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