1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6 #include <crypto/hash.h>
7 #include <linux/kernel.h>
8 #include <linux/bio.h>
9 #include <linux/blk-cgroup.h>
10 #include <linux/file.h>
11 #include <linux/fs.h>
12 #include <linux/pagemap.h>
13 #include <linux/highmem.h>
14 #include <linux/time.h>
15 #include <linux/init.h>
16 #include <linux/string.h>
17 #include <linux/backing-dev.h>
18 #include <linux/writeback.h>
19 #include <linux/compat.h>
20 #include <linux/xattr.h>
21 #include <linux/posix_acl.h>
22 #include <linux/falloc.h>
23 #include <linux/slab.h>
24 #include <linux/ratelimit.h>
25 #include <linux/btrfs.h>
26 #include <linux/blkdev.h>
27 #include <linux/posix_acl_xattr.h>
28 #include <linux/uio.h>
29 #include <linux/magic.h>
30 #include <linux/iversion.h>
31 #include <linux/swap.h>
32 #include <linux/migrate.h>
33 #include <linux/sched/mm.h>
34 #include <linux/iomap.h>
35 #include <asm/unaligned.h>
36 #include <linux/fsverity.h>
37 #include "misc.h"
38 #include "ctree.h"
39 #include "disk-io.h"
40 #include "transaction.h"
41 #include "btrfs_inode.h"
42 #include "print-tree.h"
43 #include "ordered-data.h"
44 #include "xattr.h"
45 #include "tree-log.h"
46 #include "bio.h"
47 #include "compression.h"
48 #include "locking.h"
49 #include "free-space-cache.h"
50 #include "props.h"
51 #include "qgroup.h"
52 #include "delalloc-space.h"
53 #include "block-group.h"
54 #include "space-info.h"
55 #include "zoned.h"
56 #include "subpage.h"
57 #include "inode-item.h"
58 #include "fs.h"
59 #include "accessors.h"
60 #include "extent-tree.h"
61 #include "root-tree.h"
62 #include "defrag.h"
63 #include "dir-item.h"
64 #include "file-item.h"
65 #include "uuid-tree.h"
66 #include "ioctl.h"
67 #include "file.h"
68 #include "acl.h"
69 #include "relocation.h"
70 #include "verity.h"
71 #include "super.h"
72 #include "orphan.h"
73 #include "backref.h"
74
75 struct btrfs_iget_args {
76 u64 ino;
77 struct btrfs_root *root;
78 };
79
80 struct btrfs_dio_data {
81 ssize_t submitted;
82 struct extent_changeset *data_reserved;
83 struct btrfs_ordered_extent *ordered;
84 bool data_space_reserved;
85 bool nocow_done;
86 };
87
88 struct btrfs_dio_private {
89 /* Range of I/O */
90 u64 file_offset;
91 u32 bytes;
92
93 /* This must be last */
94 struct btrfs_bio bbio;
95 };
96
97 static struct bio_set btrfs_dio_bioset;
98
99 struct btrfs_rename_ctx {
100 /* Output field. Stores the index number of the old directory entry. */
101 u64 index;
102 };
103
104 /*
105 * Used by data_reloc_print_warning_inode() to pass needed info for filename
106 * resolution and output of error message.
107 */
108 struct data_reloc_warn {
109 struct btrfs_path path;
110 struct btrfs_fs_info *fs_info;
111 u64 extent_item_size;
112 u64 logical;
113 int mirror_num;
114 };
115
116 static const struct inode_operations btrfs_dir_inode_operations;
117 static const struct inode_operations btrfs_symlink_inode_operations;
118 static const struct inode_operations btrfs_special_inode_operations;
119 static const struct inode_operations btrfs_file_inode_operations;
120 static const struct address_space_operations btrfs_aops;
121 static const struct file_operations btrfs_dir_file_operations;
122
123 static struct kmem_cache *btrfs_inode_cachep;
124
125 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
126 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback);
127
128 static noinline int run_delalloc_cow(struct btrfs_inode *inode,
129 struct page *locked_page, u64 start,
130 u64 end, struct writeback_control *wbc,
131 bool pages_dirty);
132 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
133 u64 len, u64 orig_start, u64 block_start,
134 u64 block_len, u64 orig_block_len,
135 u64 ram_bytes, int compress_type,
136 int type);
137
data_reloc_print_warning_inode(u64 inum,u64 offset,u64 num_bytes,u64 root,void * warn_ctx)138 static int data_reloc_print_warning_inode(u64 inum, u64 offset, u64 num_bytes,
139 u64 root, void *warn_ctx)
140 {
141 struct data_reloc_warn *warn = warn_ctx;
142 struct btrfs_fs_info *fs_info = warn->fs_info;
143 struct extent_buffer *eb;
144 struct btrfs_inode_item *inode_item;
145 struct inode_fs_paths *ipath = NULL;
146 struct btrfs_root *local_root;
147 struct btrfs_key key;
148 unsigned int nofs_flag;
149 u32 nlink;
150 int ret;
151
152 local_root = btrfs_get_fs_root(fs_info, root, true);
153 if (IS_ERR(local_root)) {
154 ret = PTR_ERR(local_root);
155 goto err;
156 }
157
158 /* This makes the path point to (inum INODE_ITEM ioff). */
159 key.objectid = inum;
160 key.type = BTRFS_INODE_ITEM_KEY;
161 key.offset = 0;
162
163 ret = btrfs_search_slot(NULL, local_root, &key, &warn->path, 0, 0);
164 if (ret) {
165 btrfs_put_root(local_root);
166 btrfs_release_path(&warn->path);
167 goto err;
168 }
169
170 eb = warn->path.nodes[0];
171 inode_item = btrfs_item_ptr(eb, warn->path.slots[0], struct btrfs_inode_item);
172 nlink = btrfs_inode_nlink(eb, inode_item);
173 btrfs_release_path(&warn->path);
174
175 nofs_flag = memalloc_nofs_save();
176 ipath = init_ipath(4096, local_root, &warn->path);
177 memalloc_nofs_restore(nofs_flag);
178 if (IS_ERR(ipath)) {
179 btrfs_put_root(local_root);
180 ret = PTR_ERR(ipath);
181 ipath = NULL;
182 /*
183 * -ENOMEM, not a critical error, just output an generic error
184 * without filename.
185 */
186 btrfs_warn(fs_info,
187 "checksum error at logical %llu mirror %u root %llu, inode %llu offset %llu",
188 warn->logical, warn->mirror_num, root, inum, offset);
189 return ret;
190 }
191 ret = paths_from_inode(inum, ipath);
192 if (ret < 0)
193 goto err;
194
195 /*
196 * We deliberately ignore the bit ipath might have been too small to
197 * hold all of the paths here
198 */
199 for (int i = 0; i < ipath->fspath->elem_cnt; i++) {
200 btrfs_warn(fs_info,
201 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu length %u links %u (path: %s)",
202 warn->logical, warn->mirror_num, root, inum, offset,
203 fs_info->sectorsize, nlink,
204 (char *)(unsigned long)ipath->fspath->val[i]);
205 }
206
207 btrfs_put_root(local_root);
208 free_ipath(ipath);
209 return 0;
210
211 err:
212 btrfs_warn(fs_info,
213 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu, path resolving failed with ret=%d",
214 warn->logical, warn->mirror_num, root, inum, offset, ret);
215
216 free_ipath(ipath);
217 return ret;
218 }
219
220 /*
221 * Do extra user-friendly error output (e.g. lookup all the affected files).
222 *
223 * Return true if we succeeded doing the backref lookup.
224 * Return false if such lookup failed, and has to fallback to the old error message.
225 */
print_data_reloc_error(const struct btrfs_inode * inode,u64 file_off,const u8 * csum,const u8 * csum_expected,int mirror_num)226 static void print_data_reloc_error(const struct btrfs_inode *inode, u64 file_off,
227 const u8 *csum, const u8 *csum_expected,
228 int mirror_num)
229 {
230 struct btrfs_fs_info *fs_info = inode->root->fs_info;
231 struct btrfs_path path = { 0 };
232 struct btrfs_key found_key = { 0 };
233 struct extent_buffer *eb;
234 struct btrfs_extent_item *ei;
235 const u32 csum_size = fs_info->csum_size;
236 u64 logical;
237 u64 flags;
238 u32 item_size;
239 int ret;
240
241 mutex_lock(&fs_info->reloc_mutex);
242 logical = btrfs_get_reloc_bg_bytenr(fs_info);
243 mutex_unlock(&fs_info->reloc_mutex);
244
245 if (logical == U64_MAX) {
246 btrfs_warn_rl(fs_info, "has data reloc tree but no running relocation");
247 btrfs_warn_rl(fs_info,
248 "csum failed root %lld ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
249 inode->root->root_key.objectid, btrfs_ino(inode), file_off,
250 CSUM_FMT_VALUE(csum_size, csum),
251 CSUM_FMT_VALUE(csum_size, csum_expected),
252 mirror_num);
253 return;
254 }
255
256 logical += file_off;
257 btrfs_warn_rl(fs_info,
258 "csum failed root %lld ino %llu off %llu logical %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
259 inode->root->root_key.objectid,
260 btrfs_ino(inode), file_off, logical,
261 CSUM_FMT_VALUE(csum_size, csum),
262 CSUM_FMT_VALUE(csum_size, csum_expected),
263 mirror_num);
264
265 ret = extent_from_logical(fs_info, logical, &path, &found_key, &flags);
266 if (ret < 0) {
267 btrfs_err_rl(fs_info, "failed to lookup extent item for logical %llu: %d",
268 logical, ret);
269 return;
270 }
271 eb = path.nodes[0];
272 ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item);
273 item_size = btrfs_item_size(eb, path.slots[0]);
274 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
275 unsigned long ptr = 0;
276 u64 ref_root;
277 u8 ref_level;
278
279 while (true) {
280 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
281 item_size, &ref_root,
282 &ref_level);
283 if (ret < 0) {
284 btrfs_warn_rl(fs_info,
285 "failed to resolve tree backref for logical %llu: %d",
286 logical, ret);
287 break;
288 }
289 if (ret > 0)
290 break;
291
292 btrfs_warn_rl(fs_info,
293 "csum error at logical %llu mirror %u: metadata %s (level %d) in tree %llu",
294 logical, mirror_num,
295 (ref_level ? "node" : "leaf"),
296 ref_level, ref_root);
297 }
298 btrfs_release_path(&path);
299 } else {
300 struct btrfs_backref_walk_ctx ctx = { 0 };
301 struct data_reloc_warn reloc_warn = { 0 };
302
303 btrfs_release_path(&path);
304
305 ctx.bytenr = found_key.objectid;
306 ctx.extent_item_pos = logical - found_key.objectid;
307 ctx.fs_info = fs_info;
308
309 reloc_warn.logical = logical;
310 reloc_warn.extent_item_size = found_key.offset;
311 reloc_warn.mirror_num = mirror_num;
312 reloc_warn.fs_info = fs_info;
313
314 iterate_extent_inodes(&ctx, true,
315 data_reloc_print_warning_inode, &reloc_warn);
316 }
317 }
318
btrfs_print_data_csum_error(struct btrfs_inode * inode,u64 logical_start,u8 * csum,u8 * csum_expected,int mirror_num)319 static void __cold btrfs_print_data_csum_error(struct btrfs_inode *inode,
320 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
321 {
322 struct btrfs_root *root = inode->root;
323 const u32 csum_size = root->fs_info->csum_size;
324
325 /* For data reloc tree, it's better to do a backref lookup instead. */
326 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
327 return print_data_reloc_error(inode, logical_start, csum,
328 csum_expected, mirror_num);
329
330 /* Output without objectid, which is more meaningful */
331 if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID) {
332 btrfs_warn_rl(root->fs_info,
333 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
334 root->root_key.objectid, btrfs_ino(inode),
335 logical_start,
336 CSUM_FMT_VALUE(csum_size, csum),
337 CSUM_FMT_VALUE(csum_size, csum_expected),
338 mirror_num);
339 } else {
340 btrfs_warn_rl(root->fs_info,
341 "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
342 root->root_key.objectid, btrfs_ino(inode),
343 logical_start,
344 CSUM_FMT_VALUE(csum_size, csum),
345 CSUM_FMT_VALUE(csum_size, csum_expected),
346 mirror_num);
347 }
348 }
349
350 /*
351 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
352 *
353 * ilock_flags can have the following bit set:
354 *
355 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
356 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
357 * return -EAGAIN
358 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
359 */
btrfs_inode_lock(struct btrfs_inode * inode,unsigned int ilock_flags)360 int btrfs_inode_lock(struct btrfs_inode *inode, unsigned int ilock_flags)
361 {
362 if (ilock_flags & BTRFS_ILOCK_SHARED) {
363 if (ilock_flags & BTRFS_ILOCK_TRY) {
364 if (!inode_trylock_shared(&inode->vfs_inode))
365 return -EAGAIN;
366 else
367 return 0;
368 }
369 inode_lock_shared(&inode->vfs_inode);
370 } else {
371 if (ilock_flags & BTRFS_ILOCK_TRY) {
372 if (!inode_trylock(&inode->vfs_inode))
373 return -EAGAIN;
374 else
375 return 0;
376 }
377 inode_lock(&inode->vfs_inode);
378 }
379 if (ilock_flags & BTRFS_ILOCK_MMAP)
380 down_write(&inode->i_mmap_lock);
381 return 0;
382 }
383
384 /*
385 * btrfs_inode_unlock - unock inode i_rwsem
386 *
387 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
388 * to decide whether the lock acquired is shared or exclusive.
389 */
btrfs_inode_unlock(struct btrfs_inode * inode,unsigned int ilock_flags)390 void btrfs_inode_unlock(struct btrfs_inode *inode, unsigned int ilock_flags)
391 {
392 if (ilock_flags & BTRFS_ILOCK_MMAP)
393 up_write(&inode->i_mmap_lock);
394 if (ilock_flags & BTRFS_ILOCK_SHARED)
395 inode_unlock_shared(&inode->vfs_inode);
396 else
397 inode_unlock(&inode->vfs_inode);
398 }
399
400 /*
401 * Cleanup all submitted ordered extents in specified range to handle errors
402 * from the btrfs_run_delalloc_range() callback.
403 *
404 * NOTE: caller must ensure that when an error happens, it can not call
405 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
406 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
407 * to be released, which we want to happen only when finishing the ordered
408 * extent (btrfs_finish_ordered_io()).
409 */
btrfs_cleanup_ordered_extents(struct btrfs_inode * inode,struct page * locked_page,u64 offset,u64 bytes)410 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
411 struct page *locked_page,
412 u64 offset, u64 bytes)
413 {
414 unsigned long index = offset >> PAGE_SHIFT;
415 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
416 u64 page_start = 0, page_end = 0;
417 struct page *page;
418
419 if (locked_page) {
420 page_start = page_offset(locked_page);
421 page_end = page_start + PAGE_SIZE - 1;
422 }
423
424 while (index <= end_index) {
425 /*
426 * For locked page, we will call btrfs_mark_ordered_io_finished
427 * through btrfs_mark_ordered_io_finished() on it
428 * in run_delalloc_range() for the error handling, which will
429 * clear page Ordered and run the ordered extent accounting.
430 *
431 * Here we can't just clear the Ordered bit, or
432 * btrfs_mark_ordered_io_finished() would skip the accounting
433 * for the page range, and the ordered extent will never finish.
434 */
435 if (locked_page && index == (page_start >> PAGE_SHIFT)) {
436 index++;
437 continue;
438 }
439 page = find_get_page(inode->vfs_inode.i_mapping, index);
440 index++;
441 if (!page)
442 continue;
443
444 /*
445 * Here we just clear all Ordered bits for every page in the
446 * range, then btrfs_mark_ordered_io_finished() will handle
447 * the ordered extent accounting for the range.
448 */
449 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
450 offset, bytes);
451 put_page(page);
452 }
453
454 if (locked_page) {
455 /* The locked page covers the full range, nothing needs to be done */
456 if (bytes + offset <= page_start + PAGE_SIZE)
457 return;
458 /*
459 * In case this page belongs to the delalloc range being
460 * instantiated then skip it, since the first page of a range is
461 * going to be properly cleaned up by the caller of
462 * run_delalloc_range
463 */
464 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
465 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
466 offset = page_offset(locked_page) + PAGE_SIZE;
467 }
468 }
469
470 return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false);
471 }
472
473 static int btrfs_dirty_inode(struct btrfs_inode *inode);
474
btrfs_init_inode_security(struct btrfs_trans_handle * trans,struct btrfs_new_inode_args * args)475 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
476 struct btrfs_new_inode_args *args)
477 {
478 int err;
479
480 if (args->default_acl) {
481 err = __btrfs_set_acl(trans, args->inode, args->default_acl,
482 ACL_TYPE_DEFAULT);
483 if (err)
484 return err;
485 }
486 if (args->acl) {
487 err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
488 if (err)
489 return err;
490 }
491 if (!args->default_acl && !args->acl)
492 cache_no_acl(args->inode);
493 return btrfs_xattr_security_init(trans, args->inode, args->dir,
494 &args->dentry->d_name);
495 }
496
497 /*
498 * this does all the hard work for inserting an inline extent into
499 * the btree. The caller should have done a btrfs_drop_extents so that
500 * no overlapping inline items exist in the btree
501 */
insert_inline_extent(struct btrfs_trans_handle * trans,struct btrfs_path * path,struct btrfs_inode * inode,bool extent_inserted,size_t size,size_t compressed_size,int compress_type,struct page ** compressed_pages,bool update_i_size)502 static int insert_inline_extent(struct btrfs_trans_handle *trans,
503 struct btrfs_path *path,
504 struct btrfs_inode *inode, bool extent_inserted,
505 size_t size, size_t compressed_size,
506 int compress_type,
507 struct page **compressed_pages,
508 bool update_i_size)
509 {
510 struct btrfs_root *root = inode->root;
511 struct extent_buffer *leaf;
512 struct page *page = NULL;
513 char *kaddr;
514 unsigned long ptr;
515 struct btrfs_file_extent_item *ei;
516 int ret;
517 size_t cur_size = size;
518 u64 i_size;
519
520 ASSERT((compressed_size > 0 && compressed_pages) ||
521 (compressed_size == 0 && !compressed_pages));
522
523 if (compressed_size && compressed_pages)
524 cur_size = compressed_size;
525
526 if (!extent_inserted) {
527 struct btrfs_key key;
528 size_t datasize;
529
530 key.objectid = btrfs_ino(inode);
531 key.offset = 0;
532 key.type = BTRFS_EXTENT_DATA_KEY;
533
534 datasize = btrfs_file_extent_calc_inline_size(cur_size);
535 ret = btrfs_insert_empty_item(trans, root, path, &key,
536 datasize);
537 if (ret)
538 goto fail;
539 }
540 leaf = path->nodes[0];
541 ei = btrfs_item_ptr(leaf, path->slots[0],
542 struct btrfs_file_extent_item);
543 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
544 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
545 btrfs_set_file_extent_encryption(leaf, ei, 0);
546 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
547 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
548 ptr = btrfs_file_extent_inline_start(ei);
549
550 if (compress_type != BTRFS_COMPRESS_NONE) {
551 struct page *cpage;
552 int i = 0;
553 while (compressed_size > 0) {
554 cpage = compressed_pages[i];
555 cur_size = min_t(unsigned long, compressed_size,
556 PAGE_SIZE);
557
558 kaddr = kmap_local_page(cpage);
559 write_extent_buffer(leaf, kaddr, ptr, cur_size);
560 kunmap_local(kaddr);
561
562 i++;
563 ptr += cur_size;
564 compressed_size -= cur_size;
565 }
566 btrfs_set_file_extent_compression(leaf, ei,
567 compress_type);
568 } else {
569 page = find_get_page(inode->vfs_inode.i_mapping, 0);
570 btrfs_set_file_extent_compression(leaf, ei, 0);
571 kaddr = kmap_local_page(page);
572 write_extent_buffer(leaf, kaddr, ptr, size);
573 kunmap_local(kaddr);
574 put_page(page);
575 }
576 btrfs_mark_buffer_dirty(trans, leaf);
577 btrfs_release_path(path);
578
579 /*
580 * We align size to sectorsize for inline extents just for simplicity
581 * sake.
582 */
583 ret = btrfs_inode_set_file_extent_range(inode, 0,
584 ALIGN(size, root->fs_info->sectorsize));
585 if (ret)
586 goto fail;
587
588 /*
589 * We're an inline extent, so nobody can extend the file past i_size
590 * without locking a page we already have locked.
591 *
592 * We must do any i_size and inode updates before we unlock the pages.
593 * Otherwise we could end up racing with unlink.
594 */
595 i_size = i_size_read(&inode->vfs_inode);
596 if (update_i_size && size > i_size) {
597 i_size_write(&inode->vfs_inode, size);
598 i_size = size;
599 }
600 inode->disk_i_size = i_size;
601
602 fail:
603 return ret;
604 }
605
606
607 /*
608 * conditionally insert an inline extent into the file. This
609 * does the checks required to make sure the data is small enough
610 * to fit as an inline extent.
611 */
cow_file_range_inline(struct btrfs_inode * inode,u64 size,size_t compressed_size,int compress_type,struct page ** compressed_pages,bool update_i_size)612 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
613 size_t compressed_size,
614 int compress_type,
615 struct page **compressed_pages,
616 bool update_i_size)
617 {
618 struct btrfs_drop_extents_args drop_args = { 0 };
619 struct btrfs_root *root = inode->root;
620 struct btrfs_fs_info *fs_info = root->fs_info;
621 struct btrfs_trans_handle *trans;
622 u64 data_len = (compressed_size ?: size);
623 int ret;
624 struct btrfs_path *path;
625
626 /*
627 * We can create an inline extent if it ends at or beyond the current
628 * i_size, is no larger than a sector (decompressed), and the (possibly
629 * compressed) data fits in a leaf and the configured maximum inline
630 * size.
631 */
632 if (size < i_size_read(&inode->vfs_inode) ||
633 size > fs_info->sectorsize ||
634 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
635 data_len > fs_info->max_inline)
636 return 1;
637
638 path = btrfs_alloc_path();
639 if (!path)
640 return -ENOMEM;
641
642 trans = btrfs_join_transaction(root);
643 if (IS_ERR(trans)) {
644 btrfs_free_path(path);
645 return PTR_ERR(trans);
646 }
647 trans->block_rsv = &inode->block_rsv;
648
649 drop_args.path = path;
650 drop_args.start = 0;
651 drop_args.end = fs_info->sectorsize;
652 drop_args.drop_cache = true;
653 drop_args.replace_extent = true;
654 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
655 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
656 if (ret) {
657 btrfs_abort_transaction(trans, ret);
658 goto out;
659 }
660
661 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
662 size, compressed_size, compress_type,
663 compressed_pages, update_i_size);
664 if (ret && ret != -ENOSPC) {
665 btrfs_abort_transaction(trans, ret);
666 goto out;
667 } else if (ret == -ENOSPC) {
668 ret = 1;
669 goto out;
670 }
671
672 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
673 ret = btrfs_update_inode(trans, root, inode);
674 if (ret && ret != -ENOSPC) {
675 btrfs_abort_transaction(trans, ret);
676 goto out;
677 } else if (ret == -ENOSPC) {
678 ret = 1;
679 goto out;
680 }
681
682 btrfs_set_inode_full_sync(inode);
683 out:
684 /*
685 * Don't forget to free the reserved space, as for inlined extent
686 * it won't count as data extent, free them directly here.
687 * And at reserve time, it's always aligned to page size, so
688 * just free one page here.
689 */
690 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE, NULL);
691 btrfs_free_path(path);
692 btrfs_end_transaction(trans);
693 return ret;
694 }
695
696 struct async_extent {
697 u64 start;
698 u64 ram_size;
699 u64 compressed_size;
700 struct page **pages;
701 unsigned long nr_pages;
702 int compress_type;
703 struct list_head list;
704 };
705
706 struct async_chunk {
707 struct btrfs_inode *inode;
708 struct page *locked_page;
709 u64 start;
710 u64 end;
711 blk_opf_t write_flags;
712 struct list_head extents;
713 struct cgroup_subsys_state *blkcg_css;
714 struct btrfs_work work;
715 struct async_cow *async_cow;
716 };
717
718 struct async_cow {
719 atomic_t num_chunks;
720 struct async_chunk chunks[];
721 };
722
add_async_extent(struct async_chunk * cow,u64 start,u64 ram_size,u64 compressed_size,struct page ** pages,unsigned long nr_pages,int compress_type)723 static noinline int add_async_extent(struct async_chunk *cow,
724 u64 start, u64 ram_size,
725 u64 compressed_size,
726 struct page **pages,
727 unsigned long nr_pages,
728 int compress_type)
729 {
730 struct async_extent *async_extent;
731
732 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
733 if (!async_extent)
734 return -ENOMEM;
735 async_extent->start = start;
736 async_extent->ram_size = ram_size;
737 async_extent->compressed_size = compressed_size;
738 async_extent->pages = pages;
739 async_extent->nr_pages = nr_pages;
740 async_extent->compress_type = compress_type;
741 list_add_tail(&async_extent->list, &cow->extents);
742 return 0;
743 }
744
745 /*
746 * Check if the inode needs to be submitted to compression, based on mount
747 * options, defragmentation, properties or heuristics.
748 */
inode_need_compress(struct btrfs_inode * inode,u64 start,u64 end)749 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
750 u64 end)
751 {
752 struct btrfs_fs_info *fs_info = inode->root->fs_info;
753
754 if (!btrfs_inode_can_compress(inode)) {
755 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
756 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
757 btrfs_ino(inode));
758 return 0;
759 }
760 /*
761 * Special check for subpage.
762 *
763 * We lock the full page then run each delalloc range in the page, thus
764 * for the following case, we will hit some subpage specific corner case:
765 *
766 * 0 32K 64K
767 * | |///////| |///////|
768 * \- A \- B
769 *
770 * In above case, both range A and range B will try to unlock the full
771 * page [0, 64K), causing the one finished later will have page
772 * unlocked already, triggering various page lock requirement BUG_ON()s.
773 *
774 * So here we add an artificial limit that subpage compression can only
775 * if the range is fully page aligned.
776 *
777 * In theory we only need to ensure the first page is fully covered, but
778 * the tailing partial page will be locked until the full compression
779 * finishes, delaying the write of other range.
780 *
781 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
782 * first to prevent any submitted async extent to unlock the full page.
783 * By this, we can ensure for subpage case that only the last async_cow
784 * will unlock the full page.
785 */
786 if (fs_info->sectorsize < PAGE_SIZE) {
787 if (!PAGE_ALIGNED(start) ||
788 !PAGE_ALIGNED(end + 1))
789 return 0;
790 }
791
792 /* force compress */
793 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
794 return 1;
795 /* defrag ioctl */
796 if (inode->defrag_compress)
797 return 1;
798 /* bad compression ratios */
799 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
800 return 0;
801 if (btrfs_test_opt(fs_info, COMPRESS) ||
802 inode->flags & BTRFS_INODE_COMPRESS ||
803 inode->prop_compress)
804 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
805 return 0;
806 }
807
inode_should_defrag(struct btrfs_inode * inode,u64 start,u64 end,u64 num_bytes,u32 small_write)808 static inline void inode_should_defrag(struct btrfs_inode *inode,
809 u64 start, u64 end, u64 num_bytes, u32 small_write)
810 {
811 /* If this is a small write inside eof, kick off a defrag */
812 if (num_bytes < small_write &&
813 (start > 0 || end + 1 < inode->disk_i_size))
814 btrfs_add_inode_defrag(NULL, inode, small_write);
815 }
816
817 /*
818 * Work queue call back to started compression on a file and pages.
819 *
820 * This is done inside an ordered work queue, and the compression is spread
821 * across many cpus. The actual IO submission is step two, and the ordered work
822 * queue takes care of making sure that happens in the same order things were
823 * put onto the queue by writepages and friends.
824 *
825 * If this code finds it can't get good compression, it puts an entry onto the
826 * work queue to write the uncompressed bytes. This makes sure that both
827 * compressed inodes and uncompressed inodes are written in the same order that
828 * the flusher thread sent them down.
829 */
compress_file_range(struct btrfs_work * work)830 static void compress_file_range(struct btrfs_work *work)
831 {
832 struct async_chunk *async_chunk =
833 container_of(work, struct async_chunk, work);
834 struct btrfs_inode *inode = async_chunk->inode;
835 struct btrfs_fs_info *fs_info = inode->root->fs_info;
836 struct address_space *mapping = inode->vfs_inode.i_mapping;
837 u64 blocksize = fs_info->sectorsize;
838 u64 start = async_chunk->start;
839 u64 end = async_chunk->end;
840 u64 actual_end;
841 u64 i_size;
842 int ret = 0;
843 struct page **pages;
844 unsigned long nr_pages;
845 unsigned long total_compressed = 0;
846 unsigned long total_in = 0;
847 unsigned int poff;
848 int i;
849 int compress_type = fs_info->compress_type;
850
851 inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
852
853 /*
854 * We need to call clear_page_dirty_for_io on each page in the range.
855 * Otherwise applications with the file mmap'd can wander in and change
856 * the page contents while we are compressing them.
857 */
858 extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end);
859
860 /*
861 * We need to save i_size before now because it could change in between
862 * us evaluating the size and assigning it. This is because we lock and
863 * unlock the page in truncate and fallocate, and then modify the i_size
864 * later on.
865 *
866 * The barriers are to emulate READ_ONCE, remove that once i_size_read
867 * does that for us.
868 */
869 barrier();
870 i_size = i_size_read(&inode->vfs_inode);
871 barrier();
872 actual_end = min_t(u64, i_size, end + 1);
873 again:
874 pages = NULL;
875 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
876 nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES);
877
878 /*
879 * we don't want to send crud past the end of i_size through
880 * compression, that's just a waste of CPU time. So, if the
881 * end of the file is before the start of our current
882 * requested range of bytes, we bail out to the uncompressed
883 * cleanup code that can deal with all of this.
884 *
885 * It isn't really the fastest way to fix things, but this is a
886 * very uncommon corner.
887 */
888 if (actual_end <= start)
889 goto cleanup_and_bail_uncompressed;
890
891 total_compressed = actual_end - start;
892
893 /*
894 * Skip compression for a small file range(<=blocksize) that
895 * isn't an inline extent, since it doesn't save disk space at all.
896 */
897 if (total_compressed <= blocksize &&
898 (start > 0 || end + 1 < inode->disk_i_size))
899 goto cleanup_and_bail_uncompressed;
900
901 /*
902 * For subpage case, we require full page alignment for the sector
903 * aligned range.
904 * Thus we must also check against @actual_end, not just @end.
905 */
906 if (blocksize < PAGE_SIZE) {
907 if (!PAGE_ALIGNED(start) ||
908 !PAGE_ALIGNED(round_up(actual_end, blocksize)))
909 goto cleanup_and_bail_uncompressed;
910 }
911
912 total_compressed = min_t(unsigned long, total_compressed,
913 BTRFS_MAX_UNCOMPRESSED);
914 total_in = 0;
915 ret = 0;
916
917 /*
918 * We do compression for mount -o compress and when the inode has not
919 * been flagged as NOCOMPRESS. This flag can change at any time if we
920 * discover bad compression ratios.
921 */
922 if (!inode_need_compress(inode, start, end))
923 goto cleanup_and_bail_uncompressed;
924
925 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
926 if (!pages) {
927 /*
928 * Memory allocation failure is not a fatal error, we can fall
929 * back to uncompressed code.
930 */
931 goto cleanup_and_bail_uncompressed;
932 }
933
934 if (inode->defrag_compress)
935 compress_type = inode->defrag_compress;
936 else if (inode->prop_compress)
937 compress_type = inode->prop_compress;
938
939 /* Compression level is applied here. */
940 ret = btrfs_compress_pages(compress_type | (fs_info->compress_level << 4),
941 mapping, start, pages, &nr_pages, &total_in,
942 &total_compressed);
943 if (ret)
944 goto mark_incompressible;
945
946 /*
947 * Zero the tail end of the last page, as we might be sending it down
948 * to disk.
949 */
950 poff = offset_in_page(total_compressed);
951 if (poff)
952 memzero_page(pages[nr_pages - 1], poff, PAGE_SIZE - poff);
953
954 /*
955 * Try to create an inline extent.
956 *
957 * If we didn't compress the entire range, try to create an uncompressed
958 * inline extent, else a compressed one.
959 *
960 * Check cow_file_range() for why we don't even try to create inline
961 * extent for the subpage case.
962 */
963 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
964 if (total_in < actual_end) {
965 ret = cow_file_range_inline(inode, actual_end, 0,
966 BTRFS_COMPRESS_NONE, NULL,
967 false);
968 } else {
969 ret = cow_file_range_inline(inode, actual_end,
970 total_compressed,
971 compress_type, pages,
972 false);
973 }
974 if (ret <= 0) {
975 unsigned long clear_flags = EXTENT_DELALLOC |
976 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
977 EXTENT_DO_ACCOUNTING;
978
979 if (ret < 0)
980 mapping_set_error(mapping, -EIO);
981
982 /*
983 * inline extent creation worked or returned error,
984 * we don't need to create any more async work items.
985 * Unlock and free up our temp pages.
986 *
987 * We use DO_ACCOUNTING here because we need the
988 * delalloc_release_metadata to be done _after_ we drop
989 * our outstanding extent for clearing delalloc for this
990 * range.
991 */
992 extent_clear_unlock_delalloc(inode, start, end,
993 NULL,
994 clear_flags,
995 PAGE_UNLOCK |
996 PAGE_START_WRITEBACK |
997 PAGE_END_WRITEBACK);
998 goto free_pages;
999 }
1000 }
1001
1002 /*
1003 * We aren't doing an inline extent. Round the compressed size up to a
1004 * block size boundary so the allocator does sane things.
1005 */
1006 total_compressed = ALIGN(total_compressed, blocksize);
1007
1008 /*
1009 * One last check to make sure the compression is really a win, compare
1010 * the page count read with the blocks on disk, compression must free at
1011 * least one sector.
1012 */
1013 total_in = round_up(total_in, fs_info->sectorsize);
1014 if (total_compressed + blocksize > total_in)
1015 goto mark_incompressible;
1016
1017 /*
1018 * The async work queues will take care of doing actual allocation on
1019 * disk for these compressed pages, and will submit the bios.
1020 */
1021 ret = add_async_extent(async_chunk, start, total_in, total_compressed, pages,
1022 nr_pages, compress_type);
1023 BUG_ON(ret);
1024 if (start + total_in < end) {
1025 start += total_in;
1026 cond_resched();
1027 goto again;
1028 }
1029 return;
1030
1031 mark_incompressible:
1032 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && !inode->prop_compress)
1033 inode->flags |= BTRFS_INODE_NOCOMPRESS;
1034 cleanup_and_bail_uncompressed:
1035 ret = add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
1036 BTRFS_COMPRESS_NONE);
1037 BUG_ON(ret);
1038 free_pages:
1039 if (pages) {
1040 for (i = 0; i < nr_pages; i++) {
1041 WARN_ON(pages[i]->mapping);
1042 put_page(pages[i]);
1043 }
1044 kfree(pages);
1045 }
1046 }
1047
free_async_extent_pages(struct async_extent * async_extent)1048 static void free_async_extent_pages(struct async_extent *async_extent)
1049 {
1050 int i;
1051
1052 if (!async_extent->pages)
1053 return;
1054
1055 for (i = 0; i < async_extent->nr_pages; i++) {
1056 WARN_ON(async_extent->pages[i]->mapping);
1057 put_page(async_extent->pages[i]);
1058 }
1059 kfree(async_extent->pages);
1060 async_extent->nr_pages = 0;
1061 async_extent->pages = NULL;
1062 }
1063
submit_uncompressed_range(struct btrfs_inode * inode,struct async_extent * async_extent,struct page * locked_page)1064 static void submit_uncompressed_range(struct btrfs_inode *inode,
1065 struct async_extent *async_extent,
1066 struct page *locked_page)
1067 {
1068 u64 start = async_extent->start;
1069 u64 end = async_extent->start + async_extent->ram_size - 1;
1070 int ret;
1071 struct writeback_control wbc = {
1072 .sync_mode = WB_SYNC_ALL,
1073 .range_start = start,
1074 .range_end = end,
1075 .no_cgroup_owner = 1,
1076 };
1077
1078 wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode);
1079 ret = run_delalloc_cow(inode, locked_page, start, end, &wbc, false);
1080 wbc_detach_inode(&wbc);
1081 if (ret < 0) {
1082 btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
1083 if (locked_page) {
1084 const u64 page_start = page_offset(locked_page);
1085
1086 set_page_writeback(locked_page);
1087 end_page_writeback(locked_page);
1088 btrfs_mark_ordered_io_finished(inode, locked_page,
1089 page_start, PAGE_SIZE,
1090 !ret);
1091 mapping_set_error(locked_page->mapping, ret);
1092 unlock_page(locked_page);
1093 }
1094 }
1095 }
1096
submit_one_async_extent(struct async_chunk * async_chunk,struct async_extent * async_extent,u64 * alloc_hint)1097 static void submit_one_async_extent(struct async_chunk *async_chunk,
1098 struct async_extent *async_extent,
1099 u64 *alloc_hint)
1100 {
1101 struct btrfs_inode *inode = async_chunk->inode;
1102 struct extent_io_tree *io_tree = &inode->io_tree;
1103 struct btrfs_root *root = inode->root;
1104 struct btrfs_fs_info *fs_info = root->fs_info;
1105 struct btrfs_ordered_extent *ordered;
1106 struct btrfs_key ins;
1107 struct page *locked_page = NULL;
1108 struct extent_map *em;
1109 int ret = 0;
1110 u64 start = async_extent->start;
1111 u64 end = async_extent->start + async_extent->ram_size - 1;
1112
1113 if (async_chunk->blkcg_css)
1114 kthread_associate_blkcg(async_chunk->blkcg_css);
1115
1116 /*
1117 * If async_chunk->locked_page is in the async_extent range, we need to
1118 * handle it.
1119 */
1120 if (async_chunk->locked_page) {
1121 u64 locked_page_start = page_offset(async_chunk->locked_page);
1122 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
1123
1124 if (!(start >= locked_page_end || end <= locked_page_start))
1125 locked_page = async_chunk->locked_page;
1126 }
1127 lock_extent(io_tree, start, end, NULL);
1128
1129 if (async_extent->compress_type == BTRFS_COMPRESS_NONE) {
1130 submit_uncompressed_range(inode, async_extent, locked_page);
1131 goto done;
1132 }
1133
1134 ret = btrfs_reserve_extent(root, async_extent->ram_size,
1135 async_extent->compressed_size,
1136 async_extent->compressed_size,
1137 0, *alloc_hint, &ins, 1, 1);
1138 if (ret) {
1139 /*
1140 * We can't reserve contiguous space for the compressed size.
1141 * Unlikely, but it's possible that we could have enough
1142 * non-contiguous space for the uncompressed size instead. So
1143 * fall back to uncompressed.
1144 */
1145 submit_uncompressed_range(inode, async_extent, locked_page);
1146 goto done;
1147 }
1148
1149 /* Here we're doing allocation and writeback of the compressed pages */
1150 em = create_io_em(inode, start,
1151 async_extent->ram_size, /* len */
1152 start, /* orig_start */
1153 ins.objectid, /* block_start */
1154 ins.offset, /* block_len */
1155 ins.offset, /* orig_block_len */
1156 async_extent->ram_size, /* ram_bytes */
1157 async_extent->compress_type,
1158 BTRFS_ORDERED_COMPRESSED);
1159 if (IS_ERR(em)) {
1160 ret = PTR_ERR(em);
1161 goto out_free_reserve;
1162 }
1163 free_extent_map(em);
1164
1165 ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */
1166 async_extent->ram_size, /* num_bytes */
1167 async_extent->ram_size, /* ram_bytes */
1168 ins.objectid, /* disk_bytenr */
1169 ins.offset, /* disk_num_bytes */
1170 0, /* offset */
1171 1 << BTRFS_ORDERED_COMPRESSED,
1172 async_extent->compress_type);
1173 if (IS_ERR(ordered)) {
1174 btrfs_drop_extent_map_range(inode, start, end, false);
1175 ret = PTR_ERR(ordered);
1176 goto out_free_reserve;
1177 }
1178 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1179
1180 /* Clear dirty, set writeback and unlock the pages. */
1181 extent_clear_unlock_delalloc(inode, start, end,
1182 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1183 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1184 btrfs_submit_compressed_write(ordered,
1185 async_extent->pages, /* compressed_pages */
1186 async_extent->nr_pages,
1187 async_chunk->write_flags, true);
1188 *alloc_hint = ins.objectid + ins.offset;
1189 done:
1190 if (async_chunk->blkcg_css)
1191 kthread_associate_blkcg(NULL);
1192 kfree(async_extent);
1193 return;
1194
1195 out_free_reserve:
1196 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1197 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1198 mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1199 extent_clear_unlock_delalloc(inode, start, end,
1200 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1201 EXTENT_DELALLOC_NEW |
1202 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1203 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1204 PAGE_END_WRITEBACK);
1205 free_async_extent_pages(async_extent);
1206 if (async_chunk->blkcg_css)
1207 kthread_associate_blkcg(NULL);
1208 btrfs_debug(fs_info,
1209 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1210 root->root_key.objectid, btrfs_ino(inode), start,
1211 async_extent->ram_size, ret);
1212 kfree(async_extent);
1213 }
1214
get_extent_allocation_hint(struct btrfs_inode * inode,u64 start,u64 num_bytes)1215 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1216 u64 num_bytes)
1217 {
1218 struct extent_map_tree *em_tree = &inode->extent_tree;
1219 struct extent_map *em;
1220 u64 alloc_hint = 0;
1221
1222 read_lock(&em_tree->lock);
1223 em = search_extent_mapping(em_tree, start, num_bytes);
1224 if (em) {
1225 /*
1226 * if block start isn't an actual block number then find the
1227 * first block in this inode and use that as a hint. If that
1228 * block is also bogus then just don't worry about it.
1229 */
1230 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1231 free_extent_map(em);
1232 em = search_extent_mapping(em_tree, 0, 0);
1233 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1234 alloc_hint = em->block_start;
1235 if (em)
1236 free_extent_map(em);
1237 } else {
1238 alloc_hint = em->block_start;
1239 free_extent_map(em);
1240 }
1241 }
1242 read_unlock(&em_tree->lock);
1243
1244 return alloc_hint;
1245 }
1246
1247 /*
1248 * when extent_io.c finds a delayed allocation range in the file,
1249 * the call backs end up in this code. The basic idea is to
1250 * allocate extents on disk for the range, and create ordered data structs
1251 * in ram to track those extents.
1252 *
1253 * locked_page is the page that writepage had locked already. We use
1254 * it to make sure we don't do extra locks or unlocks.
1255 *
1256 * When this function fails, it unlocks all pages except @locked_page.
1257 *
1258 * When this function successfully creates an inline extent, it returns 1 and
1259 * unlocks all pages including locked_page and starts I/O on them.
1260 * (In reality inline extents are limited to a single page, so locked_page is
1261 * the only page handled anyway).
1262 *
1263 * When this function succeed and creates a normal extent, the page locking
1264 * status depends on the passed in flags:
1265 *
1266 * - If @keep_locked is set, all pages are kept locked.
1267 * - Else all pages except for @locked_page are unlocked.
1268 *
1269 * When a failure happens in the second or later iteration of the
1270 * while-loop, the ordered extents created in previous iterations are kept
1271 * intact. So, the caller must clean them up by calling
1272 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1273 * example.
1274 */
cow_file_range(struct btrfs_inode * inode,struct page * locked_page,u64 start,u64 end,u64 * done_offset,bool keep_locked,bool no_inline)1275 static noinline int cow_file_range(struct btrfs_inode *inode,
1276 struct page *locked_page, u64 start, u64 end,
1277 u64 *done_offset,
1278 bool keep_locked, bool no_inline)
1279 {
1280 struct btrfs_root *root = inode->root;
1281 struct btrfs_fs_info *fs_info = root->fs_info;
1282 u64 alloc_hint = 0;
1283 u64 orig_start = start;
1284 u64 num_bytes;
1285 unsigned long ram_size;
1286 u64 cur_alloc_size = 0;
1287 u64 min_alloc_size;
1288 u64 blocksize = fs_info->sectorsize;
1289 struct btrfs_key ins;
1290 struct extent_map *em;
1291 unsigned clear_bits;
1292 unsigned long page_ops;
1293 bool extent_reserved = false;
1294 int ret = 0;
1295
1296 if (btrfs_is_free_space_inode(inode)) {
1297 ret = -EINVAL;
1298 goto out_unlock;
1299 }
1300
1301 num_bytes = ALIGN(end - start + 1, blocksize);
1302 num_bytes = max(blocksize, num_bytes);
1303 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1304
1305 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1306
1307 /*
1308 * Due to the page size limit, for subpage we can only trigger the
1309 * writeback for the dirty sectors of page, that means data writeback
1310 * is doing more writeback than what we want.
1311 *
1312 * This is especially unexpected for some call sites like fallocate,
1313 * where we only increase i_size after everything is done.
1314 * This means we can trigger inline extent even if we didn't want to.
1315 * So here we skip inline extent creation completely.
1316 */
1317 if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) {
1318 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1319 end + 1);
1320
1321 /* lets try to make an inline extent */
1322 ret = cow_file_range_inline(inode, actual_end, 0,
1323 BTRFS_COMPRESS_NONE, NULL, false);
1324 if (ret == 0) {
1325 /*
1326 * We use DO_ACCOUNTING here because we need the
1327 * delalloc_release_metadata to be run _after_ we drop
1328 * our outstanding extent for clearing delalloc for this
1329 * range.
1330 */
1331 extent_clear_unlock_delalloc(inode, start, end,
1332 locked_page,
1333 EXTENT_LOCKED | EXTENT_DELALLOC |
1334 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1335 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1336 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1337 /*
1338 * locked_page is locked by the caller of
1339 * writepage_delalloc(), not locked by
1340 * __process_pages_contig().
1341 *
1342 * We can't let __process_pages_contig() to unlock it,
1343 * as it doesn't have any subpage::writers recorded.
1344 *
1345 * Here we manually unlock the page, since the caller
1346 * can't determine if it's an inline extent or a
1347 * compressed extent.
1348 */
1349 unlock_page(locked_page);
1350 ret = 1;
1351 goto done;
1352 } else if (ret < 0) {
1353 goto out_unlock;
1354 }
1355 }
1356
1357 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1358
1359 /*
1360 * Relocation relies on the relocated extents to have exactly the same
1361 * size as the original extents. Normally writeback for relocation data
1362 * extents follows a NOCOW path because relocation preallocates the
1363 * extents. However, due to an operation such as scrub turning a block
1364 * group to RO mode, it may fallback to COW mode, so we must make sure
1365 * an extent allocated during COW has exactly the requested size and can
1366 * not be split into smaller extents, otherwise relocation breaks and
1367 * fails during the stage where it updates the bytenr of file extent
1368 * items.
1369 */
1370 if (btrfs_is_data_reloc_root(root))
1371 min_alloc_size = num_bytes;
1372 else
1373 min_alloc_size = fs_info->sectorsize;
1374
1375 while (num_bytes > 0) {
1376 struct btrfs_ordered_extent *ordered;
1377
1378 cur_alloc_size = num_bytes;
1379 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1380 min_alloc_size, 0, alloc_hint,
1381 &ins, 1, 1);
1382 if (ret == -EAGAIN) {
1383 /*
1384 * btrfs_reserve_extent only returns -EAGAIN for zoned
1385 * file systems, which is an indication that there are
1386 * no active zones to allocate from at the moment.
1387 *
1388 * If this is the first loop iteration, wait for at
1389 * least one zone to finish before retrying the
1390 * allocation. Otherwise ask the caller to write out
1391 * the already allocated blocks before coming back to
1392 * us, or return -ENOSPC if it can't handle retries.
1393 */
1394 ASSERT(btrfs_is_zoned(fs_info));
1395 if (start == orig_start) {
1396 wait_on_bit_io(&inode->root->fs_info->flags,
1397 BTRFS_FS_NEED_ZONE_FINISH,
1398 TASK_UNINTERRUPTIBLE);
1399 continue;
1400 }
1401 if (done_offset) {
1402 *done_offset = start - 1;
1403 return 0;
1404 }
1405 ret = -ENOSPC;
1406 }
1407 if (ret < 0)
1408 goto out_unlock;
1409 cur_alloc_size = ins.offset;
1410 extent_reserved = true;
1411
1412 ram_size = ins.offset;
1413 em = create_io_em(inode, start, ins.offset, /* len */
1414 start, /* orig_start */
1415 ins.objectid, /* block_start */
1416 ins.offset, /* block_len */
1417 ins.offset, /* orig_block_len */
1418 ram_size, /* ram_bytes */
1419 BTRFS_COMPRESS_NONE, /* compress_type */
1420 BTRFS_ORDERED_REGULAR /* type */);
1421 if (IS_ERR(em)) {
1422 ret = PTR_ERR(em);
1423 goto out_reserve;
1424 }
1425 free_extent_map(em);
1426
1427 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1428 ram_size, ins.objectid, cur_alloc_size,
1429 0, 1 << BTRFS_ORDERED_REGULAR,
1430 BTRFS_COMPRESS_NONE);
1431 if (IS_ERR(ordered)) {
1432 ret = PTR_ERR(ordered);
1433 goto out_drop_extent_cache;
1434 }
1435
1436 if (btrfs_is_data_reloc_root(root)) {
1437 ret = btrfs_reloc_clone_csums(ordered);
1438
1439 /*
1440 * Only drop cache here, and process as normal.
1441 *
1442 * We must not allow extent_clear_unlock_delalloc()
1443 * at out_unlock label to free meta of this ordered
1444 * extent, as its meta should be freed by
1445 * btrfs_finish_ordered_io().
1446 *
1447 * So we must continue until @start is increased to
1448 * skip current ordered extent.
1449 */
1450 if (ret)
1451 btrfs_drop_extent_map_range(inode, start,
1452 start + ram_size - 1,
1453 false);
1454 }
1455 btrfs_put_ordered_extent(ordered);
1456
1457 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1458
1459 /*
1460 * We're not doing compressed IO, don't unlock the first page
1461 * (which the caller expects to stay locked), don't clear any
1462 * dirty bits and don't set any writeback bits
1463 *
1464 * Do set the Ordered (Private2) bit so we know this page was
1465 * properly setup for writepage.
1466 */
1467 page_ops = (keep_locked ? 0 : PAGE_UNLOCK);
1468 page_ops |= PAGE_SET_ORDERED;
1469
1470 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1471 locked_page,
1472 EXTENT_LOCKED | EXTENT_DELALLOC,
1473 page_ops);
1474 if (num_bytes < cur_alloc_size)
1475 num_bytes = 0;
1476 else
1477 num_bytes -= cur_alloc_size;
1478 alloc_hint = ins.objectid + ins.offset;
1479 start += cur_alloc_size;
1480 extent_reserved = false;
1481
1482 /*
1483 * btrfs_reloc_clone_csums() error, since start is increased
1484 * extent_clear_unlock_delalloc() at out_unlock label won't
1485 * free metadata of current ordered extent, we're OK to exit.
1486 */
1487 if (ret)
1488 goto out_unlock;
1489 }
1490 done:
1491 if (done_offset)
1492 *done_offset = end;
1493 return ret;
1494
1495 out_drop_extent_cache:
1496 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1497 out_reserve:
1498 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1499 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1500 out_unlock:
1501 /*
1502 * Now, we have three regions to clean up:
1503 *
1504 * |-------(1)----|---(2)---|-------------(3)----------|
1505 * `- orig_start `- start `- start + cur_alloc_size `- end
1506 *
1507 * We process each region below.
1508 */
1509
1510 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1511 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1512 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1513
1514 /*
1515 * For the range (1). We have already instantiated the ordered extents
1516 * for this region. They are cleaned up by
1517 * btrfs_cleanup_ordered_extents() in e.g,
1518 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1519 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1520 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1521 * function.
1522 *
1523 * However, in case of @keep_locked, we still need to unlock the pages
1524 * (except @locked_page) to ensure all the pages are unlocked.
1525 */
1526 if (keep_locked && orig_start < start) {
1527 if (!locked_page)
1528 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1529 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1530 locked_page, 0, page_ops);
1531 }
1532
1533 /*
1534 * For the range (2). If we reserved an extent for our delalloc range
1535 * (or a subrange) and failed to create the respective ordered extent,
1536 * then it means that when we reserved the extent we decremented the
1537 * extent's size from the data space_info's bytes_may_use counter and
1538 * incremented the space_info's bytes_reserved counter by the same
1539 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1540 * to decrement again the data space_info's bytes_may_use counter,
1541 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1542 */
1543 if (extent_reserved) {
1544 extent_clear_unlock_delalloc(inode, start,
1545 start + cur_alloc_size - 1,
1546 locked_page,
1547 clear_bits,
1548 page_ops);
1549 start += cur_alloc_size;
1550 }
1551
1552 /*
1553 * For the range (3). We never touched the region. In addition to the
1554 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1555 * space_info's bytes_may_use counter, reserved in
1556 * btrfs_check_data_free_space().
1557 */
1558 if (start < end) {
1559 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1560 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1561 clear_bits, page_ops);
1562 }
1563 return ret;
1564 }
1565
1566 /*
1567 * Phase two of compressed writeback. This is the ordered portion of the code,
1568 * which only gets called in the order the work was queued. We walk all the
1569 * async extents created by compress_file_range and send them down to the disk.
1570 */
submit_compressed_extents(struct btrfs_work * work)1571 static noinline void submit_compressed_extents(struct btrfs_work *work)
1572 {
1573 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1574 work);
1575 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1576 struct async_extent *async_extent;
1577 unsigned long nr_pages;
1578 u64 alloc_hint = 0;
1579
1580 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1581 PAGE_SHIFT;
1582
1583 while (!list_empty(&async_chunk->extents)) {
1584 async_extent = list_entry(async_chunk->extents.next,
1585 struct async_extent, list);
1586 list_del(&async_extent->list);
1587 submit_one_async_extent(async_chunk, async_extent, &alloc_hint);
1588 }
1589
1590 /* atomic_sub_return implies a barrier */
1591 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1592 5 * SZ_1M)
1593 cond_wake_up_nomb(&fs_info->async_submit_wait);
1594 }
1595
async_cow_free(struct btrfs_work * work)1596 static noinline void async_cow_free(struct btrfs_work *work)
1597 {
1598 struct async_chunk *async_chunk;
1599 struct async_cow *async_cow;
1600
1601 async_chunk = container_of(work, struct async_chunk, work);
1602 btrfs_add_delayed_iput(async_chunk->inode);
1603 if (async_chunk->blkcg_css)
1604 css_put(async_chunk->blkcg_css);
1605
1606 async_cow = async_chunk->async_cow;
1607 if (atomic_dec_and_test(&async_cow->num_chunks))
1608 kvfree(async_cow);
1609 }
1610
run_delalloc_compressed(struct btrfs_inode * inode,struct page * locked_page,u64 start,u64 end,struct writeback_control * wbc)1611 static bool run_delalloc_compressed(struct btrfs_inode *inode,
1612 struct page *locked_page, u64 start,
1613 u64 end, struct writeback_control *wbc)
1614 {
1615 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1616 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1617 struct async_cow *ctx;
1618 struct async_chunk *async_chunk;
1619 unsigned long nr_pages;
1620 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1621 int i;
1622 unsigned nofs_flag;
1623 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1624
1625 nofs_flag = memalloc_nofs_save();
1626 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1627 memalloc_nofs_restore(nofs_flag);
1628 if (!ctx)
1629 return false;
1630
1631 unlock_extent(&inode->io_tree, start, end, NULL);
1632 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1633
1634 async_chunk = ctx->chunks;
1635 atomic_set(&ctx->num_chunks, num_chunks);
1636
1637 for (i = 0; i < num_chunks; i++) {
1638 u64 cur_end = min(end, start + SZ_512K - 1);
1639
1640 /*
1641 * igrab is called higher up in the call chain, take only the
1642 * lightweight reference for the callback lifetime
1643 */
1644 ihold(&inode->vfs_inode);
1645 async_chunk[i].async_cow = ctx;
1646 async_chunk[i].inode = inode;
1647 async_chunk[i].start = start;
1648 async_chunk[i].end = cur_end;
1649 async_chunk[i].write_flags = write_flags;
1650 INIT_LIST_HEAD(&async_chunk[i].extents);
1651
1652 /*
1653 * The locked_page comes all the way from writepage and its
1654 * the original page we were actually given. As we spread
1655 * this large delalloc region across multiple async_chunk
1656 * structs, only the first struct needs a pointer to locked_page
1657 *
1658 * This way we don't need racey decisions about who is supposed
1659 * to unlock it.
1660 */
1661 if (locked_page) {
1662 /*
1663 * Depending on the compressibility, the pages might or
1664 * might not go through async. We want all of them to
1665 * be accounted against wbc once. Let's do it here
1666 * before the paths diverge. wbc accounting is used
1667 * only for foreign writeback detection and doesn't
1668 * need full accuracy. Just account the whole thing
1669 * against the first page.
1670 */
1671 wbc_account_cgroup_owner(wbc, locked_page,
1672 cur_end - start);
1673 async_chunk[i].locked_page = locked_page;
1674 locked_page = NULL;
1675 } else {
1676 async_chunk[i].locked_page = NULL;
1677 }
1678
1679 if (blkcg_css != blkcg_root_css) {
1680 css_get(blkcg_css);
1681 async_chunk[i].blkcg_css = blkcg_css;
1682 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1683 } else {
1684 async_chunk[i].blkcg_css = NULL;
1685 }
1686
1687 btrfs_init_work(&async_chunk[i].work, compress_file_range,
1688 submit_compressed_extents, async_cow_free);
1689
1690 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1691 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1692
1693 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1694
1695 start = cur_end + 1;
1696 }
1697 return true;
1698 }
1699
1700 /*
1701 * Run the delalloc range from start to end, and write back any dirty pages
1702 * covered by the range.
1703 */
run_delalloc_cow(struct btrfs_inode * inode,struct page * locked_page,u64 start,u64 end,struct writeback_control * wbc,bool pages_dirty)1704 static noinline int run_delalloc_cow(struct btrfs_inode *inode,
1705 struct page *locked_page, u64 start,
1706 u64 end, struct writeback_control *wbc,
1707 bool pages_dirty)
1708 {
1709 u64 done_offset = end;
1710 int ret;
1711
1712 while (start <= end) {
1713 ret = cow_file_range(inode, locked_page, start, end, &done_offset,
1714 true, false);
1715 if (ret)
1716 return ret;
1717 extent_write_locked_range(&inode->vfs_inode, locked_page, start,
1718 done_offset, wbc, pages_dirty);
1719 start = done_offset + 1;
1720 }
1721
1722 return 1;
1723 }
1724
csum_exist_in_range(struct btrfs_fs_info * fs_info,u64 bytenr,u64 num_bytes,bool nowait)1725 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1726 u64 bytenr, u64 num_bytes, bool nowait)
1727 {
1728 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1729 struct btrfs_ordered_sum *sums;
1730 int ret;
1731 LIST_HEAD(list);
1732
1733 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1734 &list, 0, nowait);
1735 if (ret == 0 && list_empty(&list))
1736 return 0;
1737
1738 while (!list_empty(&list)) {
1739 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1740 list_del(&sums->list);
1741 kfree(sums);
1742 }
1743 if (ret < 0)
1744 return ret;
1745 return 1;
1746 }
1747
fallback_to_cow(struct btrfs_inode * inode,struct page * locked_page,const u64 start,const u64 end)1748 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1749 const u64 start, const u64 end)
1750 {
1751 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1752 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1753 const u64 range_bytes = end + 1 - start;
1754 struct extent_io_tree *io_tree = &inode->io_tree;
1755 u64 range_start = start;
1756 u64 count;
1757 int ret;
1758
1759 /*
1760 * If EXTENT_NORESERVE is set it means that when the buffered write was
1761 * made we had not enough available data space and therefore we did not
1762 * reserve data space for it, since we though we could do NOCOW for the
1763 * respective file range (either there is prealloc extent or the inode
1764 * has the NOCOW bit set).
1765 *
1766 * However when we need to fallback to COW mode (because for example the
1767 * block group for the corresponding extent was turned to RO mode by a
1768 * scrub or relocation) we need to do the following:
1769 *
1770 * 1) We increment the bytes_may_use counter of the data space info.
1771 * If COW succeeds, it allocates a new data extent and after doing
1772 * that it decrements the space info's bytes_may_use counter and
1773 * increments its bytes_reserved counter by the same amount (we do
1774 * this at btrfs_add_reserved_bytes()). So we need to increment the
1775 * bytes_may_use counter to compensate (when space is reserved at
1776 * buffered write time, the bytes_may_use counter is incremented);
1777 *
1778 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1779 * that if the COW path fails for any reason, it decrements (through
1780 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1781 * data space info, which we incremented in the step above.
1782 *
1783 * If we need to fallback to cow and the inode corresponds to a free
1784 * space cache inode or an inode of the data relocation tree, we must
1785 * also increment bytes_may_use of the data space_info for the same
1786 * reason. Space caches and relocated data extents always get a prealloc
1787 * extent for them, however scrub or balance may have set the block
1788 * group that contains that extent to RO mode and therefore force COW
1789 * when starting writeback.
1790 */
1791 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1792 EXTENT_NORESERVE, 0, NULL);
1793 if (count > 0 || is_space_ino || is_reloc_ino) {
1794 u64 bytes = count;
1795 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1796 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1797
1798 if (is_space_ino || is_reloc_ino)
1799 bytes = range_bytes;
1800
1801 spin_lock(&sinfo->lock);
1802 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1803 spin_unlock(&sinfo->lock);
1804
1805 if (count > 0)
1806 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1807 NULL);
1808 }
1809
1810 /*
1811 * Don't try to create inline extents, as a mix of inline extent that
1812 * is written out and unlocked directly and a normal NOCOW extent
1813 * doesn't work.
1814 */
1815 ret = cow_file_range(inode, locked_page, start, end, NULL, false, true);
1816 ASSERT(ret != 1);
1817 return ret;
1818 }
1819
1820 struct can_nocow_file_extent_args {
1821 /* Input fields. */
1822
1823 /* Start file offset of the range we want to NOCOW. */
1824 u64 start;
1825 /* End file offset (inclusive) of the range we want to NOCOW. */
1826 u64 end;
1827 bool writeback_path;
1828 bool strict;
1829 /*
1830 * Free the path passed to can_nocow_file_extent() once it's not needed
1831 * anymore.
1832 */
1833 bool free_path;
1834
1835 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1836
1837 u64 disk_bytenr;
1838 u64 disk_num_bytes;
1839 u64 extent_offset;
1840 /* Number of bytes that can be written to in NOCOW mode. */
1841 u64 num_bytes;
1842 };
1843
1844 /*
1845 * Check if we can NOCOW the file extent that the path points to.
1846 * This function may return with the path released, so the caller should check
1847 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1848 *
1849 * Returns: < 0 on error
1850 * 0 if we can not NOCOW
1851 * 1 if we can NOCOW
1852 */
can_nocow_file_extent(struct btrfs_path * path,struct btrfs_key * key,struct btrfs_inode * inode,struct can_nocow_file_extent_args * args)1853 static int can_nocow_file_extent(struct btrfs_path *path,
1854 struct btrfs_key *key,
1855 struct btrfs_inode *inode,
1856 struct can_nocow_file_extent_args *args)
1857 {
1858 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1859 struct extent_buffer *leaf = path->nodes[0];
1860 struct btrfs_root *root = inode->root;
1861 struct btrfs_file_extent_item *fi;
1862 u64 extent_end;
1863 u8 extent_type;
1864 int can_nocow = 0;
1865 int ret = 0;
1866 bool nowait = path->nowait;
1867
1868 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1869 extent_type = btrfs_file_extent_type(leaf, fi);
1870
1871 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1872 goto out;
1873
1874 /* Can't access these fields unless we know it's not an inline extent. */
1875 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1876 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1877 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1878
1879 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1880 extent_type == BTRFS_FILE_EXTENT_REG)
1881 goto out;
1882
1883 /*
1884 * If the extent was created before the generation where the last snapshot
1885 * for its subvolume was created, then this implies the extent is shared,
1886 * hence we must COW.
1887 */
1888 if (!args->strict &&
1889 btrfs_file_extent_generation(leaf, fi) <=
1890 btrfs_root_last_snapshot(&root->root_item))
1891 goto out;
1892
1893 /* An explicit hole, must COW. */
1894 if (args->disk_bytenr == 0)
1895 goto out;
1896
1897 /* Compressed/encrypted/encoded extents must be COWed. */
1898 if (btrfs_file_extent_compression(leaf, fi) ||
1899 btrfs_file_extent_encryption(leaf, fi) ||
1900 btrfs_file_extent_other_encoding(leaf, fi))
1901 goto out;
1902
1903 extent_end = btrfs_file_extent_end(path);
1904
1905 /*
1906 * The following checks can be expensive, as they need to take other
1907 * locks and do btree or rbtree searches, so release the path to avoid
1908 * blocking other tasks for too long.
1909 */
1910 btrfs_release_path(path);
1911
1912 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1913 key->offset - args->extent_offset,
1914 args->disk_bytenr, args->strict, path);
1915 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1916 if (ret != 0)
1917 goto out;
1918
1919 if (args->free_path) {
1920 /*
1921 * We don't need the path anymore, plus through the
1922 * csum_exist_in_range() call below we will end up allocating
1923 * another path. So free the path to avoid unnecessary extra
1924 * memory usage.
1925 */
1926 btrfs_free_path(path);
1927 path = NULL;
1928 }
1929
1930 /* If there are pending snapshots for this root, we must COW. */
1931 if (args->writeback_path && !is_freespace_inode &&
1932 atomic_read(&root->snapshot_force_cow))
1933 goto out;
1934
1935 args->disk_bytenr += args->extent_offset;
1936 args->disk_bytenr += args->start - key->offset;
1937 args->num_bytes = min(args->end + 1, extent_end) - args->start;
1938
1939 /*
1940 * Force COW if csums exist in the range. This ensures that csums for a
1941 * given extent are either valid or do not exist.
1942 */
1943 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
1944 nowait);
1945 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1946 if (ret != 0)
1947 goto out;
1948
1949 can_nocow = 1;
1950 out:
1951 if (args->free_path && path)
1952 btrfs_free_path(path);
1953
1954 return ret < 0 ? ret : can_nocow;
1955 }
1956
1957 /*
1958 * when nowcow writeback call back. This checks for snapshots or COW copies
1959 * of the extents that exist in the file, and COWs the file as required.
1960 *
1961 * If no cow copies or snapshots exist, we write directly to the existing
1962 * blocks on disk
1963 */
run_delalloc_nocow(struct btrfs_inode * inode,struct page * locked_page,const u64 start,const u64 end)1964 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1965 struct page *locked_page,
1966 const u64 start, const u64 end)
1967 {
1968 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1969 struct btrfs_root *root = inode->root;
1970 struct btrfs_path *path;
1971 u64 cow_start = (u64)-1;
1972 u64 cur_offset = start;
1973 int ret;
1974 bool check_prev = true;
1975 u64 ino = btrfs_ino(inode);
1976 struct can_nocow_file_extent_args nocow_args = { 0 };
1977
1978 /*
1979 * Normally on a zoned device we're only doing COW writes, but in case
1980 * of relocation on a zoned filesystem serializes I/O so that we're only
1981 * writing sequentially and can end up here as well.
1982 */
1983 ASSERT(!btrfs_is_zoned(fs_info) || btrfs_is_data_reloc_root(root));
1984
1985 path = btrfs_alloc_path();
1986 if (!path) {
1987 ret = -ENOMEM;
1988 goto error;
1989 }
1990
1991 nocow_args.end = end;
1992 nocow_args.writeback_path = true;
1993
1994 while (1) {
1995 struct btrfs_block_group *nocow_bg = NULL;
1996 struct btrfs_ordered_extent *ordered;
1997 struct btrfs_key found_key;
1998 struct btrfs_file_extent_item *fi;
1999 struct extent_buffer *leaf;
2000 u64 extent_end;
2001 u64 ram_bytes;
2002 u64 nocow_end;
2003 int extent_type;
2004 bool is_prealloc;
2005
2006 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2007 cur_offset, 0);
2008 if (ret < 0)
2009 goto error;
2010
2011 /*
2012 * If there is no extent for our range when doing the initial
2013 * search, then go back to the previous slot as it will be the
2014 * one containing the search offset
2015 */
2016 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2017 leaf = path->nodes[0];
2018 btrfs_item_key_to_cpu(leaf, &found_key,
2019 path->slots[0] - 1);
2020 if (found_key.objectid == ino &&
2021 found_key.type == BTRFS_EXTENT_DATA_KEY)
2022 path->slots[0]--;
2023 }
2024 check_prev = false;
2025 next_slot:
2026 /* Go to next leaf if we have exhausted the current one */
2027 leaf = path->nodes[0];
2028 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2029 ret = btrfs_next_leaf(root, path);
2030 if (ret < 0)
2031 goto error;
2032 if (ret > 0)
2033 break;
2034 leaf = path->nodes[0];
2035 }
2036
2037 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2038
2039 /* Didn't find anything for our INO */
2040 if (found_key.objectid > ino)
2041 break;
2042 /*
2043 * Keep searching until we find an EXTENT_ITEM or there are no
2044 * more extents for this inode
2045 */
2046 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2047 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2048 path->slots[0]++;
2049 goto next_slot;
2050 }
2051
2052 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2053 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2054 found_key.offset > end)
2055 break;
2056
2057 /*
2058 * If the found extent starts after requested offset, then
2059 * adjust extent_end to be right before this extent begins
2060 */
2061 if (found_key.offset > cur_offset) {
2062 extent_end = found_key.offset;
2063 extent_type = 0;
2064 goto must_cow;
2065 }
2066
2067 /*
2068 * Found extent which begins before our range and potentially
2069 * intersect it
2070 */
2071 fi = btrfs_item_ptr(leaf, path->slots[0],
2072 struct btrfs_file_extent_item);
2073 extent_type = btrfs_file_extent_type(leaf, fi);
2074 /* If this is triggered then we have a memory corruption. */
2075 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2076 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2077 ret = -EUCLEAN;
2078 goto error;
2079 }
2080 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2081 extent_end = btrfs_file_extent_end(path);
2082
2083 /*
2084 * If the extent we got ends before our current offset, skip to
2085 * the next extent.
2086 */
2087 if (extent_end <= cur_offset) {
2088 path->slots[0]++;
2089 goto next_slot;
2090 }
2091
2092 nocow_args.start = cur_offset;
2093 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2094 if (ret < 0)
2095 goto error;
2096 if (ret == 0)
2097 goto must_cow;
2098
2099 ret = 0;
2100 nocow_bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2101 if (!nocow_bg) {
2102 must_cow:
2103 /*
2104 * If we can't perform NOCOW writeback for the range,
2105 * then record the beginning of the range that needs to
2106 * be COWed. It will be written out before the next
2107 * NOCOW range if we find one, or when exiting this
2108 * loop.
2109 */
2110 if (cow_start == (u64)-1)
2111 cow_start = cur_offset;
2112 cur_offset = extent_end;
2113 if (cur_offset > end)
2114 break;
2115 if (!path->nodes[0])
2116 continue;
2117 path->slots[0]++;
2118 goto next_slot;
2119 }
2120
2121 /*
2122 * COW range from cow_start to found_key.offset - 1. As the key
2123 * will contain the beginning of the first extent that can be
2124 * NOCOW, following one which needs to be COW'ed
2125 */
2126 if (cow_start != (u64)-1) {
2127 ret = fallback_to_cow(inode, locked_page,
2128 cow_start, found_key.offset - 1);
2129 cow_start = (u64)-1;
2130 if (ret) {
2131 btrfs_dec_nocow_writers(nocow_bg);
2132 goto error;
2133 }
2134 }
2135
2136 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2137 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2138 if (is_prealloc) {
2139 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2140 struct extent_map *em;
2141
2142 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2143 orig_start,
2144 nocow_args.disk_bytenr, /* block_start */
2145 nocow_args.num_bytes, /* block_len */
2146 nocow_args.disk_num_bytes, /* orig_block_len */
2147 ram_bytes, BTRFS_COMPRESS_NONE,
2148 BTRFS_ORDERED_PREALLOC);
2149 if (IS_ERR(em)) {
2150 btrfs_dec_nocow_writers(nocow_bg);
2151 ret = PTR_ERR(em);
2152 goto error;
2153 }
2154 free_extent_map(em);
2155 }
2156
2157 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2158 nocow_args.num_bytes, nocow_args.num_bytes,
2159 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2160 is_prealloc
2161 ? (1 << BTRFS_ORDERED_PREALLOC)
2162 : (1 << BTRFS_ORDERED_NOCOW),
2163 BTRFS_COMPRESS_NONE);
2164 btrfs_dec_nocow_writers(nocow_bg);
2165 if (IS_ERR(ordered)) {
2166 if (is_prealloc) {
2167 btrfs_drop_extent_map_range(inode, cur_offset,
2168 nocow_end, false);
2169 }
2170 ret = PTR_ERR(ordered);
2171 goto error;
2172 }
2173
2174 if (btrfs_is_data_reloc_root(root))
2175 /*
2176 * Error handled later, as we must prevent
2177 * extent_clear_unlock_delalloc() in error handler
2178 * from freeing metadata of created ordered extent.
2179 */
2180 ret = btrfs_reloc_clone_csums(ordered);
2181 btrfs_put_ordered_extent(ordered);
2182
2183 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2184 locked_page, EXTENT_LOCKED |
2185 EXTENT_DELALLOC |
2186 EXTENT_CLEAR_DATA_RESV,
2187 PAGE_UNLOCK | PAGE_SET_ORDERED);
2188
2189 cur_offset = extent_end;
2190
2191 /*
2192 * btrfs_reloc_clone_csums() error, now we're OK to call error
2193 * handler, as metadata for created ordered extent will only
2194 * be freed by btrfs_finish_ordered_io().
2195 */
2196 if (ret)
2197 goto error;
2198 if (cur_offset > end)
2199 break;
2200 }
2201 btrfs_release_path(path);
2202
2203 if (cur_offset <= end && cow_start == (u64)-1)
2204 cow_start = cur_offset;
2205
2206 if (cow_start != (u64)-1) {
2207 cur_offset = end;
2208 ret = fallback_to_cow(inode, locked_page, cow_start, end);
2209 cow_start = (u64)-1;
2210 if (ret)
2211 goto error;
2212 }
2213
2214 btrfs_free_path(path);
2215 return 0;
2216
2217 error:
2218 /*
2219 * If an error happened while a COW region is outstanding, cur_offset
2220 * needs to be reset to cow_start to ensure the COW region is unlocked
2221 * as well.
2222 */
2223 if (cow_start != (u64)-1)
2224 cur_offset = cow_start;
2225 if (cur_offset < end)
2226 extent_clear_unlock_delalloc(inode, cur_offset, end,
2227 locked_page, EXTENT_LOCKED |
2228 EXTENT_DELALLOC | EXTENT_DEFRAG |
2229 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2230 PAGE_START_WRITEBACK |
2231 PAGE_END_WRITEBACK);
2232 btrfs_free_path(path);
2233 return ret;
2234 }
2235
should_nocow(struct btrfs_inode * inode,u64 start,u64 end)2236 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2237 {
2238 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2239 if (inode->defrag_bytes &&
2240 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2241 0, NULL))
2242 return false;
2243 return true;
2244 }
2245 return false;
2246 }
2247
2248 /*
2249 * Function to process delayed allocation (create CoW) for ranges which are
2250 * being touched for the first time.
2251 */
btrfs_run_delalloc_range(struct btrfs_inode * inode,struct page * locked_page,u64 start,u64 end,struct writeback_control * wbc)2252 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2253 u64 start, u64 end, struct writeback_control *wbc)
2254 {
2255 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2256 int ret;
2257
2258 /*
2259 * The range must cover part of the @locked_page, or a return of 1
2260 * can confuse the caller.
2261 */
2262 ASSERT(!(end <= page_offset(locked_page) ||
2263 start >= page_offset(locked_page) + PAGE_SIZE));
2264
2265 if (should_nocow(inode, start, end)) {
2266 ret = run_delalloc_nocow(inode, locked_page, start, end);
2267 goto out;
2268 }
2269
2270 if (btrfs_inode_can_compress(inode) &&
2271 inode_need_compress(inode, start, end) &&
2272 run_delalloc_compressed(inode, locked_page, start, end, wbc))
2273 return 1;
2274
2275 if (zoned)
2276 ret = run_delalloc_cow(inode, locked_page, start, end, wbc,
2277 true);
2278 else
2279 ret = cow_file_range(inode, locked_page, start, end, NULL,
2280 false, false);
2281
2282 out:
2283 if (ret < 0)
2284 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2285 end - start + 1);
2286 return ret;
2287 }
2288
btrfs_split_delalloc_extent(struct btrfs_inode * inode,struct extent_state * orig,u64 split)2289 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2290 struct extent_state *orig, u64 split)
2291 {
2292 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2293 u64 size;
2294
2295 /* not delalloc, ignore it */
2296 if (!(orig->state & EXTENT_DELALLOC))
2297 return;
2298
2299 size = orig->end - orig->start + 1;
2300 if (size > fs_info->max_extent_size) {
2301 u32 num_extents;
2302 u64 new_size;
2303
2304 /*
2305 * See the explanation in btrfs_merge_delalloc_extent, the same
2306 * applies here, just in reverse.
2307 */
2308 new_size = orig->end - split + 1;
2309 num_extents = count_max_extents(fs_info, new_size);
2310 new_size = split - orig->start;
2311 num_extents += count_max_extents(fs_info, new_size);
2312 if (count_max_extents(fs_info, size) >= num_extents)
2313 return;
2314 }
2315
2316 spin_lock(&inode->lock);
2317 btrfs_mod_outstanding_extents(inode, 1);
2318 spin_unlock(&inode->lock);
2319 }
2320
2321 /*
2322 * Handle merged delayed allocation extents so we can keep track of new extents
2323 * that are just merged onto old extents, such as when we are doing sequential
2324 * writes, so we can properly account for the metadata space we'll need.
2325 */
btrfs_merge_delalloc_extent(struct btrfs_inode * inode,struct extent_state * new,struct extent_state * other)2326 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2327 struct extent_state *other)
2328 {
2329 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2330 u64 new_size, old_size;
2331 u32 num_extents;
2332
2333 /* not delalloc, ignore it */
2334 if (!(other->state & EXTENT_DELALLOC))
2335 return;
2336
2337 if (new->start > other->start)
2338 new_size = new->end - other->start + 1;
2339 else
2340 new_size = other->end - new->start + 1;
2341
2342 /* we're not bigger than the max, unreserve the space and go */
2343 if (new_size <= fs_info->max_extent_size) {
2344 spin_lock(&inode->lock);
2345 btrfs_mod_outstanding_extents(inode, -1);
2346 spin_unlock(&inode->lock);
2347 return;
2348 }
2349
2350 /*
2351 * We have to add up either side to figure out how many extents were
2352 * accounted for before we merged into one big extent. If the number of
2353 * extents we accounted for is <= the amount we need for the new range
2354 * then we can return, otherwise drop. Think of it like this
2355 *
2356 * [ 4k][MAX_SIZE]
2357 *
2358 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2359 * need 2 outstanding extents, on one side we have 1 and the other side
2360 * we have 1 so they are == and we can return. But in this case
2361 *
2362 * [MAX_SIZE+4k][MAX_SIZE+4k]
2363 *
2364 * Each range on their own accounts for 2 extents, but merged together
2365 * they are only 3 extents worth of accounting, so we need to drop in
2366 * this case.
2367 */
2368 old_size = other->end - other->start + 1;
2369 num_extents = count_max_extents(fs_info, old_size);
2370 old_size = new->end - new->start + 1;
2371 num_extents += count_max_extents(fs_info, old_size);
2372 if (count_max_extents(fs_info, new_size) >= num_extents)
2373 return;
2374
2375 spin_lock(&inode->lock);
2376 btrfs_mod_outstanding_extents(inode, -1);
2377 spin_unlock(&inode->lock);
2378 }
2379
btrfs_add_delalloc_inodes(struct btrfs_root * root,struct btrfs_inode * inode)2380 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2381 struct btrfs_inode *inode)
2382 {
2383 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2384
2385 spin_lock(&root->delalloc_lock);
2386 if (list_empty(&inode->delalloc_inodes)) {
2387 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2388 set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags);
2389 root->nr_delalloc_inodes++;
2390 if (root->nr_delalloc_inodes == 1) {
2391 spin_lock(&fs_info->delalloc_root_lock);
2392 BUG_ON(!list_empty(&root->delalloc_root));
2393 list_add_tail(&root->delalloc_root,
2394 &fs_info->delalloc_roots);
2395 spin_unlock(&fs_info->delalloc_root_lock);
2396 }
2397 }
2398 spin_unlock(&root->delalloc_lock);
2399 }
2400
__btrfs_del_delalloc_inode(struct btrfs_root * root,struct btrfs_inode * inode)2401 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2402 struct btrfs_inode *inode)
2403 {
2404 struct btrfs_fs_info *fs_info = root->fs_info;
2405
2406 if (!list_empty(&inode->delalloc_inodes)) {
2407 list_del_init(&inode->delalloc_inodes);
2408 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2409 &inode->runtime_flags);
2410 root->nr_delalloc_inodes--;
2411 if (!root->nr_delalloc_inodes) {
2412 ASSERT(list_empty(&root->delalloc_inodes));
2413 spin_lock(&fs_info->delalloc_root_lock);
2414 BUG_ON(list_empty(&root->delalloc_root));
2415 list_del_init(&root->delalloc_root);
2416 spin_unlock(&fs_info->delalloc_root_lock);
2417 }
2418 }
2419 }
2420
btrfs_del_delalloc_inode(struct btrfs_root * root,struct btrfs_inode * inode)2421 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2422 struct btrfs_inode *inode)
2423 {
2424 spin_lock(&root->delalloc_lock);
2425 __btrfs_del_delalloc_inode(root, inode);
2426 spin_unlock(&root->delalloc_lock);
2427 }
2428
2429 /*
2430 * Properly track delayed allocation bytes in the inode and to maintain the
2431 * list of inodes that have pending delalloc work to be done.
2432 */
btrfs_set_delalloc_extent(struct btrfs_inode * inode,struct extent_state * state,u32 bits)2433 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2434 u32 bits)
2435 {
2436 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2437
2438 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2439 WARN_ON(1);
2440 /*
2441 * set_bit and clear bit hooks normally require _irqsave/restore
2442 * but in this case, we are only testing for the DELALLOC
2443 * bit, which is only set or cleared with irqs on
2444 */
2445 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2446 struct btrfs_root *root = inode->root;
2447 u64 len = state->end + 1 - state->start;
2448 u32 num_extents = count_max_extents(fs_info, len);
2449 bool do_list = !btrfs_is_free_space_inode(inode);
2450
2451 spin_lock(&inode->lock);
2452 btrfs_mod_outstanding_extents(inode, num_extents);
2453 spin_unlock(&inode->lock);
2454
2455 /* For sanity tests */
2456 if (btrfs_is_testing(fs_info))
2457 return;
2458
2459 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2460 fs_info->delalloc_batch);
2461 spin_lock(&inode->lock);
2462 inode->delalloc_bytes += len;
2463 if (bits & EXTENT_DEFRAG)
2464 inode->defrag_bytes += len;
2465 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2466 &inode->runtime_flags))
2467 btrfs_add_delalloc_inodes(root, inode);
2468 spin_unlock(&inode->lock);
2469 }
2470
2471 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2472 (bits & EXTENT_DELALLOC_NEW)) {
2473 spin_lock(&inode->lock);
2474 inode->new_delalloc_bytes += state->end + 1 - state->start;
2475 spin_unlock(&inode->lock);
2476 }
2477 }
2478
2479 /*
2480 * Once a range is no longer delalloc this function ensures that proper
2481 * accounting happens.
2482 */
btrfs_clear_delalloc_extent(struct btrfs_inode * inode,struct extent_state * state,u32 bits)2483 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2484 struct extent_state *state, u32 bits)
2485 {
2486 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2487 u64 len = state->end + 1 - state->start;
2488 u32 num_extents = count_max_extents(fs_info, len);
2489
2490 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2491 spin_lock(&inode->lock);
2492 inode->defrag_bytes -= len;
2493 spin_unlock(&inode->lock);
2494 }
2495
2496 /*
2497 * set_bit and clear bit hooks normally require _irqsave/restore
2498 * but in this case, we are only testing for the DELALLOC
2499 * bit, which is only set or cleared with irqs on
2500 */
2501 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2502 struct btrfs_root *root = inode->root;
2503 bool do_list = !btrfs_is_free_space_inode(inode);
2504
2505 spin_lock(&inode->lock);
2506 btrfs_mod_outstanding_extents(inode, -num_extents);
2507 spin_unlock(&inode->lock);
2508
2509 /*
2510 * We don't reserve metadata space for space cache inodes so we
2511 * don't need to call delalloc_release_metadata if there is an
2512 * error.
2513 */
2514 if (bits & EXTENT_CLEAR_META_RESV &&
2515 root != fs_info->tree_root)
2516 btrfs_delalloc_release_metadata(inode, len, true);
2517
2518 /* For sanity tests. */
2519 if (btrfs_is_testing(fs_info))
2520 return;
2521
2522 if (!btrfs_is_data_reloc_root(root) &&
2523 do_list && !(state->state & EXTENT_NORESERVE) &&
2524 (bits & EXTENT_CLEAR_DATA_RESV))
2525 btrfs_free_reserved_data_space_noquota(fs_info, len);
2526
2527 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2528 fs_info->delalloc_batch);
2529 spin_lock(&inode->lock);
2530 inode->delalloc_bytes -= len;
2531 if (do_list && inode->delalloc_bytes == 0 &&
2532 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2533 &inode->runtime_flags))
2534 btrfs_del_delalloc_inode(root, inode);
2535 spin_unlock(&inode->lock);
2536 }
2537
2538 if ((state->state & EXTENT_DELALLOC_NEW) &&
2539 (bits & EXTENT_DELALLOC_NEW)) {
2540 spin_lock(&inode->lock);
2541 ASSERT(inode->new_delalloc_bytes >= len);
2542 inode->new_delalloc_bytes -= len;
2543 if (bits & EXTENT_ADD_INODE_BYTES)
2544 inode_add_bytes(&inode->vfs_inode, len);
2545 spin_unlock(&inode->lock);
2546 }
2547 }
2548
btrfs_extract_ordered_extent(struct btrfs_bio * bbio,struct btrfs_ordered_extent * ordered)2549 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2550 struct btrfs_ordered_extent *ordered)
2551 {
2552 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2553 u64 len = bbio->bio.bi_iter.bi_size;
2554 struct btrfs_ordered_extent *new;
2555 int ret;
2556
2557 /* Must always be called for the beginning of an ordered extent. */
2558 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2559 return -EINVAL;
2560
2561 /* No need to split if the ordered extent covers the entire bio. */
2562 if (ordered->disk_num_bytes == len) {
2563 refcount_inc(&ordered->refs);
2564 bbio->ordered = ordered;
2565 return 0;
2566 }
2567
2568 /*
2569 * Don't split the extent_map for NOCOW extents, as we're writing into
2570 * a pre-existing one.
2571 */
2572 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2573 ret = split_extent_map(bbio->inode, bbio->file_offset,
2574 ordered->num_bytes, len,
2575 ordered->disk_bytenr);
2576 if (ret)
2577 return ret;
2578 }
2579
2580 new = btrfs_split_ordered_extent(ordered, len);
2581 if (IS_ERR(new))
2582 return PTR_ERR(new);
2583 bbio->ordered = new;
2584 return 0;
2585 }
2586
2587 /*
2588 * given a list of ordered sums record them in the inode. This happens
2589 * at IO completion time based on sums calculated at bio submission time.
2590 */
add_pending_csums(struct btrfs_trans_handle * trans,struct list_head * list)2591 static int add_pending_csums(struct btrfs_trans_handle *trans,
2592 struct list_head *list)
2593 {
2594 struct btrfs_ordered_sum *sum;
2595 struct btrfs_root *csum_root = NULL;
2596 int ret;
2597
2598 list_for_each_entry(sum, list, list) {
2599 trans->adding_csums = true;
2600 if (!csum_root)
2601 csum_root = btrfs_csum_root(trans->fs_info,
2602 sum->logical);
2603 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2604 trans->adding_csums = false;
2605 if (ret)
2606 return ret;
2607 }
2608 return 0;
2609 }
2610
btrfs_find_new_delalloc_bytes(struct btrfs_inode * inode,const u64 start,const u64 len,struct extent_state ** cached_state)2611 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2612 const u64 start,
2613 const u64 len,
2614 struct extent_state **cached_state)
2615 {
2616 u64 search_start = start;
2617 const u64 end = start + len - 1;
2618
2619 while (search_start < end) {
2620 const u64 search_len = end - search_start + 1;
2621 struct extent_map *em;
2622 u64 em_len;
2623 int ret = 0;
2624
2625 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2626 if (IS_ERR(em))
2627 return PTR_ERR(em);
2628
2629 if (em->block_start != EXTENT_MAP_HOLE)
2630 goto next;
2631
2632 em_len = em->len;
2633 if (em->start < search_start)
2634 em_len -= search_start - em->start;
2635 if (em_len > search_len)
2636 em_len = search_len;
2637
2638 ret = set_extent_bit(&inode->io_tree, search_start,
2639 search_start + em_len - 1,
2640 EXTENT_DELALLOC_NEW, cached_state);
2641 next:
2642 search_start = extent_map_end(em);
2643 free_extent_map(em);
2644 if (ret)
2645 return ret;
2646 }
2647 return 0;
2648 }
2649
btrfs_set_extent_delalloc(struct btrfs_inode * inode,u64 start,u64 end,unsigned int extra_bits,struct extent_state ** cached_state)2650 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2651 unsigned int extra_bits,
2652 struct extent_state **cached_state)
2653 {
2654 WARN_ON(PAGE_ALIGNED(end));
2655
2656 if (start >= i_size_read(&inode->vfs_inode) &&
2657 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2658 /*
2659 * There can't be any extents following eof in this case so just
2660 * set the delalloc new bit for the range directly.
2661 */
2662 extra_bits |= EXTENT_DELALLOC_NEW;
2663 } else {
2664 int ret;
2665
2666 ret = btrfs_find_new_delalloc_bytes(inode, start,
2667 end + 1 - start,
2668 cached_state);
2669 if (ret)
2670 return ret;
2671 }
2672
2673 return set_extent_bit(&inode->io_tree, start, end,
2674 EXTENT_DELALLOC | extra_bits, cached_state);
2675 }
2676
2677 /* see btrfs_writepage_start_hook for details on why this is required */
2678 struct btrfs_writepage_fixup {
2679 struct page *page;
2680 struct btrfs_inode *inode;
2681 struct btrfs_work work;
2682 };
2683
btrfs_writepage_fixup_worker(struct btrfs_work * work)2684 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2685 {
2686 struct btrfs_writepage_fixup *fixup =
2687 container_of(work, struct btrfs_writepage_fixup, work);
2688 struct btrfs_ordered_extent *ordered;
2689 struct extent_state *cached_state = NULL;
2690 struct extent_changeset *data_reserved = NULL;
2691 struct page *page = fixup->page;
2692 struct btrfs_inode *inode = fixup->inode;
2693 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2694 u64 page_start = page_offset(page);
2695 u64 page_end = page_offset(page) + PAGE_SIZE - 1;
2696 int ret = 0;
2697 bool free_delalloc_space = true;
2698
2699 /*
2700 * This is similar to page_mkwrite, we need to reserve the space before
2701 * we take the page lock.
2702 */
2703 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2704 PAGE_SIZE);
2705 again:
2706 lock_page(page);
2707
2708 /*
2709 * Before we queued this fixup, we took a reference on the page.
2710 * page->mapping may go NULL, but it shouldn't be moved to a different
2711 * address space.
2712 */
2713 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2714 /*
2715 * Unfortunately this is a little tricky, either
2716 *
2717 * 1) We got here and our page had already been dealt with and
2718 * we reserved our space, thus ret == 0, so we need to just
2719 * drop our space reservation and bail. This can happen the
2720 * first time we come into the fixup worker, or could happen
2721 * while waiting for the ordered extent.
2722 * 2) Our page was already dealt with, but we happened to get an
2723 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2724 * this case we obviously don't have anything to release, but
2725 * because the page was already dealt with we don't want to
2726 * mark the page with an error, so make sure we're resetting
2727 * ret to 0. This is why we have this check _before_ the ret
2728 * check, because we do not want to have a surprise ENOSPC
2729 * when the page was already properly dealt with.
2730 */
2731 if (!ret) {
2732 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2733 btrfs_delalloc_release_space(inode, data_reserved,
2734 page_start, PAGE_SIZE,
2735 true);
2736 }
2737 ret = 0;
2738 goto out_page;
2739 }
2740
2741 /*
2742 * We can't mess with the page state unless it is locked, so now that
2743 * it is locked bail if we failed to make our space reservation.
2744 */
2745 if (ret)
2746 goto out_page;
2747
2748 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2749
2750 /* already ordered? We're done */
2751 if (PageOrdered(page))
2752 goto out_reserved;
2753
2754 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2755 if (ordered) {
2756 unlock_extent(&inode->io_tree, page_start, page_end,
2757 &cached_state);
2758 unlock_page(page);
2759 btrfs_start_ordered_extent(ordered);
2760 btrfs_put_ordered_extent(ordered);
2761 goto again;
2762 }
2763
2764 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2765 &cached_state);
2766 if (ret)
2767 goto out_reserved;
2768
2769 /*
2770 * Everything went as planned, we're now the owner of a dirty page with
2771 * delayed allocation bits set and space reserved for our COW
2772 * destination.
2773 *
2774 * The page was dirty when we started, nothing should have cleaned it.
2775 */
2776 BUG_ON(!PageDirty(page));
2777 free_delalloc_space = false;
2778 out_reserved:
2779 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2780 if (free_delalloc_space)
2781 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2782 PAGE_SIZE, true);
2783 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2784 out_page:
2785 if (ret) {
2786 /*
2787 * We hit ENOSPC or other errors. Update the mapping and page
2788 * to reflect the errors and clean the page.
2789 */
2790 mapping_set_error(page->mapping, ret);
2791 btrfs_mark_ordered_io_finished(inode, page, page_start,
2792 PAGE_SIZE, !ret);
2793 clear_page_dirty_for_io(page);
2794 }
2795 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
2796 unlock_page(page);
2797 put_page(page);
2798 kfree(fixup);
2799 extent_changeset_free(data_reserved);
2800 /*
2801 * As a precaution, do a delayed iput in case it would be the last iput
2802 * that could need flushing space. Recursing back to fixup worker would
2803 * deadlock.
2804 */
2805 btrfs_add_delayed_iput(inode);
2806 }
2807
2808 /*
2809 * There are a few paths in the higher layers of the kernel that directly
2810 * set the page dirty bit without asking the filesystem if it is a
2811 * good idea. This causes problems because we want to make sure COW
2812 * properly happens and the data=ordered rules are followed.
2813 *
2814 * In our case any range that doesn't have the ORDERED bit set
2815 * hasn't been properly setup for IO. We kick off an async process
2816 * to fix it up. The async helper will wait for ordered extents, set
2817 * the delalloc bit and make it safe to write the page.
2818 */
btrfs_writepage_cow_fixup(struct page * page)2819 int btrfs_writepage_cow_fixup(struct page *page)
2820 {
2821 struct inode *inode = page->mapping->host;
2822 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2823 struct btrfs_writepage_fixup *fixup;
2824
2825 /* This page has ordered extent covering it already */
2826 if (PageOrdered(page))
2827 return 0;
2828
2829 /*
2830 * PageChecked is set below when we create a fixup worker for this page,
2831 * don't try to create another one if we're already PageChecked()
2832 *
2833 * The extent_io writepage code will redirty the page if we send back
2834 * EAGAIN.
2835 */
2836 if (PageChecked(page))
2837 return -EAGAIN;
2838
2839 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2840 if (!fixup)
2841 return -EAGAIN;
2842
2843 /*
2844 * We are already holding a reference to this inode from
2845 * write_cache_pages. We need to hold it because the space reservation
2846 * takes place outside of the page lock, and we can't trust
2847 * page->mapping outside of the page lock.
2848 */
2849 ihold(inode);
2850 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
2851 get_page(page);
2852 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2853 fixup->page = page;
2854 fixup->inode = BTRFS_I(inode);
2855 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2856
2857 return -EAGAIN;
2858 }
2859
insert_reserved_file_extent(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,u64 file_pos,struct btrfs_file_extent_item * stack_fi,const bool update_inode_bytes,u64 qgroup_reserved)2860 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2861 struct btrfs_inode *inode, u64 file_pos,
2862 struct btrfs_file_extent_item *stack_fi,
2863 const bool update_inode_bytes,
2864 u64 qgroup_reserved)
2865 {
2866 struct btrfs_root *root = inode->root;
2867 const u64 sectorsize = root->fs_info->sectorsize;
2868 struct btrfs_path *path;
2869 struct extent_buffer *leaf;
2870 struct btrfs_key ins;
2871 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2872 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2873 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2874 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2875 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2876 struct btrfs_drop_extents_args drop_args = { 0 };
2877 int ret;
2878
2879 path = btrfs_alloc_path();
2880 if (!path)
2881 return -ENOMEM;
2882
2883 /*
2884 * we may be replacing one extent in the tree with another.
2885 * The new extent is pinned in the extent map, and we don't want
2886 * to drop it from the cache until it is completely in the btree.
2887 *
2888 * So, tell btrfs_drop_extents to leave this extent in the cache.
2889 * the caller is expected to unpin it and allow it to be merged
2890 * with the others.
2891 */
2892 drop_args.path = path;
2893 drop_args.start = file_pos;
2894 drop_args.end = file_pos + num_bytes;
2895 drop_args.replace_extent = true;
2896 drop_args.extent_item_size = sizeof(*stack_fi);
2897 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2898 if (ret)
2899 goto out;
2900
2901 if (!drop_args.extent_inserted) {
2902 ins.objectid = btrfs_ino(inode);
2903 ins.offset = file_pos;
2904 ins.type = BTRFS_EXTENT_DATA_KEY;
2905
2906 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2907 sizeof(*stack_fi));
2908 if (ret)
2909 goto out;
2910 }
2911 leaf = path->nodes[0];
2912 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2913 write_extent_buffer(leaf, stack_fi,
2914 btrfs_item_ptr_offset(leaf, path->slots[0]),
2915 sizeof(struct btrfs_file_extent_item));
2916
2917 btrfs_mark_buffer_dirty(trans, leaf);
2918 btrfs_release_path(path);
2919
2920 /*
2921 * If we dropped an inline extent here, we know the range where it is
2922 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2923 * number of bytes only for that range containing the inline extent.
2924 * The remaining of the range will be processed when clearning the
2925 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2926 */
2927 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2928 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2929
2930 inline_size = drop_args.bytes_found - inline_size;
2931 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2932 drop_args.bytes_found -= inline_size;
2933 num_bytes -= sectorsize;
2934 }
2935
2936 if (update_inode_bytes)
2937 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2938
2939 ins.objectid = disk_bytenr;
2940 ins.offset = disk_num_bytes;
2941 ins.type = BTRFS_EXTENT_ITEM_KEY;
2942
2943 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2944 if (ret)
2945 goto out;
2946
2947 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2948 file_pos - offset,
2949 qgroup_reserved, &ins);
2950 out:
2951 btrfs_free_path(path);
2952
2953 return ret;
2954 }
2955
btrfs_release_delalloc_bytes(struct btrfs_fs_info * fs_info,u64 start,u64 len)2956 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2957 u64 start, u64 len)
2958 {
2959 struct btrfs_block_group *cache;
2960
2961 cache = btrfs_lookup_block_group(fs_info, start);
2962 ASSERT(cache);
2963
2964 spin_lock(&cache->lock);
2965 cache->delalloc_bytes -= len;
2966 spin_unlock(&cache->lock);
2967
2968 btrfs_put_block_group(cache);
2969 }
2970
insert_ordered_extent_file_extent(struct btrfs_trans_handle * trans,struct btrfs_ordered_extent * oe)2971 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2972 struct btrfs_ordered_extent *oe)
2973 {
2974 struct btrfs_file_extent_item stack_fi;
2975 bool update_inode_bytes;
2976 u64 num_bytes = oe->num_bytes;
2977 u64 ram_bytes = oe->ram_bytes;
2978
2979 memset(&stack_fi, 0, sizeof(stack_fi));
2980 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2981 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2982 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2983 oe->disk_num_bytes);
2984 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
2985 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
2986 num_bytes = oe->truncated_len;
2987 ram_bytes = num_bytes;
2988 }
2989 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
2990 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
2991 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2992 /* Encryption and other encoding is reserved and all 0 */
2993
2994 /*
2995 * For delalloc, when completing an ordered extent we update the inode's
2996 * bytes when clearing the range in the inode's io tree, so pass false
2997 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2998 * except if the ordered extent was truncated.
2999 */
3000 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3001 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3002 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3003
3004 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3005 oe->file_offset, &stack_fi,
3006 update_inode_bytes, oe->qgroup_rsv);
3007 }
3008
3009 /*
3010 * As ordered data IO finishes, this gets called so we can finish
3011 * an ordered extent if the range of bytes in the file it covers are
3012 * fully written.
3013 */
btrfs_finish_one_ordered(struct btrfs_ordered_extent * ordered_extent)3014 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3015 {
3016 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3017 struct btrfs_root *root = inode->root;
3018 struct btrfs_fs_info *fs_info = root->fs_info;
3019 struct btrfs_trans_handle *trans = NULL;
3020 struct extent_io_tree *io_tree = &inode->io_tree;
3021 struct extent_state *cached_state = NULL;
3022 u64 start, end;
3023 int compress_type = 0;
3024 int ret = 0;
3025 u64 logical_len = ordered_extent->num_bytes;
3026 bool freespace_inode;
3027 bool truncated = false;
3028 bool clear_reserved_extent = true;
3029 unsigned int clear_bits = EXTENT_DEFRAG;
3030
3031 start = ordered_extent->file_offset;
3032 end = start + ordered_extent->num_bytes - 1;
3033
3034 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3035 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3036 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3037 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3038 clear_bits |= EXTENT_DELALLOC_NEW;
3039
3040 freespace_inode = btrfs_is_free_space_inode(inode);
3041 if (!freespace_inode)
3042 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3043
3044 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3045 ret = -EIO;
3046 goto out;
3047 }
3048
3049 if (btrfs_is_zoned(fs_info))
3050 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3051 ordered_extent->disk_num_bytes);
3052
3053 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3054 truncated = true;
3055 logical_len = ordered_extent->truncated_len;
3056 /* Truncated the entire extent, don't bother adding */
3057 if (!logical_len)
3058 goto out;
3059 }
3060
3061 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3062 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3063
3064 btrfs_inode_safe_disk_i_size_write(inode, 0);
3065 if (freespace_inode)
3066 trans = btrfs_join_transaction_spacecache(root);
3067 else
3068 trans = btrfs_join_transaction(root);
3069 if (IS_ERR(trans)) {
3070 ret = PTR_ERR(trans);
3071 trans = NULL;
3072 goto out;
3073 }
3074 trans->block_rsv = &inode->block_rsv;
3075 ret = btrfs_update_inode_fallback(trans, root, inode);
3076 if (ret) /* -ENOMEM or corruption */
3077 btrfs_abort_transaction(trans, ret);
3078 goto out;
3079 }
3080
3081 clear_bits |= EXTENT_LOCKED;
3082 lock_extent(io_tree, start, end, &cached_state);
3083
3084 if (freespace_inode)
3085 trans = btrfs_join_transaction_spacecache(root);
3086 else
3087 trans = btrfs_join_transaction(root);
3088 if (IS_ERR(trans)) {
3089 ret = PTR_ERR(trans);
3090 trans = NULL;
3091 goto out;
3092 }
3093
3094 trans->block_rsv = &inode->block_rsv;
3095
3096 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3097 compress_type = ordered_extent->compress_type;
3098 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3099 BUG_ON(compress_type);
3100 ret = btrfs_mark_extent_written(trans, inode,
3101 ordered_extent->file_offset,
3102 ordered_extent->file_offset +
3103 logical_len);
3104 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3105 ordered_extent->disk_num_bytes);
3106 } else {
3107 BUG_ON(root == fs_info->tree_root);
3108 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3109 if (!ret) {
3110 clear_reserved_extent = false;
3111 btrfs_release_delalloc_bytes(fs_info,
3112 ordered_extent->disk_bytenr,
3113 ordered_extent->disk_num_bytes);
3114 }
3115 }
3116 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3117 ordered_extent->num_bytes, trans->transid);
3118 if (ret < 0) {
3119 btrfs_abort_transaction(trans, ret);
3120 goto out;
3121 }
3122
3123 ret = add_pending_csums(trans, &ordered_extent->list);
3124 if (ret) {
3125 btrfs_abort_transaction(trans, ret);
3126 goto out;
3127 }
3128
3129 /*
3130 * If this is a new delalloc range, clear its new delalloc flag to
3131 * update the inode's number of bytes. This needs to be done first
3132 * before updating the inode item.
3133 */
3134 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3135 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3136 clear_extent_bit(&inode->io_tree, start, end,
3137 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3138 &cached_state);
3139
3140 btrfs_inode_safe_disk_i_size_write(inode, 0);
3141 ret = btrfs_update_inode_fallback(trans, root, inode);
3142 if (ret) { /* -ENOMEM or corruption */
3143 btrfs_abort_transaction(trans, ret);
3144 goto out;
3145 }
3146 ret = 0;
3147 out:
3148 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3149 &cached_state);
3150
3151 if (trans)
3152 btrfs_end_transaction(trans);
3153
3154 if (ret || truncated) {
3155 u64 unwritten_start = start;
3156
3157 /*
3158 * If we failed to finish this ordered extent for any reason we
3159 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3160 * extent, and mark the inode with the error if it wasn't
3161 * already set. Any error during writeback would have already
3162 * set the mapping error, so we need to set it if we're the ones
3163 * marking this ordered extent as failed.
3164 */
3165 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3166 &ordered_extent->flags))
3167 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3168
3169 if (truncated)
3170 unwritten_start += logical_len;
3171 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3172
3173 /*
3174 * Drop extent maps for the part of the extent we didn't write.
3175 *
3176 * We have an exception here for the free_space_inode, this is
3177 * because when we do btrfs_get_extent() on the free space inode
3178 * we will search the commit root. If this is a new block group
3179 * we won't find anything, and we will trip over the assert in
3180 * writepage where we do ASSERT(em->block_start !=
3181 * EXTENT_MAP_HOLE).
3182 *
3183 * Theoretically we could also skip this for any NOCOW extent as
3184 * we don't mess with the extent map tree in the NOCOW case, but
3185 * for now simply skip this if we are the free space inode.
3186 */
3187 if (!btrfs_is_free_space_inode(inode))
3188 btrfs_drop_extent_map_range(inode, unwritten_start,
3189 end, false);
3190
3191 /*
3192 * If the ordered extent had an IOERR or something else went
3193 * wrong we need to return the space for this ordered extent
3194 * back to the allocator. We only free the extent in the
3195 * truncated case if we didn't write out the extent at all.
3196 *
3197 * If we made it past insert_reserved_file_extent before we
3198 * errored out then we don't need to do this as the accounting
3199 * has already been done.
3200 */
3201 if ((ret || !logical_len) &&
3202 clear_reserved_extent &&
3203 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3204 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3205 /*
3206 * Discard the range before returning it back to the
3207 * free space pool
3208 */
3209 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3210 btrfs_discard_extent(fs_info,
3211 ordered_extent->disk_bytenr,
3212 ordered_extent->disk_num_bytes,
3213 NULL);
3214 btrfs_free_reserved_extent(fs_info,
3215 ordered_extent->disk_bytenr,
3216 ordered_extent->disk_num_bytes, 1);
3217 /*
3218 * Actually free the qgroup rsv which was released when
3219 * the ordered extent was created.
3220 */
3221 btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid,
3222 ordered_extent->qgroup_rsv,
3223 BTRFS_QGROUP_RSV_DATA);
3224 }
3225 }
3226
3227 /*
3228 * This needs to be done to make sure anybody waiting knows we are done
3229 * updating everything for this ordered extent.
3230 */
3231 btrfs_remove_ordered_extent(inode, ordered_extent);
3232
3233 /* once for us */
3234 btrfs_put_ordered_extent(ordered_extent);
3235 /* once for the tree */
3236 btrfs_put_ordered_extent(ordered_extent);
3237
3238 return ret;
3239 }
3240
btrfs_finish_ordered_io(struct btrfs_ordered_extent * ordered)3241 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3242 {
3243 if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) &&
3244 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
3245 btrfs_finish_ordered_zoned(ordered);
3246 return btrfs_finish_one_ordered(ordered);
3247 }
3248
3249 /*
3250 * Verify the checksum for a single sector without any extra action that depend
3251 * on the type of I/O.
3252 */
btrfs_check_sector_csum(struct btrfs_fs_info * fs_info,struct page * page,u32 pgoff,u8 * csum,const u8 * const csum_expected)3253 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3254 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3255 {
3256 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3257 char *kaddr;
3258
3259 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3260
3261 shash->tfm = fs_info->csum_shash;
3262
3263 kaddr = kmap_local_page(page) + pgoff;
3264 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3265 kunmap_local(kaddr);
3266
3267 if (memcmp(csum, csum_expected, fs_info->csum_size))
3268 return -EIO;
3269 return 0;
3270 }
3271
3272 /*
3273 * Verify the checksum of a single data sector.
3274 *
3275 * @bbio: btrfs_io_bio which contains the csum
3276 * @dev: device the sector is on
3277 * @bio_offset: offset to the beginning of the bio (in bytes)
3278 * @bv: bio_vec to check
3279 *
3280 * Check if the checksum on a data block is valid. When a checksum mismatch is
3281 * detected, report the error and fill the corrupted range with zero.
3282 *
3283 * Return %true if the sector is ok or had no checksum to start with, else %false.
3284 */
btrfs_data_csum_ok(struct btrfs_bio * bbio,struct btrfs_device * dev,u32 bio_offset,struct bio_vec * bv)3285 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3286 u32 bio_offset, struct bio_vec *bv)
3287 {
3288 struct btrfs_inode *inode = bbio->inode;
3289 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3290 u64 file_offset = bbio->file_offset + bio_offset;
3291 u64 end = file_offset + bv->bv_len - 1;
3292 u8 *csum_expected;
3293 u8 csum[BTRFS_CSUM_SIZE];
3294
3295 ASSERT(bv->bv_len == fs_info->sectorsize);
3296
3297 if (!bbio->csum)
3298 return true;
3299
3300 if (btrfs_is_data_reloc_root(inode->root) &&
3301 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3302 1, NULL)) {
3303 /* Skip the range without csum for data reloc inode */
3304 clear_extent_bits(&inode->io_tree, file_offset, end,
3305 EXTENT_NODATASUM);
3306 return true;
3307 }
3308
3309 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3310 fs_info->csum_size;
3311 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3312 csum_expected))
3313 goto zeroit;
3314 return true;
3315
3316 zeroit:
3317 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3318 bbio->mirror_num);
3319 if (dev)
3320 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3321 memzero_bvec(bv);
3322 return false;
3323 }
3324
3325 /*
3326 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3327 *
3328 * @inode: The inode we want to perform iput on
3329 *
3330 * This function uses the generic vfs_inode::i_count to track whether we should
3331 * just decrement it (in case it's > 1) or if this is the last iput then link
3332 * the inode to the delayed iput machinery. Delayed iputs are processed at
3333 * transaction commit time/superblock commit/cleaner kthread.
3334 */
btrfs_add_delayed_iput(struct btrfs_inode * inode)3335 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3336 {
3337 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3338 unsigned long flags;
3339
3340 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3341 return;
3342
3343 atomic_inc(&fs_info->nr_delayed_iputs);
3344 /*
3345 * Need to be irq safe here because we can be called from either an irq
3346 * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3347 * context.
3348 */
3349 spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3350 ASSERT(list_empty(&inode->delayed_iput));
3351 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3352 spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3353 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3354 wake_up_process(fs_info->cleaner_kthread);
3355 }
3356
run_delayed_iput_locked(struct btrfs_fs_info * fs_info,struct btrfs_inode * inode)3357 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3358 struct btrfs_inode *inode)
3359 {
3360 list_del_init(&inode->delayed_iput);
3361 spin_unlock_irq(&fs_info->delayed_iput_lock);
3362 iput(&inode->vfs_inode);
3363 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3364 wake_up(&fs_info->delayed_iputs_wait);
3365 spin_lock_irq(&fs_info->delayed_iput_lock);
3366 }
3367
btrfs_run_delayed_iput(struct btrfs_fs_info * fs_info,struct btrfs_inode * inode)3368 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3369 struct btrfs_inode *inode)
3370 {
3371 if (!list_empty(&inode->delayed_iput)) {
3372 spin_lock_irq(&fs_info->delayed_iput_lock);
3373 if (!list_empty(&inode->delayed_iput))
3374 run_delayed_iput_locked(fs_info, inode);
3375 spin_unlock_irq(&fs_info->delayed_iput_lock);
3376 }
3377 }
3378
btrfs_run_delayed_iputs(struct btrfs_fs_info * fs_info)3379 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3380 {
3381 /*
3382 * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3383 * calls btrfs_add_delayed_iput() and that needs to lock
3384 * fs_info->delayed_iput_lock. So we need to disable irqs here to
3385 * prevent a deadlock.
3386 */
3387 spin_lock_irq(&fs_info->delayed_iput_lock);
3388 while (!list_empty(&fs_info->delayed_iputs)) {
3389 struct btrfs_inode *inode;
3390
3391 inode = list_first_entry(&fs_info->delayed_iputs,
3392 struct btrfs_inode, delayed_iput);
3393 run_delayed_iput_locked(fs_info, inode);
3394 if (need_resched()) {
3395 spin_unlock_irq(&fs_info->delayed_iput_lock);
3396 cond_resched();
3397 spin_lock_irq(&fs_info->delayed_iput_lock);
3398 }
3399 }
3400 spin_unlock_irq(&fs_info->delayed_iput_lock);
3401 }
3402
3403 /*
3404 * Wait for flushing all delayed iputs
3405 *
3406 * @fs_info: the filesystem
3407 *
3408 * This will wait on any delayed iputs that are currently running with KILLABLE
3409 * set. Once they are all done running we will return, unless we are killed in
3410 * which case we return EINTR. This helps in user operations like fallocate etc
3411 * that might get blocked on the iputs.
3412 *
3413 * Return EINTR if we were killed, 0 if nothing's pending
3414 */
btrfs_wait_on_delayed_iputs(struct btrfs_fs_info * fs_info)3415 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3416 {
3417 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3418 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3419 if (ret)
3420 return -EINTR;
3421 return 0;
3422 }
3423
3424 /*
3425 * This creates an orphan entry for the given inode in case something goes wrong
3426 * in the middle of an unlink.
3427 */
btrfs_orphan_add(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)3428 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3429 struct btrfs_inode *inode)
3430 {
3431 int ret;
3432
3433 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3434 if (ret && ret != -EEXIST) {
3435 btrfs_abort_transaction(trans, ret);
3436 return ret;
3437 }
3438
3439 return 0;
3440 }
3441
3442 /*
3443 * We have done the delete so we can go ahead and remove the orphan item for
3444 * this particular inode.
3445 */
btrfs_orphan_del(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)3446 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3447 struct btrfs_inode *inode)
3448 {
3449 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3450 }
3451
3452 /*
3453 * this cleans up any orphans that may be left on the list from the last use
3454 * of this root.
3455 */
btrfs_orphan_cleanup(struct btrfs_root * root)3456 int btrfs_orphan_cleanup(struct btrfs_root *root)
3457 {
3458 struct btrfs_fs_info *fs_info = root->fs_info;
3459 struct btrfs_path *path;
3460 struct extent_buffer *leaf;
3461 struct btrfs_key key, found_key;
3462 struct btrfs_trans_handle *trans;
3463 struct inode *inode;
3464 u64 last_objectid = 0;
3465 int ret = 0, nr_unlink = 0;
3466
3467 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3468 return 0;
3469
3470 path = btrfs_alloc_path();
3471 if (!path) {
3472 ret = -ENOMEM;
3473 goto out;
3474 }
3475 path->reada = READA_BACK;
3476
3477 key.objectid = BTRFS_ORPHAN_OBJECTID;
3478 key.type = BTRFS_ORPHAN_ITEM_KEY;
3479 key.offset = (u64)-1;
3480
3481 while (1) {
3482 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3483 if (ret < 0)
3484 goto out;
3485
3486 /*
3487 * if ret == 0 means we found what we were searching for, which
3488 * is weird, but possible, so only screw with path if we didn't
3489 * find the key and see if we have stuff that matches
3490 */
3491 if (ret > 0) {
3492 ret = 0;
3493 if (path->slots[0] == 0)
3494 break;
3495 path->slots[0]--;
3496 }
3497
3498 /* pull out the item */
3499 leaf = path->nodes[0];
3500 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3501
3502 /* make sure the item matches what we want */
3503 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3504 break;
3505 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3506 break;
3507
3508 /* release the path since we're done with it */
3509 btrfs_release_path(path);
3510
3511 /*
3512 * this is where we are basically btrfs_lookup, without the
3513 * crossing root thing. we store the inode number in the
3514 * offset of the orphan item.
3515 */
3516
3517 if (found_key.offset == last_objectid) {
3518 /*
3519 * We found the same inode as before. This means we were
3520 * not able to remove its items via eviction triggered
3521 * by an iput(). A transaction abort may have happened,
3522 * due to -ENOSPC for example, so try to grab the error
3523 * that lead to a transaction abort, if any.
3524 */
3525 btrfs_err(fs_info,
3526 "Error removing orphan entry, stopping orphan cleanup");
3527 ret = BTRFS_FS_ERROR(fs_info) ?: -EINVAL;
3528 goto out;
3529 }
3530
3531 last_objectid = found_key.offset;
3532
3533 found_key.objectid = found_key.offset;
3534 found_key.type = BTRFS_INODE_ITEM_KEY;
3535 found_key.offset = 0;
3536 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3537 if (IS_ERR(inode)) {
3538 ret = PTR_ERR(inode);
3539 inode = NULL;
3540 if (ret != -ENOENT)
3541 goto out;
3542 }
3543
3544 if (!inode && root == fs_info->tree_root) {
3545 struct btrfs_root *dead_root;
3546 int is_dead_root = 0;
3547
3548 /*
3549 * This is an orphan in the tree root. Currently these
3550 * could come from 2 sources:
3551 * a) a root (snapshot/subvolume) deletion in progress
3552 * b) a free space cache inode
3553 * We need to distinguish those two, as the orphan item
3554 * for a root must not get deleted before the deletion
3555 * of the snapshot/subvolume's tree completes.
3556 *
3557 * btrfs_find_orphan_roots() ran before us, which has
3558 * found all deleted roots and loaded them into
3559 * fs_info->fs_roots_radix. So here we can find if an
3560 * orphan item corresponds to a deleted root by looking
3561 * up the root from that radix tree.
3562 */
3563
3564 spin_lock(&fs_info->fs_roots_radix_lock);
3565 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3566 (unsigned long)found_key.objectid);
3567 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3568 is_dead_root = 1;
3569 spin_unlock(&fs_info->fs_roots_radix_lock);
3570
3571 if (is_dead_root) {
3572 /* prevent this orphan from being found again */
3573 key.offset = found_key.objectid - 1;
3574 continue;
3575 }
3576
3577 }
3578
3579 /*
3580 * If we have an inode with links, there are a couple of
3581 * possibilities:
3582 *
3583 * 1. We were halfway through creating fsverity metadata for the
3584 * file. In that case, the orphan item represents incomplete
3585 * fsverity metadata which must be cleaned up with
3586 * btrfs_drop_verity_items and deleting the orphan item.
3587
3588 * 2. Old kernels (before v3.12) used to create an
3589 * orphan item for truncate indicating that there were possibly
3590 * extent items past i_size that needed to be deleted. In v3.12,
3591 * truncate was changed to update i_size in sync with the extent
3592 * items, but the (useless) orphan item was still created. Since
3593 * v4.18, we don't create the orphan item for truncate at all.
3594 *
3595 * So, this item could mean that we need to do a truncate, but
3596 * only if this filesystem was last used on a pre-v3.12 kernel
3597 * and was not cleanly unmounted. The odds of that are quite
3598 * slim, and it's a pain to do the truncate now, so just delete
3599 * the orphan item.
3600 *
3601 * It's also possible that this orphan item was supposed to be
3602 * deleted but wasn't. The inode number may have been reused,
3603 * but either way, we can delete the orphan item.
3604 */
3605 if (!inode || inode->i_nlink) {
3606 if (inode) {
3607 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3608 iput(inode);
3609 inode = NULL;
3610 if (ret)
3611 goto out;
3612 }
3613 trans = btrfs_start_transaction(root, 1);
3614 if (IS_ERR(trans)) {
3615 ret = PTR_ERR(trans);
3616 goto out;
3617 }
3618 btrfs_debug(fs_info, "auto deleting %Lu",
3619 found_key.objectid);
3620 ret = btrfs_del_orphan_item(trans, root,
3621 found_key.objectid);
3622 btrfs_end_transaction(trans);
3623 if (ret)
3624 goto out;
3625 continue;
3626 }
3627
3628 nr_unlink++;
3629
3630 /* this will do delete_inode and everything for us */
3631 iput(inode);
3632 }
3633 /* release the path since we're done with it */
3634 btrfs_release_path(path);
3635
3636 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3637 trans = btrfs_join_transaction(root);
3638 if (!IS_ERR(trans))
3639 btrfs_end_transaction(trans);
3640 }
3641
3642 if (nr_unlink)
3643 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3644
3645 out:
3646 if (ret)
3647 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3648 btrfs_free_path(path);
3649 return ret;
3650 }
3651
3652 /*
3653 * very simple check to peek ahead in the leaf looking for xattrs. If we
3654 * don't find any xattrs, we know there can't be any acls.
3655 *
3656 * slot is the slot the inode is in, objectid is the objectid of the inode
3657 */
acls_after_inode_item(struct extent_buffer * leaf,int slot,u64 objectid,int * first_xattr_slot)3658 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3659 int slot, u64 objectid,
3660 int *first_xattr_slot)
3661 {
3662 u32 nritems = btrfs_header_nritems(leaf);
3663 struct btrfs_key found_key;
3664 static u64 xattr_access = 0;
3665 static u64 xattr_default = 0;
3666 int scanned = 0;
3667
3668 if (!xattr_access) {
3669 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3670 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3671 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3672 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3673 }
3674
3675 slot++;
3676 *first_xattr_slot = -1;
3677 while (slot < nritems) {
3678 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3679
3680 /* we found a different objectid, there must not be acls */
3681 if (found_key.objectid != objectid)
3682 return 0;
3683
3684 /* we found an xattr, assume we've got an acl */
3685 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3686 if (*first_xattr_slot == -1)
3687 *first_xattr_slot = slot;
3688 if (found_key.offset == xattr_access ||
3689 found_key.offset == xattr_default)
3690 return 1;
3691 }
3692
3693 /*
3694 * we found a key greater than an xattr key, there can't
3695 * be any acls later on
3696 */
3697 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3698 return 0;
3699
3700 slot++;
3701 scanned++;
3702
3703 /*
3704 * it goes inode, inode backrefs, xattrs, extents,
3705 * so if there are a ton of hard links to an inode there can
3706 * be a lot of backrefs. Don't waste time searching too hard,
3707 * this is just an optimization
3708 */
3709 if (scanned >= 8)
3710 break;
3711 }
3712 /* we hit the end of the leaf before we found an xattr or
3713 * something larger than an xattr. We have to assume the inode
3714 * has acls
3715 */
3716 if (*first_xattr_slot == -1)
3717 *first_xattr_slot = slot;
3718 return 1;
3719 }
3720
3721 /*
3722 * read an inode from the btree into the in-memory inode
3723 */
btrfs_read_locked_inode(struct inode * inode,struct btrfs_path * in_path)3724 static int btrfs_read_locked_inode(struct inode *inode,
3725 struct btrfs_path *in_path)
3726 {
3727 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3728 struct btrfs_path *path = in_path;
3729 struct extent_buffer *leaf;
3730 struct btrfs_inode_item *inode_item;
3731 struct btrfs_root *root = BTRFS_I(inode)->root;
3732 struct btrfs_key location;
3733 unsigned long ptr;
3734 int maybe_acls;
3735 u32 rdev;
3736 int ret;
3737 bool filled = false;
3738 int first_xattr_slot;
3739
3740 ret = btrfs_fill_inode(inode, &rdev);
3741 if (!ret)
3742 filled = true;
3743
3744 if (!path) {
3745 path = btrfs_alloc_path();
3746 if (!path)
3747 return -ENOMEM;
3748 }
3749
3750 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3751
3752 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3753 if (ret) {
3754 if (path != in_path)
3755 btrfs_free_path(path);
3756 return ret;
3757 }
3758
3759 leaf = path->nodes[0];
3760
3761 if (filled)
3762 goto cache_index;
3763
3764 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3765 struct btrfs_inode_item);
3766 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3767 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3768 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3769 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3770 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3771 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3772 round_up(i_size_read(inode), fs_info->sectorsize));
3773
3774 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3775 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3776
3777 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3778 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3779
3780 inode_set_ctime(inode, btrfs_timespec_sec(leaf, &inode_item->ctime),
3781 btrfs_timespec_nsec(leaf, &inode_item->ctime));
3782
3783 BTRFS_I(inode)->i_otime.tv_sec =
3784 btrfs_timespec_sec(leaf, &inode_item->otime);
3785 BTRFS_I(inode)->i_otime.tv_nsec =
3786 btrfs_timespec_nsec(leaf, &inode_item->otime);
3787
3788 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3789 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3790 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3791
3792 inode_set_iversion_queried(inode,
3793 btrfs_inode_sequence(leaf, inode_item));
3794 inode->i_generation = BTRFS_I(inode)->generation;
3795 inode->i_rdev = 0;
3796 rdev = btrfs_inode_rdev(leaf, inode_item);
3797
3798 BTRFS_I(inode)->index_cnt = (u64)-1;
3799 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3800 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3801
3802 cache_index:
3803 /*
3804 * If we were modified in the current generation and evicted from memory
3805 * and then re-read we need to do a full sync since we don't have any
3806 * idea about which extents were modified before we were evicted from
3807 * cache.
3808 *
3809 * This is required for both inode re-read from disk and delayed inode
3810 * in delayed_nodes_tree.
3811 */
3812 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3813 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3814 &BTRFS_I(inode)->runtime_flags);
3815
3816 /*
3817 * We don't persist the id of the transaction where an unlink operation
3818 * against the inode was last made. So here we assume the inode might
3819 * have been evicted, and therefore the exact value of last_unlink_trans
3820 * lost, and set it to last_trans to avoid metadata inconsistencies
3821 * between the inode and its parent if the inode is fsync'ed and the log
3822 * replayed. For example, in the scenario:
3823 *
3824 * touch mydir/foo
3825 * ln mydir/foo mydir/bar
3826 * sync
3827 * unlink mydir/bar
3828 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3829 * xfs_io -c fsync mydir/foo
3830 * <power failure>
3831 * mount fs, triggers fsync log replay
3832 *
3833 * We must make sure that when we fsync our inode foo we also log its
3834 * parent inode, otherwise after log replay the parent still has the
3835 * dentry with the "bar" name but our inode foo has a link count of 1
3836 * and doesn't have an inode ref with the name "bar" anymore.
3837 *
3838 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3839 * but it guarantees correctness at the expense of occasional full
3840 * transaction commits on fsync if our inode is a directory, or if our
3841 * inode is not a directory, logging its parent unnecessarily.
3842 */
3843 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3844
3845 /*
3846 * Same logic as for last_unlink_trans. We don't persist the generation
3847 * of the last transaction where this inode was used for a reflink
3848 * operation, so after eviction and reloading the inode we must be
3849 * pessimistic and assume the last transaction that modified the inode.
3850 */
3851 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3852
3853 path->slots[0]++;
3854 if (inode->i_nlink != 1 ||
3855 path->slots[0] >= btrfs_header_nritems(leaf))
3856 goto cache_acl;
3857
3858 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3859 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3860 goto cache_acl;
3861
3862 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3863 if (location.type == BTRFS_INODE_REF_KEY) {
3864 struct btrfs_inode_ref *ref;
3865
3866 ref = (struct btrfs_inode_ref *)ptr;
3867 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3868 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3869 struct btrfs_inode_extref *extref;
3870
3871 extref = (struct btrfs_inode_extref *)ptr;
3872 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3873 extref);
3874 }
3875 cache_acl:
3876 /*
3877 * try to precache a NULL acl entry for files that don't have
3878 * any xattrs or acls
3879 */
3880 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3881 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3882 if (first_xattr_slot != -1) {
3883 path->slots[0] = first_xattr_slot;
3884 ret = btrfs_load_inode_props(inode, path);
3885 if (ret)
3886 btrfs_err(fs_info,
3887 "error loading props for ino %llu (root %llu): %d",
3888 btrfs_ino(BTRFS_I(inode)),
3889 root->root_key.objectid, ret);
3890 }
3891 if (path != in_path)
3892 btrfs_free_path(path);
3893
3894 if (!maybe_acls)
3895 cache_no_acl(inode);
3896
3897 switch (inode->i_mode & S_IFMT) {
3898 case S_IFREG:
3899 inode->i_mapping->a_ops = &btrfs_aops;
3900 inode->i_fop = &btrfs_file_operations;
3901 inode->i_op = &btrfs_file_inode_operations;
3902 break;
3903 case S_IFDIR:
3904 inode->i_fop = &btrfs_dir_file_operations;
3905 inode->i_op = &btrfs_dir_inode_operations;
3906 break;
3907 case S_IFLNK:
3908 inode->i_op = &btrfs_symlink_inode_operations;
3909 inode_nohighmem(inode);
3910 inode->i_mapping->a_ops = &btrfs_aops;
3911 break;
3912 default:
3913 inode->i_op = &btrfs_special_inode_operations;
3914 init_special_inode(inode, inode->i_mode, rdev);
3915 break;
3916 }
3917
3918 btrfs_sync_inode_flags_to_i_flags(inode);
3919 return 0;
3920 }
3921
3922 /*
3923 * given a leaf and an inode, copy the inode fields into the leaf
3924 */
fill_inode_item(struct btrfs_trans_handle * trans,struct extent_buffer * leaf,struct btrfs_inode_item * item,struct inode * inode)3925 static void fill_inode_item(struct btrfs_trans_handle *trans,
3926 struct extent_buffer *leaf,
3927 struct btrfs_inode_item *item,
3928 struct inode *inode)
3929 {
3930 struct btrfs_map_token token;
3931 u64 flags;
3932
3933 btrfs_init_map_token(&token, leaf);
3934
3935 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3936 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3937 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3938 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3939 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3940
3941 btrfs_set_token_timespec_sec(&token, &item->atime,
3942 inode->i_atime.tv_sec);
3943 btrfs_set_token_timespec_nsec(&token, &item->atime,
3944 inode->i_atime.tv_nsec);
3945
3946 btrfs_set_token_timespec_sec(&token, &item->mtime,
3947 inode->i_mtime.tv_sec);
3948 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3949 inode->i_mtime.tv_nsec);
3950
3951 btrfs_set_token_timespec_sec(&token, &item->ctime,
3952 inode_get_ctime(inode).tv_sec);
3953 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3954 inode_get_ctime(inode).tv_nsec);
3955
3956 btrfs_set_token_timespec_sec(&token, &item->otime,
3957 BTRFS_I(inode)->i_otime.tv_sec);
3958 btrfs_set_token_timespec_nsec(&token, &item->otime,
3959 BTRFS_I(inode)->i_otime.tv_nsec);
3960
3961 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3962 btrfs_set_token_inode_generation(&token, item,
3963 BTRFS_I(inode)->generation);
3964 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3965 btrfs_set_token_inode_transid(&token, item, trans->transid);
3966 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3967 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3968 BTRFS_I(inode)->ro_flags);
3969 btrfs_set_token_inode_flags(&token, item, flags);
3970 btrfs_set_token_inode_block_group(&token, item, 0);
3971 }
3972
3973 /*
3974 * copy everything in the in-memory inode into the btree.
3975 */
btrfs_update_inode_item(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode)3976 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3977 struct btrfs_root *root,
3978 struct btrfs_inode *inode)
3979 {
3980 struct btrfs_inode_item *inode_item;
3981 struct btrfs_path *path;
3982 struct extent_buffer *leaf;
3983 int ret;
3984
3985 path = btrfs_alloc_path();
3986 if (!path)
3987 return -ENOMEM;
3988
3989 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
3990 if (ret) {
3991 if (ret > 0)
3992 ret = -ENOENT;
3993 goto failed;
3994 }
3995
3996 leaf = path->nodes[0];
3997 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3998 struct btrfs_inode_item);
3999
4000 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4001 btrfs_mark_buffer_dirty(trans, leaf);
4002 btrfs_set_inode_last_trans(trans, inode);
4003 ret = 0;
4004 failed:
4005 btrfs_free_path(path);
4006 return ret;
4007 }
4008
4009 /*
4010 * copy everything in the in-memory inode into the btree.
4011 */
btrfs_update_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode)4012 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4013 struct btrfs_root *root,
4014 struct btrfs_inode *inode)
4015 {
4016 struct btrfs_fs_info *fs_info = root->fs_info;
4017 int ret;
4018
4019 /*
4020 * If the inode is a free space inode, we can deadlock during commit
4021 * if we put it into the delayed code.
4022 *
4023 * The data relocation inode should also be directly updated
4024 * without delay
4025 */
4026 if (!btrfs_is_free_space_inode(inode)
4027 && !btrfs_is_data_reloc_root(root)
4028 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4029 btrfs_update_root_times(trans, root);
4030
4031 ret = btrfs_delayed_update_inode(trans, root, inode);
4032 if (!ret)
4033 btrfs_set_inode_last_trans(trans, inode);
4034 return ret;
4035 }
4036
4037 return btrfs_update_inode_item(trans, root, inode);
4038 }
4039
btrfs_update_inode_fallback(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode)4040 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4041 struct btrfs_root *root, struct btrfs_inode *inode)
4042 {
4043 int ret;
4044
4045 ret = btrfs_update_inode(trans, root, inode);
4046 if (ret == -ENOSPC)
4047 return btrfs_update_inode_item(trans, root, inode);
4048 return ret;
4049 }
4050
4051 /*
4052 * unlink helper that gets used here in inode.c and in the tree logging
4053 * recovery code. It remove a link in a directory with a given name, and
4054 * also drops the back refs in the inode to the directory
4055 */
__btrfs_unlink_inode(struct btrfs_trans_handle * trans,struct btrfs_inode * dir,struct btrfs_inode * inode,const struct fscrypt_str * name,struct btrfs_rename_ctx * rename_ctx)4056 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4057 struct btrfs_inode *dir,
4058 struct btrfs_inode *inode,
4059 const struct fscrypt_str *name,
4060 struct btrfs_rename_ctx *rename_ctx)
4061 {
4062 struct btrfs_root *root = dir->root;
4063 struct btrfs_fs_info *fs_info = root->fs_info;
4064 struct btrfs_path *path;
4065 int ret = 0;
4066 struct btrfs_dir_item *di;
4067 u64 index;
4068 u64 ino = btrfs_ino(inode);
4069 u64 dir_ino = btrfs_ino(dir);
4070
4071 path = btrfs_alloc_path();
4072 if (!path) {
4073 ret = -ENOMEM;
4074 goto out;
4075 }
4076
4077 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4078 if (IS_ERR_OR_NULL(di)) {
4079 ret = di ? PTR_ERR(di) : -ENOENT;
4080 goto err;
4081 }
4082 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4083 if (ret)
4084 goto err;
4085 btrfs_release_path(path);
4086
4087 /*
4088 * If we don't have dir index, we have to get it by looking up
4089 * the inode ref, since we get the inode ref, remove it directly,
4090 * it is unnecessary to do delayed deletion.
4091 *
4092 * But if we have dir index, needn't search inode ref to get it.
4093 * Since the inode ref is close to the inode item, it is better
4094 * that we delay to delete it, and just do this deletion when
4095 * we update the inode item.
4096 */
4097 if (inode->dir_index) {
4098 ret = btrfs_delayed_delete_inode_ref(inode);
4099 if (!ret) {
4100 index = inode->dir_index;
4101 goto skip_backref;
4102 }
4103 }
4104
4105 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4106 if (ret) {
4107 btrfs_info(fs_info,
4108 "failed to delete reference to %.*s, inode %llu parent %llu",
4109 name->len, name->name, ino, dir_ino);
4110 btrfs_abort_transaction(trans, ret);
4111 goto err;
4112 }
4113 skip_backref:
4114 if (rename_ctx)
4115 rename_ctx->index = index;
4116
4117 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4118 if (ret) {
4119 btrfs_abort_transaction(trans, ret);
4120 goto err;
4121 }
4122
4123 /*
4124 * If we are in a rename context, we don't need to update anything in the
4125 * log. That will be done later during the rename by btrfs_log_new_name().
4126 * Besides that, doing it here would only cause extra unnecessary btree
4127 * operations on the log tree, increasing latency for applications.
4128 */
4129 if (!rename_ctx) {
4130 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4131 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4132 }
4133
4134 /*
4135 * If we have a pending delayed iput we could end up with the final iput
4136 * being run in btrfs-cleaner context. If we have enough of these built
4137 * up we can end up burning a lot of time in btrfs-cleaner without any
4138 * way to throttle the unlinks. Since we're currently holding a ref on
4139 * the inode we can run the delayed iput here without any issues as the
4140 * final iput won't be done until after we drop the ref we're currently
4141 * holding.
4142 */
4143 btrfs_run_delayed_iput(fs_info, inode);
4144 err:
4145 btrfs_free_path(path);
4146 if (ret)
4147 goto out;
4148
4149 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4150 inode_inc_iversion(&inode->vfs_inode);
4151 inode_set_ctime_current(&inode->vfs_inode);
4152 inode_inc_iversion(&dir->vfs_inode);
4153 inode_set_ctime_current(&inode->vfs_inode);
4154 dir->vfs_inode.i_mtime = inode_set_ctime_current(&dir->vfs_inode);
4155 ret = btrfs_update_inode(trans, root, dir);
4156 out:
4157 return ret;
4158 }
4159
btrfs_unlink_inode(struct btrfs_trans_handle * trans,struct btrfs_inode * dir,struct btrfs_inode * inode,const struct fscrypt_str * name)4160 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4161 struct btrfs_inode *dir, struct btrfs_inode *inode,
4162 const struct fscrypt_str *name)
4163 {
4164 int ret;
4165
4166 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4167 if (!ret) {
4168 drop_nlink(&inode->vfs_inode);
4169 ret = btrfs_update_inode(trans, inode->root, inode);
4170 }
4171 return ret;
4172 }
4173
4174 /*
4175 * helper to start transaction for unlink and rmdir.
4176 *
4177 * unlink and rmdir are special in btrfs, they do not always free space, so
4178 * if we cannot make our reservations the normal way try and see if there is
4179 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4180 * allow the unlink to occur.
4181 */
__unlink_start_trans(struct btrfs_inode * dir)4182 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4183 {
4184 struct btrfs_root *root = dir->root;
4185
4186 return btrfs_start_transaction_fallback_global_rsv(root,
4187 BTRFS_UNLINK_METADATA_UNITS);
4188 }
4189
btrfs_unlink(struct inode * dir,struct dentry * dentry)4190 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4191 {
4192 struct btrfs_trans_handle *trans;
4193 struct inode *inode = d_inode(dentry);
4194 int ret;
4195 struct fscrypt_name fname;
4196
4197 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4198 if (ret)
4199 return ret;
4200
4201 /* This needs to handle no-key deletions later on */
4202
4203 trans = __unlink_start_trans(BTRFS_I(dir));
4204 if (IS_ERR(trans)) {
4205 ret = PTR_ERR(trans);
4206 goto fscrypt_free;
4207 }
4208
4209 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4210 false);
4211
4212 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4213 &fname.disk_name);
4214 if (ret)
4215 goto end_trans;
4216
4217 if (inode->i_nlink == 0) {
4218 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4219 if (ret)
4220 goto end_trans;
4221 }
4222
4223 end_trans:
4224 btrfs_end_transaction(trans);
4225 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4226 fscrypt_free:
4227 fscrypt_free_filename(&fname);
4228 return ret;
4229 }
4230
btrfs_unlink_subvol(struct btrfs_trans_handle * trans,struct btrfs_inode * dir,struct dentry * dentry)4231 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4232 struct btrfs_inode *dir, struct dentry *dentry)
4233 {
4234 struct btrfs_root *root = dir->root;
4235 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4236 struct btrfs_path *path;
4237 struct extent_buffer *leaf;
4238 struct btrfs_dir_item *di;
4239 struct btrfs_key key;
4240 u64 index;
4241 int ret;
4242 u64 objectid;
4243 u64 dir_ino = btrfs_ino(dir);
4244 struct fscrypt_name fname;
4245
4246 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4247 if (ret)
4248 return ret;
4249
4250 /* This needs to handle no-key deletions later on */
4251
4252 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4253 objectid = inode->root->root_key.objectid;
4254 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4255 objectid = inode->location.objectid;
4256 } else {
4257 WARN_ON(1);
4258 fscrypt_free_filename(&fname);
4259 return -EINVAL;
4260 }
4261
4262 path = btrfs_alloc_path();
4263 if (!path) {
4264 ret = -ENOMEM;
4265 goto out;
4266 }
4267
4268 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4269 &fname.disk_name, -1);
4270 if (IS_ERR_OR_NULL(di)) {
4271 ret = di ? PTR_ERR(di) : -ENOENT;
4272 goto out;
4273 }
4274
4275 leaf = path->nodes[0];
4276 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4277 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4278 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4279 if (ret) {
4280 btrfs_abort_transaction(trans, ret);
4281 goto out;
4282 }
4283 btrfs_release_path(path);
4284
4285 /*
4286 * This is a placeholder inode for a subvolume we didn't have a
4287 * reference to at the time of the snapshot creation. In the meantime
4288 * we could have renamed the real subvol link into our snapshot, so
4289 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4290 * Instead simply lookup the dir_index_item for this entry so we can
4291 * remove it. Otherwise we know we have a ref to the root and we can
4292 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4293 */
4294 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4295 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4296 if (IS_ERR(di)) {
4297 ret = PTR_ERR(di);
4298 btrfs_abort_transaction(trans, ret);
4299 goto out;
4300 }
4301
4302 leaf = path->nodes[0];
4303 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4304 index = key.offset;
4305 btrfs_release_path(path);
4306 } else {
4307 ret = btrfs_del_root_ref(trans, objectid,
4308 root->root_key.objectid, dir_ino,
4309 &index, &fname.disk_name);
4310 if (ret) {
4311 btrfs_abort_transaction(trans, ret);
4312 goto out;
4313 }
4314 }
4315
4316 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4317 if (ret) {
4318 btrfs_abort_transaction(trans, ret);
4319 goto out;
4320 }
4321
4322 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4323 inode_inc_iversion(&dir->vfs_inode);
4324 dir->vfs_inode.i_mtime = inode_set_ctime_current(&dir->vfs_inode);
4325 ret = btrfs_update_inode_fallback(trans, root, dir);
4326 if (ret)
4327 btrfs_abort_transaction(trans, ret);
4328 out:
4329 btrfs_free_path(path);
4330 fscrypt_free_filename(&fname);
4331 return ret;
4332 }
4333
4334 /*
4335 * Helper to check if the subvolume references other subvolumes or if it's
4336 * default.
4337 */
may_destroy_subvol(struct btrfs_root * root)4338 static noinline int may_destroy_subvol(struct btrfs_root *root)
4339 {
4340 struct btrfs_fs_info *fs_info = root->fs_info;
4341 struct btrfs_path *path;
4342 struct btrfs_dir_item *di;
4343 struct btrfs_key key;
4344 struct fscrypt_str name = FSTR_INIT("default", 7);
4345 u64 dir_id;
4346 int ret;
4347
4348 path = btrfs_alloc_path();
4349 if (!path)
4350 return -ENOMEM;
4351
4352 /* Make sure this root isn't set as the default subvol */
4353 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4354 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4355 dir_id, &name, 0);
4356 if (di && !IS_ERR(di)) {
4357 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4358 if (key.objectid == root->root_key.objectid) {
4359 ret = -EPERM;
4360 btrfs_err(fs_info,
4361 "deleting default subvolume %llu is not allowed",
4362 key.objectid);
4363 goto out;
4364 }
4365 btrfs_release_path(path);
4366 }
4367
4368 key.objectid = root->root_key.objectid;
4369 key.type = BTRFS_ROOT_REF_KEY;
4370 key.offset = (u64)-1;
4371
4372 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4373 if (ret < 0)
4374 goto out;
4375 if (ret == 0) {
4376 /*
4377 * Key with offset -1 found, there would have to exist a root
4378 * with such id, but this is out of valid range.
4379 */
4380 ret = -EUCLEAN;
4381 goto out;
4382 }
4383
4384 ret = 0;
4385 if (path->slots[0] > 0) {
4386 path->slots[0]--;
4387 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4388 if (key.objectid == root->root_key.objectid &&
4389 key.type == BTRFS_ROOT_REF_KEY)
4390 ret = -ENOTEMPTY;
4391 }
4392 out:
4393 btrfs_free_path(path);
4394 return ret;
4395 }
4396
4397 /* Delete all dentries for inodes belonging to the root */
btrfs_prune_dentries(struct btrfs_root * root)4398 static void btrfs_prune_dentries(struct btrfs_root *root)
4399 {
4400 struct btrfs_fs_info *fs_info = root->fs_info;
4401 struct rb_node *node;
4402 struct rb_node *prev;
4403 struct btrfs_inode *entry;
4404 struct inode *inode;
4405 u64 objectid = 0;
4406
4407 if (!BTRFS_FS_ERROR(fs_info))
4408 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4409
4410 spin_lock(&root->inode_lock);
4411 again:
4412 node = root->inode_tree.rb_node;
4413 prev = NULL;
4414 while (node) {
4415 prev = node;
4416 entry = rb_entry(node, struct btrfs_inode, rb_node);
4417
4418 if (objectid < btrfs_ino(entry))
4419 node = node->rb_left;
4420 else if (objectid > btrfs_ino(entry))
4421 node = node->rb_right;
4422 else
4423 break;
4424 }
4425 if (!node) {
4426 while (prev) {
4427 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4428 if (objectid <= btrfs_ino(entry)) {
4429 node = prev;
4430 break;
4431 }
4432 prev = rb_next(prev);
4433 }
4434 }
4435 while (node) {
4436 entry = rb_entry(node, struct btrfs_inode, rb_node);
4437 objectid = btrfs_ino(entry) + 1;
4438 inode = igrab(&entry->vfs_inode);
4439 if (inode) {
4440 spin_unlock(&root->inode_lock);
4441 if (atomic_read(&inode->i_count) > 1)
4442 d_prune_aliases(inode);
4443 /*
4444 * btrfs_drop_inode will have it removed from the inode
4445 * cache when its usage count hits zero.
4446 */
4447 iput(inode);
4448 cond_resched();
4449 spin_lock(&root->inode_lock);
4450 goto again;
4451 }
4452
4453 if (cond_resched_lock(&root->inode_lock))
4454 goto again;
4455
4456 node = rb_next(node);
4457 }
4458 spin_unlock(&root->inode_lock);
4459 }
4460
btrfs_delete_subvolume(struct btrfs_inode * dir,struct dentry * dentry)4461 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4462 {
4463 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4464 struct btrfs_root *root = dir->root;
4465 struct inode *inode = d_inode(dentry);
4466 struct btrfs_root *dest = BTRFS_I(inode)->root;
4467 struct btrfs_trans_handle *trans;
4468 struct btrfs_block_rsv block_rsv;
4469 u64 root_flags;
4470 u64 qgroup_reserved = 0;
4471 int ret;
4472
4473 down_write(&fs_info->subvol_sem);
4474
4475 /*
4476 * Don't allow to delete a subvolume with send in progress. This is
4477 * inside the inode lock so the error handling that has to drop the bit
4478 * again is not run concurrently.
4479 */
4480 spin_lock(&dest->root_item_lock);
4481 if (dest->send_in_progress) {
4482 spin_unlock(&dest->root_item_lock);
4483 btrfs_warn(fs_info,
4484 "attempt to delete subvolume %llu during send",
4485 dest->root_key.objectid);
4486 ret = -EPERM;
4487 goto out_up_write;
4488 }
4489 if (atomic_read(&dest->nr_swapfiles)) {
4490 spin_unlock(&dest->root_item_lock);
4491 btrfs_warn(fs_info,
4492 "attempt to delete subvolume %llu with active swapfile",
4493 root->root_key.objectid);
4494 ret = -EPERM;
4495 goto out_up_write;
4496 }
4497 root_flags = btrfs_root_flags(&dest->root_item);
4498 btrfs_set_root_flags(&dest->root_item,
4499 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4500 spin_unlock(&dest->root_item_lock);
4501
4502 ret = may_destroy_subvol(dest);
4503 if (ret)
4504 goto out_undead;
4505
4506 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4507 /*
4508 * One for dir inode,
4509 * two for dir entries,
4510 * two for root ref/backref.
4511 */
4512 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4513 if (ret)
4514 goto out_undead;
4515 qgroup_reserved = block_rsv.qgroup_rsv_reserved;
4516
4517 trans = btrfs_start_transaction(root, 0);
4518 if (IS_ERR(trans)) {
4519 ret = PTR_ERR(trans);
4520 goto out_release;
4521 }
4522 ret = btrfs_record_root_in_trans(trans, root);
4523 if (ret) {
4524 btrfs_abort_transaction(trans, ret);
4525 goto out_end_trans;
4526 }
4527 btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
4528 qgroup_reserved = 0;
4529 trans->block_rsv = &block_rsv;
4530 trans->bytes_reserved = block_rsv.size;
4531
4532 btrfs_record_snapshot_destroy(trans, dir);
4533
4534 ret = btrfs_unlink_subvol(trans, dir, dentry);
4535 if (ret) {
4536 btrfs_abort_transaction(trans, ret);
4537 goto out_end_trans;
4538 }
4539
4540 ret = btrfs_record_root_in_trans(trans, dest);
4541 if (ret) {
4542 btrfs_abort_transaction(trans, ret);
4543 goto out_end_trans;
4544 }
4545
4546 memset(&dest->root_item.drop_progress, 0,
4547 sizeof(dest->root_item.drop_progress));
4548 btrfs_set_root_drop_level(&dest->root_item, 0);
4549 btrfs_set_root_refs(&dest->root_item, 0);
4550
4551 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4552 ret = btrfs_insert_orphan_item(trans,
4553 fs_info->tree_root,
4554 dest->root_key.objectid);
4555 if (ret) {
4556 btrfs_abort_transaction(trans, ret);
4557 goto out_end_trans;
4558 }
4559 }
4560
4561 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4562 BTRFS_UUID_KEY_SUBVOL,
4563 dest->root_key.objectid);
4564 if (ret && ret != -ENOENT) {
4565 btrfs_abort_transaction(trans, ret);
4566 goto out_end_trans;
4567 }
4568 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4569 ret = btrfs_uuid_tree_remove(trans,
4570 dest->root_item.received_uuid,
4571 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4572 dest->root_key.objectid);
4573 if (ret && ret != -ENOENT) {
4574 btrfs_abort_transaction(trans, ret);
4575 goto out_end_trans;
4576 }
4577 }
4578
4579 free_anon_bdev(dest->anon_dev);
4580 dest->anon_dev = 0;
4581 out_end_trans:
4582 trans->block_rsv = NULL;
4583 trans->bytes_reserved = 0;
4584 ret = btrfs_end_transaction(trans);
4585 inode->i_flags |= S_DEAD;
4586 out_release:
4587 btrfs_block_rsv_release(fs_info, &block_rsv, (u64)-1, NULL);
4588 if (qgroup_reserved)
4589 btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved);
4590 out_undead:
4591 if (ret) {
4592 spin_lock(&dest->root_item_lock);
4593 root_flags = btrfs_root_flags(&dest->root_item);
4594 btrfs_set_root_flags(&dest->root_item,
4595 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4596 spin_unlock(&dest->root_item_lock);
4597 }
4598 out_up_write:
4599 up_write(&fs_info->subvol_sem);
4600 if (!ret) {
4601 d_invalidate(dentry);
4602 btrfs_prune_dentries(dest);
4603 ASSERT(dest->send_in_progress == 0);
4604 }
4605
4606 return ret;
4607 }
4608
btrfs_rmdir(struct inode * dir,struct dentry * dentry)4609 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4610 {
4611 struct inode *inode = d_inode(dentry);
4612 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4613 int err = 0;
4614 struct btrfs_trans_handle *trans;
4615 u64 last_unlink_trans;
4616 struct fscrypt_name fname;
4617
4618 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4619 return -ENOTEMPTY;
4620 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4621 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4622 btrfs_err(fs_info,
4623 "extent tree v2 doesn't support snapshot deletion yet");
4624 return -EOPNOTSUPP;
4625 }
4626 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4627 }
4628
4629 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4630 if (err)
4631 return err;
4632
4633 /* This needs to handle no-key deletions later on */
4634
4635 trans = __unlink_start_trans(BTRFS_I(dir));
4636 if (IS_ERR(trans)) {
4637 err = PTR_ERR(trans);
4638 goto out_notrans;
4639 }
4640
4641 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4642 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4643 goto out;
4644 }
4645
4646 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4647 if (err)
4648 goto out;
4649
4650 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4651
4652 /* now the directory is empty */
4653 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4654 &fname.disk_name);
4655 if (!err) {
4656 btrfs_i_size_write(BTRFS_I(inode), 0);
4657 /*
4658 * Propagate the last_unlink_trans value of the deleted dir to
4659 * its parent directory. This is to prevent an unrecoverable
4660 * log tree in the case we do something like this:
4661 * 1) create dir foo
4662 * 2) create snapshot under dir foo
4663 * 3) delete the snapshot
4664 * 4) rmdir foo
4665 * 5) mkdir foo
4666 * 6) fsync foo or some file inside foo
4667 */
4668 if (last_unlink_trans >= trans->transid)
4669 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4670 }
4671 out:
4672 btrfs_end_transaction(trans);
4673 out_notrans:
4674 btrfs_btree_balance_dirty(fs_info);
4675 fscrypt_free_filename(&fname);
4676
4677 return err;
4678 }
4679
4680 /*
4681 * btrfs_truncate_block - read, zero a chunk and write a block
4682 * @inode - inode that we're zeroing
4683 * @from - the offset to start zeroing
4684 * @len - the length to zero, 0 to zero the entire range respective to the
4685 * offset
4686 * @front - zero up to the offset instead of from the offset on
4687 *
4688 * This will find the block for the "from" offset and cow the block and zero the
4689 * part we want to zero. This is used with truncate and hole punching.
4690 */
btrfs_truncate_block(struct btrfs_inode * inode,loff_t from,loff_t len,int front)4691 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4692 int front)
4693 {
4694 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4695 struct address_space *mapping = inode->vfs_inode.i_mapping;
4696 struct extent_io_tree *io_tree = &inode->io_tree;
4697 struct btrfs_ordered_extent *ordered;
4698 struct extent_state *cached_state = NULL;
4699 struct extent_changeset *data_reserved = NULL;
4700 bool only_release_metadata = false;
4701 u32 blocksize = fs_info->sectorsize;
4702 pgoff_t index = from >> PAGE_SHIFT;
4703 unsigned offset = from & (blocksize - 1);
4704 struct page *page;
4705 gfp_t mask = btrfs_alloc_write_mask(mapping);
4706 size_t write_bytes = blocksize;
4707 int ret = 0;
4708 u64 block_start;
4709 u64 block_end;
4710
4711 if (IS_ALIGNED(offset, blocksize) &&
4712 (!len || IS_ALIGNED(len, blocksize)))
4713 goto out;
4714
4715 block_start = round_down(from, blocksize);
4716 block_end = block_start + blocksize - 1;
4717
4718 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4719 blocksize, false);
4720 if (ret < 0) {
4721 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4722 /* For nocow case, no need to reserve data space */
4723 only_release_metadata = true;
4724 } else {
4725 goto out;
4726 }
4727 }
4728 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4729 if (ret < 0) {
4730 if (!only_release_metadata)
4731 btrfs_free_reserved_data_space(inode, data_reserved,
4732 block_start, blocksize);
4733 goto out;
4734 }
4735 again:
4736 page = find_or_create_page(mapping, index, mask);
4737 if (!page) {
4738 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4739 blocksize, true);
4740 btrfs_delalloc_release_extents(inode, blocksize);
4741 ret = -ENOMEM;
4742 goto out;
4743 }
4744
4745 if (!PageUptodate(page)) {
4746 ret = btrfs_read_folio(NULL, page_folio(page));
4747 lock_page(page);
4748 if (page->mapping != mapping) {
4749 unlock_page(page);
4750 put_page(page);
4751 goto again;
4752 }
4753 if (!PageUptodate(page)) {
4754 ret = -EIO;
4755 goto out_unlock;
4756 }
4757 }
4758
4759 /*
4760 * We unlock the page after the io is completed and then re-lock it
4761 * above. release_folio() could have come in between that and cleared
4762 * PagePrivate(), but left the page in the mapping. Set the page mapped
4763 * here to make sure it's properly set for the subpage stuff.
4764 */
4765 ret = set_page_extent_mapped(page);
4766 if (ret < 0)
4767 goto out_unlock;
4768
4769 wait_on_page_writeback(page);
4770
4771 lock_extent(io_tree, block_start, block_end, &cached_state);
4772
4773 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4774 if (ordered) {
4775 unlock_extent(io_tree, block_start, block_end, &cached_state);
4776 unlock_page(page);
4777 put_page(page);
4778 btrfs_start_ordered_extent(ordered);
4779 btrfs_put_ordered_extent(ordered);
4780 goto again;
4781 }
4782
4783 clear_extent_bit(&inode->io_tree, block_start, block_end,
4784 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4785 &cached_state);
4786
4787 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4788 &cached_state);
4789 if (ret) {
4790 unlock_extent(io_tree, block_start, block_end, &cached_state);
4791 goto out_unlock;
4792 }
4793
4794 if (offset != blocksize) {
4795 if (!len)
4796 len = blocksize - offset;
4797 if (front)
4798 memzero_page(page, (block_start - page_offset(page)),
4799 offset);
4800 else
4801 memzero_page(page, (block_start - page_offset(page)) + offset,
4802 len);
4803 }
4804 btrfs_page_clear_checked(fs_info, page, block_start,
4805 block_end + 1 - block_start);
4806 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4807 unlock_extent(io_tree, block_start, block_end, &cached_state);
4808
4809 if (only_release_metadata)
4810 set_extent_bit(&inode->io_tree, block_start, block_end,
4811 EXTENT_NORESERVE, NULL);
4812
4813 out_unlock:
4814 if (ret) {
4815 if (only_release_metadata)
4816 btrfs_delalloc_release_metadata(inode, blocksize, true);
4817 else
4818 btrfs_delalloc_release_space(inode, data_reserved,
4819 block_start, blocksize, true);
4820 }
4821 btrfs_delalloc_release_extents(inode, blocksize);
4822 unlock_page(page);
4823 put_page(page);
4824 out:
4825 if (only_release_metadata)
4826 btrfs_check_nocow_unlock(inode);
4827 extent_changeset_free(data_reserved);
4828 return ret;
4829 }
4830
maybe_insert_hole(struct btrfs_root * root,struct btrfs_inode * inode,u64 offset,u64 len)4831 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4832 u64 offset, u64 len)
4833 {
4834 struct btrfs_fs_info *fs_info = root->fs_info;
4835 struct btrfs_trans_handle *trans;
4836 struct btrfs_drop_extents_args drop_args = { 0 };
4837 int ret;
4838
4839 /*
4840 * If NO_HOLES is enabled, we don't need to do anything.
4841 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4842 * or btrfs_update_inode() will be called, which guarantee that the next
4843 * fsync will know this inode was changed and needs to be logged.
4844 */
4845 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4846 return 0;
4847
4848 /*
4849 * 1 - for the one we're dropping
4850 * 1 - for the one we're adding
4851 * 1 - for updating the inode.
4852 */
4853 trans = btrfs_start_transaction(root, 3);
4854 if (IS_ERR(trans))
4855 return PTR_ERR(trans);
4856
4857 drop_args.start = offset;
4858 drop_args.end = offset + len;
4859 drop_args.drop_cache = true;
4860
4861 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4862 if (ret) {
4863 btrfs_abort_transaction(trans, ret);
4864 btrfs_end_transaction(trans);
4865 return ret;
4866 }
4867
4868 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4869 if (ret) {
4870 btrfs_abort_transaction(trans, ret);
4871 } else {
4872 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4873 btrfs_update_inode(trans, root, inode);
4874 }
4875 btrfs_end_transaction(trans);
4876 return ret;
4877 }
4878
4879 /*
4880 * This function puts in dummy file extents for the area we're creating a hole
4881 * for. So if we are truncating this file to a larger size we need to insert
4882 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4883 * the range between oldsize and size
4884 */
btrfs_cont_expand(struct btrfs_inode * inode,loff_t oldsize,loff_t size)4885 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4886 {
4887 struct btrfs_root *root = inode->root;
4888 struct btrfs_fs_info *fs_info = root->fs_info;
4889 struct extent_io_tree *io_tree = &inode->io_tree;
4890 struct extent_map *em = NULL;
4891 struct extent_state *cached_state = NULL;
4892 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4893 u64 block_end = ALIGN(size, fs_info->sectorsize);
4894 u64 last_byte;
4895 u64 cur_offset;
4896 u64 hole_size;
4897 int err = 0;
4898
4899 /*
4900 * If our size started in the middle of a block we need to zero out the
4901 * rest of the block before we expand the i_size, otherwise we could
4902 * expose stale data.
4903 */
4904 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4905 if (err)
4906 return err;
4907
4908 if (size <= hole_start)
4909 return 0;
4910
4911 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4912 &cached_state);
4913 cur_offset = hole_start;
4914 while (1) {
4915 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4916 block_end - cur_offset);
4917 if (IS_ERR(em)) {
4918 err = PTR_ERR(em);
4919 em = NULL;
4920 break;
4921 }
4922 last_byte = min(extent_map_end(em), block_end);
4923 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4924 hole_size = last_byte - cur_offset;
4925
4926 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4927 struct extent_map *hole_em;
4928
4929 err = maybe_insert_hole(root, inode, cur_offset,
4930 hole_size);
4931 if (err)
4932 break;
4933
4934 err = btrfs_inode_set_file_extent_range(inode,
4935 cur_offset, hole_size);
4936 if (err)
4937 break;
4938
4939 hole_em = alloc_extent_map();
4940 if (!hole_em) {
4941 btrfs_drop_extent_map_range(inode, cur_offset,
4942 cur_offset + hole_size - 1,
4943 false);
4944 btrfs_set_inode_full_sync(inode);
4945 goto next;
4946 }
4947 hole_em->start = cur_offset;
4948 hole_em->len = hole_size;
4949 hole_em->orig_start = cur_offset;
4950
4951 hole_em->block_start = EXTENT_MAP_HOLE;
4952 hole_em->block_len = 0;
4953 hole_em->orig_block_len = 0;
4954 hole_em->ram_bytes = hole_size;
4955 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4956 hole_em->generation = fs_info->generation;
4957
4958 err = btrfs_replace_extent_map_range(inode, hole_em, true);
4959 free_extent_map(hole_em);
4960 } else {
4961 err = btrfs_inode_set_file_extent_range(inode,
4962 cur_offset, hole_size);
4963 if (err)
4964 break;
4965 }
4966 next:
4967 free_extent_map(em);
4968 em = NULL;
4969 cur_offset = last_byte;
4970 if (cur_offset >= block_end)
4971 break;
4972 }
4973 free_extent_map(em);
4974 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
4975 return err;
4976 }
4977
btrfs_setsize(struct inode * inode,struct iattr * attr)4978 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4979 {
4980 struct btrfs_root *root = BTRFS_I(inode)->root;
4981 struct btrfs_trans_handle *trans;
4982 loff_t oldsize = i_size_read(inode);
4983 loff_t newsize = attr->ia_size;
4984 int mask = attr->ia_valid;
4985 int ret;
4986
4987 /*
4988 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4989 * special case where we need to update the times despite not having
4990 * these flags set. For all other operations the VFS set these flags
4991 * explicitly if it wants a timestamp update.
4992 */
4993 if (newsize != oldsize) {
4994 inode_inc_iversion(inode);
4995 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
4996 inode->i_mtime = inode_set_ctime_current(inode);
4997 }
4998 }
4999
5000 if (newsize > oldsize) {
5001 /*
5002 * Don't do an expanding truncate while snapshotting is ongoing.
5003 * This is to ensure the snapshot captures a fully consistent
5004 * state of this file - if the snapshot captures this expanding
5005 * truncation, it must capture all writes that happened before
5006 * this truncation.
5007 */
5008 btrfs_drew_write_lock(&root->snapshot_lock);
5009 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5010 if (ret) {
5011 btrfs_drew_write_unlock(&root->snapshot_lock);
5012 return ret;
5013 }
5014
5015 trans = btrfs_start_transaction(root, 1);
5016 if (IS_ERR(trans)) {
5017 btrfs_drew_write_unlock(&root->snapshot_lock);
5018 return PTR_ERR(trans);
5019 }
5020
5021 i_size_write(inode, newsize);
5022 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5023 pagecache_isize_extended(inode, oldsize, newsize);
5024 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5025 btrfs_drew_write_unlock(&root->snapshot_lock);
5026 btrfs_end_transaction(trans);
5027 } else {
5028 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5029
5030 if (btrfs_is_zoned(fs_info)) {
5031 ret = btrfs_wait_ordered_range(inode,
5032 ALIGN(newsize, fs_info->sectorsize),
5033 (u64)-1);
5034 if (ret)
5035 return ret;
5036 }
5037
5038 /*
5039 * We're truncating a file that used to have good data down to
5040 * zero. Make sure any new writes to the file get on disk
5041 * on close.
5042 */
5043 if (newsize == 0)
5044 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5045 &BTRFS_I(inode)->runtime_flags);
5046
5047 truncate_setsize(inode, newsize);
5048
5049 inode_dio_wait(inode);
5050
5051 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5052 if (ret && inode->i_nlink) {
5053 int err;
5054
5055 /*
5056 * Truncate failed, so fix up the in-memory size. We
5057 * adjusted disk_i_size down as we removed extents, so
5058 * wait for disk_i_size to be stable and then update the
5059 * in-memory size to match.
5060 */
5061 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5062 if (err)
5063 return err;
5064 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5065 }
5066 }
5067
5068 return ret;
5069 }
5070
btrfs_setattr(struct mnt_idmap * idmap,struct dentry * dentry,struct iattr * attr)5071 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5072 struct iattr *attr)
5073 {
5074 struct inode *inode = d_inode(dentry);
5075 struct btrfs_root *root = BTRFS_I(inode)->root;
5076 int err;
5077
5078 if (btrfs_root_readonly(root))
5079 return -EROFS;
5080
5081 err = setattr_prepare(idmap, dentry, attr);
5082 if (err)
5083 return err;
5084
5085 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5086 err = btrfs_setsize(inode, attr);
5087 if (err)
5088 return err;
5089 }
5090
5091 if (attr->ia_valid) {
5092 setattr_copy(idmap, inode, attr);
5093 inode_inc_iversion(inode);
5094 err = btrfs_dirty_inode(BTRFS_I(inode));
5095
5096 if (!err && attr->ia_valid & ATTR_MODE)
5097 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5098 }
5099
5100 return err;
5101 }
5102
5103 /*
5104 * While truncating the inode pages during eviction, we get the VFS
5105 * calling btrfs_invalidate_folio() against each folio of the inode. This
5106 * is slow because the calls to btrfs_invalidate_folio() result in a
5107 * huge amount of calls to lock_extent() and clear_extent_bit(),
5108 * which keep merging and splitting extent_state structures over and over,
5109 * wasting lots of time.
5110 *
5111 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5112 * skip all those expensive operations on a per folio basis and do only
5113 * the ordered io finishing, while we release here the extent_map and
5114 * extent_state structures, without the excessive merging and splitting.
5115 */
evict_inode_truncate_pages(struct inode * inode)5116 static void evict_inode_truncate_pages(struct inode *inode)
5117 {
5118 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5119 struct rb_node *node;
5120
5121 ASSERT(inode->i_state & I_FREEING);
5122 truncate_inode_pages_final(&inode->i_data);
5123
5124 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5125
5126 /*
5127 * Keep looping until we have no more ranges in the io tree.
5128 * We can have ongoing bios started by readahead that have
5129 * their endio callback (extent_io.c:end_bio_extent_readpage)
5130 * still in progress (unlocked the pages in the bio but did not yet
5131 * unlocked the ranges in the io tree). Therefore this means some
5132 * ranges can still be locked and eviction started because before
5133 * submitting those bios, which are executed by a separate task (work
5134 * queue kthread), inode references (inode->i_count) were not taken
5135 * (which would be dropped in the end io callback of each bio).
5136 * Therefore here we effectively end up waiting for those bios and
5137 * anyone else holding locked ranges without having bumped the inode's
5138 * reference count - if we don't do it, when they access the inode's
5139 * io_tree to unlock a range it may be too late, leading to an
5140 * use-after-free issue.
5141 */
5142 spin_lock(&io_tree->lock);
5143 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5144 struct extent_state *state;
5145 struct extent_state *cached_state = NULL;
5146 u64 start;
5147 u64 end;
5148 unsigned state_flags;
5149
5150 node = rb_first(&io_tree->state);
5151 state = rb_entry(node, struct extent_state, rb_node);
5152 start = state->start;
5153 end = state->end;
5154 state_flags = state->state;
5155 spin_unlock(&io_tree->lock);
5156
5157 lock_extent(io_tree, start, end, &cached_state);
5158
5159 /*
5160 * If still has DELALLOC flag, the extent didn't reach disk,
5161 * and its reserved space won't be freed by delayed_ref.
5162 * So we need to free its reserved space here.
5163 * (Refer to comment in btrfs_invalidate_folio, case 2)
5164 *
5165 * Note, end is the bytenr of last byte, so we need + 1 here.
5166 */
5167 if (state_flags & EXTENT_DELALLOC)
5168 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5169 end - start + 1, NULL);
5170
5171 clear_extent_bit(io_tree, start, end,
5172 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5173 &cached_state);
5174
5175 cond_resched();
5176 spin_lock(&io_tree->lock);
5177 }
5178 spin_unlock(&io_tree->lock);
5179 }
5180
evict_refill_and_join(struct btrfs_root * root,struct btrfs_block_rsv * rsv)5181 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5182 struct btrfs_block_rsv *rsv)
5183 {
5184 struct btrfs_fs_info *fs_info = root->fs_info;
5185 struct btrfs_trans_handle *trans;
5186 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5187 int ret;
5188
5189 /*
5190 * Eviction should be taking place at some place safe because of our
5191 * delayed iputs. However the normal flushing code will run delayed
5192 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5193 *
5194 * We reserve the delayed_refs_extra here again because we can't use
5195 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5196 * above. We reserve our extra bit here because we generate a ton of
5197 * delayed refs activity by truncating.
5198 *
5199 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5200 * if we fail to make this reservation we can re-try without the
5201 * delayed_refs_extra so we can make some forward progress.
5202 */
5203 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5204 BTRFS_RESERVE_FLUSH_EVICT);
5205 if (ret) {
5206 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5207 BTRFS_RESERVE_FLUSH_EVICT);
5208 if (ret) {
5209 btrfs_warn(fs_info,
5210 "could not allocate space for delete; will truncate on mount");
5211 return ERR_PTR(-ENOSPC);
5212 }
5213 delayed_refs_extra = 0;
5214 }
5215
5216 trans = btrfs_join_transaction(root);
5217 if (IS_ERR(trans))
5218 return trans;
5219
5220 if (delayed_refs_extra) {
5221 trans->block_rsv = &fs_info->trans_block_rsv;
5222 trans->bytes_reserved = delayed_refs_extra;
5223 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5224 delayed_refs_extra, true);
5225 }
5226 return trans;
5227 }
5228
btrfs_evict_inode(struct inode * inode)5229 void btrfs_evict_inode(struct inode *inode)
5230 {
5231 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5232 struct btrfs_trans_handle *trans;
5233 struct btrfs_root *root = BTRFS_I(inode)->root;
5234 struct btrfs_block_rsv *rsv = NULL;
5235 int ret;
5236
5237 trace_btrfs_inode_evict(inode);
5238
5239 if (!root) {
5240 fsverity_cleanup_inode(inode);
5241 clear_inode(inode);
5242 return;
5243 }
5244
5245 evict_inode_truncate_pages(inode);
5246
5247 if (inode->i_nlink &&
5248 ((btrfs_root_refs(&root->root_item) != 0 &&
5249 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5250 btrfs_is_free_space_inode(BTRFS_I(inode))))
5251 goto out;
5252
5253 if (is_bad_inode(inode))
5254 goto out;
5255
5256 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5257 goto out;
5258
5259 if (inode->i_nlink > 0) {
5260 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5261 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5262 goto out;
5263 }
5264
5265 /*
5266 * This makes sure the inode item in tree is uptodate and the space for
5267 * the inode update is released.
5268 */
5269 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5270 if (ret)
5271 goto out;
5272
5273 /*
5274 * This drops any pending insert or delete operations we have for this
5275 * inode. We could have a delayed dir index deletion queued up, but
5276 * we're removing the inode completely so that'll be taken care of in
5277 * the truncate.
5278 */
5279 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5280
5281 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5282 if (!rsv)
5283 goto out;
5284 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5285 rsv->failfast = true;
5286
5287 btrfs_i_size_write(BTRFS_I(inode), 0);
5288
5289 while (1) {
5290 struct btrfs_truncate_control control = {
5291 .inode = BTRFS_I(inode),
5292 .ino = btrfs_ino(BTRFS_I(inode)),
5293 .new_size = 0,
5294 .min_type = 0,
5295 };
5296
5297 trans = evict_refill_and_join(root, rsv);
5298 if (IS_ERR(trans))
5299 goto out;
5300
5301 trans->block_rsv = rsv;
5302
5303 ret = btrfs_truncate_inode_items(trans, root, &control);
5304 trans->block_rsv = &fs_info->trans_block_rsv;
5305 btrfs_end_transaction(trans);
5306 /*
5307 * We have not added new delayed items for our inode after we
5308 * have flushed its delayed items, so no need to throttle on
5309 * delayed items. However we have modified extent buffers.
5310 */
5311 btrfs_btree_balance_dirty_nodelay(fs_info);
5312 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5313 goto out;
5314 else if (!ret)
5315 break;
5316 }
5317
5318 /*
5319 * Errors here aren't a big deal, it just means we leave orphan items in
5320 * the tree. They will be cleaned up on the next mount. If the inode
5321 * number gets reused, cleanup deletes the orphan item without doing
5322 * anything, and unlink reuses the existing orphan item.
5323 *
5324 * If it turns out that we are dropping too many of these, we might want
5325 * to add a mechanism for retrying these after a commit.
5326 */
5327 trans = evict_refill_and_join(root, rsv);
5328 if (!IS_ERR(trans)) {
5329 trans->block_rsv = rsv;
5330 btrfs_orphan_del(trans, BTRFS_I(inode));
5331 trans->block_rsv = &fs_info->trans_block_rsv;
5332 btrfs_end_transaction(trans);
5333 }
5334
5335 out:
5336 btrfs_free_block_rsv(fs_info, rsv);
5337 /*
5338 * If we didn't successfully delete, the orphan item will still be in
5339 * the tree and we'll retry on the next mount. Again, we might also want
5340 * to retry these periodically in the future.
5341 */
5342 btrfs_remove_delayed_node(BTRFS_I(inode));
5343 fsverity_cleanup_inode(inode);
5344 clear_inode(inode);
5345 }
5346
5347 /*
5348 * Return the key found in the dir entry in the location pointer, fill @type
5349 * with BTRFS_FT_*, and return 0.
5350 *
5351 * If no dir entries were found, returns -ENOENT.
5352 * If found a corrupted location in dir entry, returns -EUCLEAN.
5353 */
btrfs_inode_by_name(struct btrfs_inode * dir,struct dentry * dentry,struct btrfs_key * location,u8 * type)5354 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5355 struct btrfs_key *location, u8 *type)
5356 {
5357 struct btrfs_dir_item *di;
5358 struct btrfs_path *path;
5359 struct btrfs_root *root = dir->root;
5360 int ret = 0;
5361 struct fscrypt_name fname;
5362
5363 path = btrfs_alloc_path();
5364 if (!path)
5365 return -ENOMEM;
5366
5367 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5368 if (ret < 0)
5369 goto out;
5370 /*
5371 * fscrypt_setup_filename() should never return a positive value, but
5372 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5373 */
5374 ASSERT(ret == 0);
5375
5376 /* This needs to handle no-key deletions later on */
5377
5378 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5379 &fname.disk_name, 0);
5380 if (IS_ERR_OR_NULL(di)) {
5381 ret = di ? PTR_ERR(di) : -ENOENT;
5382 goto out;
5383 }
5384
5385 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5386 if (location->type != BTRFS_INODE_ITEM_KEY &&
5387 location->type != BTRFS_ROOT_ITEM_KEY) {
5388 ret = -EUCLEAN;
5389 btrfs_warn(root->fs_info,
5390 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5391 __func__, fname.disk_name.name, btrfs_ino(dir),
5392 location->objectid, location->type, location->offset);
5393 }
5394 if (!ret)
5395 *type = btrfs_dir_ftype(path->nodes[0], di);
5396 out:
5397 fscrypt_free_filename(&fname);
5398 btrfs_free_path(path);
5399 return ret;
5400 }
5401
5402 /*
5403 * when we hit a tree root in a directory, the btrfs part of the inode
5404 * needs to be changed to reflect the root directory of the tree root. This
5405 * is kind of like crossing a mount point.
5406 */
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)5407 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5408 struct btrfs_inode *dir,
5409 struct dentry *dentry,
5410 struct btrfs_key *location,
5411 struct btrfs_root **sub_root)
5412 {
5413 struct btrfs_path *path;
5414 struct btrfs_root *new_root;
5415 struct btrfs_root_ref *ref;
5416 struct extent_buffer *leaf;
5417 struct btrfs_key key;
5418 int ret;
5419 int err = 0;
5420 struct fscrypt_name fname;
5421
5422 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5423 if (ret)
5424 return ret;
5425
5426 path = btrfs_alloc_path();
5427 if (!path) {
5428 err = -ENOMEM;
5429 goto out;
5430 }
5431
5432 err = -ENOENT;
5433 key.objectid = dir->root->root_key.objectid;
5434 key.type = BTRFS_ROOT_REF_KEY;
5435 key.offset = location->objectid;
5436
5437 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5438 if (ret) {
5439 if (ret < 0)
5440 err = ret;
5441 goto out;
5442 }
5443
5444 leaf = path->nodes[0];
5445 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5446 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5447 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5448 goto out;
5449
5450 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5451 (unsigned long)(ref + 1), fname.disk_name.len);
5452 if (ret)
5453 goto out;
5454
5455 btrfs_release_path(path);
5456
5457 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5458 if (IS_ERR(new_root)) {
5459 err = PTR_ERR(new_root);
5460 goto out;
5461 }
5462
5463 *sub_root = new_root;
5464 location->objectid = btrfs_root_dirid(&new_root->root_item);
5465 location->type = BTRFS_INODE_ITEM_KEY;
5466 location->offset = 0;
5467 err = 0;
5468 out:
5469 btrfs_free_path(path);
5470 fscrypt_free_filename(&fname);
5471 return err;
5472 }
5473
inode_tree_add(struct btrfs_inode * inode)5474 static void inode_tree_add(struct btrfs_inode *inode)
5475 {
5476 struct btrfs_root *root = inode->root;
5477 struct btrfs_inode *entry;
5478 struct rb_node **p;
5479 struct rb_node *parent;
5480 struct rb_node *new = &inode->rb_node;
5481 u64 ino = btrfs_ino(inode);
5482
5483 if (inode_unhashed(&inode->vfs_inode))
5484 return;
5485 parent = NULL;
5486 spin_lock(&root->inode_lock);
5487 p = &root->inode_tree.rb_node;
5488 while (*p) {
5489 parent = *p;
5490 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5491
5492 if (ino < btrfs_ino(entry))
5493 p = &parent->rb_left;
5494 else if (ino > btrfs_ino(entry))
5495 p = &parent->rb_right;
5496 else {
5497 WARN_ON(!(entry->vfs_inode.i_state &
5498 (I_WILL_FREE | I_FREEING)));
5499 rb_replace_node(parent, new, &root->inode_tree);
5500 RB_CLEAR_NODE(parent);
5501 spin_unlock(&root->inode_lock);
5502 return;
5503 }
5504 }
5505 rb_link_node(new, parent, p);
5506 rb_insert_color(new, &root->inode_tree);
5507 spin_unlock(&root->inode_lock);
5508 }
5509
inode_tree_del(struct btrfs_inode * inode)5510 static void inode_tree_del(struct btrfs_inode *inode)
5511 {
5512 struct btrfs_root *root = inode->root;
5513 int empty = 0;
5514
5515 spin_lock(&root->inode_lock);
5516 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5517 rb_erase(&inode->rb_node, &root->inode_tree);
5518 RB_CLEAR_NODE(&inode->rb_node);
5519 empty = RB_EMPTY_ROOT(&root->inode_tree);
5520 }
5521 spin_unlock(&root->inode_lock);
5522
5523 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5524 spin_lock(&root->inode_lock);
5525 empty = RB_EMPTY_ROOT(&root->inode_tree);
5526 spin_unlock(&root->inode_lock);
5527 if (empty)
5528 btrfs_add_dead_root(root);
5529 }
5530 }
5531
5532
btrfs_init_locked_inode(struct inode * inode,void * p)5533 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5534 {
5535 struct btrfs_iget_args *args = p;
5536
5537 inode->i_ino = args->ino;
5538 BTRFS_I(inode)->location.objectid = args->ino;
5539 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5540 BTRFS_I(inode)->location.offset = 0;
5541 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5542 BUG_ON(args->root && !BTRFS_I(inode)->root);
5543
5544 if (args->root && args->root == args->root->fs_info->tree_root &&
5545 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5546 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5547 &BTRFS_I(inode)->runtime_flags);
5548 return 0;
5549 }
5550
btrfs_find_actor(struct inode * inode,void * opaque)5551 static int btrfs_find_actor(struct inode *inode, void *opaque)
5552 {
5553 struct btrfs_iget_args *args = opaque;
5554
5555 return args->ino == BTRFS_I(inode)->location.objectid &&
5556 args->root == BTRFS_I(inode)->root;
5557 }
5558
btrfs_iget_locked(struct super_block * s,u64 ino,struct btrfs_root * root)5559 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5560 struct btrfs_root *root)
5561 {
5562 struct inode *inode;
5563 struct btrfs_iget_args args;
5564 unsigned long hashval = btrfs_inode_hash(ino, root);
5565
5566 args.ino = ino;
5567 args.root = root;
5568
5569 inode = iget5_locked(s, hashval, btrfs_find_actor,
5570 btrfs_init_locked_inode,
5571 (void *)&args);
5572 return inode;
5573 }
5574
5575 /*
5576 * Get an inode object given its inode number and corresponding root.
5577 * Path can be preallocated to prevent recursing back to iget through
5578 * allocator. NULL is also valid but may require an additional allocation
5579 * later.
5580 */
btrfs_iget_path(struct super_block * s,u64 ino,struct btrfs_root * root,struct btrfs_path * path)5581 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5582 struct btrfs_root *root, struct btrfs_path *path)
5583 {
5584 struct inode *inode;
5585
5586 inode = btrfs_iget_locked(s, ino, root);
5587 if (!inode)
5588 return ERR_PTR(-ENOMEM);
5589
5590 if (inode->i_state & I_NEW) {
5591 int ret;
5592
5593 ret = btrfs_read_locked_inode(inode, path);
5594 if (!ret) {
5595 inode_tree_add(BTRFS_I(inode));
5596 unlock_new_inode(inode);
5597 } else {
5598 iget_failed(inode);
5599 /*
5600 * ret > 0 can come from btrfs_search_slot called by
5601 * btrfs_read_locked_inode, this means the inode item
5602 * was not found.
5603 */
5604 if (ret > 0)
5605 ret = -ENOENT;
5606 inode = ERR_PTR(ret);
5607 }
5608 }
5609
5610 return inode;
5611 }
5612
btrfs_iget(struct super_block * s,u64 ino,struct btrfs_root * root)5613 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5614 {
5615 return btrfs_iget_path(s, ino, root, NULL);
5616 }
5617
new_simple_dir(struct inode * dir,struct btrfs_key * key,struct btrfs_root * root)5618 static struct inode *new_simple_dir(struct inode *dir,
5619 struct btrfs_key *key,
5620 struct btrfs_root *root)
5621 {
5622 struct inode *inode = new_inode(dir->i_sb);
5623
5624 if (!inode)
5625 return ERR_PTR(-ENOMEM);
5626
5627 BTRFS_I(inode)->root = btrfs_grab_root(root);
5628 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5629 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5630
5631 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5632 /*
5633 * We only need lookup, the rest is read-only and there's no inode
5634 * associated with the dentry
5635 */
5636 inode->i_op = &simple_dir_inode_operations;
5637 inode->i_opflags &= ~IOP_XATTR;
5638 inode->i_fop = &simple_dir_operations;
5639 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5640 inode->i_mtime = inode_set_ctime_current(inode);
5641 inode->i_atime = dir->i_atime;
5642 BTRFS_I(inode)->i_otime = inode->i_mtime;
5643 inode->i_uid = dir->i_uid;
5644 inode->i_gid = dir->i_gid;
5645
5646 return inode;
5647 }
5648
5649 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5650 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5651 static_assert(BTRFS_FT_DIR == FT_DIR);
5652 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5653 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5654 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5655 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5656 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5657
btrfs_inode_type(struct inode * inode)5658 static inline u8 btrfs_inode_type(struct inode *inode)
5659 {
5660 return fs_umode_to_ftype(inode->i_mode);
5661 }
5662
btrfs_lookup_dentry(struct inode * dir,struct dentry * dentry)5663 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5664 {
5665 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5666 struct inode *inode;
5667 struct btrfs_root *root = BTRFS_I(dir)->root;
5668 struct btrfs_root *sub_root = root;
5669 struct btrfs_key location = { 0 };
5670 u8 di_type = 0;
5671 int ret = 0;
5672
5673 if (dentry->d_name.len > BTRFS_NAME_LEN)
5674 return ERR_PTR(-ENAMETOOLONG);
5675
5676 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5677 if (ret < 0)
5678 return ERR_PTR(ret);
5679
5680 if (location.type == BTRFS_INODE_ITEM_KEY) {
5681 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5682 if (IS_ERR(inode))
5683 return inode;
5684
5685 /* Do extra check against inode mode with di_type */
5686 if (btrfs_inode_type(inode) != di_type) {
5687 btrfs_crit(fs_info,
5688 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5689 inode->i_mode, btrfs_inode_type(inode),
5690 di_type);
5691 iput(inode);
5692 return ERR_PTR(-EUCLEAN);
5693 }
5694 return inode;
5695 }
5696
5697 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5698 &location, &sub_root);
5699 if (ret < 0) {
5700 if (ret != -ENOENT)
5701 inode = ERR_PTR(ret);
5702 else
5703 inode = new_simple_dir(dir, &location, root);
5704 } else {
5705 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5706 btrfs_put_root(sub_root);
5707
5708 if (IS_ERR(inode))
5709 return inode;
5710
5711 down_read(&fs_info->cleanup_work_sem);
5712 if (!sb_rdonly(inode->i_sb))
5713 ret = btrfs_orphan_cleanup(sub_root);
5714 up_read(&fs_info->cleanup_work_sem);
5715 if (ret) {
5716 iput(inode);
5717 inode = ERR_PTR(ret);
5718 }
5719 }
5720
5721 return inode;
5722 }
5723
btrfs_dentry_delete(const struct dentry * dentry)5724 static int btrfs_dentry_delete(const struct dentry *dentry)
5725 {
5726 struct btrfs_root *root;
5727 struct inode *inode = d_inode(dentry);
5728
5729 if (!inode && !IS_ROOT(dentry))
5730 inode = d_inode(dentry->d_parent);
5731
5732 if (inode) {
5733 root = BTRFS_I(inode)->root;
5734 if (btrfs_root_refs(&root->root_item) == 0)
5735 return 1;
5736
5737 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5738 return 1;
5739 }
5740 return 0;
5741 }
5742
btrfs_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)5743 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5744 unsigned int flags)
5745 {
5746 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5747
5748 if (inode == ERR_PTR(-ENOENT))
5749 inode = NULL;
5750 return d_splice_alias(inode, dentry);
5751 }
5752
5753 /*
5754 * Find the highest existing sequence number in a directory and then set the
5755 * in-memory index_cnt variable to the first free sequence number.
5756 */
btrfs_set_inode_index_count(struct btrfs_inode * inode)5757 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5758 {
5759 struct btrfs_root *root = inode->root;
5760 struct btrfs_key key, found_key;
5761 struct btrfs_path *path;
5762 struct extent_buffer *leaf;
5763 int ret;
5764
5765 key.objectid = btrfs_ino(inode);
5766 key.type = BTRFS_DIR_INDEX_KEY;
5767 key.offset = (u64)-1;
5768
5769 path = btrfs_alloc_path();
5770 if (!path)
5771 return -ENOMEM;
5772
5773 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5774 if (ret < 0)
5775 goto out;
5776 /* FIXME: we should be able to handle this */
5777 if (ret == 0)
5778 goto out;
5779 ret = 0;
5780
5781 if (path->slots[0] == 0) {
5782 inode->index_cnt = BTRFS_DIR_START_INDEX;
5783 goto out;
5784 }
5785
5786 path->slots[0]--;
5787
5788 leaf = path->nodes[0];
5789 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5790
5791 if (found_key.objectid != btrfs_ino(inode) ||
5792 found_key.type != BTRFS_DIR_INDEX_KEY) {
5793 inode->index_cnt = BTRFS_DIR_START_INDEX;
5794 goto out;
5795 }
5796
5797 inode->index_cnt = found_key.offset + 1;
5798 out:
5799 btrfs_free_path(path);
5800 return ret;
5801 }
5802
btrfs_get_dir_last_index(struct btrfs_inode * dir,u64 * index)5803 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5804 {
5805 int ret = 0;
5806
5807 btrfs_inode_lock(dir, 0);
5808 if (dir->index_cnt == (u64)-1) {
5809 ret = btrfs_inode_delayed_dir_index_count(dir);
5810 if (ret) {
5811 ret = btrfs_set_inode_index_count(dir);
5812 if (ret)
5813 goto out;
5814 }
5815 }
5816
5817 /* index_cnt is the index number of next new entry, so decrement it. */
5818 *index = dir->index_cnt - 1;
5819 out:
5820 btrfs_inode_unlock(dir, 0);
5821
5822 return ret;
5823 }
5824
5825 /*
5826 * All this infrastructure exists because dir_emit can fault, and we are holding
5827 * the tree lock when doing readdir. For now just allocate a buffer and copy
5828 * our information into that, and then dir_emit from the buffer. This is
5829 * similar to what NFS does, only we don't keep the buffer around in pagecache
5830 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5831 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5832 * tree lock.
5833 */
btrfs_opendir(struct inode * inode,struct file * file)5834 static int btrfs_opendir(struct inode *inode, struct file *file)
5835 {
5836 struct btrfs_file_private *private;
5837 u64 last_index;
5838 int ret;
5839
5840 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5841 if (ret)
5842 return ret;
5843
5844 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5845 if (!private)
5846 return -ENOMEM;
5847 private->last_index = last_index;
5848 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5849 if (!private->filldir_buf) {
5850 kfree(private);
5851 return -ENOMEM;
5852 }
5853 file->private_data = private;
5854 return 0;
5855 }
5856
btrfs_dir_llseek(struct file * file,loff_t offset,int whence)5857 static loff_t btrfs_dir_llseek(struct file *file, loff_t offset, int whence)
5858 {
5859 struct btrfs_file_private *private = file->private_data;
5860 int ret;
5861
5862 ret = btrfs_get_dir_last_index(BTRFS_I(file_inode(file)),
5863 &private->last_index);
5864 if (ret)
5865 return ret;
5866
5867 return generic_file_llseek(file, offset, whence);
5868 }
5869
5870 struct dir_entry {
5871 u64 ino;
5872 u64 offset;
5873 unsigned type;
5874 int name_len;
5875 };
5876
btrfs_filldir(void * addr,int entries,struct dir_context * ctx)5877 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5878 {
5879 while (entries--) {
5880 struct dir_entry *entry = addr;
5881 char *name = (char *)(entry + 1);
5882
5883 ctx->pos = get_unaligned(&entry->offset);
5884 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5885 get_unaligned(&entry->ino),
5886 get_unaligned(&entry->type)))
5887 return 1;
5888 addr += sizeof(struct dir_entry) +
5889 get_unaligned(&entry->name_len);
5890 ctx->pos++;
5891 }
5892 return 0;
5893 }
5894
btrfs_real_readdir(struct file * file,struct dir_context * ctx)5895 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5896 {
5897 struct inode *inode = file_inode(file);
5898 struct btrfs_root *root = BTRFS_I(inode)->root;
5899 struct btrfs_file_private *private = file->private_data;
5900 struct btrfs_dir_item *di;
5901 struct btrfs_key key;
5902 struct btrfs_key found_key;
5903 struct btrfs_path *path;
5904 void *addr;
5905 LIST_HEAD(ins_list);
5906 LIST_HEAD(del_list);
5907 int ret;
5908 char *name_ptr;
5909 int name_len;
5910 int entries = 0;
5911 int total_len = 0;
5912 bool put = false;
5913 struct btrfs_key location;
5914
5915 if (!dir_emit_dots(file, ctx))
5916 return 0;
5917
5918 path = btrfs_alloc_path();
5919 if (!path)
5920 return -ENOMEM;
5921
5922 addr = private->filldir_buf;
5923 path->reada = READA_FORWARD;
5924
5925 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
5926 &ins_list, &del_list);
5927
5928 again:
5929 key.type = BTRFS_DIR_INDEX_KEY;
5930 key.offset = ctx->pos;
5931 key.objectid = btrfs_ino(BTRFS_I(inode));
5932
5933 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5934 struct dir_entry *entry;
5935 struct extent_buffer *leaf = path->nodes[0];
5936 u8 ftype;
5937
5938 if (found_key.objectid != key.objectid)
5939 break;
5940 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5941 break;
5942 if (found_key.offset < ctx->pos)
5943 continue;
5944 if (found_key.offset > private->last_index)
5945 break;
5946 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5947 continue;
5948 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5949 name_len = btrfs_dir_name_len(leaf, di);
5950 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5951 PAGE_SIZE) {
5952 btrfs_release_path(path);
5953 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5954 if (ret)
5955 goto nopos;
5956 addr = private->filldir_buf;
5957 entries = 0;
5958 total_len = 0;
5959 goto again;
5960 }
5961
5962 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
5963 entry = addr;
5964 name_ptr = (char *)(entry + 1);
5965 read_extent_buffer(leaf, name_ptr,
5966 (unsigned long)(di + 1), name_len);
5967 put_unaligned(name_len, &entry->name_len);
5968 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
5969 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5970 put_unaligned(location.objectid, &entry->ino);
5971 put_unaligned(found_key.offset, &entry->offset);
5972 entries++;
5973 addr += sizeof(struct dir_entry) + name_len;
5974 total_len += sizeof(struct dir_entry) + name_len;
5975 }
5976 /* Catch error encountered during iteration */
5977 if (ret < 0)
5978 goto err;
5979
5980 btrfs_release_path(path);
5981
5982 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5983 if (ret)
5984 goto nopos;
5985
5986 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5987 if (ret)
5988 goto nopos;
5989
5990 /*
5991 * Stop new entries from being returned after we return the last
5992 * entry.
5993 *
5994 * New directory entries are assigned a strictly increasing
5995 * offset. This means that new entries created during readdir
5996 * are *guaranteed* to be seen in the future by that readdir.
5997 * This has broken buggy programs which operate on names as
5998 * they're returned by readdir. Until we re-use freed offsets
5999 * we have this hack to stop new entries from being returned
6000 * under the assumption that they'll never reach this huge
6001 * offset.
6002 *
6003 * This is being careful not to overflow 32bit loff_t unless the
6004 * last entry requires it because doing so has broken 32bit apps
6005 * in the past.
6006 */
6007 if (ctx->pos >= INT_MAX)
6008 ctx->pos = LLONG_MAX;
6009 else
6010 ctx->pos = INT_MAX;
6011 nopos:
6012 ret = 0;
6013 err:
6014 if (put)
6015 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6016 btrfs_free_path(path);
6017 return ret;
6018 }
6019
6020 /*
6021 * This is somewhat expensive, updating the tree every time the
6022 * inode changes. But, it is most likely to find the inode in cache.
6023 * FIXME, needs more benchmarking...there are no reasons other than performance
6024 * to keep or drop this code.
6025 */
btrfs_dirty_inode(struct btrfs_inode * inode)6026 static int btrfs_dirty_inode(struct btrfs_inode *inode)
6027 {
6028 struct btrfs_root *root = inode->root;
6029 struct btrfs_fs_info *fs_info = root->fs_info;
6030 struct btrfs_trans_handle *trans;
6031 int ret;
6032
6033 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6034 return 0;
6035
6036 trans = btrfs_join_transaction(root);
6037 if (IS_ERR(trans))
6038 return PTR_ERR(trans);
6039
6040 ret = btrfs_update_inode(trans, root, inode);
6041 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6042 /* whoops, lets try again with the full transaction */
6043 btrfs_end_transaction(trans);
6044 trans = btrfs_start_transaction(root, 1);
6045 if (IS_ERR(trans))
6046 return PTR_ERR(trans);
6047
6048 ret = btrfs_update_inode(trans, root, inode);
6049 }
6050 btrfs_end_transaction(trans);
6051 if (inode->delayed_node)
6052 btrfs_balance_delayed_items(fs_info);
6053
6054 return ret;
6055 }
6056
6057 /*
6058 * This is a copy of file_update_time. We need this so we can return error on
6059 * ENOSPC for updating the inode in the case of file write and mmap writes.
6060 */
btrfs_update_time(struct inode * inode,int flags)6061 static int btrfs_update_time(struct inode *inode, int flags)
6062 {
6063 struct btrfs_root *root = BTRFS_I(inode)->root;
6064 bool dirty = flags & ~S_VERSION;
6065
6066 if (btrfs_root_readonly(root))
6067 return -EROFS;
6068
6069 dirty = inode_update_timestamps(inode, flags);
6070 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6071 }
6072
6073 /*
6074 * helper to find a free sequence number in a given directory. This current
6075 * code is very simple, later versions will do smarter things in the btree
6076 */
btrfs_set_inode_index(struct btrfs_inode * dir,u64 * index)6077 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6078 {
6079 int ret = 0;
6080
6081 if (dir->index_cnt == (u64)-1) {
6082 ret = btrfs_inode_delayed_dir_index_count(dir);
6083 if (ret) {
6084 ret = btrfs_set_inode_index_count(dir);
6085 if (ret)
6086 return ret;
6087 }
6088 }
6089
6090 *index = dir->index_cnt;
6091 dir->index_cnt++;
6092
6093 return ret;
6094 }
6095
btrfs_insert_inode_locked(struct inode * inode)6096 static int btrfs_insert_inode_locked(struct inode *inode)
6097 {
6098 struct btrfs_iget_args args;
6099
6100 args.ino = BTRFS_I(inode)->location.objectid;
6101 args.root = BTRFS_I(inode)->root;
6102
6103 return insert_inode_locked4(inode,
6104 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6105 btrfs_find_actor, &args);
6106 }
6107
btrfs_new_inode_prepare(struct btrfs_new_inode_args * args,unsigned int * trans_num_items)6108 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6109 unsigned int *trans_num_items)
6110 {
6111 struct inode *dir = args->dir;
6112 struct inode *inode = args->inode;
6113 int ret;
6114
6115 if (!args->orphan) {
6116 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6117 &args->fname);
6118 if (ret)
6119 return ret;
6120 }
6121
6122 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6123 if (ret) {
6124 fscrypt_free_filename(&args->fname);
6125 return ret;
6126 }
6127
6128 /* 1 to add inode item */
6129 *trans_num_items = 1;
6130 /* 1 to add compression property */
6131 if (BTRFS_I(dir)->prop_compress)
6132 (*trans_num_items)++;
6133 /* 1 to add default ACL xattr */
6134 if (args->default_acl)
6135 (*trans_num_items)++;
6136 /* 1 to add access ACL xattr */
6137 if (args->acl)
6138 (*trans_num_items)++;
6139 #ifdef CONFIG_SECURITY
6140 /* 1 to add LSM xattr */
6141 if (dir->i_security)
6142 (*trans_num_items)++;
6143 #endif
6144 if (args->orphan) {
6145 /* 1 to add orphan item */
6146 (*trans_num_items)++;
6147 } else {
6148 /*
6149 * 1 to add dir item
6150 * 1 to add dir index
6151 * 1 to update parent inode item
6152 *
6153 * No need for 1 unit for the inode ref item because it is
6154 * inserted in a batch together with the inode item at
6155 * btrfs_create_new_inode().
6156 */
6157 *trans_num_items += 3;
6158 }
6159 return 0;
6160 }
6161
btrfs_new_inode_args_destroy(struct btrfs_new_inode_args * args)6162 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6163 {
6164 posix_acl_release(args->acl);
6165 posix_acl_release(args->default_acl);
6166 fscrypt_free_filename(&args->fname);
6167 }
6168
6169 /*
6170 * Inherit flags from the parent inode.
6171 *
6172 * Currently only the compression flags and the cow flags are inherited.
6173 */
btrfs_inherit_iflags(struct btrfs_inode * inode,struct btrfs_inode * dir)6174 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6175 {
6176 unsigned int flags;
6177
6178 flags = dir->flags;
6179
6180 if (flags & BTRFS_INODE_NOCOMPRESS) {
6181 inode->flags &= ~BTRFS_INODE_COMPRESS;
6182 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6183 } else if (flags & BTRFS_INODE_COMPRESS) {
6184 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6185 inode->flags |= BTRFS_INODE_COMPRESS;
6186 }
6187
6188 if (flags & BTRFS_INODE_NODATACOW) {
6189 inode->flags |= BTRFS_INODE_NODATACOW;
6190 if (S_ISREG(inode->vfs_inode.i_mode))
6191 inode->flags |= BTRFS_INODE_NODATASUM;
6192 }
6193
6194 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6195 }
6196
btrfs_create_new_inode(struct btrfs_trans_handle * trans,struct btrfs_new_inode_args * args)6197 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6198 struct btrfs_new_inode_args *args)
6199 {
6200 struct inode *dir = args->dir;
6201 struct inode *inode = args->inode;
6202 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6203 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6204 struct btrfs_root *root;
6205 struct btrfs_inode_item *inode_item;
6206 struct btrfs_key *location;
6207 struct btrfs_path *path;
6208 u64 objectid;
6209 struct btrfs_inode_ref *ref;
6210 struct btrfs_key key[2];
6211 u32 sizes[2];
6212 struct btrfs_item_batch batch;
6213 unsigned long ptr;
6214 int ret;
6215
6216 path = btrfs_alloc_path();
6217 if (!path)
6218 return -ENOMEM;
6219
6220 if (!args->subvol)
6221 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6222 root = BTRFS_I(inode)->root;
6223
6224 ret = btrfs_get_free_objectid(root, &objectid);
6225 if (ret)
6226 goto out;
6227 inode->i_ino = objectid;
6228
6229 if (args->orphan) {
6230 /*
6231 * O_TMPFILE, set link count to 0, so that after this point, we
6232 * fill in an inode item with the correct link count.
6233 */
6234 set_nlink(inode, 0);
6235 } else {
6236 trace_btrfs_inode_request(dir);
6237
6238 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6239 if (ret)
6240 goto out;
6241 }
6242 /* index_cnt is ignored for everything but a dir. */
6243 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6244 BTRFS_I(inode)->generation = trans->transid;
6245 inode->i_generation = BTRFS_I(inode)->generation;
6246
6247 /*
6248 * Subvolumes don't inherit flags from their parent directory.
6249 * Originally this was probably by accident, but we probably can't
6250 * change it now without compatibility issues.
6251 */
6252 if (!args->subvol)
6253 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6254
6255 if (S_ISREG(inode->i_mode)) {
6256 if (btrfs_test_opt(fs_info, NODATASUM))
6257 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6258 if (btrfs_test_opt(fs_info, NODATACOW))
6259 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6260 BTRFS_INODE_NODATASUM;
6261 }
6262
6263 location = &BTRFS_I(inode)->location;
6264 location->objectid = objectid;
6265 location->offset = 0;
6266 location->type = BTRFS_INODE_ITEM_KEY;
6267
6268 ret = btrfs_insert_inode_locked(inode);
6269 if (ret < 0) {
6270 if (!args->orphan)
6271 BTRFS_I(dir)->index_cnt--;
6272 goto out;
6273 }
6274
6275 /*
6276 * We could have gotten an inode number from somebody who was fsynced
6277 * and then removed in this same transaction, so let's just set full
6278 * sync since it will be a full sync anyway and this will blow away the
6279 * old info in the log.
6280 */
6281 btrfs_set_inode_full_sync(BTRFS_I(inode));
6282
6283 key[0].objectid = objectid;
6284 key[0].type = BTRFS_INODE_ITEM_KEY;
6285 key[0].offset = 0;
6286
6287 sizes[0] = sizeof(struct btrfs_inode_item);
6288
6289 if (!args->orphan) {
6290 /*
6291 * Start new inodes with an inode_ref. This is slightly more
6292 * efficient for small numbers of hard links since they will
6293 * be packed into one item. Extended refs will kick in if we
6294 * add more hard links than can fit in the ref item.
6295 */
6296 key[1].objectid = objectid;
6297 key[1].type = BTRFS_INODE_REF_KEY;
6298 if (args->subvol) {
6299 key[1].offset = objectid;
6300 sizes[1] = 2 + sizeof(*ref);
6301 } else {
6302 key[1].offset = btrfs_ino(BTRFS_I(dir));
6303 sizes[1] = name->len + sizeof(*ref);
6304 }
6305 }
6306
6307 batch.keys = &key[0];
6308 batch.data_sizes = &sizes[0];
6309 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6310 batch.nr = args->orphan ? 1 : 2;
6311 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6312 if (ret != 0) {
6313 btrfs_abort_transaction(trans, ret);
6314 goto discard;
6315 }
6316
6317 inode->i_mtime = inode_set_ctime_current(inode);
6318 inode->i_atime = inode->i_mtime;
6319 BTRFS_I(inode)->i_otime = inode->i_mtime;
6320
6321 /*
6322 * We're going to fill the inode item now, so at this point the inode
6323 * must be fully initialized.
6324 */
6325
6326 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6327 struct btrfs_inode_item);
6328 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6329 sizeof(*inode_item));
6330 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6331
6332 if (!args->orphan) {
6333 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6334 struct btrfs_inode_ref);
6335 ptr = (unsigned long)(ref + 1);
6336 if (args->subvol) {
6337 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6338 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6339 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6340 } else {
6341 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6342 name->len);
6343 btrfs_set_inode_ref_index(path->nodes[0], ref,
6344 BTRFS_I(inode)->dir_index);
6345 write_extent_buffer(path->nodes[0], name->name, ptr,
6346 name->len);
6347 }
6348 }
6349
6350 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
6351 /*
6352 * We don't need the path anymore, plus inheriting properties, adding
6353 * ACLs, security xattrs, orphan item or adding the link, will result in
6354 * allocating yet another path. So just free our path.
6355 */
6356 btrfs_free_path(path);
6357 path = NULL;
6358
6359 if (args->subvol) {
6360 struct inode *parent;
6361
6362 /*
6363 * Subvolumes inherit properties from their parent subvolume,
6364 * not the directory they were created in.
6365 */
6366 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6367 BTRFS_I(dir)->root);
6368 if (IS_ERR(parent)) {
6369 ret = PTR_ERR(parent);
6370 } else {
6371 ret = btrfs_inode_inherit_props(trans, inode, parent);
6372 iput(parent);
6373 }
6374 } else {
6375 ret = btrfs_inode_inherit_props(trans, inode, dir);
6376 }
6377 if (ret) {
6378 btrfs_err(fs_info,
6379 "error inheriting props for ino %llu (root %llu): %d",
6380 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6381 ret);
6382 }
6383
6384 /*
6385 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6386 * probably a bug.
6387 */
6388 if (!args->subvol) {
6389 ret = btrfs_init_inode_security(trans, args);
6390 if (ret) {
6391 btrfs_abort_transaction(trans, ret);
6392 goto discard;
6393 }
6394 }
6395
6396 inode_tree_add(BTRFS_I(inode));
6397
6398 trace_btrfs_inode_new(inode);
6399 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6400
6401 btrfs_update_root_times(trans, root);
6402
6403 if (args->orphan) {
6404 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6405 } else {
6406 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6407 0, BTRFS_I(inode)->dir_index);
6408 }
6409 if (ret) {
6410 btrfs_abort_transaction(trans, ret);
6411 goto discard;
6412 }
6413
6414 return 0;
6415
6416 discard:
6417 /*
6418 * discard_new_inode() calls iput(), but the caller owns the reference
6419 * to the inode.
6420 */
6421 ihold(inode);
6422 discard_new_inode(inode);
6423 out:
6424 btrfs_free_path(path);
6425 return ret;
6426 }
6427
6428 /*
6429 * utility function to add 'inode' into 'parent_inode' with
6430 * a give name and a given sequence number.
6431 * if 'add_backref' is true, also insert a backref from the
6432 * inode to the parent directory.
6433 */
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)6434 int btrfs_add_link(struct btrfs_trans_handle *trans,
6435 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6436 const struct fscrypt_str *name, int add_backref, u64 index)
6437 {
6438 int ret = 0;
6439 struct btrfs_key key;
6440 struct btrfs_root *root = parent_inode->root;
6441 u64 ino = btrfs_ino(inode);
6442 u64 parent_ino = btrfs_ino(parent_inode);
6443
6444 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6445 memcpy(&key, &inode->root->root_key, sizeof(key));
6446 } else {
6447 key.objectid = ino;
6448 key.type = BTRFS_INODE_ITEM_KEY;
6449 key.offset = 0;
6450 }
6451
6452 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6453 ret = btrfs_add_root_ref(trans, key.objectid,
6454 root->root_key.objectid, parent_ino,
6455 index, name);
6456 } else if (add_backref) {
6457 ret = btrfs_insert_inode_ref(trans, root, name,
6458 ino, parent_ino, index);
6459 }
6460
6461 /* Nothing to clean up yet */
6462 if (ret)
6463 return ret;
6464
6465 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6466 btrfs_inode_type(&inode->vfs_inode), index);
6467 if (ret == -EEXIST || ret == -EOVERFLOW)
6468 goto fail_dir_item;
6469 else if (ret) {
6470 btrfs_abort_transaction(trans, ret);
6471 return ret;
6472 }
6473
6474 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6475 name->len * 2);
6476 inode_inc_iversion(&parent_inode->vfs_inode);
6477 /*
6478 * If we are replaying a log tree, we do not want to update the mtime
6479 * and ctime of the parent directory with the current time, since the
6480 * log replay procedure is responsible for setting them to their correct
6481 * values (the ones it had when the fsync was done).
6482 */
6483 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags))
6484 parent_inode->vfs_inode.i_mtime =
6485 inode_set_ctime_current(&parent_inode->vfs_inode);
6486
6487 ret = btrfs_update_inode(trans, root, parent_inode);
6488 if (ret)
6489 btrfs_abort_transaction(trans, ret);
6490 return ret;
6491
6492 fail_dir_item:
6493 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6494 u64 local_index;
6495 int err;
6496 err = btrfs_del_root_ref(trans, key.objectid,
6497 root->root_key.objectid, parent_ino,
6498 &local_index, name);
6499 if (err)
6500 btrfs_abort_transaction(trans, err);
6501 } else if (add_backref) {
6502 u64 local_index;
6503 int err;
6504
6505 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6506 &local_index);
6507 if (err)
6508 btrfs_abort_transaction(trans, err);
6509 }
6510
6511 /* Return the original error code */
6512 return ret;
6513 }
6514
btrfs_create_common(struct inode * dir,struct dentry * dentry,struct inode * inode)6515 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6516 struct inode *inode)
6517 {
6518 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6519 struct btrfs_root *root = BTRFS_I(dir)->root;
6520 struct btrfs_new_inode_args new_inode_args = {
6521 .dir = dir,
6522 .dentry = dentry,
6523 .inode = inode,
6524 };
6525 unsigned int trans_num_items;
6526 struct btrfs_trans_handle *trans;
6527 int err;
6528
6529 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6530 if (err)
6531 goto out_inode;
6532
6533 trans = btrfs_start_transaction(root, trans_num_items);
6534 if (IS_ERR(trans)) {
6535 err = PTR_ERR(trans);
6536 goto out_new_inode_args;
6537 }
6538
6539 err = btrfs_create_new_inode(trans, &new_inode_args);
6540 if (!err)
6541 d_instantiate_new(dentry, inode);
6542
6543 btrfs_end_transaction(trans);
6544 btrfs_btree_balance_dirty(fs_info);
6545 out_new_inode_args:
6546 btrfs_new_inode_args_destroy(&new_inode_args);
6547 out_inode:
6548 if (err)
6549 iput(inode);
6550 return err;
6551 }
6552
btrfs_mknod(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode,dev_t rdev)6553 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6554 struct dentry *dentry, umode_t mode, dev_t rdev)
6555 {
6556 struct inode *inode;
6557
6558 inode = new_inode(dir->i_sb);
6559 if (!inode)
6560 return -ENOMEM;
6561 inode_init_owner(idmap, inode, dir, mode);
6562 inode->i_op = &btrfs_special_inode_operations;
6563 init_special_inode(inode, inode->i_mode, rdev);
6564 return btrfs_create_common(dir, dentry, inode);
6565 }
6566
btrfs_create(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode,bool excl)6567 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6568 struct dentry *dentry, umode_t mode, bool excl)
6569 {
6570 struct inode *inode;
6571
6572 inode = new_inode(dir->i_sb);
6573 if (!inode)
6574 return -ENOMEM;
6575 inode_init_owner(idmap, inode, dir, mode);
6576 inode->i_fop = &btrfs_file_operations;
6577 inode->i_op = &btrfs_file_inode_operations;
6578 inode->i_mapping->a_ops = &btrfs_aops;
6579 return btrfs_create_common(dir, dentry, inode);
6580 }
6581
btrfs_link(struct dentry * old_dentry,struct inode * dir,struct dentry * dentry)6582 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6583 struct dentry *dentry)
6584 {
6585 struct btrfs_trans_handle *trans = NULL;
6586 struct btrfs_root *root = BTRFS_I(dir)->root;
6587 struct inode *inode = d_inode(old_dentry);
6588 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6589 struct fscrypt_name fname;
6590 u64 index;
6591 int err;
6592 int drop_inode = 0;
6593
6594 /* do not allow sys_link's with other subvols of the same device */
6595 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6596 return -EXDEV;
6597
6598 if (inode->i_nlink >= BTRFS_LINK_MAX)
6599 return -EMLINK;
6600
6601 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6602 if (err)
6603 goto fail;
6604
6605 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6606 if (err)
6607 goto fail;
6608
6609 /*
6610 * 2 items for inode and inode ref
6611 * 2 items for dir items
6612 * 1 item for parent inode
6613 * 1 item for orphan item deletion if O_TMPFILE
6614 */
6615 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6616 if (IS_ERR(trans)) {
6617 err = PTR_ERR(trans);
6618 trans = NULL;
6619 goto fail;
6620 }
6621
6622 /* There are several dir indexes for this inode, clear the cache. */
6623 BTRFS_I(inode)->dir_index = 0ULL;
6624 inc_nlink(inode);
6625 inode_inc_iversion(inode);
6626 inode_set_ctime_current(inode);
6627 ihold(inode);
6628 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6629
6630 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6631 &fname.disk_name, 1, index);
6632
6633 if (err) {
6634 drop_inode = 1;
6635 } else {
6636 struct dentry *parent = dentry->d_parent;
6637
6638 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6639 if (err)
6640 goto fail;
6641 if (inode->i_nlink == 1) {
6642 /*
6643 * If new hard link count is 1, it's a file created
6644 * with open(2) O_TMPFILE flag.
6645 */
6646 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6647 if (err)
6648 goto fail;
6649 }
6650 d_instantiate(dentry, inode);
6651 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6652 }
6653
6654 fail:
6655 fscrypt_free_filename(&fname);
6656 if (trans)
6657 btrfs_end_transaction(trans);
6658 if (drop_inode) {
6659 inode_dec_link_count(inode);
6660 iput(inode);
6661 }
6662 btrfs_btree_balance_dirty(fs_info);
6663 return err;
6664 }
6665
btrfs_mkdir(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode)6666 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6667 struct dentry *dentry, umode_t mode)
6668 {
6669 struct inode *inode;
6670
6671 inode = new_inode(dir->i_sb);
6672 if (!inode)
6673 return -ENOMEM;
6674 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6675 inode->i_op = &btrfs_dir_inode_operations;
6676 inode->i_fop = &btrfs_dir_file_operations;
6677 return btrfs_create_common(dir, dentry, inode);
6678 }
6679
uncompress_inline(struct btrfs_path * path,struct page * page,struct btrfs_file_extent_item * item)6680 static noinline int uncompress_inline(struct btrfs_path *path,
6681 struct page *page,
6682 struct btrfs_file_extent_item *item)
6683 {
6684 int ret;
6685 struct extent_buffer *leaf = path->nodes[0];
6686 char *tmp;
6687 size_t max_size;
6688 unsigned long inline_size;
6689 unsigned long ptr;
6690 int compress_type;
6691
6692 compress_type = btrfs_file_extent_compression(leaf, item);
6693 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6694 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6695 tmp = kmalloc(inline_size, GFP_NOFS);
6696 if (!tmp)
6697 return -ENOMEM;
6698 ptr = btrfs_file_extent_inline_start(item);
6699
6700 read_extent_buffer(leaf, tmp, ptr, inline_size);
6701
6702 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6703 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6704
6705 /*
6706 * decompression code contains a memset to fill in any space between the end
6707 * of the uncompressed data and the end of max_size in case the decompressed
6708 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6709 * the end of an inline extent and the beginning of the next block, so we
6710 * cover that region here.
6711 */
6712
6713 if (max_size < PAGE_SIZE)
6714 memzero_page(page, max_size, PAGE_SIZE - max_size);
6715 kfree(tmp);
6716 return ret;
6717 }
6718
read_inline_extent(struct btrfs_inode * inode,struct btrfs_path * path,struct page * page)6719 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6720 struct page *page)
6721 {
6722 struct btrfs_file_extent_item *fi;
6723 void *kaddr;
6724 size_t copy_size;
6725
6726 if (!page || PageUptodate(page))
6727 return 0;
6728
6729 ASSERT(page_offset(page) == 0);
6730
6731 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6732 struct btrfs_file_extent_item);
6733 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6734 return uncompress_inline(path, page, fi);
6735
6736 copy_size = min_t(u64, PAGE_SIZE,
6737 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6738 kaddr = kmap_local_page(page);
6739 read_extent_buffer(path->nodes[0], kaddr,
6740 btrfs_file_extent_inline_start(fi), copy_size);
6741 kunmap_local(kaddr);
6742 if (copy_size < PAGE_SIZE)
6743 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6744 return 0;
6745 }
6746
6747 /*
6748 * Lookup the first extent overlapping a range in a file.
6749 *
6750 * @inode: file to search in
6751 * @page: page to read extent data into if the extent is inline
6752 * @pg_offset: offset into @page to copy to
6753 * @start: file offset
6754 * @len: length of range starting at @start
6755 *
6756 * Return the first &struct extent_map which overlaps the given range, reading
6757 * it from the B-tree and caching it if necessary. Note that there may be more
6758 * extents which overlap the given range after the returned extent_map.
6759 *
6760 * If @page is not NULL and the extent is inline, this also reads the extent
6761 * data directly into the page and marks the extent up to date in the io_tree.
6762 *
6763 * Return: ERR_PTR on error, non-NULL extent_map on success.
6764 */
btrfs_get_extent(struct btrfs_inode * inode,struct page * page,size_t pg_offset,u64 start,u64 len)6765 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6766 struct page *page, size_t pg_offset,
6767 u64 start, u64 len)
6768 {
6769 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6770 int ret = 0;
6771 u64 extent_start = 0;
6772 u64 extent_end = 0;
6773 u64 objectid = btrfs_ino(inode);
6774 int extent_type = -1;
6775 struct btrfs_path *path = NULL;
6776 struct btrfs_root *root = inode->root;
6777 struct btrfs_file_extent_item *item;
6778 struct extent_buffer *leaf;
6779 struct btrfs_key found_key;
6780 struct extent_map *em = NULL;
6781 struct extent_map_tree *em_tree = &inode->extent_tree;
6782
6783 read_lock(&em_tree->lock);
6784 em = lookup_extent_mapping(em_tree, start, len);
6785 read_unlock(&em_tree->lock);
6786
6787 if (em) {
6788 if (em->start > start || em->start + em->len <= start)
6789 free_extent_map(em);
6790 else if (em->block_start == EXTENT_MAP_INLINE && page)
6791 free_extent_map(em);
6792 else
6793 goto out;
6794 }
6795 em = alloc_extent_map();
6796 if (!em) {
6797 ret = -ENOMEM;
6798 goto out;
6799 }
6800 em->start = EXTENT_MAP_HOLE;
6801 em->orig_start = EXTENT_MAP_HOLE;
6802 em->len = (u64)-1;
6803 em->block_len = (u64)-1;
6804
6805 path = btrfs_alloc_path();
6806 if (!path) {
6807 ret = -ENOMEM;
6808 goto out;
6809 }
6810
6811 /* Chances are we'll be called again, so go ahead and do readahead */
6812 path->reada = READA_FORWARD;
6813
6814 /*
6815 * The same explanation in load_free_space_cache applies here as well,
6816 * we only read when we're loading the free space cache, and at that
6817 * point the commit_root has everything we need.
6818 */
6819 if (btrfs_is_free_space_inode(inode)) {
6820 path->search_commit_root = 1;
6821 path->skip_locking = 1;
6822 }
6823
6824 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6825 if (ret < 0) {
6826 goto out;
6827 } else if (ret > 0) {
6828 if (path->slots[0] == 0)
6829 goto not_found;
6830 path->slots[0]--;
6831 ret = 0;
6832 }
6833
6834 leaf = path->nodes[0];
6835 item = btrfs_item_ptr(leaf, path->slots[0],
6836 struct btrfs_file_extent_item);
6837 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6838 if (found_key.objectid != objectid ||
6839 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6840 /*
6841 * If we backup past the first extent we want to move forward
6842 * and see if there is an extent in front of us, otherwise we'll
6843 * say there is a hole for our whole search range which can
6844 * cause problems.
6845 */
6846 extent_end = start;
6847 goto next;
6848 }
6849
6850 extent_type = btrfs_file_extent_type(leaf, item);
6851 extent_start = found_key.offset;
6852 extent_end = btrfs_file_extent_end(path);
6853 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6854 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6855 /* Only regular file could have regular/prealloc extent */
6856 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6857 ret = -EUCLEAN;
6858 btrfs_crit(fs_info,
6859 "regular/prealloc extent found for non-regular inode %llu",
6860 btrfs_ino(inode));
6861 goto out;
6862 }
6863 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6864 extent_start);
6865 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6866 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6867 path->slots[0],
6868 extent_start);
6869 }
6870 next:
6871 if (start >= extent_end) {
6872 path->slots[0]++;
6873 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6874 ret = btrfs_next_leaf(root, path);
6875 if (ret < 0)
6876 goto out;
6877 else if (ret > 0)
6878 goto not_found;
6879
6880 leaf = path->nodes[0];
6881 }
6882 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6883 if (found_key.objectid != objectid ||
6884 found_key.type != BTRFS_EXTENT_DATA_KEY)
6885 goto not_found;
6886 if (start + len <= found_key.offset)
6887 goto not_found;
6888 if (start > found_key.offset)
6889 goto next;
6890
6891 /* New extent overlaps with existing one */
6892 em->start = start;
6893 em->orig_start = start;
6894 em->len = found_key.offset - start;
6895 em->block_start = EXTENT_MAP_HOLE;
6896 goto insert;
6897 }
6898
6899 btrfs_extent_item_to_extent_map(inode, path, item, em);
6900
6901 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6902 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6903 goto insert;
6904 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6905 /*
6906 * Inline extent can only exist at file offset 0. This is
6907 * ensured by tree-checker and inline extent creation path.
6908 * Thus all members representing file offsets should be zero.
6909 */
6910 ASSERT(pg_offset == 0);
6911 ASSERT(extent_start == 0);
6912 ASSERT(em->start == 0);
6913
6914 /*
6915 * btrfs_extent_item_to_extent_map() should have properly
6916 * initialized em members already.
6917 *
6918 * Other members are not utilized for inline extents.
6919 */
6920 ASSERT(em->block_start == EXTENT_MAP_INLINE);
6921 ASSERT(em->len == fs_info->sectorsize);
6922
6923 ret = read_inline_extent(inode, path, page);
6924 if (ret < 0)
6925 goto out;
6926 goto insert;
6927 }
6928 not_found:
6929 em->start = start;
6930 em->orig_start = start;
6931 em->len = len;
6932 em->block_start = EXTENT_MAP_HOLE;
6933 insert:
6934 ret = 0;
6935 btrfs_release_path(path);
6936 if (em->start > start || extent_map_end(em) <= start) {
6937 btrfs_err(fs_info,
6938 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6939 em->start, em->len, start, len);
6940 ret = -EIO;
6941 goto out;
6942 }
6943
6944 write_lock(&em_tree->lock);
6945 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6946 write_unlock(&em_tree->lock);
6947 out:
6948 btrfs_free_path(path);
6949
6950 trace_btrfs_get_extent(root, inode, em);
6951
6952 if (ret) {
6953 free_extent_map(em);
6954 return ERR_PTR(ret);
6955 }
6956 return em;
6957 }
6958
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)6959 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
6960 struct btrfs_dio_data *dio_data,
6961 const u64 start,
6962 const u64 len,
6963 const u64 orig_start,
6964 const u64 block_start,
6965 const u64 block_len,
6966 const u64 orig_block_len,
6967 const u64 ram_bytes,
6968 const int type)
6969 {
6970 struct extent_map *em = NULL;
6971 struct btrfs_ordered_extent *ordered;
6972
6973 if (type != BTRFS_ORDERED_NOCOW) {
6974 em = create_io_em(inode, start, len, orig_start, block_start,
6975 block_len, orig_block_len, ram_bytes,
6976 BTRFS_COMPRESS_NONE, /* compress_type */
6977 type);
6978 if (IS_ERR(em))
6979 goto out;
6980 }
6981 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
6982 block_start, block_len, 0,
6983 (1 << type) |
6984 (1 << BTRFS_ORDERED_DIRECT),
6985 BTRFS_COMPRESS_NONE);
6986 if (IS_ERR(ordered)) {
6987 if (em) {
6988 free_extent_map(em);
6989 btrfs_drop_extent_map_range(inode, start,
6990 start + len - 1, false);
6991 }
6992 em = ERR_CAST(ordered);
6993 } else {
6994 ASSERT(!dio_data->ordered);
6995 dio_data->ordered = ordered;
6996 }
6997 out:
6998
6999 return em;
7000 }
7001
btrfs_new_extent_direct(struct btrfs_inode * inode,struct btrfs_dio_data * dio_data,u64 start,u64 len)7002 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7003 struct btrfs_dio_data *dio_data,
7004 u64 start, u64 len)
7005 {
7006 struct btrfs_root *root = inode->root;
7007 struct btrfs_fs_info *fs_info = root->fs_info;
7008 struct extent_map *em;
7009 struct btrfs_key ins;
7010 u64 alloc_hint;
7011 int ret;
7012
7013 alloc_hint = get_extent_allocation_hint(inode, start, len);
7014 again:
7015 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7016 0, alloc_hint, &ins, 1, 1);
7017 if (ret == -EAGAIN) {
7018 ASSERT(btrfs_is_zoned(fs_info));
7019 wait_on_bit_io(&inode->root->fs_info->flags, BTRFS_FS_NEED_ZONE_FINISH,
7020 TASK_UNINTERRUPTIBLE);
7021 goto again;
7022 }
7023 if (ret)
7024 return ERR_PTR(ret);
7025
7026 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
7027 ins.objectid, ins.offset, ins.offset,
7028 ins.offset, BTRFS_ORDERED_REGULAR);
7029 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7030 if (IS_ERR(em))
7031 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7032 1);
7033
7034 return em;
7035 }
7036
btrfs_extent_readonly(struct btrfs_fs_info * fs_info,u64 bytenr)7037 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7038 {
7039 struct btrfs_block_group *block_group;
7040 bool readonly = false;
7041
7042 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7043 if (!block_group || block_group->ro)
7044 readonly = true;
7045 if (block_group)
7046 btrfs_put_block_group(block_group);
7047 return readonly;
7048 }
7049
7050 /*
7051 * Check if we can do nocow write into the range [@offset, @offset + @len)
7052 *
7053 * @offset: File offset
7054 * @len: The length to write, will be updated to the nocow writeable
7055 * range
7056 * @orig_start: (optional) Return the original file offset of the file extent
7057 * @orig_len: (optional) Return the original on-disk length of the file extent
7058 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7059 * @strict: if true, omit optimizations that might force us into unnecessary
7060 * cow. e.g., don't trust generation number.
7061 *
7062 * Return:
7063 * >0 and update @len if we can do nocow write
7064 * 0 if we can't do nocow write
7065 * <0 if error happened
7066 *
7067 * NOTE: This only checks the file extents, caller is responsible to wait for
7068 * any ordered extents.
7069 */
can_nocow_extent(struct inode * inode,u64 offset,u64 * len,u64 * orig_start,u64 * orig_block_len,u64 * ram_bytes,bool nowait,bool strict)7070 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7071 u64 *orig_start, u64 *orig_block_len,
7072 u64 *ram_bytes, bool nowait, bool strict)
7073 {
7074 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7075 struct can_nocow_file_extent_args nocow_args = { 0 };
7076 struct btrfs_path *path;
7077 int ret;
7078 struct extent_buffer *leaf;
7079 struct btrfs_root *root = BTRFS_I(inode)->root;
7080 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7081 struct btrfs_file_extent_item *fi;
7082 struct btrfs_key key;
7083 int found_type;
7084
7085 path = btrfs_alloc_path();
7086 if (!path)
7087 return -ENOMEM;
7088 path->nowait = nowait;
7089
7090 ret = btrfs_lookup_file_extent(NULL, root, path,
7091 btrfs_ino(BTRFS_I(inode)), offset, 0);
7092 if (ret < 0)
7093 goto out;
7094
7095 if (ret == 1) {
7096 if (path->slots[0] == 0) {
7097 /* can't find the item, must cow */
7098 ret = 0;
7099 goto out;
7100 }
7101 path->slots[0]--;
7102 }
7103 ret = 0;
7104 leaf = path->nodes[0];
7105 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7106 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7107 key.type != BTRFS_EXTENT_DATA_KEY) {
7108 /* not our file or wrong item type, must cow */
7109 goto out;
7110 }
7111
7112 if (key.offset > offset) {
7113 /* Wrong offset, must cow */
7114 goto out;
7115 }
7116
7117 if (btrfs_file_extent_end(path) <= offset)
7118 goto out;
7119
7120 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7121 found_type = btrfs_file_extent_type(leaf, fi);
7122 if (ram_bytes)
7123 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7124
7125 nocow_args.start = offset;
7126 nocow_args.end = offset + *len - 1;
7127 nocow_args.strict = strict;
7128 nocow_args.free_path = true;
7129
7130 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7131 /* can_nocow_file_extent() has freed the path. */
7132 path = NULL;
7133
7134 if (ret != 1) {
7135 /* Treat errors as not being able to NOCOW. */
7136 ret = 0;
7137 goto out;
7138 }
7139
7140 ret = 0;
7141 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7142 goto out;
7143
7144 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7145 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7146 u64 range_end;
7147
7148 range_end = round_up(offset + nocow_args.num_bytes,
7149 root->fs_info->sectorsize) - 1;
7150 ret = test_range_bit(io_tree, offset, range_end,
7151 EXTENT_DELALLOC, 0, NULL);
7152 if (ret) {
7153 ret = -EAGAIN;
7154 goto out;
7155 }
7156 }
7157
7158 if (orig_start)
7159 *orig_start = key.offset - nocow_args.extent_offset;
7160 if (orig_block_len)
7161 *orig_block_len = nocow_args.disk_num_bytes;
7162
7163 *len = nocow_args.num_bytes;
7164 ret = 1;
7165 out:
7166 btrfs_free_path(path);
7167 return ret;
7168 }
7169
lock_extent_direct(struct inode * inode,u64 lockstart,u64 lockend,struct extent_state ** cached_state,unsigned int iomap_flags)7170 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7171 struct extent_state **cached_state,
7172 unsigned int iomap_flags)
7173 {
7174 const bool writing = (iomap_flags & IOMAP_WRITE);
7175 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7176 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7177 struct btrfs_ordered_extent *ordered;
7178 int ret = 0;
7179
7180 while (1) {
7181 if (nowait) {
7182 if (!try_lock_extent(io_tree, lockstart, lockend,
7183 cached_state))
7184 return -EAGAIN;
7185 } else {
7186 lock_extent(io_tree, lockstart, lockend, cached_state);
7187 }
7188 /*
7189 * We're concerned with the entire range that we're going to be
7190 * doing DIO to, so we need to make sure there's no ordered
7191 * extents in this range.
7192 */
7193 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7194 lockend - lockstart + 1);
7195
7196 /*
7197 * We need to make sure there are no buffered pages in this
7198 * range either, we could have raced between the invalidate in
7199 * generic_file_direct_write and locking the extent. The
7200 * invalidate needs to happen so that reads after a write do not
7201 * get stale data.
7202 */
7203 if (!ordered &&
7204 (!writing || !filemap_range_has_page(inode->i_mapping,
7205 lockstart, lockend)))
7206 break;
7207
7208 unlock_extent(io_tree, lockstart, lockend, cached_state);
7209
7210 if (ordered) {
7211 if (nowait) {
7212 btrfs_put_ordered_extent(ordered);
7213 ret = -EAGAIN;
7214 break;
7215 }
7216 /*
7217 * If we are doing a DIO read and the ordered extent we
7218 * found is for a buffered write, we can not wait for it
7219 * to complete and retry, because if we do so we can
7220 * deadlock with concurrent buffered writes on page
7221 * locks. This happens only if our DIO read covers more
7222 * than one extent map, if at this point has already
7223 * created an ordered extent for a previous extent map
7224 * and locked its range in the inode's io tree, and a
7225 * concurrent write against that previous extent map's
7226 * range and this range started (we unlock the ranges
7227 * in the io tree only when the bios complete and
7228 * buffered writes always lock pages before attempting
7229 * to lock range in the io tree).
7230 */
7231 if (writing ||
7232 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7233 btrfs_start_ordered_extent(ordered);
7234 else
7235 ret = nowait ? -EAGAIN : -ENOTBLK;
7236 btrfs_put_ordered_extent(ordered);
7237 } else {
7238 /*
7239 * We could trigger writeback for this range (and wait
7240 * for it to complete) and then invalidate the pages for
7241 * this range (through invalidate_inode_pages2_range()),
7242 * but that can lead us to a deadlock with a concurrent
7243 * call to readahead (a buffered read or a defrag call
7244 * triggered a readahead) on a page lock due to an
7245 * ordered dio extent we created before but did not have
7246 * yet a corresponding bio submitted (whence it can not
7247 * complete), which makes readahead wait for that
7248 * ordered extent to complete while holding a lock on
7249 * that page.
7250 */
7251 ret = nowait ? -EAGAIN : -ENOTBLK;
7252 }
7253
7254 if (ret)
7255 break;
7256
7257 cond_resched();
7258 }
7259
7260 return ret;
7261 }
7262
7263 /* 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)7264 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7265 u64 len, u64 orig_start, u64 block_start,
7266 u64 block_len, u64 orig_block_len,
7267 u64 ram_bytes, int compress_type,
7268 int type)
7269 {
7270 struct extent_map *em;
7271 int ret;
7272
7273 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7274 type == BTRFS_ORDERED_COMPRESSED ||
7275 type == BTRFS_ORDERED_NOCOW ||
7276 type == BTRFS_ORDERED_REGULAR);
7277
7278 em = alloc_extent_map();
7279 if (!em)
7280 return ERR_PTR(-ENOMEM);
7281
7282 em->start = start;
7283 em->orig_start = orig_start;
7284 em->len = len;
7285 em->block_len = block_len;
7286 em->block_start = block_start;
7287 em->orig_block_len = orig_block_len;
7288 em->ram_bytes = ram_bytes;
7289 em->generation = -1;
7290 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7291 if (type == BTRFS_ORDERED_PREALLOC) {
7292 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7293 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7294 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7295 em->compress_type = compress_type;
7296 }
7297
7298 ret = btrfs_replace_extent_map_range(inode, em, true);
7299 if (ret) {
7300 free_extent_map(em);
7301 return ERR_PTR(ret);
7302 }
7303
7304 /* em got 2 refs now, callers needs to do free_extent_map once. */
7305 return em;
7306 }
7307
7308
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)7309 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7310 struct inode *inode,
7311 struct btrfs_dio_data *dio_data,
7312 u64 start, u64 *lenp,
7313 unsigned int iomap_flags)
7314 {
7315 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7316 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7317 struct extent_map *em = *map;
7318 int type;
7319 u64 block_start, orig_start, orig_block_len, ram_bytes;
7320 struct btrfs_block_group *bg;
7321 bool can_nocow = false;
7322 bool space_reserved = false;
7323 u64 len = *lenp;
7324 u64 prev_len;
7325 int ret = 0;
7326
7327 /*
7328 * We don't allocate a new extent in the following cases
7329 *
7330 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7331 * existing extent.
7332 * 2) The extent is marked as PREALLOC. We're good to go here and can
7333 * just use the extent.
7334 *
7335 */
7336 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7337 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7338 em->block_start != EXTENT_MAP_HOLE)) {
7339 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7340 type = BTRFS_ORDERED_PREALLOC;
7341 else
7342 type = BTRFS_ORDERED_NOCOW;
7343 len = min(len, em->len - (start - em->start));
7344 block_start = em->block_start + (start - em->start);
7345
7346 if (can_nocow_extent(inode, start, &len, &orig_start,
7347 &orig_block_len, &ram_bytes, false, false) == 1) {
7348 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7349 if (bg)
7350 can_nocow = true;
7351 }
7352 }
7353
7354 prev_len = len;
7355 if (can_nocow) {
7356 struct extent_map *em2;
7357
7358 /* We can NOCOW, so only need to reserve metadata space. */
7359 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7360 nowait);
7361 if (ret < 0) {
7362 /* Our caller expects us to free the input extent map. */
7363 free_extent_map(em);
7364 *map = NULL;
7365 btrfs_dec_nocow_writers(bg);
7366 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7367 ret = -EAGAIN;
7368 goto out;
7369 }
7370 space_reserved = true;
7371
7372 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7373 orig_start, block_start,
7374 len, orig_block_len,
7375 ram_bytes, type);
7376 btrfs_dec_nocow_writers(bg);
7377 if (type == BTRFS_ORDERED_PREALLOC) {
7378 free_extent_map(em);
7379 *map = em2;
7380 em = em2;
7381 }
7382
7383 if (IS_ERR(em2)) {
7384 ret = PTR_ERR(em2);
7385 goto out;
7386 }
7387
7388 dio_data->nocow_done = true;
7389 } else {
7390 /* Our caller expects us to free the input extent map. */
7391 free_extent_map(em);
7392 *map = NULL;
7393
7394 if (nowait) {
7395 ret = -EAGAIN;
7396 goto out;
7397 }
7398
7399 /*
7400 * If we could not allocate data space before locking the file
7401 * range and we can't do a NOCOW write, then we have to fail.
7402 */
7403 if (!dio_data->data_space_reserved) {
7404 ret = -ENOSPC;
7405 goto out;
7406 }
7407
7408 /*
7409 * We have to COW and we have already reserved data space before,
7410 * so now we reserve only metadata.
7411 */
7412 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7413 false);
7414 if (ret < 0)
7415 goto out;
7416 space_reserved = true;
7417
7418 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7419 if (IS_ERR(em)) {
7420 ret = PTR_ERR(em);
7421 goto out;
7422 }
7423 *map = em;
7424 len = min(len, em->len - (start - em->start));
7425 if (len < prev_len)
7426 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7427 prev_len - len, true);
7428 }
7429
7430 /*
7431 * We have created our ordered extent, so we can now release our reservation
7432 * for an outstanding extent.
7433 */
7434 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7435
7436 /*
7437 * Need to update the i_size under the extent lock so buffered
7438 * readers will get the updated i_size when we unlock.
7439 */
7440 if (start + len > i_size_read(inode))
7441 i_size_write(inode, start + len);
7442 out:
7443 if (ret && space_reserved) {
7444 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7445 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7446 }
7447 *lenp = len;
7448 return ret;
7449 }
7450
btrfs_dio_iomap_begin(struct inode * inode,loff_t start,loff_t length,unsigned int flags,struct iomap * iomap,struct iomap * srcmap)7451 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7452 loff_t length, unsigned int flags, struct iomap *iomap,
7453 struct iomap *srcmap)
7454 {
7455 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7456 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7457 struct extent_map *em;
7458 struct extent_state *cached_state = NULL;
7459 struct btrfs_dio_data *dio_data = iter->private;
7460 u64 lockstart, lockend;
7461 const bool write = !!(flags & IOMAP_WRITE);
7462 int ret = 0;
7463 u64 len = length;
7464 const u64 data_alloc_len = length;
7465 bool unlock_extents = false;
7466
7467 /*
7468 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7469 * we're NOWAIT we may submit a bio for a partial range and return
7470 * EIOCBQUEUED, which would result in an errant short read.
7471 *
7472 * The best way to handle this would be to allow for partial completions
7473 * of iocb's, so we could submit the partial bio, return and fault in
7474 * the rest of the pages, and then submit the io for the rest of the
7475 * range. However we don't have that currently, so simply return
7476 * -EAGAIN at this point so that the normal path is used.
7477 */
7478 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7479 return -EAGAIN;
7480
7481 /*
7482 * Cap the size of reads to that usually seen in buffered I/O as we need
7483 * to allocate a contiguous array for the checksums.
7484 */
7485 if (!write)
7486 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7487
7488 lockstart = start;
7489 lockend = start + len - 1;
7490
7491 /*
7492 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7493 * enough if we've written compressed pages to this area, so we need to
7494 * flush the dirty pages again to make absolutely sure that any
7495 * outstanding dirty pages are on disk - the first flush only starts
7496 * compression on the data, while keeping the pages locked, so by the
7497 * time the second flush returns we know bios for the compressed pages
7498 * were submitted and finished, and the pages no longer under writeback.
7499 *
7500 * If we have a NOWAIT request and we have any pages in the range that
7501 * are locked, likely due to compression still in progress, we don't want
7502 * to block on page locks. We also don't want to block on pages marked as
7503 * dirty or under writeback (same as for the non-compression case).
7504 * iomap_dio_rw() did the same check, but after that and before we got
7505 * here, mmap'ed writes may have happened or buffered reads started
7506 * (readpage() and readahead(), which lock pages), as we haven't locked
7507 * the file range yet.
7508 */
7509 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7510 &BTRFS_I(inode)->runtime_flags)) {
7511 if (flags & IOMAP_NOWAIT) {
7512 if (filemap_range_needs_writeback(inode->i_mapping,
7513 lockstart, lockend))
7514 return -EAGAIN;
7515 } else {
7516 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7517 start + length - 1);
7518 if (ret)
7519 return ret;
7520 }
7521 }
7522
7523 memset(dio_data, 0, sizeof(*dio_data));
7524
7525 /*
7526 * We always try to allocate data space and must do it before locking
7527 * the file range, to avoid deadlocks with concurrent writes to the same
7528 * range if the range has several extents and the writes don't expand the
7529 * current i_size (the inode lock is taken in shared mode). If we fail to
7530 * allocate data space here we continue and later, after locking the
7531 * file range, we fail with ENOSPC only if we figure out we can not do a
7532 * NOCOW write.
7533 */
7534 if (write && !(flags & IOMAP_NOWAIT)) {
7535 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7536 &dio_data->data_reserved,
7537 start, data_alloc_len, false);
7538 if (!ret)
7539 dio_data->data_space_reserved = true;
7540 else if (ret && !(BTRFS_I(inode)->flags &
7541 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7542 goto err;
7543 }
7544
7545 /*
7546 * If this errors out it's because we couldn't invalidate pagecache for
7547 * this range and we need to fallback to buffered IO, or we are doing a
7548 * NOWAIT read/write and we need to block.
7549 */
7550 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7551 if (ret < 0)
7552 goto err;
7553
7554 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7555 if (IS_ERR(em)) {
7556 ret = PTR_ERR(em);
7557 goto unlock_err;
7558 }
7559
7560 /*
7561 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7562 * io. INLINE is special, and we could probably kludge it in here, but
7563 * it's still buffered so for safety lets just fall back to the generic
7564 * buffered path.
7565 *
7566 * For COMPRESSED we _have_ to read the entire extent in so we can
7567 * decompress it, so there will be buffering required no matter what we
7568 * do, so go ahead and fallback to buffered.
7569 *
7570 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7571 * to buffered IO. Don't blame me, this is the price we pay for using
7572 * the generic code.
7573 */
7574 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7575 em->block_start == EXTENT_MAP_INLINE) {
7576 free_extent_map(em);
7577 /*
7578 * If we are in a NOWAIT context, return -EAGAIN in order to
7579 * fallback to buffered IO. This is not only because we can
7580 * block with buffered IO (no support for NOWAIT semantics at
7581 * the moment) but also to avoid returning short reads to user
7582 * space - this happens if we were able to read some data from
7583 * previous non-compressed extents and then when we fallback to
7584 * buffered IO, at btrfs_file_read_iter() by calling
7585 * filemap_read(), we fail to fault in pages for the read buffer,
7586 * in which case filemap_read() returns a short read (the number
7587 * of bytes previously read is > 0, so it does not return -EFAULT).
7588 */
7589 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7590 goto unlock_err;
7591 }
7592
7593 len = min(len, em->len - (start - em->start));
7594
7595 /*
7596 * If we have a NOWAIT request and the range contains multiple extents
7597 * (or a mix of extents and holes), then we return -EAGAIN to make the
7598 * caller fallback to a context where it can do a blocking (without
7599 * NOWAIT) request. This way we avoid doing partial IO and returning
7600 * success to the caller, which is not optimal for writes and for reads
7601 * it can result in unexpected behaviour for an application.
7602 *
7603 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7604 * iomap_dio_rw(), we can end up returning less data then what the caller
7605 * asked for, resulting in an unexpected, and incorrect, short read.
7606 * That is, the caller asked to read N bytes and we return less than that,
7607 * which is wrong unless we are crossing EOF. This happens if we get a
7608 * page fault error when trying to fault in pages for the buffer that is
7609 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7610 * have previously submitted bios for other extents in the range, in
7611 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7612 * those bios have completed by the time we get the page fault error,
7613 * which we return back to our caller - we should only return EIOCBQUEUED
7614 * after we have submitted bios for all the extents in the range.
7615 */
7616 if ((flags & IOMAP_NOWAIT) && len < length) {
7617 free_extent_map(em);
7618 ret = -EAGAIN;
7619 goto unlock_err;
7620 }
7621
7622 if (write) {
7623 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7624 start, &len, flags);
7625 if (ret < 0)
7626 goto unlock_err;
7627 unlock_extents = true;
7628 /* Recalc len in case the new em is smaller than requested */
7629 len = min(len, em->len - (start - em->start));
7630 if (dio_data->data_space_reserved) {
7631 u64 release_offset;
7632 u64 release_len = 0;
7633
7634 if (dio_data->nocow_done) {
7635 release_offset = start;
7636 release_len = data_alloc_len;
7637 } else if (len < data_alloc_len) {
7638 release_offset = start + len;
7639 release_len = data_alloc_len - len;
7640 }
7641
7642 if (release_len > 0)
7643 btrfs_free_reserved_data_space(BTRFS_I(inode),
7644 dio_data->data_reserved,
7645 release_offset,
7646 release_len);
7647 }
7648 } else {
7649 /*
7650 * We need to unlock only the end area that we aren't using.
7651 * The rest is going to be unlocked by the endio routine.
7652 */
7653 lockstart = start + len;
7654 if (lockstart < lockend)
7655 unlock_extents = true;
7656 }
7657
7658 if (unlock_extents)
7659 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7660 &cached_state);
7661 else
7662 free_extent_state(cached_state);
7663
7664 /*
7665 * Translate extent map information to iomap.
7666 * We trim the extents (and move the addr) even though iomap code does
7667 * that, since we have locked only the parts we are performing I/O in.
7668 */
7669 if ((em->block_start == EXTENT_MAP_HOLE) ||
7670 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7671 iomap->addr = IOMAP_NULL_ADDR;
7672 iomap->type = IOMAP_HOLE;
7673 } else {
7674 iomap->addr = em->block_start + (start - em->start);
7675 iomap->type = IOMAP_MAPPED;
7676 }
7677 iomap->offset = start;
7678 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7679 iomap->length = len;
7680 free_extent_map(em);
7681
7682 return 0;
7683
7684 unlock_err:
7685 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7686 &cached_state);
7687 err:
7688 if (dio_data->data_space_reserved) {
7689 btrfs_free_reserved_data_space(BTRFS_I(inode),
7690 dio_data->data_reserved,
7691 start, data_alloc_len);
7692 extent_changeset_free(dio_data->data_reserved);
7693 }
7694
7695 return ret;
7696 }
7697
btrfs_dio_iomap_end(struct inode * inode,loff_t pos,loff_t length,ssize_t written,unsigned int flags,struct iomap * iomap)7698 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7699 ssize_t written, unsigned int flags, struct iomap *iomap)
7700 {
7701 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7702 struct btrfs_dio_data *dio_data = iter->private;
7703 size_t submitted = dio_data->submitted;
7704 const bool write = !!(flags & IOMAP_WRITE);
7705 int ret = 0;
7706
7707 if (!write && (iomap->type == IOMAP_HOLE)) {
7708 /* If reading from a hole, unlock and return */
7709 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7710 NULL);
7711 return 0;
7712 }
7713
7714 if (submitted < length) {
7715 pos += submitted;
7716 length -= submitted;
7717 if (write)
7718 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7719 pos, length, false);
7720 else
7721 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7722 pos + length - 1, NULL);
7723 ret = -ENOTBLK;
7724 }
7725 if (write) {
7726 btrfs_put_ordered_extent(dio_data->ordered);
7727 dio_data->ordered = NULL;
7728 }
7729
7730 if (write)
7731 extent_changeset_free(dio_data->data_reserved);
7732 return ret;
7733 }
7734
btrfs_dio_end_io(struct btrfs_bio * bbio)7735 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7736 {
7737 struct btrfs_dio_private *dip =
7738 container_of(bbio, struct btrfs_dio_private, bbio);
7739 struct btrfs_inode *inode = bbio->inode;
7740 struct bio *bio = &bbio->bio;
7741
7742 if (bio->bi_status) {
7743 btrfs_warn(inode->root->fs_info,
7744 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7745 btrfs_ino(inode), bio->bi_opf,
7746 dip->file_offset, dip->bytes, bio->bi_status);
7747 }
7748
7749 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7750 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7751 dip->file_offset, dip->bytes,
7752 !bio->bi_status);
7753 } else {
7754 unlock_extent(&inode->io_tree, dip->file_offset,
7755 dip->file_offset + dip->bytes - 1, NULL);
7756 }
7757
7758 bbio->bio.bi_private = bbio->private;
7759 iomap_dio_bio_end_io(bio);
7760 }
7761
btrfs_dio_submit_io(const struct iomap_iter * iter,struct bio * bio,loff_t file_offset)7762 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7763 loff_t file_offset)
7764 {
7765 struct btrfs_bio *bbio = btrfs_bio(bio);
7766 struct btrfs_dio_private *dip =
7767 container_of(bbio, struct btrfs_dio_private, bbio);
7768 struct btrfs_dio_data *dio_data = iter->private;
7769
7770 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7771 btrfs_dio_end_io, bio->bi_private);
7772 bbio->inode = BTRFS_I(iter->inode);
7773 bbio->file_offset = file_offset;
7774
7775 dip->file_offset = file_offset;
7776 dip->bytes = bio->bi_iter.bi_size;
7777
7778 dio_data->submitted += bio->bi_iter.bi_size;
7779
7780 /*
7781 * Check if we are doing a partial write. If we are, we need to split
7782 * the ordered extent to match the submitted bio. Hang on to the
7783 * remaining unfinishable ordered_extent in dio_data so that it can be
7784 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7785 * remaining pages is blocked on the outstanding ordered extent.
7786 */
7787 if (iter->flags & IOMAP_WRITE) {
7788 int ret;
7789
7790 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7791 if (ret) {
7792 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7793 file_offset, dip->bytes,
7794 !ret);
7795 bio->bi_status = errno_to_blk_status(ret);
7796 iomap_dio_bio_end_io(bio);
7797 return;
7798 }
7799 }
7800
7801 btrfs_submit_bio(bbio, 0);
7802 }
7803
7804 static const struct iomap_ops btrfs_dio_iomap_ops = {
7805 .iomap_begin = btrfs_dio_iomap_begin,
7806 .iomap_end = btrfs_dio_iomap_end,
7807 };
7808
7809 static const struct iomap_dio_ops btrfs_dio_ops = {
7810 .submit_io = btrfs_dio_submit_io,
7811 .bio_set = &btrfs_dio_bioset,
7812 };
7813
btrfs_dio_read(struct kiocb * iocb,struct iov_iter * iter,size_t done_before)7814 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7815 {
7816 struct btrfs_dio_data data = { 0 };
7817
7818 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7819 IOMAP_DIO_PARTIAL, &data, done_before);
7820 }
7821
btrfs_dio_write(struct kiocb * iocb,struct iov_iter * iter,size_t done_before)7822 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7823 size_t done_before)
7824 {
7825 struct btrfs_dio_data data = { 0 };
7826
7827 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7828 IOMAP_DIO_PARTIAL, &data, done_before);
7829 }
7830
btrfs_fiemap(struct inode * inode,struct fiemap_extent_info * fieinfo,u64 start,u64 len)7831 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7832 u64 start, u64 len)
7833 {
7834 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
7835 int ret;
7836
7837 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7838 if (ret)
7839 return ret;
7840
7841 /*
7842 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7843 * file range (0 to LLONG_MAX), but that is not enough if we have
7844 * compression enabled. The first filemap_fdatawrite_range() only kicks
7845 * in the compression of data (in an async thread) and will return
7846 * before the compression is done and writeback is started. A second
7847 * filemap_fdatawrite_range() is needed to wait for the compression to
7848 * complete and writeback to start. We also need to wait for ordered
7849 * extents to complete, because our fiemap implementation uses mainly
7850 * file extent items to list the extents, searching for extent maps
7851 * only for file ranges with holes or prealloc extents to figure out
7852 * if we have delalloc in those ranges.
7853 */
7854 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7855 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7856 if (ret)
7857 return ret;
7858 }
7859
7860 btrfs_inode_lock(btrfs_inode, BTRFS_ILOCK_SHARED);
7861
7862 /*
7863 * We did an initial flush to avoid holding the inode's lock while
7864 * triggering writeback and waiting for the completion of IO and ordered
7865 * extents. Now after we locked the inode we do it again, because it's
7866 * possible a new write may have happened in between those two steps.
7867 */
7868 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7869 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7870 if (ret) {
7871 btrfs_inode_unlock(btrfs_inode, BTRFS_ILOCK_SHARED);
7872 return ret;
7873 }
7874 }
7875
7876 ret = extent_fiemap(btrfs_inode, fieinfo, start, len);
7877 btrfs_inode_unlock(btrfs_inode, BTRFS_ILOCK_SHARED);
7878
7879 return ret;
7880 }
7881
btrfs_writepages(struct address_space * mapping,struct writeback_control * wbc)7882 static int btrfs_writepages(struct address_space *mapping,
7883 struct writeback_control *wbc)
7884 {
7885 return extent_writepages(mapping, wbc);
7886 }
7887
btrfs_readahead(struct readahead_control * rac)7888 static void btrfs_readahead(struct readahead_control *rac)
7889 {
7890 extent_readahead(rac);
7891 }
7892
7893 /*
7894 * For release_folio() and invalidate_folio() we have a race window where
7895 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7896 * If we continue to release/invalidate the page, we could cause use-after-free
7897 * for subpage spinlock. So this function is to spin and wait for subpage
7898 * spinlock.
7899 */
wait_subpage_spinlock(struct page * page)7900 static void wait_subpage_spinlock(struct page *page)
7901 {
7902 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7903 struct btrfs_subpage *subpage;
7904
7905 if (!btrfs_is_subpage(fs_info, page))
7906 return;
7907
7908 ASSERT(PagePrivate(page) && page->private);
7909 subpage = (struct btrfs_subpage *)page->private;
7910
7911 /*
7912 * This may look insane as we just acquire the spinlock and release it,
7913 * without doing anything. But we just want to make sure no one is
7914 * still holding the subpage spinlock.
7915 * And since the page is not dirty nor writeback, and we have page
7916 * locked, the only possible way to hold a spinlock is from the endio
7917 * function to clear page writeback.
7918 *
7919 * Here we just acquire the spinlock so that all existing callers
7920 * should exit and we're safe to release/invalidate the page.
7921 */
7922 spin_lock_irq(&subpage->lock);
7923 spin_unlock_irq(&subpage->lock);
7924 }
7925
__btrfs_release_folio(struct folio * folio,gfp_t gfp_flags)7926 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7927 {
7928 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
7929
7930 if (ret == 1) {
7931 wait_subpage_spinlock(&folio->page);
7932 clear_page_extent_mapped(&folio->page);
7933 }
7934 return ret;
7935 }
7936
btrfs_release_folio(struct folio * folio,gfp_t gfp_flags)7937 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7938 {
7939 if (folio_test_writeback(folio) || folio_test_dirty(folio))
7940 return false;
7941 return __btrfs_release_folio(folio, gfp_flags);
7942 }
7943
7944 #ifdef CONFIG_MIGRATION
btrfs_migrate_folio(struct address_space * mapping,struct folio * dst,struct folio * src,enum migrate_mode mode)7945 static int btrfs_migrate_folio(struct address_space *mapping,
7946 struct folio *dst, struct folio *src,
7947 enum migrate_mode mode)
7948 {
7949 int ret = filemap_migrate_folio(mapping, dst, src, mode);
7950
7951 if (ret != MIGRATEPAGE_SUCCESS)
7952 return ret;
7953
7954 if (folio_test_ordered(src)) {
7955 folio_clear_ordered(src);
7956 folio_set_ordered(dst);
7957 }
7958
7959 return MIGRATEPAGE_SUCCESS;
7960 }
7961 #else
7962 #define btrfs_migrate_folio NULL
7963 #endif
7964
btrfs_invalidate_folio(struct folio * folio,size_t offset,size_t length)7965 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
7966 size_t length)
7967 {
7968 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
7969 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7970 struct extent_io_tree *tree = &inode->io_tree;
7971 struct extent_state *cached_state = NULL;
7972 u64 page_start = folio_pos(folio);
7973 u64 page_end = page_start + folio_size(folio) - 1;
7974 u64 cur;
7975 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
7976
7977 /*
7978 * We have folio locked so no new ordered extent can be created on this
7979 * page, nor bio can be submitted for this folio.
7980 *
7981 * But already submitted bio can still be finished on this folio.
7982 * Furthermore, endio function won't skip folio which has Ordered
7983 * (Private2) already cleared, so it's possible for endio and
7984 * invalidate_folio to do the same ordered extent accounting twice
7985 * on one folio.
7986 *
7987 * So here we wait for any submitted bios to finish, so that we won't
7988 * do double ordered extent accounting on the same folio.
7989 */
7990 folio_wait_writeback(folio);
7991 wait_subpage_spinlock(&folio->page);
7992
7993 /*
7994 * For subpage case, we have call sites like
7995 * btrfs_punch_hole_lock_range() which passes range not aligned to
7996 * sectorsize.
7997 * If the range doesn't cover the full folio, we don't need to and
7998 * shouldn't clear page extent mapped, as folio->private can still
7999 * record subpage dirty bits for other part of the range.
8000 *
8001 * For cases that invalidate the full folio even the range doesn't
8002 * cover the full folio, like invalidating the last folio, we're
8003 * still safe to wait for ordered extent to finish.
8004 */
8005 if (!(offset == 0 && length == folio_size(folio))) {
8006 btrfs_release_folio(folio, GFP_NOFS);
8007 return;
8008 }
8009
8010 if (!inode_evicting)
8011 lock_extent(tree, page_start, page_end, &cached_state);
8012
8013 cur = page_start;
8014 while (cur < page_end) {
8015 struct btrfs_ordered_extent *ordered;
8016 u64 range_end;
8017 u32 range_len;
8018 u32 extra_flags = 0;
8019
8020 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8021 page_end + 1 - cur);
8022 if (!ordered) {
8023 range_end = page_end;
8024 /*
8025 * No ordered extent covering this range, we are safe
8026 * to delete all extent states in the range.
8027 */
8028 extra_flags = EXTENT_CLEAR_ALL_BITS;
8029 goto next;
8030 }
8031 if (ordered->file_offset > cur) {
8032 /*
8033 * There is a range between [cur, oe->file_offset) not
8034 * covered by any ordered extent.
8035 * We are safe to delete all extent states, and handle
8036 * the ordered extent in the next iteration.
8037 */
8038 range_end = ordered->file_offset - 1;
8039 extra_flags = EXTENT_CLEAR_ALL_BITS;
8040 goto next;
8041 }
8042
8043 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8044 page_end);
8045 ASSERT(range_end + 1 - cur < U32_MAX);
8046 range_len = range_end + 1 - cur;
8047 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8048 /*
8049 * If Ordered (Private2) is cleared, it means endio has
8050 * already been executed for the range.
8051 * We can't delete the extent states as
8052 * btrfs_finish_ordered_io() may still use some of them.
8053 */
8054 goto next;
8055 }
8056 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8057
8058 /*
8059 * IO on this page will never be started, so we need to account
8060 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8061 * here, must leave that up for the ordered extent completion.
8062 *
8063 * This will also unlock the range for incoming
8064 * btrfs_finish_ordered_io().
8065 */
8066 if (!inode_evicting)
8067 clear_extent_bit(tree, cur, range_end,
8068 EXTENT_DELALLOC |
8069 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8070 EXTENT_DEFRAG, &cached_state);
8071
8072 spin_lock_irq(&inode->ordered_tree.lock);
8073 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8074 ordered->truncated_len = min(ordered->truncated_len,
8075 cur - ordered->file_offset);
8076 spin_unlock_irq(&inode->ordered_tree.lock);
8077
8078 /*
8079 * If the ordered extent has finished, we're safe to delete all
8080 * the extent states of the range, otherwise
8081 * btrfs_finish_ordered_io() will get executed by endio for
8082 * other pages, so we can't delete extent states.
8083 */
8084 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8085 cur, range_end + 1 - cur)) {
8086 btrfs_finish_ordered_io(ordered);
8087 /*
8088 * The ordered extent has finished, now we're again
8089 * safe to delete all extent states of the range.
8090 */
8091 extra_flags = EXTENT_CLEAR_ALL_BITS;
8092 }
8093 next:
8094 if (ordered)
8095 btrfs_put_ordered_extent(ordered);
8096 /*
8097 * Qgroup reserved space handler
8098 * Sector(s) here will be either:
8099 *
8100 * 1) Already written to disk or bio already finished
8101 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8102 * Qgroup will be handled by its qgroup_record then.
8103 * btrfs_qgroup_free_data() call will do nothing here.
8104 *
8105 * 2) Not written to disk yet
8106 * Then btrfs_qgroup_free_data() call will clear the
8107 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8108 * reserved data space.
8109 * Since the IO will never happen for this page.
8110 */
8111 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur, NULL);
8112 if (!inode_evicting) {
8113 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8114 EXTENT_DELALLOC | EXTENT_UPTODATE |
8115 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8116 extra_flags, &cached_state);
8117 }
8118 cur = range_end + 1;
8119 }
8120 /*
8121 * We have iterated through all ordered extents of the page, the page
8122 * should not have Ordered (Private2) anymore, or the above iteration
8123 * did something wrong.
8124 */
8125 ASSERT(!folio_test_ordered(folio));
8126 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8127 if (!inode_evicting)
8128 __btrfs_release_folio(folio, GFP_NOFS);
8129 clear_page_extent_mapped(&folio->page);
8130 }
8131
8132 /*
8133 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8134 * called from a page fault handler when a page is first dirtied. Hence we must
8135 * be careful to check for EOF conditions here. We set the page up correctly
8136 * for a written page which means we get ENOSPC checking when writing into
8137 * holes and correct delalloc and unwritten extent mapping on filesystems that
8138 * support these features.
8139 *
8140 * We are not allowed to take the i_mutex here so we have to play games to
8141 * protect against truncate races as the page could now be beyond EOF. Because
8142 * truncate_setsize() writes the inode size before removing pages, once we have
8143 * the page lock we can determine safely if the page is beyond EOF. If it is not
8144 * beyond EOF, then the page is guaranteed safe against truncation until we
8145 * unlock the page.
8146 */
btrfs_page_mkwrite(struct vm_fault * vmf)8147 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8148 {
8149 struct page *page = vmf->page;
8150 struct inode *inode = file_inode(vmf->vma->vm_file);
8151 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8152 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8153 struct btrfs_ordered_extent *ordered;
8154 struct extent_state *cached_state = NULL;
8155 struct extent_changeset *data_reserved = NULL;
8156 unsigned long zero_start;
8157 loff_t size;
8158 vm_fault_t ret;
8159 int ret2;
8160 int reserved = 0;
8161 u64 reserved_space;
8162 u64 page_start;
8163 u64 page_end;
8164 u64 end;
8165
8166 reserved_space = PAGE_SIZE;
8167
8168 sb_start_pagefault(inode->i_sb);
8169 page_start = page_offset(page);
8170 page_end = page_start + PAGE_SIZE - 1;
8171 end = page_end;
8172
8173 /*
8174 * Reserving delalloc space after obtaining the page lock can lead to
8175 * deadlock. For example, if a dirty page is locked by this function
8176 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8177 * dirty page write out, then the btrfs_writepages() function could
8178 * end up waiting indefinitely to get a lock on the page currently
8179 * being processed by btrfs_page_mkwrite() function.
8180 */
8181 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8182 page_start, reserved_space);
8183 if (!ret2) {
8184 ret2 = file_update_time(vmf->vma->vm_file);
8185 reserved = 1;
8186 }
8187 if (ret2) {
8188 ret = vmf_error(ret2);
8189 if (reserved)
8190 goto out;
8191 goto out_noreserve;
8192 }
8193
8194 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8195 again:
8196 down_read(&BTRFS_I(inode)->i_mmap_lock);
8197 lock_page(page);
8198 size = i_size_read(inode);
8199
8200 if ((page->mapping != inode->i_mapping) ||
8201 (page_start >= size)) {
8202 /* page got truncated out from underneath us */
8203 goto out_unlock;
8204 }
8205 wait_on_page_writeback(page);
8206
8207 lock_extent(io_tree, page_start, page_end, &cached_state);
8208 ret2 = set_page_extent_mapped(page);
8209 if (ret2 < 0) {
8210 ret = vmf_error(ret2);
8211 unlock_extent(io_tree, page_start, page_end, &cached_state);
8212 goto out_unlock;
8213 }
8214
8215 /*
8216 * we can't set the delalloc bits if there are pending ordered
8217 * extents. Drop our locks and wait for them to finish
8218 */
8219 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8220 PAGE_SIZE);
8221 if (ordered) {
8222 unlock_extent(io_tree, page_start, page_end, &cached_state);
8223 unlock_page(page);
8224 up_read(&BTRFS_I(inode)->i_mmap_lock);
8225 btrfs_start_ordered_extent(ordered);
8226 btrfs_put_ordered_extent(ordered);
8227 goto again;
8228 }
8229
8230 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8231 reserved_space = round_up(size - page_start,
8232 fs_info->sectorsize);
8233 if (reserved_space < PAGE_SIZE) {
8234 end = page_start + reserved_space - 1;
8235 btrfs_delalloc_release_space(BTRFS_I(inode),
8236 data_reserved, page_start,
8237 PAGE_SIZE - reserved_space, true);
8238 }
8239 }
8240
8241 /*
8242 * page_mkwrite gets called when the page is firstly dirtied after it's
8243 * faulted in, but write(2) could also dirty a page and set delalloc
8244 * bits, thus in this case for space account reason, we still need to
8245 * clear any delalloc bits within this page range since we have to
8246 * reserve data&meta space before lock_page() (see above comments).
8247 */
8248 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8249 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8250 EXTENT_DEFRAG, &cached_state);
8251
8252 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8253 &cached_state);
8254 if (ret2) {
8255 unlock_extent(io_tree, page_start, page_end, &cached_state);
8256 ret = VM_FAULT_SIGBUS;
8257 goto out_unlock;
8258 }
8259
8260 /* page is wholly or partially inside EOF */
8261 if (page_start + PAGE_SIZE > size)
8262 zero_start = offset_in_page(size);
8263 else
8264 zero_start = PAGE_SIZE;
8265
8266 if (zero_start != PAGE_SIZE)
8267 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8268
8269 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8270 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8271 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8272
8273 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8274
8275 unlock_extent(io_tree, page_start, page_end, &cached_state);
8276 up_read(&BTRFS_I(inode)->i_mmap_lock);
8277
8278 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8279 sb_end_pagefault(inode->i_sb);
8280 extent_changeset_free(data_reserved);
8281 return VM_FAULT_LOCKED;
8282
8283 out_unlock:
8284 unlock_page(page);
8285 up_read(&BTRFS_I(inode)->i_mmap_lock);
8286 out:
8287 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8288 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8289 reserved_space, (ret != 0));
8290 out_noreserve:
8291 sb_end_pagefault(inode->i_sb);
8292 extent_changeset_free(data_reserved);
8293 return ret;
8294 }
8295
btrfs_truncate(struct btrfs_inode * inode,bool skip_writeback)8296 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8297 {
8298 struct btrfs_truncate_control control = {
8299 .inode = inode,
8300 .ino = btrfs_ino(inode),
8301 .min_type = BTRFS_EXTENT_DATA_KEY,
8302 .clear_extent_range = true,
8303 };
8304 struct btrfs_root *root = inode->root;
8305 struct btrfs_fs_info *fs_info = root->fs_info;
8306 struct btrfs_block_rsv *rsv;
8307 int ret;
8308 struct btrfs_trans_handle *trans;
8309 u64 mask = fs_info->sectorsize - 1;
8310 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8311
8312 if (!skip_writeback) {
8313 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8314 inode->vfs_inode.i_size & (~mask),
8315 (u64)-1);
8316 if (ret)
8317 return ret;
8318 }
8319
8320 /*
8321 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8322 * things going on here:
8323 *
8324 * 1) We need to reserve space to update our inode.
8325 *
8326 * 2) We need to have something to cache all the space that is going to
8327 * be free'd up by the truncate operation, but also have some slack
8328 * space reserved in case it uses space during the truncate (thank you
8329 * very much snapshotting).
8330 *
8331 * And we need these to be separate. The fact is we can use a lot of
8332 * space doing the truncate, and we have no earthly idea how much space
8333 * we will use, so we need the truncate reservation to be separate so it
8334 * doesn't end up using space reserved for updating the inode. We also
8335 * need to be able to stop the transaction and start a new one, which
8336 * means we need to be able to update the inode several times, and we
8337 * have no idea of knowing how many times that will be, so we can't just
8338 * reserve 1 item for the entirety of the operation, so that has to be
8339 * done separately as well.
8340 *
8341 * So that leaves us with
8342 *
8343 * 1) rsv - for the truncate reservation, which we will steal from the
8344 * transaction reservation.
8345 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8346 * updating the inode.
8347 */
8348 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8349 if (!rsv)
8350 return -ENOMEM;
8351 rsv->size = min_size;
8352 rsv->failfast = true;
8353
8354 /*
8355 * 1 for the truncate slack space
8356 * 1 for updating the inode.
8357 */
8358 trans = btrfs_start_transaction(root, 2);
8359 if (IS_ERR(trans)) {
8360 ret = PTR_ERR(trans);
8361 goto out;
8362 }
8363
8364 /* Migrate the slack space for the truncate to our reserve */
8365 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8366 min_size, false);
8367 /*
8368 * We have reserved 2 metadata units when we started the transaction and
8369 * min_size matches 1 unit, so this should never fail, but if it does,
8370 * it's not critical we just fail truncation.
8371 */
8372 if (WARN_ON(ret)) {
8373 btrfs_end_transaction(trans);
8374 goto out;
8375 }
8376
8377 trans->block_rsv = rsv;
8378
8379 while (1) {
8380 struct extent_state *cached_state = NULL;
8381 const u64 new_size = inode->vfs_inode.i_size;
8382 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8383
8384 control.new_size = new_size;
8385 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8386 /*
8387 * We want to drop from the next block forward in case this new
8388 * size is not block aligned since we will be keeping the last
8389 * block of the extent just the way it is.
8390 */
8391 btrfs_drop_extent_map_range(inode,
8392 ALIGN(new_size, fs_info->sectorsize),
8393 (u64)-1, false);
8394
8395 ret = btrfs_truncate_inode_items(trans, root, &control);
8396
8397 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8398 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8399
8400 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8401
8402 trans->block_rsv = &fs_info->trans_block_rsv;
8403 if (ret != -ENOSPC && ret != -EAGAIN)
8404 break;
8405
8406 ret = btrfs_update_inode(trans, root, inode);
8407 if (ret)
8408 break;
8409
8410 btrfs_end_transaction(trans);
8411 btrfs_btree_balance_dirty(fs_info);
8412
8413 trans = btrfs_start_transaction(root, 2);
8414 if (IS_ERR(trans)) {
8415 ret = PTR_ERR(trans);
8416 trans = NULL;
8417 break;
8418 }
8419
8420 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8421 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8422 rsv, min_size, false);
8423 /*
8424 * We have reserved 2 metadata units when we started the
8425 * transaction and min_size matches 1 unit, so this should never
8426 * fail, but if it does, it's not critical we just fail truncation.
8427 */
8428 if (WARN_ON(ret))
8429 break;
8430
8431 trans->block_rsv = rsv;
8432 }
8433
8434 /*
8435 * We can't call btrfs_truncate_block inside a trans handle as we could
8436 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8437 * know we've truncated everything except the last little bit, and can
8438 * do btrfs_truncate_block and then update the disk_i_size.
8439 */
8440 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8441 btrfs_end_transaction(trans);
8442 btrfs_btree_balance_dirty(fs_info);
8443
8444 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8445 if (ret)
8446 goto out;
8447 trans = btrfs_start_transaction(root, 1);
8448 if (IS_ERR(trans)) {
8449 ret = PTR_ERR(trans);
8450 goto out;
8451 }
8452 btrfs_inode_safe_disk_i_size_write(inode, 0);
8453 }
8454
8455 if (trans) {
8456 int ret2;
8457
8458 trans->block_rsv = &fs_info->trans_block_rsv;
8459 ret2 = btrfs_update_inode(trans, root, inode);
8460 if (ret2 && !ret)
8461 ret = ret2;
8462
8463 ret2 = btrfs_end_transaction(trans);
8464 if (ret2 && !ret)
8465 ret = ret2;
8466 btrfs_btree_balance_dirty(fs_info);
8467 }
8468 out:
8469 btrfs_free_block_rsv(fs_info, rsv);
8470 /*
8471 * So if we truncate and then write and fsync we normally would just
8472 * write the extents that changed, which is a problem if we need to
8473 * first truncate that entire inode. So set this flag so we write out
8474 * all of the extents in the inode to the sync log so we're completely
8475 * safe.
8476 *
8477 * If no extents were dropped or trimmed we don't need to force the next
8478 * fsync to truncate all the inode's items from the log and re-log them
8479 * all. This means the truncate operation did not change the file size,
8480 * or changed it to a smaller size but there was only an implicit hole
8481 * between the old i_size and the new i_size, and there were no prealloc
8482 * extents beyond i_size to drop.
8483 */
8484 if (control.extents_found > 0)
8485 btrfs_set_inode_full_sync(inode);
8486
8487 return ret;
8488 }
8489
btrfs_new_subvol_inode(struct mnt_idmap * idmap,struct inode * dir)8490 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8491 struct inode *dir)
8492 {
8493 struct inode *inode;
8494
8495 inode = new_inode(dir->i_sb);
8496 if (inode) {
8497 /*
8498 * Subvolumes don't inherit the sgid bit or the parent's gid if
8499 * the parent's sgid bit is set. This is probably a bug.
8500 */
8501 inode_init_owner(idmap, inode, NULL,
8502 S_IFDIR | (~current_umask() & S_IRWXUGO));
8503 inode->i_op = &btrfs_dir_inode_operations;
8504 inode->i_fop = &btrfs_dir_file_operations;
8505 }
8506 return inode;
8507 }
8508
btrfs_alloc_inode(struct super_block * sb)8509 struct inode *btrfs_alloc_inode(struct super_block *sb)
8510 {
8511 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8512 struct btrfs_inode *ei;
8513 struct inode *inode;
8514
8515 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8516 if (!ei)
8517 return NULL;
8518
8519 ei->root = NULL;
8520 ei->generation = 0;
8521 ei->last_trans = 0;
8522 ei->last_sub_trans = 0;
8523 ei->logged_trans = 0;
8524 ei->delalloc_bytes = 0;
8525 ei->new_delalloc_bytes = 0;
8526 ei->defrag_bytes = 0;
8527 ei->disk_i_size = 0;
8528 ei->flags = 0;
8529 ei->ro_flags = 0;
8530 ei->csum_bytes = 0;
8531 ei->index_cnt = (u64)-1;
8532 ei->dir_index = 0;
8533 ei->last_unlink_trans = 0;
8534 ei->last_reflink_trans = 0;
8535 ei->last_log_commit = 0;
8536
8537 spin_lock_init(&ei->lock);
8538 ei->outstanding_extents = 0;
8539 if (sb->s_magic != BTRFS_TEST_MAGIC)
8540 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8541 BTRFS_BLOCK_RSV_DELALLOC);
8542 ei->runtime_flags = 0;
8543 ei->prop_compress = BTRFS_COMPRESS_NONE;
8544 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8545
8546 ei->delayed_node = NULL;
8547
8548 ei->i_otime.tv_sec = 0;
8549 ei->i_otime.tv_nsec = 0;
8550
8551 inode = &ei->vfs_inode;
8552 extent_map_tree_init(&ei->extent_tree);
8553 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8554 ei->io_tree.inode = ei;
8555 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8556 IO_TREE_INODE_FILE_EXTENT);
8557 mutex_init(&ei->log_mutex);
8558 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8559 INIT_LIST_HEAD(&ei->delalloc_inodes);
8560 INIT_LIST_HEAD(&ei->delayed_iput);
8561 RB_CLEAR_NODE(&ei->rb_node);
8562 init_rwsem(&ei->i_mmap_lock);
8563
8564 return inode;
8565 }
8566
8567 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
btrfs_test_destroy_inode(struct inode * inode)8568 void btrfs_test_destroy_inode(struct inode *inode)
8569 {
8570 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8571 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8572 }
8573 #endif
8574
btrfs_free_inode(struct inode * inode)8575 void btrfs_free_inode(struct inode *inode)
8576 {
8577 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8578 }
8579
btrfs_destroy_inode(struct inode * vfs_inode)8580 void btrfs_destroy_inode(struct inode *vfs_inode)
8581 {
8582 struct btrfs_ordered_extent *ordered;
8583 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8584 struct btrfs_root *root = inode->root;
8585 bool freespace_inode;
8586
8587 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8588 WARN_ON(vfs_inode->i_data.nrpages);
8589 WARN_ON(inode->block_rsv.reserved);
8590 WARN_ON(inode->block_rsv.size);
8591 WARN_ON(inode->outstanding_extents);
8592 if (!S_ISDIR(vfs_inode->i_mode)) {
8593 WARN_ON(inode->delalloc_bytes);
8594 WARN_ON(inode->new_delalloc_bytes);
8595 }
8596 WARN_ON(inode->csum_bytes);
8597 WARN_ON(inode->defrag_bytes);
8598
8599 /*
8600 * This can happen where we create an inode, but somebody else also
8601 * created the same inode and we need to destroy the one we already
8602 * created.
8603 */
8604 if (!root)
8605 return;
8606
8607 /*
8608 * If this is a free space inode do not take the ordered extents lockdep
8609 * map.
8610 */
8611 freespace_inode = btrfs_is_free_space_inode(inode);
8612
8613 while (1) {
8614 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8615 if (!ordered)
8616 break;
8617 else {
8618 btrfs_err(root->fs_info,
8619 "found ordered extent %llu %llu on inode cleanup",
8620 ordered->file_offset, ordered->num_bytes);
8621
8622 if (!freespace_inode)
8623 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8624
8625 btrfs_remove_ordered_extent(inode, ordered);
8626 btrfs_put_ordered_extent(ordered);
8627 btrfs_put_ordered_extent(ordered);
8628 }
8629 }
8630 btrfs_qgroup_check_reserved_leak(inode);
8631 inode_tree_del(inode);
8632 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8633 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8634 btrfs_put_root(inode->root);
8635 }
8636
btrfs_drop_inode(struct inode * inode)8637 int btrfs_drop_inode(struct inode *inode)
8638 {
8639 struct btrfs_root *root = BTRFS_I(inode)->root;
8640
8641 if (root == NULL)
8642 return 1;
8643
8644 /* the snap/subvol tree is on deleting */
8645 if (btrfs_root_refs(&root->root_item) == 0)
8646 return 1;
8647 else
8648 return generic_drop_inode(inode);
8649 }
8650
init_once(void * foo)8651 static void init_once(void *foo)
8652 {
8653 struct btrfs_inode *ei = foo;
8654
8655 inode_init_once(&ei->vfs_inode);
8656 }
8657
btrfs_destroy_cachep(void)8658 void __cold btrfs_destroy_cachep(void)
8659 {
8660 /*
8661 * Make sure all delayed rcu free inodes are flushed before we
8662 * destroy cache.
8663 */
8664 rcu_barrier();
8665 bioset_exit(&btrfs_dio_bioset);
8666 kmem_cache_destroy(btrfs_inode_cachep);
8667 }
8668
btrfs_init_cachep(void)8669 int __init btrfs_init_cachep(void)
8670 {
8671 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8672 sizeof(struct btrfs_inode), 0,
8673 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8674 init_once);
8675 if (!btrfs_inode_cachep)
8676 goto fail;
8677
8678 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8679 offsetof(struct btrfs_dio_private, bbio.bio),
8680 BIOSET_NEED_BVECS))
8681 goto fail;
8682
8683 return 0;
8684 fail:
8685 btrfs_destroy_cachep();
8686 return -ENOMEM;
8687 }
8688
btrfs_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int flags)8689 static int btrfs_getattr(struct mnt_idmap *idmap,
8690 const struct path *path, struct kstat *stat,
8691 u32 request_mask, unsigned int flags)
8692 {
8693 u64 delalloc_bytes;
8694 u64 inode_bytes;
8695 struct inode *inode = d_inode(path->dentry);
8696 u32 blocksize = btrfs_sb(inode->i_sb)->sectorsize;
8697 u32 bi_flags = BTRFS_I(inode)->flags;
8698 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8699
8700 stat->result_mask |= STATX_BTIME;
8701 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8702 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8703 if (bi_flags & BTRFS_INODE_APPEND)
8704 stat->attributes |= STATX_ATTR_APPEND;
8705 if (bi_flags & BTRFS_INODE_COMPRESS)
8706 stat->attributes |= STATX_ATTR_COMPRESSED;
8707 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8708 stat->attributes |= STATX_ATTR_IMMUTABLE;
8709 if (bi_flags & BTRFS_INODE_NODUMP)
8710 stat->attributes |= STATX_ATTR_NODUMP;
8711 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8712 stat->attributes |= STATX_ATTR_VERITY;
8713
8714 stat->attributes_mask |= (STATX_ATTR_APPEND |
8715 STATX_ATTR_COMPRESSED |
8716 STATX_ATTR_IMMUTABLE |
8717 STATX_ATTR_NODUMP);
8718
8719 generic_fillattr(idmap, request_mask, inode, stat);
8720 stat->dev = BTRFS_I(inode)->root->anon_dev;
8721
8722 spin_lock(&BTRFS_I(inode)->lock);
8723 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8724 inode_bytes = inode_get_bytes(inode);
8725 spin_unlock(&BTRFS_I(inode)->lock);
8726 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8727 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8728 return 0;
8729 }
8730
btrfs_rename_exchange(struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry)8731 static int btrfs_rename_exchange(struct inode *old_dir,
8732 struct dentry *old_dentry,
8733 struct inode *new_dir,
8734 struct dentry *new_dentry)
8735 {
8736 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8737 struct btrfs_trans_handle *trans;
8738 unsigned int trans_num_items;
8739 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8740 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8741 struct inode *new_inode = new_dentry->d_inode;
8742 struct inode *old_inode = old_dentry->d_inode;
8743 struct btrfs_rename_ctx old_rename_ctx;
8744 struct btrfs_rename_ctx new_rename_ctx;
8745 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8746 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8747 u64 old_idx = 0;
8748 u64 new_idx = 0;
8749 int ret;
8750 int ret2;
8751 bool need_abort = false;
8752 struct fscrypt_name old_fname, new_fname;
8753 struct fscrypt_str *old_name, *new_name;
8754
8755 /*
8756 * For non-subvolumes allow exchange only within one subvolume, in the
8757 * same inode namespace. Two subvolumes (represented as directory) can
8758 * be exchanged as they're a logical link and have a fixed inode number.
8759 */
8760 if (root != dest &&
8761 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8762 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8763 return -EXDEV;
8764
8765 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8766 if (ret)
8767 return ret;
8768
8769 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8770 if (ret) {
8771 fscrypt_free_filename(&old_fname);
8772 return ret;
8773 }
8774
8775 old_name = &old_fname.disk_name;
8776 new_name = &new_fname.disk_name;
8777
8778 /* close the race window with snapshot create/destroy ioctl */
8779 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8780 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8781 down_read(&fs_info->subvol_sem);
8782
8783 /*
8784 * For each inode:
8785 * 1 to remove old dir item
8786 * 1 to remove old dir index
8787 * 1 to add new dir item
8788 * 1 to add new dir index
8789 * 1 to update parent inode
8790 *
8791 * If the parents are the same, we only need to account for one
8792 */
8793 trans_num_items = (old_dir == new_dir ? 9 : 10);
8794 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8795 /*
8796 * 1 to remove old root ref
8797 * 1 to remove old root backref
8798 * 1 to add new root ref
8799 * 1 to add new root backref
8800 */
8801 trans_num_items += 4;
8802 } else {
8803 /*
8804 * 1 to update inode item
8805 * 1 to remove old inode ref
8806 * 1 to add new inode ref
8807 */
8808 trans_num_items += 3;
8809 }
8810 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8811 trans_num_items += 4;
8812 else
8813 trans_num_items += 3;
8814 trans = btrfs_start_transaction(root, trans_num_items);
8815 if (IS_ERR(trans)) {
8816 ret = PTR_ERR(trans);
8817 goto out_notrans;
8818 }
8819
8820 if (dest != root) {
8821 ret = btrfs_record_root_in_trans(trans, dest);
8822 if (ret)
8823 goto out_fail;
8824 }
8825
8826 /*
8827 * We need to find a free sequence number both in the source and
8828 * in the destination directory for the exchange.
8829 */
8830 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8831 if (ret)
8832 goto out_fail;
8833 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8834 if (ret)
8835 goto out_fail;
8836
8837 BTRFS_I(old_inode)->dir_index = 0ULL;
8838 BTRFS_I(new_inode)->dir_index = 0ULL;
8839
8840 /* Reference for the source. */
8841 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8842 /* force full log commit if subvolume involved. */
8843 btrfs_set_log_full_commit(trans);
8844 } else {
8845 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8846 btrfs_ino(BTRFS_I(new_dir)),
8847 old_idx);
8848 if (ret)
8849 goto out_fail;
8850 need_abort = true;
8851 }
8852
8853 /* And now for the dest. */
8854 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8855 /* force full log commit if subvolume involved. */
8856 btrfs_set_log_full_commit(trans);
8857 } else {
8858 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8859 btrfs_ino(BTRFS_I(old_dir)),
8860 new_idx);
8861 if (ret) {
8862 if (need_abort)
8863 btrfs_abort_transaction(trans, ret);
8864 goto out_fail;
8865 }
8866 }
8867
8868 /* Update inode version and ctime/mtime. */
8869 inode_inc_iversion(old_dir);
8870 inode_inc_iversion(new_dir);
8871 inode_inc_iversion(old_inode);
8872 inode_inc_iversion(new_inode);
8873 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
8874
8875 if (old_dentry->d_parent != new_dentry->d_parent) {
8876 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8877 BTRFS_I(old_inode), true);
8878 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8879 BTRFS_I(new_inode), true);
8880 }
8881
8882 /* src is a subvolume */
8883 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8884 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8885 } else { /* src is an inode */
8886 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8887 BTRFS_I(old_dentry->d_inode),
8888 old_name, &old_rename_ctx);
8889 if (!ret)
8890 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
8891 }
8892 if (ret) {
8893 btrfs_abort_transaction(trans, ret);
8894 goto out_fail;
8895 }
8896
8897 /* dest is a subvolume */
8898 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8899 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8900 } else { /* dest is an inode */
8901 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8902 BTRFS_I(new_dentry->d_inode),
8903 new_name, &new_rename_ctx);
8904 if (!ret)
8905 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
8906 }
8907 if (ret) {
8908 btrfs_abort_transaction(trans, ret);
8909 goto out_fail;
8910 }
8911
8912 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8913 new_name, 0, old_idx);
8914 if (ret) {
8915 btrfs_abort_transaction(trans, ret);
8916 goto out_fail;
8917 }
8918
8919 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8920 old_name, 0, new_idx);
8921 if (ret) {
8922 btrfs_abort_transaction(trans, ret);
8923 goto out_fail;
8924 }
8925
8926 if (old_inode->i_nlink == 1)
8927 BTRFS_I(old_inode)->dir_index = old_idx;
8928 if (new_inode->i_nlink == 1)
8929 BTRFS_I(new_inode)->dir_index = new_idx;
8930
8931 /*
8932 * Now pin the logs of the roots. We do it to ensure that no other task
8933 * can sync the logs while we are in progress with the rename, because
8934 * that could result in an inconsistency in case any of the inodes that
8935 * are part of this rename operation were logged before.
8936 */
8937 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8938 btrfs_pin_log_trans(root);
8939 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8940 btrfs_pin_log_trans(dest);
8941
8942 /* Do the log updates for all inodes. */
8943 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8944 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
8945 old_rename_ctx.index, new_dentry->d_parent);
8946 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8947 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
8948 new_rename_ctx.index, old_dentry->d_parent);
8949
8950 /* Now unpin the logs. */
8951 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8952 btrfs_end_log_trans(root);
8953 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8954 btrfs_end_log_trans(dest);
8955 out_fail:
8956 ret2 = btrfs_end_transaction(trans);
8957 ret = ret ? ret : ret2;
8958 out_notrans:
8959 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
8960 old_ino == BTRFS_FIRST_FREE_OBJECTID)
8961 up_read(&fs_info->subvol_sem);
8962
8963 fscrypt_free_filename(&new_fname);
8964 fscrypt_free_filename(&old_fname);
8965 return ret;
8966 }
8967
new_whiteout_inode(struct mnt_idmap * idmap,struct inode * dir)8968 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
8969 struct inode *dir)
8970 {
8971 struct inode *inode;
8972
8973 inode = new_inode(dir->i_sb);
8974 if (inode) {
8975 inode_init_owner(idmap, inode, dir,
8976 S_IFCHR | WHITEOUT_MODE);
8977 inode->i_op = &btrfs_special_inode_operations;
8978 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
8979 }
8980 return inode;
8981 }
8982
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)8983 static int btrfs_rename(struct mnt_idmap *idmap,
8984 struct inode *old_dir, struct dentry *old_dentry,
8985 struct inode *new_dir, struct dentry *new_dentry,
8986 unsigned int flags)
8987 {
8988 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8989 struct btrfs_new_inode_args whiteout_args = {
8990 .dir = old_dir,
8991 .dentry = old_dentry,
8992 };
8993 struct btrfs_trans_handle *trans;
8994 unsigned int trans_num_items;
8995 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8996 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8997 struct inode *new_inode = d_inode(new_dentry);
8998 struct inode *old_inode = d_inode(old_dentry);
8999 struct btrfs_rename_ctx rename_ctx;
9000 u64 index = 0;
9001 int ret;
9002 int ret2;
9003 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9004 struct fscrypt_name old_fname, new_fname;
9005
9006 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9007 return -EPERM;
9008
9009 /* we only allow rename subvolume link between subvolumes */
9010 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9011 return -EXDEV;
9012
9013 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9014 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9015 return -ENOTEMPTY;
9016
9017 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9018 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9019 return -ENOTEMPTY;
9020
9021 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
9022 if (ret)
9023 return ret;
9024
9025 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
9026 if (ret) {
9027 fscrypt_free_filename(&old_fname);
9028 return ret;
9029 }
9030
9031 /* check for collisions, even if the name isn't there */
9032 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
9033 if (ret) {
9034 if (ret == -EEXIST) {
9035 /* we shouldn't get
9036 * eexist without a new_inode */
9037 if (WARN_ON(!new_inode)) {
9038 goto out_fscrypt_names;
9039 }
9040 } else {
9041 /* maybe -EOVERFLOW */
9042 goto out_fscrypt_names;
9043 }
9044 }
9045 ret = 0;
9046
9047 /*
9048 * we're using rename to replace one file with another. Start IO on it
9049 * now so we don't add too much work to the end of the transaction
9050 */
9051 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9052 filemap_flush(old_inode->i_mapping);
9053
9054 if (flags & RENAME_WHITEOUT) {
9055 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9056 if (!whiteout_args.inode) {
9057 ret = -ENOMEM;
9058 goto out_fscrypt_names;
9059 }
9060 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9061 if (ret)
9062 goto out_whiteout_inode;
9063 } else {
9064 /* 1 to update the old parent inode. */
9065 trans_num_items = 1;
9066 }
9067
9068 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9069 /* Close the race window with snapshot create/destroy ioctl */
9070 down_read(&fs_info->subvol_sem);
9071 /*
9072 * 1 to remove old root ref
9073 * 1 to remove old root backref
9074 * 1 to add new root ref
9075 * 1 to add new root backref
9076 */
9077 trans_num_items += 4;
9078 } else {
9079 /*
9080 * 1 to update inode
9081 * 1 to remove old inode ref
9082 * 1 to add new inode ref
9083 */
9084 trans_num_items += 3;
9085 }
9086 /*
9087 * 1 to remove old dir item
9088 * 1 to remove old dir index
9089 * 1 to add new dir item
9090 * 1 to add new dir index
9091 */
9092 trans_num_items += 4;
9093 /* 1 to update new parent inode if it's not the same as the old parent */
9094 if (new_dir != old_dir)
9095 trans_num_items++;
9096 if (new_inode) {
9097 /*
9098 * 1 to update inode
9099 * 1 to remove inode ref
9100 * 1 to remove dir item
9101 * 1 to remove dir index
9102 * 1 to possibly add orphan item
9103 */
9104 trans_num_items += 5;
9105 }
9106 trans = btrfs_start_transaction(root, trans_num_items);
9107 if (IS_ERR(trans)) {
9108 ret = PTR_ERR(trans);
9109 goto out_notrans;
9110 }
9111
9112 if (dest != root) {
9113 ret = btrfs_record_root_in_trans(trans, dest);
9114 if (ret)
9115 goto out_fail;
9116 }
9117
9118 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9119 if (ret)
9120 goto out_fail;
9121
9122 BTRFS_I(old_inode)->dir_index = 0ULL;
9123 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9124 /* force full log commit if subvolume involved. */
9125 btrfs_set_log_full_commit(trans);
9126 } else {
9127 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9128 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9129 index);
9130 if (ret)
9131 goto out_fail;
9132 }
9133
9134 inode_inc_iversion(old_dir);
9135 inode_inc_iversion(new_dir);
9136 inode_inc_iversion(old_inode);
9137 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
9138
9139 if (old_dentry->d_parent != new_dentry->d_parent)
9140 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9141 BTRFS_I(old_inode), true);
9142
9143 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9144 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9145 } else {
9146 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9147 BTRFS_I(d_inode(old_dentry)),
9148 &old_fname.disk_name, &rename_ctx);
9149 if (!ret)
9150 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9151 }
9152 if (ret) {
9153 btrfs_abort_transaction(trans, ret);
9154 goto out_fail;
9155 }
9156
9157 if (new_inode) {
9158 inode_inc_iversion(new_inode);
9159 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9160 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9161 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9162 BUG_ON(new_inode->i_nlink == 0);
9163 } else {
9164 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9165 BTRFS_I(d_inode(new_dentry)),
9166 &new_fname.disk_name);
9167 }
9168 if (!ret && new_inode->i_nlink == 0)
9169 ret = btrfs_orphan_add(trans,
9170 BTRFS_I(d_inode(new_dentry)));
9171 if (ret) {
9172 btrfs_abort_transaction(trans, ret);
9173 goto out_fail;
9174 }
9175 }
9176
9177 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9178 &new_fname.disk_name, 0, index);
9179 if (ret) {
9180 btrfs_abort_transaction(trans, ret);
9181 goto out_fail;
9182 }
9183
9184 if (old_inode->i_nlink == 1)
9185 BTRFS_I(old_inode)->dir_index = index;
9186
9187 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9188 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9189 rename_ctx.index, new_dentry->d_parent);
9190
9191 if (flags & RENAME_WHITEOUT) {
9192 ret = btrfs_create_new_inode(trans, &whiteout_args);
9193 if (ret) {
9194 btrfs_abort_transaction(trans, ret);
9195 goto out_fail;
9196 } else {
9197 unlock_new_inode(whiteout_args.inode);
9198 iput(whiteout_args.inode);
9199 whiteout_args.inode = NULL;
9200 }
9201 }
9202 out_fail:
9203 ret2 = btrfs_end_transaction(trans);
9204 ret = ret ? ret : ret2;
9205 out_notrans:
9206 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9207 up_read(&fs_info->subvol_sem);
9208 if (flags & RENAME_WHITEOUT)
9209 btrfs_new_inode_args_destroy(&whiteout_args);
9210 out_whiteout_inode:
9211 if (flags & RENAME_WHITEOUT)
9212 iput(whiteout_args.inode);
9213 out_fscrypt_names:
9214 fscrypt_free_filename(&old_fname);
9215 fscrypt_free_filename(&new_fname);
9216 return ret;
9217 }
9218
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)9219 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9220 struct dentry *old_dentry, struct inode *new_dir,
9221 struct dentry *new_dentry, unsigned int flags)
9222 {
9223 int ret;
9224
9225 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9226 return -EINVAL;
9227
9228 if (flags & RENAME_EXCHANGE)
9229 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9230 new_dentry);
9231 else
9232 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9233 new_dentry, flags);
9234
9235 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9236
9237 return ret;
9238 }
9239
9240 struct btrfs_delalloc_work {
9241 struct inode *inode;
9242 struct completion completion;
9243 struct list_head list;
9244 struct btrfs_work work;
9245 };
9246
btrfs_run_delalloc_work(struct btrfs_work * work)9247 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9248 {
9249 struct btrfs_delalloc_work *delalloc_work;
9250 struct inode *inode;
9251
9252 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9253 work);
9254 inode = delalloc_work->inode;
9255 filemap_flush(inode->i_mapping);
9256 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9257 &BTRFS_I(inode)->runtime_flags))
9258 filemap_flush(inode->i_mapping);
9259
9260 iput(inode);
9261 complete(&delalloc_work->completion);
9262 }
9263
btrfs_alloc_delalloc_work(struct inode * inode)9264 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9265 {
9266 struct btrfs_delalloc_work *work;
9267
9268 work = kmalloc(sizeof(*work), GFP_NOFS);
9269 if (!work)
9270 return NULL;
9271
9272 init_completion(&work->completion);
9273 INIT_LIST_HEAD(&work->list);
9274 work->inode = inode;
9275 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9276
9277 return work;
9278 }
9279
9280 /*
9281 * some fairly slow code that needs optimization. This walks the list
9282 * of all the inodes with pending delalloc and forces them to disk.
9283 */
start_delalloc_inodes(struct btrfs_root * root,struct writeback_control * wbc,bool snapshot,bool in_reclaim_context)9284 static int start_delalloc_inodes(struct btrfs_root *root,
9285 struct writeback_control *wbc, bool snapshot,
9286 bool in_reclaim_context)
9287 {
9288 struct btrfs_inode *binode;
9289 struct inode *inode;
9290 struct btrfs_delalloc_work *work, *next;
9291 LIST_HEAD(works);
9292 LIST_HEAD(splice);
9293 int ret = 0;
9294 bool full_flush = wbc->nr_to_write == LONG_MAX;
9295
9296 mutex_lock(&root->delalloc_mutex);
9297 spin_lock(&root->delalloc_lock);
9298 list_splice_init(&root->delalloc_inodes, &splice);
9299 while (!list_empty(&splice)) {
9300 binode = list_entry(splice.next, struct btrfs_inode,
9301 delalloc_inodes);
9302
9303 list_move_tail(&binode->delalloc_inodes,
9304 &root->delalloc_inodes);
9305
9306 if (in_reclaim_context &&
9307 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9308 continue;
9309
9310 inode = igrab(&binode->vfs_inode);
9311 if (!inode) {
9312 cond_resched_lock(&root->delalloc_lock);
9313 continue;
9314 }
9315 spin_unlock(&root->delalloc_lock);
9316
9317 if (snapshot)
9318 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9319 &binode->runtime_flags);
9320 if (full_flush) {
9321 work = btrfs_alloc_delalloc_work(inode);
9322 if (!work) {
9323 iput(inode);
9324 ret = -ENOMEM;
9325 goto out;
9326 }
9327 list_add_tail(&work->list, &works);
9328 btrfs_queue_work(root->fs_info->flush_workers,
9329 &work->work);
9330 } else {
9331 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9332 btrfs_add_delayed_iput(BTRFS_I(inode));
9333 if (ret || wbc->nr_to_write <= 0)
9334 goto out;
9335 }
9336 cond_resched();
9337 spin_lock(&root->delalloc_lock);
9338 }
9339 spin_unlock(&root->delalloc_lock);
9340
9341 out:
9342 list_for_each_entry_safe(work, next, &works, list) {
9343 list_del_init(&work->list);
9344 wait_for_completion(&work->completion);
9345 kfree(work);
9346 }
9347
9348 if (!list_empty(&splice)) {
9349 spin_lock(&root->delalloc_lock);
9350 list_splice_tail(&splice, &root->delalloc_inodes);
9351 spin_unlock(&root->delalloc_lock);
9352 }
9353 mutex_unlock(&root->delalloc_mutex);
9354 return ret;
9355 }
9356
btrfs_start_delalloc_snapshot(struct btrfs_root * root,bool in_reclaim_context)9357 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9358 {
9359 struct writeback_control wbc = {
9360 .nr_to_write = LONG_MAX,
9361 .sync_mode = WB_SYNC_NONE,
9362 .range_start = 0,
9363 .range_end = LLONG_MAX,
9364 };
9365 struct btrfs_fs_info *fs_info = root->fs_info;
9366
9367 if (BTRFS_FS_ERROR(fs_info))
9368 return -EROFS;
9369
9370 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9371 }
9372
btrfs_start_delalloc_roots(struct btrfs_fs_info * fs_info,long nr,bool in_reclaim_context)9373 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9374 bool in_reclaim_context)
9375 {
9376 struct writeback_control wbc = {
9377 .nr_to_write = nr,
9378 .sync_mode = WB_SYNC_NONE,
9379 .range_start = 0,
9380 .range_end = LLONG_MAX,
9381 };
9382 struct btrfs_root *root;
9383 LIST_HEAD(splice);
9384 int ret;
9385
9386 if (BTRFS_FS_ERROR(fs_info))
9387 return -EROFS;
9388
9389 mutex_lock(&fs_info->delalloc_root_mutex);
9390 spin_lock(&fs_info->delalloc_root_lock);
9391 list_splice_init(&fs_info->delalloc_roots, &splice);
9392 while (!list_empty(&splice)) {
9393 /*
9394 * Reset nr_to_write here so we know that we're doing a full
9395 * flush.
9396 */
9397 if (nr == LONG_MAX)
9398 wbc.nr_to_write = LONG_MAX;
9399
9400 root = list_first_entry(&splice, struct btrfs_root,
9401 delalloc_root);
9402 root = btrfs_grab_root(root);
9403 BUG_ON(!root);
9404 list_move_tail(&root->delalloc_root,
9405 &fs_info->delalloc_roots);
9406 spin_unlock(&fs_info->delalloc_root_lock);
9407
9408 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9409 btrfs_put_root(root);
9410 if (ret < 0 || wbc.nr_to_write <= 0)
9411 goto out;
9412 spin_lock(&fs_info->delalloc_root_lock);
9413 }
9414 spin_unlock(&fs_info->delalloc_root_lock);
9415
9416 ret = 0;
9417 out:
9418 if (!list_empty(&splice)) {
9419 spin_lock(&fs_info->delalloc_root_lock);
9420 list_splice_tail(&splice, &fs_info->delalloc_roots);
9421 spin_unlock(&fs_info->delalloc_root_lock);
9422 }
9423 mutex_unlock(&fs_info->delalloc_root_mutex);
9424 return ret;
9425 }
9426
btrfs_symlink(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,const char * symname)9427 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9428 struct dentry *dentry, const char *symname)
9429 {
9430 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9431 struct btrfs_trans_handle *trans;
9432 struct btrfs_root *root = BTRFS_I(dir)->root;
9433 struct btrfs_path *path;
9434 struct btrfs_key key;
9435 struct inode *inode;
9436 struct btrfs_new_inode_args new_inode_args = {
9437 .dir = dir,
9438 .dentry = dentry,
9439 };
9440 unsigned int trans_num_items;
9441 int err;
9442 int name_len;
9443 int datasize;
9444 unsigned long ptr;
9445 struct btrfs_file_extent_item *ei;
9446 struct extent_buffer *leaf;
9447
9448 name_len = strlen(symname);
9449 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9450 return -ENAMETOOLONG;
9451
9452 inode = new_inode(dir->i_sb);
9453 if (!inode)
9454 return -ENOMEM;
9455 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9456 inode->i_op = &btrfs_symlink_inode_operations;
9457 inode_nohighmem(inode);
9458 inode->i_mapping->a_ops = &btrfs_aops;
9459 btrfs_i_size_write(BTRFS_I(inode), name_len);
9460 inode_set_bytes(inode, name_len);
9461
9462 new_inode_args.inode = inode;
9463 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9464 if (err)
9465 goto out_inode;
9466 /* 1 additional item for the inline extent */
9467 trans_num_items++;
9468
9469 trans = btrfs_start_transaction(root, trans_num_items);
9470 if (IS_ERR(trans)) {
9471 err = PTR_ERR(trans);
9472 goto out_new_inode_args;
9473 }
9474
9475 err = btrfs_create_new_inode(trans, &new_inode_args);
9476 if (err)
9477 goto out;
9478
9479 path = btrfs_alloc_path();
9480 if (!path) {
9481 err = -ENOMEM;
9482 btrfs_abort_transaction(trans, err);
9483 discard_new_inode(inode);
9484 inode = NULL;
9485 goto out;
9486 }
9487 key.objectid = btrfs_ino(BTRFS_I(inode));
9488 key.offset = 0;
9489 key.type = BTRFS_EXTENT_DATA_KEY;
9490 datasize = btrfs_file_extent_calc_inline_size(name_len);
9491 err = btrfs_insert_empty_item(trans, root, path, &key,
9492 datasize);
9493 if (err) {
9494 btrfs_abort_transaction(trans, err);
9495 btrfs_free_path(path);
9496 discard_new_inode(inode);
9497 inode = NULL;
9498 goto out;
9499 }
9500 leaf = path->nodes[0];
9501 ei = btrfs_item_ptr(leaf, path->slots[0],
9502 struct btrfs_file_extent_item);
9503 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9504 btrfs_set_file_extent_type(leaf, ei,
9505 BTRFS_FILE_EXTENT_INLINE);
9506 btrfs_set_file_extent_encryption(leaf, ei, 0);
9507 btrfs_set_file_extent_compression(leaf, ei, 0);
9508 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9509 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9510
9511 ptr = btrfs_file_extent_inline_start(ei);
9512 write_extent_buffer(leaf, symname, ptr, name_len);
9513 btrfs_mark_buffer_dirty(trans, leaf);
9514 btrfs_free_path(path);
9515
9516 d_instantiate_new(dentry, inode);
9517 err = 0;
9518 out:
9519 btrfs_end_transaction(trans);
9520 btrfs_btree_balance_dirty(fs_info);
9521 out_new_inode_args:
9522 btrfs_new_inode_args_destroy(&new_inode_args);
9523 out_inode:
9524 if (err)
9525 iput(inode);
9526 return err;
9527 }
9528
insert_prealloc_file_extent(struct btrfs_trans_handle * trans_in,struct btrfs_inode * inode,struct btrfs_key * ins,u64 file_offset)9529 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9530 struct btrfs_trans_handle *trans_in,
9531 struct btrfs_inode *inode,
9532 struct btrfs_key *ins,
9533 u64 file_offset)
9534 {
9535 struct btrfs_file_extent_item stack_fi;
9536 struct btrfs_replace_extent_info extent_info;
9537 struct btrfs_trans_handle *trans = trans_in;
9538 struct btrfs_path *path;
9539 u64 start = ins->objectid;
9540 u64 len = ins->offset;
9541 u64 qgroup_released = 0;
9542 int ret;
9543
9544 memset(&stack_fi, 0, sizeof(stack_fi));
9545
9546 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9547 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9548 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9549 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9550 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9551 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9552 /* Encryption and other encoding is reserved and all 0 */
9553
9554 ret = btrfs_qgroup_release_data(inode, file_offset, len, &qgroup_released);
9555 if (ret < 0)
9556 return ERR_PTR(ret);
9557
9558 if (trans) {
9559 ret = insert_reserved_file_extent(trans, inode,
9560 file_offset, &stack_fi,
9561 true, qgroup_released);
9562 if (ret)
9563 goto free_qgroup;
9564 return trans;
9565 }
9566
9567 extent_info.disk_offset = start;
9568 extent_info.disk_len = len;
9569 extent_info.data_offset = 0;
9570 extent_info.data_len = len;
9571 extent_info.file_offset = file_offset;
9572 extent_info.extent_buf = (char *)&stack_fi;
9573 extent_info.is_new_extent = true;
9574 extent_info.update_times = true;
9575 extent_info.qgroup_reserved = qgroup_released;
9576 extent_info.insertions = 0;
9577
9578 path = btrfs_alloc_path();
9579 if (!path) {
9580 ret = -ENOMEM;
9581 goto free_qgroup;
9582 }
9583
9584 ret = btrfs_replace_file_extents(inode, path, file_offset,
9585 file_offset + len - 1, &extent_info,
9586 &trans);
9587 btrfs_free_path(path);
9588 if (ret)
9589 goto free_qgroup;
9590 return trans;
9591
9592 free_qgroup:
9593 /*
9594 * We have released qgroup data range at the beginning of the function,
9595 * and normally qgroup_released bytes will be freed when committing
9596 * transaction.
9597 * But if we error out early, we have to free what we have released
9598 * or we leak qgroup data reservation.
9599 */
9600 btrfs_qgroup_free_refroot(inode->root->fs_info,
9601 inode->root->root_key.objectid, qgroup_released,
9602 BTRFS_QGROUP_RSV_DATA);
9603 return ERR_PTR(ret);
9604 }
9605
__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)9606 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9607 u64 start, u64 num_bytes, u64 min_size,
9608 loff_t actual_len, u64 *alloc_hint,
9609 struct btrfs_trans_handle *trans)
9610 {
9611 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9612 struct extent_map *em;
9613 struct btrfs_root *root = BTRFS_I(inode)->root;
9614 struct btrfs_key ins;
9615 u64 cur_offset = start;
9616 u64 clear_offset = start;
9617 u64 i_size;
9618 u64 cur_bytes;
9619 u64 last_alloc = (u64)-1;
9620 int ret = 0;
9621 bool own_trans = true;
9622 u64 end = start + num_bytes - 1;
9623
9624 if (trans)
9625 own_trans = false;
9626 while (num_bytes > 0) {
9627 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9628 cur_bytes = max(cur_bytes, min_size);
9629 /*
9630 * If we are severely fragmented we could end up with really
9631 * small allocations, so if the allocator is returning small
9632 * chunks lets make its job easier by only searching for those
9633 * sized chunks.
9634 */
9635 cur_bytes = min(cur_bytes, last_alloc);
9636 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9637 min_size, 0, *alloc_hint, &ins, 1, 0);
9638 if (ret)
9639 break;
9640
9641 /*
9642 * We've reserved this space, and thus converted it from
9643 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9644 * from here on out we will only need to clear our reservation
9645 * for the remaining unreserved area, so advance our
9646 * clear_offset by our extent size.
9647 */
9648 clear_offset += ins.offset;
9649
9650 last_alloc = ins.offset;
9651 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9652 &ins, cur_offset);
9653 /*
9654 * Now that we inserted the prealloc extent we can finally
9655 * decrement the number of reservations in the block group.
9656 * If we did it before, we could race with relocation and have
9657 * relocation miss the reserved extent, making it fail later.
9658 */
9659 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9660 if (IS_ERR(trans)) {
9661 ret = PTR_ERR(trans);
9662 btrfs_free_reserved_extent(fs_info, ins.objectid,
9663 ins.offset, 0);
9664 break;
9665 }
9666
9667 em = alloc_extent_map();
9668 if (!em) {
9669 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9670 cur_offset + ins.offset - 1, false);
9671 btrfs_set_inode_full_sync(BTRFS_I(inode));
9672 goto next;
9673 }
9674
9675 em->start = cur_offset;
9676 em->orig_start = cur_offset;
9677 em->len = ins.offset;
9678 em->block_start = ins.objectid;
9679 em->block_len = ins.offset;
9680 em->orig_block_len = ins.offset;
9681 em->ram_bytes = ins.offset;
9682 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9683 em->generation = trans->transid;
9684
9685 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9686 free_extent_map(em);
9687 next:
9688 num_bytes -= ins.offset;
9689 cur_offset += ins.offset;
9690 *alloc_hint = ins.objectid + ins.offset;
9691
9692 inode_inc_iversion(inode);
9693 inode_set_ctime_current(inode);
9694 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9695 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9696 (actual_len > inode->i_size) &&
9697 (cur_offset > inode->i_size)) {
9698 if (cur_offset > actual_len)
9699 i_size = actual_len;
9700 else
9701 i_size = cur_offset;
9702 i_size_write(inode, i_size);
9703 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9704 }
9705
9706 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9707
9708 if (ret) {
9709 btrfs_abort_transaction(trans, ret);
9710 if (own_trans)
9711 btrfs_end_transaction(trans);
9712 break;
9713 }
9714
9715 if (own_trans) {
9716 btrfs_end_transaction(trans);
9717 trans = NULL;
9718 }
9719 }
9720 if (clear_offset < end)
9721 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9722 end - clear_offset + 1);
9723 return ret;
9724 }
9725
btrfs_prealloc_file_range(struct inode * inode,int mode,u64 start,u64 num_bytes,u64 min_size,loff_t actual_len,u64 * alloc_hint)9726 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9727 u64 start, u64 num_bytes, u64 min_size,
9728 loff_t actual_len, u64 *alloc_hint)
9729 {
9730 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9731 min_size, actual_len, alloc_hint,
9732 NULL);
9733 }
9734
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)9735 int btrfs_prealloc_file_range_trans(struct inode *inode,
9736 struct btrfs_trans_handle *trans, int mode,
9737 u64 start, u64 num_bytes, u64 min_size,
9738 loff_t actual_len, u64 *alloc_hint)
9739 {
9740 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9741 min_size, actual_len, alloc_hint, trans);
9742 }
9743
btrfs_permission(struct mnt_idmap * idmap,struct inode * inode,int mask)9744 static int btrfs_permission(struct mnt_idmap *idmap,
9745 struct inode *inode, int mask)
9746 {
9747 struct btrfs_root *root = BTRFS_I(inode)->root;
9748 umode_t mode = inode->i_mode;
9749
9750 if (mask & MAY_WRITE &&
9751 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9752 if (btrfs_root_readonly(root))
9753 return -EROFS;
9754 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9755 return -EACCES;
9756 }
9757 return generic_permission(idmap, inode, mask);
9758 }
9759
btrfs_tmpfile(struct mnt_idmap * idmap,struct inode * dir,struct file * file,umode_t mode)9760 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9761 struct file *file, umode_t mode)
9762 {
9763 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9764 struct btrfs_trans_handle *trans;
9765 struct btrfs_root *root = BTRFS_I(dir)->root;
9766 struct inode *inode;
9767 struct btrfs_new_inode_args new_inode_args = {
9768 .dir = dir,
9769 .dentry = file->f_path.dentry,
9770 .orphan = true,
9771 };
9772 unsigned int trans_num_items;
9773 int ret;
9774
9775 inode = new_inode(dir->i_sb);
9776 if (!inode)
9777 return -ENOMEM;
9778 inode_init_owner(idmap, inode, dir, mode);
9779 inode->i_fop = &btrfs_file_operations;
9780 inode->i_op = &btrfs_file_inode_operations;
9781 inode->i_mapping->a_ops = &btrfs_aops;
9782
9783 new_inode_args.inode = inode;
9784 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9785 if (ret)
9786 goto out_inode;
9787
9788 trans = btrfs_start_transaction(root, trans_num_items);
9789 if (IS_ERR(trans)) {
9790 ret = PTR_ERR(trans);
9791 goto out_new_inode_args;
9792 }
9793
9794 ret = btrfs_create_new_inode(trans, &new_inode_args);
9795
9796 /*
9797 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9798 * set it to 1 because d_tmpfile() will issue a warning if the count is
9799 * 0, through:
9800 *
9801 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9802 */
9803 set_nlink(inode, 1);
9804
9805 if (!ret) {
9806 d_tmpfile(file, inode);
9807 unlock_new_inode(inode);
9808 mark_inode_dirty(inode);
9809 }
9810
9811 btrfs_end_transaction(trans);
9812 btrfs_btree_balance_dirty(fs_info);
9813 out_new_inode_args:
9814 btrfs_new_inode_args_destroy(&new_inode_args);
9815 out_inode:
9816 if (ret)
9817 iput(inode);
9818 return finish_open_simple(file, ret);
9819 }
9820
btrfs_set_range_writeback(struct btrfs_inode * inode,u64 start,u64 end)9821 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9822 {
9823 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9824 unsigned long index = start >> PAGE_SHIFT;
9825 unsigned long end_index = end >> PAGE_SHIFT;
9826 struct page *page;
9827 u32 len;
9828
9829 ASSERT(end + 1 - start <= U32_MAX);
9830 len = end + 1 - start;
9831 while (index <= end_index) {
9832 page = find_get_page(inode->vfs_inode.i_mapping, index);
9833 ASSERT(page); /* Pages should be in the extent_io_tree */
9834
9835 btrfs_page_set_writeback(fs_info, page, start, len);
9836 put_page(page);
9837 index++;
9838 }
9839 }
9840
btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info * fs_info,int compress_type)9841 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9842 int compress_type)
9843 {
9844 switch (compress_type) {
9845 case BTRFS_COMPRESS_NONE:
9846 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9847 case BTRFS_COMPRESS_ZLIB:
9848 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9849 case BTRFS_COMPRESS_LZO:
9850 /*
9851 * The LZO format depends on the sector size. 64K is the maximum
9852 * sector size that we support.
9853 */
9854 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9855 return -EINVAL;
9856 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9857 (fs_info->sectorsize_bits - 12);
9858 case BTRFS_COMPRESS_ZSTD:
9859 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9860 default:
9861 return -EUCLEAN;
9862 }
9863 }
9864
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)9865 static ssize_t btrfs_encoded_read_inline(
9866 struct kiocb *iocb,
9867 struct iov_iter *iter, u64 start,
9868 u64 lockend,
9869 struct extent_state **cached_state,
9870 u64 extent_start, size_t count,
9871 struct btrfs_ioctl_encoded_io_args *encoded,
9872 bool *unlocked)
9873 {
9874 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9875 struct btrfs_root *root = inode->root;
9876 struct btrfs_fs_info *fs_info = root->fs_info;
9877 struct extent_io_tree *io_tree = &inode->io_tree;
9878 struct btrfs_path *path;
9879 struct extent_buffer *leaf;
9880 struct btrfs_file_extent_item *item;
9881 u64 ram_bytes;
9882 unsigned long ptr;
9883 void *tmp;
9884 ssize_t ret;
9885
9886 path = btrfs_alloc_path();
9887 if (!path) {
9888 ret = -ENOMEM;
9889 goto out;
9890 }
9891 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9892 extent_start, 0);
9893 if (ret) {
9894 if (ret > 0) {
9895 /* The extent item disappeared? */
9896 ret = -EIO;
9897 }
9898 goto out;
9899 }
9900 leaf = path->nodes[0];
9901 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9902
9903 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9904 ptr = btrfs_file_extent_inline_start(item);
9905
9906 encoded->len = min_t(u64, extent_start + ram_bytes,
9907 inode->vfs_inode.i_size) - iocb->ki_pos;
9908 ret = btrfs_encoded_io_compression_from_extent(fs_info,
9909 btrfs_file_extent_compression(leaf, item));
9910 if (ret < 0)
9911 goto out;
9912 encoded->compression = ret;
9913 if (encoded->compression) {
9914 size_t inline_size;
9915
9916 inline_size = btrfs_file_extent_inline_item_len(leaf,
9917 path->slots[0]);
9918 if (inline_size > count) {
9919 ret = -ENOBUFS;
9920 goto out;
9921 }
9922 count = inline_size;
9923 encoded->unencoded_len = ram_bytes;
9924 encoded->unencoded_offset = iocb->ki_pos - extent_start;
9925 } else {
9926 count = min_t(u64, count, encoded->len);
9927 encoded->len = count;
9928 encoded->unencoded_len = count;
9929 ptr += iocb->ki_pos - extent_start;
9930 }
9931
9932 tmp = kmalloc(count, GFP_NOFS);
9933 if (!tmp) {
9934 ret = -ENOMEM;
9935 goto out;
9936 }
9937 read_extent_buffer(leaf, tmp, ptr, count);
9938 btrfs_release_path(path);
9939 unlock_extent(io_tree, start, lockend, cached_state);
9940 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
9941 *unlocked = true;
9942
9943 ret = copy_to_iter(tmp, count, iter);
9944 if (ret != count)
9945 ret = -EFAULT;
9946 kfree(tmp);
9947 out:
9948 btrfs_free_path(path);
9949 return ret;
9950 }
9951
9952 struct btrfs_encoded_read_private {
9953 wait_queue_head_t wait;
9954 atomic_t pending;
9955 blk_status_t status;
9956 };
9957
btrfs_encoded_read_endio(struct btrfs_bio * bbio)9958 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
9959 {
9960 struct btrfs_encoded_read_private *priv = bbio->private;
9961
9962 if (bbio->bio.bi_status) {
9963 /*
9964 * The memory barrier implied by the atomic_dec_return() here
9965 * pairs with the memory barrier implied by the
9966 * atomic_dec_return() or io_wait_event() in
9967 * btrfs_encoded_read_regular_fill_pages() to ensure that this
9968 * write is observed before the load of status in
9969 * btrfs_encoded_read_regular_fill_pages().
9970 */
9971 WRITE_ONCE(priv->status, bbio->bio.bi_status);
9972 }
9973 if (!atomic_dec_return(&priv->pending))
9974 wake_up(&priv->wait);
9975 bio_put(&bbio->bio);
9976 }
9977
btrfs_encoded_read_regular_fill_pages(struct btrfs_inode * inode,u64 file_offset,u64 disk_bytenr,u64 disk_io_size,struct page ** pages)9978 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
9979 u64 file_offset, u64 disk_bytenr,
9980 u64 disk_io_size, struct page **pages)
9981 {
9982 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9983 struct btrfs_encoded_read_private priv = {
9984 .pending = ATOMIC_INIT(1),
9985 };
9986 unsigned long i = 0;
9987 struct btrfs_bio *bbio;
9988
9989 init_waitqueue_head(&priv.wait);
9990
9991 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9992 btrfs_encoded_read_endio, &priv);
9993 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9994 bbio->inode = inode;
9995
9996 do {
9997 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
9998
9999 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
10000 atomic_inc(&priv.pending);
10001 btrfs_submit_bio(bbio, 0);
10002
10003 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10004 btrfs_encoded_read_endio, &priv);
10005 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10006 bbio->inode = inode;
10007 continue;
10008 }
10009
10010 i++;
10011 disk_bytenr += bytes;
10012 disk_io_size -= bytes;
10013 } while (disk_io_size);
10014
10015 atomic_inc(&priv.pending);
10016 btrfs_submit_bio(bbio, 0);
10017
10018 if (atomic_dec_return(&priv.pending))
10019 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10020 /* See btrfs_encoded_read_endio() for ordering. */
10021 return blk_status_to_errno(READ_ONCE(priv.status));
10022 }
10023
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)10024 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10025 struct iov_iter *iter,
10026 u64 start, u64 lockend,
10027 struct extent_state **cached_state,
10028 u64 disk_bytenr, u64 disk_io_size,
10029 size_t count, bool compressed,
10030 bool *unlocked)
10031 {
10032 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10033 struct extent_io_tree *io_tree = &inode->io_tree;
10034 struct page **pages;
10035 unsigned long nr_pages, i;
10036 u64 cur;
10037 size_t page_offset;
10038 ssize_t ret;
10039
10040 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10041 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10042 if (!pages)
10043 return -ENOMEM;
10044 ret = btrfs_alloc_page_array(nr_pages, pages);
10045 if (ret) {
10046 ret = -ENOMEM;
10047 goto out;
10048 }
10049
10050 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10051 disk_io_size, pages);
10052 if (ret)
10053 goto out;
10054
10055 unlock_extent(io_tree, start, lockend, cached_state);
10056 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10057 *unlocked = true;
10058
10059 if (compressed) {
10060 i = 0;
10061 page_offset = 0;
10062 } else {
10063 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10064 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10065 }
10066 cur = 0;
10067 while (cur < count) {
10068 size_t bytes = min_t(size_t, count - cur,
10069 PAGE_SIZE - page_offset);
10070
10071 if (copy_page_to_iter(pages[i], page_offset, bytes,
10072 iter) != bytes) {
10073 ret = -EFAULT;
10074 goto out;
10075 }
10076 i++;
10077 cur += bytes;
10078 page_offset = 0;
10079 }
10080 ret = count;
10081 out:
10082 for (i = 0; i < nr_pages; i++) {
10083 if (pages[i])
10084 __free_page(pages[i]);
10085 }
10086 kfree(pages);
10087 return ret;
10088 }
10089
btrfs_encoded_read(struct kiocb * iocb,struct iov_iter * iter,struct btrfs_ioctl_encoded_io_args * encoded)10090 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10091 struct btrfs_ioctl_encoded_io_args *encoded)
10092 {
10093 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10094 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10095 struct extent_io_tree *io_tree = &inode->io_tree;
10096 ssize_t ret;
10097 size_t count = iov_iter_count(iter);
10098 u64 start, lockend, disk_bytenr, disk_io_size;
10099 struct extent_state *cached_state = NULL;
10100 struct extent_map *em;
10101 bool unlocked = false;
10102
10103 file_accessed(iocb->ki_filp);
10104
10105 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10106
10107 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10108 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10109 return 0;
10110 }
10111 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10112 /*
10113 * We don't know how long the extent containing iocb->ki_pos is, but if
10114 * it's compressed we know that it won't be longer than this.
10115 */
10116 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10117
10118 for (;;) {
10119 struct btrfs_ordered_extent *ordered;
10120
10121 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10122 lockend - start + 1);
10123 if (ret)
10124 goto out_unlock_inode;
10125 lock_extent(io_tree, start, lockend, &cached_state);
10126 ordered = btrfs_lookup_ordered_range(inode, start,
10127 lockend - start + 1);
10128 if (!ordered)
10129 break;
10130 btrfs_put_ordered_extent(ordered);
10131 unlock_extent(io_tree, start, lockend, &cached_state);
10132 cond_resched();
10133 }
10134
10135 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10136 if (IS_ERR(em)) {
10137 ret = PTR_ERR(em);
10138 goto out_unlock_extent;
10139 }
10140
10141 if (em->block_start == EXTENT_MAP_INLINE) {
10142 u64 extent_start = em->start;
10143
10144 /*
10145 * For inline extents we get everything we need out of the
10146 * extent item.
10147 */
10148 free_extent_map(em);
10149 em = NULL;
10150 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10151 &cached_state, extent_start,
10152 count, encoded, &unlocked);
10153 goto out;
10154 }
10155
10156 /*
10157 * We only want to return up to EOF even if the extent extends beyond
10158 * that.
10159 */
10160 encoded->len = min_t(u64, extent_map_end(em),
10161 inode->vfs_inode.i_size) - iocb->ki_pos;
10162 if (em->block_start == EXTENT_MAP_HOLE ||
10163 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10164 disk_bytenr = EXTENT_MAP_HOLE;
10165 count = min_t(u64, count, encoded->len);
10166 encoded->len = count;
10167 encoded->unencoded_len = count;
10168 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10169 disk_bytenr = em->block_start;
10170 /*
10171 * Bail if the buffer isn't large enough to return the whole
10172 * compressed extent.
10173 */
10174 if (em->block_len > count) {
10175 ret = -ENOBUFS;
10176 goto out_em;
10177 }
10178 disk_io_size = em->block_len;
10179 count = em->block_len;
10180 encoded->unencoded_len = em->ram_bytes;
10181 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10182 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10183 em->compress_type);
10184 if (ret < 0)
10185 goto out_em;
10186 encoded->compression = ret;
10187 } else {
10188 disk_bytenr = em->block_start + (start - em->start);
10189 if (encoded->len > count)
10190 encoded->len = count;
10191 /*
10192 * Don't read beyond what we locked. This also limits the page
10193 * allocations that we'll do.
10194 */
10195 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10196 count = start + disk_io_size - iocb->ki_pos;
10197 encoded->len = count;
10198 encoded->unencoded_len = count;
10199 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10200 }
10201 free_extent_map(em);
10202 em = NULL;
10203
10204 if (disk_bytenr == EXTENT_MAP_HOLE) {
10205 unlock_extent(io_tree, start, lockend, &cached_state);
10206 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10207 unlocked = true;
10208 ret = iov_iter_zero(count, iter);
10209 if (ret != count)
10210 ret = -EFAULT;
10211 } else {
10212 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10213 &cached_state, disk_bytenr,
10214 disk_io_size, count,
10215 encoded->compression,
10216 &unlocked);
10217 }
10218
10219 out:
10220 if (ret >= 0)
10221 iocb->ki_pos += encoded->len;
10222 out_em:
10223 free_extent_map(em);
10224 out_unlock_extent:
10225 if (!unlocked)
10226 unlock_extent(io_tree, start, lockend, &cached_state);
10227 out_unlock_inode:
10228 if (!unlocked)
10229 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10230 return ret;
10231 }
10232
btrfs_do_encoded_write(struct kiocb * iocb,struct iov_iter * from,const struct btrfs_ioctl_encoded_io_args * encoded)10233 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10234 const struct btrfs_ioctl_encoded_io_args *encoded)
10235 {
10236 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10237 struct btrfs_root *root = inode->root;
10238 struct btrfs_fs_info *fs_info = root->fs_info;
10239 struct extent_io_tree *io_tree = &inode->io_tree;
10240 struct extent_changeset *data_reserved = NULL;
10241 struct extent_state *cached_state = NULL;
10242 struct btrfs_ordered_extent *ordered;
10243 int compression;
10244 size_t orig_count;
10245 u64 start, end;
10246 u64 num_bytes, ram_bytes, disk_num_bytes;
10247 unsigned long nr_pages, i;
10248 struct page **pages;
10249 struct btrfs_key ins;
10250 bool extent_reserved = false;
10251 struct extent_map *em;
10252 ssize_t ret;
10253
10254 switch (encoded->compression) {
10255 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10256 compression = BTRFS_COMPRESS_ZLIB;
10257 break;
10258 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10259 compression = BTRFS_COMPRESS_ZSTD;
10260 break;
10261 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10262 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10263 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10264 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10265 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10266 /* The sector size must match for LZO. */
10267 if (encoded->compression -
10268 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10269 fs_info->sectorsize_bits)
10270 return -EINVAL;
10271 compression = BTRFS_COMPRESS_LZO;
10272 break;
10273 default:
10274 return -EINVAL;
10275 }
10276 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10277 return -EINVAL;
10278
10279 /*
10280 * Compressed extents should always have checksums, so error out if we
10281 * have a NOCOW file or inode was created while mounted with NODATASUM.
10282 */
10283 if (inode->flags & BTRFS_INODE_NODATASUM)
10284 return -EINVAL;
10285
10286 orig_count = iov_iter_count(from);
10287
10288 /* The extent size must be sane. */
10289 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10290 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10291 return -EINVAL;
10292
10293 /*
10294 * The compressed data must be smaller than the decompressed data.
10295 *
10296 * It's of course possible for data to compress to larger or the same
10297 * size, but the buffered I/O path falls back to no compression for such
10298 * data, and we don't want to break any assumptions by creating these
10299 * extents.
10300 *
10301 * Note that this is less strict than the current check we have that the
10302 * compressed data must be at least one sector smaller than the
10303 * decompressed data. We only want to enforce the weaker requirement
10304 * from old kernels that it is at least one byte smaller.
10305 */
10306 if (orig_count >= encoded->unencoded_len)
10307 return -EINVAL;
10308
10309 /* The extent must start on a sector boundary. */
10310 start = iocb->ki_pos;
10311 if (!IS_ALIGNED(start, fs_info->sectorsize))
10312 return -EINVAL;
10313
10314 /*
10315 * The extent must end on a sector boundary. However, we allow a write
10316 * which ends at or extends i_size to have an unaligned length; we round
10317 * up the extent size and set i_size to the unaligned end.
10318 */
10319 if (start + encoded->len < inode->vfs_inode.i_size &&
10320 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10321 return -EINVAL;
10322
10323 /* Finally, the offset in the unencoded data must be sector-aligned. */
10324 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10325 return -EINVAL;
10326
10327 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10328 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10329 end = start + num_bytes - 1;
10330
10331 /*
10332 * If the extent cannot be inline, the compressed data on disk must be
10333 * sector-aligned. For convenience, we extend it with zeroes if it
10334 * isn't.
10335 */
10336 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10337 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10338 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10339 if (!pages)
10340 return -ENOMEM;
10341 for (i = 0; i < nr_pages; i++) {
10342 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10343 char *kaddr;
10344
10345 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10346 if (!pages[i]) {
10347 ret = -ENOMEM;
10348 goto out_pages;
10349 }
10350 kaddr = kmap_local_page(pages[i]);
10351 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10352 kunmap_local(kaddr);
10353 ret = -EFAULT;
10354 goto out_pages;
10355 }
10356 if (bytes < PAGE_SIZE)
10357 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10358 kunmap_local(kaddr);
10359 }
10360
10361 for (;;) {
10362 struct btrfs_ordered_extent *ordered;
10363
10364 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10365 if (ret)
10366 goto out_pages;
10367 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10368 start >> PAGE_SHIFT,
10369 end >> PAGE_SHIFT);
10370 if (ret)
10371 goto out_pages;
10372 lock_extent(io_tree, start, end, &cached_state);
10373 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10374 if (!ordered &&
10375 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10376 break;
10377 if (ordered)
10378 btrfs_put_ordered_extent(ordered);
10379 unlock_extent(io_tree, start, end, &cached_state);
10380 cond_resched();
10381 }
10382
10383 /*
10384 * We don't use the higher-level delalloc space functions because our
10385 * num_bytes and disk_num_bytes are different.
10386 */
10387 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10388 if (ret)
10389 goto out_unlock;
10390 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10391 if (ret)
10392 goto out_free_data_space;
10393 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10394 false);
10395 if (ret)
10396 goto out_qgroup_free_data;
10397
10398 /* Try an inline extent first. */
10399 if (start == 0 && encoded->unencoded_len == encoded->len &&
10400 encoded->unencoded_offset == 0) {
10401 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10402 compression, pages, true);
10403 if (ret <= 0) {
10404 if (ret == 0)
10405 ret = orig_count;
10406 goto out_delalloc_release;
10407 }
10408 }
10409
10410 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10411 disk_num_bytes, 0, 0, &ins, 1, 1);
10412 if (ret)
10413 goto out_delalloc_release;
10414 extent_reserved = true;
10415
10416 em = create_io_em(inode, start, num_bytes,
10417 start - encoded->unencoded_offset, ins.objectid,
10418 ins.offset, ins.offset, ram_bytes, compression,
10419 BTRFS_ORDERED_COMPRESSED);
10420 if (IS_ERR(em)) {
10421 ret = PTR_ERR(em);
10422 goto out_free_reserved;
10423 }
10424 free_extent_map(em);
10425
10426 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10427 ins.objectid, ins.offset,
10428 encoded->unencoded_offset,
10429 (1 << BTRFS_ORDERED_ENCODED) |
10430 (1 << BTRFS_ORDERED_COMPRESSED),
10431 compression);
10432 if (IS_ERR(ordered)) {
10433 btrfs_drop_extent_map_range(inode, start, end, false);
10434 ret = PTR_ERR(ordered);
10435 goto out_free_reserved;
10436 }
10437 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10438
10439 if (start + encoded->len > inode->vfs_inode.i_size)
10440 i_size_write(&inode->vfs_inode, start + encoded->len);
10441
10442 unlock_extent(io_tree, start, end, &cached_state);
10443
10444 btrfs_delalloc_release_extents(inode, num_bytes);
10445
10446 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10447 ret = orig_count;
10448 goto out;
10449
10450 out_free_reserved:
10451 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10452 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10453 out_delalloc_release:
10454 btrfs_delalloc_release_extents(inode, num_bytes);
10455 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10456 out_qgroup_free_data:
10457 if (ret < 0)
10458 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes, NULL);
10459 out_free_data_space:
10460 /*
10461 * If btrfs_reserve_extent() succeeded, then we already decremented
10462 * bytes_may_use.
10463 */
10464 if (!extent_reserved)
10465 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10466 out_unlock:
10467 unlock_extent(io_tree, start, end, &cached_state);
10468 out_pages:
10469 for (i = 0; i < nr_pages; i++) {
10470 if (pages[i])
10471 __free_page(pages[i]);
10472 }
10473 kvfree(pages);
10474 out:
10475 if (ret >= 0)
10476 iocb->ki_pos += encoded->len;
10477 return ret;
10478 }
10479
10480 #ifdef CONFIG_SWAP
10481 /*
10482 * Add an entry indicating a block group or device which is pinned by a
10483 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10484 * negative errno on failure.
10485 */
btrfs_add_swapfile_pin(struct inode * inode,void * ptr,bool is_block_group)10486 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10487 bool is_block_group)
10488 {
10489 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10490 struct btrfs_swapfile_pin *sp, *entry;
10491 struct rb_node **p;
10492 struct rb_node *parent = NULL;
10493
10494 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10495 if (!sp)
10496 return -ENOMEM;
10497 sp->ptr = ptr;
10498 sp->inode = inode;
10499 sp->is_block_group = is_block_group;
10500 sp->bg_extent_count = 1;
10501
10502 spin_lock(&fs_info->swapfile_pins_lock);
10503 p = &fs_info->swapfile_pins.rb_node;
10504 while (*p) {
10505 parent = *p;
10506 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10507 if (sp->ptr < entry->ptr ||
10508 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10509 p = &(*p)->rb_left;
10510 } else if (sp->ptr > entry->ptr ||
10511 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10512 p = &(*p)->rb_right;
10513 } else {
10514 if (is_block_group)
10515 entry->bg_extent_count++;
10516 spin_unlock(&fs_info->swapfile_pins_lock);
10517 kfree(sp);
10518 return 1;
10519 }
10520 }
10521 rb_link_node(&sp->node, parent, p);
10522 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10523 spin_unlock(&fs_info->swapfile_pins_lock);
10524 return 0;
10525 }
10526
10527 /* Free all of the entries pinned by this swapfile. */
btrfs_free_swapfile_pins(struct inode * inode)10528 static void btrfs_free_swapfile_pins(struct inode *inode)
10529 {
10530 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10531 struct btrfs_swapfile_pin *sp;
10532 struct rb_node *node, *next;
10533
10534 spin_lock(&fs_info->swapfile_pins_lock);
10535 node = rb_first(&fs_info->swapfile_pins);
10536 while (node) {
10537 next = rb_next(node);
10538 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10539 if (sp->inode == inode) {
10540 rb_erase(&sp->node, &fs_info->swapfile_pins);
10541 if (sp->is_block_group) {
10542 btrfs_dec_block_group_swap_extents(sp->ptr,
10543 sp->bg_extent_count);
10544 btrfs_put_block_group(sp->ptr);
10545 }
10546 kfree(sp);
10547 }
10548 node = next;
10549 }
10550 spin_unlock(&fs_info->swapfile_pins_lock);
10551 }
10552
10553 struct btrfs_swap_info {
10554 u64 start;
10555 u64 block_start;
10556 u64 block_len;
10557 u64 lowest_ppage;
10558 u64 highest_ppage;
10559 unsigned long nr_pages;
10560 int nr_extents;
10561 };
10562
btrfs_add_swap_extent(struct swap_info_struct * sis,struct btrfs_swap_info * bsi)10563 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10564 struct btrfs_swap_info *bsi)
10565 {
10566 unsigned long nr_pages;
10567 unsigned long max_pages;
10568 u64 first_ppage, first_ppage_reported, next_ppage;
10569 int ret;
10570
10571 /*
10572 * Our swapfile may have had its size extended after the swap header was
10573 * written. In that case activating the swapfile should not go beyond
10574 * the max size set in the swap header.
10575 */
10576 if (bsi->nr_pages >= sis->max)
10577 return 0;
10578
10579 max_pages = sis->max - bsi->nr_pages;
10580 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10581 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10582
10583 if (first_ppage >= next_ppage)
10584 return 0;
10585 nr_pages = next_ppage - first_ppage;
10586 nr_pages = min(nr_pages, max_pages);
10587
10588 first_ppage_reported = first_ppage;
10589 if (bsi->start == 0)
10590 first_ppage_reported++;
10591 if (bsi->lowest_ppage > first_ppage_reported)
10592 bsi->lowest_ppage = first_ppage_reported;
10593 if (bsi->highest_ppage < (next_ppage - 1))
10594 bsi->highest_ppage = next_ppage - 1;
10595
10596 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10597 if (ret < 0)
10598 return ret;
10599 bsi->nr_extents += ret;
10600 bsi->nr_pages += nr_pages;
10601 return 0;
10602 }
10603
btrfs_swap_deactivate(struct file * file)10604 static void btrfs_swap_deactivate(struct file *file)
10605 {
10606 struct inode *inode = file_inode(file);
10607
10608 btrfs_free_swapfile_pins(inode);
10609 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10610 }
10611
btrfs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)10612 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10613 sector_t *span)
10614 {
10615 struct inode *inode = file_inode(file);
10616 struct btrfs_root *root = BTRFS_I(inode)->root;
10617 struct btrfs_fs_info *fs_info = root->fs_info;
10618 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10619 struct extent_state *cached_state = NULL;
10620 struct extent_map *em = NULL;
10621 struct btrfs_device *device = NULL;
10622 struct btrfs_swap_info bsi = {
10623 .lowest_ppage = (sector_t)-1ULL,
10624 };
10625 int ret = 0;
10626 u64 isize;
10627 u64 start;
10628
10629 /*
10630 * If the swap file was just created, make sure delalloc is done. If the
10631 * file changes again after this, the user is doing something stupid and
10632 * we don't really care.
10633 */
10634 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10635 if (ret)
10636 return ret;
10637
10638 /*
10639 * The inode is locked, so these flags won't change after we check them.
10640 */
10641 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10642 btrfs_warn(fs_info, "swapfile must not be compressed");
10643 return -EINVAL;
10644 }
10645 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10646 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10647 return -EINVAL;
10648 }
10649 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10650 btrfs_warn(fs_info, "swapfile must not be checksummed");
10651 return -EINVAL;
10652 }
10653
10654 /*
10655 * Balance or device remove/replace/resize can move stuff around from
10656 * under us. The exclop protection makes sure they aren't running/won't
10657 * run concurrently while we are mapping the swap extents, and
10658 * fs_info->swapfile_pins prevents them from running while the swap
10659 * file is active and moving the extents. Note that this also prevents
10660 * a concurrent device add which isn't actually necessary, but it's not
10661 * really worth the trouble to allow it.
10662 */
10663 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10664 btrfs_warn(fs_info,
10665 "cannot activate swapfile while exclusive operation is running");
10666 return -EBUSY;
10667 }
10668
10669 /*
10670 * Prevent snapshot creation while we are activating the swap file.
10671 * We do not want to race with snapshot creation. If snapshot creation
10672 * already started before we bumped nr_swapfiles from 0 to 1 and
10673 * completes before the first write into the swap file after it is
10674 * activated, than that write would fallback to COW.
10675 */
10676 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10677 btrfs_exclop_finish(fs_info);
10678 btrfs_warn(fs_info,
10679 "cannot activate swapfile because snapshot creation is in progress");
10680 return -EINVAL;
10681 }
10682 /*
10683 * Snapshots can create extents which require COW even if NODATACOW is
10684 * set. We use this counter to prevent snapshots. We must increment it
10685 * before walking the extents because we don't want a concurrent
10686 * snapshot to run after we've already checked the extents.
10687 *
10688 * It is possible that subvolume is marked for deletion but still not
10689 * removed yet. To prevent this race, we check the root status before
10690 * activating the swapfile.
10691 */
10692 spin_lock(&root->root_item_lock);
10693 if (btrfs_root_dead(root)) {
10694 spin_unlock(&root->root_item_lock);
10695
10696 btrfs_exclop_finish(fs_info);
10697 btrfs_warn(fs_info,
10698 "cannot activate swapfile because subvolume %llu is being deleted",
10699 root->root_key.objectid);
10700 return -EPERM;
10701 }
10702 atomic_inc(&root->nr_swapfiles);
10703 spin_unlock(&root->root_item_lock);
10704
10705 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10706
10707 lock_extent(io_tree, 0, isize - 1, &cached_state);
10708 start = 0;
10709 while (start < isize) {
10710 u64 logical_block_start, physical_block_start;
10711 struct btrfs_block_group *bg;
10712 u64 len = isize - start;
10713
10714 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10715 if (IS_ERR(em)) {
10716 ret = PTR_ERR(em);
10717 goto out;
10718 }
10719
10720 if (em->block_start == EXTENT_MAP_HOLE) {
10721 btrfs_warn(fs_info, "swapfile must not have holes");
10722 ret = -EINVAL;
10723 goto out;
10724 }
10725 if (em->block_start == EXTENT_MAP_INLINE) {
10726 /*
10727 * It's unlikely we'll ever actually find ourselves
10728 * here, as a file small enough to fit inline won't be
10729 * big enough to store more than the swap header, but in
10730 * case something changes in the future, let's catch it
10731 * here rather than later.
10732 */
10733 btrfs_warn(fs_info, "swapfile must not be inline");
10734 ret = -EINVAL;
10735 goto out;
10736 }
10737 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10738 btrfs_warn(fs_info, "swapfile must not be compressed");
10739 ret = -EINVAL;
10740 goto out;
10741 }
10742
10743 logical_block_start = em->block_start + (start - em->start);
10744 len = min(len, em->len - (start - em->start));
10745 free_extent_map(em);
10746 em = NULL;
10747
10748 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10749 if (ret < 0) {
10750 goto out;
10751 } else if (ret) {
10752 ret = 0;
10753 } else {
10754 btrfs_warn(fs_info,
10755 "swapfile must not be copy-on-write");
10756 ret = -EINVAL;
10757 goto out;
10758 }
10759
10760 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10761 if (IS_ERR(em)) {
10762 ret = PTR_ERR(em);
10763 goto out;
10764 }
10765
10766 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10767 btrfs_warn(fs_info,
10768 "swapfile must have single data profile");
10769 ret = -EINVAL;
10770 goto out;
10771 }
10772
10773 if (device == NULL) {
10774 device = em->map_lookup->stripes[0].dev;
10775 ret = btrfs_add_swapfile_pin(inode, device, false);
10776 if (ret == 1)
10777 ret = 0;
10778 else if (ret)
10779 goto out;
10780 } else if (device != em->map_lookup->stripes[0].dev) {
10781 btrfs_warn(fs_info, "swapfile must be on one device");
10782 ret = -EINVAL;
10783 goto out;
10784 }
10785
10786 physical_block_start = (em->map_lookup->stripes[0].physical +
10787 (logical_block_start - em->start));
10788 len = min(len, em->len - (logical_block_start - em->start));
10789 free_extent_map(em);
10790 em = NULL;
10791
10792 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10793 if (!bg) {
10794 btrfs_warn(fs_info,
10795 "could not find block group containing swapfile");
10796 ret = -EINVAL;
10797 goto out;
10798 }
10799
10800 if (!btrfs_inc_block_group_swap_extents(bg)) {
10801 btrfs_warn(fs_info,
10802 "block group for swapfile at %llu is read-only%s",
10803 bg->start,
10804 atomic_read(&fs_info->scrubs_running) ?
10805 " (scrub running)" : "");
10806 btrfs_put_block_group(bg);
10807 ret = -EINVAL;
10808 goto out;
10809 }
10810
10811 ret = btrfs_add_swapfile_pin(inode, bg, true);
10812 if (ret) {
10813 btrfs_put_block_group(bg);
10814 if (ret == 1)
10815 ret = 0;
10816 else
10817 goto out;
10818 }
10819
10820 if (bsi.block_len &&
10821 bsi.block_start + bsi.block_len == physical_block_start) {
10822 bsi.block_len += len;
10823 } else {
10824 if (bsi.block_len) {
10825 ret = btrfs_add_swap_extent(sis, &bsi);
10826 if (ret)
10827 goto out;
10828 }
10829 bsi.start = start;
10830 bsi.block_start = physical_block_start;
10831 bsi.block_len = len;
10832 }
10833
10834 start += len;
10835 }
10836
10837 if (bsi.block_len)
10838 ret = btrfs_add_swap_extent(sis, &bsi);
10839
10840 out:
10841 if (!IS_ERR_OR_NULL(em))
10842 free_extent_map(em);
10843
10844 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10845
10846 if (ret)
10847 btrfs_swap_deactivate(file);
10848
10849 btrfs_drew_write_unlock(&root->snapshot_lock);
10850
10851 btrfs_exclop_finish(fs_info);
10852
10853 if (ret)
10854 return ret;
10855
10856 if (device)
10857 sis->bdev = device->bdev;
10858 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10859 sis->max = bsi.nr_pages;
10860 sis->pages = bsi.nr_pages - 1;
10861 sis->highest_bit = bsi.nr_pages - 1;
10862 return bsi.nr_extents;
10863 }
10864 #else
btrfs_swap_deactivate(struct file * file)10865 static void btrfs_swap_deactivate(struct file *file)
10866 {
10867 }
10868
btrfs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)10869 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10870 sector_t *span)
10871 {
10872 return -EOPNOTSUPP;
10873 }
10874 #endif
10875
10876 /*
10877 * Update the number of bytes used in the VFS' inode. When we replace extents in
10878 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10879 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10880 * always get a correct value.
10881 */
btrfs_update_inode_bytes(struct btrfs_inode * inode,const u64 add_bytes,const u64 del_bytes)10882 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10883 const u64 add_bytes,
10884 const u64 del_bytes)
10885 {
10886 if (add_bytes == del_bytes)
10887 return;
10888
10889 spin_lock(&inode->lock);
10890 if (del_bytes > 0)
10891 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10892 if (add_bytes > 0)
10893 inode_add_bytes(&inode->vfs_inode, add_bytes);
10894 spin_unlock(&inode->lock);
10895 }
10896
10897 /*
10898 * Verify that there are no ordered extents for a given file range.
10899 *
10900 * @inode: The target inode.
10901 * @start: Start offset of the file range, should be sector size aligned.
10902 * @end: End offset (inclusive) of the file range, its value +1 should be
10903 * sector size aligned.
10904 *
10905 * This should typically be used for cases where we locked an inode's VFS lock in
10906 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10907 * we have flushed all delalloc in the range, we have waited for all ordered
10908 * extents in the range to complete and finally we have locked the file range in
10909 * the inode's io_tree.
10910 */
btrfs_assert_inode_range_clean(struct btrfs_inode * inode,u64 start,u64 end)10911 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10912 {
10913 struct btrfs_root *root = inode->root;
10914 struct btrfs_ordered_extent *ordered;
10915
10916 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10917 return;
10918
10919 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10920 if (ordered) {
10921 btrfs_err(root->fs_info,
10922 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10923 start, end, btrfs_ino(inode), root->root_key.objectid,
10924 ordered->file_offset,
10925 ordered->file_offset + ordered->num_bytes - 1);
10926 btrfs_put_ordered_extent(ordered);
10927 }
10928
10929 ASSERT(ordered == NULL);
10930 }
10931
10932 static const struct inode_operations btrfs_dir_inode_operations = {
10933 .getattr = btrfs_getattr,
10934 .lookup = btrfs_lookup,
10935 .create = btrfs_create,
10936 .unlink = btrfs_unlink,
10937 .link = btrfs_link,
10938 .mkdir = btrfs_mkdir,
10939 .rmdir = btrfs_rmdir,
10940 .rename = btrfs_rename2,
10941 .symlink = btrfs_symlink,
10942 .setattr = btrfs_setattr,
10943 .mknod = btrfs_mknod,
10944 .listxattr = btrfs_listxattr,
10945 .permission = btrfs_permission,
10946 .get_inode_acl = btrfs_get_acl,
10947 .set_acl = btrfs_set_acl,
10948 .update_time = btrfs_update_time,
10949 .tmpfile = btrfs_tmpfile,
10950 .fileattr_get = btrfs_fileattr_get,
10951 .fileattr_set = btrfs_fileattr_set,
10952 };
10953
10954 static const struct file_operations btrfs_dir_file_operations = {
10955 .llseek = btrfs_dir_llseek,
10956 .read = generic_read_dir,
10957 .iterate_shared = btrfs_real_readdir,
10958 .open = btrfs_opendir,
10959 .unlocked_ioctl = btrfs_ioctl,
10960 #ifdef CONFIG_COMPAT
10961 .compat_ioctl = btrfs_compat_ioctl,
10962 #endif
10963 .release = btrfs_release_file,
10964 .fsync = btrfs_sync_file,
10965 };
10966
10967 /*
10968 * btrfs doesn't support the bmap operation because swapfiles
10969 * use bmap to make a mapping of extents in the file. They assume
10970 * these extents won't change over the life of the file and they
10971 * use the bmap result to do IO directly to the drive.
10972 *
10973 * the btrfs bmap call would return logical addresses that aren't
10974 * suitable for IO and they also will change frequently as COW
10975 * operations happen. So, swapfile + btrfs == corruption.
10976 *
10977 * For now we're avoiding this by dropping bmap.
10978 */
10979 static const struct address_space_operations btrfs_aops = {
10980 .read_folio = btrfs_read_folio,
10981 .writepages = btrfs_writepages,
10982 .readahead = btrfs_readahead,
10983 .invalidate_folio = btrfs_invalidate_folio,
10984 .release_folio = btrfs_release_folio,
10985 .migrate_folio = btrfs_migrate_folio,
10986 .dirty_folio = filemap_dirty_folio,
10987 .error_remove_page = generic_error_remove_page,
10988 .swap_activate = btrfs_swap_activate,
10989 .swap_deactivate = btrfs_swap_deactivate,
10990 };
10991
10992 static const struct inode_operations btrfs_file_inode_operations = {
10993 .getattr = btrfs_getattr,
10994 .setattr = btrfs_setattr,
10995 .listxattr = btrfs_listxattr,
10996 .permission = btrfs_permission,
10997 .fiemap = btrfs_fiemap,
10998 .get_inode_acl = btrfs_get_acl,
10999 .set_acl = btrfs_set_acl,
11000 .update_time = btrfs_update_time,
11001 .fileattr_get = btrfs_fileattr_get,
11002 .fileattr_set = btrfs_fileattr_set,
11003 };
11004 static const struct inode_operations btrfs_special_inode_operations = {
11005 .getattr = btrfs_getattr,
11006 .setattr = btrfs_setattr,
11007 .permission = btrfs_permission,
11008 .listxattr = btrfs_listxattr,
11009 .get_inode_acl = btrfs_get_acl,
11010 .set_acl = btrfs_set_acl,
11011 .update_time = btrfs_update_time,
11012 };
11013 static const struct inode_operations btrfs_symlink_inode_operations = {
11014 .get_link = page_get_link,
11015 .getattr = btrfs_getattr,
11016 .setattr = btrfs_setattr,
11017 .permission = btrfs_permission,
11018 .listxattr = btrfs_listxattr,
11019 .update_time = btrfs_update_time,
11020 };
11021
11022 const struct dentry_operations btrfs_dentry_operations = {
11023 .d_delete = btrfs_dentry_delete,
11024 };
11025