonly(root))
9756 return -EROFS;
9757 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9758 return -EACCES;
9759 }
9760 return generic_permission(idmap, inode, mask);
9761 }
9762
btrfs_tmpfile(struct mnt_idmap * idmap,struct inode * dir,struct file * file,umode_t mode)9763 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9764 struct file *file, umode_t mode)
9765 {
9766 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9767 struct btrfs_trans_handle *trans;
9768 struct btrfs_root *root = BTRFS_I(dir)->root;
9769 struct inode *inode;
9770 struct btrfs_new_inode_args new_inode_args = {
9771 .dir = dir,
9772 .dentry = file->f_path.dentry,
9773 .orphan = true,
9774 };
9775 unsigned int trans_num_items;
9776 int ret;
9777
9778 inode = new_inode(dir->i_sb);
9779 if (!inode)
9780 return -ENOMEM;
9781 inode_init_owner(idmap, inode, dir, mode);
9782 inode->i_fop = &btrfs_file_operations;
9783 inode->i_op = &btrfs_file_inode_operations;
9784 inode->i_mapping->a_ops = &btrfs_aops;
9785
9786 new_inode_args.inode = inode;
9787 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9788 if (ret)
9789 goto out_inode;
9790
9791 trans = btrfs_start_transaction(root, trans_num_items);
9792 if (IS_ERR(trans)) {
9793 ret = PTR_ERR(trans);
9794 goto out_new_inode_args;
9795 }
9796
9797 ret = btrfs_create_new_inode(trans, &new_inode_args);
9798
9799 /*
9800 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9801 * set it to 1 because d_tmpfile() will issue a warning if the count is
9802 * 0, through:
9803 *
9804 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9805 */
9806 set_nlink(inode, 1);
9807
9808 if (!ret) {
9809 d_tmpfile(file, inode);
9810 unlock_new_inode(inode);
9811 mark_inode_dirty(inode);
9812 }
9813
9814 btrfs_end_transaction(trans);
9815 btrfs_btree_balance_dirty(fs_info);
9816 out_new_inode_args:
9817 btrfs_new_inode_args_destroy(&new_inode_args);
9818 out_inode:
9819 if (ret)
9820 iput(inode);
9821 return finish_open_simple(file, ret);
9822 }
9823
btrfs_set_range_writeback(struct btrfs_inode * inode,u64 start,u64 end)9824 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9825 {
9826 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9827 unsigned long index = start >> PAGE_SHIFT;
9828 unsigned long end_index = end >> PAGE_SHIFT;
9829 struct page *page;
9830 u32 len;
9831
9832 ASSERT(end + 1 - start <= U32_MAX);
9833 len = end + 1 - start;
9834 while (index <= end_index) {
9835 page = find_get_page(inode->vfs_inode.i_mapping, index);
9836 ASSERT(page); /* Pages should be in the extent_io_tree */
9837
9838 btrfs_page_set_writeback(fs_info, page, start, len);
9839 put_page(page);
9840 index++;
9841 }
9842 }
9843
btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info * fs_info,int compress_type)9844 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9845 int compress_type)
9846 {
9847 switch (compress_type) {
9848 case BTRFS_COMPRESS_NONE:
9849 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9850 case BTRFS_COMPRESS_ZLIB:
9851 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9852 case BTRFS_COMPRESS_LZO:
9853 /*
9854 * The LZO format depends on the sector size. 64K is the maximum
9855 * sector size that we support.
9856 */
9857 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9858 return -EINVAL;
9859 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9860 (fs_info->sectorsize_bits - 12);
9861 case BTRFS_COMPRESS_ZSTD:
9862 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9863 default:
9864 return -EUCLEAN;
9865 }
9866 }
9867
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)9868 static ssize_t btrfs_encoded_read_inline(
9869 struct kiocb *iocb,
9870 struct iov_iter *iter, u64 start,
9871 u64 lockend,
9872 struct extent_state **cached_state,
9873 u64 extent_start, size_t count,
9874 struct btrfs_ioctl_encoded_io_args *encoded,
9875 bool *unlocked)
9876 {
9877 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9878 struct btrfs_root *root = inode->root;
9879 struct btrfs_fs_info *fs_info = root->fs_info;
9880 struct extent_io_tree *io_tree = &inode->io_tree;
9881 struct btrfs_path *path;
9882 struct extent_buffer *leaf;
9883 struct btrfs_file_extent_item *item;
9884 u64 ram_bytes;
9885 unsigned long ptr;
9886 void *tmp;
9887 ssize_t ret;
9888
9889 path = btrfs_alloc_path();
9890 if (!path) {
9891 ret = -ENOMEM;
9892 goto out;
9893 }
9894 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9895 extent_start, 0);
9896 if (ret) {
9897 if (ret > 0) {
9898 /* The extent item disappeared? */
9899 ret = -EIO;
9900 }
9901 goto out;
9902 }
9903 leaf = path->nodes[0];
9904 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9905
9906 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9907 ptr = btrfs_file_extent_inline_start(item);
9908
9909 encoded->len = min_t(u64, extent_start + ram_bytes,
9910 inode->vfs_inode.i_size) - iocb->ki_pos;
9911 ret = btrfs_encoded_io_compression_from_extent(fs_info,
9912 btrfs_file_extent_compression(leaf, item));
9913 if (ret < 0)
9914 goto out;
9915 encoded->compression = ret;
9916 if (encoded->compression) {
9917 size_t inline_size;
9918
9919 inline_size = btrfs_file_extent_inline_item_len(leaf,
9920 path->slots[0]);
9921 if (inline_size > count) {
9922 ret = -ENOBUFS;
9923 goto out;
9924 }
9925 count = inline_size;
9926 encoded->unencoded_len = ram_bytes;
9927 encoded->unencoded_offset = iocb->ki_pos - extent_start;
9928 } else {
9929 count = min_t(u64, count, encoded->len);
9930 encoded->len = count;
9931 encoded->unencoded_len = count;
9932 ptr += iocb->ki_pos - extent_start;
9933 }
9934
9935 tmp = kmalloc(count, GFP_NOFS);
9936 if (!tmp) {
9937 ret = -ENOMEM;
9938 goto out;
9939 }
9940 read_extent_buffer(leaf, tmp, ptr, count);
9941 btrfs_release_path(path);
9942 unlock_extent(io_tree, start, lockend, cached_state);
9943 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
9944 *unlocked = true;
9945
9946 ret = copy_to_iter(tmp, count, iter);
9947 if (ret != count)
9948 ret = -EFAULT;
9949 kfree(tmp);
9950 out:
9951 btrfs_free_path(path);
9952 return ret;
9953 }
9954
9955 struct btrfs_encoded_read_private {
9956 wait_queue_head_t wait;
9957 atomic_t pending;
9958 blk_status_t status;
9959 };
9960
btrfs_encoded_read_endio(struct btrfs_bio * bbio)9961 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
9962 {
9963 struct btrfs_encoded_read_private *priv = bbio->private;
9964
9965 if (bbio->bio.bi_status) {
9966 /*
9967 * The memory barrier implied by the atomic_dec_return() here
9968 * pairs with the memory barrier implied by the
9969 * atomic_dec_return() or io_wait_event() in
9970 * btrfs_encoded_read_regular_fill_pages() to ensure that this
9971 * write is observed before the load of status in
9972 * btrfs_encoded_read_regular_fill_pages().
9973 */
9974 WRITE_ONCE(priv->status, bbio->bio.bi_status);
9975 }
9976 if (!atomic_dec_return(&priv->pending))
9977 wake_up(&priv->wait);
9978 bio_put(&bbio->bio);
9979 }
9980
btrfs_encoded_read_regular_fill_pages(struct btrfs_inode * inode,u64 file_offset,u64 disk_bytenr,u64 disk_io_size,struct page ** pages)9981 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
9982 u64 file_offset, u64 disk_bytenr,
9983 u64 disk_io_size, struct page **pages)
9984 {
9985 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9986 struct btrfs_encoded_read_private priv = {
9987 .pending = ATOMIC_INIT(1),
9988 };
9989 unsigned long i = 0;
9990 struct btrfs_bio *bbio;
9991
9992 init_waitqueue_head(&priv.wait);
9993
9994 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9995 btrfs_encoded_read_endio, &priv);
9996 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9997 bbio->inode = inode;
9998
9999 do {
10000 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
10001
10002 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
10003 atomic_inc(&priv.pending);
10004 btrfs_submit_bio(bbio, 0);
10005
10006 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10007 btrfs_encoded_read_endio, &priv);
10008 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10009 bbio->inode = inode;
10010 continue;
10011 }
10012
10013 i++;
10014 disk_bytenr += bytes;
10015 disk_io_size -= bytes;
10016 } while (disk_io_size);
10017
10018 atomic_inc(&priv.pending);
10019 btrfs_submit_bio(bbio, 0);
10020
10021 if (atomic_dec_return(&priv.pending))
10022 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10023 /* See btrfs_encoded_read_endio() for ordering. */
10024 return blk_status_to_errno(READ_ONCE(priv.status));
10025 }
10026
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)10027 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10028 struct iov_iter *iter,
10029 u64 start, u64 lockend,
10030 struct extent_state **cached_state,
10031 u64 disk_bytenr, u64 disk_io_size,
10032 size_t count, bool compressed,
10033 bool *unlocked)
10034 {
10035 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10036 struct extent_io_tree *io_tree = &inode->io_tree;
10037 struct page **pages;
10038 unsigned long nr_pages, i;
10039 u64 cur;
10040 size_t page_offset;
10041 ssize_t ret;
10042
10043 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10044 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10045 if (!pages)
10046 return -ENOMEM;
10047 ret = btrfs_alloc_page_array(nr_pages, pages);
10048 if (ret) {
10049 ret = -ENOMEM;
10050 goto out;
10051 }
10052
10053 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10054 disk_io_size, pages);
10055 if (ret)
10056 goto out;
10057
10058 unlock_extent(io_tree, start, lockend, cached_state);
10059 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10060 *unlocked = true;
10061
10062 if (compressed) {
10063 i = 0;
10064 page_offset = 0;
10065 } else {
10066 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10067 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10068 }
10069 cur = 0;
10070 while (cur < count) {
10071 size_t bytes = min_t(size_t, count - cur,
10072 PAGE_SIZE - page_offset);
10073
10074 if (copy_page_to_iter(pages[i], page_offset, bytes,
10075 iter) != bytes) {
10076 ret = -EFAULT;
10077 goto out;
10078 }
10079 i++;
10080 cur += bytes;
10081 page_offset = 0;
10082 }
10083 ret = count;
10084 out:
10085 for (i = 0; i < nr_pages; i++) {
10086 if (pages[i])
10087 __free_page(pages[i]);
10088 }
10089 kfree(pages);
10090 return ret;
10091 }
10092
btrfs_encoded_read(struct kiocb * iocb,struct iov_iter * iter,struct btrfs_ioctl_encoded_io_args * encoded)10093 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10094 struct btrfs_ioctl_encoded_io_args *encoded)
10095 {
10096 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10097 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10098 struct extent_io_tree *io_tree = &inode->io_tree;
10099 ssize_t ret;
10100 size_t count = iov_iter_count(iter);
10101 u64 start, lockend, disk_bytenr, disk_io_size;
10102 struct extent_state *cached_state = NULL;
10103 struct extent_map *em;
10104 bool unlocked = false;
10105
10106 file_accessed(iocb->ki_filp);
10107
10108 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10109
10110 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10111 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10112 return 0;
10113 }
10114 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10115 /*
10116 * We don't know how long the extent containing iocb->ki_pos is, but if
10117 * it's compressed we know that it won't be longer than this.
10118 */
10119 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10120
10121 for (;;) {
10122 struct btrfs_ordered_extent *ordered;
10123
10124 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10125 lockend - start + 1);
10126 if (ret)
10127 goto out_unlock_inode;
10128 lock_extent(io_tree, start, lockend, &cached_state);
10129 ordered = btrfs_lookup_ordered_range(inode, start,
10130 lockend - start + 1);
10131 if (!ordered)
10132 break;
10133 btrfs_put_ordered_extent(ordered);
10134 unlock_extent(io_tree, start, lockend, &cached_state);
10135 cond_resched();
10136 }
10137
10138 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10139 if (IS_ERR(em)) {
10140 ret = PTR_ERR(em);
10141 goto out_unlock_extent;
10142 }
10143
10144 if (em->block_start == EXTENT_MAP_INLINE) {
10145 u64 extent_start = em->start;
10146
10147 /*
10148 * For inline extents we get everything we need out of the
10149 * extent item.
10150 */
10151 free_extent_map(em);
10152 em = NULL;
10153 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10154 &cached_state, extent_start,
10155 count, encoded, &unlocked);
10156 goto out;
10157 }
10158
10159 /*
10160 * We only want to return up to EOF even if the extent extends beyond
10161 * that.
10162 */
10163 encoded->len = min_t(u64, extent_map_end(em),
10164 inode->vfs_inode.i_size) - iocb->ki_pos;
10165 if (em->block_start == EXTENT_MAP_HOLE ||
10166 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10167 disk_bytenr = EXTENT_MAP_HOLE;
10168 count = min_t(u64, count, encoded->len);
10169 encoded->len = count;
10170 encoded->unencoded_len = count;
10171 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10172 disk_bytenr = em->block_start;
10173 /*
10174 * Bail if the buffer isn't large enough to return the whole
10175 * compressed extent.
10176 */
10177 if (em->block_len > count) {
10178 ret = -ENOBUFS;
10179 goto out_em;
10180 }
10181 disk_io_size = em->block_len;
10182 count = em->block_len;
10183 encoded->unencoded_len = em->ram_bytes;
10184 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10185 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10186 em->compress_type);
10187 if (ret < 0)
10188 goto out_em;
10189 encoded->compression = ret;
10190 } else {
10191 disk_bytenr = em->block_start + (start - em->start);
10192 if (encoded->len > count)
10193 encoded->len = count;
10194 /*
10195 * Don't read beyond what we locked. This also limits the page
10196 * allocations that we'll do.
10197 */
10198 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10199 count = start + disk_io_size - iocb->ki_pos;
10200 encoded->len = count;
10201 encoded->unencoded_len = count;
10202 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10203 }
10204 free_extent_map(em);
10205 em = NULL;
10206
10207 if (disk_bytenr == EXTENT_MAP_HOLE) {
10208 unlock_extent(io_tree, start, lockend, &cached_state);
10209 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10210 unlocked = true;
10211 ret = iov_iter_zero(count, iter);
10212 if (ret != count)
10213 ret = -EFAULT;
10214 } else {
10215 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10216 &cached_state, disk_bytenr,
10217 disk_io_size, count,
10218 encoded->compression,
10219 &unlocked);
10220 }
10221
10222 out:
10223 if (ret >= 0)
10224 iocb->ki_pos += encoded->len;
10225 out_em:
10226 free_extent_map(em);
10227 out_unlock_extent:
10228 if (!unlocked)
10229 unlock_extent(io_tree, start, lockend, &cached_state);
10230 out_unlock_inode:
10231 if (!unlocked)
10232 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10233 return ret;
10234 }
10235
btrfs_do_encoded_write(struct kiocb * iocb,struct iov_iter * from,const struct btrfs_ioctl_encoded_io_args * encoded)10236 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10237 const struct btrfs_ioctl_encoded_io_args *encoded)
10238 {
10239 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10240 struct btrfs_root *root = inode->root;
10241 struct btrfs_fs_info *fs_info = root->fs_info;
10242 struct extent_io_tree *io_tree = &inode->io_tree;
10243 struct extent_changeset *data_reserved = NULL;
10244 struct extent_state *cached_state = NULL;
10245 struct btrfs_ordered_extent *ordered;
10246 int compression;
10247 size_t orig_count;
10248 u64 start, end;
10249 u64 num_bytes, ram_bytes, disk_num_bytes;
10250 unsigned long nr_pages, i;
10251 struct page **pages;
10252 struct btrfs_key ins;
10253 bool extent_reserved = false;
10254 struct extent_map *em;
10255 ssize_t ret;
10256
10257 switch (encoded->compression) {
10258 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10259 compression = BTRFS_COMPRESS_ZLIB;
10260 break;
10261 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10262 compression = BTRFS_COMPRESS_ZSTD;
10263 break;
10264 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10265 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10266 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10267 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10268 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10269 /* The sector size must match for LZO. */
10270 if (encoded->compression -
10271 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10272 fs_info->sectorsize_bits)
10273 return -EINVAL;
10274 compression = BTRFS_COMPRESS_LZO;
10275 break;
10276 default:
10277 return -EINVAL;
10278 }
10279 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10280 return -EINVAL;
10281
10282 /*
10283 * Compressed extents should always have checksums, so error out if we
10284 * have a NOCOW file or inode was created while mounted with NODATASUM.
10285 */
10286 if (inode->flags & BTRFS_INODE_NODATASUM)
10287 return -EINVAL;
10288
10289 orig_count = iov_iter_count(from);
10290
10291 /* The extent size must be sane. */
10292 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10293 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10294 return -EINVAL;
10295
10296 /*
10297 * The compressed data must be smaller than the decompressed data.
10298 *
10299 * It's of course possible for data to compress to larger or the same
10300 * size, but the buffered I/O path falls back to no compression for such
10301 * data, and we don't want to break any assumptions by creating these
10302 * extents.
10303 *
10304 * Note that this is less strict than the current check we have that the
10305 * compressed data must be at least one sector smaller than the
10306 * decompressed data. We only want to enforce the weaker requirement
10307 * from old kernels that it is at least one byte smaller.
10308 */
10309 if (orig_count >= encoded->unencoded_len)
10310 return -EINVAL;
10311
10312 /* The extent must start on a sector boundary. */
10313 start = iocb->ki_pos;
10314 if (!IS_ALIGNED(start, fs_info->sectorsize))
10315 return -EINVAL;
10316
10317 /*
10318 * The extent must end on a sector boundary. However, we allow a write
10319 * which ends at or extends i_size to have an unaligned length; we round
10320 * up the extent size and set i_size to the unaligned end.
10321 */
10322 if (start + encoded->len < inode->vfs_inode.i_size &&
10323 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10324 return -EINVAL;
10325
10326 /* Finally, the offset in the unencoded data must be sector-aligned. */
10327 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10328 return -EINVAL;
10329
10330 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10331 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10332 end = start + num_bytes - 1;
10333
10334 /*
10335 * If the extent cannot be inline, the compressed data on disk must be
10336 * sector-aligned. For convenience, we extend it with zeroes if it
10337 * isn't.
10338 */
10339 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10340 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10341 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10342 if (!pages)
10343 return -ENOMEM;
10344 for (i = 0; i < nr_pages; i++) {
10345 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10346 char *kaddr;
10347
10348 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10349 if (!pages[i]) {
10350 ret = -ENOMEM;
10351 goto out_pages;
10352 }
10353 kaddr = kmap_local_page(pages[i]);
10354 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10355 kunmap_local(kaddr);
10356 ret = -EFAULT;
10357 goto out_pages;
10358 }
10359 if (bytes < PAGE_SIZE)
10360 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10361 kunmap_local(kaddr);
10362 }
10363
10364 for (;;) {
10365 struct btrfs_ordered_extent *ordered;
10366
10367 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10368 if (ret)
10369 goto out_pages;
10370 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10371 start >> PAGE_SHIFT,
10372 end >> PAGE_SHIFT);
10373 if (ret)
10374 goto out_pages;
10375 lock_extent(io_tree, start, end, &cached_state);
10376 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10377 if (!ordered &&
10378 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10379 break;
10380 if (ordered)
10381 btrfs_put_ordered_extent(ordered);
10382 unlock_extent(io_tree, start, end, &cached_state);
10383 cond_resched();
10384 }
10385
10386 /*
10387 * We don't use the higher-level delalloc space functions because our
10388 * num_bytes and disk_num_bytes are different.
10389 */
10390 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10391 if (ret)
10392 goto out_unlock;
10393 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10394 if (ret)
10395 goto out_free_data_space;
10396 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10397 false);
10398 if (ret)
10399 goto out_qgroup_free_data;
10400
10401 /* Try an inline extent first. */
10402 if (start == 0 && encoded->unencoded_len == encoded->len &&
10403 encoded->unencoded_offset == 0) {
10404 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10405 compression, pages, true);
10406 if (ret <= 0) {
10407 if (ret == 0)
10408 ret = orig_count;
10409 goto out_delalloc_release;
10410 }
10411 }
10412
10413 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10414 disk_num_bytes, 0, 0, &ins, 1, 1);
10415 if (ret)
10416 goto out_delalloc_release;
10417 extent_reserved = true;
10418
10419 em = create_io_em(inode, start, num_bytes,
10420 start - encoded->unencoded_offset, ins.objectid,
10421 ins.offset, ins.offset, ram_bytes, compression,
10422 BTRFS_ORDERED_COMPRESSED);
10423 if (IS_ERR(em)) {
10424 ret = PTR_ERR(em);
10425 goto out_free_reserved;
10426 }
10427 free_extent_map(em);
10428
10429 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10430 ins.objectid, ins.offset,
10431 encoded->unencoded_offset,
10432 (1 << BTRFS_ORDERED_ENCODED) |
10433 (1 << BTRFS_ORDERED_COMPRESSED),
10434 compression);
10435 if (IS_ERR(ordered)) {
10436 btrfs_drop_extent_map_range(inode, start, end, false);
10437 ret = PTR_ERR(ordered);
10438 goto out_free_reserved;
10439 }
10440 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10441
10442 if (start + encoded->len > inode->vfs_inode.i_size)
10443 i_size_write(&inode->vfs_inode, start + encoded->len);
10444
10445 unlock_extent(io_tree, start, end, &cached_state);
10446
10447 btrfs_delalloc_release_extents(inode, num_bytes);
10448
10449 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10450 ret = orig_count;
10451 goto out;
10452
10453 out_free_reserved:
10454 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10455 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10456 out_delalloc_release:
10457 btrfs_delalloc_release_extents(inode, num_bytes);
10458 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10459 out_qgroup_free_data:
10460 if (ret < 0)
10461 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes, NULL);
10462 out_free_data_space:
10463 /*
10464 * If btrfs_reserve_extent() succeeded, then we already decremented
10465 * bytes_may_use.
10466 */
10467 if (!extent_reserved)
10468 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10469 out_unlock:
10470 unlock_extent(io_tree, start, end, &cached_state);
10471 out_pages:
10472 for (i = 0; i < nr_pages; i++) {
10473 if (pages[i])
10474 __free_page(pages[i]);
10475 }
10476 kvfree(pages);
10477 out:
10478 if (ret >= 0)
10479 iocb->ki_pos += encoded->len;
10480 return ret;
10481 }
10482
10483 #ifdef CONFIG_SWAP
10484 /*
10485 * Add an entry indicating a block group or device which is pinned by a
10486 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10487 * negative errno on failure.
10488 */
btrfs_add_swapfile_pin(struct inode * inode,void * ptr,bool is_block_group)10489 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10490 bool is_block_group)
10491 {
10492 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10493 struct btrfs_swapfile_pin *sp, *entry;
10494 struct rb_node **p;
10495 struct rb_node *parent = NULL;
10496
10497 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10498 if (!sp)
10499 return -ENOMEM;
10500 sp->ptr = ptr;
10501 sp->inode = inode;
10502 sp->is_block_group = is_block_group;
10503 sp->bg_extent_count = 1;
10504
10505 spin_lock(&fs_info->swapfile_pins_lock);
10506 p = &fs_info->swapfile_pins.rb_node;
10507 while (*p) {
10508 parent = *p;
10509 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10510 if (sp->ptr < entry->ptr ||
10511 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10512 p = &(*p)->rb_left;
10513 } else if (sp->ptr > entry->ptr ||
10514 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10515 p = &(*p)->rb_right;
10516 } else {
10517 if (is_block_group)
10518 entry->bg_extent_count++;
10519 spin_unlock(&fs_info->swapfile_pins_lock);
10520 kfree(sp);
10521 return 1;
10522 }
10523 }
10524 rb_link_node(&sp->node, parent, p);
10525 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10526 spin_unlock(&fs_info->swapfile_pins_lock);
10527 return 0;
10528 }
10529
10530 /* Free all of the entries pinned by this swapfile. */
btrfs_free_swapfile_pins(struct inode * inode)10531 static void btrfs_free_swapfile_pins(struct inode *inode)
10532 {
10533 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10534 struct btrfs_swapfile_pin *sp;
10535 struct rb_node *node, *next;
10536
10537 spin_lock(&fs_info->swapfile_pins_lock);
10538 node = rb_first(&fs_info->swapfile_pins);
10539 while (node) {
10540 next = rb_next(node);
10541 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10542 if (sp->inode == inode) {
10543 rb_erase(&sp->node, &fs_info->swapfile_pins);
10544 if (sp->is_block_group) {
10545 btrfs_dec_block_group_swap_extents(sp->ptr,
10546 sp->bg_extent_count);
10547 btrfs_put_block_group(sp->ptr);
10548 }
10549 kfree(sp);
10550 }
10551 node = next;
10552 }
10553 spin_unlock(&fs_info->swapfile_pins_lock);
10554 }
10555
10556 struct btrfs_swap_info {
10557 u64 start;
10558 u64 block_start;
10559 u64 block_len;
10560 u64 lowest_ppage;
10561 u64 highest_ppage;
10562 unsigned long nr_pages;
10563 int nr_extents;
10564 };
10565
btrfs_add_swap_extent(struct swap_info_struct * sis,struct btrfs_swap_info * bsi)10566 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10567 struct btrfs_swap_info *bsi)
10568 {
10569 unsigned long nr_pages;
10570 unsigned long max_pages;
10571 u64 first_ppage, first_ppage_reported, next_ppage;
10572 int ret;
10573
10574 /*
10575 * Our swapfile may have had its size extended after the swap header was
10576 * written. In that case activating the swapfile should not go beyond
10577 * the max size set in the swap header.
10578 */
10579 if (bsi->nr_pages >= sis->max)
10580 return 0;
10581
10582 max_pages = sis->max - bsi->nr_pages;
10583 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10584 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10585
10586 if (first_ppage >= next_ppage)
10587 return 0;
10588 nr_pages = next_ppage - first_ppage;
10589 nr_pages = min(nr_pages, max_pages);
10590
10591 first_ppage_reported = first_ppage;
10592 if (bsi->start == 0)
10593 first_ppage_reported++;
10594 if (bsi->lowest_ppage > first_ppage_reported)
10595 bsi->lowest_ppage = first_ppage_reported;
10596 if (bsi->highest_ppage < (next_ppage - 1))
10597 bsi->highest_ppage = next_ppage - 1;
10598
10599 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10600 if (ret < 0)
10601 return ret;
10602 bsi->nr_extents += ret;
10603 bsi->nr_pages += nr_pages;
10604 return 0;
10605 }
10606
btrfs_swap_deactivate(struct file * file)10607 static void btrfs_swap_deactivate(struct file *file)
10608 {
10609 struct inode *inode = file_inode(file);
10610
10611 btrfs_free_swapfile_pins(inode);
10612 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10613 }
10614
btrfs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)10615 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10616 sector_t *span)
10617 {
10618 struct inode *inode = file_inode(file);
10619 struct btrfs_root *root = BTRFS_I(inode)->root;
10620 struct btrfs_fs_info *fs_info = root->fs_info;
10621 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10622 struct extent_state *cached_state = NULL;
10623 struct extent_map *em = NULL;
10624 struct btrfs_device *device = NULL;
10625 struct btrfs_swap_info bsi = {
10626 .lowest_ppage = (sector_t)-1ULL,
10627 };
10628 int ret = 0;
10629 u64 isize;
10630 u64 start;
10631
10632 /*
10633 * If the swap file was just created, make sure delalloc is done. If the
10634 * file changes again after this, the user is doing something stupid and
10635 * we don't really care.
10636 */
10637 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10638 if (ret)
10639 return ret;
10640
10641 /*
10642 * The inode is locked, so these flags won't change after we check them.
10643 */
10644 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10645 btrfs_warn(fs_info, "swapfile must not be compressed");
10646 return -EINVAL;
10647 }
10648 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10649 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10650 return -EINVAL;
10651 }
10652 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10653 btrfs_warn(fs_info, "swapfile must not be checksummed");
10654 return -EINVAL;
10655 }
10656
10657 /*
10658 * Balance or device remove/replace/resize can move stuff around from
10659 * under us. The exclop protection makes sure they aren't running/won't
10660 * run concurrently while we are mapping the swap extents, and
10661 * fs_info->swapfile_pins prevents them from running while the swap
10662 * file is active and moving the extents. Note that this also prevents
10663 * a concurrent device add which isn't actually necessary, but it's not
10664 * really worth the trouble to allow it.
10665 */
10666 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10667 btrfs_warn(fs_info,
10668 "cannot activate swapfile while exclusive operation is running");
10669 return -EBUSY;
10670 }
10671
10672 /*
10673 * Prevent snapshot creation while we are activating the swap file.
10674 * We do not want to race with snapshot creation. If snapshot creation
10675 * already started before we bumped nr_swapfiles from 0 to 1 and
10676 * completes before the first write into the swap file after it is
10677 * activated, than that write would fallback to COW.
10678 */
10679 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10680 btrfs_exclop_finish(fs_info);
10681 btrfs_warn(fs_info,
10682 "cannot activate swapfile because snapshot creation is in progress");
10683 return -EINVAL;
10684 }
10685 /*
10686 * Snapshots can create extents which require COW even if NODATACOW is
10687 * set. We use this counter to prevent snapshots. We must increment it
10688 * before walking the extents because we don't want a concurrent
10689 * snapshot to run after we've already checked the extents.
10690 *
10691 * It is possible that subvolume is marked for deletion but still not
10692 * removed yet. To prevent this race, we check the root status before
10693 * activating the swapfile.
10694 */
10695 spin_lock(&root->root_item_lock);
10696 if (btrfs_root_dead(root)) {
10697 spin_unlock(&root->root_item_lock);
10698
10699 btrfs_exclop_finish(fs_info);
10700 btrfs_warn(fs_info,
10701 "cannot activate swapfile because subvolume %llu is being deleted",
10702 root->root_key.objectid);
10703 return -EPERM;
10704 }
10705 atomic_inc(&root->nr_swapfiles);
10706 spin_unlock(&root->root_item_lock);
10707
10708 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10709
10710 lock_extent(io_tree, 0, isize - 1, &cached_state);
10711 start = 0;
10712 while (start < isize) {
10713 u64 logical_block_start, physical_block_start;
10714 struct btrfs_block_group *bg;
10715 u64 len = isize - start;
10716
10717 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10718 if (IS_ERR(em)) {
10719 ret = PTR_ERR(em);
10720 goto out;
10721 }
10722
10723 if (em->block_start == EXTENT_MAP_HOLE) {
10724 btrfs_warn(fs_info, "swapfile must not have holes");
10725 ret = -EINVAL;
10726 goto out;
10727 }
10728 if (em->block_start == EXTENT_MAP_INLINE) {
10729 /*
10730 * It's unlikely we'll ever actually find ourselves
10731 * here, as a file small enough to fit inline won't be
10732 * big enough to store more than the swap header, but in
10733 * case something changes in the future, let's catch it
10734 * here rather than later.
10735 */
10736 btrfs_warn(fs_info, "swapfile must not be inline");
10737 ret = -EINVAL;
10738 goto out;
10739 }
10740 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10741 btrfs_warn(fs_info, "swapfile must not be compressed");
10742 ret = -EINVAL;
10743 goto out;
10744 }
10745
10746 logical_block_start = em->block_start + (start - em->start);
10747 len = min(len, em->len - (start - em->start));
10748 free_extent_map(em);
10749 em = NULL;
10750
10751 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10752 if (ret < 0) {
10753 goto out;
10754 } else if (ret) {
10755 ret = 0;
10756 } else {
10757 btrfs_warn(fs_info,
10758 "swapfile must not be copy-on-write");
10759 ret = -EINVAL;
10760 goto out;
10761 }
10762
10763 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10764 if (IS_ERR(em)) {
10765 ret = PTR_ERR(em);
10766 goto out;
10767 }
10768
10769 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10770 btrfs_warn(fs_info,
10771 "swapfile must have single data profile");
10772 ret = -EINVAL;
10773 goto out;
10774 }
10775
10776 if (device == NULL) {
10777 device = em->map_lookup->stripes[0].dev;
10778 ret = btrfs_add_swapfile_pin(inode, device, false);
10779 if (ret == 1)
10780 ret = 0;
10781 else if (ret)
10782 goto out;
10783 } else if (device != em->map_lookup->stripes[0].dev) {
10784 btrfs_warn(fs_info, "swapfile must be on one device");
10785 ret = -EINVAL;
10786 goto out;
10787 }
10788
10789 physical_block_start = (em->map_lookup->stripes[0].physical +
10790 (logical_block_start - em->start));
10791 len = min(len, em->len - (logical_block_start - em->start));
10792 free_extent_map(em);
10793 em = NULL;
10794
10795 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10796 if (!bg) {
10797 btrfs_warn(fs_info,
10798 "could not find block group containing swapfile");
10799 ret = -EINVAL;
10800 goto out;
10801 }
10802
10803 if (!btrfs_inc_block_group_swap_extents(bg)) {
10804 btrfs_warn(fs_info,
10805 "block group for swapfile at %llu is read-only%s",
10806 bg->start,
10807 atomic_read(&fs_info->scrubs_running) ?
10808 " (scrub running)" : "");
10809 btrfs_put_block_group(bg);
10810 ret = -EINVAL;
10811 goto out;
10812 }
10813
10814 ret = btrfs_add_swapfile_pin(inode, bg, true);
10815 if (ret) {
10816 btrfs_put_block_group(bg);
10817 if (ret == 1)
10818 ret = 0;
10819 else
10820 goto out;
10821 }
10822
10823 if (bsi.block_len &&
10824 bsi.block_start + bsi.block_len == physical_block_start) {
10825 bsi.block_len += len;
10826 } else {
10827 if (bsi.block_len) {
10828 ret = btrfs_add_swap_extent(sis, &bsi);
10829 if (ret)
10830 goto out;
10831 }
10832 bsi.start = start;
10833 bsi.block_start = physical_block_start;
10834 bsi.block_len = len;
10835 }
10836
10837 start += len;
10838 }
10839
10840 if (bsi.block_len)
10841 ret = btrfs_add_swap_extent(sis, &bsi);
10842
10843 out:
10844 if (!IS_ERR_OR_NULL(em))
10845 free_extent_map(em);
10846
10847 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10848
10849 if (ret)
10850 btrfs_swap_deactivate(file);
10851
10852 btrfs_drew_write_unlock(&root->snapshot_lock);
10853
10854 btrfs_exclop_finish(fs_info);
10855
10856 if (ret)
10857 return ret;
10858
10859 if (device)
10860 sis->bdev = device->bdev;
10861 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10862 sis->max = bsi.nr_pages;
10863 sis->pages = bsi.nr_pages - 1;
10864 sis->highest_bit = bsi.nr_pages - 1;
10865 return bsi.nr_extents;
10866 }
10867 #else
btrfs_swap_deactivate(struct file * file)10868 static void btrfs_swap_deactivate(struct file *file)
10869 {
10870 }
10871
btrfs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)10872 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10873 sector_t *span)
10874 {
10875 return -EOPNOTSUPP;
10876 }
10877 #endif
10878
10879 /*
10880 * Update the number of bytes used in the VFS' inode. When we replace extents in
10881 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10882 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10883 * always get a correct value.
10884 */
btrfs_update_inode_bytes(struct btrfs_inode * inode,const u64 add_bytes,const u64 del_bytes)10885 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10886 const u64 add_bytes,
10887 const u64 del_bytes)
10888 {
10889 if (add_bytes == del_bytes)
10890 return;
10891
10892 spin_lock(&inode->lock);
10893 if (del_bytes > 0)
10894 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10895 if (add_bytes > 0)
10896 inode_add_bytes(&inode->vfs_inode, add_bytes);
10897 spin_unlock(&inode->lock);
10898 }
10899
10900 /*
10901 * Verify that there are no ordered extents for a given file range.
10902 *
10903 * @inode: The target inode.
10904 * @start: Start offset of the file range, should be sector size aligned.
10905 * @end: End offset (inclusive) of the file range, its value +1 should be
10906 * sector size aligned.
10907 *
10908 * This should typically be used for cases where we locked an inode's VFS lock in
10909 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10910 * we have flushed all delalloc in the range, we have waited for all ordered
10911 * extents in the range to complete and finally we have locked the file range in
10912 * the inode's io_tree.
10913 */
btrfs_assert_inode_range_clean(struct btrfs_inode * inode,u64 start,u64 end)10914 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10915 {
10916 struct btrfs_root *root = inode->root;
10917 struct btrfs_ordered_extent *ordered;
10918
10919 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10920 return;
10921
10922 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10923 if (ordered) {
10924 btrfs_err(root->fs_info,
10925 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10926 start, end, btrfs_ino(inode), root->root_key.objectid,
10927 ordered->file_offset,
10928 ordered->file_offset + ordered->num_bytes - 1);
10929 btrfs_put_ordered_extent(ordered);
10930 }
10931
10932 ASSERT(ordered == NULL);
10933 }
10934
10935 static const struct inode_operations btrfs_dir_inode_operations = {
10936 .getattr = btrfs_getattr,
10937 .lookup = btrfs_lookup,
10938 .create = btrfs_create,
10939 .unlink = btrfs_unlink,
10940 .link = btrfs_link,
10941 .mkdir = btrfs_mkdir,
10942 .rmdir = btrfs_rmdir,
10943 .rename = btrfs_rename2,
10944 .symlink = btrfs_symlink,
10945 .setattr = btrfs_setattr,
10946 .mknod = btrfs_mknod,
10947 .listxattr = btrfs_listxattr,
10948 .permission = btrfs_permission,
10949 .get_inode_acl = btrfs_get_acl,
10950 .set_acl = btrfs_set_acl,
10951 .update_time = btrfs_update_time,
10952 .tmpfile = btrfs_tmpfile,
10953 .fileattr_get = btrfs_fileattr_get,
10954 .fileattr_set = btrfs_fileattr_set,
10955 };
10956
10957 static const struct file_operations btrfs_dir_file_operations = {
10958 .llseek = btrfs_dir_llseek,
10959 .read = generic_read_dir,
10960 .iterate_shared = btrfs_real_readdir,
10961 .open = btrfs_opendir,
10962 .unlocked_ioctl = btrfs_ioctl,
10963 #ifdef CONFIG_COMPAT
10964 .compat_ioctl = btrfs_compat_ioctl,
10965 #endif
10966 .release = btrfs_release_file,
10967 .fsync = btrfs_sync_file,
10968 };
10969
10970 /*
10971 * btrfs doesn't support the bmap operation because swapfiles
10972 * use bmap to make a mapping of extents in the file. They assume
10973 * these extents won't change over the life of the file and they
10974 * use the bmap result to do IO directly to the drive.
10975 *
10976 * the btrfs bmap call would return logical addresses that aren't
10977 * suitable for IO and they also will change frequently as COW
10978 * operations happen. So, swapfile + btrfs == corruption.
10979 *
10980 * For now we're avoiding this by dropping bmap.
10981 */
10982 static const struct address_space_operations btrfs_aops = {
10983 .read_folio = btrfs_read_folio,
10984 .writepages = btrfs_writepages,
10985 .readahead = btrfs_readahead,
10986 .invalidate_folio = btrfs_invalidate_folio,
10987 .release_folio = btrfs_release_folio,
10988 .migrate_folio = btrfs_migrate_folio,
10989 .dirty_folio = filemap_dirty_folio,
10990 .error_remove_page = generic_error_remove_page,
10991 .swap_activate = btrfs_swap_activate,
10992 .swap_deactivate = btrfs_swap_deactivate,
10993 };
10994
10995 static const struct inode_operations btrfs_file_inode_operations = {
10996 .getattr = btrfs_getattr,
10997 .setattr = btrfs_setattr,
10998 .listxattr = btrfs_listxattr,
10999 .permission = btrfs_permission,
11000 .fiemap = btrfs_fiemap,
11001 .get_inode_acl = btrfs_get_acl,
11002 .set_acl = btrfs_set_acl,
11003 .update_time = btrfs_update_time,
11004 .fileattr_get = btrfs_fileattr_get,
11005 .fileattr_set = btrfs_fileattr_set,
11006 };
11007 static const struct inode_operations btrfs_special_inode_operations = {
11008 .getattr = btrfs_getattr,
11009 .setattr = btrfs_setattr,
11010 .permission = btrfs_permission,
11011 .listxattr = btrfs_listxattr,
11012 .get_inode_acl = btrfs_get_acl,
11013 .set_acl = btrfs_set_acl,
11014 .update_time = btrfs_update_time,
11015 };
11016 static const struct inode_operations btrfs_symlink_inode_operations = {
11017 .get_link = page_get_link,
11018 .getattr = btrfs_getattr,
11019 .setattr = btrfs_setattr,
11020 .permission = btrfs_permission,
11021 .listxattr = btrfs_listxattr,
11022 .update_time = btrfs_update_time,
11023 };
11024
11025 const struct dentry_operations btrfs_dentry_operations = {
11026 .d_delete = btrfs_dentry_delete,
11027 };
11028