Line data Source code
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 : static noinline int cow_file_range(struct btrfs_inode *inode,
128 : struct page *locked_page,
129 : u64 start, u64 end, int *page_started,
130 : unsigned long *nr_written, int unlock,
131 : u64 *done_offset);
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 :
138 0 : static int data_reloc_print_warning_inode(u64 inum, u64 offset, u64 num_bytes,
139 : u64 root, void *warn_ctx)
140 : {
141 0 : struct data_reloc_warn *warn = warn_ctx;
142 0 : struct btrfs_fs_info *fs_info = warn->fs_info;
143 0 : struct extent_buffer *eb;
144 0 : struct btrfs_inode_item *inode_item;
145 0 : struct inode_fs_paths *ipath = NULL;
146 0 : struct btrfs_root *local_root;
147 0 : struct btrfs_key key;
148 0 : unsigned int nofs_flag;
149 0 : u32 nlink;
150 0 : int ret;
151 :
152 0 : local_root = btrfs_get_fs_root(fs_info, root, true);
153 0 : if (IS_ERR(local_root)) {
154 0 : ret = PTR_ERR(local_root);
155 0 : goto err;
156 : }
157 :
158 : /* This makes the path point to (inum INODE_ITEM ioff). */
159 0 : key.objectid = inum;
160 0 : key.type = BTRFS_INODE_ITEM_KEY;
161 0 : key.offset = 0;
162 :
163 0 : ret = btrfs_search_slot(NULL, local_root, &key, &warn->path, 0, 0);
164 0 : if (ret) {
165 0 : btrfs_put_root(local_root);
166 0 : btrfs_release_path(&warn->path);
167 0 : goto err;
168 : }
169 :
170 0 : eb = warn->path.nodes[0];
171 0 : inode_item = btrfs_item_ptr(eb, warn->path.slots[0], struct btrfs_inode_item);
172 0 : nlink = btrfs_inode_nlink(eb, inode_item);
173 0 : btrfs_release_path(&warn->path);
174 :
175 0 : nofs_flag = memalloc_nofs_save();
176 0 : ipath = init_ipath(4096, local_root, &warn->path);
177 0 : memalloc_nofs_restore(nofs_flag);
178 0 : if (IS_ERR(ipath)) {
179 0 : btrfs_put_root(local_root);
180 0 : ret = PTR_ERR(ipath);
181 0 : ipath = NULL;
182 : /*
183 : * -ENOMEM, not a critical error, just output an generic error
184 : * without filename.
185 : */
186 0 : 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 0 : return ret;
190 : }
191 0 : ret = paths_from_inode(inum, ipath);
192 0 : if (ret < 0)
193 0 : 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 0 : for (int i = 0; i < ipath->fspath->elem_cnt; i++) {
200 0 : 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 0 : btrfs_put_root(local_root);
208 0 : free_ipath(ipath);
209 0 : return 0;
210 :
211 0 : err:
212 0 : 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 0 : free_ipath(ipath);
217 0 : 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 : */
226 0 : 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 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
231 0 : struct btrfs_path path = { 0 };
232 0 : struct btrfs_key found_key = { 0 };
233 0 : struct extent_buffer *eb;
234 0 : struct btrfs_extent_item *ei;
235 0 : const u32 csum_size = fs_info->csum_size;
236 0 : u64 logical;
237 0 : u64 flags;
238 0 : u32 item_size;
239 0 : int ret;
240 :
241 0 : mutex_lock(&fs_info->reloc_mutex);
242 0 : logical = btrfs_get_reloc_bg_bytenr(fs_info);
243 0 : mutex_unlock(&fs_info->reloc_mutex);
244 :
245 0 : if (logical == U64_MAX) {
246 0 : btrfs_warn_rl(fs_info, "has data reloc tree but no running relocation");
247 0 : 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 0 : return;
254 : }
255 :
256 0 : logical += file_off;
257 0 : 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 0 : ret = extent_from_logical(fs_info, logical, &path, &found_key, &flags);
266 0 : if (ret < 0) {
267 0 : btrfs_err_rl(fs_info, "failed to lookup extent item for logical %llu: %d",
268 : logical, ret);
269 0 : return;
270 : }
271 0 : eb = path.nodes[0];
272 0 : ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item);
273 0 : item_size = btrfs_item_size(eb, path.slots[0]);
274 0 : if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
275 0 : unsigned long ptr = 0;
276 0 : u64 ref_root;
277 0 : u8 ref_level;
278 :
279 0 : while (true) {
280 0 : ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
281 : item_size, &ref_root,
282 : &ref_level);
283 0 : if (ret < 0) {
284 0 : btrfs_warn_rl(fs_info,
285 : "failed to resolve tree backref for logical %llu: %d",
286 : logical, ret);
287 : break;
288 : }
289 0 : if (ret > 0)
290 : break;
291 :
292 0 : 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 0 : btrfs_release_path(&path);
299 : } else {
300 0 : struct btrfs_backref_walk_ctx ctx = { 0 };
301 0 : struct data_reloc_warn reloc_warn = { 0 };
302 :
303 0 : btrfs_release_path(&path);
304 :
305 0 : ctx.bytenr = found_key.objectid;
306 0 : ctx.extent_item_pos = logical - found_key.objectid;
307 0 : ctx.fs_info = fs_info;
308 :
309 0 : reloc_warn.logical = logical;
310 0 : reloc_warn.extent_item_size = found_key.offset;
311 0 : reloc_warn.mirror_num = mirror_num;
312 0 : reloc_warn.fs_info = fs_info;
313 :
314 0 : iterate_extent_inodes(&ctx, true,
315 : data_reloc_print_warning_inode, &reloc_warn);
316 : }
317 : }
318 :
319 0 : 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 0 : struct btrfs_root *root = inode->root;
323 0 : 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 0 : if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
327 0 : return print_data_reloc_error(inode, logical_start, csum,
328 : csum_expected, mirror_num);
329 :
330 : /* Output without objectid, which is more meaningful */
331 0 : if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID) {
332 0 : 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 0 : 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 : */
360 0 : int btrfs_inode_lock(struct btrfs_inode *inode, unsigned int ilock_flags)
361 : {
362 0 : if (ilock_flags & BTRFS_ILOCK_SHARED) {
363 0 : if (ilock_flags & BTRFS_ILOCK_TRY) {
364 0 : if (!inode_trylock_shared(&inode->vfs_inode))
365 : return -EAGAIN;
366 : else
367 0 : return 0;
368 : }
369 0 : inode_lock_shared(&inode->vfs_inode);
370 : } else {
371 0 : if (ilock_flags & BTRFS_ILOCK_TRY) {
372 0 : if (!inode_trylock(&inode->vfs_inode))
373 : return -EAGAIN;
374 : else
375 0 : return 0;
376 : }
377 0 : inode_lock(&inode->vfs_inode);
378 : }
379 0 : if (ilock_flags & BTRFS_ILOCK_MMAP)
380 0 : 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 : */
390 0 : void btrfs_inode_unlock(struct btrfs_inode *inode, unsigned int ilock_flags)
391 : {
392 0 : if (ilock_flags & BTRFS_ILOCK_MMAP)
393 0 : up_write(&inode->i_mmap_lock);
394 0 : if (ilock_flags & BTRFS_ILOCK_SHARED)
395 0 : inode_unlock_shared(&inode->vfs_inode);
396 : else
397 0 : inode_unlock(&inode->vfs_inode);
398 0 : }
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 : */
410 0 : static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
411 : struct page *locked_page,
412 : u64 offset, u64 bytes)
413 : {
414 0 : unsigned long index = offset >> PAGE_SHIFT;
415 0 : unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
416 0 : u64 page_start = 0, page_end = 0;
417 0 : struct page *page;
418 :
419 0 : if (locked_page) {
420 0 : page_start = page_offset(locked_page);
421 0 : page_end = page_start + PAGE_SIZE - 1;
422 : }
423 :
424 0 : while (index <= end_index) {
425 : /*
426 : * For locked page, we will call end_extent_writepage() on it
427 : * in run_delalloc_range() for the error handling. That
428 : * end_extent_writepage() function will call
429 : * btrfs_mark_ordered_io_finished() to clear page Ordered and
430 : * run the ordered extent accounting.
431 : *
432 : * Here we can't just clear the Ordered bit, or
433 : * btrfs_mark_ordered_io_finished() would skip the accounting
434 : * for the page range, and the ordered extent will never finish.
435 : */
436 0 : if (locked_page && index == (page_start >> PAGE_SHIFT)) {
437 0 : index++;
438 0 : continue;
439 : }
440 0 : page = find_get_page(inode->vfs_inode.i_mapping, index);
441 0 : index++;
442 0 : if (!page)
443 0 : continue;
444 :
445 : /*
446 : * Here we just clear all Ordered bits for every page in the
447 : * range, then btrfs_mark_ordered_io_finished() will handle
448 : * the ordered extent accounting for the range.
449 : */
450 0 : btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
451 : offset, bytes);
452 0 : put_page(page);
453 : }
454 :
455 0 : if (locked_page) {
456 : /* The locked page covers the full range, nothing needs to be done */
457 0 : if (bytes + offset <= page_start + PAGE_SIZE)
458 : return;
459 : /*
460 : * In case this page belongs to the delalloc range being
461 : * instantiated then skip it, since the first page of a range is
462 : * going to be properly cleaned up by the caller of
463 : * run_delalloc_range
464 : */
465 0 : if (page_start >= offset && page_end <= (offset + bytes - 1)) {
466 0 : bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
467 0 : offset = page_offset(locked_page) + PAGE_SIZE;
468 : }
469 : }
470 :
471 0 : return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false);
472 : }
473 :
474 : static int btrfs_dirty_inode(struct btrfs_inode *inode);
475 :
476 0 : static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
477 : struct btrfs_new_inode_args *args)
478 : {
479 0 : int err;
480 :
481 0 : if (args->default_acl) {
482 0 : err = __btrfs_set_acl(trans, args->inode, args->default_acl,
483 : ACL_TYPE_DEFAULT);
484 0 : if (err)
485 : return err;
486 : }
487 0 : if (args->acl) {
488 0 : err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
489 0 : if (err)
490 : return err;
491 : }
492 0 : if (!args->default_acl && !args->acl)
493 0 : cache_no_acl(args->inode);
494 0 : return btrfs_xattr_security_init(trans, args->inode, args->dir,
495 0 : &args->dentry->d_name);
496 : }
497 :
498 : /*
499 : * this does all the hard work for inserting an inline extent into
500 : * the btree. The caller should have done a btrfs_drop_extents so that
501 : * no overlapping inline items exist in the btree
502 : */
503 0 : static int insert_inline_extent(struct btrfs_trans_handle *trans,
504 : struct btrfs_path *path,
505 : struct btrfs_inode *inode, bool extent_inserted,
506 : size_t size, size_t compressed_size,
507 : int compress_type,
508 : struct page **compressed_pages,
509 : bool update_i_size)
510 : {
511 0 : struct btrfs_root *root = inode->root;
512 0 : struct extent_buffer *leaf;
513 0 : struct page *page = NULL;
514 0 : char *kaddr;
515 0 : unsigned long ptr;
516 0 : struct btrfs_file_extent_item *ei;
517 0 : int ret;
518 0 : size_t cur_size = size;
519 0 : u64 i_size;
520 :
521 0 : ASSERT((compressed_size > 0 && compressed_pages) ||
522 : (compressed_size == 0 && !compressed_pages));
523 :
524 0 : if (compressed_size && compressed_pages)
525 0 : cur_size = compressed_size;
526 :
527 0 : if (!extent_inserted) {
528 0 : struct btrfs_key key;
529 0 : size_t datasize;
530 :
531 0 : key.objectid = btrfs_ino(inode);
532 0 : key.offset = 0;
533 0 : key.type = BTRFS_EXTENT_DATA_KEY;
534 :
535 0 : datasize = btrfs_file_extent_calc_inline_size(cur_size);
536 0 : ret = btrfs_insert_empty_item(trans, root, path, &key,
537 : datasize);
538 0 : if (ret)
539 0 : goto fail;
540 : }
541 0 : leaf = path->nodes[0];
542 0 : ei = btrfs_item_ptr(leaf, path->slots[0],
543 : struct btrfs_file_extent_item);
544 0 : btrfs_set_file_extent_generation(leaf, ei, trans->transid);
545 0 : btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
546 0 : btrfs_set_file_extent_encryption(leaf, ei, 0);
547 0 : btrfs_set_file_extent_other_encoding(leaf, ei, 0);
548 0 : btrfs_set_file_extent_ram_bytes(leaf, ei, size);
549 0 : ptr = btrfs_file_extent_inline_start(ei);
550 :
551 0 : if (compress_type != BTRFS_COMPRESS_NONE) {
552 : struct page *cpage;
553 : int i = 0;
554 0 : while (compressed_size > 0) {
555 0 : cpage = compressed_pages[i];
556 0 : cur_size = min_t(unsigned long, compressed_size,
557 : PAGE_SIZE);
558 :
559 0 : kaddr = kmap_local_page(cpage);
560 0 : write_extent_buffer(leaf, kaddr, ptr, cur_size);
561 0 : kunmap_local(kaddr);
562 :
563 0 : i++;
564 0 : ptr += cur_size;
565 0 : compressed_size -= cur_size;
566 : }
567 0 : btrfs_set_file_extent_compression(leaf, ei,
568 : compress_type);
569 : } else {
570 0 : page = find_get_page(inode->vfs_inode.i_mapping, 0);
571 0 : btrfs_set_file_extent_compression(leaf, ei, 0);
572 0 : kaddr = kmap_local_page(page);
573 0 : write_extent_buffer(leaf, kaddr, ptr, size);
574 0 : kunmap_local(kaddr);
575 0 : put_page(page);
576 : }
577 0 : btrfs_mark_buffer_dirty(leaf);
578 0 : btrfs_release_path(path);
579 :
580 : /*
581 : * We align size to sectorsize for inline extents just for simplicity
582 : * sake.
583 : */
584 0 : ret = btrfs_inode_set_file_extent_range(inode, 0,
585 0 : ALIGN(size, root->fs_info->sectorsize));
586 0 : if (ret)
587 0 : goto fail;
588 :
589 : /*
590 : * We're an inline extent, so nobody can extend the file past i_size
591 : * without locking a page we already have locked.
592 : *
593 : * We must do any i_size and inode updates before we unlock the pages.
594 : * Otherwise we could end up racing with unlink.
595 : */
596 0 : i_size = i_size_read(&inode->vfs_inode);
597 0 : if (update_i_size && size > i_size) {
598 0 : i_size_write(&inode->vfs_inode, size);
599 0 : i_size = size;
600 : }
601 0 : inode->disk_i_size = i_size;
602 :
603 0 : fail:
604 0 : return ret;
605 : }
606 :
607 :
608 : /*
609 : * conditionally insert an inline extent into the file. This
610 : * does the checks required to make sure the data is small enough
611 : * to fit as an inline extent.
612 : */
613 0 : static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
614 : size_t compressed_size,
615 : int compress_type,
616 : struct page **compressed_pages,
617 : bool update_i_size)
618 : {
619 0 : struct btrfs_drop_extents_args drop_args = { 0 };
620 0 : struct btrfs_root *root = inode->root;
621 0 : struct btrfs_fs_info *fs_info = root->fs_info;
622 0 : struct btrfs_trans_handle *trans;
623 0 : u64 data_len = (compressed_size ?: size);
624 0 : int ret;
625 0 : struct btrfs_path *path;
626 :
627 : /*
628 : * We can create an inline extent if it ends at or beyond the current
629 : * i_size, is no larger than a sector (decompressed), and the (possibly
630 : * compressed) data fits in a leaf and the configured maximum inline
631 : * size.
632 : */
633 0 : if (size < i_size_read(&inode->vfs_inode) ||
634 0 : size > fs_info->sectorsize ||
635 0 : data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
636 0 : data_len > fs_info->max_inline)
637 : return 1;
638 :
639 0 : path = btrfs_alloc_path();
640 0 : if (!path)
641 : return -ENOMEM;
642 :
643 0 : trans = btrfs_join_transaction(root);
644 0 : if (IS_ERR(trans)) {
645 0 : btrfs_free_path(path);
646 0 : return PTR_ERR(trans);
647 : }
648 0 : trans->block_rsv = &inode->block_rsv;
649 :
650 0 : drop_args.path = path;
651 0 : drop_args.start = 0;
652 0 : drop_args.end = fs_info->sectorsize;
653 0 : drop_args.drop_cache = true;
654 0 : drop_args.replace_extent = true;
655 0 : drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
656 0 : ret = btrfs_drop_extents(trans, root, inode, &drop_args);
657 0 : if (ret) {
658 0 : btrfs_abort_transaction(trans, ret);
659 0 : goto out;
660 : }
661 :
662 0 : ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
663 : size, compressed_size, compress_type,
664 : compressed_pages, update_i_size);
665 0 : if (ret && ret != -ENOSPC) {
666 0 : btrfs_abort_transaction(trans, ret);
667 0 : goto out;
668 0 : } else if (ret == -ENOSPC) {
669 0 : ret = 1;
670 0 : goto out;
671 : }
672 :
673 0 : btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
674 0 : ret = btrfs_update_inode(trans, root, inode);
675 0 : if (ret && ret != -ENOSPC) {
676 0 : btrfs_abort_transaction(trans, ret);
677 0 : goto out;
678 0 : } else if (ret == -ENOSPC) {
679 0 : ret = 1;
680 0 : goto out;
681 : }
682 :
683 0 : btrfs_set_inode_full_sync(inode);
684 0 : out:
685 : /*
686 : * Don't forget to free the reserved space, as for inlined extent
687 : * it won't count as data extent, free them directly here.
688 : * And at reserve time, it's always aligned to page size, so
689 : * just free one page here.
690 : */
691 0 : btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
692 0 : btrfs_free_path(path);
693 0 : btrfs_end_transaction(trans);
694 0 : return ret;
695 : }
696 :
697 : struct async_extent {
698 : u64 start;
699 : u64 ram_size;
700 : u64 compressed_size;
701 : struct page **pages;
702 : unsigned long nr_pages;
703 : int compress_type;
704 : struct list_head list;
705 : };
706 :
707 : struct async_chunk {
708 : struct btrfs_inode *inode;
709 : struct page *locked_page;
710 : u64 start;
711 : u64 end;
712 : blk_opf_t write_flags;
713 : struct list_head extents;
714 : struct cgroup_subsys_state *blkcg_css;
715 : struct btrfs_work work;
716 : struct async_cow *async_cow;
717 : };
718 :
719 : struct async_cow {
720 : atomic_t num_chunks;
721 : struct async_chunk chunks[];
722 : };
723 :
724 0 : static noinline int add_async_extent(struct async_chunk *cow,
725 : u64 start, u64 ram_size,
726 : u64 compressed_size,
727 : struct page **pages,
728 : unsigned long nr_pages,
729 : int compress_type)
730 : {
731 0 : struct async_extent *async_extent;
732 :
733 0 : async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
734 0 : BUG_ON(!async_extent); /* -ENOMEM */
735 0 : async_extent->start = start;
736 0 : async_extent->ram_size = ram_size;
737 0 : async_extent->compressed_size = compressed_size;
738 0 : async_extent->pages = pages;
739 0 : async_extent->nr_pages = nr_pages;
740 0 : async_extent->compress_type = compress_type;
741 0 : list_add_tail(&async_extent->list, &cow->extents);
742 0 : 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 : */
749 0 : static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
750 : u64 end)
751 : {
752 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
753 :
754 0 : 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 0 : if (fs_info->sectorsize < PAGE_SIZE) {
787 0 : if (!PAGE_ALIGNED(start) ||
788 0 : !PAGE_ALIGNED(end + 1))
789 : return 0;
790 : }
791 :
792 : /* force compress */
793 0 : if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
794 : return 1;
795 : /* defrag ioctl */
796 0 : if (inode->defrag_compress)
797 : return 1;
798 : /* bad compression ratios */
799 0 : if (inode->flags & BTRFS_INODE_NOCOMPRESS)
800 : return 0;
801 0 : if (btrfs_test_opt(fs_info, COMPRESS) ||
802 0 : inode->flags & BTRFS_INODE_COMPRESS ||
803 0 : inode->prop_compress)
804 0 : return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
805 : return 0;
806 : }
807 :
808 0 : 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 0 : if (num_bytes < small_write &&
813 0 : (start > 0 || end + 1 < inode->disk_i_size))
814 0 : btrfs_add_inode_defrag(NULL, inode, small_write);
815 0 : }
816 :
817 : /*
818 : * we create compressed extents in two phases. The first
819 : * phase compresses a range of pages that have already been
820 : * locked (both pages and state bits are locked).
821 : *
822 : * This is done inside an ordered work queue, and the compression
823 : * is spread across many cpus. The actual IO submission is step
824 : * two, and the ordered work queue takes care of making sure that
825 : * happens in the same order things were put onto the queue by
826 : * writepages and friends.
827 : *
828 : * If this code finds it can't get good compression, it puts an
829 : * entry onto the work queue to write the uncompressed bytes. This
830 : * makes sure that both compressed inodes and uncompressed inodes
831 : * are written in the same order that the flusher thread sent them
832 : * down.
833 : */
834 0 : static noinline int compress_file_range(struct async_chunk *async_chunk)
835 : {
836 0 : struct btrfs_inode *inode = async_chunk->inode;
837 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
838 0 : struct address_space *mapping = inode->vfs_inode.i_mapping;
839 0 : u64 blocksize = fs_info->sectorsize;
840 0 : u64 start = async_chunk->start;
841 0 : u64 end = async_chunk->end;
842 0 : u64 actual_end;
843 0 : u64 i_size;
844 0 : int ret = 0;
845 0 : struct page **pages = NULL;
846 0 : unsigned long nr_pages;
847 0 : unsigned long total_compressed = 0;
848 0 : unsigned long total_in = 0;
849 0 : int i;
850 0 : int will_compress;
851 0 : int compress_type = fs_info->compress_type;
852 0 : int compressed_extents = 0;
853 0 : int redirty = 0;
854 :
855 0 : inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
856 :
857 : /*
858 : * We need to save i_size before now because it could change in between
859 : * us evaluating the size and assigning it. This is because we lock and
860 : * unlock the page in truncate and fallocate, and then modify the i_size
861 : * later on.
862 : *
863 : * The barriers are to emulate READ_ONCE, remove that once i_size_read
864 : * does that for us.
865 : */
866 0 : barrier();
867 0 : i_size = i_size_read(&inode->vfs_inode);
868 0 : barrier();
869 0 : actual_end = min_t(u64, i_size, end + 1);
870 0 : again:
871 0 : will_compress = 0;
872 0 : nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
873 0 : nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES);
874 :
875 : /*
876 : * we don't want to send crud past the end of i_size through
877 : * compression, that's just a waste of CPU time. So, if the
878 : * end of the file is before the start of our current
879 : * requested range of bytes, we bail out to the uncompressed
880 : * cleanup code that can deal with all of this.
881 : *
882 : * It isn't really the fastest way to fix things, but this is a
883 : * very uncommon corner.
884 : */
885 0 : if (actual_end <= start)
886 0 : goto cleanup_and_bail_uncompressed;
887 :
888 0 : total_compressed = actual_end - start;
889 :
890 : /*
891 : * Skip compression for a small file range(<=blocksize) that
892 : * isn't an inline extent, since it doesn't save disk space at all.
893 : */
894 0 : if (total_compressed <= blocksize &&
895 0 : (start > 0 || end + 1 < inode->disk_i_size))
896 0 : goto cleanup_and_bail_uncompressed;
897 :
898 : /*
899 : * For subpage case, we require full page alignment for the sector
900 : * aligned range.
901 : * Thus we must also check against @actual_end, not just @end.
902 : */
903 0 : if (blocksize < PAGE_SIZE) {
904 0 : if (!PAGE_ALIGNED(start) ||
905 0 : !PAGE_ALIGNED(round_up(actual_end, blocksize)))
906 0 : goto cleanup_and_bail_uncompressed;
907 : }
908 :
909 0 : total_compressed = min_t(unsigned long, total_compressed,
910 : BTRFS_MAX_UNCOMPRESSED);
911 0 : total_in = 0;
912 0 : ret = 0;
913 :
914 : /*
915 : * we do compression for mount -o compress and when the
916 : * inode has not been flagged as nocompress. This flag can
917 : * change at any time if we discover bad compression ratios.
918 : */
919 0 : if (inode_need_compress(inode, start, end)) {
920 0 : WARN_ON(pages);
921 0 : pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
922 0 : if (!pages) {
923 : /* just bail out to the uncompressed code */
924 0 : nr_pages = 0;
925 0 : goto cont;
926 : }
927 :
928 0 : if (inode->defrag_compress)
929 0 : compress_type = inode->defrag_compress;
930 0 : else if (inode->prop_compress)
931 0 : compress_type = inode->prop_compress;
932 :
933 : /*
934 : * we need to call clear_page_dirty_for_io on each
935 : * page in the range. Otherwise applications with the file
936 : * mmap'd can wander in and change the page contents while
937 : * we are compressing them.
938 : *
939 : * If the compression fails for any reason, we set the pages
940 : * dirty again later on.
941 : *
942 : * Note that the remaining part is redirtied, the start pointer
943 : * has moved, the end is the original one.
944 : */
945 0 : if (!redirty) {
946 0 : extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end);
947 0 : redirty = 1;
948 : }
949 :
950 : /* Compression level is applied here and only here */
951 0 : ret = btrfs_compress_pages(
952 0 : compress_type | (fs_info->compress_level << 4),
953 : mapping, start,
954 : pages,
955 : &nr_pages,
956 : &total_in,
957 : &total_compressed);
958 :
959 0 : if (!ret) {
960 0 : unsigned long offset = offset_in_page(total_compressed);
961 0 : struct page *page = pages[nr_pages - 1];
962 :
963 : /* zero the tail end of the last page, we might be
964 : * sending it down to disk
965 : */
966 0 : if (offset)
967 0 : memzero_page(page, offset, PAGE_SIZE - offset);
968 : will_compress = 1;
969 : }
970 : }
971 0 : cont:
972 : /*
973 : * Check cow_file_range() for why we don't even try to create inline
974 : * extent for subpage case.
975 : */
976 0 : if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
977 : /* lets try to make an inline extent */
978 0 : if (ret || total_in < actual_end) {
979 : /* we didn't compress the entire range, try
980 : * to make an uncompressed inline extent.
981 : */
982 0 : ret = cow_file_range_inline(inode, actual_end,
983 : 0, BTRFS_COMPRESS_NONE,
984 : NULL, false);
985 : } else {
986 : /* try making a compressed inline extent */
987 0 : ret = cow_file_range_inline(inode, actual_end,
988 : total_compressed,
989 : compress_type, pages,
990 : false);
991 : }
992 0 : if (ret <= 0) {
993 0 : unsigned long clear_flags = EXTENT_DELALLOC |
994 : EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
995 : EXTENT_DO_ACCOUNTING;
996 :
997 0 : if (ret < 0)
998 0 : mapping_set_error(mapping, -EIO);
999 :
1000 : /*
1001 : * inline extent creation worked or returned error,
1002 : * we don't need to create any more async work items.
1003 : * Unlock and free up our temp pages.
1004 : *
1005 : * We use DO_ACCOUNTING here because we need the
1006 : * delalloc_release_metadata to be done _after_ we drop
1007 : * our outstanding extent for clearing delalloc for this
1008 : * range.
1009 : */
1010 0 : extent_clear_unlock_delalloc(inode, start, end,
1011 : NULL,
1012 : clear_flags,
1013 : PAGE_UNLOCK |
1014 : PAGE_START_WRITEBACK |
1015 : PAGE_END_WRITEBACK);
1016 :
1017 : /*
1018 : * Ensure we only free the compressed pages if we have
1019 : * them allocated, as we can still reach here with
1020 : * inode_need_compress() == false.
1021 : */
1022 0 : if (pages) {
1023 0 : for (i = 0; i < nr_pages; i++) {
1024 0 : WARN_ON(pages[i]->mapping);
1025 0 : put_page(pages[i]);
1026 : }
1027 0 : kfree(pages);
1028 : }
1029 0 : return 0;
1030 : }
1031 : }
1032 :
1033 0 : if (will_compress) {
1034 : /*
1035 : * we aren't doing an inline extent round the compressed size
1036 : * up to a block size boundary so the allocator does sane
1037 : * things
1038 : */
1039 0 : total_compressed = ALIGN(total_compressed, blocksize);
1040 :
1041 : /*
1042 : * one last check to make sure the compression is really a
1043 : * win, compare the page count read with the blocks on disk,
1044 : * compression must free at least one sector size
1045 : */
1046 0 : total_in = round_up(total_in, fs_info->sectorsize);
1047 0 : if (total_compressed + blocksize <= total_in) {
1048 0 : compressed_extents++;
1049 :
1050 : /*
1051 : * The async work queues will take care of doing actual
1052 : * allocation on disk for these compressed pages, and
1053 : * will submit them to the elevator.
1054 : */
1055 0 : add_async_extent(async_chunk, start, total_in,
1056 : total_compressed, pages, nr_pages,
1057 : compress_type);
1058 :
1059 0 : if (start + total_in < end) {
1060 0 : start += total_in;
1061 0 : pages = NULL;
1062 0 : cond_resched();
1063 0 : goto again;
1064 : }
1065 0 : return compressed_extents;
1066 : }
1067 : }
1068 0 : if (pages) {
1069 : /*
1070 : * the compression code ran but failed to make things smaller,
1071 : * free any pages it allocated and our page pointer array
1072 : */
1073 0 : for (i = 0; i < nr_pages; i++) {
1074 0 : WARN_ON(pages[i]->mapping);
1075 0 : put_page(pages[i]);
1076 : }
1077 0 : kfree(pages);
1078 0 : pages = NULL;
1079 0 : total_compressed = 0;
1080 0 : nr_pages = 0;
1081 :
1082 : /* flag the file so we don't compress in the future */
1083 0 : if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
1084 0 : !(inode->prop_compress)) {
1085 0 : inode->flags |= BTRFS_INODE_NOCOMPRESS;
1086 : }
1087 : }
1088 0 : cleanup_and_bail_uncompressed:
1089 : /*
1090 : * No compression, but we still need to write the pages in the file
1091 : * we've been given so far. redirty the locked page if it corresponds
1092 : * to our extent and set things up for the async work queue to run
1093 : * cow_file_range to do the normal delalloc dance.
1094 : */
1095 0 : if (async_chunk->locked_page &&
1096 0 : (page_offset(async_chunk->locked_page) >= start &&
1097 : page_offset(async_chunk->locked_page)) <= end) {
1098 0 : __set_page_dirty_nobuffers(async_chunk->locked_page);
1099 : /* unlocked later on in the async handlers */
1100 : }
1101 :
1102 0 : if (redirty)
1103 0 : extent_range_redirty_for_io(&inode->vfs_inode, start, end);
1104 0 : add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
1105 : BTRFS_COMPRESS_NONE);
1106 0 : compressed_extents++;
1107 :
1108 0 : return compressed_extents;
1109 : }
1110 :
1111 0 : static void free_async_extent_pages(struct async_extent *async_extent)
1112 : {
1113 0 : int i;
1114 :
1115 0 : if (!async_extent->pages)
1116 : return;
1117 :
1118 0 : for (i = 0; i < async_extent->nr_pages; i++) {
1119 0 : WARN_ON(async_extent->pages[i]->mapping);
1120 0 : put_page(async_extent->pages[i]);
1121 : }
1122 0 : kfree(async_extent->pages);
1123 0 : async_extent->nr_pages = 0;
1124 0 : async_extent->pages = NULL;
1125 : }
1126 :
1127 0 : static int submit_uncompressed_range(struct btrfs_inode *inode,
1128 : struct async_extent *async_extent,
1129 : struct page *locked_page)
1130 : {
1131 0 : u64 start = async_extent->start;
1132 0 : u64 end = async_extent->start + async_extent->ram_size - 1;
1133 0 : unsigned long nr_written = 0;
1134 0 : int page_started = 0;
1135 0 : int ret;
1136 0 : struct writeback_control wbc = {
1137 : .sync_mode = WB_SYNC_ALL,
1138 : .range_start = start,
1139 : .range_end = end,
1140 : .no_cgroup_owner = 1,
1141 : };
1142 :
1143 : /*
1144 : * Call cow_file_range() to run the delalloc range directly, since we
1145 : * won't go to NOCOW or async path again.
1146 : *
1147 : * Also we call cow_file_range() with @unlock_page == 0, so that we
1148 : * can directly submit them without interruption.
1149 : */
1150 0 : ret = cow_file_range(inode, locked_page, start, end, &page_started,
1151 : &nr_written, 0, NULL);
1152 : /* Inline extent inserted, page gets unlocked and everything is done */
1153 0 : if (page_started)
1154 : return 0;
1155 :
1156 0 : if (ret < 0) {
1157 0 : btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
1158 0 : if (locked_page) {
1159 0 : const u64 page_start = page_offset(locked_page);
1160 0 : const u64 page_end = page_start + PAGE_SIZE - 1;
1161 :
1162 0 : set_page_writeback(locked_page);
1163 0 : end_page_writeback(locked_page);
1164 0 : end_extent_writepage(locked_page, ret, page_start, page_end);
1165 0 : unlock_page(locked_page);
1166 : }
1167 0 : return ret;
1168 : }
1169 :
1170 : /* All pages will be unlocked, including @locked_page */
1171 0 : wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode);
1172 0 : ret = extent_write_locked_range(&inode->vfs_inode, start, end, &wbc);
1173 0 : wbc_detach_inode(&wbc);
1174 0 : return ret;
1175 : }
1176 :
1177 0 : static int submit_one_async_extent(struct btrfs_inode *inode,
1178 : struct async_chunk *async_chunk,
1179 : struct async_extent *async_extent,
1180 : u64 *alloc_hint)
1181 : {
1182 0 : struct extent_io_tree *io_tree = &inode->io_tree;
1183 0 : struct btrfs_root *root = inode->root;
1184 0 : struct btrfs_fs_info *fs_info = root->fs_info;
1185 0 : struct btrfs_ordered_extent *ordered;
1186 0 : struct btrfs_key ins;
1187 0 : struct page *locked_page = NULL;
1188 0 : struct extent_map *em;
1189 0 : int ret = 0;
1190 0 : u64 start = async_extent->start;
1191 0 : u64 end = async_extent->start + async_extent->ram_size - 1;
1192 :
1193 0 : if (async_chunk->blkcg_css)
1194 0 : kthread_associate_blkcg(async_chunk->blkcg_css);
1195 :
1196 : /*
1197 : * If async_chunk->locked_page is in the async_extent range, we need to
1198 : * handle it.
1199 : */
1200 0 : if (async_chunk->locked_page) {
1201 0 : u64 locked_page_start = page_offset(async_chunk->locked_page);
1202 0 : u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
1203 :
1204 0 : if (!(start >= locked_page_end || end <= locked_page_start))
1205 0 : locked_page = async_chunk->locked_page;
1206 : }
1207 0 : lock_extent(io_tree, start, end, NULL);
1208 :
1209 : /* We have fall back to uncompressed write */
1210 0 : if (!async_extent->pages) {
1211 0 : ret = submit_uncompressed_range(inode, async_extent, locked_page);
1212 0 : goto done;
1213 : }
1214 :
1215 0 : ret = btrfs_reserve_extent(root, async_extent->ram_size,
1216 : async_extent->compressed_size,
1217 : async_extent->compressed_size,
1218 : 0, *alloc_hint, &ins, 1, 1);
1219 0 : if (ret) {
1220 0 : free_async_extent_pages(async_extent);
1221 : /*
1222 : * Here we used to try again by going back to non-compressed
1223 : * path for ENOSPC. But we can't reserve space even for
1224 : * compressed size, how could it work for uncompressed size
1225 : * which requires larger size? So here we directly go error
1226 : * path.
1227 : */
1228 0 : goto out_free;
1229 : }
1230 :
1231 : /* Here we're doing allocation and writeback of the compressed pages */
1232 0 : em = create_io_em(inode, start,
1233 : async_extent->ram_size, /* len */
1234 : start, /* orig_start */
1235 : ins.objectid, /* block_start */
1236 : ins.offset, /* block_len */
1237 : ins.offset, /* orig_block_len */
1238 : async_extent->ram_size, /* ram_bytes */
1239 : async_extent->compress_type,
1240 : BTRFS_ORDERED_COMPRESSED);
1241 0 : if (IS_ERR(em)) {
1242 0 : ret = PTR_ERR(em);
1243 0 : goto out_free_reserve;
1244 : }
1245 0 : free_extent_map(em);
1246 :
1247 0 : ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */
1248 : async_extent->ram_size, /* num_bytes */
1249 : async_extent->ram_size, /* ram_bytes */
1250 : ins.objectid, /* disk_bytenr */
1251 : ins.offset, /* disk_num_bytes */
1252 : 0, /* offset */
1253 : 1 << BTRFS_ORDERED_COMPRESSED,
1254 : async_extent->compress_type);
1255 0 : if (IS_ERR(ordered)) {
1256 0 : btrfs_drop_extent_map_range(inode, start, end, false);
1257 0 : ret = PTR_ERR(ordered);
1258 0 : goto out_free_reserve;
1259 : }
1260 0 : btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1261 :
1262 : /* Clear dirty, set writeback and unlock the pages. */
1263 0 : extent_clear_unlock_delalloc(inode, start, end,
1264 : NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1265 : PAGE_UNLOCK | PAGE_START_WRITEBACK);
1266 0 : btrfs_submit_compressed_write(ordered,
1267 : async_extent->pages, /* compressed_pages */
1268 0 : async_extent->nr_pages,
1269 : async_chunk->write_flags, true);
1270 0 : *alloc_hint = ins.objectid + ins.offset;
1271 0 : done:
1272 0 : if (async_chunk->blkcg_css)
1273 0 : kthread_associate_blkcg(NULL);
1274 0 : kfree(async_extent);
1275 0 : return ret;
1276 :
1277 0 : out_free_reserve:
1278 0 : btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1279 0 : btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1280 0 : out_free:
1281 0 : mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1282 0 : extent_clear_unlock_delalloc(inode, start, end,
1283 : NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1284 : EXTENT_DELALLOC_NEW |
1285 : EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1286 : PAGE_UNLOCK | PAGE_START_WRITEBACK |
1287 : PAGE_END_WRITEBACK);
1288 0 : free_async_extent_pages(async_extent);
1289 0 : goto done;
1290 : }
1291 :
1292 : /*
1293 : * Phase two of compressed writeback. This is the ordered portion of the code,
1294 : * which only gets called in the order the work was queued. We walk all the
1295 : * async extents created by compress_file_range and send them down to the disk.
1296 : */
1297 0 : static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
1298 : {
1299 0 : struct btrfs_inode *inode = async_chunk->inode;
1300 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
1301 0 : struct async_extent *async_extent;
1302 0 : u64 alloc_hint = 0;
1303 0 : int ret = 0;
1304 :
1305 0 : while (!list_empty(&async_chunk->extents)) {
1306 0 : u64 extent_start;
1307 0 : u64 ram_size;
1308 :
1309 0 : async_extent = list_entry(async_chunk->extents.next,
1310 : struct async_extent, list);
1311 0 : list_del(&async_extent->list);
1312 0 : extent_start = async_extent->start;
1313 0 : ram_size = async_extent->ram_size;
1314 :
1315 0 : ret = submit_one_async_extent(inode, async_chunk, async_extent,
1316 : &alloc_hint);
1317 0 : btrfs_debug(fs_info,
1318 : "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1319 : inode->root->root_key.objectid,
1320 : btrfs_ino(inode), extent_start, ram_size, ret);
1321 : }
1322 0 : }
1323 :
1324 0 : static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1325 : u64 num_bytes)
1326 : {
1327 0 : struct extent_map_tree *em_tree = &inode->extent_tree;
1328 0 : struct extent_map *em;
1329 0 : u64 alloc_hint = 0;
1330 :
1331 0 : read_lock(&em_tree->lock);
1332 0 : em = search_extent_mapping(em_tree, start, num_bytes);
1333 0 : if (em) {
1334 : /*
1335 : * if block start isn't an actual block number then find the
1336 : * first block in this inode and use that as a hint. If that
1337 : * block is also bogus then just don't worry about it.
1338 : */
1339 0 : if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1340 0 : free_extent_map(em);
1341 0 : em = search_extent_mapping(em_tree, 0, 0);
1342 0 : if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1343 0 : alloc_hint = em->block_start;
1344 0 : if (em)
1345 0 : free_extent_map(em);
1346 : } else {
1347 0 : alloc_hint = em->block_start;
1348 0 : free_extent_map(em);
1349 : }
1350 : }
1351 0 : read_unlock(&em_tree->lock);
1352 :
1353 0 : return alloc_hint;
1354 : }
1355 :
1356 : /*
1357 : * when extent_io.c finds a delayed allocation range in the file,
1358 : * the call backs end up in this code. The basic idea is to
1359 : * allocate extents on disk for the range, and create ordered data structs
1360 : * in ram to track those extents.
1361 : *
1362 : * locked_page is the page that writepage had locked already. We use
1363 : * it to make sure we don't do extra locks or unlocks.
1364 : *
1365 : * *page_started is set to one if we unlock locked_page and do everything
1366 : * required to start IO on it. It may be clean and already done with
1367 : * IO when we return.
1368 : *
1369 : * When unlock == 1, we unlock the pages in successfully allocated regions.
1370 : * When unlock == 0, we leave them locked for writing them out.
1371 : *
1372 : * However, we unlock all the pages except @locked_page in case of failure.
1373 : *
1374 : * In summary, page locking state will be as follow:
1375 : *
1376 : * - page_started == 1 (return value)
1377 : * - All the pages are unlocked. IO is started.
1378 : * - Note that this can happen only on success
1379 : * - unlock == 1
1380 : * - All the pages except @locked_page are unlocked in any case
1381 : * - unlock == 0
1382 : * - On success, all the pages are locked for writing out them
1383 : * - On failure, all the pages except @locked_page are unlocked
1384 : *
1385 : * When a failure happens in the second or later iteration of the
1386 : * while-loop, the ordered extents created in previous iterations are kept
1387 : * intact. So, the caller must clean them up by calling
1388 : * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1389 : * example.
1390 : */
1391 0 : static noinline int cow_file_range(struct btrfs_inode *inode,
1392 : struct page *locked_page,
1393 : u64 start, u64 end, int *page_started,
1394 : unsigned long *nr_written, int unlock,
1395 : u64 *done_offset)
1396 : {
1397 0 : struct btrfs_root *root = inode->root;
1398 0 : struct btrfs_fs_info *fs_info = root->fs_info;
1399 0 : u64 alloc_hint = 0;
1400 0 : u64 orig_start = start;
1401 0 : u64 num_bytes;
1402 0 : unsigned long ram_size;
1403 0 : u64 cur_alloc_size = 0;
1404 0 : u64 min_alloc_size;
1405 0 : u64 blocksize = fs_info->sectorsize;
1406 0 : struct btrfs_key ins;
1407 0 : struct extent_map *em;
1408 0 : unsigned clear_bits;
1409 0 : unsigned long page_ops;
1410 0 : bool extent_reserved = false;
1411 0 : int ret = 0;
1412 :
1413 0 : if (btrfs_is_free_space_inode(inode)) {
1414 0 : ret = -EINVAL;
1415 0 : goto out_unlock;
1416 : }
1417 :
1418 0 : num_bytes = ALIGN(end - start + 1, blocksize);
1419 0 : num_bytes = max(blocksize, num_bytes);
1420 0 : ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1421 :
1422 0 : inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1423 :
1424 : /*
1425 : * Due to the page size limit, for subpage we can only trigger the
1426 : * writeback for the dirty sectors of page, that means data writeback
1427 : * is doing more writeback than what we want.
1428 : *
1429 : * This is especially unexpected for some call sites like fallocate,
1430 : * where we only increase i_size after everything is done.
1431 : * This means we can trigger inline extent even if we didn't want to.
1432 : * So here we skip inline extent creation completely.
1433 : */
1434 0 : if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
1435 0 : u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1436 : end + 1);
1437 :
1438 : /* lets try to make an inline extent */
1439 0 : ret = cow_file_range_inline(inode, actual_end, 0,
1440 : BTRFS_COMPRESS_NONE, NULL, false);
1441 0 : if (ret == 0) {
1442 : /*
1443 : * We use DO_ACCOUNTING here because we need the
1444 : * delalloc_release_metadata to be run _after_ we drop
1445 : * our outstanding extent for clearing delalloc for this
1446 : * range.
1447 : */
1448 0 : extent_clear_unlock_delalloc(inode, start, end,
1449 : locked_page,
1450 : EXTENT_LOCKED | EXTENT_DELALLOC |
1451 : EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1452 : EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1453 : PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1454 0 : *nr_written = *nr_written +
1455 0 : (end - start + PAGE_SIZE) / PAGE_SIZE;
1456 0 : *page_started = 1;
1457 : /*
1458 : * locked_page is locked by the caller of
1459 : * writepage_delalloc(), not locked by
1460 : * __process_pages_contig().
1461 : *
1462 : * We can't let __process_pages_contig() to unlock it,
1463 : * as it doesn't have any subpage::writers recorded.
1464 : *
1465 : * Here we manually unlock the page, since the caller
1466 : * can't use page_started to determine if it's an
1467 : * inline extent or a compressed extent.
1468 : */
1469 0 : unlock_page(locked_page);
1470 0 : goto out;
1471 0 : } else if (ret < 0) {
1472 0 : goto out_unlock;
1473 : }
1474 : }
1475 :
1476 0 : alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1477 :
1478 : /*
1479 : * Relocation relies on the relocated extents to have exactly the same
1480 : * size as the original extents. Normally writeback for relocation data
1481 : * extents follows a NOCOW path because relocation preallocates the
1482 : * extents. However, due to an operation such as scrub turning a block
1483 : * group to RO mode, it may fallback to COW mode, so we must make sure
1484 : * an extent allocated during COW has exactly the requested size and can
1485 : * not be split into smaller extents, otherwise relocation breaks and
1486 : * fails during the stage where it updates the bytenr of file extent
1487 : * items.
1488 : */
1489 0 : if (btrfs_is_data_reloc_root(root))
1490 : min_alloc_size = num_bytes;
1491 : else
1492 0 : min_alloc_size = fs_info->sectorsize;
1493 :
1494 0 : while (num_bytes > 0) {
1495 0 : struct btrfs_ordered_extent *ordered;
1496 :
1497 0 : cur_alloc_size = num_bytes;
1498 0 : ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1499 : min_alloc_size, 0, alloc_hint,
1500 : &ins, 1, 1);
1501 0 : if (ret < 0)
1502 0 : goto out_unlock;
1503 0 : cur_alloc_size = ins.offset;
1504 0 : extent_reserved = true;
1505 :
1506 0 : ram_size = ins.offset;
1507 0 : em = create_io_em(inode, start, ins.offset, /* len */
1508 : start, /* orig_start */
1509 : ins.objectid, /* block_start */
1510 : ins.offset, /* block_len */
1511 : ins.offset, /* orig_block_len */
1512 : ram_size, /* ram_bytes */
1513 : BTRFS_COMPRESS_NONE, /* compress_type */
1514 : BTRFS_ORDERED_REGULAR /* type */);
1515 0 : if (IS_ERR(em)) {
1516 0 : ret = PTR_ERR(em);
1517 0 : goto out_reserve;
1518 : }
1519 0 : free_extent_map(em);
1520 :
1521 0 : ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1522 : ram_size, ins.objectid, cur_alloc_size,
1523 : 0, 1 << BTRFS_ORDERED_REGULAR,
1524 : BTRFS_COMPRESS_NONE);
1525 0 : if (IS_ERR(ordered)) {
1526 0 : ret = PTR_ERR(ordered);
1527 0 : goto out_drop_extent_cache;
1528 : }
1529 :
1530 0 : if (btrfs_is_data_reloc_root(root)) {
1531 0 : ret = btrfs_reloc_clone_csums(ordered);
1532 :
1533 : /*
1534 : * Only drop cache here, and process as normal.
1535 : *
1536 : * We must not allow extent_clear_unlock_delalloc()
1537 : * at out_unlock label to free meta of this ordered
1538 : * extent, as its meta should be freed by
1539 : * btrfs_finish_ordered_io().
1540 : *
1541 : * So we must continue until @start is increased to
1542 : * skip current ordered extent.
1543 : */
1544 0 : if (ret)
1545 0 : btrfs_drop_extent_map_range(inode, start,
1546 0 : start + ram_size - 1,
1547 : false);
1548 : }
1549 0 : btrfs_put_ordered_extent(ordered);
1550 :
1551 0 : btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1552 :
1553 : /*
1554 : * We're not doing compressed IO, don't unlock the first page
1555 : * (which the caller expects to stay locked), don't clear any
1556 : * dirty bits and don't set any writeback bits
1557 : *
1558 : * Do set the Ordered (Private2) bit so we know this page was
1559 : * properly setup for writepage.
1560 : */
1561 0 : page_ops = unlock ? PAGE_UNLOCK : 0;
1562 0 : page_ops |= PAGE_SET_ORDERED;
1563 :
1564 0 : extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1565 : locked_page,
1566 : EXTENT_LOCKED | EXTENT_DELALLOC,
1567 : page_ops);
1568 0 : if (num_bytes < cur_alloc_size)
1569 : num_bytes = 0;
1570 : else
1571 0 : num_bytes -= cur_alloc_size;
1572 0 : alloc_hint = ins.objectid + ins.offset;
1573 0 : start += cur_alloc_size;
1574 0 : extent_reserved = false;
1575 :
1576 : /*
1577 : * btrfs_reloc_clone_csums() error, since start is increased
1578 : * extent_clear_unlock_delalloc() at out_unlock label won't
1579 : * free metadata of current ordered extent, we're OK to exit.
1580 : */
1581 0 : if (ret)
1582 0 : goto out_unlock;
1583 : }
1584 0 : out:
1585 : return ret;
1586 :
1587 : out_drop_extent_cache:
1588 0 : btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1589 0 : out_reserve:
1590 0 : btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1591 0 : btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1592 0 : out_unlock:
1593 : /*
1594 : * If done_offset is non-NULL and ret == -EAGAIN, we expect the
1595 : * caller to write out the successfully allocated region and retry.
1596 : */
1597 0 : if (done_offset && ret == -EAGAIN) {
1598 0 : if (orig_start < start)
1599 0 : *done_offset = start - 1;
1600 : else
1601 0 : *done_offset = start;
1602 0 : return ret;
1603 0 : } else if (ret == -EAGAIN) {
1604 : /* Convert to -ENOSPC since the caller cannot retry. */
1605 0 : ret = -ENOSPC;
1606 : }
1607 :
1608 : /*
1609 : * Now, we have three regions to clean up:
1610 : *
1611 : * |-------(1)----|---(2)---|-------------(3)----------|
1612 : * `- orig_start `- start `- start + cur_alloc_size `- end
1613 : *
1614 : * We process each region below.
1615 : */
1616 :
1617 0 : clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1618 : EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1619 0 : page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1620 :
1621 : /*
1622 : * For the range (1). We have already instantiated the ordered extents
1623 : * for this region. They are cleaned up by
1624 : * btrfs_cleanup_ordered_extents() in e.g,
1625 : * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1626 : * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1627 : * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1628 : * function.
1629 : *
1630 : * However, in case of unlock == 0, we still need to unlock the pages
1631 : * (except @locked_page) to ensure all the pages are unlocked.
1632 : */
1633 0 : if (!unlock && orig_start < start) {
1634 0 : if (!locked_page)
1635 0 : mapping_set_error(inode->vfs_inode.i_mapping, ret);
1636 0 : extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1637 : locked_page, 0, page_ops);
1638 : }
1639 :
1640 : /*
1641 : * For the range (2). If we reserved an extent for our delalloc range
1642 : * (or a subrange) and failed to create the respective ordered extent,
1643 : * then it means that when we reserved the extent we decremented the
1644 : * extent's size from the data space_info's bytes_may_use counter and
1645 : * incremented the space_info's bytes_reserved counter by the same
1646 : * amount. We must make sure extent_clear_unlock_delalloc() does not try
1647 : * to decrement again the data space_info's bytes_may_use counter,
1648 : * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1649 : */
1650 0 : if (extent_reserved) {
1651 0 : extent_clear_unlock_delalloc(inode, start,
1652 0 : start + cur_alloc_size - 1,
1653 : locked_page,
1654 : clear_bits,
1655 : page_ops);
1656 0 : start += cur_alloc_size;
1657 0 : if (start >= end)
1658 : return ret;
1659 : }
1660 :
1661 : /*
1662 : * For the range (3). We never touched the region. In addition to the
1663 : * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1664 : * space_info's bytes_may_use counter, reserved in
1665 : * btrfs_check_data_free_space().
1666 : */
1667 0 : extent_clear_unlock_delalloc(inode, start, end, locked_page,
1668 : clear_bits | EXTENT_CLEAR_DATA_RESV,
1669 : page_ops);
1670 0 : return ret;
1671 : }
1672 :
1673 : /*
1674 : * work queue call back to started compression on a file and pages
1675 : */
1676 0 : static noinline void async_cow_start(struct btrfs_work *work)
1677 : {
1678 0 : struct async_chunk *async_chunk;
1679 0 : int compressed_extents;
1680 :
1681 0 : async_chunk = container_of(work, struct async_chunk, work);
1682 :
1683 0 : compressed_extents = compress_file_range(async_chunk);
1684 0 : if (compressed_extents == 0) {
1685 0 : btrfs_add_delayed_iput(async_chunk->inode);
1686 0 : async_chunk->inode = NULL;
1687 : }
1688 0 : }
1689 :
1690 : /*
1691 : * work queue call back to submit previously compressed pages
1692 : */
1693 0 : static noinline void async_cow_submit(struct btrfs_work *work)
1694 : {
1695 0 : struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1696 : work);
1697 0 : struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1698 0 : unsigned long nr_pages;
1699 :
1700 0 : nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1701 : PAGE_SHIFT;
1702 :
1703 : /*
1704 : * ->inode could be NULL if async_chunk_start has failed to compress,
1705 : * in which case we don't have anything to submit, yet we need to
1706 : * always adjust ->async_delalloc_pages as its paired with the init
1707 : * happening in run_delalloc_compressed
1708 : */
1709 0 : if (async_chunk->inode)
1710 0 : submit_compressed_extents(async_chunk);
1711 :
1712 : /* atomic_sub_return implies a barrier */
1713 0 : if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1714 : 5 * SZ_1M)
1715 0 : cond_wake_up_nomb(&fs_info->async_submit_wait);
1716 0 : }
1717 :
1718 0 : static noinline void async_cow_free(struct btrfs_work *work)
1719 : {
1720 0 : struct async_chunk *async_chunk;
1721 0 : struct async_cow *async_cow;
1722 :
1723 0 : async_chunk = container_of(work, struct async_chunk, work);
1724 0 : if (async_chunk->inode)
1725 0 : btrfs_add_delayed_iput(async_chunk->inode);
1726 0 : if (async_chunk->blkcg_css)
1727 0 : css_put(async_chunk->blkcg_css);
1728 :
1729 0 : async_cow = async_chunk->async_cow;
1730 0 : if (atomic_dec_and_test(&async_cow->num_chunks))
1731 0 : kvfree(async_cow);
1732 0 : }
1733 :
1734 0 : static bool run_delalloc_compressed(struct btrfs_inode *inode,
1735 : struct writeback_control *wbc,
1736 : struct page *locked_page,
1737 : u64 start, u64 end, int *page_started,
1738 : unsigned long *nr_written)
1739 : {
1740 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
1741 0 : struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1742 0 : struct async_cow *ctx;
1743 0 : struct async_chunk *async_chunk;
1744 0 : unsigned long nr_pages;
1745 0 : u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1746 0 : int i;
1747 0 : unsigned nofs_flag;
1748 0 : const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1749 :
1750 0 : nofs_flag = memalloc_nofs_save();
1751 0 : ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1752 0 : memalloc_nofs_restore(nofs_flag);
1753 0 : if (!ctx)
1754 : return false;
1755 :
1756 0 : unlock_extent(&inode->io_tree, start, end, NULL);
1757 0 : set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1758 :
1759 0 : async_chunk = ctx->chunks;
1760 0 : atomic_set(&ctx->num_chunks, num_chunks);
1761 :
1762 0 : for (i = 0; i < num_chunks; i++) {
1763 0 : u64 cur_end = min(end, start + SZ_512K - 1);
1764 :
1765 : /*
1766 : * igrab is called higher up in the call chain, take only the
1767 : * lightweight reference for the callback lifetime
1768 : */
1769 0 : ihold(&inode->vfs_inode);
1770 0 : async_chunk[i].async_cow = ctx;
1771 0 : async_chunk[i].inode = inode;
1772 0 : async_chunk[i].start = start;
1773 0 : async_chunk[i].end = cur_end;
1774 0 : async_chunk[i].write_flags = write_flags;
1775 0 : INIT_LIST_HEAD(&async_chunk[i].extents);
1776 :
1777 : /*
1778 : * The locked_page comes all the way from writepage and its
1779 : * the original page we were actually given. As we spread
1780 : * this large delalloc region across multiple async_chunk
1781 : * structs, only the first struct needs a pointer to locked_page
1782 : *
1783 : * This way we don't need racey decisions about who is supposed
1784 : * to unlock it.
1785 : */
1786 0 : if (locked_page) {
1787 : /*
1788 : * Depending on the compressibility, the pages might or
1789 : * might not go through async. We want all of them to
1790 : * be accounted against wbc once. Let's do it here
1791 : * before the paths diverge. wbc accounting is used
1792 : * only for foreign writeback detection and doesn't
1793 : * need full accuracy. Just account the whole thing
1794 : * against the first page.
1795 : */
1796 0 : wbc_account_cgroup_owner(wbc, locked_page,
1797 0 : cur_end - start);
1798 0 : async_chunk[i].locked_page = locked_page;
1799 0 : locked_page = NULL;
1800 : } else {
1801 0 : async_chunk[i].locked_page = NULL;
1802 : }
1803 :
1804 0 : if (blkcg_css != blkcg_root_css) {
1805 0 : css_get(blkcg_css);
1806 0 : async_chunk[i].blkcg_css = blkcg_css;
1807 0 : async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1808 : } else {
1809 0 : async_chunk[i].blkcg_css = NULL;
1810 : }
1811 :
1812 0 : btrfs_init_work(&async_chunk[i].work, async_cow_start,
1813 : async_cow_submit, async_cow_free);
1814 :
1815 0 : nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1816 0 : atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1817 :
1818 0 : btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1819 :
1820 0 : *nr_written += nr_pages;
1821 0 : start = cur_end + 1;
1822 : }
1823 0 : *page_started = 1;
1824 0 : return true;
1825 : }
1826 :
1827 0 : static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1828 : struct page *locked_page, u64 start,
1829 : u64 end, int *page_started,
1830 : unsigned long *nr_written,
1831 : struct writeback_control *wbc)
1832 : {
1833 0 : u64 done_offset = end;
1834 0 : int ret;
1835 0 : bool locked_page_done = false;
1836 :
1837 0 : while (start <= end) {
1838 0 : ret = cow_file_range(inode, locked_page, start, end, page_started,
1839 : nr_written, 0, &done_offset);
1840 0 : if (ret && ret != -EAGAIN)
1841 0 : return ret;
1842 :
1843 0 : if (*page_started) {
1844 : ASSERT(ret == 0);
1845 : return 0;
1846 : }
1847 :
1848 0 : if (ret == 0)
1849 0 : done_offset = end;
1850 :
1851 0 : if (done_offset == start) {
1852 0 : wait_on_bit_io(&inode->root->fs_info->flags,
1853 : BTRFS_FS_NEED_ZONE_FINISH,
1854 : TASK_UNINTERRUPTIBLE);
1855 0 : continue;
1856 : }
1857 :
1858 0 : if (!locked_page_done) {
1859 0 : __set_page_dirty_nobuffers(locked_page);
1860 0 : account_page_redirty(locked_page);
1861 : }
1862 0 : locked_page_done = true;
1863 0 : extent_write_locked_range(&inode->vfs_inode, start, done_offset,
1864 : wbc);
1865 0 : start = done_offset + 1;
1866 : }
1867 :
1868 0 : *page_started = 1;
1869 :
1870 0 : return 0;
1871 : }
1872 :
1873 0 : static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1874 : u64 bytenr, u64 num_bytes, bool nowait)
1875 : {
1876 0 : struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1877 0 : struct btrfs_ordered_sum *sums;
1878 0 : int ret;
1879 0 : LIST_HEAD(list);
1880 :
1881 0 : ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1882 : &list, 0, nowait);
1883 0 : if (ret == 0 && list_empty(&list))
1884 : return 0;
1885 :
1886 0 : while (!list_empty(&list)) {
1887 0 : sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1888 0 : list_del(&sums->list);
1889 0 : kfree(sums);
1890 : }
1891 0 : if (ret < 0)
1892 0 : return ret;
1893 : return 1;
1894 : }
1895 :
1896 0 : static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1897 : const u64 start, const u64 end,
1898 : int *page_started, unsigned long *nr_written)
1899 : {
1900 0 : const bool is_space_ino = btrfs_is_free_space_inode(inode);
1901 0 : const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1902 0 : const u64 range_bytes = end + 1 - start;
1903 0 : struct extent_io_tree *io_tree = &inode->io_tree;
1904 0 : u64 range_start = start;
1905 0 : u64 count;
1906 :
1907 : /*
1908 : * If EXTENT_NORESERVE is set it means that when the buffered write was
1909 : * made we had not enough available data space and therefore we did not
1910 : * reserve data space for it, since we though we could do NOCOW for the
1911 : * respective file range (either there is prealloc extent or the inode
1912 : * has the NOCOW bit set).
1913 : *
1914 : * However when we need to fallback to COW mode (because for example the
1915 : * block group for the corresponding extent was turned to RO mode by a
1916 : * scrub or relocation) we need to do the following:
1917 : *
1918 : * 1) We increment the bytes_may_use counter of the data space info.
1919 : * If COW succeeds, it allocates a new data extent and after doing
1920 : * that it decrements the space info's bytes_may_use counter and
1921 : * increments its bytes_reserved counter by the same amount (we do
1922 : * this at btrfs_add_reserved_bytes()). So we need to increment the
1923 : * bytes_may_use counter to compensate (when space is reserved at
1924 : * buffered write time, the bytes_may_use counter is incremented);
1925 : *
1926 : * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1927 : * that if the COW path fails for any reason, it decrements (through
1928 : * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1929 : * data space info, which we incremented in the step above.
1930 : *
1931 : * If we need to fallback to cow and the inode corresponds to a free
1932 : * space cache inode or an inode of the data relocation tree, we must
1933 : * also increment bytes_may_use of the data space_info for the same
1934 : * reason. Space caches and relocated data extents always get a prealloc
1935 : * extent for them, however scrub or balance may have set the block
1936 : * group that contains that extent to RO mode and therefore force COW
1937 : * when starting writeback.
1938 : */
1939 0 : count = count_range_bits(io_tree, &range_start, end, range_bytes,
1940 : EXTENT_NORESERVE, 0, NULL);
1941 0 : if (count > 0 || is_space_ino || is_reloc_ino) {
1942 0 : u64 bytes = count;
1943 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
1944 0 : struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1945 :
1946 0 : if (is_space_ino || is_reloc_ino)
1947 0 : bytes = range_bytes;
1948 :
1949 0 : spin_lock(&sinfo->lock);
1950 0 : btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1951 0 : spin_unlock(&sinfo->lock);
1952 :
1953 0 : if (count > 0)
1954 0 : clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1955 : NULL);
1956 : }
1957 :
1958 0 : return cow_file_range(inode, locked_page, start, end, page_started,
1959 : nr_written, 1, NULL);
1960 : }
1961 :
1962 : struct can_nocow_file_extent_args {
1963 : /* Input fields. */
1964 :
1965 : /* Start file offset of the range we want to NOCOW. */
1966 : u64 start;
1967 : /* End file offset (inclusive) of the range we want to NOCOW. */
1968 : u64 end;
1969 : bool writeback_path;
1970 : bool strict;
1971 : /*
1972 : * Free the path passed to can_nocow_file_extent() once it's not needed
1973 : * anymore.
1974 : */
1975 : bool free_path;
1976 :
1977 : /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1978 :
1979 : u64 disk_bytenr;
1980 : u64 disk_num_bytes;
1981 : u64 extent_offset;
1982 : /* Number of bytes that can be written to in NOCOW mode. */
1983 : u64 num_bytes;
1984 : };
1985 :
1986 : /*
1987 : * Check if we can NOCOW the file extent that the path points to.
1988 : * This function may return with the path released, so the caller should check
1989 : * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1990 : *
1991 : * Returns: < 0 on error
1992 : * 0 if we can not NOCOW
1993 : * 1 if we can NOCOW
1994 : */
1995 0 : static int can_nocow_file_extent(struct btrfs_path *path,
1996 : struct btrfs_key *key,
1997 : struct btrfs_inode *inode,
1998 : struct can_nocow_file_extent_args *args)
1999 : {
2000 0 : const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
2001 0 : struct extent_buffer *leaf = path->nodes[0];
2002 0 : struct btrfs_root *root = inode->root;
2003 0 : struct btrfs_file_extent_item *fi;
2004 0 : u64 extent_end;
2005 0 : u8 extent_type;
2006 0 : int can_nocow = 0;
2007 0 : int ret = 0;
2008 0 : bool nowait = path->nowait;
2009 :
2010 0 : fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
2011 0 : extent_type = btrfs_file_extent_type(leaf, fi);
2012 :
2013 0 : if (extent_type == BTRFS_FILE_EXTENT_INLINE)
2014 0 : goto out;
2015 :
2016 : /* Can't access these fields unless we know it's not an inline extent. */
2017 0 : args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2018 0 : args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
2019 0 : args->extent_offset = btrfs_file_extent_offset(leaf, fi);
2020 :
2021 0 : if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
2022 : extent_type == BTRFS_FILE_EXTENT_REG)
2023 0 : goto out;
2024 :
2025 : /*
2026 : * If the extent was created before the generation where the last snapshot
2027 : * for its subvolume was created, then this implies the extent is shared,
2028 : * hence we must COW.
2029 : */
2030 0 : if (!args->strict &&
2031 : btrfs_file_extent_generation(leaf, fi) <=
2032 : btrfs_root_last_snapshot(&root->root_item))
2033 0 : goto out;
2034 :
2035 : /* An explicit hole, must COW. */
2036 0 : if (args->disk_bytenr == 0)
2037 0 : goto out;
2038 :
2039 : /* Compressed/encrypted/encoded extents must be COWed. */
2040 0 : if (btrfs_file_extent_compression(leaf, fi) ||
2041 0 : btrfs_file_extent_encryption(leaf, fi) ||
2042 : btrfs_file_extent_other_encoding(leaf, fi))
2043 0 : goto out;
2044 :
2045 0 : extent_end = btrfs_file_extent_end(path);
2046 :
2047 : /*
2048 : * The following checks can be expensive, as they need to take other
2049 : * locks and do btree or rbtree searches, so release the path to avoid
2050 : * blocking other tasks for too long.
2051 : */
2052 0 : btrfs_release_path(path);
2053 :
2054 0 : ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
2055 0 : key->offset - args->extent_offset,
2056 : args->disk_bytenr, args->strict, path);
2057 0 : WARN_ON_ONCE(ret > 0 && is_freespace_inode);
2058 0 : if (ret != 0)
2059 0 : goto out;
2060 :
2061 0 : if (args->free_path) {
2062 : /*
2063 : * We don't need the path anymore, plus through the
2064 : * csum_exist_in_range() call below we will end up allocating
2065 : * another path. So free the path to avoid unnecessary extra
2066 : * memory usage.
2067 : */
2068 0 : btrfs_free_path(path);
2069 0 : path = NULL;
2070 : }
2071 :
2072 : /* If there are pending snapshots for this root, we must COW. */
2073 0 : if (args->writeback_path && !is_freespace_inode &&
2074 : atomic_read(&root->snapshot_force_cow))
2075 0 : goto out;
2076 :
2077 0 : args->disk_bytenr += args->extent_offset;
2078 0 : args->disk_bytenr += args->start - key->offset;
2079 0 : args->num_bytes = min(args->end + 1, extent_end) - args->start;
2080 :
2081 : /*
2082 : * Force COW if csums exist in the range. This ensures that csums for a
2083 : * given extent are either valid or do not exist.
2084 : */
2085 0 : ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
2086 : nowait);
2087 0 : WARN_ON_ONCE(ret > 0 && is_freespace_inode);
2088 0 : if (ret != 0)
2089 0 : goto out;
2090 :
2091 : can_nocow = 1;
2092 0 : out:
2093 0 : if (args->free_path && path)
2094 0 : btrfs_free_path(path);
2095 :
2096 0 : return ret < 0 ? ret : can_nocow;
2097 : }
2098 :
2099 : /*
2100 : * when nowcow writeback call back. This checks for snapshots or COW copies
2101 : * of the extents that exist in the file, and COWs the file as required.
2102 : *
2103 : * If no cow copies or snapshots exist, we write directly to the existing
2104 : * blocks on disk
2105 : */
2106 0 : static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
2107 : struct page *locked_page,
2108 : const u64 start, const u64 end,
2109 : int *page_started,
2110 : unsigned long *nr_written)
2111 : {
2112 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
2113 0 : struct btrfs_root *root = inode->root;
2114 0 : struct btrfs_path *path;
2115 0 : u64 cow_start = (u64)-1;
2116 0 : u64 cur_offset = start;
2117 0 : int ret;
2118 0 : bool check_prev = true;
2119 0 : u64 ino = btrfs_ino(inode);
2120 0 : struct btrfs_block_group *bg;
2121 0 : bool nocow = false;
2122 0 : struct can_nocow_file_extent_args nocow_args = { 0 };
2123 :
2124 0 : path = btrfs_alloc_path();
2125 0 : if (!path) {
2126 0 : extent_clear_unlock_delalloc(inode, start, end, locked_page,
2127 : EXTENT_LOCKED | EXTENT_DELALLOC |
2128 : EXTENT_DO_ACCOUNTING |
2129 : EXTENT_DEFRAG, PAGE_UNLOCK |
2130 : PAGE_START_WRITEBACK |
2131 : PAGE_END_WRITEBACK);
2132 0 : return -ENOMEM;
2133 : }
2134 :
2135 0 : nocow_args.end = end;
2136 0 : nocow_args.writeback_path = true;
2137 :
2138 0 : while (1) {
2139 0 : struct btrfs_ordered_extent *ordered;
2140 0 : struct btrfs_key found_key;
2141 0 : struct btrfs_file_extent_item *fi;
2142 0 : struct extent_buffer *leaf;
2143 0 : u64 extent_end;
2144 0 : u64 ram_bytes;
2145 0 : u64 nocow_end;
2146 0 : int extent_type;
2147 0 : bool is_prealloc;
2148 :
2149 0 : nocow = false;
2150 :
2151 0 : ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2152 : cur_offset, 0);
2153 0 : if (ret < 0)
2154 0 : goto error;
2155 :
2156 : /*
2157 : * If there is no extent for our range when doing the initial
2158 : * search, then go back to the previous slot as it will be the
2159 : * one containing the search offset
2160 : */
2161 0 : if (ret > 0 && path->slots[0] > 0 && check_prev) {
2162 0 : leaf = path->nodes[0];
2163 0 : btrfs_item_key_to_cpu(leaf, &found_key,
2164 : path->slots[0] - 1);
2165 0 : if (found_key.objectid == ino &&
2166 0 : found_key.type == BTRFS_EXTENT_DATA_KEY)
2167 0 : path->slots[0]--;
2168 : }
2169 : check_prev = false;
2170 : next_slot:
2171 : /* Go to next leaf if we have exhausted the current one */
2172 0 : leaf = path->nodes[0];
2173 0 : if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2174 0 : ret = btrfs_next_leaf(root, path);
2175 0 : if (ret < 0) {
2176 0 : if (cow_start != (u64)-1)
2177 0 : cur_offset = cow_start;
2178 0 : goto error;
2179 : }
2180 0 : if (ret > 0)
2181 : break;
2182 0 : leaf = path->nodes[0];
2183 : }
2184 :
2185 0 : btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2186 :
2187 : /* Didn't find anything for our INO */
2188 0 : if (found_key.objectid > ino)
2189 : break;
2190 : /*
2191 : * Keep searching until we find an EXTENT_ITEM or there are no
2192 : * more extents for this inode
2193 : */
2194 0 : if (WARN_ON_ONCE(found_key.objectid < ino) ||
2195 0 : found_key.type < BTRFS_EXTENT_DATA_KEY) {
2196 0 : path->slots[0]++;
2197 0 : goto next_slot;
2198 : }
2199 :
2200 : /* Found key is not EXTENT_DATA_KEY or starts after req range */
2201 0 : if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2202 0 : found_key.offset > end)
2203 : break;
2204 :
2205 : /*
2206 : * If the found extent starts after requested offset, then
2207 : * adjust extent_end to be right before this extent begins
2208 : */
2209 0 : if (found_key.offset > cur_offset) {
2210 0 : extent_end = found_key.offset;
2211 0 : extent_type = 0;
2212 0 : goto out_check;
2213 : }
2214 :
2215 : /*
2216 : * Found extent which begins before our range and potentially
2217 : * intersect it
2218 : */
2219 0 : fi = btrfs_item_ptr(leaf, path->slots[0],
2220 : struct btrfs_file_extent_item);
2221 0 : extent_type = btrfs_file_extent_type(leaf, fi);
2222 : /* If this is triggered then we have a memory corruption. */
2223 0 : ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2224 0 : if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2225 0 : ret = -EUCLEAN;
2226 0 : goto error;
2227 : }
2228 0 : ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2229 0 : extent_end = btrfs_file_extent_end(path);
2230 :
2231 : /*
2232 : * If the extent we got ends before our current offset, skip to
2233 : * the next extent.
2234 : */
2235 0 : if (extent_end <= cur_offset) {
2236 0 : path->slots[0]++;
2237 0 : goto next_slot;
2238 : }
2239 :
2240 0 : nocow_args.start = cur_offset;
2241 0 : ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2242 0 : if (ret < 0) {
2243 0 : if (cow_start != (u64)-1)
2244 0 : cur_offset = cow_start;
2245 0 : goto error;
2246 0 : } else if (ret == 0) {
2247 0 : goto out_check;
2248 : }
2249 :
2250 0 : ret = 0;
2251 0 : bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2252 0 : if (bg)
2253 0 : nocow = true;
2254 0 : out_check:
2255 : /*
2256 : * If nocow is false then record the beginning of the range
2257 : * that needs to be COWed
2258 : */
2259 0 : if (!nocow) {
2260 0 : if (cow_start == (u64)-1)
2261 0 : cow_start = cur_offset;
2262 0 : cur_offset = extent_end;
2263 0 : if (cur_offset > end)
2264 : break;
2265 0 : if (!path->nodes[0])
2266 0 : continue;
2267 0 : path->slots[0]++;
2268 0 : goto next_slot;
2269 : }
2270 :
2271 : /*
2272 : * COW range from cow_start to found_key.offset - 1. As the key
2273 : * will contain the beginning of the first extent that can be
2274 : * NOCOW, following one which needs to be COW'ed
2275 : */
2276 0 : if (cow_start != (u64)-1) {
2277 0 : ret = fallback_to_cow(inode, locked_page,
2278 0 : cow_start, found_key.offset - 1,
2279 : page_started, nr_written);
2280 0 : if (ret)
2281 0 : goto error;
2282 : cow_start = (u64)-1;
2283 : }
2284 :
2285 0 : nocow_end = cur_offset + nocow_args.num_bytes - 1;
2286 0 : is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2287 0 : if (is_prealloc) {
2288 0 : u64 orig_start = found_key.offset - nocow_args.extent_offset;
2289 0 : struct extent_map *em;
2290 :
2291 0 : em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2292 : orig_start,
2293 : nocow_args.disk_bytenr, /* block_start */
2294 : nocow_args.num_bytes, /* block_len */
2295 : nocow_args.disk_num_bytes, /* orig_block_len */
2296 : ram_bytes, BTRFS_COMPRESS_NONE,
2297 : BTRFS_ORDERED_PREALLOC);
2298 0 : if (IS_ERR(em)) {
2299 0 : ret = PTR_ERR(em);
2300 0 : goto error;
2301 : }
2302 0 : free_extent_map(em);
2303 : }
2304 :
2305 0 : ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2306 : nocow_args.num_bytes, nocow_args.num_bytes,
2307 : nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2308 : is_prealloc
2309 : ? (1 << BTRFS_ORDERED_PREALLOC)
2310 : : (1 << BTRFS_ORDERED_NOCOW),
2311 : BTRFS_COMPRESS_NONE);
2312 0 : if (IS_ERR(ordered)) {
2313 0 : if (is_prealloc) {
2314 0 : btrfs_drop_extent_map_range(inode, cur_offset,
2315 : nocow_end, false);
2316 : }
2317 0 : ret = PTR_ERR(ordered);
2318 0 : goto error;
2319 : }
2320 :
2321 0 : if (nocow) {
2322 0 : btrfs_dec_nocow_writers(bg);
2323 0 : nocow = false;
2324 : }
2325 :
2326 0 : if (btrfs_is_data_reloc_root(root))
2327 : /*
2328 : * Error handled later, as we must prevent
2329 : * extent_clear_unlock_delalloc() in error handler
2330 : * from freeing metadata of created ordered extent.
2331 : */
2332 0 : ret = btrfs_reloc_clone_csums(ordered);
2333 0 : btrfs_put_ordered_extent(ordered);
2334 :
2335 0 : extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2336 : locked_page, EXTENT_LOCKED |
2337 : EXTENT_DELALLOC |
2338 : EXTENT_CLEAR_DATA_RESV,
2339 : PAGE_UNLOCK | PAGE_SET_ORDERED);
2340 :
2341 0 : cur_offset = extent_end;
2342 :
2343 : /*
2344 : * btrfs_reloc_clone_csums() error, now we're OK to call error
2345 : * handler, as metadata for created ordered extent will only
2346 : * be freed by btrfs_finish_ordered_io().
2347 : */
2348 0 : if (ret)
2349 0 : goto error;
2350 0 : if (cur_offset > end)
2351 : break;
2352 : }
2353 0 : btrfs_release_path(path);
2354 :
2355 0 : if (cur_offset <= end && cow_start == (u64)-1)
2356 0 : cow_start = cur_offset;
2357 :
2358 0 : if (cow_start != (u64)-1) {
2359 0 : cur_offset = end;
2360 0 : ret = fallback_to_cow(inode, locked_page, cow_start, end,
2361 : page_started, nr_written);
2362 0 : if (ret)
2363 0 : goto error;
2364 : }
2365 :
2366 0 : error:
2367 0 : if (nocow)
2368 0 : btrfs_dec_nocow_writers(bg);
2369 :
2370 0 : if (ret && cur_offset < end)
2371 0 : extent_clear_unlock_delalloc(inode, cur_offset, end,
2372 : locked_page, EXTENT_LOCKED |
2373 : EXTENT_DELALLOC | EXTENT_DEFRAG |
2374 : EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2375 : PAGE_START_WRITEBACK |
2376 : PAGE_END_WRITEBACK);
2377 0 : btrfs_free_path(path);
2378 0 : return ret;
2379 : }
2380 :
2381 0 : static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2382 : {
2383 0 : if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2384 0 : if (inode->defrag_bytes &&
2385 0 : test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2386 : 0, NULL))
2387 : return false;
2388 0 : return true;
2389 : }
2390 : return false;
2391 : }
2392 :
2393 : /*
2394 : * Function to process delayed allocation (create CoW) for ranges which are
2395 : * being touched for the first time.
2396 : */
2397 0 : int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2398 : u64 start, u64 end, int *page_started, unsigned long *nr_written,
2399 : struct writeback_control *wbc)
2400 : {
2401 0 : int ret = 0;
2402 0 : const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2403 :
2404 : /*
2405 : * The range must cover part of the @locked_page, or the returned
2406 : * @page_started can confuse the caller.
2407 : */
2408 0 : ASSERT(!(end <= page_offset(locked_page) ||
2409 : start >= page_offset(locked_page) + PAGE_SIZE));
2410 :
2411 0 : if (should_nocow(inode, start, end)) {
2412 : /*
2413 : * Normally on a zoned device we're only doing COW writes, but
2414 : * in case of relocation on a zoned filesystem we have taken
2415 : * precaution, that we're only writing sequentially. It's safe
2416 : * to use run_delalloc_nocow() here, like for regular
2417 : * preallocated inodes.
2418 : */
2419 0 : ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2420 0 : ret = run_delalloc_nocow(inode, locked_page, start, end,
2421 : page_started, nr_written);
2422 0 : goto out;
2423 : }
2424 :
2425 0 : if (btrfs_inode_can_compress(inode) &&
2426 0 : inode_need_compress(inode, start, end) &&
2427 0 : run_delalloc_compressed(inode, wbc, locked_page, start,
2428 : end, page_started, nr_written))
2429 0 : goto out;
2430 :
2431 0 : if (zoned)
2432 0 : ret = run_delalloc_zoned(inode, locked_page, start, end,
2433 : page_started, nr_written, wbc);
2434 : else
2435 0 : ret = cow_file_range(inode, locked_page, start, end,
2436 : page_started, nr_written, 1, NULL);
2437 :
2438 0 : out:
2439 0 : ASSERT(ret <= 0);
2440 0 : if (ret)
2441 0 : btrfs_cleanup_ordered_extents(inode, locked_page, start,
2442 0 : end - start + 1);
2443 0 : return ret;
2444 : }
2445 :
2446 0 : void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2447 : struct extent_state *orig, u64 split)
2448 : {
2449 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
2450 0 : u64 size;
2451 :
2452 : /* not delalloc, ignore it */
2453 0 : if (!(orig->state & EXTENT_DELALLOC))
2454 : return;
2455 :
2456 0 : size = orig->end - orig->start + 1;
2457 0 : if (size > fs_info->max_extent_size) {
2458 0 : u32 num_extents;
2459 0 : u64 new_size;
2460 :
2461 : /*
2462 : * See the explanation in btrfs_merge_delalloc_extent, the same
2463 : * applies here, just in reverse.
2464 : */
2465 0 : new_size = orig->end - split + 1;
2466 0 : num_extents = count_max_extents(fs_info, new_size);
2467 0 : new_size = split - orig->start;
2468 0 : num_extents += count_max_extents(fs_info, new_size);
2469 0 : if (count_max_extents(fs_info, size) >= num_extents)
2470 : return;
2471 : }
2472 :
2473 0 : spin_lock(&inode->lock);
2474 0 : btrfs_mod_outstanding_extents(inode, 1);
2475 0 : spin_unlock(&inode->lock);
2476 : }
2477 :
2478 : /*
2479 : * Handle merged delayed allocation extents so we can keep track of new extents
2480 : * that are just merged onto old extents, such as when we are doing sequential
2481 : * writes, so we can properly account for the metadata space we'll need.
2482 : */
2483 0 : void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2484 : struct extent_state *other)
2485 : {
2486 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
2487 0 : u64 new_size, old_size;
2488 0 : u32 num_extents;
2489 :
2490 : /* not delalloc, ignore it */
2491 0 : if (!(other->state & EXTENT_DELALLOC))
2492 : return;
2493 :
2494 0 : if (new->start > other->start)
2495 0 : new_size = new->end - other->start + 1;
2496 : else
2497 0 : new_size = other->end - new->start + 1;
2498 :
2499 : /* we're not bigger than the max, unreserve the space and go */
2500 0 : if (new_size <= fs_info->max_extent_size) {
2501 0 : spin_lock(&inode->lock);
2502 0 : btrfs_mod_outstanding_extents(inode, -1);
2503 0 : spin_unlock(&inode->lock);
2504 0 : return;
2505 : }
2506 :
2507 : /*
2508 : * We have to add up either side to figure out how many extents were
2509 : * accounted for before we merged into one big extent. If the number of
2510 : * extents we accounted for is <= the amount we need for the new range
2511 : * then we can return, otherwise drop. Think of it like this
2512 : *
2513 : * [ 4k][MAX_SIZE]
2514 : *
2515 : * So we've grown the extent by a MAX_SIZE extent, this would mean we
2516 : * need 2 outstanding extents, on one side we have 1 and the other side
2517 : * we have 1 so they are == and we can return. But in this case
2518 : *
2519 : * [MAX_SIZE+4k][MAX_SIZE+4k]
2520 : *
2521 : * Each range on their own accounts for 2 extents, but merged together
2522 : * they are only 3 extents worth of accounting, so we need to drop in
2523 : * this case.
2524 : */
2525 0 : old_size = other->end - other->start + 1;
2526 0 : num_extents = count_max_extents(fs_info, old_size);
2527 0 : old_size = new->end - new->start + 1;
2528 0 : num_extents += count_max_extents(fs_info, old_size);
2529 0 : if (count_max_extents(fs_info, new_size) >= num_extents)
2530 : return;
2531 :
2532 0 : spin_lock(&inode->lock);
2533 0 : btrfs_mod_outstanding_extents(inode, -1);
2534 0 : spin_unlock(&inode->lock);
2535 : }
2536 :
2537 0 : static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2538 : struct btrfs_inode *inode)
2539 : {
2540 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
2541 :
2542 0 : spin_lock(&root->delalloc_lock);
2543 0 : if (list_empty(&inode->delalloc_inodes)) {
2544 0 : list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2545 0 : set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags);
2546 0 : root->nr_delalloc_inodes++;
2547 0 : if (root->nr_delalloc_inodes == 1) {
2548 0 : spin_lock(&fs_info->delalloc_root_lock);
2549 0 : BUG_ON(!list_empty(&root->delalloc_root));
2550 0 : list_add_tail(&root->delalloc_root,
2551 : &fs_info->delalloc_roots);
2552 0 : spin_unlock(&fs_info->delalloc_root_lock);
2553 : }
2554 : }
2555 0 : spin_unlock(&root->delalloc_lock);
2556 0 : }
2557 :
2558 0 : void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2559 : struct btrfs_inode *inode)
2560 : {
2561 0 : struct btrfs_fs_info *fs_info = root->fs_info;
2562 :
2563 0 : if (!list_empty(&inode->delalloc_inodes)) {
2564 0 : list_del_init(&inode->delalloc_inodes);
2565 0 : clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2566 0 : &inode->runtime_flags);
2567 0 : root->nr_delalloc_inodes--;
2568 0 : if (!root->nr_delalloc_inodes) {
2569 0 : ASSERT(list_empty(&root->delalloc_inodes));
2570 0 : spin_lock(&fs_info->delalloc_root_lock);
2571 0 : BUG_ON(list_empty(&root->delalloc_root));
2572 0 : list_del_init(&root->delalloc_root);
2573 0 : spin_unlock(&fs_info->delalloc_root_lock);
2574 : }
2575 : }
2576 0 : }
2577 :
2578 0 : static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2579 : struct btrfs_inode *inode)
2580 : {
2581 0 : spin_lock(&root->delalloc_lock);
2582 0 : __btrfs_del_delalloc_inode(root, inode);
2583 0 : spin_unlock(&root->delalloc_lock);
2584 0 : }
2585 :
2586 : /*
2587 : * Properly track delayed allocation bytes in the inode and to maintain the
2588 : * list of inodes that have pending delalloc work to be done.
2589 : */
2590 0 : void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2591 : u32 bits)
2592 : {
2593 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
2594 :
2595 0 : if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2596 0 : WARN_ON(1);
2597 : /*
2598 : * set_bit and clear bit hooks normally require _irqsave/restore
2599 : * but in this case, we are only testing for the DELALLOC
2600 : * bit, which is only set or cleared with irqs on
2601 : */
2602 0 : if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2603 0 : struct btrfs_root *root = inode->root;
2604 0 : u64 len = state->end + 1 - state->start;
2605 0 : u32 num_extents = count_max_extents(fs_info, len);
2606 0 : bool do_list = !btrfs_is_free_space_inode(inode);
2607 :
2608 0 : spin_lock(&inode->lock);
2609 0 : btrfs_mod_outstanding_extents(inode, num_extents);
2610 0 : spin_unlock(&inode->lock);
2611 :
2612 : /* For sanity tests */
2613 0 : if (btrfs_is_testing(fs_info))
2614 : return;
2615 :
2616 0 : percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2617 : fs_info->delalloc_batch);
2618 0 : spin_lock(&inode->lock);
2619 0 : inode->delalloc_bytes += len;
2620 0 : if (bits & EXTENT_DEFRAG)
2621 0 : inode->defrag_bytes += len;
2622 0 : if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2623 : &inode->runtime_flags))
2624 0 : btrfs_add_delalloc_inodes(root, inode);
2625 0 : spin_unlock(&inode->lock);
2626 : }
2627 :
2628 0 : if (!(state->state & EXTENT_DELALLOC_NEW) &&
2629 0 : (bits & EXTENT_DELALLOC_NEW)) {
2630 0 : spin_lock(&inode->lock);
2631 0 : inode->new_delalloc_bytes += state->end + 1 - state->start;
2632 0 : spin_unlock(&inode->lock);
2633 : }
2634 : }
2635 :
2636 : /*
2637 : * Once a range is no longer delalloc this function ensures that proper
2638 : * accounting happens.
2639 : */
2640 0 : void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2641 : struct extent_state *state, u32 bits)
2642 : {
2643 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
2644 0 : u64 len = state->end + 1 - state->start;
2645 0 : u32 num_extents = count_max_extents(fs_info, len);
2646 :
2647 0 : if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2648 0 : spin_lock(&inode->lock);
2649 0 : inode->defrag_bytes -= len;
2650 0 : spin_unlock(&inode->lock);
2651 : }
2652 :
2653 : /*
2654 : * set_bit and clear bit hooks normally require _irqsave/restore
2655 : * but in this case, we are only testing for the DELALLOC
2656 : * bit, which is only set or cleared with irqs on
2657 : */
2658 0 : if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2659 0 : struct btrfs_root *root = inode->root;
2660 0 : bool do_list = !btrfs_is_free_space_inode(inode);
2661 :
2662 0 : spin_lock(&inode->lock);
2663 0 : btrfs_mod_outstanding_extents(inode, -num_extents);
2664 0 : spin_unlock(&inode->lock);
2665 :
2666 : /*
2667 : * We don't reserve metadata space for space cache inodes so we
2668 : * don't need to call delalloc_release_metadata if there is an
2669 : * error.
2670 : */
2671 0 : if (bits & EXTENT_CLEAR_META_RESV &&
2672 0 : root != fs_info->tree_root)
2673 0 : btrfs_delalloc_release_metadata(inode, len, false);
2674 :
2675 : /* For sanity tests. */
2676 0 : if (btrfs_is_testing(fs_info))
2677 : return;
2678 :
2679 0 : if (!btrfs_is_data_reloc_root(root) &&
2680 0 : do_list && !(state->state & EXTENT_NORESERVE) &&
2681 0 : (bits & EXTENT_CLEAR_DATA_RESV))
2682 0 : btrfs_free_reserved_data_space_noquota(fs_info, len);
2683 :
2684 0 : percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2685 : fs_info->delalloc_batch);
2686 0 : spin_lock(&inode->lock);
2687 0 : inode->delalloc_bytes -= len;
2688 0 : if (do_list && inode->delalloc_bytes == 0 &&
2689 0 : test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2690 : &inode->runtime_flags))
2691 0 : btrfs_del_delalloc_inode(root, inode);
2692 0 : spin_unlock(&inode->lock);
2693 : }
2694 :
2695 0 : if ((state->state & EXTENT_DELALLOC_NEW) &&
2696 0 : (bits & EXTENT_DELALLOC_NEW)) {
2697 0 : spin_lock(&inode->lock);
2698 0 : ASSERT(inode->new_delalloc_bytes >= len);
2699 0 : inode->new_delalloc_bytes -= len;
2700 0 : if (bits & EXTENT_ADD_INODE_BYTES)
2701 0 : inode_add_bytes(&inode->vfs_inode, len);
2702 0 : spin_unlock(&inode->lock);
2703 : }
2704 : }
2705 :
2706 0 : static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2707 : struct btrfs_ordered_extent *ordered)
2708 : {
2709 0 : u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2710 0 : u64 len = bbio->bio.bi_iter.bi_size;
2711 0 : struct btrfs_ordered_extent *new;
2712 0 : int ret;
2713 :
2714 : /* Must always be called for the beginning of an ordered extent. */
2715 0 : if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2716 : return -EINVAL;
2717 :
2718 : /* No need to split if the ordered extent covers the entire bio. */
2719 0 : if (ordered->disk_num_bytes == len) {
2720 0 : refcount_inc(&ordered->refs);
2721 0 : bbio->ordered = ordered;
2722 0 : return 0;
2723 : }
2724 :
2725 : /*
2726 : * Don't split the extent_map for NOCOW extents, as we're writing into
2727 : * a pre-existing one.
2728 : */
2729 0 : if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2730 0 : ret = split_extent_map(bbio->inode, bbio->file_offset,
2731 : ordered->num_bytes, len,
2732 : ordered->disk_bytenr);
2733 0 : if (ret)
2734 : return ret;
2735 : }
2736 :
2737 0 : new = btrfs_split_ordered_extent(ordered, len);
2738 0 : if (IS_ERR(new))
2739 0 : return PTR_ERR(new);
2740 0 : bbio->ordered = new;
2741 0 : return 0;
2742 : }
2743 :
2744 : /*
2745 : * given a list of ordered sums record them in the inode. This happens
2746 : * at IO completion time based on sums calculated at bio submission time.
2747 : */
2748 0 : static int add_pending_csums(struct btrfs_trans_handle *trans,
2749 : struct list_head *list)
2750 : {
2751 0 : struct btrfs_ordered_sum *sum;
2752 0 : struct btrfs_root *csum_root = NULL;
2753 0 : int ret;
2754 :
2755 0 : list_for_each_entry(sum, list, list) {
2756 0 : trans->adding_csums = true;
2757 0 : if (!csum_root)
2758 0 : csum_root = btrfs_csum_root(trans->fs_info,
2759 : sum->logical);
2760 0 : ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2761 0 : trans->adding_csums = false;
2762 0 : if (ret)
2763 0 : return ret;
2764 : }
2765 : return 0;
2766 : }
2767 :
2768 0 : static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2769 : const u64 start,
2770 : const u64 len,
2771 : struct extent_state **cached_state)
2772 : {
2773 0 : u64 search_start = start;
2774 0 : const u64 end = start + len - 1;
2775 :
2776 0 : while (search_start < end) {
2777 0 : const u64 search_len = end - search_start + 1;
2778 0 : struct extent_map *em;
2779 0 : u64 em_len;
2780 0 : int ret = 0;
2781 :
2782 0 : em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2783 0 : if (IS_ERR(em))
2784 0 : return PTR_ERR(em);
2785 :
2786 0 : if (em->block_start != EXTENT_MAP_HOLE)
2787 0 : goto next;
2788 :
2789 0 : em_len = em->len;
2790 0 : if (em->start < search_start)
2791 0 : em_len -= search_start - em->start;
2792 0 : if (em_len > search_len)
2793 : em_len = search_len;
2794 :
2795 0 : ret = set_extent_bit(&inode->io_tree, search_start,
2796 0 : search_start + em_len - 1,
2797 : EXTENT_DELALLOC_NEW, cached_state);
2798 0 : next:
2799 0 : search_start = extent_map_end(em);
2800 0 : free_extent_map(em);
2801 0 : if (ret)
2802 0 : return ret;
2803 : }
2804 : return 0;
2805 : }
2806 :
2807 0 : int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2808 : unsigned int extra_bits,
2809 : struct extent_state **cached_state)
2810 : {
2811 0 : WARN_ON(PAGE_ALIGNED(end));
2812 :
2813 0 : if (start >= i_size_read(&inode->vfs_inode) &&
2814 0 : !(inode->flags & BTRFS_INODE_PREALLOC)) {
2815 : /*
2816 : * There can't be any extents following eof in this case so just
2817 : * set the delalloc new bit for the range directly.
2818 : */
2819 0 : extra_bits |= EXTENT_DELALLOC_NEW;
2820 : } else {
2821 0 : int ret;
2822 :
2823 0 : ret = btrfs_find_new_delalloc_bytes(inode, start,
2824 0 : end + 1 - start,
2825 : cached_state);
2826 0 : if (ret)
2827 : return ret;
2828 : }
2829 :
2830 0 : return set_extent_bit(&inode->io_tree, start, end,
2831 : EXTENT_DELALLOC | extra_bits, cached_state);
2832 : }
2833 :
2834 : /* see btrfs_writepage_start_hook for details on why this is required */
2835 : struct btrfs_writepage_fixup {
2836 : struct page *page;
2837 : struct btrfs_inode *inode;
2838 : struct btrfs_work work;
2839 : };
2840 :
2841 0 : static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2842 : {
2843 0 : struct btrfs_writepage_fixup *fixup;
2844 0 : struct btrfs_ordered_extent *ordered;
2845 0 : struct extent_state *cached_state = NULL;
2846 0 : struct extent_changeset *data_reserved = NULL;
2847 0 : struct page *page;
2848 0 : struct btrfs_inode *inode;
2849 0 : u64 page_start;
2850 0 : u64 page_end;
2851 0 : int ret = 0;
2852 0 : bool free_delalloc_space = true;
2853 :
2854 0 : fixup = container_of(work, struct btrfs_writepage_fixup, work);
2855 0 : page = fixup->page;
2856 0 : inode = fixup->inode;
2857 0 : page_start = page_offset(page);
2858 0 : page_end = page_offset(page) + PAGE_SIZE - 1;
2859 :
2860 : /*
2861 : * This is similar to page_mkwrite, we need to reserve the space before
2862 : * we take the page lock.
2863 : */
2864 0 : ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2865 : PAGE_SIZE);
2866 0 : again:
2867 0 : lock_page(page);
2868 :
2869 : /*
2870 : * Before we queued this fixup, we took a reference on the page.
2871 : * page->mapping may go NULL, but it shouldn't be moved to a different
2872 : * address space.
2873 : */
2874 0 : if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2875 : /*
2876 : * Unfortunately this is a little tricky, either
2877 : *
2878 : * 1) We got here and our page had already been dealt with and
2879 : * we reserved our space, thus ret == 0, so we need to just
2880 : * drop our space reservation and bail. This can happen the
2881 : * first time we come into the fixup worker, or could happen
2882 : * while waiting for the ordered extent.
2883 : * 2) Our page was already dealt with, but we happened to get an
2884 : * ENOSPC above from the btrfs_delalloc_reserve_space. In
2885 : * this case we obviously don't have anything to release, but
2886 : * because the page was already dealt with we don't want to
2887 : * mark the page with an error, so make sure we're resetting
2888 : * ret to 0. This is why we have this check _before_ the ret
2889 : * check, because we do not want to have a surprise ENOSPC
2890 : * when the page was already properly dealt with.
2891 : */
2892 0 : if (!ret) {
2893 0 : btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2894 0 : btrfs_delalloc_release_space(inode, data_reserved,
2895 : page_start, PAGE_SIZE,
2896 : true);
2897 : }
2898 0 : ret = 0;
2899 0 : goto out_page;
2900 : }
2901 :
2902 : /*
2903 : * We can't mess with the page state unless it is locked, so now that
2904 : * it is locked bail if we failed to make our space reservation.
2905 : */
2906 0 : if (ret)
2907 0 : goto out_page;
2908 :
2909 0 : lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2910 :
2911 : /* already ordered? We're done */
2912 0 : if (PageOrdered(page))
2913 0 : goto out_reserved;
2914 :
2915 0 : ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2916 0 : if (ordered) {
2917 0 : unlock_extent(&inode->io_tree, page_start, page_end,
2918 : &cached_state);
2919 0 : unlock_page(page);
2920 0 : btrfs_start_ordered_extent(ordered);
2921 0 : btrfs_put_ordered_extent(ordered);
2922 0 : goto again;
2923 : }
2924 :
2925 0 : ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2926 : &cached_state);
2927 0 : if (ret)
2928 0 : goto out_reserved;
2929 :
2930 : /*
2931 : * Everything went as planned, we're now the owner of a dirty page with
2932 : * delayed allocation bits set and space reserved for our COW
2933 : * destination.
2934 : *
2935 : * The page was dirty when we started, nothing should have cleaned it.
2936 : */
2937 0 : BUG_ON(!PageDirty(page));
2938 : free_delalloc_space = false;
2939 0 : out_reserved:
2940 0 : btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2941 0 : if (free_delalloc_space)
2942 0 : btrfs_delalloc_release_space(inode, data_reserved, page_start,
2943 : PAGE_SIZE, true);
2944 0 : unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2945 0 : out_page:
2946 0 : if (ret) {
2947 : /*
2948 : * We hit ENOSPC or other errors. Update the mapping and page
2949 : * to reflect the errors and clean the page.
2950 : */
2951 0 : mapping_set_error(page->mapping, ret);
2952 0 : end_extent_writepage(page, ret, page_start, page_end);
2953 0 : clear_page_dirty_for_io(page);
2954 : }
2955 0 : btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE);
2956 0 : unlock_page(page);
2957 0 : put_page(page);
2958 0 : kfree(fixup);
2959 0 : extent_changeset_free(data_reserved);
2960 : /*
2961 : * As a precaution, do a delayed iput in case it would be the last iput
2962 : * that could need flushing space. Recursing back to fixup worker would
2963 : * deadlock.
2964 : */
2965 0 : btrfs_add_delayed_iput(inode);
2966 0 : }
2967 :
2968 : /*
2969 : * There are a few paths in the higher layers of the kernel that directly
2970 : * set the page dirty bit without asking the filesystem if it is a
2971 : * good idea. This causes problems because we want to make sure COW
2972 : * properly happens and the data=ordered rules are followed.
2973 : *
2974 : * In our case any range that doesn't have the ORDERED bit set
2975 : * hasn't been properly setup for IO. We kick off an async process
2976 : * to fix it up. The async helper will wait for ordered extents, set
2977 : * the delalloc bit and make it safe to write the page.
2978 : */
2979 0 : int btrfs_writepage_cow_fixup(struct page *page)
2980 : {
2981 0 : struct inode *inode = page->mapping->host;
2982 0 : struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2983 0 : struct btrfs_writepage_fixup *fixup;
2984 :
2985 : /* This page has ordered extent covering it already */
2986 0 : if (PageOrdered(page))
2987 : return 0;
2988 :
2989 : /*
2990 : * PageChecked is set below when we create a fixup worker for this page,
2991 : * don't try to create another one if we're already PageChecked()
2992 : *
2993 : * The extent_io writepage code will redirty the page if we send back
2994 : * EAGAIN.
2995 : */
2996 0 : if (PageChecked(page))
2997 : return -EAGAIN;
2998 :
2999 0 : fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
3000 0 : if (!fixup)
3001 : return -EAGAIN;
3002 :
3003 : /*
3004 : * We are already holding a reference to this inode from
3005 : * write_cache_pages. We need to hold it because the space reservation
3006 : * takes place outside of the page lock, and we can't trust
3007 : * page->mapping outside of the page lock.
3008 : */
3009 0 : ihold(inode);
3010 0 : btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
3011 0 : get_page(page);
3012 0 : btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
3013 0 : fixup->page = page;
3014 0 : fixup->inode = BTRFS_I(inode);
3015 0 : btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
3016 :
3017 0 : return -EAGAIN;
3018 : }
3019 :
3020 0 : static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
3021 : struct btrfs_inode *inode, u64 file_pos,
3022 : struct btrfs_file_extent_item *stack_fi,
3023 : const bool update_inode_bytes,
3024 : u64 qgroup_reserved)
3025 : {
3026 0 : struct btrfs_root *root = inode->root;
3027 0 : const u64 sectorsize = root->fs_info->sectorsize;
3028 0 : struct btrfs_path *path;
3029 0 : struct extent_buffer *leaf;
3030 0 : struct btrfs_key ins;
3031 0 : u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
3032 0 : u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
3033 0 : u64 offset = btrfs_stack_file_extent_offset(stack_fi);
3034 0 : u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
3035 0 : u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
3036 0 : struct btrfs_drop_extents_args drop_args = { 0 };
3037 0 : int ret;
3038 :
3039 0 : path = btrfs_alloc_path();
3040 0 : if (!path)
3041 : return -ENOMEM;
3042 :
3043 : /*
3044 : * we may be replacing one extent in the tree with another.
3045 : * The new extent is pinned in the extent map, and we don't want
3046 : * to drop it from the cache until it is completely in the btree.
3047 : *
3048 : * So, tell btrfs_drop_extents to leave this extent in the cache.
3049 : * the caller is expected to unpin it and allow it to be merged
3050 : * with the others.
3051 : */
3052 0 : drop_args.path = path;
3053 0 : drop_args.start = file_pos;
3054 0 : drop_args.end = file_pos + num_bytes;
3055 0 : drop_args.replace_extent = true;
3056 0 : drop_args.extent_item_size = sizeof(*stack_fi);
3057 0 : ret = btrfs_drop_extents(trans, root, inode, &drop_args);
3058 0 : if (ret)
3059 0 : goto out;
3060 :
3061 0 : if (!drop_args.extent_inserted) {
3062 0 : ins.objectid = btrfs_ino(inode);
3063 0 : ins.offset = file_pos;
3064 0 : ins.type = BTRFS_EXTENT_DATA_KEY;
3065 :
3066 0 : ret = btrfs_insert_empty_item(trans, root, path, &ins,
3067 : sizeof(*stack_fi));
3068 0 : if (ret)
3069 0 : goto out;
3070 : }
3071 0 : leaf = path->nodes[0];
3072 0 : btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
3073 0 : write_extent_buffer(leaf, stack_fi,
3074 0 : btrfs_item_ptr_offset(leaf, path->slots[0]),
3075 : sizeof(struct btrfs_file_extent_item));
3076 :
3077 0 : btrfs_mark_buffer_dirty(leaf);
3078 0 : btrfs_release_path(path);
3079 :
3080 : /*
3081 : * If we dropped an inline extent here, we know the range where it is
3082 : * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
3083 : * number of bytes only for that range containing the inline extent.
3084 : * The remaining of the range will be processed when clearning the
3085 : * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
3086 : */
3087 0 : if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
3088 0 : u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
3089 :
3090 0 : inline_size = drop_args.bytes_found - inline_size;
3091 0 : btrfs_update_inode_bytes(inode, sectorsize, inline_size);
3092 0 : drop_args.bytes_found -= inline_size;
3093 0 : num_bytes -= sectorsize;
3094 : }
3095 :
3096 0 : if (update_inode_bytes)
3097 0 : btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
3098 :
3099 0 : ins.objectid = disk_bytenr;
3100 0 : ins.offset = disk_num_bytes;
3101 0 : ins.type = BTRFS_EXTENT_ITEM_KEY;
3102 :
3103 0 : ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
3104 0 : if (ret)
3105 0 : goto out;
3106 :
3107 0 : ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
3108 : file_pos - offset,
3109 : qgroup_reserved, &ins);
3110 0 : out:
3111 0 : btrfs_free_path(path);
3112 :
3113 0 : return ret;
3114 : }
3115 :
3116 0 : static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
3117 : u64 start, u64 len)
3118 : {
3119 0 : struct btrfs_block_group *cache;
3120 :
3121 0 : cache = btrfs_lookup_block_group(fs_info, start);
3122 0 : ASSERT(cache);
3123 :
3124 0 : spin_lock(&cache->lock);
3125 0 : cache->delalloc_bytes -= len;
3126 0 : spin_unlock(&cache->lock);
3127 :
3128 0 : btrfs_put_block_group(cache);
3129 0 : }
3130 :
3131 0 : static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3132 : struct btrfs_ordered_extent *oe)
3133 : {
3134 0 : struct btrfs_file_extent_item stack_fi;
3135 0 : bool update_inode_bytes;
3136 0 : u64 num_bytes = oe->num_bytes;
3137 0 : u64 ram_bytes = oe->ram_bytes;
3138 :
3139 0 : memset(&stack_fi, 0, sizeof(stack_fi));
3140 0 : btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3141 0 : btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3142 0 : btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3143 : oe->disk_num_bytes);
3144 0 : btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3145 0 : if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3146 0 : num_bytes = oe->truncated_len;
3147 0 : ram_bytes = num_bytes;
3148 : }
3149 0 : btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3150 0 : btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3151 0 : btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3152 : /* Encryption and other encoding is reserved and all 0 */
3153 :
3154 : /*
3155 : * For delalloc, when completing an ordered extent we update the inode's
3156 : * bytes when clearing the range in the inode's io tree, so pass false
3157 : * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3158 : * except if the ordered extent was truncated.
3159 : */
3160 0 : update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3161 0 : test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3162 0 : test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3163 :
3164 0 : return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3165 : oe->file_offset, &stack_fi,
3166 0 : update_inode_bytes, oe->qgroup_rsv);
3167 : }
3168 :
3169 : /*
3170 : * As ordered data IO finishes, this gets called so we can finish
3171 : * an ordered extent if the range of bytes in the file it covers are
3172 : * fully written.
3173 : */
3174 0 : int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3175 : {
3176 0 : struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3177 0 : struct btrfs_root *root = inode->root;
3178 0 : struct btrfs_fs_info *fs_info = root->fs_info;
3179 0 : struct btrfs_trans_handle *trans = NULL;
3180 0 : struct extent_io_tree *io_tree = &inode->io_tree;
3181 0 : struct extent_state *cached_state = NULL;
3182 0 : u64 start, end;
3183 0 : int compress_type = 0;
3184 0 : int ret = 0;
3185 0 : u64 logical_len = ordered_extent->num_bytes;
3186 0 : bool freespace_inode;
3187 0 : bool truncated = false;
3188 0 : bool clear_reserved_extent = true;
3189 0 : unsigned int clear_bits = EXTENT_DEFRAG;
3190 :
3191 0 : start = ordered_extent->file_offset;
3192 0 : end = start + ordered_extent->num_bytes - 1;
3193 :
3194 0 : if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3195 0 : !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3196 0 : !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3197 0 : !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3198 0 : clear_bits |= EXTENT_DELALLOC_NEW;
3199 :
3200 0 : freespace_inode = btrfs_is_free_space_inode(inode);
3201 0 : if (!freespace_inode)
3202 0 : btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3203 :
3204 0 : if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3205 0 : ret = -EIO;
3206 0 : goto out;
3207 : }
3208 :
3209 0 : if (btrfs_is_zoned(fs_info))
3210 0 : btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3211 : ordered_extent->disk_num_bytes);
3212 :
3213 0 : if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3214 0 : truncated = true;
3215 0 : logical_len = ordered_extent->truncated_len;
3216 : /* Truncated the entire extent, don't bother adding */
3217 0 : if (!logical_len)
3218 0 : goto out;
3219 : }
3220 :
3221 0 : if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3222 0 : BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3223 :
3224 0 : btrfs_inode_safe_disk_i_size_write(inode, 0);
3225 0 : if (freespace_inode)
3226 0 : trans = btrfs_join_transaction_spacecache(root);
3227 : else
3228 0 : trans = btrfs_join_transaction(root);
3229 0 : if (IS_ERR(trans)) {
3230 0 : ret = PTR_ERR(trans);
3231 0 : trans = NULL;
3232 0 : goto out;
3233 : }
3234 0 : trans->block_rsv = &inode->block_rsv;
3235 0 : ret = btrfs_update_inode_fallback(trans, root, inode);
3236 0 : if (ret) /* -ENOMEM or corruption */
3237 0 : btrfs_abort_transaction(trans, ret);
3238 0 : goto out;
3239 : }
3240 :
3241 0 : clear_bits |= EXTENT_LOCKED;
3242 0 : lock_extent(io_tree, start, end, &cached_state);
3243 :
3244 0 : if (freespace_inode)
3245 0 : trans = btrfs_join_transaction_spacecache(root);
3246 : else
3247 0 : trans = btrfs_join_transaction(root);
3248 0 : if (IS_ERR(trans)) {
3249 0 : ret = PTR_ERR(trans);
3250 0 : trans = NULL;
3251 0 : goto out;
3252 : }
3253 :
3254 0 : trans->block_rsv = &inode->block_rsv;
3255 :
3256 0 : if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3257 0 : compress_type = ordered_extent->compress_type;
3258 0 : if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3259 0 : BUG_ON(compress_type);
3260 0 : ret = btrfs_mark_extent_written(trans, inode,
3261 : ordered_extent->file_offset,
3262 : ordered_extent->file_offset +
3263 : logical_len);
3264 0 : btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3265 : ordered_extent->disk_num_bytes);
3266 : } else {
3267 0 : BUG_ON(root == fs_info->tree_root);
3268 0 : ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3269 0 : if (!ret) {
3270 0 : clear_reserved_extent = false;
3271 0 : btrfs_release_delalloc_bytes(fs_info,
3272 : ordered_extent->disk_bytenr,
3273 : ordered_extent->disk_num_bytes);
3274 : }
3275 : }
3276 0 : unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3277 : ordered_extent->num_bytes, trans->transid);
3278 0 : if (ret < 0) {
3279 0 : btrfs_abort_transaction(trans, ret);
3280 0 : goto out;
3281 : }
3282 :
3283 0 : ret = add_pending_csums(trans, &ordered_extent->list);
3284 0 : if (ret) {
3285 0 : btrfs_abort_transaction(trans, ret);
3286 0 : goto out;
3287 : }
3288 :
3289 : /*
3290 : * If this is a new delalloc range, clear its new delalloc flag to
3291 : * update the inode's number of bytes. This needs to be done first
3292 : * before updating the inode item.
3293 : */
3294 0 : if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3295 0 : !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3296 0 : clear_extent_bit(&inode->io_tree, start, end,
3297 : EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3298 : &cached_state);
3299 :
3300 0 : btrfs_inode_safe_disk_i_size_write(inode, 0);
3301 0 : ret = btrfs_update_inode_fallback(trans, root, inode);
3302 0 : if (ret) { /* -ENOMEM or corruption */
3303 0 : btrfs_abort_transaction(trans, ret);
3304 0 : goto out;
3305 : }
3306 : ret = 0;
3307 0 : out:
3308 0 : clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3309 : &cached_state);
3310 :
3311 0 : if (trans)
3312 0 : btrfs_end_transaction(trans);
3313 :
3314 0 : if (ret || truncated) {
3315 0 : u64 unwritten_start = start;
3316 :
3317 : /*
3318 : * If we failed to finish this ordered extent for any reason we
3319 : * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3320 : * extent, and mark the inode with the error if it wasn't
3321 : * already set. Any error during writeback would have already
3322 : * set the mapping error, so we need to set it if we're the ones
3323 : * marking this ordered extent as failed.
3324 : */
3325 0 : if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3326 : &ordered_extent->flags))
3327 0 : mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3328 :
3329 0 : if (truncated)
3330 0 : unwritten_start += logical_len;
3331 0 : clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3332 :
3333 : /* Drop extent maps for the part of the extent we didn't write. */
3334 0 : btrfs_drop_extent_map_range(inode, unwritten_start, end, false);
3335 :
3336 : /*
3337 : * If the ordered extent had an IOERR or something else went
3338 : * wrong we need to return the space for this ordered extent
3339 : * back to the allocator. We only free the extent in the
3340 : * truncated case if we didn't write out the extent at all.
3341 : *
3342 : * If we made it past insert_reserved_file_extent before we
3343 : * errored out then we don't need to do this as the accounting
3344 : * has already been done.
3345 : */
3346 0 : if ((ret || !logical_len) &&
3347 0 : clear_reserved_extent &&
3348 0 : !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3349 0 : !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3350 : /*
3351 : * Discard the range before returning it back to the
3352 : * free space pool
3353 : */
3354 0 : if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3355 0 : btrfs_discard_extent(fs_info,
3356 : ordered_extent->disk_bytenr,
3357 : ordered_extent->disk_num_bytes,
3358 : NULL);
3359 0 : btrfs_free_reserved_extent(fs_info,
3360 : ordered_extent->disk_bytenr,
3361 : ordered_extent->disk_num_bytes, 1);
3362 : }
3363 : }
3364 :
3365 : /*
3366 : * This needs to be done to make sure anybody waiting knows we are done
3367 : * updating everything for this ordered extent.
3368 : */
3369 0 : btrfs_remove_ordered_extent(inode, ordered_extent);
3370 :
3371 : /* once for us */
3372 0 : btrfs_put_ordered_extent(ordered_extent);
3373 : /* once for the tree */
3374 0 : btrfs_put_ordered_extent(ordered_extent);
3375 :
3376 0 : return ret;
3377 : }
3378 :
3379 0 : int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3380 : {
3381 0 : if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) &&
3382 0 : !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
3383 0 : btrfs_finish_ordered_zoned(ordered);
3384 0 : return btrfs_finish_one_ordered(ordered);
3385 : }
3386 :
3387 0 : void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3388 : struct page *page, u64 start,
3389 : u64 end, bool uptodate)
3390 : {
3391 0 : trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3392 :
3393 0 : btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start, uptodate);
3394 0 : }
3395 :
3396 : /*
3397 : * Verify the checksum for a single sector without any extra action that depend
3398 : * on the type of I/O.
3399 : */
3400 0 : int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3401 : u32 pgoff, u8 *csum, const u8 * const csum_expected)
3402 : {
3403 0 : SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3404 0 : char *kaddr;
3405 :
3406 0 : ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3407 :
3408 0 : shash->tfm = fs_info->csum_shash;
3409 :
3410 0 : kaddr = kmap_local_page(page) + pgoff;
3411 0 : crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3412 0 : kunmap_local(kaddr);
3413 :
3414 0 : if (memcmp(csum, csum_expected, fs_info->csum_size))
3415 0 : return -EIO;
3416 : return 0;
3417 : }
3418 :
3419 : /*
3420 : * Verify the checksum of a single data sector.
3421 : *
3422 : * @bbio: btrfs_io_bio which contains the csum
3423 : * @dev: device the sector is on
3424 : * @bio_offset: offset to the beginning of the bio (in bytes)
3425 : * @bv: bio_vec to check
3426 : *
3427 : * Check if the checksum on a data block is valid. When a checksum mismatch is
3428 : * detected, report the error and fill the corrupted range with zero.
3429 : *
3430 : * Return %true if the sector is ok or had no checksum to start with, else %false.
3431 : */
3432 0 : bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3433 : u32 bio_offset, struct bio_vec *bv)
3434 : {
3435 0 : struct btrfs_inode *inode = bbio->inode;
3436 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
3437 0 : u64 file_offset = bbio->file_offset + bio_offset;
3438 0 : u64 end = file_offset + bv->bv_len - 1;
3439 0 : u8 *csum_expected;
3440 0 : u8 csum[BTRFS_CSUM_SIZE];
3441 :
3442 0 : ASSERT(bv->bv_len == fs_info->sectorsize);
3443 :
3444 0 : if (!bbio->csum)
3445 : return true;
3446 :
3447 0 : if (btrfs_is_data_reloc_root(inode->root) &&
3448 0 : test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3449 : 1, NULL)) {
3450 : /* Skip the range without csum for data reloc inode */
3451 0 : clear_extent_bits(&inode->io_tree, file_offset, end,
3452 : EXTENT_NODATASUM);
3453 0 : return true;
3454 : }
3455 :
3456 0 : csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3457 0 : fs_info->csum_size;
3458 0 : if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3459 : csum_expected))
3460 0 : goto zeroit;
3461 : return true;
3462 :
3463 : zeroit:
3464 0 : btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3465 0 : bbio->mirror_num);
3466 0 : if (dev)
3467 0 : btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3468 0 : memzero_bvec(bv);
3469 0 : return false;
3470 : }
3471 :
3472 : /*
3473 : * btrfs_add_delayed_iput - perform a delayed iput on @inode
3474 : *
3475 : * @inode: The inode we want to perform iput on
3476 : *
3477 : * This function uses the generic vfs_inode::i_count to track whether we should
3478 : * just decrement it (in case it's > 1) or if this is the last iput then link
3479 : * the inode to the delayed iput machinery. Delayed iputs are processed at
3480 : * transaction commit time/superblock commit/cleaner kthread.
3481 : */
3482 0 : void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3483 : {
3484 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
3485 0 : unsigned long flags;
3486 :
3487 0 : if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3488 : return;
3489 :
3490 0 : atomic_inc(&fs_info->nr_delayed_iputs);
3491 : /*
3492 : * Need to be irq safe here because we can be called from either an irq
3493 : * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3494 : * context.
3495 : */
3496 0 : spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3497 0 : ASSERT(list_empty(&inode->delayed_iput));
3498 0 : list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3499 0 : spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3500 0 : if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3501 0 : wake_up_process(fs_info->cleaner_kthread);
3502 : }
3503 :
3504 0 : static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3505 : struct btrfs_inode *inode)
3506 : {
3507 0 : list_del_init(&inode->delayed_iput);
3508 0 : spin_unlock_irq(&fs_info->delayed_iput_lock);
3509 0 : iput(&inode->vfs_inode);
3510 0 : if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3511 0 : wake_up(&fs_info->delayed_iputs_wait);
3512 0 : spin_lock_irq(&fs_info->delayed_iput_lock);
3513 0 : }
3514 :
3515 0 : static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3516 : struct btrfs_inode *inode)
3517 : {
3518 0 : if (!list_empty(&inode->delayed_iput)) {
3519 0 : spin_lock_irq(&fs_info->delayed_iput_lock);
3520 0 : if (!list_empty(&inode->delayed_iput))
3521 0 : run_delayed_iput_locked(fs_info, inode);
3522 0 : spin_unlock_irq(&fs_info->delayed_iput_lock);
3523 : }
3524 0 : }
3525 :
3526 0 : void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3527 : {
3528 : /*
3529 : * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3530 : * calls btrfs_add_delayed_iput() and that needs to lock
3531 : * fs_info->delayed_iput_lock. So we need to disable irqs here to
3532 : * prevent a deadlock.
3533 : */
3534 0 : spin_lock_irq(&fs_info->delayed_iput_lock);
3535 0 : while (!list_empty(&fs_info->delayed_iputs)) {
3536 0 : struct btrfs_inode *inode;
3537 :
3538 0 : inode = list_first_entry(&fs_info->delayed_iputs,
3539 : struct btrfs_inode, delayed_iput);
3540 0 : run_delayed_iput_locked(fs_info, inode);
3541 0 : if (need_resched()) {
3542 0 : spin_unlock_irq(&fs_info->delayed_iput_lock);
3543 0 : cond_resched();
3544 0 : spin_lock_irq(&fs_info->delayed_iput_lock);
3545 : }
3546 : }
3547 0 : spin_unlock_irq(&fs_info->delayed_iput_lock);
3548 0 : }
3549 :
3550 : /*
3551 : * Wait for flushing all delayed iputs
3552 : *
3553 : * @fs_info: the filesystem
3554 : *
3555 : * This will wait on any delayed iputs that are currently running with KILLABLE
3556 : * set. Once they are all done running we will return, unless we are killed in
3557 : * which case we return EINTR. This helps in user operations like fallocate etc
3558 : * that might get blocked on the iputs.
3559 : *
3560 : * Return EINTR if we were killed, 0 if nothing's pending
3561 : */
3562 0 : int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3563 : {
3564 0 : int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3565 : atomic_read(&fs_info->nr_delayed_iputs) == 0);
3566 0 : if (ret)
3567 0 : return -EINTR;
3568 : return 0;
3569 : }
3570 :
3571 : /*
3572 : * This creates an orphan entry for the given inode in case something goes wrong
3573 : * in the middle of an unlink.
3574 : */
3575 0 : int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3576 : struct btrfs_inode *inode)
3577 : {
3578 0 : int ret;
3579 :
3580 0 : ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3581 0 : if (ret && ret != -EEXIST) {
3582 0 : btrfs_abort_transaction(trans, ret);
3583 0 : return ret;
3584 : }
3585 :
3586 : return 0;
3587 : }
3588 :
3589 : /*
3590 : * We have done the delete so we can go ahead and remove the orphan item for
3591 : * this particular inode.
3592 : */
3593 : static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3594 : struct btrfs_inode *inode)
3595 : {
3596 0 : return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3597 : }
3598 :
3599 : /*
3600 : * this cleans up any orphans that may be left on the list from the last use
3601 : * of this root.
3602 : */
3603 0 : int btrfs_orphan_cleanup(struct btrfs_root *root)
3604 : {
3605 0 : struct btrfs_fs_info *fs_info = root->fs_info;
3606 0 : struct btrfs_path *path;
3607 0 : struct extent_buffer *leaf;
3608 0 : struct btrfs_key key, found_key;
3609 0 : struct btrfs_trans_handle *trans;
3610 0 : struct inode *inode;
3611 0 : u64 last_objectid = 0;
3612 0 : int ret = 0, nr_unlink = 0;
3613 :
3614 0 : if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3615 : return 0;
3616 :
3617 0 : path = btrfs_alloc_path();
3618 0 : if (!path) {
3619 0 : ret = -ENOMEM;
3620 0 : goto out;
3621 : }
3622 0 : path->reada = READA_BACK;
3623 :
3624 0 : key.objectid = BTRFS_ORPHAN_OBJECTID;
3625 0 : key.type = BTRFS_ORPHAN_ITEM_KEY;
3626 0 : key.offset = (u64)-1;
3627 :
3628 0 : while (1) {
3629 0 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3630 0 : if (ret < 0)
3631 0 : goto out;
3632 :
3633 : /*
3634 : * if ret == 0 means we found what we were searching for, which
3635 : * is weird, but possible, so only screw with path if we didn't
3636 : * find the key and see if we have stuff that matches
3637 : */
3638 0 : if (ret > 0) {
3639 0 : ret = 0;
3640 0 : if (path->slots[0] == 0)
3641 : break;
3642 0 : path->slots[0]--;
3643 : }
3644 :
3645 : /* pull out the item */
3646 0 : leaf = path->nodes[0];
3647 0 : btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3648 :
3649 : /* make sure the item matches what we want */
3650 0 : if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3651 : break;
3652 0 : if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3653 : break;
3654 :
3655 : /* release the path since we're done with it */
3656 0 : btrfs_release_path(path);
3657 :
3658 : /*
3659 : * this is where we are basically btrfs_lookup, without the
3660 : * crossing root thing. we store the inode number in the
3661 : * offset of the orphan item.
3662 : */
3663 :
3664 0 : if (found_key.offset == last_objectid) {
3665 0 : btrfs_err(fs_info,
3666 : "Error removing orphan entry, stopping orphan cleanup");
3667 0 : ret = -EINVAL;
3668 0 : goto out;
3669 : }
3670 :
3671 0 : last_objectid = found_key.offset;
3672 :
3673 0 : found_key.objectid = found_key.offset;
3674 0 : found_key.type = BTRFS_INODE_ITEM_KEY;
3675 0 : found_key.offset = 0;
3676 0 : inode = btrfs_iget(fs_info->sb, last_objectid, root);
3677 0 : if (IS_ERR(inode)) {
3678 0 : ret = PTR_ERR(inode);
3679 0 : inode = NULL;
3680 0 : if (ret != -ENOENT)
3681 0 : goto out;
3682 : }
3683 :
3684 0 : if (!inode && root == fs_info->tree_root) {
3685 0 : struct btrfs_root *dead_root;
3686 0 : int is_dead_root = 0;
3687 :
3688 : /*
3689 : * This is an orphan in the tree root. Currently these
3690 : * could come from 2 sources:
3691 : * a) a root (snapshot/subvolume) deletion in progress
3692 : * b) a free space cache inode
3693 : * We need to distinguish those two, as the orphan item
3694 : * for a root must not get deleted before the deletion
3695 : * of the snapshot/subvolume's tree completes.
3696 : *
3697 : * btrfs_find_orphan_roots() ran before us, which has
3698 : * found all deleted roots and loaded them into
3699 : * fs_info->fs_roots_radix. So here we can find if an
3700 : * orphan item corresponds to a deleted root by looking
3701 : * up the root from that radix tree.
3702 : */
3703 :
3704 0 : spin_lock(&fs_info->fs_roots_radix_lock);
3705 0 : dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3706 0 : (unsigned long)found_key.objectid);
3707 0 : if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3708 0 : is_dead_root = 1;
3709 0 : spin_unlock(&fs_info->fs_roots_radix_lock);
3710 :
3711 0 : if (is_dead_root) {
3712 : /* prevent this orphan from being found again */
3713 0 : key.offset = found_key.objectid - 1;
3714 0 : continue;
3715 : }
3716 :
3717 : }
3718 :
3719 : /*
3720 : * If we have an inode with links, there are a couple of
3721 : * possibilities:
3722 : *
3723 : * 1. We were halfway through creating fsverity metadata for the
3724 : * file. In that case, the orphan item represents incomplete
3725 : * fsverity metadata which must be cleaned up with
3726 : * btrfs_drop_verity_items and deleting the orphan item.
3727 :
3728 : * 2. Old kernels (before v3.12) used to create an
3729 : * orphan item for truncate indicating that there were possibly
3730 : * extent items past i_size that needed to be deleted. In v3.12,
3731 : * truncate was changed to update i_size in sync with the extent
3732 : * items, but the (useless) orphan item was still created. Since
3733 : * v4.18, we don't create the orphan item for truncate at all.
3734 : *
3735 : * So, this item could mean that we need to do a truncate, but
3736 : * only if this filesystem was last used on a pre-v3.12 kernel
3737 : * and was not cleanly unmounted. The odds of that are quite
3738 : * slim, and it's a pain to do the truncate now, so just delete
3739 : * the orphan item.
3740 : *
3741 : * It's also possible that this orphan item was supposed to be
3742 : * deleted but wasn't. The inode number may have been reused,
3743 : * but either way, we can delete the orphan item.
3744 : */
3745 0 : if (!inode || inode->i_nlink) {
3746 0 : if (inode) {
3747 0 : ret = btrfs_drop_verity_items(BTRFS_I(inode));
3748 0 : iput(inode);
3749 0 : inode = NULL;
3750 0 : if (ret)
3751 : goto out;
3752 : }
3753 0 : trans = btrfs_start_transaction(root, 1);
3754 0 : if (IS_ERR(trans)) {
3755 0 : ret = PTR_ERR(trans);
3756 0 : goto out;
3757 : }
3758 0 : btrfs_debug(fs_info, "auto deleting %Lu",
3759 : found_key.objectid);
3760 0 : ret = btrfs_del_orphan_item(trans, root,
3761 : found_key.objectid);
3762 0 : btrfs_end_transaction(trans);
3763 0 : if (ret)
3764 0 : goto out;
3765 0 : continue;
3766 : }
3767 :
3768 0 : nr_unlink++;
3769 :
3770 : /* this will do delete_inode and everything for us */
3771 0 : iput(inode);
3772 : }
3773 : /* release the path since we're done with it */
3774 0 : btrfs_release_path(path);
3775 :
3776 0 : if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3777 0 : trans = btrfs_join_transaction(root);
3778 0 : if (!IS_ERR(trans))
3779 0 : btrfs_end_transaction(trans);
3780 : }
3781 :
3782 : if (nr_unlink)
3783 : btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3784 :
3785 0 : out:
3786 0 : if (ret)
3787 0 : btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3788 0 : btrfs_free_path(path);
3789 0 : return ret;
3790 : }
3791 :
3792 : /*
3793 : * very simple check to peek ahead in the leaf looking for xattrs. If we
3794 : * don't find any xattrs, we know there can't be any acls.
3795 : *
3796 : * slot is the slot the inode is in, objectid is the objectid of the inode
3797 : */
3798 0 : static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3799 : int slot, u64 objectid,
3800 : int *first_xattr_slot)
3801 : {
3802 0 : u32 nritems = btrfs_header_nritems(leaf);
3803 0 : struct btrfs_key found_key;
3804 0 : static u64 xattr_access = 0;
3805 0 : static u64 xattr_default = 0;
3806 0 : int scanned = 0;
3807 :
3808 0 : if (!xattr_access) {
3809 0 : xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3810 : strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3811 0 : xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3812 : strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3813 : }
3814 :
3815 0 : slot++;
3816 0 : *first_xattr_slot = -1;
3817 0 : while (slot < nritems) {
3818 0 : btrfs_item_key_to_cpu(leaf, &found_key, slot);
3819 :
3820 : /* we found a different objectid, there must not be acls */
3821 0 : if (found_key.objectid != objectid)
3822 : return 0;
3823 :
3824 : /* we found an xattr, assume we've got an acl */
3825 0 : if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3826 0 : if (*first_xattr_slot == -1)
3827 0 : *first_xattr_slot = slot;
3828 0 : if (found_key.offset == xattr_access ||
3829 0 : found_key.offset == xattr_default)
3830 : return 1;
3831 : }
3832 :
3833 : /*
3834 : * we found a key greater than an xattr key, there can't
3835 : * be any acls later on
3836 : */
3837 0 : if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3838 : return 0;
3839 :
3840 0 : slot++;
3841 0 : scanned++;
3842 :
3843 : /*
3844 : * it goes inode, inode backrefs, xattrs, extents,
3845 : * so if there are a ton of hard links to an inode there can
3846 : * be a lot of backrefs. Don't waste time searching too hard,
3847 : * this is just an optimization
3848 : */
3849 0 : if (scanned >= 8)
3850 : break;
3851 : }
3852 : /* we hit the end of the leaf before we found an xattr or
3853 : * something larger than an xattr. We have to assume the inode
3854 : * has acls
3855 : */
3856 0 : if (*first_xattr_slot == -1)
3857 0 : *first_xattr_slot = slot;
3858 : return 1;
3859 : }
3860 :
3861 : /*
3862 : * read an inode from the btree into the in-memory inode
3863 : */
3864 0 : static int btrfs_read_locked_inode(struct inode *inode,
3865 : struct btrfs_path *in_path)
3866 : {
3867 0 : struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3868 0 : struct btrfs_path *path = in_path;
3869 0 : struct extent_buffer *leaf;
3870 0 : struct btrfs_inode_item *inode_item;
3871 0 : struct btrfs_root *root = BTRFS_I(inode)->root;
3872 0 : struct btrfs_key location;
3873 0 : unsigned long ptr;
3874 0 : int maybe_acls;
3875 0 : u32 rdev;
3876 0 : int ret;
3877 0 : bool filled = false;
3878 0 : int first_xattr_slot;
3879 :
3880 0 : ret = btrfs_fill_inode(inode, &rdev);
3881 0 : if (!ret)
3882 0 : filled = true;
3883 :
3884 0 : if (!path) {
3885 0 : path = btrfs_alloc_path();
3886 0 : if (!path)
3887 : return -ENOMEM;
3888 : }
3889 :
3890 0 : memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3891 :
3892 0 : ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3893 0 : if (ret) {
3894 0 : if (path != in_path)
3895 0 : btrfs_free_path(path);
3896 0 : return ret;
3897 : }
3898 :
3899 0 : leaf = path->nodes[0];
3900 :
3901 0 : if (filled)
3902 0 : goto cache_index;
3903 :
3904 0 : inode_item = btrfs_item_ptr(leaf, path->slots[0],
3905 : struct btrfs_inode_item);
3906 0 : inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3907 0 : set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3908 0 : i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3909 0 : i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3910 0 : btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3911 0 : btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3912 0 : round_up(i_size_read(inode), fs_info->sectorsize));
3913 :
3914 0 : inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3915 0 : inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3916 :
3917 0 : inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3918 0 : inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3919 :
3920 0 : inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3921 0 : inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3922 :
3923 0 : BTRFS_I(inode)->i_otime.tv_sec =
3924 0 : btrfs_timespec_sec(leaf, &inode_item->otime);
3925 0 : BTRFS_I(inode)->i_otime.tv_nsec =
3926 0 : btrfs_timespec_nsec(leaf, &inode_item->otime);
3927 :
3928 0 : inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3929 0 : BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3930 0 : BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3931 :
3932 0 : inode_set_iversion_queried(inode,
3933 : btrfs_inode_sequence(leaf, inode_item));
3934 0 : inode->i_generation = BTRFS_I(inode)->generation;
3935 0 : inode->i_rdev = 0;
3936 0 : rdev = btrfs_inode_rdev(leaf, inode_item);
3937 :
3938 0 : BTRFS_I(inode)->index_cnt = (u64)-1;
3939 0 : btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3940 : &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3941 :
3942 0 : cache_index:
3943 : /*
3944 : * If we were modified in the current generation and evicted from memory
3945 : * and then re-read we need to do a full sync since we don't have any
3946 : * idea about which extents were modified before we were evicted from
3947 : * cache.
3948 : *
3949 : * This is required for both inode re-read from disk and delayed inode
3950 : * in delayed_nodes_tree.
3951 : */
3952 0 : if (BTRFS_I(inode)->last_trans == fs_info->generation)
3953 0 : set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3954 0 : &BTRFS_I(inode)->runtime_flags);
3955 :
3956 : /*
3957 : * We don't persist the id of the transaction where an unlink operation
3958 : * against the inode was last made. So here we assume the inode might
3959 : * have been evicted, and therefore the exact value of last_unlink_trans
3960 : * lost, and set it to last_trans to avoid metadata inconsistencies
3961 : * between the inode and its parent if the inode is fsync'ed and the log
3962 : * replayed. For example, in the scenario:
3963 : *
3964 : * touch mydir/foo
3965 : * ln mydir/foo mydir/bar
3966 : * sync
3967 : * unlink mydir/bar
3968 : * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3969 : * xfs_io -c fsync mydir/foo
3970 : * <power failure>
3971 : * mount fs, triggers fsync log replay
3972 : *
3973 : * We must make sure that when we fsync our inode foo we also log its
3974 : * parent inode, otherwise after log replay the parent still has the
3975 : * dentry with the "bar" name but our inode foo has a link count of 1
3976 : * and doesn't have an inode ref with the name "bar" anymore.
3977 : *
3978 : * Setting last_unlink_trans to last_trans is a pessimistic approach,
3979 : * but it guarantees correctness at the expense of occasional full
3980 : * transaction commits on fsync if our inode is a directory, or if our
3981 : * inode is not a directory, logging its parent unnecessarily.
3982 : */
3983 0 : BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3984 :
3985 : /*
3986 : * Same logic as for last_unlink_trans. We don't persist the generation
3987 : * of the last transaction where this inode was used for a reflink
3988 : * operation, so after eviction and reloading the inode we must be
3989 : * pessimistic and assume the last transaction that modified the inode.
3990 : */
3991 0 : BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3992 :
3993 0 : path->slots[0]++;
3994 0 : if (inode->i_nlink != 1 ||
3995 0 : path->slots[0] >= btrfs_header_nritems(leaf))
3996 0 : goto cache_acl;
3997 :
3998 0 : btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3999 0 : if (location.objectid != btrfs_ino(BTRFS_I(inode)))
4000 0 : goto cache_acl;
4001 :
4002 0 : ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4003 0 : if (location.type == BTRFS_INODE_REF_KEY) {
4004 0 : struct btrfs_inode_ref *ref;
4005 :
4006 0 : ref = (struct btrfs_inode_ref *)ptr;
4007 0 : BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
4008 0 : } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
4009 0 : struct btrfs_inode_extref *extref;
4010 :
4011 0 : extref = (struct btrfs_inode_extref *)ptr;
4012 0 : BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
4013 : extref);
4014 : }
4015 0 : cache_acl:
4016 : /*
4017 : * try to precache a NULL acl entry for files that don't have
4018 : * any xattrs or acls
4019 : */
4020 0 : maybe_acls = acls_after_inode_item(leaf, path->slots[0],
4021 : btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
4022 0 : if (first_xattr_slot != -1) {
4023 0 : path->slots[0] = first_xattr_slot;
4024 0 : ret = btrfs_load_inode_props(inode, path);
4025 0 : if (ret)
4026 0 : btrfs_err(fs_info,
4027 : "error loading props for ino %llu (root %llu): %d",
4028 : btrfs_ino(BTRFS_I(inode)),
4029 : root->root_key.objectid, ret);
4030 : }
4031 0 : if (path != in_path)
4032 0 : btrfs_free_path(path);
4033 :
4034 0 : if (!maybe_acls)
4035 0 : cache_no_acl(inode);
4036 :
4037 0 : switch (inode->i_mode & S_IFMT) {
4038 0 : case S_IFREG:
4039 0 : inode->i_mapping->a_ops = &btrfs_aops;
4040 0 : inode->i_fop = &btrfs_file_operations;
4041 0 : inode->i_op = &btrfs_file_inode_operations;
4042 0 : break;
4043 0 : case S_IFDIR:
4044 0 : inode->i_fop = &btrfs_dir_file_operations;
4045 0 : inode->i_op = &btrfs_dir_inode_operations;
4046 0 : break;
4047 0 : case S_IFLNK:
4048 0 : inode->i_op = &btrfs_symlink_inode_operations;
4049 0 : inode_nohighmem(inode);
4050 0 : inode->i_mapping->a_ops = &btrfs_aops;
4051 0 : break;
4052 0 : default:
4053 0 : inode->i_op = &btrfs_special_inode_operations;
4054 0 : init_special_inode(inode, inode->i_mode, rdev);
4055 0 : break;
4056 : }
4057 :
4058 0 : btrfs_sync_inode_flags_to_i_flags(inode);
4059 0 : return 0;
4060 : }
4061 :
4062 : /*
4063 : * given a leaf and an inode, copy the inode fields into the leaf
4064 : */
4065 0 : static void fill_inode_item(struct btrfs_trans_handle *trans,
4066 : struct extent_buffer *leaf,
4067 : struct btrfs_inode_item *item,
4068 : struct inode *inode)
4069 : {
4070 0 : struct btrfs_map_token token;
4071 0 : u64 flags;
4072 :
4073 0 : btrfs_init_map_token(&token, leaf);
4074 :
4075 0 : btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4076 0 : btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4077 0 : btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
4078 0 : btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4079 0 : btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4080 :
4081 0 : btrfs_set_token_timespec_sec(&token, &item->atime,
4082 0 : inode->i_atime.tv_sec);
4083 0 : btrfs_set_token_timespec_nsec(&token, &item->atime,
4084 0 : inode->i_atime.tv_nsec);
4085 :
4086 0 : btrfs_set_token_timespec_sec(&token, &item->mtime,
4087 0 : inode->i_mtime.tv_sec);
4088 0 : btrfs_set_token_timespec_nsec(&token, &item->mtime,
4089 0 : inode->i_mtime.tv_nsec);
4090 :
4091 0 : btrfs_set_token_timespec_sec(&token, &item->ctime,
4092 0 : inode->i_ctime.tv_sec);
4093 0 : btrfs_set_token_timespec_nsec(&token, &item->ctime,
4094 0 : inode->i_ctime.tv_nsec);
4095 :
4096 0 : btrfs_set_token_timespec_sec(&token, &item->otime,
4097 0 : BTRFS_I(inode)->i_otime.tv_sec);
4098 0 : btrfs_set_token_timespec_nsec(&token, &item->otime,
4099 0 : BTRFS_I(inode)->i_otime.tv_nsec);
4100 :
4101 0 : btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
4102 0 : btrfs_set_token_inode_generation(&token, item,
4103 : BTRFS_I(inode)->generation);
4104 0 : btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4105 0 : btrfs_set_token_inode_transid(&token, item, trans->transid);
4106 0 : btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4107 0 : flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4108 : BTRFS_I(inode)->ro_flags);
4109 0 : btrfs_set_token_inode_flags(&token, item, flags);
4110 0 : btrfs_set_token_inode_block_group(&token, item, 0);
4111 0 : }
4112 :
4113 : /*
4114 : * copy everything in the in-memory inode into the btree.
4115 : */
4116 0 : static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4117 : struct btrfs_root *root,
4118 : struct btrfs_inode *inode)
4119 : {
4120 0 : struct btrfs_inode_item *inode_item;
4121 0 : struct btrfs_path *path;
4122 0 : struct extent_buffer *leaf;
4123 0 : int ret;
4124 :
4125 0 : path = btrfs_alloc_path();
4126 0 : if (!path)
4127 : return -ENOMEM;
4128 :
4129 0 : ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4130 0 : if (ret) {
4131 0 : if (ret > 0)
4132 0 : ret = -ENOENT;
4133 0 : goto failed;
4134 : }
4135 :
4136 0 : leaf = path->nodes[0];
4137 0 : inode_item = btrfs_item_ptr(leaf, path->slots[0],
4138 : struct btrfs_inode_item);
4139 :
4140 0 : fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4141 0 : btrfs_mark_buffer_dirty(leaf);
4142 0 : btrfs_set_inode_last_trans(trans, inode);
4143 0 : ret = 0;
4144 0 : failed:
4145 0 : btrfs_free_path(path);
4146 0 : return ret;
4147 : }
4148 :
4149 : /*
4150 : * copy everything in the in-memory inode into the btree.
4151 : */
4152 0 : noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4153 : struct btrfs_root *root,
4154 : struct btrfs_inode *inode)
4155 : {
4156 0 : struct btrfs_fs_info *fs_info = root->fs_info;
4157 0 : int ret;
4158 :
4159 : /*
4160 : * If the inode is a free space inode, we can deadlock during commit
4161 : * if we put it into the delayed code.
4162 : *
4163 : * The data relocation inode should also be directly updated
4164 : * without delay
4165 : */
4166 0 : if (!btrfs_is_free_space_inode(inode)
4167 0 : && !btrfs_is_data_reloc_root(root)
4168 0 : && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4169 0 : btrfs_update_root_times(trans, root);
4170 :
4171 0 : ret = btrfs_delayed_update_inode(trans, root, inode);
4172 0 : if (!ret)
4173 0 : btrfs_set_inode_last_trans(trans, inode);
4174 0 : return ret;
4175 : }
4176 :
4177 0 : return btrfs_update_inode_item(trans, root, inode);
4178 : }
4179 :
4180 0 : int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4181 : struct btrfs_root *root, struct btrfs_inode *inode)
4182 : {
4183 0 : int ret;
4184 :
4185 0 : ret = btrfs_update_inode(trans, root, inode);
4186 0 : if (ret == -ENOSPC)
4187 0 : return btrfs_update_inode_item(trans, root, inode);
4188 : return ret;
4189 : }
4190 :
4191 : /*
4192 : * unlink helper that gets used here in inode.c and in the tree logging
4193 : * recovery code. It remove a link in a directory with a given name, and
4194 : * also drops the back refs in the inode to the directory
4195 : */
4196 0 : static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4197 : struct btrfs_inode *dir,
4198 : struct btrfs_inode *inode,
4199 : const struct fscrypt_str *name,
4200 : struct btrfs_rename_ctx *rename_ctx)
4201 : {
4202 0 : struct btrfs_root *root = dir->root;
4203 0 : struct btrfs_fs_info *fs_info = root->fs_info;
4204 0 : struct btrfs_path *path;
4205 0 : int ret = 0;
4206 0 : struct btrfs_dir_item *di;
4207 0 : u64 index;
4208 0 : u64 ino = btrfs_ino(inode);
4209 0 : u64 dir_ino = btrfs_ino(dir);
4210 :
4211 0 : path = btrfs_alloc_path();
4212 0 : if (!path) {
4213 0 : ret = -ENOMEM;
4214 0 : goto out;
4215 : }
4216 :
4217 0 : di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4218 0 : if (IS_ERR_OR_NULL(di)) {
4219 0 : ret = di ? PTR_ERR(di) : -ENOENT;
4220 0 : goto err;
4221 : }
4222 0 : ret = btrfs_delete_one_dir_name(trans, root, path, di);
4223 0 : if (ret)
4224 0 : goto err;
4225 0 : btrfs_release_path(path);
4226 :
4227 : /*
4228 : * If we don't have dir index, we have to get it by looking up
4229 : * the inode ref, since we get the inode ref, remove it directly,
4230 : * it is unnecessary to do delayed deletion.
4231 : *
4232 : * But if we have dir index, needn't search inode ref to get it.
4233 : * Since the inode ref is close to the inode item, it is better
4234 : * that we delay to delete it, and just do this deletion when
4235 : * we update the inode item.
4236 : */
4237 0 : if (inode->dir_index) {
4238 0 : ret = btrfs_delayed_delete_inode_ref(inode);
4239 0 : if (!ret) {
4240 0 : index = inode->dir_index;
4241 0 : goto skip_backref;
4242 : }
4243 : }
4244 :
4245 0 : ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4246 0 : if (ret) {
4247 0 : btrfs_info(fs_info,
4248 : "failed to delete reference to %.*s, inode %llu parent %llu",
4249 : name->len, name->name, ino, dir_ino);
4250 0 : btrfs_abort_transaction(trans, ret);
4251 0 : goto err;
4252 : }
4253 0 : skip_backref:
4254 0 : if (rename_ctx)
4255 0 : rename_ctx->index = index;
4256 :
4257 0 : ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4258 0 : if (ret) {
4259 0 : btrfs_abort_transaction(trans, ret);
4260 0 : goto err;
4261 : }
4262 :
4263 : /*
4264 : * If we are in a rename context, we don't need to update anything in the
4265 : * log. That will be done later during the rename by btrfs_log_new_name().
4266 : * Besides that, doing it here would only cause extra unnecessary btree
4267 : * operations on the log tree, increasing latency for applications.
4268 : */
4269 0 : if (!rename_ctx) {
4270 0 : btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4271 0 : btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4272 : }
4273 :
4274 : /*
4275 : * If we have a pending delayed iput we could end up with the final iput
4276 : * being run in btrfs-cleaner context. If we have enough of these built
4277 : * up we can end up burning a lot of time in btrfs-cleaner without any
4278 : * way to throttle the unlinks. Since we're currently holding a ref on
4279 : * the inode we can run the delayed iput here without any issues as the
4280 : * final iput won't be done until after we drop the ref we're currently
4281 : * holding.
4282 : */
4283 0 : btrfs_run_delayed_iput(fs_info, inode);
4284 0 : err:
4285 0 : btrfs_free_path(path);
4286 0 : if (ret)
4287 0 : goto out;
4288 :
4289 0 : btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4290 0 : inode_inc_iversion(&inode->vfs_inode);
4291 0 : inode_inc_iversion(&dir->vfs_inode);
4292 0 : inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4293 0 : dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime;
4294 0 : dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime;
4295 0 : ret = btrfs_update_inode(trans, root, dir);
4296 0 : out:
4297 0 : return ret;
4298 : }
4299 :
4300 0 : int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4301 : struct btrfs_inode *dir, struct btrfs_inode *inode,
4302 : const struct fscrypt_str *name)
4303 : {
4304 0 : int ret;
4305 :
4306 0 : ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4307 0 : if (!ret) {
4308 0 : drop_nlink(&inode->vfs_inode);
4309 0 : ret = btrfs_update_inode(trans, inode->root, inode);
4310 : }
4311 0 : return ret;
4312 : }
4313 :
4314 : /*
4315 : * helper to start transaction for unlink and rmdir.
4316 : *
4317 : * unlink and rmdir are special in btrfs, they do not always free space, so
4318 : * if we cannot make our reservations the normal way try and see if there is
4319 : * plenty of slack room in the global reserve to migrate, otherwise we cannot
4320 : * allow the unlink to occur.
4321 : */
4322 : static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4323 : {
4324 0 : struct btrfs_root *root = dir->root;
4325 :
4326 0 : return btrfs_start_transaction_fallback_global_rsv(root,
4327 : BTRFS_UNLINK_METADATA_UNITS);
4328 : }
4329 :
4330 0 : static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4331 : {
4332 0 : struct btrfs_trans_handle *trans;
4333 0 : struct inode *inode = d_inode(dentry);
4334 0 : int ret;
4335 0 : struct fscrypt_name fname;
4336 :
4337 0 : ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4338 0 : if (ret)
4339 : return ret;
4340 :
4341 : /* This needs to handle no-key deletions later on */
4342 :
4343 0 : trans = __unlink_start_trans(BTRFS_I(dir));
4344 0 : if (IS_ERR(trans)) {
4345 0 : ret = PTR_ERR(trans);
4346 0 : goto fscrypt_free;
4347 : }
4348 :
4349 0 : btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4350 : false);
4351 :
4352 0 : ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4353 : &fname.disk_name);
4354 0 : if (ret)
4355 0 : goto end_trans;
4356 :
4357 0 : if (inode->i_nlink == 0) {
4358 0 : ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4359 0 : if (ret)
4360 0 : goto end_trans;
4361 : }
4362 :
4363 0 : end_trans:
4364 0 : btrfs_end_transaction(trans);
4365 0 : btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4366 : fscrypt_free:
4367 : fscrypt_free_filename(&fname);
4368 : return ret;
4369 : }
4370 :
4371 0 : static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4372 : struct btrfs_inode *dir, struct dentry *dentry)
4373 : {
4374 0 : struct btrfs_root *root = dir->root;
4375 0 : struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4376 0 : struct btrfs_path *path;
4377 0 : struct extent_buffer *leaf;
4378 0 : struct btrfs_dir_item *di;
4379 0 : struct btrfs_key key;
4380 0 : u64 index;
4381 0 : int ret;
4382 0 : u64 objectid;
4383 0 : u64 dir_ino = btrfs_ino(dir);
4384 0 : struct fscrypt_name fname;
4385 :
4386 0 : ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4387 0 : if (ret)
4388 : return ret;
4389 :
4390 : /* This needs to handle no-key deletions later on */
4391 :
4392 0 : if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4393 0 : objectid = inode->root->root_key.objectid;
4394 0 : } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4395 0 : objectid = inode->location.objectid;
4396 : } else {
4397 0 : WARN_ON(1);
4398 0 : fscrypt_free_filename(&fname);
4399 0 : return -EINVAL;
4400 : }
4401 :
4402 0 : path = btrfs_alloc_path();
4403 0 : if (!path) {
4404 0 : ret = -ENOMEM;
4405 0 : goto out;
4406 : }
4407 :
4408 0 : di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4409 : &fname.disk_name, -1);
4410 0 : if (IS_ERR_OR_NULL(di)) {
4411 0 : ret = di ? PTR_ERR(di) : -ENOENT;
4412 0 : goto out;
4413 : }
4414 :
4415 0 : leaf = path->nodes[0];
4416 0 : btrfs_dir_item_key_to_cpu(leaf, di, &key);
4417 0 : WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4418 0 : ret = btrfs_delete_one_dir_name(trans, root, path, di);
4419 0 : if (ret) {
4420 0 : btrfs_abort_transaction(trans, ret);
4421 0 : goto out;
4422 : }
4423 0 : btrfs_release_path(path);
4424 :
4425 : /*
4426 : * This is a placeholder inode for a subvolume we didn't have a
4427 : * reference to at the time of the snapshot creation. In the meantime
4428 : * we could have renamed the real subvol link into our snapshot, so
4429 : * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4430 : * Instead simply lookup the dir_index_item for this entry so we can
4431 : * remove it. Otherwise we know we have a ref to the root and we can
4432 : * call btrfs_del_root_ref, and it _shouldn't_ fail.
4433 : */
4434 0 : if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4435 0 : di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4436 0 : if (IS_ERR_OR_NULL(di)) {
4437 0 : if (!di)
4438 : ret = -ENOENT;
4439 : else
4440 0 : ret = PTR_ERR(di);
4441 0 : btrfs_abort_transaction(trans, ret);
4442 0 : goto out;
4443 : }
4444 :
4445 0 : leaf = path->nodes[0];
4446 0 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4447 0 : index = key.offset;
4448 0 : btrfs_release_path(path);
4449 : } else {
4450 0 : ret = btrfs_del_root_ref(trans, objectid,
4451 : root->root_key.objectid, dir_ino,
4452 : &index, &fname.disk_name);
4453 0 : if (ret) {
4454 0 : btrfs_abort_transaction(trans, ret);
4455 0 : goto out;
4456 : }
4457 : }
4458 :
4459 0 : ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4460 0 : if (ret) {
4461 0 : btrfs_abort_transaction(trans, ret);
4462 0 : goto out;
4463 : }
4464 :
4465 0 : btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4466 0 : inode_inc_iversion(&dir->vfs_inode);
4467 0 : dir->vfs_inode.i_mtime = current_time(&dir->vfs_inode);
4468 0 : dir->vfs_inode.i_ctime = dir->vfs_inode.i_mtime;
4469 0 : ret = btrfs_update_inode_fallback(trans, root, dir);
4470 0 : if (ret)
4471 0 : btrfs_abort_transaction(trans, ret);
4472 0 : out:
4473 0 : btrfs_free_path(path);
4474 0 : fscrypt_free_filename(&fname);
4475 0 : return ret;
4476 : }
4477 :
4478 : /*
4479 : * Helper to check if the subvolume references other subvolumes or if it's
4480 : * default.
4481 : */
4482 0 : static noinline int may_destroy_subvol(struct btrfs_root *root)
4483 : {
4484 0 : struct btrfs_fs_info *fs_info = root->fs_info;
4485 0 : struct btrfs_path *path;
4486 0 : struct btrfs_dir_item *di;
4487 0 : struct btrfs_key key;
4488 0 : struct fscrypt_str name = FSTR_INIT("default", 7);
4489 0 : u64 dir_id;
4490 0 : int ret;
4491 :
4492 0 : path = btrfs_alloc_path();
4493 0 : if (!path)
4494 : return -ENOMEM;
4495 :
4496 : /* Make sure this root isn't set as the default subvol */
4497 0 : dir_id = btrfs_super_root_dir(fs_info->super_copy);
4498 0 : di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4499 : dir_id, &name, 0);
4500 0 : if (di && !IS_ERR(di)) {
4501 0 : btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4502 0 : if (key.objectid == root->root_key.objectid) {
4503 0 : ret = -EPERM;
4504 0 : btrfs_err(fs_info,
4505 : "deleting default subvolume %llu is not allowed",
4506 : key.objectid);
4507 0 : goto out;
4508 : }
4509 0 : btrfs_release_path(path);
4510 : }
4511 :
4512 0 : key.objectid = root->root_key.objectid;
4513 0 : key.type = BTRFS_ROOT_REF_KEY;
4514 0 : key.offset = (u64)-1;
4515 :
4516 0 : ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4517 0 : if (ret < 0)
4518 0 : goto out;
4519 0 : BUG_ON(ret == 0);
4520 :
4521 0 : ret = 0;
4522 0 : if (path->slots[0] > 0) {
4523 0 : path->slots[0]--;
4524 0 : btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4525 0 : if (key.objectid == root->root_key.objectid &&
4526 0 : key.type == BTRFS_ROOT_REF_KEY)
4527 0 : ret = -ENOTEMPTY;
4528 : }
4529 0 : out:
4530 0 : btrfs_free_path(path);
4531 0 : return ret;
4532 : }
4533 :
4534 : /* Delete all dentries for inodes belonging to the root */
4535 0 : static void btrfs_prune_dentries(struct btrfs_root *root)
4536 : {
4537 0 : struct btrfs_fs_info *fs_info = root->fs_info;
4538 0 : struct rb_node *node;
4539 0 : struct rb_node *prev;
4540 0 : struct btrfs_inode *entry;
4541 0 : struct inode *inode;
4542 0 : u64 objectid = 0;
4543 :
4544 0 : if (!BTRFS_FS_ERROR(fs_info))
4545 0 : WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4546 :
4547 0 : spin_lock(&root->inode_lock);
4548 : again:
4549 0 : node = root->inode_tree.rb_node;
4550 0 : prev = NULL;
4551 0 : while (node) {
4552 0 : prev = node;
4553 0 : entry = rb_entry(node, struct btrfs_inode, rb_node);
4554 :
4555 0 : if (objectid < btrfs_ino(entry))
4556 0 : node = node->rb_left;
4557 0 : else if (objectid > btrfs_ino(entry))
4558 0 : node = node->rb_right;
4559 : else
4560 : break;
4561 : }
4562 0 : if (!node) {
4563 0 : while (prev) {
4564 0 : entry = rb_entry(prev, struct btrfs_inode, rb_node);
4565 0 : if (objectid <= btrfs_ino(entry)) {
4566 : node = prev;
4567 : break;
4568 : }
4569 0 : prev = rb_next(prev);
4570 : }
4571 : }
4572 0 : while (node) {
4573 0 : entry = rb_entry(node, struct btrfs_inode, rb_node);
4574 0 : objectid = btrfs_ino(entry) + 1;
4575 0 : inode = igrab(&entry->vfs_inode);
4576 0 : if (inode) {
4577 0 : spin_unlock(&root->inode_lock);
4578 0 : if (atomic_read(&inode->i_count) > 1)
4579 0 : d_prune_aliases(inode);
4580 : /*
4581 : * btrfs_drop_inode will have it removed from the inode
4582 : * cache when its usage count hits zero.
4583 : */
4584 0 : iput(inode);
4585 0 : cond_resched();
4586 0 : spin_lock(&root->inode_lock);
4587 0 : goto again;
4588 : }
4589 :
4590 0 : if (cond_resched_lock(&root->inode_lock))
4591 0 : goto again;
4592 :
4593 0 : node = rb_next(node);
4594 : }
4595 0 : spin_unlock(&root->inode_lock);
4596 0 : }
4597 :
4598 0 : int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4599 : {
4600 0 : struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4601 0 : struct btrfs_root *root = dir->root;
4602 0 : struct inode *inode = d_inode(dentry);
4603 0 : struct btrfs_root *dest = BTRFS_I(inode)->root;
4604 0 : struct btrfs_trans_handle *trans;
4605 0 : struct btrfs_block_rsv block_rsv;
4606 0 : u64 root_flags;
4607 0 : int ret;
4608 :
4609 : /*
4610 : * Don't allow to delete a subvolume with send in progress. This is
4611 : * inside the inode lock so the error handling that has to drop the bit
4612 : * again is not run concurrently.
4613 : */
4614 0 : spin_lock(&dest->root_item_lock);
4615 0 : if (dest->send_in_progress) {
4616 0 : spin_unlock(&dest->root_item_lock);
4617 0 : btrfs_warn(fs_info,
4618 : "attempt to delete subvolume %llu during send",
4619 : dest->root_key.objectid);
4620 0 : return -EPERM;
4621 : }
4622 0 : if (atomic_read(&dest->nr_swapfiles)) {
4623 0 : spin_unlock(&dest->root_item_lock);
4624 0 : btrfs_warn(fs_info,
4625 : "attempt to delete subvolume %llu with active swapfile",
4626 : root->root_key.objectid);
4627 0 : return -EPERM;
4628 : }
4629 0 : root_flags = btrfs_root_flags(&dest->root_item);
4630 0 : btrfs_set_root_flags(&dest->root_item,
4631 : root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4632 0 : spin_unlock(&dest->root_item_lock);
4633 :
4634 0 : down_write(&fs_info->subvol_sem);
4635 :
4636 0 : ret = may_destroy_subvol(dest);
4637 0 : if (ret)
4638 0 : goto out_up_write;
4639 :
4640 0 : btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4641 : /*
4642 : * One for dir inode,
4643 : * two for dir entries,
4644 : * two for root ref/backref.
4645 : */
4646 0 : ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4647 0 : if (ret)
4648 0 : goto out_up_write;
4649 :
4650 0 : trans = btrfs_start_transaction(root, 0);
4651 0 : if (IS_ERR(trans)) {
4652 0 : ret = PTR_ERR(trans);
4653 0 : goto out_release;
4654 : }
4655 0 : trans->block_rsv = &block_rsv;
4656 0 : trans->bytes_reserved = block_rsv.size;
4657 :
4658 0 : btrfs_record_snapshot_destroy(trans, dir);
4659 :
4660 0 : ret = btrfs_unlink_subvol(trans, dir, dentry);
4661 0 : if (ret) {
4662 0 : btrfs_abort_transaction(trans, ret);
4663 0 : goto out_end_trans;
4664 : }
4665 :
4666 0 : ret = btrfs_record_root_in_trans(trans, dest);
4667 0 : if (ret) {
4668 0 : btrfs_abort_transaction(trans, ret);
4669 0 : goto out_end_trans;
4670 : }
4671 :
4672 0 : memset(&dest->root_item.drop_progress, 0,
4673 : sizeof(dest->root_item.drop_progress));
4674 0 : btrfs_set_root_drop_level(&dest->root_item, 0);
4675 0 : btrfs_set_root_refs(&dest->root_item, 0);
4676 :
4677 0 : if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4678 0 : ret = btrfs_insert_orphan_item(trans,
4679 : fs_info->tree_root,
4680 : dest->root_key.objectid);
4681 0 : if (ret) {
4682 0 : btrfs_abort_transaction(trans, ret);
4683 0 : goto out_end_trans;
4684 : }
4685 : }
4686 :
4687 0 : ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4688 : BTRFS_UUID_KEY_SUBVOL,
4689 : dest->root_key.objectid);
4690 0 : if (ret && ret != -ENOENT) {
4691 0 : btrfs_abort_transaction(trans, ret);
4692 0 : goto out_end_trans;
4693 : }
4694 0 : if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4695 0 : ret = btrfs_uuid_tree_remove(trans,
4696 : dest->root_item.received_uuid,
4697 : BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4698 : dest->root_key.objectid);
4699 0 : if (ret && ret != -ENOENT) {
4700 0 : btrfs_abort_transaction(trans, ret);
4701 0 : goto out_end_trans;
4702 : }
4703 : }
4704 :
4705 0 : free_anon_bdev(dest->anon_dev);
4706 0 : dest->anon_dev = 0;
4707 0 : out_end_trans:
4708 0 : trans->block_rsv = NULL;
4709 0 : trans->bytes_reserved = 0;
4710 0 : ret = btrfs_end_transaction(trans);
4711 0 : inode->i_flags |= S_DEAD;
4712 0 : out_release:
4713 0 : btrfs_subvolume_release_metadata(root, &block_rsv);
4714 0 : out_up_write:
4715 0 : up_write(&fs_info->subvol_sem);
4716 0 : if (ret) {
4717 0 : spin_lock(&dest->root_item_lock);
4718 0 : root_flags = btrfs_root_flags(&dest->root_item);
4719 0 : btrfs_set_root_flags(&dest->root_item,
4720 : root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4721 0 : spin_unlock(&dest->root_item_lock);
4722 : } else {
4723 0 : d_invalidate(dentry);
4724 0 : btrfs_prune_dentries(dest);
4725 0 : ASSERT(dest->send_in_progress == 0);
4726 : }
4727 :
4728 : return ret;
4729 : }
4730 :
4731 0 : static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4732 : {
4733 0 : struct inode *inode = d_inode(dentry);
4734 0 : struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4735 0 : int err = 0;
4736 0 : struct btrfs_trans_handle *trans;
4737 0 : u64 last_unlink_trans;
4738 0 : struct fscrypt_name fname;
4739 :
4740 0 : if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4741 : return -ENOTEMPTY;
4742 0 : if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4743 0 : if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4744 0 : btrfs_err(fs_info,
4745 : "extent tree v2 doesn't support snapshot deletion yet");
4746 0 : return -EOPNOTSUPP;
4747 : }
4748 0 : return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4749 : }
4750 :
4751 0 : err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4752 0 : if (err)
4753 : return err;
4754 :
4755 : /* This needs to handle no-key deletions later on */
4756 :
4757 0 : trans = __unlink_start_trans(BTRFS_I(dir));
4758 0 : if (IS_ERR(trans)) {
4759 0 : err = PTR_ERR(trans);
4760 0 : goto out_notrans;
4761 : }
4762 :
4763 0 : if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4764 0 : err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4765 0 : goto out;
4766 : }
4767 :
4768 0 : err = btrfs_orphan_add(trans, BTRFS_I(inode));
4769 0 : if (err)
4770 0 : goto out;
4771 :
4772 0 : last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4773 :
4774 : /* now the directory is empty */
4775 0 : err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4776 : &fname.disk_name);
4777 0 : if (!err) {
4778 0 : btrfs_i_size_write(BTRFS_I(inode), 0);
4779 : /*
4780 : * Propagate the last_unlink_trans value of the deleted dir to
4781 : * its parent directory. This is to prevent an unrecoverable
4782 : * log tree in the case we do something like this:
4783 : * 1) create dir foo
4784 : * 2) create snapshot under dir foo
4785 : * 3) delete the snapshot
4786 : * 4) rmdir foo
4787 : * 5) mkdir foo
4788 : * 6) fsync foo or some file inside foo
4789 : */
4790 0 : if (last_unlink_trans >= trans->transid)
4791 0 : BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4792 : }
4793 0 : out:
4794 0 : btrfs_end_transaction(trans);
4795 0 : out_notrans:
4796 0 : btrfs_btree_balance_dirty(fs_info);
4797 0 : fscrypt_free_filename(&fname);
4798 :
4799 0 : return err;
4800 : }
4801 :
4802 : /*
4803 : * btrfs_truncate_block - read, zero a chunk and write a block
4804 : * @inode - inode that we're zeroing
4805 : * @from - the offset to start zeroing
4806 : * @len - the length to zero, 0 to zero the entire range respective to the
4807 : * offset
4808 : * @front - zero up to the offset instead of from the offset on
4809 : *
4810 : * This will find the block for the "from" offset and cow the block and zero the
4811 : * part we want to zero. This is used with truncate and hole punching.
4812 : */
4813 0 : int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4814 : int front)
4815 : {
4816 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
4817 0 : struct address_space *mapping = inode->vfs_inode.i_mapping;
4818 0 : struct extent_io_tree *io_tree = &inode->io_tree;
4819 0 : struct btrfs_ordered_extent *ordered;
4820 0 : struct extent_state *cached_state = NULL;
4821 0 : struct extent_changeset *data_reserved = NULL;
4822 0 : bool only_release_metadata = false;
4823 0 : u32 blocksize = fs_info->sectorsize;
4824 0 : pgoff_t index = from >> PAGE_SHIFT;
4825 0 : unsigned offset = from & (blocksize - 1);
4826 0 : struct page *page;
4827 0 : gfp_t mask = btrfs_alloc_write_mask(mapping);
4828 0 : size_t write_bytes = blocksize;
4829 0 : int ret = 0;
4830 0 : u64 block_start;
4831 0 : u64 block_end;
4832 :
4833 0 : if (IS_ALIGNED(offset, blocksize) &&
4834 0 : (!len || IS_ALIGNED(len, blocksize)))
4835 0 : goto out;
4836 :
4837 0 : block_start = round_down(from, blocksize);
4838 0 : block_end = block_start + blocksize - 1;
4839 :
4840 0 : ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4841 : blocksize, false);
4842 0 : if (ret < 0) {
4843 0 : if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4844 : /* For nocow case, no need to reserve data space */
4845 : only_release_metadata = true;
4846 : } else {
4847 0 : goto out;
4848 : }
4849 : }
4850 0 : ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4851 0 : if (ret < 0) {
4852 0 : if (!only_release_metadata)
4853 0 : btrfs_free_reserved_data_space(inode, data_reserved,
4854 : block_start, blocksize);
4855 0 : goto out;
4856 : }
4857 0 : again:
4858 0 : page = find_or_create_page(mapping, index, mask);
4859 0 : if (!page) {
4860 0 : btrfs_delalloc_release_space(inode, data_reserved, block_start,
4861 : blocksize, true);
4862 0 : btrfs_delalloc_release_extents(inode, blocksize);
4863 0 : ret = -ENOMEM;
4864 0 : goto out;
4865 : }
4866 :
4867 0 : if (!PageUptodate(page)) {
4868 0 : ret = btrfs_read_folio(NULL, page_folio(page));
4869 0 : lock_page(page);
4870 0 : if (page->mapping != mapping) {
4871 0 : unlock_page(page);
4872 0 : put_page(page);
4873 0 : goto again;
4874 : }
4875 0 : if (!PageUptodate(page)) {
4876 0 : ret = -EIO;
4877 0 : goto out_unlock;
4878 : }
4879 : }
4880 :
4881 : /*
4882 : * We unlock the page after the io is completed and then re-lock it
4883 : * above. release_folio() could have come in between that and cleared
4884 : * PagePrivate(), but left the page in the mapping. Set the page mapped
4885 : * here to make sure it's properly set for the subpage stuff.
4886 : */
4887 0 : ret = set_page_extent_mapped(page);
4888 0 : if (ret < 0)
4889 0 : goto out_unlock;
4890 :
4891 0 : wait_on_page_writeback(page);
4892 :
4893 0 : lock_extent(io_tree, block_start, block_end, &cached_state);
4894 :
4895 0 : ordered = btrfs_lookup_ordered_extent(inode, block_start);
4896 0 : if (ordered) {
4897 0 : unlock_extent(io_tree, block_start, block_end, &cached_state);
4898 0 : unlock_page(page);
4899 0 : put_page(page);
4900 0 : btrfs_start_ordered_extent(ordered);
4901 0 : btrfs_put_ordered_extent(ordered);
4902 0 : goto again;
4903 : }
4904 :
4905 0 : clear_extent_bit(&inode->io_tree, block_start, block_end,
4906 : EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4907 : &cached_state);
4908 :
4909 0 : ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4910 : &cached_state);
4911 0 : if (ret) {
4912 0 : unlock_extent(io_tree, block_start, block_end, &cached_state);
4913 0 : goto out_unlock;
4914 : }
4915 :
4916 0 : if (offset != blocksize) {
4917 0 : if (!len)
4918 0 : len = blocksize - offset;
4919 0 : if (front)
4920 0 : memzero_page(page, (block_start - page_offset(page)),
4921 : offset);
4922 : else
4923 0 : memzero_page(page, (block_start - page_offset(page)) + offset,
4924 : len);
4925 : }
4926 0 : btrfs_page_clear_checked(fs_info, page, block_start,
4927 0 : block_end + 1 - block_start);
4928 0 : btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4929 0 : unlock_extent(io_tree, block_start, block_end, &cached_state);
4930 :
4931 0 : if (only_release_metadata)
4932 0 : set_extent_bit(&inode->io_tree, block_start, block_end,
4933 : EXTENT_NORESERVE, NULL);
4934 :
4935 0 : out_unlock:
4936 0 : if (ret) {
4937 0 : if (only_release_metadata)
4938 0 : btrfs_delalloc_release_metadata(inode, blocksize, true);
4939 : else
4940 0 : btrfs_delalloc_release_space(inode, data_reserved,
4941 : block_start, blocksize, true);
4942 : }
4943 0 : btrfs_delalloc_release_extents(inode, blocksize);
4944 0 : unlock_page(page);
4945 0 : put_page(page);
4946 0 : out:
4947 0 : if (only_release_metadata)
4948 0 : btrfs_check_nocow_unlock(inode);
4949 0 : extent_changeset_free(data_reserved);
4950 0 : return ret;
4951 : }
4952 :
4953 0 : static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4954 : u64 offset, u64 len)
4955 : {
4956 0 : struct btrfs_fs_info *fs_info = root->fs_info;
4957 0 : struct btrfs_trans_handle *trans;
4958 0 : struct btrfs_drop_extents_args drop_args = { 0 };
4959 0 : int ret;
4960 :
4961 : /*
4962 : * If NO_HOLES is enabled, we don't need to do anything.
4963 : * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4964 : * or btrfs_update_inode() will be called, which guarantee that the next
4965 : * fsync will know this inode was changed and needs to be logged.
4966 : */
4967 0 : if (btrfs_fs_incompat(fs_info, NO_HOLES))
4968 : return 0;
4969 :
4970 : /*
4971 : * 1 - for the one we're dropping
4972 : * 1 - for the one we're adding
4973 : * 1 - for updating the inode.
4974 : */
4975 0 : trans = btrfs_start_transaction(root, 3);
4976 0 : if (IS_ERR(trans))
4977 0 : return PTR_ERR(trans);
4978 :
4979 0 : drop_args.start = offset;
4980 0 : drop_args.end = offset + len;
4981 0 : drop_args.drop_cache = true;
4982 :
4983 0 : ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4984 0 : if (ret) {
4985 0 : btrfs_abort_transaction(trans, ret);
4986 0 : btrfs_end_transaction(trans);
4987 0 : return ret;
4988 : }
4989 :
4990 0 : ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4991 0 : if (ret) {
4992 0 : btrfs_abort_transaction(trans, ret);
4993 : } else {
4994 0 : btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4995 0 : btrfs_update_inode(trans, root, inode);
4996 : }
4997 0 : btrfs_end_transaction(trans);
4998 0 : return ret;
4999 : }
5000 :
5001 : /*
5002 : * This function puts in dummy file extents for the area we're creating a hole
5003 : * for. So if we are truncating this file to a larger size we need to insert
5004 : * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5005 : * the range between oldsize and size
5006 : */
5007 0 : int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
5008 : {
5009 0 : struct btrfs_root *root = inode->root;
5010 0 : struct btrfs_fs_info *fs_info = root->fs_info;
5011 0 : struct extent_io_tree *io_tree = &inode->io_tree;
5012 0 : struct extent_map *em = NULL;
5013 0 : struct extent_state *cached_state = NULL;
5014 0 : u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
5015 0 : u64 block_end = ALIGN(size, fs_info->sectorsize);
5016 0 : u64 last_byte;
5017 0 : u64 cur_offset;
5018 0 : u64 hole_size;
5019 0 : int err = 0;
5020 :
5021 : /*
5022 : * If our size started in the middle of a block we need to zero out the
5023 : * rest of the block before we expand the i_size, otherwise we could
5024 : * expose stale data.
5025 : */
5026 0 : err = btrfs_truncate_block(inode, oldsize, 0, 0);
5027 0 : if (err)
5028 : return err;
5029 :
5030 0 : if (size <= hole_start)
5031 : return 0;
5032 :
5033 0 : btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
5034 : &cached_state);
5035 0 : cur_offset = hole_start;
5036 0 : while (1) {
5037 0 : em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5038 : block_end - cur_offset);
5039 0 : if (IS_ERR(em)) {
5040 0 : err = PTR_ERR(em);
5041 0 : em = NULL;
5042 0 : break;
5043 : }
5044 0 : last_byte = min(extent_map_end(em), block_end);
5045 0 : last_byte = ALIGN(last_byte, fs_info->sectorsize);
5046 0 : hole_size = last_byte - cur_offset;
5047 :
5048 0 : if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5049 0 : struct extent_map *hole_em;
5050 :
5051 0 : err = maybe_insert_hole(root, inode, cur_offset,
5052 : hole_size);
5053 0 : if (err)
5054 : break;
5055 :
5056 0 : err = btrfs_inode_set_file_extent_range(inode,
5057 : cur_offset, hole_size);
5058 0 : if (err)
5059 : break;
5060 :
5061 0 : hole_em = alloc_extent_map();
5062 0 : if (!hole_em) {
5063 0 : btrfs_drop_extent_map_range(inode, cur_offset,
5064 : cur_offset + hole_size - 1,
5065 : false);
5066 0 : btrfs_set_inode_full_sync(inode);
5067 0 : goto next;
5068 : }
5069 0 : hole_em->start = cur_offset;
5070 0 : hole_em->len = hole_size;
5071 0 : hole_em->orig_start = cur_offset;
5072 :
5073 0 : hole_em->block_start = EXTENT_MAP_HOLE;
5074 0 : hole_em->block_len = 0;
5075 0 : hole_em->orig_block_len = 0;
5076 0 : hole_em->ram_bytes = hole_size;
5077 0 : hole_em->compress_type = BTRFS_COMPRESS_NONE;
5078 0 : hole_em->generation = fs_info->generation;
5079 :
5080 0 : err = btrfs_replace_extent_map_range(inode, hole_em, true);
5081 0 : free_extent_map(hole_em);
5082 : } else {
5083 0 : err = btrfs_inode_set_file_extent_range(inode,
5084 : cur_offset, hole_size);
5085 0 : if (err)
5086 : break;
5087 : }
5088 0 : next:
5089 0 : free_extent_map(em);
5090 0 : em = NULL;
5091 0 : cur_offset = last_byte;
5092 0 : if (cur_offset >= block_end)
5093 : break;
5094 : }
5095 0 : free_extent_map(em);
5096 0 : unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
5097 0 : return err;
5098 : }
5099 :
5100 0 : static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5101 : {
5102 0 : struct btrfs_root *root = BTRFS_I(inode)->root;
5103 0 : struct btrfs_trans_handle *trans;
5104 0 : loff_t oldsize = i_size_read(inode);
5105 0 : loff_t newsize = attr->ia_size;
5106 0 : int mask = attr->ia_valid;
5107 0 : int ret;
5108 :
5109 : /*
5110 : * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5111 : * special case where we need to update the times despite not having
5112 : * these flags set. For all other operations the VFS set these flags
5113 : * explicitly if it wants a timestamp update.
5114 : */
5115 0 : if (newsize != oldsize) {
5116 0 : inode_inc_iversion(inode);
5117 0 : if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
5118 0 : inode->i_mtime = current_time(inode);
5119 0 : inode->i_ctime = inode->i_mtime;
5120 : }
5121 : }
5122 :
5123 0 : if (newsize > oldsize) {
5124 : /*
5125 : * Don't do an expanding truncate while snapshotting is ongoing.
5126 : * This is to ensure the snapshot captures a fully consistent
5127 : * state of this file - if the snapshot captures this expanding
5128 : * truncation, it must capture all writes that happened before
5129 : * this truncation.
5130 : */
5131 0 : btrfs_drew_write_lock(&root->snapshot_lock);
5132 0 : ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5133 0 : if (ret) {
5134 0 : btrfs_drew_write_unlock(&root->snapshot_lock);
5135 0 : return ret;
5136 : }
5137 :
5138 0 : trans = btrfs_start_transaction(root, 1);
5139 0 : if (IS_ERR(trans)) {
5140 0 : btrfs_drew_write_unlock(&root->snapshot_lock);
5141 0 : return PTR_ERR(trans);
5142 : }
5143 :
5144 0 : i_size_write(inode, newsize);
5145 0 : btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5146 0 : pagecache_isize_extended(inode, oldsize, newsize);
5147 0 : ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5148 0 : btrfs_drew_write_unlock(&root->snapshot_lock);
5149 0 : btrfs_end_transaction(trans);
5150 : } else {
5151 0 : struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5152 :
5153 0 : if (btrfs_is_zoned(fs_info)) {
5154 0 : ret = btrfs_wait_ordered_range(inode,
5155 0 : ALIGN(newsize, fs_info->sectorsize),
5156 : (u64)-1);
5157 0 : if (ret)
5158 : return ret;
5159 : }
5160 :
5161 : /*
5162 : * We're truncating a file that used to have good data down to
5163 : * zero. Make sure any new writes to the file get on disk
5164 : * on close.
5165 : */
5166 0 : if (newsize == 0)
5167 0 : set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5168 0 : &BTRFS_I(inode)->runtime_flags);
5169 :
5170 0 : truncate_setsize(inode, newsize);
5171 :
5172 0 : inode_dio_wait(inode);
5173 :
5174 0 : ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5175 0 : if (ret && inode->i_nlink) {
5176 0 : int err;
5177 :
5178 : /*
5179 : * Truncate failed, so fix up the in-memory size. We
5180 : * adjusted disk_i_size down as we removed extents, so
5181 : * wait for disk_i_size to be stable and then update the
5182 : * in-memory size to match.
5183 : */
5184 0 : err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5185 0 : if (err)
5186 : return err;
5187 0 : i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5188 : }
5189 : }
5190 :
5191 : return ret;
5192 : }
5193 :
5194 0 : static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5195 : struct iattr *attr)
5196 : {
5197 0 : struct inode *inode = d_inode(dentry);
5198 0 : struct btrfs_root *root = BTRFS_I(inode)->root;
5199 0 : int err;
5200 :
5201 0 : if (btrfs_root_readonly(root))
5202 : return -EROFS;
5203 :
5204 0 : err = setattr_prepare(idmap, dentry, attr);
5205 0 : if (err)
5206 : return err;
5207 :
5208 0 : if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5209 0 : err = btrfs_setsize(inode, attr);
5210 0 : if (err)
5211 : return err;
5212 : }
5213 :
5214 0 : if (attr->ia_valid) {
5215 0 : setattr_copy(idmap, inode, attr);
5216 0 : inode_inc_iversion(inode);
5217 0 : err = btrfs_dirty_inode(BTRFS_I(inode));
5218 :
5219 0 : if (!err && attr->ia_valid & ATTR_MODE)
5220 0 : err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5221 : }
5222 :
5223 : return err;
5224 : }
5225 :
5226 : /*
5227 : * While truncating the inode pages during eviction, we get the VFS
5228 : * calling btrfs_invalidate_folio() against each folio of the inode. This
5229 : * is slow because the calls to btrfs_invalidate_folio() result in a
5230 : * huge amount of calls to lock_extent() and clear_extent_bit(),
5231 : * which keep merging and splitting extent_state structures over and over,
5232 : * wasting lots of time.
5233 : *
5234 : * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5235 : * skip all those expensive operations on a per folio basis and do only
5236 : * the ordered io finishing, while we release here the extent_map and
5237 : * extent_state structures, without the excessive merging and splitting.
5238 : */
5239 0 : static void evict_inode_truncate_pages(struct inode *inode)
5240 : {
5241 0 : struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5242 0 : struct rb_node *node;
5243 :
5244 0 : ASSERT(inode->i_state & I_FREEING);
5245 0 : truncate_inode_pages_final(&inode->i_data);
5246 :
5247 0 : btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5248 :
5249 : /*
5250 : * Keep looping until we have no more ranges in the io tree.
5251 : * We can have ongoing bios started by readahead that have
5252 : * their endio callback (extent_io.c:end_bio_extent_readpage)
5253 : * still in progress (unlocked the pages in the bio but did not yet
5254 : * unlocked the ranges in the io tree). Therefore this means some
5255 : * ranges can still be locked and eviction started because before
5256 : * submitting those bios, which are executed by a separate task (work
5257 : * queue kthread), inode references (inode->i_count) were not taken
5258 : * (which would be dropped in the end io callback of each bio).
5259 : * Therefore here we effectively end up waiting for those bios and
5260 : * anyone else holding locked ranges without having bumped the inode's
5261 : * reference count - if we don't do it, when they access the inode's
5262 : * io_tree to unlock a range it may be too late, leading to an
5263 : * use-after-free issue.
5264 : */
5265 0 : spin_lock(&io_tree->lock);
5266 0 : while (!RB_EMPTY_ROOT(&io_tree->state)) {
5267 0 : struct extent_state *state;
5268 0 : struct extent_state *cached_state = NULL;
5269 0 : u64 start;
5270 0 : u64 end;
5271 0 : unsigned state_flags;
5272 :
5273 0 : node = rb_first(&io_tree->state);
5274 0 : state = rb_entry(node, struct extent_state, rb_node);
5275 0 : start = state->start;
5276 0 : end = state->end;
5277 0 : state_flags = state->state;
5278 0 : spin_unlock(&io_tree->lock);
5279 :
5280 0 : lock_extent(io_tree, start, end, &cached_state);
5281 :
5282 : /*
5283 : * If still has DELALLOC flag, the extent didn't reach disk,
5284 : * and its reserved space won't be freed by delayed_ref.
5285 : * So we need to free its reserved space here.
5286 : * (Refer to comment in btrfs_invalidate_folio, case 2)
5287 : *
5288 : * Note, end is the bytenr of last byte, so we need + 1 here.
5289 : */
5290 0 : if (state_flags & EXTENT_DELALLOC)
5291 0 : btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5292 0 : end - start + 1);
5293 :
5294 0 : clear_extent_bit(io_tree, start, end,
5295 : EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5296 : &cached_state);
5297 :
5298 0 : cond_resched();
5299 0 : spin_lock(&io_tree->lock);
5300 : }
5301 0 : spin_unlock(&io_tree->lock);
5302 0 : }
5303 :
5304 0 : static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5305 : struct btrfs_block_rsv *rsv)
5306 : {
5307 0 : struct btrfs_fs_info *fs_info = root->fs_info;
5308 0 : struct btrfs_trans_handle *trans;
5309 0 : u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5310 0 : int ret;
5311 :
5312 : /*
5313 : * Eviction should be taking place at some place safe because of our
5314 : * delayed iputs. However the normal flushing code will run delayed
5315 : * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5316 : *
5317 : * We reserve the delayed_refs_extra here again because we can't use
5318 : * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5319 : * above. We reserve our extra bit here because we generate a ton of
5320 : * delayed refs activity by truncating.
5321 : *
5322 : * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5323 : * if we fail to make this reservation we can re-try without the
5324 : * delayed_refs_extra so we can make some forward progress.
5325 : */
5326 0 : ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5327 : BTRFS_RESERVE_FLUSH_EVICT);
5328 0 : if (ret) {
5329 0 : ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5330 : BTRFS_RESERVE_FLUSH_EVICT);
5331 0 : if (ret) {
5332 0 : btrfs_warn(fs_info,
5333 : "could not allocate space for delete; will truncate on mount");
5334 0 : return ERR_PTR(-ENOSPC);
5335 : }
5336 : delayed_refs_extra = 0;
5337 : }
5338 :
5339 0 : trans = btrfs_join_transaction(root);
5340 0 : if (IS_ERR(trans))
5341 : return trans;
5342 :
5343 0 : if (delayed_refs_extra) {
5344 0 : trans->block_rsv = &fs_info->trans_block_rsv;
5345 0 : trans->bytes_reserved = delayed_refs_extra;
5346 0 : btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5347 : delayed_refs_extra, true);
5348 : }
5349 : return trans;
5350 : }
5351 :
5352 0 : void btrfs_evict_inode(struct inode *inode)
5353 : {
5354 0 : struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5355 0 : struct btrfs_trans_handle *trans;
5356 0 : struct btrfs_root *root = BTRFS_I(inode)->root;
5357 0 : struct btrfs_block_rsv *rsv = NULL;
5358 0 : int ret;
5359 :
5360 0 : trace_btrfs_inode_evict(inode);
5361 :
5362 0 : if (!root) {
5363 0 : fsverity_cleanup_inode(inode);
5364 0 : clear_inode(inode);
5365 0 : return;
5366 : }
5367 :
5368 0 : evict_inode_truncate_pages(inode);
5369 :
5370 0 : if (inode->i_nlink &&
5371 0 : ((btrfs_root_refs(&root->root_item) != 0 &&
5372 0 : root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5373 : btrfs_is_free_space_inode(BTRFS_I(inode))))
5374 0 : goto out;
5375 :
5376 0 : if (is_bad_inode(inode))
5377 0 : goto out;
5378 :
5379 0 : if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5380 0 : goto out;
5381 :
5382 0 : if (inode->i_nlink > 0) {
5383 0 : BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5384 : root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5385 0 : goto out;
5386 : }
5387 :
5388 : /*
5389 : * This makes sure the inode item in tree is uptodate and the space for
5390 : * the inode update is released.
5391 : */
5392 0 : ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5393 0 : if (ret)
5394 0 : goto out;
5395 :
5396 : /*
5397 : * This drops any pending insert or delete operations we have for this
5398 : * inode. We could have a delayed dir index deletion queued up, but
5399 : * we're removing the inode completely so that'll be taken care of in
5400 : * the truncate.
5401 : */
5402 0 : btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5403 :
5404 0 : rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5405 0 : if (!rsv)
5406 0 : goto out;
5407 0 : rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5408 0 : rsv->failfast = true;
5409 :
5410 0 : btrfs_i_size_write(BTRFS_I(inode), 0);
5411 :
5412 0 : while (1) {
5413 0 : struct btrfs_truncate_control control = {
5414 : .inode = BTRFS_I(inode),
5415 : .ino = btrfs_ino(BTRFS_I(inode)),
5416 : .new_size = 0,
5417 : .min_type = 0,
5418 : };
5419 :
5420 0 : trans = evict_refill_and_join(root, rsv);
5421 0 : if (IS_ERR(trans))
5422 0 : goto out;
5423 :
5424 0 : trans->block_rsv = rsv;
5425 :
5426 0 : ret = btrfs_truncate_inode_items(trans, root, &control);
5427 0 : trans->block_rsv = &fs_info->trans_block_rsv;
5428 0 : btrfs_end_transaction(trans);
5429 : /*
5430 : * We have not added new delayed items for our inode after we
5431 : * have flushed its delayed items, so no need to throttle on
5432 : * delayed items. However we have modified extent buffers.
5433 : */
5434 0 : btrfs_btree_balance_dirty_nodelay(fs_info);
5435 0 : if (ret && ret != -ENOSPC && ret != -EAGAIN)
5436 0 : goto out;
5437 0 : else if (!ret)
5438 : break;
5439 : }
5440 :
5441 : /*
5442 : * Errors here aren't a big deal, it just means we leave orphan items in
5443 : * the tree. They will be cleaned up on the next mount. If the inode
5444 : * number gets reused, cleanup deletes the orphan item without doing
5445 : * anything, and unlink reuses the existing orphan item.
5446 : *
5447 : * If it turns out that we are dropping too many of these, we might want
5448 : * to add a mechanism for retrying these after a commit.
5449 : */
5450 0 : trans = evict_refill_and_join(root, rsv);
5451 0 : if (!IS_ERR(trans)) {
5452 0 : trans->block_rsv = rsv;
5453 0 : btrfs_orphan_del(trans, BTRFS_I(inode));
5454 0 : trans->block_rsv = &fs_info->trans_block_rsv;
5455 0 : btrfs_end_transaction(trans);
5456 : }
5457 :
5458 0 : out:
5459 0 : btrfs_free_block_rsv(fs_info, rsv);
5460 : /*
5461 : * If we didn't successfully delete, the orphan item will still be in
5462 : * the tree and we'll retry on the next mount. Again, we might also want
5463 : * to retry these periodically in the future.
5464 : */
5465 0 : btrfs_remove_delayed_node(BTRFS_I(inode));
5466 0 : fsverity_cleanup_inode(inode);
5467 0 : clear_inode(inode);
5468 : }
5469 :
5470 : /*
5471 : * Return the key found in the dir entry in the location pointer, fill @type
5472 : * with BTRFS_FT_*, and return 0.
5473 : *
5474 : * If no dir entries were found, returns -ENOENT.
5475 : * If found a corrupted location in dir entry, returns -EUCLEAN.
5476 : */
5477 0 : static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5478 : struct btrfs_key *location, u8 *type)
5479 : {
5480 0 : struct btrfs_dir_item *di;
5481 0 : struct btrfs_path *path;
5482 0 : struct btrfs_root *root = dir->root;
5483 0 : int ret = 0;
5484 0 : struct fscrypt_name fname;
5485 :
5486 0 : path = btrfs_alloc_path();
5487 0 : if (!path)
5488 : return -ENOMEM;
5489 :
5490 0 : ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5491 0 : if (ret < 0)
5492 0 : goto out;
5493 : /*
5494 : * fscrypt_setup_filename() should never return a positive value, but
5495 : * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5496 : */
5497 0 : ASSERT(ret == 0);
5498 :
5499 : /* This needs to handle no-key deletions later on */
5500 :
5501 0 : di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5502 : &fname.disk_name, 0);
5503 0 : if (IS_ERR_OR_NULL(di)) {
5504 0 : ret = di ? PTR_ERR(di) : -ENOENT;
5505 0 : goto out;
5506 : }
5507 :
5508 0 : btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5509 0 : if (location->type != BTRFS_INODE_ITEM_KEY &&
5510 : location->type != BTRFS_ROOT_ITEM_KEY) {
5511 0 : ret = -EUCLEAN;
5512 0 : btrfs_warn(root->fs_info,
5513 : "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5514 : __func__, fname.disk_name.name, btrfs_ino(dir),
5515 : location->objectid, location->type, location->offset);
5516 : }
5517 0 : if (!ret)
5518 0 : *type = btrfs_dir_ftype(path->nodes[0], di);
5519 0 : out:
5520 0 : fscrypt_free_filename(&fname);
5521 0 : btrfs_free_path(path);
5522 0 : return ret;
5523 : }
5524 :
5525 : /*
5526 : * when we hit a tree root in a directory, the btrfs part of the inode
5527 : * needs to be changed to reflect the root directory of the tree root. This
5528 : * is kind of like crossing a mount point.
5529 : */
5530 0 : static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5531 : struct btrfs_inode *dir,
5532 : struct dentry *dentry,
5533 : struct btrfs_key *location,
5534 : struct btrfs_root **sub_root)
5535 : {
5536 0 : struct btrfs_path *path;
5537 0 : struct btrfs_root *new_root;
5538 0 : struct btrfs_root_ref *ref;
5539 0 : struct extent_buffer *leaf;
5540 0 : struct btrfs_key key;
5541 0 : int ret;
5542 0 : int err = 0;
5543 0 : struct fscrypt_name fname;
5544 :
5545 0 : ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5546 0 : if (ret)
5547 : return ret;
5548 :
5549 0 : path = btrfs_alloc_path();
5550 0 : if (!path) {
5551 0 : err = -ENOMEM;
5552 0 : goto out;
5553 : }
5554 :
5555 0 : err = -ENOENT;
5556 0 : key.objectid = dir->root->root_key.objectid;
5557 0 : key.type = BTRFS_ROOT_REF_KEY;
5558 0 : key.offset = location->objectid;
5559 :
5560 0 : ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5561 0 : if (ret) {
5562 0 : if (ret < 0)
5563 0 : err = ret;
5564 0 : goto out;
5565 : }
5566 :
5567 0 : leaf = path->nodes[0];
5568 0 : ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5569 0 : if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5570 0 : btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5571 0 : goto out;
5572 :
5573 0 : ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5574 0 : (unsigned long)(ref + 1), fname.disk_name.len);
5575 0 : if (ret)
5576 0 : goto out;
5577 :
5578 0 : btrfs_release_path(path);
5579 :
5580 0 : new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5581 0 : if (IS_ERR(new_root)) {
5582 0 : err = PTR_ERR(new_root);
5583 0 : goto out;
5584 : }
5585 :
5586 0 : *sub_root = new_root;
5587 0 : location->objectid = btrfs_root_dirid(&new_root->root_item);
5588 0 : location->type = BTRFS_INODE_ITEM_KEY;
5589 0 : location->offset = 0;
5590 0 : err = 0;
5591 0 : out:
5592 0 : btrfs_free_path(path);
5593 0 : fscrypt_free_filename(&fname);
5594 0 : return err;
5595 : }
5596 :
5597 0 : static void inode_tree_add(struct btrfs_inode *inode)
5598 : {
5599 0 : struct btrfs_root *root = inode->root;
5600 0 : struct btrfs_inode *entry;
5601 0 : struct rb_node **p;
5602 0 : struct rb_node *parent;
5603 0 : struct rb_node *new = &inode->rb_node;
5604 0 : u64 ino = btrfs_ino(inode);
5605 :
5606 0 : if (inode_unhashed(&inode->vfs_inode))
5607 : return;
5608 0 : parent = NULL;
5609 0 : spin_lock(&root->inode_lock);
5610 0 : p = &root->inode_tree.rb_node;
5611 0 : while (*p) {
5612 0 : parent = *p;
5613 0 : entry = rb_entry(parent, struct btrfs_inode, rb_node);
5614 :
5615 0 : if (ino < btrfs_ino(entry))
5616 0 : p = &parent->rb_left;
5617 0 : else if (ino > btrfs_ino(entry))
5618 0 : p = &parent->rb_right;
5619 : else {
5620 0 : WARN_ON(!(entry->vfs_inode.i_state &
5621 : (I_WILL_FREE | I_FREEING)));
5622 0 : rb_replace_node(parent, new, &root->inode_tree);
5623 0 : RB_CLEAR_NODE(parent);
5624 0 : spin_unlock(&root->inode_lock);
5625 0 : return;
5626 : }
5627 : }
5628 0 : rb_link_node(new, parent, p);
5629 0 : rb_insert_color(new, &root->inode_tree);
5630 0 : spin_unlock(&root->inode_lock);
5631 : }
5632 :
5633 0 : static void inode_tree_del(struct btrfs_inode *inode)
5634 : {
5635 0 : struct btrfs_root *root = inode->root;
5636 0 : int empty = 0;
5637 :
5638 0 : spin_lock(&root->inode_lock);
5639 0 : if (!RB_EMPTY_NODE(&inode->rb_node)) {
5640 0 : rb_erase(&inode->rb_node, &root->inode_tree);
5641 0 : RB_CLEAR_NODE(&inode->rb_node);
5642 0 : empty = RB_EMPTY_ROOT(&root->inode_tree);
5643 : }
5644 0 : spin_unlock(&root->inode_lock);
5645 :
5646 0 : if (empty && btrfs_root_refs(&root->root_item) == 0) {
5647 0 : spin_lock(&root->inode_lock);
5648 0 : empty = RB_EMPTY_ROOT(&root->inode_tree);
5649 0 : spin_unlock(&root->inode_lock);
5650 0 : if (empty)
5651 0 : btrfs_add_dead_root(root);
5652 : }
5653 0 : }
5654 :
5655 :
5656 0 : static int btrfs_init_locked_inode(struct inode *inode, void *p)
5657 : {
5658 0 : struct btrfs_iget_args *args = p;
5659 :
5660 0 : inode->i_ino = args->ino;
5661 0 : BTRFS_I(inode)->location.objectid = args->ino;
5662 0 : BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5663 0 : BTRFS_I(inode)->location.offset = 0;
5664 0 : BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5665 0 : BUG_ON(args->root && !BTRFS_I(inode)->root);
5666 :
5667 0 : if (args->root && args->root == args->root->fs_info->tree_root &&
5668 0 : args->ino != BTRFS_BTREE_INODE_OBJECTID)
5669 0 : set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5670 0 : &BTRFS_I(inode)->runtime_flags);
5671 0 : return 0;
5672 : }
5673 :
5674 0 : static int btrfs_find_actor(struct inode *inode, void *opaque)
5675 : {
5676 0 : struct btrfs_iget_args *args = opaque;
5677 :
5678 0 : return args->ino == BTRFS_I(inode)->location.objectid &&
5679 0 : args->root == BTRFS_I(inode)->root;
5680 : }
5681 :
5682 0 : static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5683 : struct btrfs_root *root)
5684 : {
5685 0 : struct inode *inode;
5686 0 : struct btrfs_iget_args args;
5687 0 : unsigned long hashval = btrfs_inode_hash(ino, root);
5688 :
5689 0 : args.ino = ino;
5690 0 : args.root = root;
5691 :
5692 0 : inode = iget5_locked(s, hashval, btrfs_find_actor,
5693 : btrfs_init_locked_inode,
5694 : (void *)&args);
5695 0 : return inode;
5696 : }
5697 :
5698 : /*
5699 : * Get an inode object given its inode number and corresponding root.
5700 : * Path can be preallocated to prevent recursing back to iget through
5701 : * allocator. NULL is also valid but may require an additional allocation
5702 : * later.
5703 : */
5704 0 : struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5705 : struct btrfs_root *root, struct btrfs_path *path)
5706 : {
5707 0 : struct inode *inode;
5708 :
5709 0 : inode = btrfs_iget_locked(s, ino, root);
5710 0 : if (!inode)
5711 : return ERR_PTR(-ENOMEM);
5712 :
5713 0 : if (inode->i_state & I_NEW) {
5714 0 : int ret;
5715 :
5716 0 : ret = btrfs_read_locked_inode(inode, path);
5717 0 : if (!ret) {
5718 0 : inode_tree_add(BTRFS_I(inode));
5719 0 : unlock_new_inode(inode);
5720 : } else {
5721 0 : iget_failed(inode);
5722 : /*
5723 : * ret > 0 can come from btrfs_search_slot called by
5724 : * btrfs_read_locked_inode, this means the inode item
5725 : * was not found.
5726 : */
5727 0 : if (ret > 0)
5728 0 : ret = -ENOENT;
5729 0 : inode = ERR_PTR(ret);
5730 : }
5731 : }
5732 :
5733 : return inode;
5734 : }
5735 :
5736 0 : struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5737 : {
5738 0 : return btrfs_iget_path(s, ino, root, NULL);
5739 : }
5740 :
5741 0 : static struct inode *new_simple_dir(struct super_block *s,
5742 : struct btrfs_key *key,
5743 : struct btrfs_root *root)
5744 : {
5745 0 : struct inode *inode = new_inode(s);
5746 :
5747 0 : if (!inode)
5748 : return ERR_PTR(-ENOMEM);
5749 :
5750 0 : BTRFS_I(inode)->root = btrfs_grab_root(root);
5751 0 : memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5752 0 : set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5753 :
5754 0 : inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5755 : /*
5756 : * We only need lookup, the rest is read-only and there's no inode
5757 : * associated with the dentry
5758 : */
5759 0 : inode->i_op = &simple_dir_inode_operations;
5760 0 : inode->i_opflags &= ~IOP_XATTR;
5761 0 : inode->i_fop = &simple_dir_operations;
5762 0 : inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5763 0 : inode->i_mtime = current_time(inode);
5764 0 : inode->i_atime = inode->i_mtime;
5765 0 : inode->i_ctime = inode->i_mtime;
5766 0 : BTRFS_I(inode)->i_otime = inode->i_mtime;
5767 :
5768 0 : return inode;
5769 : }
5770 :
5771 : static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5772 : static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5773 : static_assert(BTRFS_FT_DIR == FT_DIR);
5774 : static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5775 : static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5776 : static_assert(BTRFS_FT_FIFO == FT_FIFO);
5777 : static_assert(BTRFS_FT_SOCK == FT_SOCK);
5778 : static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5779 :
5780 : static inline u8 btrfs_inode_type(struct inode *inode)
5781 : {
5782 0 : return fs_umode_to_ftype(inode->i_mode);
5783 : }
5784 :
5785 0 : struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5786 : {
5787 0 : struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5788 0 : struct inode *inode;
5789 0 : struct btrfs_root *root = BTRFS_I(dir)->root;
5790 0 : struct btrfs_root *sub_root = root;
5791 0 : struct btrfs_key location;
5792 0 : u8 di_type = 0;
5793 0 : int ret = 0;
5794 :
5795 0 : if (dentry->d_name.len > BTRFS_NAME_LEN)
5796 : return ERR_PTR(-ENAMETOOLONG);
5797 :
5798 0 : ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5799 0 : if (ret < 0)
5800 0 : return ERR_PTR(ret);
5801 :
5802 0 : if (location.type == BTRFS_INODE_ITEM_KEY) {
5803 0 : inode = btrfs_iget(dir->i_sb, location.objectid, root);
5804 0 : if (IS_ERR(inode))
5805 : return inode;
5806 :
5807 : /* Do extra check against inode mode with di_type */
5808 0 : if (btrfs_inode_type(inode) != di_type) {
5809 0 : btrfs_crit(fs_info,
5810 : "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5811 : inode->i_mode, btrfs_inode_type(inode),
5812 : di_type);
5813 0 : iput(inode);
5814 0 : return ERR_PTR(-EUCLEAN);
5815 : }
5816 : return inode;
5817 : }
5818 :
5819 0 : ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5820 : &location, &sub_root);
5821 0 : if (ret < 0) {
5822 0 : if (ret != -ENOENT)
5823 0 : inode = ERR_PTR(ret);
5824 : else
5825 0 : inode = new_simple_dir(dir->i_sb, &location, root);
5826 : } else {
5827 0 : inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5828 0 : btrfs_put_root(sub_root);
5829 :
5830 0 : if (IS_ERR(inode))
5831 : return inode;
5832 :
5833 0 : down_read(&fs_info->cleanup_work_sem);
5834 0 : if (!sb_rdonly(inode->i_sb))
5835 0 : ret = btrfs_orphan_cleanup(sub_root);
5836 0 : up_read(&fs_info->cleanup_work_sem);
5837 0 : if (ret) {
5838 0 : iput(inode);
5839 0 : inode = ERR_PTR(ret);
5840 : }
5841 : }
5842 :
5843 : return inode;
5844 : }
5845 :
5846 0 : static int btrfs_dentry_delete(const struct dentry *dentry)
5847 : {
5848 0 : struct btrfs_root *root;
5849 0 : struct inode *inode = d_inode(dentry);
5850 :
5851 0 : if (!inode && !IS_ROOT(dentry))
5852 0 : inode = d_inode(dentry->d_parent);
5853 :
5854 0 : if (inode) {
5855 0 : root = BTRFS_I(inode)->root;
5856 0 : if (btrfs_root_refs(&root->root_item) == 0)
5857 : return 1;
5858 :
5859 0 : if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5860 0 : return 1;
5861 : }
5862 : return 0;
5863 : }
5864 :
5865 0 : static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5866 : unsigned int flags)
5867 : {
5868 0 : struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5869 :
5870 0 : if (inode == ERR_PTR(-ENOENT))
5871 0 : inode = NULL;
5872 0 : return d_splice_alias(inode, dentry);
5873 : }
5874 :
5875 : /*
5876 : * All this infrastructure exists because dir_emit can fault, and we are holding
5877 : * the tree lock when doing readdir. For now just allocate a buffer and copy
5878 : * our information into that, and then dir_emit from the buffer. This is
5879 : * similar to what NFS does, only we don't keep the buffer around in pagecache
5880 : * because I'm afraid I'll mess that up. Long term we need to make filldir do
5881 : * copy_to_user_inatomic so we don't have to worry about page faulting under the
5882 : * tree lock.
5883 : */
5884 0 : static int btrfs_opendir(struct inode *inode, struct file *file)
5885 : {
5886 0 : struct btrfs_file_private *private;
5887 :
5888 0 : private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5889 0 : if (!private)
5890 : return -ENOMEM;
5891 0 : private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5892 0 : if (!private->filldir_buf) {
5893 0 : kfree(private);
5894 0 : return -ENOMEM;
5895 : }
5896 0 : file->private_data = private;
5897 0 : return 0;
5898 : }
5899 :
5900 : struct dir_entry {
5901 : u64 ino;
5902 : u64 offset;
5903 : unsigned type;
5904 : int name_len;
5905 : };
5906 :
5907 0 : static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5908 : {
5909 0 : while (entries--) {
5910 0 : struct dir_entry *entry = addr;
5911 0 : char *name = (char *)(entry + 1);
5912 :
5913 0 : ctx->pos = get_unaligned(&entry->offset);
5914 0 : if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5915 0 : get_unaligned(&entry->ino),
5916 0 : get_unaligned(&entry->type)))
5917 : return 1;
5918 0 : addr += sizeof(struct dir_entry) +
5919 0 : get_unaligned(&entry->name_len);
5920 0 : ctx->pos++;
5921 : }
5922 : return 0;
5923 : }
5924 :
5925 0 : static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5926 : {
5927 0 : struct inode *inode = file_inode(file);
5928 0 : struct btrfs_root *root = BTRFS_I(inode)->root;
5929 0 : struct btrfs_file_private *private = file->private_data;
5930 0 : struct btrfs_dir_item *di;
5931 0 : struct btrfs_key key;
5932 0 : struct btrfs_key found_key;
5933 0 : struct btrfs_path *path;
5934 0 : void *addr;
5935 0 : struct list_head ins_list;
5936 0 : struct list_head del_list;
5937 0 : int ret;
5938 0 : char *name_ptr;
5939 0 : int name_len;
5940 0 : int entries = 0;
5941 0 : int total_len = 0;
5942 0 : bool put = false;
5943 0 : struct btrfs_key location;
5944 :
5945 0 : if (!dir_emit_dots(file, ctx))
5946 : return 0;
5947 :
5948 0 : path = btrfs_alloc_path();
5949 0 : if (!path)
5950 : return -ENOMEM;
5951 :
5952 0 : addr = private->filldir_buf;
5953 0 : path->reada = READA_FORWARD;
5954 :
5955 0 : INIT_LIST_HEAD(&ins_list);
5956 0 : INIT_LIST_HEAD(&del_list);
5957 0 : put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5958 :
5959 0 : again:
5960 0 : key.type = BTRFS_DIR_INDEX_KEY;
5961 0 : key.offset = ctx->pos;
5962 0 : key.objectid = btrfs_ino(BTRFS_I(inode));
5963 :
5964 0 : btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5965 0 : struct dir_entry *entry;
5966 0 : struct extent_buffer *leaf = path->nodes[0];
5967 0 : u8 ftype;
5968 :
5969 0 : if (found_key.objectid != key.objectid)
5970 : break;
5971 0 : if (found_key.type != BTRFS_DIR_INDEX_KEY)
5972 : break;
5973 0 : if (found_key.offset < ctx->pos)
5974 0 : continue;
5975 0 : if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5976 0 : continue;
5977 0 : di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5978 0 : name_len = btrfs_dir_name_len(leaf, di);
5979 0 : if ((total_len + sizeof(struct dir_entry) + name_len) >=
5980 : PAGE_SIZE) {
5981 0 : btrfs_release_path(path);
5982 0 : ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5983 0 : if (ret)
5984 0 : goto nopos;
5985 0 : addr = private->filldir_buf;
5986 0 : entries = 0;
5987 0 : total_len = 0;
5988 0 : goto again;
5989 : }
5990 :
5991 0 : ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
5992 0 : entry = addr;
5993 0 : name_ptr = (char *)(entry + 1);
5994 0 : read_extent_buffer(leaf, name_ptr,
5995 0 : (unsigned long)(di + 1), name_len);
5996 0 : put_unaligned(name_len, &entry->name_len);
5997 0 : put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
5998 0 : btrfs_dir_item_key_to_cpu(leaf, di, &location);
5999 0 : put_unaligned(location.objectid, &entry->ino);
6000 0 : put_unaligned(found_key.offset, &entry->offset);
6001 0 : entries++;
6002 0 : addr += sizeof(struct dir_entry) + name_len;
6003 0 : total_len += sizeof(struct dir_entry) + name_len;
6004 : }
6005 : /* Catch error encountered during iteration */
6006 0 : if (ret < 0)
6007 0 : goto err;
6008 :
6009 0 : btrfs_release_path(path);
6010 :
6011 0 : ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6012 0 : if (ret)
6013 0 : goto nopos;
6014 :
6015 0 : ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6016 0 : if (ret)
6017 0 : goto nopos;
6018 :
6019 : /*
6020 : * Stop new entries from being returned after we return the last
6021 : * entry.
6022 : *
6023 : * New directory entries are assigned a strictly increasing
6024 : * offset. This means that new entries created during readdir
6025 : * are *guaranteed* to be seen in the future by that readdir.
6026 : * This has broken buggy programs which operate on names as
6027 : * they're returned by readdir. Until we re-use freed offsets
6028 : * we have this hack to stop new entries from being returned
6029 : * under the assumption that they'll never reach this huge
6030 : * offset.
6031 : *
6032 : * This is being careful not to overflow 32bit loff_t unless the
6033 : * last entry requires it because doing so has broken 32bit apps
6034 : * in the past.
6035 : */
6036 0 : if (ctx->pos >= INT_MAX)
6037 0 : ctx->pos = LLONG_MAX;
6038 : else
6039 0 : ctx->pos = INT_MAX;
6040 : nopos:
6041 : ret = 0;
6042 0 : err:
6043 0 : if (put)
6044 0 : btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6045 0 : btrfs_free_path(path);
6046 0 : return ret;
6047 : }
6048 :
6049 : /*
6050 : * This is somewhat expensive, updating the tree every time the
6051 : * inode changes. But, it is most likely to find the inode in cache.
6052 : * FIXME, needs more benchmarking...there are no reasons other than performance
6053 : * to keep or drop this code.
6054 : */
6055 0 : static int btrfs_dirty_inode(struct btrfs_inode *inode)
6056 : {
6057 0 : struct btrfs_root *root = inode->root;
6058 0 : struct btrfs_fs_info *fs_info = root->fs_info;
6059 0 : struct btrfs_trans_handle *trans;
6060 0 : int ret;
6061 :
6062 0 : if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6063 : return 0;
6064 :
6065 0 : trans = btrfs_join_transaction(root);
6066 0 : if (IS_ERR(trans))
6067 0 : return PTR_ERR(trans);
6068 :
6069 0 : ret = btrfs_update_inode(trans, root, inode);
6070 0 : if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6071 : /* whoops, lets try again with the full transaction */
6072 0 : btrfs_end_transaction(trans);
6073 0 : trans = btrfs_start_transaction(root, 1);
6074 0 : if (IS_ERR(trans))
6075 0 : return PTR_ERR(trans);
6076 :
6077 0 : ret = btrfs_update_inode(trans, root, inode);
6078 : }
6079 0 : btrfs_end_transaction(trans);
6080 0 : if (inode->delayed_node)
6081 0 : btrfs_balance_delayed_items(fs_info);
6082 :
6083 : return ret;
6084 : }
6085 :
6086 : /*
6087 : * This is a copy of file_update_time. We need this so we can return error on
6088 : * ENOSPC for updating the inode in the case of file write and mmap writes.
6089 : */
6090 0 : static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6091 : int flags)
6092 : {
6093 0 : struct btrfs_root *root = BTRFS_I(inode)->root;
6094 0 : bool dirty = flags & ~S_VERSION;
6095 :
6096 0 : if (btrfs_root_readonly(root))
6097 : return -EROFS;
6098 :
6099 0 : if (flags & S_VERSION)
6100 0 : dirty |= inode_maybe_inc_iversion(inode, dirty);
6101 0 : if (flags & S_CTIME)
6102 0 : inode->i_ctime = *now;
6103 0 : if (flags & S_MTIME)
6104 0 : inode->i_mtime = *now;
6105 0 : if (flags & S_ATIME)
6106 0 : inode->i_atime = *now;
6107 0 : return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6108 : }
6109 :
6110 : /*
6111 : * find the highest existing sequence number in a directory
6112 : * and then set the in-memory index_cnt variable to reflect
6113 : * free sequence numbers
6114 : */
6115 0 : static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6116 : {
6117 0 : struct btrfs_root *root = inode->root;
6118 0 : struct btrfs_key key, found_key;
6119 0 : struct btrfs_path *path;
6120 0 : struct extent_buffer *leaf;
6121 0 : int ret;
6122 :
6123 0 : key.objectid = btrfs_ino(inode);
6124 0 : key.type = BTRFS_DIR_INDEX_KEY;
6125 0 : key.offset = (u64)-1;
6126 :
6127 0 : path = btrfs_alloc_path();
6128 0 : if (!path)
6129 : return -ENOMEM;
6130 :
6131 0 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6132 0 : if (ret < 0)
6133 0 : goto out;
6134 : /* FIXME: we should be able to handle this */
6135 0 : if (ret == 0)
6136 0 : goto out;
6137 0 : ret = 0;
6138 :
6139 0 : if (path->slots[0] == 0) {
6140 0 : inode->index_cnt = BTRFS_DIR_START_INDEX;
6141 0 : goto out;
6142 : }
6143 :
6144 0 : path->slots[0]--;
6145 :
6146 0 : leaf = path->nodes[0];
6147 0 : btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6148 :
6149 0 : if (found_key.objectid != btrfs_ino(inode) ||
6150 0 : found_key.type != BTRFS_DIR_INDEX_KEY) {
6151 0 : inode->index_cnt = BTRFS_DIR_START_INDEX;
6152 0 : goto out;
6153 : }
6154 :
6155 0 : inode->index_cnt = found_key.offset + 1;
6156 0 : out:
6157 0 : btrfs_free_path(path);
6158 0 : return ret;
6159 : }
6160 :
6161 : /*
6162 : * helper to find a free sequence number in a given directory. This current
6163 : * code is very simple, later versions will do smarter things in the btree
6164 : */
6165 0 : int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6166 : {
6167 0 : int ret = 0;
6168 :
6169 0 : if (dir->index_cnt == (u64)-1) {
6170 0 : ret = btrfs_inode_delayed_dir_index_count(dir);
6171 0 : if (ret) {
6172 0 : ret = btrfs_set_inode_index_count(dir);
6173 0 : if (ret)
6174 : return ret;
6175 : }
6176 : }
6177 :
6178 0 : *index = dir->index_cnt;
6179 0 : dir->index_cnt++;
6180 :
6181 0 : return ret;
6182 : }
6183 :
6184 0 : static int btrfs_insert_inode_locked(struct inode *inode)
6185 : {
6186 0 : struct btrfs_iget_args args;
6187 :
6188 0 : args.ino = BTRFS_I(inode)->location.objectid;
6189 0 : args.root = BTRFS_I(inode)->root;
6190 :
6191 0 : return insert_inode_locked4(inode,
6192 0 : btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6193 : btrfs_find_actor, &args);
6194 : }
6195 :
6196 0 : int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6197 : unsigned int *trans_num_items)
6198 : {
6199 0 : struct inode *dir = args->dir;
6200 0 : struct inode *inode = args->inode;
6201 0 : int ret;
6202 :
6203 0 : if (!args->orphan) {
6204 0 : ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6205 : &args->fname);
6206 0 : if (ret)
6207 : return ret;
6208 : }
6209 :
6210 0 : ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6211 0 : if (ret) {
6212 : fscrypt_free_filename(&args->fname);
6213 : return ret;
6214 : }
6215 :
6216 : /* 1 to add inode item */
6217 0 : *trans_num_items = 1;
6218 : /* 1 to add compression property */
6219 0 : if (BTRFS_I(dir)->prop_compress)
6220 0 : (*trans_num_items)++;
6221 : /* 1 to add default ACL xattr */
6222 0 : if (args->default_acl)
6223 0 : (*trans_num_items)++;
6224 : /* 1 to add access ACL xattr */
6225 0 : if (args->acl)
6226 0 : (*trans_num_items)++;
6227 : #ifdef CONFIG_SECURITY
6228 : /* 1 to add LSM xattr */
6229 : if (dir->i_security)
6230 : (*trans_num_items)++;
6231 : #endif
6232 0 : if (args->orphan) {
6233 : /* 1 to add orphan item */
6234 0 : (*trans_num_items)++;
6235 : } else {
6236 : /*
6237 : * 1 to add dir item
6238 : * 1 to add dir index
6239 : * 1 to update parent inode item
6240 : *
6241 : * No need for 1 unit for the inode ref item because it is
6242 : * inserted in a batch together with the inode item at
6243 : * btrfs_create_new_inode().
6244 : */
6245 0 : *trans_num_items += 3;
6246 : }
6247 : return 0;
6248 : }
6249 :
6250 0 : void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6251 : {
6252 0 : posix_acl_release(args->acl);
6253 0 : posix_acl_release(args->default_acl);
6254 0 : fscrypt_free_filename(&args->fname);
6255 0 : }
6256 :
6257 : /*
6258 : * Inherit flags from the parent inode.
6259 : *
6260 : * Currently only the compression flags and the cow flags are inherited.
6261 : */
6262 0 : static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6263 : {
6264 0 : unsigned int flags;
6265 :
6266 0 : flags = dir->flags;
6267 :
6268 0 : if (flags & BTRFS_INODE_NOCOMPRESS) {
6269 0 : inode->flags &= ~BTRFS_INODE_COMPRESS;
6270 0 : inode->flags |= BTRFS_INODE_NOCOMPRESS;
6271 0 : } else if (flags & BTRFS_INODE_COMPRESS) {
6272 0 : inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6273 0 : inode->flags |= BTRFS_INODE_COMPRESS;
6274 : }
6275 :
6276 0 : if (flags & BTRFS_INODE_NODATACOW) {
6277 0 : inode->flags |= BTRFS_INODE_NODATACOW;
6278 0 : if (S_ISREG(inode->vfs_inode.i_mode))
6279 0 : inode->flags |= BTRFS_INODE_NODATASUM;
6280 : }
6281 :
6282 0 : btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6283 0 : }
6284 :
6285 0 : int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6286 : struct btrfs_new_inode_args *args)
6287 : {
6288 0 : struct inode *dir = args->dir;
6289 0 : struct inode *inode = args->inode;
6290 0 : const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6291 0 : struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6292 0 : struct btrfs_root *root;
6293 0 : struct btrfs_inode_item *inode_item;
6294 0 : struct btrfs_key *location;
6295 0 : struct btrfs_path *path;
6296 0 : u64 objectid;
6297 0 : struct btrfs_inode_ref *ref;
6298 0 : struct btrfs_key key[2];
6299 0 : u32 sizes[2];
6300 0 : struct btrfs_item_batch batch;
6301 0 : unsigned long ptr;
6302 0 : int ret;
6303 :
6304 0 : path = btrfs_alloc_path();
6305 0 : if (!path)
6306 : return -ENOMEM;
6307 :
6308 0 : if (!args->subvol)
6309 0 : BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6310 0 : root = BTRFS_I(inode)->root;
6311 :
6312 0 : ret = btrfs_get_free_objectid(root, &objectid);
6313 0 : if (ret)
6314 0 : goto out;
6315 0 : inode->i_ino = objectid;
6316 :
6317 0 : if (args->orphan) {
6318 : /*
6319 : * O_TMPFILE, set link count to 0, so that after this point, we
6320 : * fill in an inode item with the correct link count.
6321 : */
6322 0 : set_nlink(inode, 0);
6323 : } else {
6324 0 : trace_btrfs_inode_request(dir);
6325 :
6326 0 : ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6327 0 : if (ret)
6328 0 : goto out;
6329 : }
6330 : /* index_cnt is ignored for everything but a dir. */
6331 0 : BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6332 0 : BTRFS_I(inode)->generation = trans->transid;
6333 0 : inode->i_generation = BTRFS_I(inode)->generation;
6334 :
6335 : /*
6336 : * Subvolumes don't inherit flags from their parent directory.
6337 : * Originally this was probably by accident, but we probably can't
6338 : * change it now without compatibility issues.
6339 : */
6340 0 : if (!args->subvol)
6341 0 : btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6342 :
6343 0 : if (S_ISREG(inode->i_mode)) {
6344 0 : if (btrfs_test_opt(fs_info, NODATASUM))
6345 0 : BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6346 0 : if (btrfs_test_opt(fs_info, NODATACOW))
6347 0 : BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6348 : BTRFS_INODE_NODATASUM;
6349 : }
6350 :
6351 0 : location = &BTRFS_I(inode)->location;
6352 0 : location->objectid = objectid;
6353 0 : location->offset = 0;
6354 0 : location->type = BTRFS_INODE_ITEM_KEY;
6355 :
6356 0 : ret = btrfs_insert_inode_locked(inode);
6357 0 : if (ret < 0) {
6358 0 : if (!args->orphan)
6359 0 : BTRFS_I(dir)->index_cnt--;
6360 0 : goto out;
6361 : }
6362 :
6363 : /*
6364 : * We could have gotten an inode number from somebody who was fsynced
6365 : * and then removed in this same transaction, so let's just set full
6366 : * sync since it will be a full sync anyway and this will blow away the
6367 : * old info in the log.
6368 : */
6369 0 : btrfs_set_inode_full_sync(BTRFS_I(inode));
6370 :
6371 0 : key[0].objectid = objectid;
6372 0 : key[0].type = BTRFS_INODE_ITEM_KEY;
6373 0 : key[0].offset = 0;
6374 :
6375 0 : sizes[0] = sizeof(struct btrfs_inode_item);
6376 :
6377 0 : if (!args->orphan) {
6378 : /*
6379 : * Start new inodes with an inode_ref. This is slightly more
6380 : * efficient for small numbers of hard links since they will
6381 : * be packed into one item. Extended refs will kick in if we
6382 : * add more hard links than can fit in the ref item.
6383 : */
6384 0 : key[1].objectid = objectid;
6385 0 : key[1].type = BTRFS_INODE_REF_KEY;
6386 0 : if (args->subvol) {
6387 0 : key[1].offset = objectid;
6388 0 : sizes[1] = 2 + sizeof(*ref);
6389 : } else {
6390 0 : key[1].offset = btrfs_ino(BTRFS_I(dir));
6391 0 : sizes[1] = name->len + sizeof(*ref);
6392 : }
6393 : }
6394 :
6395 0 : batch.keys = &key[0];
6396 0 : batch.data_sizes = &sizes[0];
6397 0 : batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6398 0 : batch.nr = args->orphan ? 1 : 2;
6399 0 : ret = btrfs_insert_empty_items(trans, root, path, &batch);
6400 0 : if (ret != 0) {
6401 0 : btrfs_abort_transaction(trans, ret);
6402 0 : goto discard;
6403 : }
6404 :
6405 0 : inode->i_mtime = current_time(inode);
6406 0 : inode->i_atime = inode->i_mtime;
6407 0 : inode->i_ctime = inode->i_mtime;
6408 0 : BTRFS_I(inode)->i_otime = inode->i_mtime;
6409 :
6410 : /*
6411 : * We're going to fill the inode item now, so at this point the inode
6412 : * must be fully initialized.
6413 : */
6414 :
6415 0 : inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6416 : struct btrfs_inode_item);
6417 0 : memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6418 : sizeof(*inode_item));
6419 0 : fill_inode_item(trans, path->nodes[0], inode_item, inode);
6420 :
6421 0 : if (!args->orphan) {
6422 0 : ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6423 : struct btrfs_inode_ref);
6424 0 : ptr = (unsigned long)(ref + 1);
6425 0 : if (args->subvol) {
6426 0 : btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6427 0 : btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6428 0 : write_extent_buffer(path->nodes[0], "..", ptr, 2);
6429 : } else {
6430 0 : btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6431 0 : name->len);
6432 0 : btrfs_set_inode_ref_index(path->nodes[0], ref,
6433 : BTRFS_I(inode)->dir_index);
6434 0 : write_extent_buffer(path->nodes[0], name->name, ptr,
6435 0 : name->len);
6436 : }
6437 : }
6438 :
6439 0 : btrfs_mark_buffer_dirty(path->nodes[0]);
6440 : /*
6441 : * We don't need the path anymore, plus inheriting properties, adding
6442 : * ACLs, security xattrs, orphan item or adding the link, will result in
6443 : * allocating yet another path. So just free our path.
6444 : */
6445 0 : btrfs_free_path(path);
6446 0 : path = NULL;
6447 :
6448 0 : if (args->subvol) {
6449 0 : struct inode *parent;
6450 :
6451 : /*
6452 : * Subvolumes inherit properties from their parent subvolume,
6453 : * not the directory they were created in.
6454 : */
6455 0 : parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6456 : BTRFS_I(dir)->root);
6457 0 : if (IS_ERR(parent)) {
6458 0 : ret = PTR_ERR(parent);
6459 : } else {
6460 0 : ret = btrfs_inode_inherit_props(trans, inode, parent);
6461 0 : iput(parent);
6462 : }
6463 : } else {
6464 0 : ret = btrfs_inode_inherit_props(trans, inode, dir);
6465 : }
6466 0 : if (ret) {
6467 0 : btrfs_err(fs_info,
6468 : "error inheriting props for ino %llu (root %llu): %d",
6469 : btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6470 : ret);
6471 : }
6472 :
6473 : /*
6474 : * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6475 : * probably a bug.
6476 : */
6477 0 : if (!args->subvol) {
6478 0 : ret = btrfs_init_inode_security(trans, args);
6479 0 : if (ret) {
6480 0 : btrfs_abort_transaction(trans, ret);
6481 0 : goto discard;
6482 : }
6483 : }
6484 :
6485 0 : inode_tree_add(BTRFS_I(inode));
6486 :
6487 0 : trace_btrfs_inode_new(inode);
6488 0 : btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6489 :
6490 0 : btrfs_update_root_times(trans, root);
6491 :
6492 0 : if (args->orphan) {
6493 0 : ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6494 : } else {
6495 0 : ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6496 : 0, BTRFS_I(inode)->dir_index);
6497 : }
6498 0 : if (ret) {
6499 0 : btrfs_abort_transaction(trans, ret);
6500 0 : goto discard;
6501 : }
6502 :
6503 : return 0;
6504 :
6505 0 : discard:
6506 : /*
6507 : * discard_new_inode() calls iput(), but the caller owns the reference
6508 : * to the inode.
6509 : */
6510 0 : ihold(inode);
6511 0 : discard_new_inode(inode);
6512 0 : out:
6513 0 : btrfs_free_path(path);
6514 0 : return ret;
6515 : }
6516 :
6517 : /*
6518 : * utility function to add 'inode' into 'parent_inode' with
6519 : * a give name and a given sequence number.
6520 : * if 'add_backref' is true, also insert a backref from the
6521 : * inode to the parent directory.
6522 : */
6523 0 : int btrfs_add_link(struct btrfs_trans_handle *trans,
6524 : struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6525 : const struct fscrypt_str *name, int add_backref, u64 index)
6526 : {
6527 0 : int ret = 0;
6528 0 : struct btrfs_key key;
6529 0 : struct btrfs_root *root = parent_inode->root;
6530 0 : u64 ino = btrfs_ino(inode);
6531 0 : u64 parent_ino = btrfs_ino(parent_inode);
6532 :
6533 0 : if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6534 0 : memcpy(&key, &inode->root->root_key, sizeof(key));
6535 : } else {
6536 0 : key.objectid = ino;
6537 0 : key.type = BTRFS_INODE_ITEM_KEY;
6538 0 : key.offset = 0;
6539 : }
6540 :
6541 0 : if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6542 0 : ret = btrfs_add_root_ref(trans, key.objectid,
6543 : root->root_key.objectid, parent_ino,
6544 : index, name);
6545 0 : } else if (add_backref) {
6546 0 : ret = btrfs_insert_inode_ref(trans, root, name,
6547 : ino, parent_ino, index);
6548 : }
6549 :
6550 : /* Nothing to clean up yet */
6551 0 : if (ret)
6552 : return ret;
6553 :
6554 0 : ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6555 : btrfs_inode_type(&inode->vfs_inode), index);
6556 0 : if (ret == -EEXIST || ret == -EOVERFLOW)
6557 0 : goto fail_dir_item;
6558 0 : else if (ret) {
6559 0 : btrfs_abort_transaction(trans, ret);
6560 0 : return ret;
6561 : }
6562 :
6563 0 : btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6564 0 : name->len * 2);
6565 0 : inode_inc_iversion(&parent_inode->vfs_inode);
6566 : /*
6567 : * If we are replaying a log tree, we do not want to update the mtime
6568 : * and ctime of the parent directory with the current time, since the
6569 : * log replay procedure is responsible for setting them to their correct
6570 : * values (the ones it had when the fsync was done).
6571 : */
6572 0 : if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6573 0 : struct timespec64 now = current_time(&parent_inode->vfs_inode);
6574 :
6575 0 : parent_inode->vfs_inode.i_mtime = now;
6576 0 : parent_inode->vfs_inode.i_ctime = now;
6577 : }
6578 0 : ret = btrfs_update_inode(trans, root, parent_inode);
6579 0 : if (ret)
6580 0 : btrfs_abort_transaction(trans, ret);
6581 : return ret;
6582 :
6583 : fail_dir_item:
6584 0 : if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6585 0 : u64 local_index;
6586 0 : int err;
6587 0 : err = btrfs_del_root_ref(trans, key.objectid,
6588 : root->root_key.objectid, parent_ino,
6589 : &local_index, name);
6590 0 : if (err)
6591 0 : btrfs_abort_transaction(trans, err);
6592 0 : } else if (add_backref) {
6593 0 : u64 local_index;
6594 0 : int err;
6595 :
6596 0 : err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6597 : &local_index);
6598 0 : if (err)
6599 0 : btrfs_abort_transaction(trans, err);
6600 : }
6601 :
6602 : /* Return the original error code */
6603 : return ret;
6604 : }
6605 :
6606 0 : static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6607 : struct inode *inode)
6608 : {
6609 0 : struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6610 0 : struct btrfs_root *root = BTRFS_I(dir)->root;
6611 0 : struct btrfs_new_inode_args new_inode_args = {
6612 : .dir = dir,
6613 : .dentry = dentry,
6614 : .inode = inode,
6615 : };
6616 0 : unsigned int trans_num_items;
6617 0 : struct btrfs_trans_handle *trans;
6618 0 : int err;
6619 :
6620 0 : err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6621 0 : if (err)
6622 0 : goto out_inode;
6623 :
6624 0 : trans = btrfs_start_transaction(root, trans_num_items);
6625 0 : if (IS_ERR(trans)) {
6626 0 : err = PTR_ERR(trans);
6627 0 : goto out_new_inode_args;
6628 : }
6629 :
6630 0 : err = btrfs_create_new_inode(trans, &new_inode_args);
6631 0 : if (!err)
6632 0 : d_instantiate_new(dentry, inode);
6633 :
6634 0 : btrfs_end_transaction(trans);
6635 0 : btrfs_btree_balance_dirty(fs_info);
6636 0 : out_new_inode_args:
6637 0 : btrfs_new_inode_args_destroy(&new_inode_args);
6638 0 : out_inode:
6639 0 : if (err)
6640 0 : iput(inode);
6641 0 : return err;
6642 : }
6643 :
6644 0 : static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6645 : struct dentry *dentry, umode_t mode, dev_t rdev)
6646 : {
6647 0 : struct inode *inode;
6648 :
6649 0 : inode = new_inode(dir->i_sb);
6650 0 : if (!inode)
6651 : return -ENOMEM;
6652 0 : inode_init_owner(idmap, inode, dir, mode);
6653 0 : inode->i_op = &btrfs_special_inode_operations;
6654 0 : init_special_inode(inode, inode->i_mode, rdev);
6655 0 : return btrfs_create_common(dir, dentry, inode);
6656 : }
6657 :
6658 0 : static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6659 : struct dentry *dentry, umode_t mode, bool excl)
6660 : {
6661 0 : struct inode *inode;
6662 :
6663 0 : inode = new_inode(dir->i_sb);
6664 0 : if (!inode)
6665 : return -ENOMEM;
6666 0 : inode_init_owner(idmap, inode, dir, mode);
6667 0 : inode->i_fop = &btrfs_file_operations;
6668 0 : inode->i_op = &btrfs_file_inode_operations;
6669 0 : inode->i_mapping->a_ops = &btrfs_aops;
6670 0 : return btrfs_create_common(dir, dentry, inode);
6671 : }
6672 :
6673 0 : static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6674 : struct dentry *dentry)
6675 : {
6676 0 : struct btrfs_trans_handle *trans = NULL;
6677 0 : struct btrfs_root *root = BTRFS_I(dir)->root;
6678 0 : struct inode *inode = d_inode(old_dentry);
6679 0 : struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6680 0 : struct fscrypt_name fname;
6681 0 : u64 index;
6682 0 : int err;
6683 0 : int drop_inode = 0;
6684 :
6685 : /* do not allow sys_link's with other subvols of the same device */
6686 0 : if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6687 : return -EXDEV;
6688 :
6689 0 : if (inode->i_nlink >= BTRFS_LINK_MAX)
6690 : return -EMLINK;
6691 :
6692 0 : err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6693 0 : if (err)
6694 0 : goto fail;
6695 :
6696 0 : err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6697 0 : if (err)
6698 0 : goto fail;
6699 :
6700 : /*
6701 : * 2 items for inode and inode ref
6702 : * 2 items for dir items
6703 : * 1 item for parent inode
6704 : * 1 item for orphan item deletion if O_TMPFILE
6705 : */
6706 0 : trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6707 0 : if (IS_ERR(trans)) {
6708 0 : err = PTR_ERR(trans);
6709 0 : trans = NULL;
6710 0 : goto fail;
6711 : }
6712 :
6713 : /* There are several dir indexes for this inode, clear the cache. */
6714 0 : BTRFS_I(inode)->dir_index = 0ULL;
6715 0 : inc_nlink(inode);
6716 0 : inode_inc_iversion(inode);
6717 0 : inode->i_ctime = current_time(inode);
6718 0 : ihold(inode);
6719 0 : set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6720 :
6721 0 : err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6722 : &fname.disk_name, 1, index);
6723 :
6724 0 : if (err) {
6725 : drop_inode = 1;
6726 : } else {
6727 0 : struct dentry *parent = dentry->d_parent;
6728 :
6729 0 : err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6730 0 : if (err)
6731 0 : goto fail;
6732 0 : if (inode->i_nlink == 1) {
6733 : /*
6734 : * If new hard link count is 1, it's a file created
6735 : * with open(2) O_TMPFILE flag.
6736 : */
6737 0 : err = btrfs_orphan_del(trans, BTRFS_I(inode));
6738 0 : if (err)
6739 0 : goto fail;
6740 : }
6741 0 : d_instantiate(dentry, inode);
6742 0 : btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6743 : }
6744 :
6745 0 : fail:
6746 0 : fscrypt_free_filename(&fname);
6747 0 : if (trans)
6748 0 : btrfs_end_transaction(trans);
6749 0 : if (drop_inode) {
6750 0 : inode_dec_link_count(inode);
6751 0 : iput(inode);
6752 : }
6753 0 : btrfs_btree_balance_dirty(fs_info);
6754 0 : return err;
6755 : }
6756 :
6757 0 : static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6758 : struct dentry *dentry, umode_t mode)
6759 : {
6760 0 : struct inode *inode;
6761 :
6762 0 : inode = new_inode(dir->i_sb);
6763 0 : if (!inode)
6764 : return -ENOMEM;
6765 0 : inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6766 0 : inode->i_op = &btrfs_dir_inode_operations;
6767 0 : inode->i_fop = &btrfs_dir_file_operations;
6768 0 : return btrfs_create_common(dir, dentry, inode);
6769 : }
6770 :
6771 0 : static noinline int uncompress_inline(struct btrfs_path *path,
6772 : struct page *page,
6773 : struct btrfs_file_extent_item *item)
6774 : {
6775 0 : int ret;
6776 0 : struct extent_buffer *leaf = path->nodes[0];
6777 0 : char *tmp;
6778 0 : size_t max_size;
6779 0 : unsigned long inline_size;
6780 0 : unsigned long ptr;
6781 0 : int compress_type;
6782 :
6783 0 : compress_type = btrfs_file_extent_compression(leaf, item);
6784 0 : max_size = btrfs_file_extent_ram_bytes(leaf, item);
6785 0 : inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6786 0 : tmp = kmalloc(inline_size, GFP_NOFS);
6787 0 : if (!tmp)
6788 : return -ENOMEM;
6789 0 : ptr = btrfs_file_extent_inline_start(item);
6790 :
6791 0 : read_extent_buffer(leaf, tmp, ptr, inline_size);
6792 :
6793 0 : max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6794 0 : ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6795 :
6796 : /*
6797 : * decompression code contains a memset to fill in any space between the end
6798 : * of the uncompressed data and the end of max_size in case the decompressed
6799 : * data ends up shorter than ram_bytes. That doesn't cover the hole between
6800 : * the end of an inline extent and the beginning of the next block, so we
6801 : * cover that region here.
6802 : */
6803 :
6804 0 : if (max_size < PAGE_SIZE)
6805 0 : memzero_page(page, max_size, PAGE_SIZE - max_size);
6806 0 : kfree(tmp);
6807 0 : return ret;
6808 : }
6809 :
6810 0 : static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6811 : struct page *page)
6812 : {
6813 0 : struct btrfs_file_extent_item *fi;
6814 0 : void *kaddr;
6815 0 : size_t copy_size;
6816 :
6817 0 : if (!page || PageUptodate(page))
6818 0 : return 0;
6819 :
6820 0 : ASSERT(page_offset(page) == 0);
6821 :
6822 0 : fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6823 : struct btrfs_file_extent_item);
6824 0 : if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6825 0 : return uncompress_inline(path, page, fi);
6826 :
6827 0 : copy_size = min_t(u64, PAGE_SIZE,
6828 : btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6829 0 : kaddr = kmap_local_page(page);
6830 0 : read_extent_buffer(path->nodes[0], kaddr,
6831 : btrfs_file_extent_inline_start(fi), copy_size);
6832 0 : kunmap_local(kaddr);
6833 0 : if (copy_size < PAGE_SIZE)
6834 0 : memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6835 : return 0;
6836 : }
6837 :
6838 : /*
6839 : * Lookup the first extent overlapping a range in a file.
6840 : *
6841 : * @inode: file to search in
6842 : * @page: page to read extent data into if the extent is inline
6843 : * @pg_offset: offset into @page to copy to
6844 : * @start: file offset
6845 : * @len: length of range starting at @start
6846 : *
6847 : * Return the first &struct extent_map which overlaps the given range, reading
6848 : * it from the B-tree and caching it if necessary. Note that there may be more
6849 : * extents which overlap the given range after the returned extent_map.
6850 : *
6851 : * If @page is not NULL and the extent is inline, this also reads the extent
6852 : * data directly into the page and marks the extent up to date in the io_tree.
6853 : *
6854 : * Return: ERR_PTR on error, non-NULL extent_map on success.
6855 : */
6856 0 : struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6857 : struct page *page, size_t pg_offset,
6858 : u64 start, u64 len)
6859 : {
6860 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
6861 0 : int ret = 0;
6862 0 : u64 extent_start = 0;
6863 0 : u64 extent_end = 0;
6864 0 : u64 objectid = btrfs_ino(inode);
6865 0 : int extent_type = -1;
6866 0 : struct btrfs_path *path = NULL;
6867 0 : struct btrfs_root *root = inode->root;
6868 0 : struct btrfs_file_extent_item *item;
6869 0 : struct extent_buffer *leaf;
6870 0 : struct btrfs_key found_key;
6871 0 : struct extent_map *em = NULL;
6872 0 : struct extent_map_tree *em_tree = &inode->extent_tree;
6873 :
6874 0 : read_lock(&em_tree->lock);
6875 0 : em = lookup_extent_mapping(em_tree, start, len);
6876 0 : read_unlock(&em_tree->lock);
6877 :
6878 0 : if (em) {
6879 0 : if (em->start > start || em->start + em->len <= start)
6880 0 : free_extent_map(em);
6881 0 : else if (em->block_start == EXTENT_MAP_INLINE && page)
6882 0 : free_extent_map(em);
6883 : else
6884 0 : goto out;
6885 : }
6886 0 : em = alloc_extent_map();
6887 0 : if (!em) {
6888 0 : ret = -ENOMEM;
6889 0 : goto out;
6890 : }
6891 0 : em->start = EXTENT_MAP_HOLE;
6892 0 : em->orig_start = EXTENT_MAP_HOLE;
6893 0 : em->len = (u64)-1;
6894 0 : em->block_len = (u64)-1;
6895 :
6896 0 : path = btrfs_alloc_path();
6897 0 : if (!path) {
6898 0 : ret = -ENOMEM;
6899 0 : goto out;
6900 : }
6901 :
6902 : /* Chances are we'll be called again, so go ahead and do readahead */
6903 0 : path->reada = READA_FORWARD;
6904 :
6905 : /*
6906 : * The same explanation in load_free_space_cache applies here as well,
6907 : * we only read when we're loading the free space cache, and at that
6908 : * point the commit_root has everything we need.
6909 : */
6910 0 : if (btrfs_is_free_space_inode(inode)) {
6911 0 : path->search_commit_root = 1;
6912 0 : path->skip_locking = 1;
6913 : }
6914 :
6915 0 : ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6916 0 : if (ret < 0) {
6917 0 : goto out;
6918 0 : } else if (ret > 0) {
6919 0 : if (path->slots[0] == 0)
6920 0 : goto not_found;
6921 0 : path->slots[0]--;
6922 0 : ret = 0;
6923 : }
6924 :
6925 0 : leaf = path->nodes[0];
6926 0 : item = btrfs_item_ptr(leaf, path->slots[0],
6927 : struct btrfs_file_extent_item);
6928 0 : btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6929 0 : if (found_key.objectid != objectid ||
6930 0 : found_key.type != BTRFS_EXTENT_DATA_KEY) {
6931 : /*
6932 : * If we backup past the first extent we want to move forward
6933 : * and see if there is an extent in front of us, otherwise we'll
6934 : * say there is a hole for our whole search range which can
6935 : * cause problems.
6936 : */
6937 0 : extent_end = start;
6938 0 : goto next;
6939 : }
6940 :
6941 0 : extent_type = btrfs_file_extent_type(leaf, item);
6942 0 : extent_start = found_key.offset;
6943 0 : extent_end = btrfs_file_extent_end(path);
6944 0 : if (extent_type == BTRFS_FILE_EXTENT_REG ||
6945 : extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6946 : /* Only regular file could have regular/prealloc extent */
6947 0 : if (!S_ISREG(inode->vfs_inode.i_mode)) {
6948 0 : ret = -EUCLEAN;
6949 0 : btrfs_crit(fs_info,
6950 : "regular/prealloc extent found for non-regular inode %llu",
6951 : btrfs_ino(inode));
6952 0 : goto out;
6953 : }
6954 0 : trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6955 : extent_start);
6956 0 : } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6957 0 : trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6958 : path->slots[0],
6959 : extent_start);
6960 : }
6961 0 : next:
6962 0 : if (start >= extent_end) {
6963 0 : path->slots[0]++;
6964 0 : if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6965 0 : ret = btrfs_next_leaf(root, path);
6966 0 : if (ret < 0)
6967 0 : goto out;
6968 0 : else if (ret > 0)
6969 0 : goto not_found;
6970 :
6971 0 : leaf = path->nodes[0];
6972 : }
6973 0 : btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6974 0 : if (found_key.objectid != objectid ||
6975 0 : found_key.type != BTRFS_EXTENT_DATA_KEY)
6976 0 : goto not_found;
6977 0 : if (start + len <= found_key.offset)
6978 0 : goto not_found;
6979 0 : if (start > found_key.offset)
6980 0 : goto next;
6981 :
6982 : /* New extent overlaps with existing one */
6983 0 : em->start = start;
6984 0 : em->orig_start = start;
6985 0 : em->len = found_key.offset - start;
6986 0 : em->block_start = EXTENT_MAP_HOLE;
6987 0 : goto insert;
6988 : }
6989 :
6990 0 : btrfs_extent_item_to_extent_map(inode, path, item, em);
6991 :
6992 0 : if (extent_type == BTRFS_FILE_EXTENT_REG ||
6993 : extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6994 0 : goto insert;
6995 0 : } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6996 : /*
6997 : * Inline extent can only exist at file offset 0. This is
6998 : * ensured by tree-checker and inline extent creation path.
6999 : * Thus all members representing file offsets should be zero.
7000 : */
7001 0 : ASSERT(pg_offset == 0);
7002 0 : ASSERT(extent_start == 0);
7003 0 : ASSERT(em->start == 0);
7004 :
7005 : /*
7006 : * btrfs_extent_item_to_extent_map() should have properly
7007 : * initialized em members already.
7008 : *
7009 : * Other members are not utilized for inline extents.
7010 : */
7011 0 : ASSERT(em->block_start == EXTENT_MAP_INLINE);
7012 0 : ASSERT(em->len == fs_info->sectorsize);
7013 :
7014 0 : ret = read_inline_extent(inode, path, page);
7015 0 : if (ret < 0)
7016 0 : goto out;
7017 0 : goto insert;
7018 : }
7019 0 : not_found:
7020 0 : em->start = start;
7021 0 : em->orig_start = start;
7022 0 : em->len = len;
7023 0 : em->block_start = EXTENT_MAP_HOLE;
7024 0 : insert:
7025 0 : ret = 0;
7026 0 : btrfs_release_path(path);
7027 0 : if (em->start > start || extent_map_end(em) <= start) {
7028 0 : btrfs_err(fs_info,
7029 : "bad extent! em: [%llu %llu] passed [%llu %llu]",
7030 : em->start, em->len, start, len);
7031 0 : ret = -EIO;
7032 0 : goto out;
7033 : }
7034 :
7035 0 : write_lock(&em_tree->lock);
7036 0 : ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7037 0 : write_unlock(&em_tree->lock);
7038 0 : out:
7039 0 : btrfs_free_path(path);
7040 :
7041 0 : trace_btrfs_get_extent(root, inode, em);
7042 :
7043 0 : if (ret) {
7044 0 : free_extent_map(em);
7045 0 : return ERR_PTR(ret);
7046 : }
7047 0 : return em;
7048 : }
7049 :
7050 0 : static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7051 : struct btrfs_dio_data *dio_data,
7052 : const u64 start,
7053 : const u64 len,
7054 : const u64 orig_start,
7055 : const u64 block_start,
7056 : const u64 block_len,
7057 : const u64 orig_block_len,
7058 : const u64 ram_bytes,
7059 : const int type)
7060 : {
7061 0 : struct extent_map *em = NULL;
7062 0 : struct btrfs_ordered_extent *ordered;
7063 :
7064 0 : if (type != BTRFS_ORDERED_NOCOW) {
7065 0 : em = create_io_em(inode, start, len, orig_start, block_start,
7066 : block_len, orig_block_len, ram_bytes,
7067 : BTRFS_COMPRESS_NONE, /* compress_type */
7068 : type);
7069 0 : if (IS_ERR(em))
7070 0 : goto out;
7071 : }
7072 0 : ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
7073 : block_start, block_len, 0,
7074 0 : (1 << type) |
7075 : (1 << BTRFS_ORDERED_DIRECT),
7076 : BTRFS_COMPRESS_NONE);
7077 0 : if (IS_ERR(ordered)) {
7078 0 : if (em) {
7079 0 : free_extent_map(em);
7080 0 : btrfs_drop_extent_map_range(inode, start,
7081 0 : start + len - 1, false);
7082 : }
7083 : em = ERR_CAST(ordered);
7084 : } else {
7085 0 : ASSERT(!dio_data->ordered);
7086 0 : dio_data->ordered = ordered;
7087 : }
7088 0 : out:
7089 :
7090 0 : return em;
7091 : }
7092 :
7093 0 : static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7094 : struct btrfs_dio_data *dio_data,
7095 : u64 start, u64 len)
7096 : {
7097 0 : struct btrfs_root *root = inode->root;
7098 0 : struct btrfs_fs_info *fs_info = root->fs_info;
7099 0 : struct extent_map *em;
7100 0 : struct btrfs_key ins;
7101 0 : u64 alloc_hint;
7102 0 : int ret;
7103 :
7104 0 : alloc_hint = get_extent_allocation_hint(inode, start, len);
7105 0 : ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7106 : 0, alloc_hint, &ins, 1, 1);
7107 0 : if (ret)
7108 0 : return ERR_PTR(ret);
7109 :
7110 0 : em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
7111 : ins.objectid, ins.offset, ins.offset,
7112 : ins.offset, BTRFS_ORDERED_REGULAR);
7113 0 : btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7114 0 : if (IS_ERR(em))
7115 0 : btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7116 : 1);
7117 :
7118 : return em;
7119 : }
7120 :
7121 0 : static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7122 : {
7123 0 : struct btrfs_block_group *block_group;
7124 0 : bool readonly = false;
7125 :
7126 0 : block_group = btrfs_lookup_block_group(fs_info, bytenr);
7127 0 : if (!block_group || block_group->ro)
7128 0 : readonly = true;
7129 0 : if (block_group)
7130 0 : btrfs_put_block_group(block_group);
7131 0 : return readonly;
7132 : }
7133 :
7134 : /*
7135 : * Check if we can do nocow write into the range [@offset, @offset + @len)
7136 : *
7137 : * @offset: File offset
7138 : * @len: The length to write, will be updated to the nocow writeable
7139 : * range
7140 : * @orig_start: (optional) Return the original file offset of the file extent
7141 : * @orig_len: (optional) Return the original on-disk length of the file extent
7142 : * @ram_bytes: (optional) Return the ram_bytes of the file extent
7143 : * @strict: if true, omit optimizations that might force us into unnecessary
7144 : * cow. e.g., don't trust generation number.
7145 : *
7146 : * Return:
7147 : * >0 and update @len if we can do nocow write
7148 : * 0 if we can't do nocow write
7149 : * <0 if error happened
7150 : *
7151 : * NOTE: This only checks the file extents, caller is responsible to wait for
7152 : * any ordered extents.
7153 : */
7154 0 : noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7155 : u64 *orig_start, u64 *orig_block_len,
7156 : u64 *ram_bytes, bool nowait, bool strict)
7157 : {
7158 0 : struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7159 0 : struct can_nocow_file_extent_args nocow_args = { 0 };
7160 0 : struct btrfs_path *path;
7161 0 : int ret;
7162 0 : struct extent_buffer *leaf;
7163 0 : struct btrfs_root *root = BTRFS_I(inode)->root;
7164 0 : struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7165 0 : struct btrfs_file_extent_item *fi;
7166 0 : struct btrfs_key key;
7167 0 : int found_type;
7168 :
7169 0 : path = btrfs_alloc_path();
7170 0 : if (!path)
7171 : return -ENOMEM;
7172 0 : path->nowait = nowait;
7173 :
7174 0 : ret = btrfs_lookup_file_extent(NULL, root, path,
7175 : btrfs_ino(BTRFS_I(inode)), offset, 0);
7176 0 : if (ret < 0)
7177 0 : goto out;
7178 :
7179 0 : if (ret == 1) {
7180 0 : if (path->slots[0] == 0) {
7181 : /* can't find the item, must cow */
7182 0 : ret = 0;
7183 0 : goto out;
7184 : }
7185 0 : path->slots[0]--;
7186 : }
7187 0 : ret = 0;
7188 0 : leaf = path->nodes[0];
7189 0 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7190 0 : if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7191 0 : key.type != BTRFS_EXTENT_DATA_KEY) {
7192 : /* not our file or wrong item type, must cow */
7193 0 : goto out;
7194 : }
7195 :
7196 0 : if (key.offset > offset) {
7197 : /* Wrong offset, must cow */
7198 0 : goto out;
7199 : }
7200 :
7201 0 : if (btrfs_file_extent_end(path) <= offset)
7202 0 : goto out;
7203 :
7204 0 : fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7205 0 : found_type = btrfs_file_extent_type(leaf, fi);
7206 0 : if (ram_bytes)
7207 0 : *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7208 :
7209 0 : nocow_args.start = offset;
7210 0 : nocow_args.end = offset + *len - 1;
7211 0 : nocow_args.strict = strict;
7212 0 : nocow_args.free_path = true;
7213 :
7214 0 : ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7215 : /* can_nocow_file_extent() has freed the path. */
7216 0 : path = NULL;
7217 :
7218 0 : if (ret != 1) {
7219 : /* Treat errors as not being able to NOCOW. */
7220 0 : ret = 0;
7221 0 : goto out;
7222 : }
7223 :
7224 0 : ret = 0;
7225 0 : if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7226 0 : goto out;
7227 :
7228 0 : if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7229 : found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7230 0 : u64 range_end;
7231 :
7232 0 : range_end = round_up(offset + nocow_args.num_bytes,
7233 : root->fs_info->sectorsize) - 1;
7234 0 : ret = test_range_bit(io_tree, offset, range_end,
7235 : EXTENT_DELALLOC, 0, NULL);
7236 0 : if (ret) {
7237 0 : ret = -EAGAIN;
7238 0 : goto out;
7239 : }
7240 : }
7241 :
7242 0 : if (orig_start)
7243 0 : *orig_start = key.offset - nocow_args.extent_offset;
7244 0 : if (orig_block_len)
7245 0 : *orig_block_len = nocow_args.disk_num_bytes;
7246 :
7247 0 : *len = nocow_args.num_bytes;
7248 0 : ret = 1;
7249 0 : out:
7250 0 : btrfs_free_path(path);
7251 0 : return ret;
7252 : }
7253 :
7254 0 : static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7255 : struct extent_state **cached_state,
7256 : unsigned int iomap_flags)
7257 : {
7258 0 : const bool writing = (iomap_flags & IOMAP_WRITE);
7259 0 : const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7260 0 : struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7261 0 : struct btrfs_ordered_extent *ordered;
7262 0 : int ret = 0;
7263 :
7264 0 : while (1) {
7265 0 : if (nowait) {
7266 0 : if (!try_lock_extent(io_tree, lockstart, lockend,
7267 : cached_state))
7268 : return -EAGAIN;
7269 : } else {
7270 0 : lock_extent(io_tree, lockstart, lockend, cached_state);
7271 : }
7272 : /*
7273 : * We're concerned with the entire range that we're going to be
7274 : * doing DIO to, so we need to make sure there's no ordered
7275 : * extents in this range.
7276 : */
7277 0 : ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7278 0 : lockend - lockstart + 1);
7279 :
7280 : /*
7281 : * We need to make sure there are no buffered pages in this
7282 : * range either, we could have raced between the invalidate in
7283 : * generic_file_direct_write and locking the extent. The
7284 : * invalidate needs to happen so that reads after a write do not
7285 : * get stale data.
7286 : */
7287 0 : if (!ordered &&
7288 0 : (!writing || !filemap_range_has_page(inode->i_mapping,
7289 : lockstart, lockend)))
7290 : break;
7291 :
7292 0 : unlock_extent(io_tree, lockstart, lockend, cached_state);
7293 :
7294 0 : if (ordered) {
7295 0 : if (nowait) {
7296 0 : btrfs_put_ordered_extent(ordered);
7297 0 : ret = -EAGAIN;
7298 0 : break;
7299 : }
7300 : /*
7301 : * If we are doing a DIO read and the ordered extent we
7302 : * found is for a buffered write, we can not wait for it
7303 : * to complete and retry, because if we do so we can
7304 : * deadlock with concurrent buffered writes on page
7305 : * locks. This happens only if our DIO read covers more
7306 : * than one extent map, if at this point has already
7307 : * created an ordered extent for a previous extent map
7308 : * and locked its range in the inode's io tree, and a
7309 : * concurrent write against that previous extent map's
7310 : * range and this range started (we unlock the ranges
7311 : * in the io tree only when the bios complete and
7312 : * buffered writes always lock pages before attempting
7313 : * to lock range in the io tree).
7314 : */
7315 0 : if (writing ||
7316 0 : test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7317 0 : btrfs_start_ordered_extent(ordered);
7318 : else
7319 : ret = nowait ? -EAGAIN : -ENOTBLK;
7320 0 : btrfs_put_ordered_extent(ordered);
7321 : } else {
7322 : /*
7323 : * We could trigger writeback for this range (and wait
7324 : * for it to complete) and then invalidate the pages for
7325 : * this range (through invalidate_inode_pages2_range()),
7326 : * but that can lead us to a deadlock with a concurrent
7327 : * call to readahead (a buffered read or a defrag call
7328 : * triggered a readahead) on a page lock due to an
7329 : * ordered dio extent we created before but did not have
7330 : * yet a corresponding bio submitted (whence it can not
7331 : * complete), which makes readahead wait for that
7332 : * ordered extent to complete while holding a lock on
7333 : * that page.
7334 : */
7335 0 : ret = nowait ? -EAGAIN : -ENOTBLK;
7336 : }
7337 :
7338 0 : if (ret)
7339 : break;
7340 :
7341 0 : cond_resched();
7342 : }
7343 :
7344 : return ret;
7345 : }
7346 :
7347 : /* The callers of this must take lock_extent() */
7348 0 : static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7349 : u64 len, u64 orig_start, u64 block_start,
7350 : u64 block_len, u64 orig_block_len,
7351 : u64 ram_bytes, int compress_type,
7352 : int type)
7353 : {
7354 0 : struct extent_map *em;
7355 0 : int ret;
7356 :
7357 0 : ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7358 : type == BTRFS_ORDERED_COMPRESSED ||
7359 : type == BTRFS_ORDERED_NOCOW ||
7360 : type == BTRFS_ORDERED_REGULAR);
7361 :
7362 0 : em = alloc_extent_map();
7363 0 : if (!em)
7364 : return ERR_PTR(-ENOMEM);
7365 :
7366 0 : em->start = start;
7367 0 : em->orig_start = orig_start;
7368 0 : em->len = len;
7369 0 : em->block_len = block_len;
7370 0 : em->block_start = block_start;
7371 0 : em->orig_block_len = orig_block_len;
7372 0 : em->ram_bytes = ram_bytes;
7373 0 : em->generation = -1;
7374 0 : set_bit(EXTENT_FLAG_PINNED, &em->flags);
7375 0 : if (type == BTRFS_ORDERED_PREALLOC) {
7376 0 : set_bit(EXTENT_FLAG_FILLING, &em->flags);
7377 0 : } else if (type == BTRFS_ORDERED_COMPRESSED) {
7378 0 : set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7379 0 : em->compress_type = compress_type;
7380 : }
7381 :
7382 0 : ret = btrfs_replace_extent_map_range(inode, em, true);
7383 0 : if (ret) {
7384 0 : free_extent_map(em);
7385 0 : return ERR_PTR(ret);
7386 : }
7387 :
7388 : /* em got 2 refs now, callers needs to do free_extent_map once. */
7389 : return em;
7390 : }
7391 :
7392 :
7393 0 : static int btrfs_get_blocks_direct_write(struct extent_map **map,
7394 : struct inode *inode,
7395 : struct btrfs_dio_data *dio_data,
7396 : u64 start, u64 *lenp,
7397 : unsigned int iomap_flags)
7398 : {
7399 0 : const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7400 0 : struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7401 0 : struct extent_map *em = *map;
7402 0 : int type;
7403 0 : u64 block_start, orig_start, orig_block_len, ram_bytes;
7404 0 : struct btrfs_block_group *bg;
7405 0 : bool can_nocow = false;
7406 0 : bool space_reserved = false;
7407 0 : u64 len = *lenp;
7408 0 : u64 prev_len;
7409 0 : int ret = 0;
7410 :
7411 : /*
7412 : * We don't allocate a new extent in the following cases
7413 : *
7414 : * 1) The inode is marked as NODATACOW. In this case we'll just use the
7415 : * existing extent.
7416 : * 2) The extent is marked as PREALLOC. We're good to go here and can
7417 : * just use the extent.
7418 : *
7419 : */
7420 0 : if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7421 0 : ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7422 0 : em->block_start != EXTENT_MAP_HOLE)) {
7423 0 : if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7424 : type = BTRFS_ORDERED_PREALLOC;
7425 : else
7426 0 : type = BTRFS_ORDERED_NOCOW;
7427 0 : len = min(len, em->len - (start - em->start));
7428 0 : block_start = em->block_start + (start - em->start);
7429 :
7430 0 : if (can_nocow_extent(inode, start, &len, &orig_start,
7431 : &orig_block_len, &ram_bytes, false, false) == 1) {
7432 0 : bg = btrfs_inc_nocow_writers(fs_info, block_start);
7433 0 : if (bg)
7434 0 : can_nocow = true;
7435 : }
7436 : }
7437 :
7438 0 : prev_len = len;
7439 0 : if (can_nocow) {
7440 0 : struct extent_map *em2;
7441 :
7442 : /* We can NOCOW, so only need to reserve metadata space. */
7443 0 : ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7444 : nowait);
7445 0 : if (ret < 0) {
7446 : /* Our caller expects us to free the input extent map. */
7447 0 : free_extent_map(em);
7448 0 : *map = NULL;
7449 0 : btrfs_dec_nocow_writers(bg);
7450 0 : if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7451 0 : ret = -EAGAIN;
7452 0 : goto out;
7453 : }
7454 0 : space_reserved = true;
7455 :
7456 0 : em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7457 : orig_start, block_start,
7458 : len, orig_block_len,
7459 : ram_bytes, type);
7460 0 : btrfs_dec_nocow_writers(bg);
7461 0 : if (type == BTRFS_ORDERED_PREALLOC) {
7462 0 : free_extent_map(em);
7463 0 : *map = em2;
7464 0 : em = em2;
7465 : }
7466 :
7467 0 : if (IS_ERR(em2)) {
7468 0 : ret = PTR_ERR(em2);
7469 0 : goto out;
7470 : }
7471 :
7472 0 : dio_data->nocow_done = true;
7473 : } else {
7474 : /* Our caller expects us to free the input extent map. */
7475 0 : free_extent_map(em);
7476 0 : *map = NULL;
7477 :
7478 0 : if (nowait) {
7479 0 : ret = -EAGAIN;
7480 0 : goto out;
7481 : }
7482 :
7483 : /*
7484 : * If we could not allocate data space before locking the file
7485 : * range and we can't do a NOCOW write, then we have to fail.
7486 : */
7487 0 : if (!dio_data->data_space_reserved) {
7488 0 : ret = -ENOSPC;
7489 0 : goto out;
7490 : }
7491 :
7492 : /*
7493 : * We have to COW and we have already reserved data space before,
7494 : * so now we reserve only metadata.
7495 : */
7496 0 : ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7497 : false);
7498 0 : if (ret < 0)
7499 0 : goto out;
7500 0 : space_reserved = true;
7501 :
7502 0 : em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7503 0 : if (IS_ERR(em)) {
7504 0 : ret = PTR_ERR(em);
7505 0 : goto out;
7506 : }
7507 0 : *map = em;
7508 0 : len = min(len, em->len - (start - em->start));
7509 0 : if (len < prev_len)
7510 0 : btrfs_delalloc_release_metadata(BTRFS_I(inode),
7511 : prev_len - len, true);
7512 : }
7513 :
7514 : /*
7515 : * We have created our ordered extent, so we can now release our reservation
7516 : * for an outstanding extent.
7517 : */
7518 0 : btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7519 :
7520 : /*
7521 : * Need to update the i_size under the extent lock so buffered
7522 : * readers will get the updated i_size when we unlock.
7523 : */
7524 0 : if (start + len > i_size_read(inode))
7525 0 : i_size_write(inode, start + len);
7526 0 : out:
7527 0 : if (ret && space_reserved) {
7528 0 : btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7529 0 : btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7530 : }
7531 0 : *lenp = len;
7532 0 : return ret;
7533 : }
7534 :
7535 0 : static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7536 : loff_t length, unsigned int flags, struct iomap *iomap,
7537 : struct iomap *srcmap)
7538 : {
7539 0 : struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7540 0 : struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7541 0 : struct extent_map *em;
7542 0 : struct extent_state *cached_state = NULL;
7543 0 : struct btrfs_dio_data *dio_data = iter->private;
7544 0 : u64 lockstart, lockend;
7545 0 : const bool write = !!(flags & IOMAP_WRITE);
7546 0 : int ret = 0;
7547 0 : u64 len = length;
7548 0 : const u64 data_alloc_len = length;
7549 0 : bool unlock_extents = false;
7550 :
7551 : /*
7552 : * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7553 : * we're NOWAIT we may submit a bio for a partial range and return
7554 : * EIOCBQUEUED, which would result in an errant short read.
7555 : *
7556 : * The best way to handle this would be to allow for partial completions
7557 : * of iocb's, so we could submit the partial bio, return and fault in
7558 : * the rest of the pages, and then submit the io for the rest of the
7559 : * range. However we don't have that currently, so simply return
7560 : * -EAGAIN at this point so that the normal path is used.
7561 : */
7562 0 : if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7563 : return -EAGAIN;
7564 :
7565 : /*
7566 : * Cap the size of reads to that usually seen in buffered I/O as we need
7567 : * to allocate a contiguous array for the checksums.
7568 : */
7569 0 : if (!write)
7570 0 : len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7571 :
7572 0 : lockstart = start;
7573 0 : lockend = start + len - 1;
7574 :
7575 : /*
7576 : * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7577 : * enough if we've written compressed pages to this area, so we need to
7578 : * flush the dirty pages again to make absolutely sure that any
7579 : * outstanding dirty pages are on disk - the first flush only starts
7580 : * compression on the data, while keeping the pages locked, so by the
7581 : * time the second flush returns we know bios for the compressed pages
7582 : * were submitted and finished, and the pages no longer under writeback.
7583 : *
7584 : * If we have a NOWAIT request and we have any pages in the range that
7585 : * are locked, likely due to compression still in progress, we don't want
7586 : * to block on page locks. We also don't want to block on pages marked as
7587 : * dirty or under writeback (same as for the non-compression case).
7588 : * iomap_dio_rw() did the same check, but after that and before we got
7589 : * here, mmap'ed writes may have happened or buffered reads started
7590 : * (readpage() and readahead(), which lock pages), as we haven't locked
7591 : * the file range yet.
7592 : */
7593 0 : if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7594 : &BTRFS_I(inode)->runtime_flags)) {
7595 0 : if (flags & IOMAP_NOWAIT) {
7596 0 : if (filemap_range_needs_writeback(inode->i_mapping,
7597 : lockstart, lockend))
7598 : return -EAGAIN;
7599 : } else {
7600 0 : ret = filemap_fdatawrite_range(inode->i_mapping, start,
7601 0 : start + length - 1);
7602 0 : if (ret)
7603 : return ret;
7604 : }
7605 : }
7606 :
7607 0 : memset(dio_data, 0, sizeof(*dio_data));
7608 :
7609 : /*
7610 : * We always try to allocate data space and must do it before locking
7611 : * the file range, to avoid deadlocks with concurrent writes to the same
7612 : * range if the range has several extents and the writes don't expand the
7613 : * current i_size (the inode lock is taken in shared mode). If we fail to
7614 : * allocate data space here we continue and later, after locking the
7615 : * file range, we fail with ENOSPC only if we figure out we can not do a
7616 : * NOCOW write.
7617 : */
7618 0 : if (write && !(flags & IOMAP_NOWAIT)) {
7619 0 : ret = btrfs_check_data_free_space(BTRFS_I(inode),
7620 : &dio_data->data_reserved,
7621 : start, data_alloc_len, false);
7622 0 : if (!ret)
7623 0 : dio_data->data_space_reserved = true;
7624 0 : else if (ret && !(BTRFS_I(inode)->flags &
7625 : (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7626 0 : goto err;
7627 : }
7628 :
7629 : /*
7630 : * If this errors out it's because we couldn't invalidate pagecache for
7631 : * this range and we need to fallback to buffered IO, or we are doing a
7632 : * NOWAIT read/write and we need to block.
7633 : */
7634 0 : ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7635 0 : if (ret < 0)
7636 0 : goto err;
7637 :
7638 0 : em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7639 0 : if (IS_ERR(em)) {
7640 0 : ret = PTR_ERR(em);
7641 0 : goto unlock_err;
7642 : }
7643 :
7644 : /*
7645 : * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7646 : * io. INLINE is special, and we could probably kludge it in here, but
7647 : * it's still buffered so for safety lets just fall back to the generic
7648 : * buffered path.
7649 : *
7650 : * For COMPRESSED we _have_ to read the entire extent in so we can
7651 : * decompress it, so there will be buffering required no matter what we
7652 : * do, so go ahead and fallback to buffered.
7653 : *
7654 : * We return -ENOTBLK because that's what makes DIO go ahead and go back
7655 : * to buffered IO. Don't blame me, this is the price we pay for using
7656 : * the generic code.
7657 : */
7658 0 : if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7659 0 : em->block_start == EXTENT_MAP_INLINE) {
7660 0 : free_extent_map(em);
7661 : /*
7662 : * If we are in a NOWAIT context, return -EAGAIN in order to
7663 : * fallback to buffered IO. This is not only because we can
7664 : * block with buffered IO (no support for NOWAIT semantics at
7665 : * the moment) but also to avoid returning short reads to user
7666 : * space - this happens if we were able to read some data from
7667 : * previous non-compressed extents and then when we fallback to
7668 : * buffered IO, at btrfs_file_read_iter() by calling
7669 : * filemap_read(), we fail to fault in pages for the read buffer,
7670 : * in which case filemap_read() returns a short read (the number
7671 : * of bytes previously read is > 0, so it does not return -EFAULT).
7672 : */
7673 0 : ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7674 0 : goto unlock_err;
7675 : }
7676 :
7677 0 : len = min(len, em->len - (start - em->start));
7678 :
7679 : /*
7680 : * If we have a NOWAIT request and the range contains multiple extents
7681 : * (or a mix of extents and holes), then we return -EAGAIN to make the
7682 : * caller fallback to a context where it can do a blocking (without
7683 : * NOWAIT) request. This way we avoid doing partial IO and returning
7684 : * success to the caller, which is not optimal for writes and for reads
7685 : * it can result in unexpected behaviour for an application.
7686 : *
7687 : * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7688 : * iomap_dio_rw(), we can end up returning less data then what the caller
7689 : * asked for, resulting in an unexpected, and incorrect, short read.
7690 : * That is, the caller asked to read N bytes and we return less than that,
7691 : * which is wrong unless we are crossing EOF. This happens if we get a
7692 : * page fault error when trying to fault in pages for the buffer that is
7693 : * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7694 : * have previously submitted bios for other extents in the range, in
7695 : * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7696 : * those bios have completed by the time we get the page fault error,
7697 : * which we return back to our caller - we should only return EIOCBQUEUED
7698 : * after we have submitted bios for all the extents in the range.
7699 : */
7700 0 : if ((flags & IOMAP_NOWAIT) && len < length) {
7701 0 : free_extent_map(em);
7702 0 : ret = -EAGAIN;
7703 0 : goto unlock_err;
7704 : }
7705 :
7706 0 : if (write) {
7707 0 : ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7708 : start, &len, flags);
7709 0 : if (ret < 0)
7710 0 : goto unlock_err;
7711 0 : unlock_extents = true;
7712 : /* Recalc len in case the new em is smaller than requested */
7713 0 : len = min(len, em->len - (start - em->start));
7714 0 : if (dio_data->data_space_reserved) {
7715 0 : u64 release_offset;
7716 0 : u64 release_len = 0;
7717 :
7718 0 : if (dio_data->nocow_done) {
7719 : release_offset = start;
7720 : release_len = data_alloc_len;
7721 0 : } else if (len < data_alloc_len) {
7722 0 : release_offset = start + len;
7723 0 : release_len = data_alloc_len - len;
7724 : }
7725 :
7726 0 : if (release_len > 0)
7727 0 : btrfs_free_reserved_data_space(BTRFS_I(inode),
7728 : dio_data->data_reserved,
7729 : release_offset,
7730 : release_len);
7731 : }
7732 : } else {
7733 : /*
7734 : * We need to unlock only the end area that we aren't using.
7735 : * The rest is going to be unlocked by the endio routine.
7736 : */
7737 0 : lockstart = start + len;
7738 0 : if (lockstart < lockend)
7739 : unlock_extents = true;
7740 : }
7741 :
7742 0 : if (unlock_extents)
7743 0 : unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7744 : &cached_state);
7745 : else
7746 0 : free_extent_state(cached_state);
7747 :
7748 : /*
7749 : * Translate extent map information to iomap.
7750 : * We trim the extents (and move the addr) even though iomap code does
7751 : * that, since we have locked only the parts we are performing I/O in.
7752 : */
7753 0 : if ((em->block_start == EXTENT_MAP_HOLE) ||
7754 0 : (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7755 0 : iomap->addr = IOMAP_NULL_ADDR;
7756 0 : iomap->type = IOMAP_HOLE;
7757 : } else {
7758 0 : iomap->addr = em->block_start + (start - em->start);
7759 0 : iomap->type = IOMAP_MAPPED;
7760 : }
7761 0 : iomap->offset = start;
7762 0 : iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7763 0 : iomap->length = len;
7764 0 : free_extent_map(em);
7765 :
7766 0 : return 0;
7767 :
7768 0 : unlock_err:
7769 0 : unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7770 : &cached_state);
7771 0 : err:
7772 0 : if (dio_data->data_space_reserved) {
7773 0 : btrfs_free_reserved_data_space(BTRFS_I(inode),
7774 : dio_data->data_reserved,
7775 : start, data_alloc_len);
7776 0 : extent_changeset_free(dio_data->data_reserved);
7777 : }
7778 :
7779 : return ret;
7780 : }
7781 :
7782 0 : static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7783 : ssize_t written, unsigned int flags, struct iomap *iomap)
7784 : {
7785 0 : struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7786 0 : struct btrfs_dio_data *dio_data = iter->private;
7787 0 : size_t submitted = dio_data->submitted;
7788 0 : const bool write = !!(flags & IOMAP_WRITE);
7789 0 : int ret = 0;
7790 :
7791 0 : if (!write && (iomap->type == IOMAP_HOLE)) {
7792 : /* If reading from a hole, unlock and return */
7793 0 : unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7794 : NULL);
7795 0 : return 0;
7796 : }
7797 :
7798 0 : if (submitted < length) {
7799 0 : pos += submitted;
7800 0 : length -= submitted;
7801 0 : if (write)
7802 0 : btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7803 : pos, length, false);
7804 : else
7805 0 : unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7806 0 : pos + length - 1, NULL);
7807 : ret = -ENOTBLK;
7808 : }
7809 0 : if (write) {
7810 0 : btrfs_put_ordered_extent(dio_data->ordered);
7811 0 : dio_data->ordered = NULL;
7812 : }
7813 :
7814 0 : if (write)
7815 0 : extent_changeset_free(dio_data->data_reserved);
7816 : return ret;
7817 : }
7818 :
7819 0 : static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7820 : {
7821 0 : struct btrfs_dio_private *dip =
7822 0 : container_of(bbio, struct btrfs_dio_private, bbio);
7823 0 : struct btrfs_inode *inode = bbio->inode;
7824 0 : struct bio *bio = &bbio->bio;
7825 :
7826 0 : if (bio->bi_status) {
7827 0 : btrfs_warn(inode->root->fs_info,
7828 : "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7829 : btrfs_ino(inode), bio->bi_opf,
7830 : dip->file_offset, dip->bytes, bio->bi_status);
7831 : }
7832 :
7833 0 : if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7834 0 : btrfs_finish_ordered_extent(bbio->ordered, NULL,
7835 0 : dip->file_offset, dip->bytes,
7836 0 : !bio->bi_status);
7837 : } else {
7838 0 : unlock_extent(&inode->io_tree, dip->file_offset,
7839 0 : dip->file_offset + dip->bytes - 1, NULL);
7840 : }
7841 :
7842 0 : bbio->bio.bi_private = bbio->private;
7843 0 : iomap_dio_bio_end_io(bio);
7844 0 : }
7845 :
7846 0 : static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7847 : loff_t file_offset)
7848 : {
7849 0 : struct btrfs_bio *bbio = btrfs_bio(bio);
7850 0 : struct btrfs_dio_private *dip =
7851 0 : container_of(bbio, struct btrfs_dio_private, bbio);
7852 0 : struct btrfs_dio_data *dio_data = iter->private;
7853 :
7854 0 : btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7855 : btrfs_dio_end_io, bio->bi_private);
7856 0 : bbio->inode = BTRFS_I(iter->inode);
7857 0 : bbio->file_offset = file_offset;
7858 :
7859 0 : dip->file_offset = file_offset;
7860 0 : dip->bytes = bio->bi_iter.bi_size;
7861 :
7862 0 : dio_data->submitted += bio->bi_iter.bi_size;
7863 :
7864 : /*
7865 : * Check if we are doing a partial write. If we are, we need to split
7866 : * the ordered extent to match the submitted bio. Hang on to the
7867 : * remaining unfinishable ordered_extent in dio_data so that it can be
7868 : * cancelled in iomap_end to avoid a deadlock wherein faulting the
7869 : * remaining pages is blocked on the outstanding ordered extent.
7870 : */
7871 0 : if (iter->flags & IOMAP_WRITE) {
7872 0 : int ret;
7873 :
7874 0 : ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7875 0 : if (ret) {
7876 0 : btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7877 0 : file_offset, dip->bytes,
7878 : !ret);
7879 0 : bio->bi_status = errno_to_blk_status(ret);
7880 0 : iomap_dio_bio_end_io(bio);
7881 0 : return;
7882 : }
7883 : }
7884 :
7885 0 : btrfs_submit_bio(bbio, 0);
7886 : }
7887 :
7888 : static const struct iomap_ops btrfs_dio_iomap_ops = {
7889 : .iomap_begin = btrfs_dio_iomap_begin,
7890 : .iomap_end = btrfs_dio_iomap_end,
7891 : };
7892 :
7893 : static const struct iomap_dio_ops btrfs_dio_ops = {
7894 : .submit_io = btrfs_dio_submit_io,
7895 : .bio_set = &btrfs_dio_bioset,
7896 : };
7897 :
7898 0 : ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7899 : {
7900 0 : struct btrfs_dio_data data = { 0 };
7901 :
7902 0 : return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7903 : IOMAP_DIO_PARTIAL, &data, done_before);
7904 : }
7905 :
7906 0 : struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7907 : size_t done_before)
7908 : {
7909 0 : struct btrfs_dio_data data = { 0 };
7910 :
7911 0 : return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7912 : IOMAP_DIO_PARTIAL, &data, done_before);
7913 : }
7914 :
7915 0 : static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7916 : u64 start, u64 len)
7917 : {
7918 0 : int ret;
7919 :
7920 0 : ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7921 0 : if (ret)
7922 : return ret;
7923 :
7924 : /*
7925 : * fiemap_prep() called filemap_write_and_wait() for the whole possible
7926 : * file range (0 to LLONG_MAX), but that is not enough if we have
7927 : * compression enabled. The first filemap_fdatawrite_range() only kicks
7928 : * in the compression of data (in an async thread) and will return
7929 : * before the compression is done and writeback is started. A second
7930 : * filemap_fdatawrite_range() is needed to wait for the compression to
7931 : * complete and writeback to start. We also need to wait for ordered
7932 : * extents to complete, because our fiemap implementation uses mainly
7933 : * file extent items to list the extents, searching for extent maps
7934 : * only for file ranges with holes or prealloc extents to figure out
7935 : * if we have delalloc in those ranges.
7936 : */
7937 0 : if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7938 0 : ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7939 0 : if (ret)
7940 : return ret;
7941 : }
7942 :
7943 0 : return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
7944 : }
7945 :
7946 0 : static int btrfs_writepages(struct address_space *mapping,
7947 : struct writeback_control *wbc)
7948 : {
7949 0 : return extent_writepages(mapping, wbc);
7950 : }
7951 :
7952 0 : static void btrfs_readahead(struct readahead_control *rac)
7953 : {
7954 0 : extent_readahead(rac);
7955 0 : }
7956 :
7957 : /*
7958 : * For release_folio() and invalidate_folio() we have a race window where
7959 : * folio_end_writeback() is called but the subpage spinlock is not yet released.
7960 : * If we continue to release/invalidate the page, we could cause use-after-free
7961 : * for subpage spinlock. So this function is to spin and wait for subpage
7962 : * spinlock.
7963 : */
7964 0 : static void wait_subpage_spinlock(struct page *page)
7965 : {
7966 0 : struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7967 0 : struct btrfs_subpage *subpage;
7968 :
7969 0 : if (!btrfs_is_subpage(fs_info, page))
7970 : return;
7971 :
7972 0 : ASSERT(PagePrivate(page) && page->private);
7973 0 : subpage = (struct btrfs_subpage *)page->private;
7974 :
7975 : /*
7976 : * This may look insane as we just acquire the spinlock and release it,
7977 : * without doing anything. But we just want to make sure no one is
7978 : * still holding the subpage spinlock.
7979 : * And since the page is not dirty nor writeback, and we have page
7980 : * locked, the only possible way to hold a spinlock is from the endio
7981 : * function to clear page writeback.
7982 : *
7983 : * Here we just acquire the spinlock so that all existing callers
7984 : * should exit and we're safe to release/invalidate the page.
7985 : */
7986 0 : spin_lock_irq(&subpage->lock);
7987 0 : spin_unlock_irq(&subpage->lock);
7988 : }
7989 :
7990 0 : static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7991 : {
7992 0 : int ret = try_release_extent_mapping(&folio->page, gfp_flags);
7993 :
7994 0 : if (ret == 1) {
7995 0 : wait_subpage_spinlock(&folio->page);
7996 0 : clear_page_extent_mapped(&folio->page);
7997 : }
7998 0 : return ret;
7999 : }
8000 :
8001 0 : static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8002 : {
8003 0 : if (folio_test_writeback(folio) || folio_test_dirty(folio))
8004 : return false;
8005 0 : return __btrfs_release_folio(folio, gfp_flags);
8006 : }
8007 :
8008 : #ifdef CONFIG_MIGRATION
8009 0 : static int btrfs_migrate_folio(struct address_space *mapping,
8010 : struct folio *dst, struct folio *src,
8011 : enum migrate_mode mode)
8012 : {
8013 0 : int ret = filemap_migrate_folio(mapping, dst, src, mode);
8014 :
8015 0 : if (ret != MIGRATEPAGE_SUCCESS)
8016 : return ret;
8017 :
8018 0 : if (folio_test_ordered(src)) {
8019 0 : folio_clear_ordered(src);
8020 0 : folio_set_ordered(dst);
8021 : }
8022 :
8023 : return MIGRATEPAGE_SUCCESS;
8024 : }
8025 : #else
8026 : #define btrfs_migrate_folio NULL
8027 : #endif
8028 :
8029 0 : static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8030 : size_t length)
8031 : {
8032 0 : struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
8033 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
8034 0 : struct extent_io_tree *tree = &inode->io_tree;
8035 0 : struct extent_state *cached_state = NULL;
8036 0 : u64 page_start = folio_pos(folio);
8037 0 : u64 page_end = page_start + folio_size(folio) - 1;
8038 0 : u64 cur;
8039 0 : int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8040 :
8041 : /*
8042 : * We have folio locked so no new ordered extent can be created on this
8043 : * page, nor bio can be submitted for this folio.
8044 : *
8045 : * But already submitted bio can still be finished on this folio.
8046 : * Furthermore, endio function won't skip folio which has Ordered
8047 : * (Private2) already cleared, so it's possible for endio and
8048 : * invalidate_folio to do the same ordered extent accounting twice
8049 : * on one folio.
8050 : *
8051 : * So here we wait for any submitted bios to finish, so that we won't
8052 : * do double ordered extent accounting on the same folio.
8053 : */
8054 0 : folio_wait_writeback(folio);
8055 0 : wait_subpage_spinlock(&folio->page);
8056 :
8057 : /*
8058 : * For subpage case, we have call sites like
8059 : * btrfs_punch_hole_lock_range() which passes range not aligned to
8060 : * sectorsize.
8061 : * If the range doesn't cover the full folio, we don't need to and
8062 : * shouldn't clear page extent mapped, as folio->private can still
8063 : * record subpage dirty bits for other part of the range.
8064 : *
8065 : * For cases that invalidate the full folio even the range doesn't
8066 : * cover the full folio, like invalidating the last folio, we're
8067 : * still safe to wait for ordered extent to finish.
8068 : */
8069 0 : if (!(offset == 0 && length == folio_size(folio))) {
8070 0 : btrfs_release_folio(folio, GFP_NOFS);
8071 0 : return;
8072 : }
8073 :
8074 0 : if (!inode_evicting)
8075 0 : lock_extent(tree, page_start, page_end, &cached_state);
8076 :
8077 : cur = page_start;
8078 0 : while (cur < page_end) {
8079 0 : struct btrfs_ordered_extent *ordered;
8080 0 : u64 range_end;
8081 0 : u32 range_len;
8082 0 : u32 extra_flags = 0;
8083 :
8084 0 : ordered = btrfs_lookup_first_ordered_range(inode, cur,
8085 0 : page_end + 1 - cur);
8086 0 : if (!ordered) {
8087 0 : range_end = page_end;
8088 : /*
8089 : * No ordered extent covering this range, we are safe
8090 : * to delete all extent states in the range.
8091 : */
8092 0 : extra_flags = EXTENT_CLEAR_ALL_BITS;
8093 0 : goto next;
8094 : }
8095 0 : if (ordered->file_offset > cur) {
8096 : /*
8097 : * There is a range between [cur, oe->file_offset) not
8098 : * covered by any ordered extent.
8099 : * We are safe to delete all extent states, and handle
8100 : * the ordered extent in the next iteration.
8101 : */
8102 0 : range_end = ordered->file_offset - 1;
8103 0 : extra_flags = EXTENT_CLEAR_ALL_BITS;
8104 0 : goto next;
8105 : }
8106 :
8107 0 : range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8108 : page_end);
8109 0 : ASSERT(range_end + 1 - cur < U32_MAX);
8110 0 : range_len = range_end + 1 - cur;
8111 0 : if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8112 : /*
8113 : * If Ordered (Private2) is cleared, it means endio has
8114 : * already been executed for the range.
8115 : * We can't delete the extent states as
8116 : * btrfs_finish_ordered_io() may still use some of them.
8117 : */
8118 0 : goto next;
8119 : }
8120 0 : btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8121 :
8122 : /*
8123 : * IO on this page will never be started, so we need to account
8124 : * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8125 : * here, must leave that up for the ordered extent completion.
8126 : *
8127 : * This will also unlock the range for incoming
8128 : * btrfs_finish_ordered_io().
8129 : */
8130 0 : if (!inode_evicting)
8131 0 : clear_extent_bit(tree, cur, range_end,
8132 : EXTENT_DELALLOC |
8133 : EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8134 : EXTENT_DEFRAG, &cached_state);
8135 :
8136 0 : spin_lock_irq(&inode->ordered_tree.lock);
8137 0 : set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8138 0 : ordered->truncated_len = min(ordered->truncated_len,
8139 : cur - ordered->file_offset);
8140 0 : spin_unlock_irq(&inode->ordered_tree.lock);
8141 :
8142 : /*
8143 : * If the ordered extent has finished, we're safe to delete all
8144 : * the extent states of the range, otherwise
8145 : * btrfs_finish_ordered_io() will get executed by endio for
8146 : * other pages, so we can't delete extent states.
8147 : */
8148 0 : if (btrfs_dec_test_ordered_pending(inode, &ordered,
8149 : cur, range_end + 1 - cur)) {
8150 0 : btrfs_finish_ordered_io(ordered);
8151 : /*
8152 : * The ordered extent has finished, now we're again
8153 : * safe to delete all extent states of the range.
8154 : */
8155 0 : extra_flags = EXTENT_CLEAR_ALL_BITS;
8156 : }
8157 0 : next:
8158 0 : if (ordered)
8159 0 : btrfs_put_ordered_extent(ordered);
8160 : /*
8161 : * Qgroup reserved space handler
8162 : * Sector(s) here will be either:
8163 : *
8164 : * 1) Already written to disk or bio already finished
8165 : * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8166 : * Qgroup will be handled by its qgroup_record then.
8167 : * btrfs_qgroup_free_data() call will do nothing here.
8168 : *
8169 : * 2) Not written to disk yet
8170 : * Then btrfs_qgroup_free_data() call will clear the
8171 : * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8172 : * reserved data space.
8173 : * Since the IO will never happen for this page.
8174 : */
8175 0 : btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8176 0 : if (!inode_evicting) {
8177 0 : clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8178 : EXTENT_DELALLOC | EXTENT_UPTODATE |
8179 : EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8180 : extra_flags, &cached_state);
8181 : }
8182 0 : cur = range_end + 1;
8183 : }
8184 : /*
8185 : * We have iterated through all ordered extents of the page, the page
8186 : * should not have Ordered (Private2) anymore, or the above iteration
8187 : * did something wrong.
8188 : */
8189 0 : ASSERT(!folio_test_ordered(folio));
8190 0 : btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8191 0 : if (!inode_evicting)
8192 0 : __btrfs_release_folio(folio, GFP_NOFS);
8193 0 : clear_page_extent_mapped(&folio->page);
8194 : }
8195 :
8196 : /*
8197 : * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8198 : * called from a page fault handler when a page is first dirtied. Hence we must
8199 : * be careful to check for EOF conditions here. We set the page up correctly
8200 : * for a written page which means we get ENOSPC checking when writing into
8201 : * holes and correct delalloc and unwritten extent mapping on filesystems that
8202 : * support these features.
8203 : *
8204 : * We are not allowed to take the i_mutex here so we have to play games to
8205 : * protect against truncate races as the page could now be beyond EOF. Because
8206 : * truncate_setsize() writes the inode size before removing pages, once we have
8207 : * the page lock we can determine safely if the page is beyond EOF. If it is not
8208 : * beyond EOF, then the page is guaranteed safe against truncation until we
8209 : * unlock the page.
8210 : */
8211 0 : vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8212 : {
8213 0 : struct page *page = vmf->page;
8214 0 : struct inode *inode = file_inode(vmf->vma->vm_file);
8215 0 : struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8216 0 : struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8217 0 : struct btrfs_ordered_extent *ordered;
8218 0 : struct extent_state *cached_state = NULL;
8219 0 : struct extent_changeset *data_reserved = NULL;
8220 0 : unsigned long zero_start;
8221 0 : loff_t size;
8222 0 : vm_fault_t ret;
8223 0 : int ret2;
8224 0 : int reserved = 0;
8225 0 : u64 reserved_space;
8226 0 : u64 page_start;
8227 0 : u64 page_end;
8228 0 : u64 end;
8229 :
8230 0 : reserved_space = PAGE_SIZE;
8231 :
8232 0 : sb_start_pagefault(inode->i_sb);
8233 0 : page_start = page_offset(page);
8234 0 : page_end = page_start + PAGE_SIZE - 1;
8235 0 : end = page_end;
8236 :
8237 : /*
8238 : * Reserving delalloc space after obtaining the page lock can lead to
8239 : * deadlock. For example, if a dirty page is locked by this function
8240 : * and the call to btrfs_delalloc_reserve_space() ends up triggering
8241 : * dirty page write out, then the btrfs_writepages() function could
8242 : * end up waiting indefinitely to get a lock on the page currently
8243 : * being processed by btrfs_page_mkwrite() function.
8244 : */
8245 0 : ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8246 : page_start, reserved_space);
8247 0 : if (!ret2) {
8248 0 : ret2 = file_update_time(vmf->vma->vm_file);
8249 0 : reserved = 1;
8250 : }
8251 0 : if (ret2) {
8252 0 : ret = vmf_error(ret2);
8253 0 : if (reserved)
8254 0 : goto out;
8255 0 : goto out_noreserve;
8256 : }
8257 :
8258 : ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8259 0 : again:
8260 0 : down_read(&BTRFS_I(inode)->i_mmap_lock);
8261 0 : lock_page(page);
8262 0 : size = i_size_read(inode);
8263 :
8264 0 : if ((page->mapping != inode->i_mapping) ||
8265 0 : (page_start >= size)) {
8266 : /* page got truncated out from underneath us */
8267 0 : goto out_unlock;
8268 : }
8269 0 : wait_on_page_writeback(page);
8270 :
8271 0 : lock_extent(io_tree, page_start, page_end, &cached_state);
8272 0 : ret2 = set_page_extent_mapped(page);
8273 0 : if (ret2 < 0) {
8274 0 : ret = vmf_error(ret2);
8275 0 : unlock_extent(io_tree, page_start, page_end, &cached_state);
8276 0 : goto out_unlock;
8277 : }
8278 :
8279 : /*
8280 : * we can't set the delalloc bits if there are pending ordered
8281 : * extents. Drop our locks and wait for them to finish
8282 : */
8283 0 : ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8284 : PAGE_SIZE);
8285 0 : if (ordered) {
8286 0 : unlock_extent(io_tree, page_start, page_end, &cached_state);
8287 0 : unlock_page(page);
8288 0 : up_read(&BTRFS_I(inode)->i_mmap_lock);
8289 0 : btrfs_start_ordered_extent(ordered);
8290 0 : btrfs_put_ordered_extent(ordered);
8291 0 : goto again;
8292 : }
8293 :
8294 0 : if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8295 0 : reserved_space = round_up(size - page_start,
8296 : fs_info->sectorsize);
8297 0 : if (reserved_space < PAGE_SIZE) {
8298 0 : end = page_start + reserved_space - 1;
8299 0 : btrfs_delalloc_release_space(BTRFS_I(inode),
8300 : data_reserved, page_start,
8301 : PAGE_SIZE - reserved_space, true);
8302 : }
8303 : }
8304 :
8305 : /*
8306 : * page_mkwrite gets called when the page is firstly dirtied after it's
8307 : * faulted in, but write(2) could also dirty a page and set delalloc
8308 : * bits, thus in this case for space account reason, we still need to
8309 : * clear any delalloc bits within this page range since we have to
8310 : * reserve data&meta space before lock_page() (see above comments).
8311 : */
8312 0 : clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8313 : EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8314 : EXTENT_DEFRAG, &cached_state);
8315 :
8316 0 : ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8317 : &cached_state);
8318 0 : if (ret2) {
8319 0 : unlock_extent(io_tree, page_start, page_end, &cached_state);
8320 0 : ret = VM_FAULT_SIGBUS;
8321 0 : goto out_unlock;
8322 : }
8323 :
8324 : /* page is wholly or partially inside EOF */
8325 0 : if (page_start + PAGE_SIZE > size)
8326 0 : zero_start = offset_in_page(size);
8327 : else
8328 : zero_start = PAGE_SIZE;
8329 :
8330 0 : if (zero_start != PAGE_SIZE)
8331 0 : memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8332 :
8333 0 : btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8334 0 : btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8335 0 : btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8336 :
8337 0 : btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8338 :
8339 0 : unlock_extent(io_tree, page_start, page_end, &cached_state);
8340 0 : up_read(&BTRFS_I(inode)->i_mmap_lock);
8341 :
8342 0 : btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8343 0 : sb_end_pagefault(inode->i_sb);
8344 0 : extent_changeset_free(data_reserved);
8345 0 : return VM_FAULT_LOCKED;
8346 :
8347 0 : out_unlock:
8348 0 : unlock_page(page);
8349 0 : up_read(&BTRFS_I(inode)->i_mmap_lock);
8350 0 : out:
8351 0 : btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8352 0 : btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8353 : reserved_space, (ret != 0));
8354 0 : out_noreserve:
8355 0 : sb_end_pagefault(inode->i_sb);
8356 0 : extent_changeset_free(data_reserved);
8357 0 : return ret;
8358 : }
8359 :
8360 0 : static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8361 : {
8362 0 : struct btrfs_truncate_control control = {
8363 : .inode = inode,
8364 : .ino = btrfs_ino(inode),
8365 : .min_type = BTRFS_EXTENT_DATA_KEY,
8366 : .clear_extent_range = true,
8367 : };
8368 0 : struct btrfs_root *root = inode->root;
8369 0 : struct btrfs_fs_info *fs_info = root->fs_info;
8370 0 : struct btrfs_block_rsv *rsv;
8371 0 : int ret;
8372 0 : struct btrfs_trans_handle *trans;
8373 0 : u64 mask = fs_info->sectorsize - 1;
8374 0 : const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8375 :
8376 0 : if (!skip_writeback) {
8377 0 : ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8378 0 : inode->vfs_inode.i_size & (~mask),
8379 : (u64)-1);
8380 0 : if (ret)
8381 : return ret;
8382 : }
8383 :
8384 : /*
8385 : * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8386 : * things going on here:
8387 : *
8388 : * 1) We need to reserve space to update our inode.
8389 : *
8390 : * 2) We need to have something to cache all the space that is going to
8391 : * be free'd up by the truncate operation, but also have some slack
8392 : * space reserved in case it uses space during the truncate (thank you
8393 : * very much snapshotting).
8394 : *
8395 : * And we need these to be separate. The fact is we can use a lot of
8396 : * space doing the truncate, and we have no earthly idea how much space
8397 : * we will use, so we need the truncate reservation to be separate so it
8398 : * doesn't end up using space reserved for updating the inode. We also
8399 : * need to be able to stop the transaction and start a new one, which
8400 : * means we need to be able to update the inode several times, and we
8401 : * have no idea of knowing how many times that will be, so we can't just
8402 : * reserve 1 item for the entirety of the operation, so that has to be
8403 : * done separately as well.
8404 : *
8405 : * So that leaves us with
8406 : *
8407 : * 1) rsv - for the truncate reservation, which we will steal from the
8408 : * transaction reservation.
8409 : * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8410 : * updating the inode.
8411 : */
8412 0 : rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8413 0 : if (!rsv)
8414 : return -ENOMEM;
8415 0 : rsv->size = min_size;
8416 0 : rsv->failfast = true;
8417 :
8418 : /*
8419 : * 1 for the truncate slack space
8420 : * 1 for updating the inode.
8421 : */
8422 0 : trans = btrfs_start_transaction(root, 2);
8423 0 : if (IS_ERR(trans)) {
8424 0 : ret = PTR_ERR(trans);
8425 0 : goto out;
8426 : }
8427 :
8428 : /* Migrate the slack space for the truncate to our reserve */
8429 0 : ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8430 : min_size, false);
8431 : /*
8432 : * We have reserved 2 metadata units when we started the transaction and
8433 : * min_size matches 1 unit, so this should never fail, but if it does,
8434 : * it's not critical we just fail truncation.
8435 : */
8436 0 : if (WARN_ON(ret)) {
8437 0 : btrfs_end_transaction(trans);
8438 0 : goto out;
8439 : }
8440 :
8441 0 : trans->block_rsv = rsv;
8442 :
8443 0 : while (1) {
8444 0 : struct extent_state *cached_state = NULL;
8445 0 : const u64 new_size = inode->vfs_inode.i_size;
8446 0 : const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8447 :
8448 0 : control.new_size = new_size;
8449 0 : lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8450 : /*
8451 : * We want to drop from the next block forward in case this new
8452 : * size is not block aligned since we will be keeping the last
8453 : * block of the extent just the way it is.
8454 : */
8455 0 : btrfs_drop_extent_map_range(inode,
8456 0 : ALIGN(new_size, fs_info->sectorsize),
8457 : (u64)-1, false);
8458 :
8459 0 : ret = btrfs_truncate_inode_items(trans, root, &control);
8460 :
8461 0 : inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8462 0 : btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8463 :
8464 0 : unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8465 :
8466 0 : trans->block_rsv = &fs_info->trans_block_rsv;
8467 0 : if (ret != -ENOSPC && ret != -EAGAIN)
8468 : break;
8469 :
8470 0 : ret = btrfs_update_inode(trans, root, inode);
8471 0 : if (ret)
8472 : break;
8473 :
8474 0 : btrfs_end_transaction(trans);
8475 0 : btrfs_btree_balance_dirty(fs_info);
8476 :
8477 0 : trans = btrfs_start_transaction(root, 2);
8478 0 : if (IS_ERR(trans)) {
8479 0 : ret = PTR_ERR(trans);
8480 0 : trans = NULL;
8481 0 : break;
8482 : }
8483 :
8484 0 : btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8485 0 : ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8486 : rsv, min_size, false);
8487 : /*
8488 : * We have reserved 2 metadata units when we started the
8489 : * transaction and min_size matches 1 unit, so this should never
8490 : * fail, but if it does, it's not critical we just fail truncation.
8491 : */
8492 0 : if (WARN_ON(ret))
8493 : break;
8494 :
8495 0 : trans->block_rsv = rsv;
8496 : }
8497 :
8498 : /*
8499 : * We can't call btrfs_truncate_block inside a trans handle as we could
8500 : * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8501 : * know we've truncated everything except the last little bit, and can
8502 : * do btrfs_truncate_block and then update the disk_i_size.
8503 : */
8504 0 : if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8505 0 : btrfs_end_transaction(trans);
8506 0 : btrfs_btree_balance_dirty(fs_info);
8507 :
8508 0 : ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8509 0 : if (ret)
8510 0 : goto out;
8511 0 : trans = btrfs_start_transaction(root, 1);
8512 0 : if (IS_ERR(trans)) {
8513 0 : ret = PTR_ERR(trans);
8514 0 : goto out;
8515 : }
8516 0 : btrfs_inode_safe_disk_i_size_write(inode, 0);
8517 : }
8518 :
8519 0 : if (trans) {
8520 0 : int ret2;
8521 :
8522 0 : trans->block_rsv = &fs_info->trans_block_rsv;
8523 0 : ret2 = btrfs_update_inode(trans, root, inode);
8524 0 : if (ret2 && !ret)
8525 0 : ret = ret2;
8526 :
8527 0 : ret2 = btrfs_end_transaction(trans);
8528 0 : if (ret2 && !ret)
8529 0 : ret = ret2;
8530 0 : btrfs_btree_balance_dirty(fs_info);
8531 : }
8532 0 : out:
8533 0 : btrfs_free_block_rsv(fs_info, rsv);
8534 : /*
8535 : * So if we truncate and then write and fsync we normally would just
8536 : * write the extents that changed, which is a problem if we need to
8537 : * first truncate that entire inode. So set this flag so we write out
8538 : * all of the extents in the inode to the sync log so we're completely
8539 : * safe.
8540 : *
8541 : * If no extents were dropped or trimmed we don't need to force the next
8542 : * fsync to truncate all the inode's items from the log and re-log them
8543 : * all. This means the truncate operation did not change the file size,
8544 : * or changed it to a smaller size but there was only an implicit hole
8545 : * between the old i_size and the new i_size, and there were no prealloc
8546 : * extents beyond i_size to drop.
8547 : */
8548 0 : if (control.extents_found > 0)
8549 0 : btrfs_set_inode_full_sync(inode);
8550 :
8551 : return ret;
8552 : }
8553 :
8554 0 : struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8555 : struct inode *dir)
8556 : {
8557 0 : struct inode *inode;
8558 :
8559 0 : inode = new_inode(dir->i_sb);
8560 0 : if (inode) {
8561 : /*
8562 : * Subvolumes don't inherit the sgid bit or the parent's gid if
8563 : * the parent's sgid bit is set. This is probably a bug.
8564 : */
8565 0 : inode_init_owner(idmap, inode, NULL,
8566 0 : S_IFDIR | (~current_umask() & S_IRWXUGO));
8567 0 : inode->i_op = &btrfs_dir_inode_operations;
8568 0 : inode->i_fop = &btrfs_dir_file_operations;
8569 : }
8570 0 : return inode;
8571 : }
8572 :
8573 0 : struct inode *btrfs_alloc_inode(struct super_block *sb)
8574 : {
8575 0 : struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8576 0 : struct btrfs_inode *ei;
8577 0 : struct inode *inode;
8578 :
8579 0 : ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8580 0 : if (!ei)
8581 : return NULL;
8582 :
8583 0 : ei->root = NULL;
8584 0 : ei->generation = 0;
8585 0 : ei->last_trans = 0;
8586 0 : ei->last_sub_trans = 0;
8587 0 : ei->logged_trans = 0;
8588 0 : ei->delalloc_bytes = 0;
8589 0 : ei->new_delalloc_bytes = 0;
8590 0 : ei->defrag_bytes = 0;
8591 0 : ei->disk_i_size = 0;
8592 0 : ei->flags = 0;
8593 0 : ei->ro_flags = 0;
8594 0 : ei->csum_bytes = 0;
8595 0 : ei->index_cnt = (u64)-1;
8596 0 : ei->dir_index = 0;
8597 0 : ei->last_unlink_trans = 0;
8598 0 : ei->last_reflink_trans = 0;
8599 0 : ei->last_log_commit = 0;
8600 :
8601 0 : spin_lock_init(&ei->lock);
8602 0 : ei->outstanding_extents = 0;
8603 0 : if (sb->s_magic != BTRFS_TEST_MAGIC)
8604 0 : btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8605 : BTRFS_BLOCK_RSV_DELALLOC);
8606 0 : ei->runtime_flags = 0;
8607 0 : ei->prop_compress = BTRFS_COMPRESS_NONE;
8608 0 : ei->defrag_compress = BTRFS_COMPRESS_NONE;
8609 :
8610 0 : ei->delayed_node = NULL;
8611 :
8612 0 : ei->i_otime.tv_sec = 0;
8613 0 : ei->i_otime.tv_nsec = 0;
8614 :
8615 0 : inode = &ei->vfs_inode;
8616 0 : extent_map_tree_init(&ei->extent_tree);
8617 0 : extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8618 0 : ei->io_tree.inode = ei;
8619 0 : extent_io_tree_init(fs_info, &ei->file_extent_tree,
8620 : IO_TREE_INODE_FILE_EXTENT);
8621 0 : mutex_init(&ei->log_mutex);
8622 0 : btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8623 0 : INIT_LIST_HEAD(&ei->delalloc_inodes);
8624 0 : INIT_LIST_HEAD(&ei->delayed_iput);
8625 0 : RB_CLEAR_NODE(&ei->rb_node);
8626 0 : init_rwsem(&ei->i_mmap_lock);
8627 :
8628 0 : return inode;
8629 : }
8630 :
8631 : #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8632 : void btrfs_test_destroy_inode(struct inode *inode)
8633 : {
8634 : btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8635 : kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8636 : }
8637 : #endif
8638 :
8639 0 : void btrfs_free_inode(struct inode *inode)
8640 : {
8641 0 : kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8642 0 : }
8643 :
8644 0 : void btrfs_destroy_inode(struct inode *vfs_inode)
8645 : {
8646 0 : struct btrfs_ordered_extent *ordered;
8647 0 : struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8648 0 : struct btrfs_root *root = inode->root;
8649 0 : bool freespace_inode;
8650 :
8651 0 : WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8652 0 : WARN_ON(vfs_inode->i_data.nrpages);
8653 0 : WARN_ON(inode->block_rsv.reserved);
8654 0 : WARN_ON(inode->block_rsv.size);
8655 0 : WARN_ON(inode->outstanding_extents);
8656 0 : if (!S_ISDIR(vfs_inode->i_mode)) {
8657 0 : WARN_ON(inode->delalloc_bytes);
8658 0 : WARN_ON(inode->new_delalloc_bytes);
8659 : }
8660 0 : WARN_ON(inode->csum_bytes);
8661 0 : WARN_ON(inode->defrag_bytes);
8662 :
8663 : /*
8664 : * This can happen where we create an inode, but somebody else also
8665 : * created the same inode and we need to destroy the one we already
8666 : * created.
8667 : */
8668 0 : if (!root)
8669 : return;
8670 :
8671 : /*
8672 : * If this is a free space inode do not take the ordered extents lockdep
8673 : * map.
8674 : */
8675 0 : freespace_inode = btrfs_is_free_space_inode(inode);
8676 :
8677 0 : while (1) {
8678 0 : ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8679 0 : if (!ordered)
8680 : break;
8681 : else {
8682 0 : btrfs_err(root->fs_info,
8683 : "found ordered extent %llu %llu on inode cleanup",
8684 : ordered->file_offset, ordered->num_bytes);
8685 :
8686 0 : if (!freespace_inode)
8687 0 : btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8688 :
8689 0 : btrfs_remove_ordered_extent(inode, ordered);
8690 0 : btrfs_put_ordered_extent(ordered);
8691 0 : btrfs_put_ordered_extent(ordered);
8692 : }
8693 : }
8694 0 : btrfs_qgroup_check_reserved_leak(inode);
8695 0 : inode_tree_del(inode);
8696 0 : btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8697 0 : btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8698 0 : btrfs_put_root(inode->root);
8699 : }
8700 :
8701 0 : int btrfs_drop_inode(struct inode *inode)
8702 : {
8703 0 : struct btrfs_root *root = BTRFS_I(inode)->root;
8704 :
8705 0 : if (root == NULL)
8706 : return 1;
8707 :
8708 : /* the snap/subvol tree is on deleting */
8709 0 : if (btrfs_root_refs(&root->root_item) == 0)
8710 : return 1;
8711 : else
8712 0 : return generic_drop_inode(inode);
8713 : }
8714 :
8715 0 : static void init_once(void *foo)
8716 : {
8717 0 : struct btrfs_inode *ei = foo;
8718 :
8719 0 : inode_init_once(&ei->vfs_inode);
8720 0 : }
8721 :
8722 0 : void __cold btrfs_destroy_cachep(void)
8723 : {
8724 : /*
8725 : * Make sure all delayed rcu free inodes are flushed before we
8726 : * destroy cache.
8727 : */
8728 0 : rcu_barrier();
8729 0 : bioset_exit(&btrfs_dio_bioset);
8730 0 : kmem_cache_destroy(btrfs_inode_cachep);
8731 0 : }
8732 :
8733 2 : int __init btrfs_init_cachep(void)
8734 : {
8735 2 : btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8736 : sizeof(struct btrfs_inode), 0,
8737 : SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8738 : init_once);
8739 2 : if (!btrfs_inode_cachep)
8740 0 : goto fail;
8741 :
8742 2 : if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8743 : offsetof(struct btrfs_dio_private, bbio.bio),
8744 : BIOSET_NEED_BVECS))
8745 0 : goto fail;
8746 :
8747 : return 0;
8748 0 : fail:
8749 0 : btrfs_destroy_cachep();
8750 0 : return -ENOMEM;
8751 : }
8752 :
8753 0 : static int btrfs_getattr(struct mnt_idmap *idmap,
8754 : const struct path *path, struct kstat *stat,
8755 : u32 request_mask, unsigned int flags)
8756 : {
8757 0 : u64 delalloc_bytes;
8758 0 : u64 inode_bytes;
8759 0 : struct inode *inode = d_inode(path->dentry);
8760 0 : u32 blocksize = inode->i_sb->s_blocksize;
8761 0 : u32 bi_flags = BTRFS_I(inode)->flags;
8762 0 : u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8763 :
8764 0 : stat->result_mask |= STATX_BTIME;
8765 0 : stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8766 0 : stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8767 0 : if (bi_flags & BTRFS_INODE_APPEND)
8768 0 : stat->attributes |= STATX_ATTR_APPEND;
8769 0 : if (bi_flags & BTRFS_INODE_COMPRESS)
8770 0 : stat->attributes |= STATX_ATTR_COMPRESSED;
8771 0 : if (bi_flags & BTRFS_INODE_IMMUTABLE)
8772 0 : stat->attributes |= STATX_ATTR_IMMUTABLE;
8773 0 : if (bi_flags & BTRFS_INODE_NODUMP)
8774 0 : stat->attributes |= STATX_ATTR_NODUMP;
8775 0 : if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8776 0 : stat->attributes |= STATX_ATTR_VERITY;
8777 :
8778 0 : stat->attributes_mask |= (STATX_ATTR_APPEND |
8779 : STATX_ATTR_COMPRESSED |
8780 : STATX_ATTR_IMMUTABLE |
8781 : STATX_ATTR_NODUMP);
8782 :
8783 0 : generic_fillattr(idmap, inode, stat);
8784 0 : stat->dev = BTRFS_I(inode)->root->anon_dev;
8785 :
8786 0 : spin_lock(&BTRFS_I(inode)->lock);
8787 0 : delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8788 0 : inode_bytes = inode_get_bytes(inode);
8789 0 : spin_unlock(&BTRFS_I(inode)->lock);
8790 0 : stat->blocks = (ALIGN(inode_bytes, blocksize) +
8791 0 : ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8792 0 : return 0;
8793 : }
8794 :
8795 0 : static int btrfs_rename_exchange(struct inode *old_dir,
8796 : struct dentry *old_dentry,
8797 : struct inode *new_dir,
8798 : struct dentry *new_dentry)
8799 : {
8800 0 : struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8801 0 : struct btrfs_trans_handle *trans;
8802 0 : unsigned int trans_num_items;
8803 0 : struct btrfs_root *root = BTRFS_I(old_dir)->root;
8804 0 : struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8805 0 : struct inode *new_inode = new_dentry->d_inode;
8806 0 : struct inode *old_inode = old_dentry->d_inode;
8807 0 : struct timespec64 ctime = current_time(old_inode);
8808 0 : struct btrfs_rename_ctx old_rename_ctx;
8809 0 : struct btrfs_rename_ctx new_rename_ctx;
8810 0 : u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8811 0 : u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8812 0 : u64 old_idx = 0;
8813 0 : u64 new_idx = 0;
8814 0 : int ret;
8815 0 : int ret2;
8816 0 : bool need_abort = false;
8817 0 : struct fscrypt_name old_fname, new_fname;
8818 0 : struct fscrypt_str *old_name, *new_name;
8819 :
8820 : /*
8821 : * For non-subvolumes allow exchange only within one subvolume, in the
8822 : * same inode namespace. Two subvolumes (represented as directory) can
8823 : * be exchanged as they're a logical link and have a fixed inode number.
8824 : */
8825 0 : if (root != dest &&
8826 0 : (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8827 0 : new_ino != BTRFS_FIRST_FREE_OBJECTID))
8828 : return -EXDEV;
8829 :
8830 0 : ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8831 0 : if (ret)
8832 : return ret;
8833 :
8834 0 : ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8835 0 : if (ret) {
8836 : fscrypt_free_filename(&old_fname);
8837 : return ret;
8838 : }
8839 :
8840 0 : old_name = &old_fname.disk_name;
8841 0 : new_name = &new_fname.disk_name;
8842 :
8843 : /* close the race window with snapshot create/destroy ioctl */
8844 0 : if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8845 0 : new_ino == BTRFS_FIRST_FREE_OBJECTID)
8846 0 : down_read(&fs_info->subvol_sem);
8847 :
8848 : /*
8849 : * For each inode:
8850 : * 1 to remove old dir item
8851 : * 1 to remove old dir index
8852 : * 1 to add new dir item
8853 : * 1 to add new dir index
8854 : * 1 to update parent inode
8855 : *
8856 : * If the parents are the same, we only need to account for one
8857 : */
8858 0 : trans_num_items = (old_dir == new_dir ? 9 : 10);
8859 0 : if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8860 : /*
8861 : * 1 to remove old root ref
8862 : * 1 to remove old root backref
8863 : * 1 to add new root ref
8864 : * 1 to add new root backref
8865 : */
8866 0 : trans_num_items += 4;
8867 : } else {
8868 : /*
8869 : * 1 to update inode item
8870 : * 1 to remove old inode ref
8871 : * 1 to add new inode ref
8872 : */
8873 0 : trans_num_items += 3;
8874 : }
8875 0 : if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8876 0 : trans_num_items += 4;
8877 : else
8878 0 : trans_num_items += 3;
8879 0 : trans = btrfs_start_transaction(root, trans_num_items);
8880 0 : if (IS_ERR(trans)) {
8881 0 : ret = PTR_ERR(trans);
8882 0 : goto out_notrans;
8883 : }
8884 :
8885 0 : if (dest != root) {
8886 0 : ret = btrfs_record_root_in_trans(trans, dest);
8887 0 : if (ret)
8888 0 : goto out_fail;
8889 : }
8890 :
8891 : /*
8892 : * We need to find a free sequence number both in the source and
8893 : * in the destination directory for the exchange.
8894 : */
8895 0 : ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8896 0 : if (ret)
8897 0 : goto out_fail;
8898 0 : ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8899 0 : if (ret)
8900 0 : goto out_fail;
8901 :
8902 0 : BTRFS_I(old_inode)->dir_index = 0ULL;
8903 0 : BTRFS_I(new_inode)->dir_index = 0ULL;
8904 :
8905 : /* Reference for the source. */
8906 0 : if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8907 : /* force full log commit if subvolume involved. */
8908 0 : btrfs_set_log_full_commit(trans);
8909 : } else {
8910 0 : ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8911 : btrfs_ino(BTRFS_I(new_dir)),
8912 : old_idx);
8913 0 : if (ret)
8914 0 : goto out_fail;
8915 : need_abort = true;
8916 : }
8917 :
8918 : /* And now for the dest. */
8919 0 : if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8920 : /* force full log commit if subvolume involved. */
8921 0 : btrfs_set_log_full_commit(trans);
8922 : } else {
8923 0 : ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8924 : btrfs_ino(BTRFS_I(old_dir)),
8925 : new_idx);
8926 0 : if (ret) {
8927 0 : if (need_abort)
8928 0 : btrfs_abort_transaction(trans, ret);
8929 0 : goto out_fail;
8930 : }
8931 : }
8932 :
8933 : /* Update inode version and ctime/mtime. */
8934 0 : inode_inc_iversion(old_dir);
8935 0 : inode_inc_iversion(new_dir);
8936 0 : inode_inc_iversion(old_inode);
8937 0 : inode_inc_iversion(new_inode);
8938 0 : old_dir->i_mtime = ctime;
8939 0 : old_dir->i_ctime = ctime;
8940 0 : new_dir->i_mtime = ctime;
8941 0 : new_dir->i_ctime = ctime;
8942 0 : old_inode->i_ctime = ctime;
8943 0 : new_inode->i_ctime = ctime;
8944 :
8945 0 : if (old_dentry->d_parent != new_dentry->d_parent) {
8946 0 : btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8947 : BTRFS_I(old_inode), true);
8948 0 : btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8949 : BTRFS_I(new_inode), true);
8950 : }
8951 :
8952 : /* src is a subvolume */
8953 0 : if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8954 0 : ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8955 : } else { /* src is an inode */
8956 0 : ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8957 0 : BTRFS_I(old_dentry->d_inode),
8958 : old_name, &old_rename_ctx);
8959 0 : if (!ret)
8960 0 : ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
8961 : }
8962 0 : if (ret) {
8963 0 : btrfs_abort_transaction(trans, ret);
8964 0 : goto out_fail;
8965 : }
8966 :
8967 : /* dest is a subvolume */
8968 0 : if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8969 0 : ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8970 : } else { /* dest is an inode */
8971 0 : ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8972 0 : BTRFS_I(new_dentry->d_inode),
8973 : new_name, &new_rename_ctx);
8974 0 : if (!ret)
8975 0 : ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
8976 : }
8977 0 : if (ret) {
8978 0 : btrfs_abort_transaction(trans, ret);
8979 0 : goto out_fail;
8980 : }
8981 :
8982 0 : ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8983 : new_name, 0, old_idx);
8984 0 : if (ret) {
8985 0 : btrfs_abort_transaction(trans, ret);
8986 0 : goto out_fail;
8987 : }
8988 :
8989 0 : ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8990 : old_name, 0, new_idx);
8991 0 : if (ret) {
8992 0 : btrfs_abort_transaction(trans, ret);
8993 0 : goto out_fail;
8994 : }
8995 :
8996 0 : if (old_inode->i_nlink == 1)
8997 0 : BTRFS_I(old_inode)->dir_index = old_idx;
8998 0 : if (new_inode->i_nlink == 1)
8999 0 : BTRFS_I(new_inode)->dir_index = new_idx;
9000 :
9001 : /*
9002 : * Now pin the logs of the roots. We do it to ensure that no other task
9003 : * can sync the logs while we are in progress with the rename, because
9004 : * that could result in an inconsistency in case any of the inodes that
9005 : * are part of this rename operation were logged before.
9006 : */
9007 0 : if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9008 0 : btrfs_pin_log_trans(root);
9009 0 : if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9010 0 : btrfs_pin_log_trans(dest);
9011 :
9012 : /* Do the log updates for all inodes. */
9013 0 : if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9014 0 : btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9015 : old_rename_ctx.index, new_dentry->d_parent);
9016 0 : if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9017 0 : btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9018 : new_rename_ctx.index, old_dentry->d_parent);
9019 :
9020 : /* Now unpin the logs. */
9021 0 : if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9022 0 : btrfs_end_log_trans(root);
9023 0 : if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9024 0 : btrfs_end_log_trans(dest);
9025 0 : out_fail:
9026 0 : ret2 = btrfs_end_transaction(trans);
9027 0 : ret = ret ? ret : ret2;
9028 0 : out_notrans:
9029 0 : if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9030 : old_ino == BTRFS_FIRST_FREE_OBJECTID)
9031 0 : up_read(&fs_info->subvol_sem);
9032 :
9033 : fscrypt_free_filename(&new_fname);
9034 : fscrypt_free_filename(&old_fname);
9035 : return ret;
9036 : }
9037 :
9038 0 : static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
9039 : struct inode *dir)
9040 : {
9041 0 : struct inode *inode;
9042 :
9043 0 : inode = new_inode(dir->i_sb);
9044 0 : if (inode) {
9045 0 : inode_init_owner(idmap, inode, dir,
9046 : S_IFCHR | WHITEOUT_MODE);
9047 0 : inode->i_op = &btrfs_special_inode_operations;
9048 0 : init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9049 : }
9050 0 : return inode;
9051 : }
9052 :
9053 0 : static int btrfs_rename(struct mnt_idmap *idmap,
9054 : struct inode *old_dir, struct dentry *old_dentry,
9055 : struct inode *new_dir, struct dentry *new_dentry,
9056 : unsigned int flags)
9057 : {
9058 0 : struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9059 0 : struct btrfs_new_inode_args whiteout_args = {
9060 : .dir = old_dir,
9061 : .dentry = old_dentry,
9062 : };
9063 0 : struct btrfs_trans_handle *trans;
9064 0 : unsigned int trans_num_items;
9065 0 : struct btrfs_root *root = BTRFS_I(old_dir)->root;
9066 0 : struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9067 0 : struct inode *new_inode = d_inode(new_dentry);
9068 0 : struct inode *old_inode = d_inode(old_dentry);
9069 0 : struct btrfs_rename_ctx rename_ctx;
9070 0 : u64 index = 0;
9071 0 : int ret;
9072 0 : int ret2;
9073 0 : u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9074 0 : struct fscrypt_name old_fname, new_fname;
9075 :
9076 0 : if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9077 : return -EPERM;
9078 :
9079 : /* we only allow rename subvolume link between subvolumes */
9080 0 : if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9081 : return -EXDEV;
9082 :
9083 0 : if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9084 0 : (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9085 : return -ENOTEMPTY;
9086 :
9087 0 : if (S_ISDIR(old_inode->i_mode) && new_inode &&
9088 0 : new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9089 : return -ENOTEMPTY;
9090 :
9091 0 : ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
9092 0 : if (ret)
9093 : return ret;
9094 :
9095 0 : ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
9096 0 : if (ret) {
9097 : fscrypt_free_filename(&old_fname);
9098 : return ret;
9099 : }
9100 :
9101 : /* check for collisions, even if the name isn't there */
9102 0 : ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
9103 0 : if (ret) {
9104 0 : if (ret == -EEXIST) {
9105 : /* we shouldn't get
9106 : * eexist without a new_inode */
9107 0 : if (WARN_ON(!new_inode)) {
9108 0 : goto out_fscrypt_names;
9109 : }
9110 : } else {
9111 : /* maybe -EOVERFLOW */
9112 0 : goto out_fscrypt_names;
9113 : }
9114 : }
9115 0 : ret = 0;
9116 :
9117 : /*
9118 : * we're using rename to replace one file with another. Start IO on it
9119 : * now so we don't add too much work to the end of the transaction
9120 : */
9121 0 : if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9122 0 : filemap_flush(old_inode->i_mapping);
9123 :
9124 0 : if (flags & RENAME_WHITEOUT) {
9125 0 : whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9126 0 : if (!whiteout_args.inode) {
9127 0 : ret = -ENOMEM;
9128 0 : goto out_fscrypt_names;
9129 : }
9130 0 : ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9131 0 : if (ret)
9132 0 : goto out_whiteout_inode;
9133 : } else {
9134 : /* 1 to update the old parent inode. */
9135 0 : trans_num_items = 1;
9136 : }
9137 :
9138 0 : if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9139 : /* Close the race window with snapshot create/destroy ioctl */
9140 0 : down_read(&fs_info->subvol_sem);
9141 : /*
9142 : * 1 to remove old root ref
9143 : * 1 to remove old root backref
9144 : * 1 to add new root ref
9145 : * 1 to add new root backref
9146 : */
9147 0 : trans_num_items += 4;
9148 : } else {
9149 : /*
9150 : * 1 to update inode
9151 : * 1 to remove old inode ref
9152 : * 1 to add new inode ref
9153 : */
9154 0 : trans_num_items += 3;
9155 : }
9156 : /*
9157 : * 1 to remove old dir item
9158 : * 1 to remove old dir index
9159 : * 1 to add new dir item
9160 : * 1 to add new dir index
9161 : */
9162 0 : trans_num_items += 4;
9163 : /* 1 to update new parent inode if it's not the same as the old parent */
9164 0 : if (new_dir != old_dir)
9165 0 : trans_num_items++;
9166 0 : if (new_inode) {
9167 : /*
9168 : * 1 to update inode
9169 : * 1 to remove inode ref
9170 : * 1 to remove dir item
9171 : * 1 to remove dir index
9172 : * 1 to possibly add orphan item
9173 : */
9174 0 : trans_num_items += 5;
9175 : }
9176 0 : trans = btrfs_start_transaction(root, trans_num_items);
9177 0 : if (IS_ERR(trans)) {
9178 0 : ret = PTR_ERR(trans);
9179 0 : goto out_notrans;
9180 : }
9181 :
9182 0 : if (dest != root) {
9183 0 : ret = btrfs_record_root_in_trans(trans, dest);
9184 0 : if (ret)
9185 0 : goto out_fail;
9186 : }
9187 :
9188 0 : ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9189 0 : if (ret)
9190 0 : goto out_fail;
9191 :
9192 0 : BTRFS_I(old_inode)->dir_index = 0ULL;
9193 0 : if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9194 : /* force full log commit if subvolume involved. */
9195 0 : btrfs_set_log_full_commit(trans);
9196 : } else {
9197 0 : ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9198 : old_ino, btrfs_ino(BTRFS_I(new_dir)),
9199 : index);
9200 0 : if (ret)
9201 0 : goto out_fail;
9202 : }
9203 :
9204 0 : inode_inc_iversion(old_dir);
9205 0 : inode_inc_iversion(new_dir);
9206 0 : inode_inc_iversion(old_inode);
9207 0 : old_dir->i_mtime = current_time(old_dir);
9208 0 : old_dir->i_ctime = old_dir->i_mtime;
9209 0 : new_dir->i_mtime = old_dir->i_mtime;
9210 0 : new_dir->i_ctime = old_dir->i_mtime;
9211 0 : old_inode->i_ctime = old_dir->i_mtime;
9212 :
9213 0 : if (old_dentry->d_parent != new_dentry->d_parent)
9214 0 : btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9215 : BTRFS_I(old_inode), true);
9216 :
9217 0 : if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9218 0 : ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9219 : } else {
9220 0 : ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9221 : BTRFS_I(d_inode(old_dentry)),
9222 : &old_fname.disk_name, &rename_ctx);
9223 0 : if (!ret)
9224 0 : ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9225 : }
9226 0 : if (ret) {
9227 0 : btrfs_abort_transaction(trans, ret);
9228 0 : goto out_fail;
9229 : }
9230 :
9231 0 : if (new_inode) {
9232 0 : inode_inc_iversion(new_inode);
9233 0 : new_inode->i_ctime = current_time(new_inode);
9234 0 : if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9235 : BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9236 0 : ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9237 0 : BUG_ON(new_inode->i_nlink == 0);
9238 : } else {
9239 0 : ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9240 : BTRFS_I(d_inode(new_dentry)),
9241 : &new_fname.disk_name);
9242 : }
9243 0 : if (!ret && new_inode->i_nlink == 0)
9244 0 : ret = btrfs_orphan_add(trans,
9245 : BTRFS_I(d_inode(new_dentry)));
9246 0 : if (ret) {
9247 0 : btrfs_abort_transaction(trans, ret);
9248 0 : goto out_fail;
9249 : }
9250 : }
9251 :
9252 0 : ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9253 : &new_fname.disk_name, 0, index);
9254 0 : if (ret) {
9255 0 : btrfs_abort_transaction(trans, ret);
9256 0 : goto out_fail;
9257 : }
9258 :
9259 0 : if (old_inode->i_nlink == 1)
9260 0 : BTRFS_I(old_inode)->dir_index = index;
9261 :
9262 0 : if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9263 0 : btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9264 : rename_ctx.index, new_dentry->d_parent);
9265 :
9266 0 : if (flags & RENAME_WHITEOUT) {
9267 0 : ret = btrfs_create_new_inode(trans, &whiteout_args);
9268 0 : if (ret) {
9269 0 : btrfs_abort_transaction(trans, ret);
9270 0 : goto out_fail;
9271 : } else {
9272 0 : unlock_new_inode(whiteout_args.inode);
9273 0 : iput(whiteout_args.inode);
9274 0 : whiteout_args.inode = NULL;
9275 : }
9276 : }
9277 0 : out_fail:
9278 0 : ret2 = btrfs_end_transaction(trans);
9279 0 : ret = ret ? ret : ret2;
9280 0 : out_notrans:
9281 0 : if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9282 0 : up_read(&fs_info->subvol_sem);
9283 0 : if (flags & RENAME_WHITEOUT)
9284 0 : btrfs_new_inode_args_destroy(&whiteout_args);
9285 0 : out_whiteout_inode:
9286 0 : if (flags & RENAME_WHITEOUT)
9287 0 : iput(whiteout_args.inode);
9288 0 : out_fscrypt_names:
9289 : fscrypt_free_filename(&old_fname);
9290 : fscrypt_free_filename(&new_fname);
9291 : return ret;
9292 : }
9293 :
9294 0 : static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9295 : struct dentry *old_dentry, struct inode *new_dir,
9296 : struct dentry *new_dentry, unsigned int flags)
9297 : {
9298 0 : int ret;
9299 :
9300 0 : if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9301 : return -EINVAL;
9302 :
9303 0 : if (flags & RENAME_EXCHANGE)
9304 0 : ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9305 : new_dentry);
9306 : else
9307 0 : ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9308 : new_dentry, flags);
9309 :
9310 0 : btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9311 :
9312 0 : return ret;
9313 : }
9314 :
9315 : struct btrfs_delalloc_work {
9316 : struct inode *inode;
9317 : struct completion completion;
9318 : struct list_head list;
9319 : struct btrfs_work work;
9320 : };
9321 :
9322 0 : static void btrfs_run_delalloc_work(struct btrfs_work *work)
9323 : {
9324 0 : struct btrfs_delalloc_work *delalloc_work;
9325 0 : struct inode *inode;
9326 :
9327 0 : delalloc_work = container_of(work, struct btrfs_delalloc_work,
9328 : work);
9329 0 : inode = delalloc_work->inode;
9330 0 : filemap_flush(inode->i_mapping);
9331 0 : if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9332 : &BTRFS_I(inode)->runtime_flags))
9333 0 : filemap_flush(inode->i_mapping);
9334 :
9335 0 : iput(inode);
9336 0 : complete(&delalloc_work->completion);
9337 0 : }
9338 :
9339 0 : static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9340 : {
9341 0 : struct btrfs_delalloc_work *work;
9342 :
9343 0 : work = kmalloc(sizeof(*work), GFP_NOFS);
9344 0 : if (!work)
9345 : return NULL;
9346 :
9347 0 : init_completion(&work->completion);
9348 0 : INIT_LIST_HEAD(&work->list);
9349 0 : work->inode = inode;
9350 0 : btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9351 :
9352 0 : return work;
9353 : }
9354 :
9355 : /*
9356 : * some fairly slow code that needs optimization. This walks the list
9357 : * of all the inodes with pending delalloc and forces them to disk.
9358 : */
9359 0 : static int start_delalloc_inodes(struct btrfs_root *root,
9360 : struct writeback_control *wbc, bool snapshot,
9361 : bool in_reclaim_context)
9362 : {
9363 0 : struct btrfs_inode *binode;
9364 0 : struct inode *inode;
9365 0 : struct btrfs_delalloc_work *work, *next;
9366 0 : struct list_head works;
9367 0 : struct list_head splice;
9368 0 : int ret = 0;
9369 0 : bool full_flush = wbc->nr_to_write == LONG_MAX;
9370 :
9371 0 : INIT_LIST_HEAD(&works);
9372 0 : INIT_LIST_HEAD(&splice);
9373 :
9374 0 : mutex_lock(&root->delalloc_mutex);
9375 0 : spin_lock(&root->delalloc_lock);
9376 0 : list_splice_init(&root->delalloc_inodes, &splice);
9377 0 : while (!list_empty(&splice)) {
9378 0 : binode = list_entry(splice.next, struct btrfs_inode,
9379 : delalloc_inodes);
9380 :
9381 0 : list_move_tail(&binode->delalloc_inodes,
9382 : &root->delalloc_inodes);
9383 :
9384 0 : if (in_reclaim_context &&
9385 0 : test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9386 0 : continue;
9387 :
9388 0 : inode = igrab(&binode->vfs_inode);
9389 0 : if (!inode) {
9390 0 : cond_resched_lock(&root->delalloc_lock);
9391 0 : continue;
9392 : }
9393 0 : spin_unlock(&root->delalloc_lock);
9394 :
9395 0 : if (snapshot)
9396 0 : set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9397 0 : &binode->runtime_flags);
9398 0 : if (full_flush) {
9399 0 : work = btrfs_alloc_delalloc_work(inode);
9400 0 : if (!work) {
9401 0 : iput(inode);
9402 0 : ret = -ENOMEM;
9403 0 : goto out;
9404 : }
9405 0 : list_add_tail(&work->list, &works);
9406 0 : btrfs_queue_work(root->fs_info->flush_workers,
9407 : &work->work);
9408 : } else {
9409 0 : ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9410 0 : btrfs_add_delayed_iput(BTRFS_I(inode));
9411 0 : if (ret || wbc->nr_to_write <= 0)
9412 0 : goto out;
9413 : }
9414 0 : cond_resched();
9415 0 : spin_lock(&root->delalloc_lock);
9416 : }
9417 0 : spin_unlock(&root->delalloc_lock);
9418 :
9419 0 : out:
9420 0 : list_for_each_entry_safe(work, next, &works, list) {
9421 0 : list_del_init(&work->list);
9422 0 : wait_for_completion(&work->completion);
9423 0 : kfree(work);
9424 : }
9425 :
9426 0 : if (!list_empty(&splice)) {
9427 0 : spin_lock(&root->delalloc_lock);
9428 0 : list_splice_tail(&splice, &root->delalloc_inodes);
9429 0 : spin_unlock(&root->delalloc_lock);
9430 : }
9431 0 : mutex_unlock(&root->delalloc_mutex);
9432 0 : return ret;
9433 : }
9434 :
9435 0 : int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9436 : {
9437 0 : struct writeback_control wbc = {
9438 : .nr_to_write = LONG_MAX,
9439 : .sync_mode = WB_SYNC_NONE,
9440 : .range_start = 0,
9441 : .range_end = LLONG_MAX,
9442 : };
9443 0 : struct btrfs_fs_info *fs_info = root->fs_info;
9444 :
9445 0 : if (BTRFS_FS_ERROR(fs_info))
9446 : return -EROFS;
9447 :
9448 0 : return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9449 : }
9450 :
9451 0 : int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9452 : bool in_reclaim_context)
9453 : {
9454 0 : struct writeback_control wbc = {
9455 : .nr_to_write = nr,
9456 : .sync_mode = WB_SYNC_NONE,
9457 : .range_start = 0,
9458 : .range_end = LLONG_MAX,
9459 : };
9460 0 : struct btrfs_root *root;
9461 0 : struct list_head splice;
9462 0 : int ret;
9463 :
9464 0 : if (BTRFS_FS_ERROR(fs_info))
9465 : return -EROFS;
9466 :
9467 0 : INIT_LIST_HEAD(&splice);
9468 :
9469 0 : mutex_lock(&fs_info->delalloc_root_mutex);
9470 0 : spin_lock(&fs_info->delalloc_root_lock);
9471 0 : list_splice_init(&fs_info->delalloc_roots, &splice);
9472 0 : while (!list_empty(&splice)) {
9473 : /*
9474 : * Reset nr_to_write here so we know that we're doing a full
9475 : * flush.
9476 : */
9477 0 : if (nr == LONG_MAX)
9478 0 : wbc.nr_to_write = LONG_MAX;
9479 :
9480 0 : root = list_first_entry(&splice, struct btrfs_root,
9481 : delalloc_root);
9482 0 : root = btrfs_grab_root(root);
9483 0 : BUG_ON(!root);
9484 0 : list_move_tail(&root->delalloc_root,
9485 : &fs_info->delalloc_roots);
9486 0 : spin_unlock(&fs_info->delalloc_root_lock);
9487 :
9488 0 : ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9489 0 : btrfs_put_root(root);
9490 0 : if (ret < 0 || wbc.nr_to_write <= 0)
9491 0 : goto out;
9492 0 : spin_lock(&fs_info->delalloc_root_lock);
9493 : }
9494 0 : spin_unlock(&fs_info->delalloc_root_lock);
9495 :
9496 0 : ret = 0;
9497 0 : out:
9498 0 : if (!list_empty(&splice)) {
9499 0 : spin_lock(&fs_info->delalloc_root_lock);
9500 0 : list_splice_tail(&splice, &fs_info->delalloc_roots);
9501 0 : spin_unlock(&fs_info->delalloc_root_lock);
9502 : }
9503 0 : mutex_unlock(&fs_info->delalloc_root_mutex);
9504 0 : return ret;
9505 : }
9506 :
9507 0 : static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9508 : struct dentry *dentry, const char *symname)
9509 : {
9510 0 : struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9511 0 : struct btrfs_trans_handle *trans;
9512 0 : struct btrfs_root *root = BTRFS_I(dir)->root;
9513 0 : struct btrfs_path *path;
9514 0 : struct btrfs_key key;
9515 0 : struct inode *inode;
9516 0 : struct btrfs_new_inode_args new_inode_args = {
9517 : .dir = dir,
9518 : .dentry = dentry,
9519 : };
9520 0 : unsigned int trans_num_items;
9521 0 : int err;
9522 0 : int name_len;
9523 0 : int datasize;
9524 0 : unsigned long ptr;
9525 0 : struct btrfs_file_extent_item *ei;
9526 0 : struct extent_buffer *leaf;
9527 :
9528 0 : name_len = strlen(symname);
9529 0 : if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9530 : return -ENAMETOOLONG;
9531 :
9532 0 : inode = new_inode(dir->i_sb);
9533 0 : if (!inode)
9534 : return -ENOMEM;
9535 0 : inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9536 0 : inode->i_op = &btrfs_symlink_inode_operations;
9537 0 : inode_nohighmem(inode);
9538 0 : inode->i_mapping->a_ops = &btrfs_aops;
9539 0 : btrfs_i_size_write(BTRFS_I(inode), name_len);
9540 0 : inode_set_bytes(inode, name_len);
9541 :
9542 0 : new_inode_args.inode = inode;
9543 0 : err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9544 0 : if (err)
9545 0 : goto out_inode;
9546 : /* 1 additional item for the inline extent */
9547 0 : trans_num_items++;
9548 :
9549 0 : trans = btrfs_start_transaction(root, trans_num_items);
9550 0 : if (IS_ERR(trans)) {
9551 0 : err = PTR_ERR(trans);
9552 0 : goto out_new_inode_args;
9553 : }
9554 :
9555 0 : err = btrfs_create_new_inode(trans, &new_inode_args);
9556 0 : if (err)
9557 0 : goto out;
9558 :
9559 0 : path = btrfs_alloc_path();
9560 0 : if (!path) {
9561 0 : err = -ENOMEM;
9562 0 : btrfs_abort_transaction(trans, err);
9563 0 : discard_new_inode(inode);
9564 0 : inode = NULL;
9565 0 : goto out;
9566 : }
9567 0 : key.objectid = btrfs_ino(BTRFS_I(inode));
9568 0 : key.offset = 0;
9569 0 : key.type = BTRFS_EXTENT_DATA_KEY;
9570 0 : datasize = btrfs_file_extent_calc_inline_size(name_len);
9571 0 : err = btrfs_insert_empty_item(trans, root, path, &key,
9572 : datasize);
9573 0 : if (err) {
9574 0 : btrfs_abort_transaction(trans, err);
9575 0 : btrfs_free_path(path);
9576 0 : discard_new_inode(inode);
9577 0 : inode = NULL;
9578 0 : goto out;
9579 : }
9580 0 : leaf = path->nodes[0];
9581 0 : ei = btrfs_item_ptr(leaf, path->slots[0],
9582 : struct btrfs_file_extent_item);
9583 0 : btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9584 0 : btrfs_set_file_extent_type(leaf, ei,
9585 : BTRFS_FILE_EXTENT_INLINE);
9586 0 : btrfs_set_file_extent_encryption(leaf, ei, 0);
9587 0 : btrfs_set_file_extent_compression(leaf, ei, 0);
9588 0 : btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9589 0 : btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9590 :
9591 0 : ptr = btrfs_file_extent_inline_start(ei);
9592 0 : write_extent_buffer(leaf, symname, ptr, name_len);
9593 0 : btrfs_mark_buffer_dirty(leaf);
9594 0 : btrfs_free_path(path);
9595 :
9596 0 : d_instantiate_new(dentry, inode);
9597 0 : err = 0;
9598 0 : out:
9599 0 : btrfs_end_transaction(trans);
9600 0 : btrfs_btree_balance_dirty(fs_info);
9601 0 : out_new_inode_args:
9602 0 : btrfs_new_inode_args_destroy(&new_inode_args);
9603 0 : out_inode:
9604 0 : if (err)
9605 0 : iput(inode);
9606 : return err;
9607 : }
9608 :
9609 0 : static struct btrfs_trans_handle *insert_prealloc_file_extent(
9610 : struct btrfs_trans_handle *trans_in,
9611 : struct btrfs_inode *inode,
9612 : struct btrfs_key *ins,
9613 : u64 file_offset)
9614 : {
9615 0 : struct btrfs_file_extent_item stack_fi;
9616 0 : struct btrfs_replace_extent_info extent_info;
9617 0 : struct btrfs_trans_handle *trans = trans_in;
9618 0 : struct btrfs_path *path;
9619 0 : u64 start = ins->objectid;
9620 0 : u64 len = ins->offset;
9621 0 : int qgroup_released;
9622 0 : int ret;
9623 :
9624 0 : memset(&stack_fi, 0, sizeof(stack_fi));
9625 :
9626 0 : btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9627 0 : btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9628 0 : btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9629 0 : btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9630 0 : btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9631 0 : btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9632 : /* Encryption and other encoding is reserved and all 0 */
9633 :
9634 0 : qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9635 0 : if (qgroup_released < 0)
9636 0 : return ERR_PTR(qgroup_released);
9637 :
9638 0 : if (trans) {
9639 0 : ret = insert_reserved_file_extent(trans, inode,
9640 : file_offset, &stack_fi,
9641 : true, qgroup_released);
9642 0 : if (ret)
9643 0 : goto free_qgroup;
9644 0 : return trans;
9645 : }
9646 :
9647 0 : extent_info.disk_offset = start;
9648 0 : extent_info.disk_len = len;
9649 0 : extent_info.data_offset = 0;
9650 0 : extent_info.data_len = len;
9651 0 : extent_info.file_offset = file_offset;
9652 0 : extent_info.extent_buf = (char *)&stack_fi;
9653 0 : extent_info.is_new_extent = true;
9654 0 : extent_info.update_times = true;
9655 0 : extent_info.qgroup_reserved = qgroup_released;
9656 0 : extent_info.insertions = 0;
9657 :
9658 0 : path = btrfs_alloc_path();
9659 0 : if (!path) {
9660 0 : ret = -ENOMEM;
9661 0 : goto free_qgroup;
9662 : }
9663 :
9664 0 : ret = btrfs_replace_file_extents(inode, path, file_offset,
9665 0 : file_offset + len - 1, &extent_info,
9666 : &trans);
9667 0 : btrfs_free_path(path);
9668 0 : if (ret)
9669 0 : goto free_qgroup;
9670 0 : return trans;
9671 :
9672 0 : free_qgroup:
9673 : /*
9674 : * We have released qgroup data range at the beginning of the function,
9675 : * and normally qgroup_released bytes will be freed when committing
9676 : * transaction.
9677 : * But if we error out early, we have to free what we have released
9678 : * or we leak qgroup data reservation.
9679 : */
9680 0 : btrfs_qgroup_free_refroot(inode->root->fs_info,
9681 : inode->root->root_key.objectid, qgroup_released,
9682 : BTRFS_QGROUP_RSV_DATA);
9683 0 : return ERR_PTR(ret);
9684 : }
9685 :
9686 0 : static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9687 : u64 start, u64 num_bytes, u64 min_size,
9688 : loff_t actual_len, u64 *alloc_hint,
9689 : struct btrfs_trans_handle *trans)
9690 : {
9691 0 : struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9692 0 : struct extent_map *em;
9693 0 : struct btrfs_root *root = BTRFS_I(inode)->root;
9694 0 : struct btrfs_key ins;
9695 0 : u64 cur_offset = start;
9696 0 : u64 clear_offset = start;
9697 0 : u64 i_size;
9698 0 : u64 cur_bytes;
9699 0 : u64 last_alloc = (u64)-1;
9700 0 : int ret = 0;
9701 0 : bool own_trans = true;
9702 0 : u64 end = start + num_bytes - 1;
9703 :
9704 0 : if (trans)
9705 0 : own_trans = false;
9706 0 : while (num_bytes > 0) {
9707 0 : cur_bytes = min_t(u64, num_bytes, SZ_256M);
9708 0 : cur_bytes = max(cur_bytes, min_size);
9709 : /*
9710 : * If we are severely fragmented we could end up with really
9711 : * small allocations, so if the allocator is returning small
9712 : * chunks lets make its job easier by only searching for those
9713 : * sized chunks.
9714 : */
9715 0 : cur_bytes = min(cur_bytes, last_alloc);
9716 0 : ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9717 : min_size, 0, *alloc_hint, &ins, 1, 0);
9718 0 : if (ret)
9719 : break;
9720 :
9721 : /*
9722 : * We've reserved this space, and thus converted it from
9723 : * ->bytes_may_use to ->bytes_reserved. Any error that happens
9724 : * from here on out we will only need to clear our reservation
9725 : * for the remaining unreserved area, so advance our
9726 : * clear_offset by our extent size.
9727 : */
9728 0 : clear_offset += ins.offset;
9729 :
9730 0 : last_alloc = ins.offset;
9731 0 : trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9732 : &ins, cur_offset);
9733 : /*
9734 : * Now that we inserted the prealloc extent we can finally
9735 : * decrement the number of reservations in the block group.
9736 : * If we did it before, we could race with relocation and have
9737 : * relocation miss the reserved extent, making it fail later.
9738 : */
9739 0 : btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9740 0 : if (IS_ERR(trans)) {
9741 0 : ret = PTR_ERR(trans);
9742 0 : btrfs_free_reserved_extent(fs_info, ins.objectid,
9743 : ins.offset, 0);
9744 0 : break;
9745 : }
9746 :
9747 0 : em = alloc_extent_map();
9748 0 : if (!em) {
9749 0 : btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9750 0 : cur_offset + ins.offset - 1, false);
9751 0 : btrfs_set_inode_full_sync(BTRFS_I(inode));
9752 0 : goto next;
9753 : }
9754 :
9755 0 : em->start = cur_offset;
9756 0 : em->orig_start = cur_offset;
9757 0 : em->len = ins.offset;
9758 0 : em->block_start = ins.objectid;
9759 0 : em->block_len = ins.offset;
9760 0 : em->orig_block_len = ins.offset;
9761 0 : em->ram_bytes = ins.offset;
9762 0 : set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9763 0 : em->generation = trans->transid;
9764 :
9765 0 : ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9766 0 : free_extent_map(em);
9767 0 : next:
9768 0 : num_bytes -= ins.offset;
9769 0 : cur_offset += ins.offset;
9770 0 : *alloc_hint = ins.objectid + ins.offset;
9771 :
9772 0 : inode_inc_iversion(inode);
9773 0 : inode->i_ctime = current_time(inode);
9774 0 : BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9775 0 : if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9776 0 : (actual_len > inode->i_size) &&
9777 0 : (cur_offset > inode->i_size)) {
9778 0 : if (cur_offset > actual_len)
9779 : i_size = actual_len;
9780 : else
9781 : i_size = cur_offset;
9782 0 : i_size_write(inode, i_size);
9783 0 : btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9784 : }
9785 :
9786 0 : ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9787 :
9788 0 : if (ret) {
9789 0 : btrfs_abort_transaction(trans, ret);
9790 0 : if (own_trans)
9791 0 : btrfs_end_transaction(trans);
9792 : break;
9793 : }
9794 :
9795 0 : if (own_trans) {
9796 0 : btrfs_end_transaction(trans);
9797 0 : trans = NULL;
9798 : }
9799 : }
9800 0 : if (clear_offset < end)
9801 0 : btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9802 0 : end - clear_offset + 1);
9803 0 : return ret;
9804 : }
9805 :
9806 0 : int btrfs_prealloc_file_range(struct inode *inode, int mode,
9807 : u64 start, u64 num_bytes, u64 min_size,
9808 : loff_t actual_len, u64 *alloc_hint)
9809 : {
9810 0 : return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9811 : min_size, actual_len, alloc_hint,
9812 : NULL);
9813 : }
9814 :
9815 0 : int btrfs_prealloc_file_range_trans(struct inode *inode,
9816 : struct btrfs_trans_handle *trans, int mode,
9817 : u64 start, u64 num_bytes, u64 min_size,
9818 : loff_t actual_len, u64 *alloc_hint)
9819 : {
9820 0 : return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9821 : min_size, actual_len, alloc_hint, trans);
9822 : }
9823 :
9824 0 : static int btrfs_permission(struct mnt_idmap *idmap,
9825 : struct inode *inode, int mask)
9826 : {
9827 0 : struct btrfs_root *root = BTRFS_I(inode)->root;
9828 0 : umode_t mode = inode->i_mode;
9829 :
9830 0 : if (mask & MAY_WRITE &&
9831 0 : (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9832 0 : if (btrfs_root_readonly(root))
9833 : return -EROFS;
9834 0 : if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9835 : return -EACCES;
9836 : }
9837 0 : return generic_permission(idmap, inode, mask);
9838 : }
9839 :
9840 0 : static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9841 : struct file *file, umode_t mode)
9842 : {
9843 0 : struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9844 0 : struct btrfs_trans_handle *trans;
9845 0 : struct btrfs_root *root = BTRFS_I(dir)->root;
9846 0 : struct inode *inode;
9847 0 : struct btrfs_new_inode_args new_inode_args = {
9848 : .dir = dir,
9849 0 : .dentry = file->f_path.dentry,
9850 : .orphan = true,
9851 : };
9852 0 : unsigned int trans_num_items;
9853 0 : int ret;
9854 :
9855 0 : inode = new_inode(dir->i_sb);
9856 0 : if (!inode)
9857 : return -ENOMEM;
9858 0 : inode_init_owner(idmap, inode, dir, mode);
9859 0 : inode->i_fop = &btrfs_file_operations;
9860 0 : inode->i_op = &btrfs_file_inode_operations;
9861 0 : inode->i_mapping->a_ops = &btrfs_aops;
9862 :
9863 0 : new_inode_args.inode = inode;
9864 0 : ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9865 0 : if (ret)
9866 0 : goto out_inode;
9867 :
9868 0 : trans = btrfs_start_transaction(root, trans_num_items);
9869 0 : if (IS_ERR(trans)) {
9870 0 : ret = PTR_ERR(trans);
9871 0 : goto out_new_inode_args;
9872 : }
9873 :
9874 0 : ret = btrfs_create_new_inode(trans, &new_inode_args);
9875 :
9876 : /*
9877 : * We set number of links to 0 in btrfs_create_new_inode(), and here we
9878 : * set it to 1 because d_tmpfile() will issue a warning if the count is
9879 : * 0, through:
9880 : *
9881 : * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9882 : */
9883 0 : set_nlink(inode, 1);
9884 :
9885 0 : if (!ret) {
9886 0 : d_tmpfile(file, inode);
9887 0 : unlock_new_inode(inode);
9888 0 : mark_inode_dirty(inode);
9889 : }
9890 :
9891 0 : btrfs_end_transaction(trans);
9892 0 : btrfs_btree_balance_dirty(fs_info);
9893 0 : out_new_inode_args:
9894 0 : btrfs_new_inode_args_destroy(&new_inode_args);
9895 0 : out_inode:
9896 0 : if (ret)
9897 0 : iput(inode);
9898 0 : return finish_open_simple(file, ret);
9899 : }
9900 :
9901 0 : void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9902 : {
9903 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
9904 0 : unsigned long index = start >> PAGE_SHIFT;
9905 0 : unsigned long end_index = end >> PAGE_SHIFT;
9906 0 : struct page *page;
9907 0 : u32 len;
9908 :
9909 0 : ASSERT(end + 1 - start <= U32_MAX);
9910 0 : len = end + 1 - start;
9911 0 : while (index <= end_index) {
9912 0 : page = find_get_page(inode->vfs_inode.i_mapping, index);
9913 0 : ASSERT(page); /* Pages should be in the extent_io_tree */
9914 :
9915 0 : btrfs_page_set_writeback(fs_info, page, start, len);
9916 0 : put_page(page);
9917 0 : index++;
9918 : }
9919 0 : }
9920 :
9921 0 : int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9922 : int compress_type)
9923 : {
9924 0 : switch (compress_type) {
9925 : case BTRFS_COMPRESS_NONE:
9926 : return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9927 0 : case BTRFS_COMPRESS_ZLIB:
9928 0 : return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9929 0 : case BTRFS_COMPRESS_LZO:
9930 : /*
9931 : * The LZO format depends on the sector size. 64K is the maximum
9932 : * sector size that we support.
9933 : */
9934 0 : if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9935 : return -EINVAL;
9936 0 : return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9937 0 : (fs_info->sectorsize_bits - 12);
9938 0 : case BTRFS_COMPRESS_ZSTD:
9939 0 : return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9940 0 : default:
9941 0 : return -EUCLEAN;
9942 : }
9943 : }
9944 :
9945 0 : static ssize_t btrfs_encoded_read_inline(
9946 : struct kiocb *iocb,
9947 : struct iov_iter *iter, u64 start,
9948 : u64 lockend,
9949 : struct extent_state **cached_state,
9950 : u64 extent_start, size_t count,
9951 : struct btrfs_ioctl_encoded_io_args *encoded,
9952 : bool *unlocked)
9953 : {
9954 0 : struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9955 0 : struct btrfs_root *root = inode->root;
9956 0 : struct btrfs_fs_info *fs_info = root->fs_info;
9957 0 : struct extent_io_tree *io_tree = &inode->io_tree;
9958 0 : struct btrfs_path *path;
9959 0 : struct extent_buffer *leaf;
9960 0 : struct btrfs_file_extent_item *item;
9961 0 : u64 ram_bytes;
9962 0 : unsigned long ptr;
9963 0 : void *tmp;
9964 0 : ssize_t ret;
9965 :
9966 0 : path = btrfs_alloc_path();
9967 0 : if (!path) {
9968 0 : ret = -ENOMEM;
9969 0 : goto out;
9970 : }
9971 0 : ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9972 : extent_start, 0);
9973 0 : if (ret) {
9974 0 : if (ret > 0) {
9975 : /* The extent item disappeared? */
9976 0 : ret = -EIO;
9977 : }
9978 0 : goto out;
9979 : }
9980 0 : leaf = path->nodes[0];
9981 0 : item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9982 :
9983 0 : ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9984 0 : ptr = btrfs_file_extent_inline_start(item);
9985 :
9986 0 : encoded->len = min_t(u64, extent_start + ram_bytes,
9987 0 : inode->vfs_inode.i_size) - iocb->ki_pos;
9988 0 : ret = btrfs_encoded_io_compression_from_extent(fs_info,
9989 : btrfs_file_extent_compression(leaf, item));
9990 0 : if (ret < 0)
9991 0 : goto out;
9992 0 : encoded->compression = ret;
9993 0 : if (encoded->compression) {
9994 0 : size_t inline_size;
9995 :
9996 0 : inline_size = btrfs_file_extent_inline_item_len(leaf,
9997 : path->slots[0]);
9998 0 : if (inline_size > count) {
9999 0 : ret = -ENOBUFS;
10000 0 : goto out;
10001 : }
10002 0 : count = inline_size;
10003 0 : encoded->unencoded_len = ram_bytes;
10004 0 : encoded->unencoded_offset = iocb->ki_pos - extent_start;
10005 : } else {
10006 0 : count = min_t(u64, count, encoded->len);
10007 0 : encoded->len = count;
10008 0 : encoded->unencoded_len = count;
10009 0 : ptr += iocb->ki_pos - extent_start;
10010 : }
10011 :
10012 0 : tmp = kmalloc(count, GFP_NOFS);
10013 0 : if (!tmp) {
10014 0 : ret = -ENOMEM;
10015 0 : goto out;
10016 : }
10017 0 : read_extent_buffer(leaf, tmp, ptr, count);
10018 0 : btrfs_release_path(path);
10019 0 : unlock_extent(io_tree, start, lockend, cached_state);
10020 0 : btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10021 0 : *unlocked = true;
10022 :
10023 0 : ret = copy_to_iter(tmp, count, iter);
10024 0 : if (ret != count)
10025 0 : ret = -EFAULT;
10026 0 : kfree(tmp);
10027 0 : out:
10028 0 : btrfs_free_path(path);
10029 0 : return ret;
10030 : }
10031 :
10032 : struct btrfs_encoded_read_private {
10033 : wait_queue_head_t wait;
10034 : atomic_t pending;
10035 : blk_status_t status;
10036 : };
10037 :
10038 0 : static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
10039 : {
10040 0 : struct btrfs_encoded_read_private *priv = bbio->private;
10041 :
10042 0 : if (bbio->bio.bi_status) {
10043 : /*
10044 : * The memory barrier implied by the atomic_dec_return() here
10045 : * pairs with the memory barrier implied by the
10046 : * atomic_dec_return() or io_wait_event() in
10047 : * btrfs_encoded_read_regular_fill_pages() to ensure that this
10048 : * write is observed before the load of status in
10049 : * btrfs_encoded_read_regular_fill_pages().
10050 : */
10051 0 : WRITE_ONCE(priv->status, bbio->bio.bi_status);
10052 : }
10053 0 : if (!atomic_dec_return(&priv->pending))
10054 0 : wake_up(&priv->wait);
10055 0 : bio_put(&bbio->bio);
10056 0 : }
10057 :
10058 0 : int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10059 : u64 file_offset, u64 disk_bytenr,
10060 : u64 disk_io_size, struct page **pages)
10061 : {
10062 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
10063 0 : struct btrfs_encoded_read_private priv = {
10064 : .pending = ATOMIC_INIT(1),
10065 : };
10066 0 : unsigned long i = 0;
10067 0 : struct btrfs_bio *bbio;
10068 :
10069 0 : init_waitqueue_head(&priv.wait);
10070 :
10071 0 : bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10072 : btrfs_encoded_read_endio, &priv);
10073 0 : bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10074 0 : bbio->inode = inode;
10075 :
10076 0 : do {
10077 0 : size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
10078 :
10079 0 : if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
10080 0 : atomic_inc(&priv.pending);
10081 0 : btrfs_submit_bio(bbio, 0);
10082 :
10083 0 : bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10084 : btrfs_encoded_read_endio, &priv);
10085 0 : bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10086 0 : bbio->inode = inode;
10087 0 : continue;
10088 : }
10089 :
10090 0 : i++;
10091 0 : disk_bytenr += bytes;
10092 0 : disk_io_size -= bytes;
10093 0 : } while (disk_io_size);
10094 :
10095 0 : atomic_inc(&priv.pending);
10096 0 : btrfs_submit_bio(bbio, 0);
10097 :
10098 0 : if (atomic_dec_return(&priv.pending))
10099 0 : io_wait_event(priv.wait, !atomic_read(&priv.pending));
10100 : /* See btrfs_encoded_read_endio() for ordering. */
10101 0 : return blk_status_to_errno(READ_ONCE(priv.status));
10102 : }
10103 :
10104 0 : static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10105 : struct iov_iter *iter,
10106 : u64 start, u64 lockend,
10107 : struct extent_state **cached_state,
10108 : u64 disk_bytenr, u64 disk_io_size,
10109 : size_t count, bool compressed,
10110 : bool *unlocked)
10111 : {
10112 0 : struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10113 0 : struct extent_io_tree *io_tree = &inode->io_tree;
10114 0 : struct page **pages;
10115 0 : unsigned long nr_pages, i;
10116 0 : u64 cur;
10117 0 : size_t page_offset;
10118 0 : ssize_t ret;
10119 :
10120 0 : nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10121 0 : pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10122 0 : if (!pages)
10123 : return -ENOMEM;
10124 0 : ret = btrfs_alloc_page_array(nr_pages, pages);
10125 0 : if (ret) {
10126 0 : ret = -ENOMEM;
10127 0 : goto out;
10128 : }
10129 :
10130 0 : ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10131 : disk_io_size, pages);
10132 0 : if (ret)
10133 0 : goto out;
10134 :
10135 0 : unlock_extent(io_tree, start, lockend, cached_state);
10136 0 : btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10137 0 : *unlocked = true;
10138 :
10139 0 : if (compressed) {
10140 : i = 0;
10141 : page_offset = 0;
10142 : } else {
10143 0 : i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10144 0 : page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10145 : }
10146 : cur = 0;
10147 0 : while (cur < count) {
10148 0 : size_t bytes = min_t(size_t, count - cur,
10149 : PAGE_SIZE - page_offset);
10150 :
10151 0 : if (copy_page_to_iter(pages[i], page_offset, bytes,
10152 : iter) != bytes) {
10153 0 : ret = -EFAULT;
10154 0 : goto out;
10155 : }
10156 0 : i++;
10157 0 : cur += bytes;
10158 0 : page_offset = 0;
10159 : }
10160 0 : ret = count;
10161 0 : out:
10162 0 : for (i = 0; i < nr_pages; i++) {
10163 0 : if (pages[i])
10164 0 : __free_page(pages[i]);
10165 : }
10166 0 : kfree(pages);
10167 0 : return ret;
10168 : }
10169 :
10170 0 : ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10171 : struct btrfs_ioctl_encoded_io_args *encoded)
10172 : {
10173 0 : struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10174 0 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
10175 0 : struct extent_io_tree *io_tree = &inode->io_tree;
10176 0 : ssize_t ret;
10177 0 : size_t count = iov_iter_count(iter);
10178 0 : u64 start, lockend, disk_bytenr, disk_io_size;
10179 0 : struct extent_state *cached_state = NULL;
10180 0 : struct extent_map *em;
10181 0 : bool unlocked = false;
10182 :
10183 0 : file_accessed(iocb->ki_filp);
10184 :
10185 0 : btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10186 :
10187 0 : if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10188 0 : btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10189 0 : return 0;
10190 : }
10191 0 : start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10192 : /*
10193 : * We don't know how long the extent containing iocb->ki_pos is, but if
10194 : * it's compressed we know that it won't be longer than this.
10195 : */
10196 0 : lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10197 :
10198 0 : for (;;) {
10199 0 : struct btrfs_ordered_extent *ordered;
10200 :
10201 0 : ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10202 : lockend - start + 1);
10203 0 : if (ret)
10204 0 : goto out_unlock_inode;
10205 0 : lock_extent(io_tree, start, lockend, &cached_state);
10206 0 : ordered = btrfs_lookup_ordered_range(inode, start,
10207 : lockend - start + 1);
10208 0 : if (!ordered)
10209 : break;
10210 0 : btrfs_put_ordered_extent(ordered);
10211 0 : unlock_extent(io_tree, start, lockend, &cached_state);
10212 0 : cond_resched();
10213 : }
10214 :
10215 0 : em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10216 0 : if (IS_ERR(em)) {
10217 0 : ret = PTR_ERR(em);
10218 0 : goto out_unlock_extent;
10219 : }
10220 :
10221 0 : if (em->block_start == EXTENT_MAP_INLINE) {
10222 0 : u64 extent_start = em->start;
10223 :
10224 : /*
10225 : * For inline extents we get everything we need out of the
10226 : * extent item.
10227 : */
10228 0 : free_extent_map(em);
10229 0 : em = NULL;
10230 0 : ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10231 : &cached_state, extent_start,
10232 : count, encoded, &unlocked);
10233 0 : goto out;
10234 : }
10235 :
10236 : /*
10237 : * We only want to return up to EOF even if the extent extends beyond
10238 : * that.
10239 : */
10240 0 : encoded->len = min_t(u64, extent_map_end(em),
10241 0 : inode->vfs_inode.i_size) - iocb->ki_pos;
10242 0 : if (em->block_start == EXTENT_MAP_HOLE ||
10243 0 : test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10244 0 : disk_bytenr = EXTENT_MAP_HOLE;
10245 0 : count = min_t(u64, count, encoded->len);
10246 0 : encoded->len = count;
10247 0 : encoded->unencoded_len = count;
10248 0 : } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10249 0 : disk_bytenr = em->block_start;
10250 : /*
10251 : * Bail if the buffer isn't large enough to return the whole
10252 : * compressed extent.
10253 : */
10254 0 : if (em->block_len > count) {
10255 0 : ret = -ENOBUFS;
10256 0 : goto out_em;
10257 : }
10258 0 : disk_io_size = em->block_len;
10259 0 : count = em->block_len;
10260 0 : encoded->unencoded_len = em->ram_bytes;
10261 0 : encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10262 0 : ret = btrfs_encoded_io_compression_from_extent(fs_info,
10263 0 : em->compress_type);
10264 0 : if (ret < 0)
10265 0 : goto out_em;
10266 0 : encoded->compression = ret;
10267 : } else {
10268 0 : disk_bytenr = em->block_start + (start - em->start);
10269 0 : if (encoded->len > count)
10270 0 : encoded->len = count;
10271 : /*
10272 : * Don't read beyond what we locked. This also limits the page
10273 : * allocations that we'll do.
10274 : */
10275 0 : disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10276 0 : count = start + disk_io_size - iocb->ki_pos;
10277 0 : encoded->len = count;
10278 0 : encoded->unencoded_len = count;
10279 0 : disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10280 : }
10281 0 : free_extent_map(em);
10282 0 : em = NULL;
10283 :
10284 0 : if (disk_bytenr == EXTENT_MAP_HOLE) {
10285 0 : unlock_extent(io_tree, start, lockend, &cached_state);
10286 0 : btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10287 0 : unlocked = true;
10288 0 : ret = iov_iter_zero(count, iter);
10289 0 : if (ret != count)
10290 : ret = -EFAULT;
10291 : } else {
10292 0 : ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10293 : &cached_state, disk_bytenr,
10294 : disk_io_size, count,
10295 0 : encoded->compression,
10296 : &unlocked);
10297 : }
10298 :
10299 0 : out:
10300 0 : if (ret >= 0)
10301 0 : iocb->ki_pos += encoded->len;
10302 0 : out_em:
10303 0 : free_extent_map(em);
10304 0 : out_unlock_extent:
10305 0 : if (!unlocked)
10306 0 : unlock_extent(io_tree, start, lockend, &cached_state);
10307 0 : out_unlock_inode:
10308 0 : if (!unlocked)
10309 0 : btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10310 : return ret;
10311 : }
10312 :
10313 0 : ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10314 : const struct btrfs_ioctl_encoded_io_args *encoded)
10315 : {
10316 0 : struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10317 0 : struct btrfs_root *root = inode->root;
10318 0 : struct btrfs_fs_info *fs_info = root->fs_info;
10319 0 : struct extent_io_tree *io_tree = &inode->io_tree;
10320 0 : struct extent_changeset *data_reserved = NULL;
10321 0 : struct extent_state *cached_state = NULL;
10322 0 : struct btrfs_ordered_extent *ordered;
10323 0 : int compression;
10324 0 : size_t orig_count;
10325 0 : u64 start, end;
10326 0 : u64 num_bytes, ram_bytes, disk_num_bytes;
10327 0 : unsigned long nr_pages, i;
10328 0 : struct page **pages;
10329 0 : struct btrfs_key ins;
10330 0 : bool extent_reserved = false;
10331 0 : struct extent_map *em;
10332 0 : ssize_t ret;
10333 :
10334 0 : switch (encoded->compression) {
10335 : case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10336 : compression = BTRFS_COMPRESS_ZLIB;
10337 : break;
10338 0 : case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10339 0 : compression = BTRFS_COMPRESS_ZSTD;
10340 0 : break;
10341 0 : case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10342 : case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10343 : case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10344 : case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10345 : case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10346 : /* The sector size must match for LZO. */
10347 0 : if (encoded->compression -
10348 0 : BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10349 0 : fs_info->sectorsize_bits)
10350 : return -EINVAL;
10351 : compression = BTRFS_COMPRESS_LZO;
10352 : break;
10353 : default:
10354 : return -EINVAL;
10355 : }
10356 0 : if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10357 : return -EINVAL;
10358 :
10359 0 : orig_count = iov_iter_count(from);
10360 :
10361 : /* The extent size must be sane. */
10362 0 : if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10363 0 : orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10364 : return -EINVAL;
10365 :
10366 : /*
10367 : * The compressed data must be smaller than the decompressed data.
10368 : *
10369 : * It's of course possible for data to compress to larger or the same
10370 : * size, but the buffered I/O path falls back to no compression for such
10371 : * data, and we don't want to break any assumptions by creating these
10372 : * extents.
10373 : *
10374 : * Note that this is less strict than the current check we have that the
10375 : * compressed data must be at least one sector smaller than the
10376 : * decompressed data. We only want to enforce the weaker requirement
10377 : * from old kernels that it is at least one byte smaller.
10378 : */
10379 0 : if (orig_count >= encoded->unencoded_len)
10380 : return -EINVAL;
10381 :
10382 : /* The extent must start on a sector boundary. */
10383 0 : start = iocb->ki_pos;
10384 0 : if (!IS_ALIGNED(start, fs_info->sectorsize))
10385 : return -EINVAL;
10386 :
10387 : /*
10388 : * The extent must end on a sector boundary. However, we allow a write
10389 : * which ends at or extends i_size to have an unaligned length; we round
10390 : * up the extent size and set i_size to the unaligned end.
10391 : */
10392 0 : if (start + encoded->len < inode->vfs_inode.i_size &&
10393 0 : !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10394 : return -EINVAL;
10395 :
10396 : /* Finally, the offset in the unencoded data must be sector-aligned. */
10397 0 : if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10398 : return -EINVAL;
10399 :
10400 0 : num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10401 0 : ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10402 0 : end = start + num_bytes - 1;
10403 :
10404 : /*
10405 : * If the extent cannot be inline, the compressed data on disk must be
10406 : * sector-aligned. For convenience, we extend it with zeroes if it
10407 : * isn't.
10408 : */
10409 0 : disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10410 0 : nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10411 0 : pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10412 0 : if (!pages)
10413 : return -ENOMEM;
10414 0 : for (i = 0; i < nr_pages; i++) {
10415 0 : size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10416 0 : char *kaddr;
10417 :
10418 0 : pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10419 0 : if (!pages[i]) {
10420 0 : ret = -ENOMEM;
10421 0 : goto out_pages;
10422 : }
10423 0 : kaddr = kmap_local_page(pages[i]);
10424 0 : if (copy_from_iter(kaddr, bytes, from) != bytes) {
10425 0 : kunmap_local(kaddr);
10426 0 : ret = -EFAULT;
10427 0 : goto out_pages;
10428 : }
10429 0 : if (bytes < PAGE_SIZE)
10430 0 : memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10431 0 : kunmap_local(kaddr);
10432 : }
10433 :
10434 0 : for (;;) {
10435 0 : struct btrfs_ordered_extent *ordered;
10436 :
10437 0 : ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10438 0 : if (ret)
10439 0 : goto out_pages;
10440 0 : ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10441 0 : start >> PAGE_SHIFT,
10442 0 : end >> PAGE_SHIFT);
10443 0 : if (ret)
10444 0 : goto out_pages;
10445 0 : lock_extent(io_tree, start, end, &cached_state);
10446 0 : ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10447 0 : if (!ordered &&
10448 0 : !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10449 : break;
10450 0 : if (ordered)
10451 0 : btrfs_put_ordered_extent(ordered);
10452 0 : unlock_extent(io_tree, start, end, &cached_state);
10453 0 : cond_resched();
10454 : }
10455 :
10456 : /*
10457 : * We don't use the higher-level delalloc space functions because our
10458 : * num_bytes and disk_num_bytes are different.
10459 : */
10460 0 : ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10461 0 : if (ret)
10462 0 : goto out_unlock;
10463 0 : ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10464 0 : if (ret)
10465 0 : goto out_free_data_space;
10466 0 : ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10467 : false);
10468 0 : if (ret)
10469 0 : goto out_qgroup_free_data;
10470 :
10471 : /* Try an inline extent first. */
10472 0 : if (start == 0 && encoded->unencoded_len == encoded->len &&
10473 0 : encoded->unencoded_offset == 0) {
10474 0 : ret = cow_file_range_inline(inode, encoded->len, orig_count,
10475 : compression, pages, true);
10476 0 : if (ret <= 0) {
10477 0 : if (ret == 0)
10478 0 : ret = orig_count;
10479 0 : goto out_delalloc_release;
10480 : }
10481 : }
10482 :
10483 0 : ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10484 : disk_num_bytes, 0, 0, &ins, 1, 1);
10485 0 : if (ret)
10486 0 : goto out_delalloc_release;
10487 0 : extent_reserved = true;
10488 :
10489 0 : em = create_io_em(inode, start, num_bytes,
10490 0 : start - encoded->unencoded_offset, ins.objectid,
10491 : ins.offset, ins.offset, ram_bytes, compression,
10492 : BTRFS_ORDERED_COMPRESSED);
10493 0 : if (IS_ERR(em)) {
10494 0 : ret = PTR_ERR(em);
10495 0 : goto out_free_reserved;
10496 : }
10497 0 : free_extent_map(em);
10498 :
10499 0 : ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10500 : ins.objectid, ins.offset,
10501 0 : encoded->unencoded_offset,
10502 : (1 << BTRFS_ORDERED_ENCODED) |
10503 : (1 << BTRFS_ORDERED_COMPRESSED),
10504 : compression);
10505 0 : if (IS_ERR(ordered)) {
10506 0 : btrfs_drop_extent_map_range(inode, start, end, false);
10507 0 : ret = PTR_ERR(ordered);
10508 0 : goto out_free_reserved;
10509 : }
10510 0 : btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10511 :
10512 0 : if (start + encoded->len > inode->vfs_inode.i_size)
10513 0 : i_size_write(&inode->vfs_inode, start + encoded->len);
10514 :
10515 0 : unlock_extent(io_tree, start, end, &cached_state);
10516 :
10517 0 : btrfs_delalloc_release_extents(inode, num_bytes);
10518 :
10519 0 : btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10520 0 : ret = orig_count;
10521 0 : goto out;
10522 :
10523 0 : out_free_reserved:
10524 0 : btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10525 0 : btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10526 0 : out_delalloc_release:
10527 0 : btrfs_delalloc_release_extents(inode, num_bytes);
10528 0 : btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10529 0 : out_qgroup_free_data:
10530 0 : if (ret < 0)
10531 0 : btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10532 0 : out_free_data_space:
10533 : /*
10534 : * If btrfs_reserve_extent() succeeded, then we already decremented
10535 : * bytes_may_use.
10536 : */
10537 0 : if (!extent_reserved)
10538 0 : btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10539 0 : out_unlock:
10540 0 : unlock_extent(io_tree, start, end, &cached_state);
10541 0 : out_pages:
10542 0 : for (i = 0; i < nr_pages; i++) {
10543 0 : if (pages[i])
10544 0 : __free_page(pages[i]);
10545 : }
10546 0 : kvfree(pages);
10547 0 : out:
10548 0 : if (ret >= 0)
10549 0 : iocb->ki_pos += encoded->len;
10550 : return ret;
10551 : }
10552 :
10553 : #ifdef CONFIG_SWAP
10554 : /*
10555 : * Add an entry indicating a block group or device which is pinned by a
10556 : * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10557 : * negative errno on failure.
10558 : */
10559 0 : static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10560 : bool is_block_group)
10561 : {
10562 0 : struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10563 0 : struct btrfs_swapfile_pin *sp, *entry;
10564 0 : struct rb_node **p;
10565 0 : struct rb_node *parent = NULL;
10566 :
10567 0 : sp = kmalloc(sizeof(*sp), GFP_NOFS);
10568 0 : if (!sp)
10569 : return -ENOMEM;
10570 0 : sp->ptr = ptr;
10571 0 : sp->inode = inode;
10572 0 : sp->is_block_group = is_block_group;
10573 0 : sp->bg_extent_count = 1;
10574 :
10575 0 : spin_lock(&fs_info->swapfile_pins_lock);
10576 0 : p = &fs_info->swapfile_pins.rb_node;
10577 0 : while (*p) {
10578 0 : parent = *p;
10579 0 : entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10580 0 : if (sp->ptr < entry->ptr ||
10581 0 : (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10582 0 : p = &(*p)->rb_left;
10583 0 : } else if (sp->ptr > entry->ptr ||
10584 0 : (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10585 0 : p = &(*p)->rb_right;
10586 : } else {
10587 0 : if (is_block_group)
10588 0 : entry->bg_extent_count++;
10589 0 : spin_unlock(&fs_info->swapfile_pins_lock);
10590 0 : kfree(sp);
10591 0 : return 1;
10592 : }
10593 : }
10594 0 : rb_link_node(&sp->node, parent, p);
10595 0 : rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10596 0 : spin_unlock(&fs_info->swapfile_pins_lock);
10597 0 : return 0;
10598 : }
10599 :
10600 : /* Free all of the entries pinned by this swapfile. */
10601 0 : static void btrfs_free_swapfile_pins(struct inode *inode)
10602 : {
10603 0 : struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10604 0 : struct btrfs_swapfile_pin *sp;
10605 0 : struct rb_node *node, *next;
10606 :
10607 0 : spin_lock(&fs_info->swapfile_pins_lock);
10608 0 : node = rb_first(&fs_info->swapfile_pins);
10609 0 : while (node) {
10610 0 : next = rb_next(node);
10611 0 : sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10612 0 : if (sp->inode == inode) {
10613 0 : rb_erase(&sp->node, &fs_info->swapfile_pins);
10614 0 : if (sp->is_block_group) {
10615 0 : btrfs_dec_block_group_swap_extents(sp->ptr,
10616 : sp->bg_extent_count);
10617 0 : btrfs_put_block_group(sp->ptr);
10618 : }
10619 0 : kfree(sp);
10620 : }
10621 : node = next;
10622 : }
10623 0 : spin_unlock(&fs_info->swapfile_pins_lock);
10624 0 : }
10625 :
10626 : struct btrfs_swap_info {
10627 : u64 start;
10628 : u64 block_start;
10629 : u64 block_len;
10630 : u64 lowest_ppage;
10631 : u64 highest_ppage;
10632 : unsigned long nr_pages;
10633 : int nr_extents;
10634 : };
10635 :
10636 0 : static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10637 : struct btrfs_swap_info *bsi)
10638 : {
10639 0 : unsigned long nr_pages;
10640 0 : unsigned long max_pages;
10641 0 : u64 first_ppage, first_ppage_reported, next_ppage;
10642 0 : int ret;
10643 :
10644 : /*
10645 : * Our swapfile may have had its size extended after the swap header was
10646 : * written. In that case activating the swapfile should not go beyond
10647 : * the max size set in the swap header.
10648 : */
10649 0 : if (bsi->nr_pages >= sis->max)
10650 : return 0;
10651 :
10652 0 : max_pages = sis->max - bsi->nr_pages;
10653 0 : first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10654 0 : next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10655 :
10656 0 : if (first_ppage >= next_ppage)
10657 : return 0;
10658 0 : nr_pages = next_ppage - first_ppage;
10659 0 : nr_pages = min(nr_pages, max_pages);
10660 :
10661 0 : first_ppage_reported = first_ppage;
10662 0 : if (bsi->start == 0)
10663 0 : first_ppage_reported++;
10664 0 : if (bsi->lowest_ppage > first_ppage_reported)
10665 0 : bsi->lowest_ppage = first_ppage_reported;
10666 0 : if (bsi->highest_ppage < (next_ppage - 1))
10667 0 : bsi->highest_ppage = next_ppage - 1;
10668 :
10669 0 : ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10670 0 : if (ret < 0)
10671 : return ret;
10672 0 : bsi->nr_extents += ret;
10673 0 : bsi->nr_pages += nr_pages;
10674 0 : return 0;
10675 : }
10676 :
10677 0 : static void btrfs_swap_deactivate(struct file *file)
10678 : {
10679 0 : struct inode *inode = file_inode(file);
10680 :
10681 0 : btrfs_free_swapfile_pins(inode);
10682 0 : atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10683 0 : }
10684 :
10685 0 : static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10686 : sector_t *span)
10687 : {
10688 0 : struct inode *inode = file_inode(file);
10689 0 : struct btrfs_root *root = BTRFS_I(inode)->root;
10690 0 : struct btrfs_fs_info *fs_info = root->fs_info;
10691 0 : struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10692 0 : struct extent_state *cached_state = NULL;
10693 0 : struct extent_map *em = NULL;
10694 0 : struct btrfs_device *device = NULL;
10695 0 : struct btrfs_swap_info bsi = {
10696 : .lowest_ppage = (sector_t)-1ULL,
10697 : };
10698 0 : int ret = 0;
10699 0 : u64 isize;
10700 0 : u64 start;
10701 :
10702 : /*
10703 : * If the swap file was just created, make sure delalloc is done. If the
10704 : * file changes again after this, the user is doing something stupid and
10705 : * we don't really care.
10706 : */
10707 0 : ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10708 0 : if (ret)
10709 : return ret;
10710 :
10711 : /*
10712 : * The inode is locked, so these flags won't change after we check them.
10713 : */
10714 0 : if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10715 0 : btrfs_warn(fs_info, "swapfile must not be compressed");
10716 0 : return -EINVAL;
10717 : }
10718 0 : if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10719 0 : btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10720 0 : return -EINVAL;
10721 : }
10722 0 : if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10723 0 : btrfs_warn(fs_info, "swapfile must not be checksummed");
10724 0 : return -EINVAL;
10725 : }
10726 :
10727 : /*
10728 : * Balance or device remove/replace/resize can move stuff around from
10729 : * under us. The exclop protection makes sure they aren't running/won't
10730 : * run concurrently while we are mapping the swap extents, and
10731 : * fs_info->swapfile_pins prevents them from running while the swap
10732 : * file is active and moving the extents. Note that this also prevents
10733 : * a concurrent device add which isn't actually necessary, but it's not
10734 : * really worth the trouble to allow it.
10735 : */
10736 0 : if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10737 0 : btrfs_warn(fs_info,
10738 : "cannot activate swapfile while exclusive operation is running");
10739 0 : return -EBUSY;
10740 : }
10741 :
10742 : /*
10743 : * Prevent snapshot creation while we are activating the swap file.
10744 : * We do not want to race with snapshot creation. If snapshot creation
10745 : * already started before we bumped nr_swapfiles from 0 to 1 and
10746 : * completes before the first write into the swap file after it is
10747 : * activated, than that write would fallback to COW.
10748 : */
10749 0 : if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10750 0 : btrfs_exclop_finish(fs_info);
10751 0 : btrfs_warn(fs_info,
10752 : "cannot activate swapfile because snapshot creation is in progress");
10753 0 : return -EINVAL;
10754 : }
10755 : /*
10756 : * Snapshots can create extents which require COW even if NODATACOW is
10757 : * set. We use this counter to prevent snapshots. We must increment it
10758 : * before walking the extents because we don't want a concurrent
10759 : * snapshot to run after we've already checked the extents.
10760 : *
10761 : * It is possible that subvolume is marked for deletion but still not
10762 : * removed yet. To prevent this race, we check the root status before
10763 : * activating the swapfile.
10764 : */
10765 0 : spin_lock(&root->root_item_lock);
10766 0 : if (btrfs_root_dead(root)) {
10767 0 : spin_unlock(&root->root_item_lock);
10768 :
10769 0 : btrfs_exclop_finish(fs_info);
10770 0 : btrfs_warn(fs_info,
10771 : "cannot activate swapfile because subvolume %llu is being deleted",
10772 : root->root_key.objectid);
10773 0 : return -EPERM;
10774 : }
10775 0 : atomic_inc(&root->nr_swapfiles);
10776 0 : spin_unlock(&root->root_item_lock);
10777 :
10778 0 : isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10779 :
10780 0 : lock_extent(io_tree, 0, isize - 1, &cached_state);
10781 0 : start = 0;
10782 0 : while (start < isize) {
10783 0 : u64 logical_block_start, physical_block_start;
10784 0 : struct btrfs_block_group *bg;
10785 0 : u64 len = isize - start;
10786 :
10787 0 : em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10788 0 : if (IS_ERR(em)) {
10789 0 : ret = PTR_ERR(em);
10790 0 : goto out;
10791 : }
10792 :
10793 0 : if (em->block_start == EXTENT_MAP_HOLE) {
10794 0 : btrfs_warn(fs_info, "swapfile must not have holes");
10795 0 : ret = -EINVAL;
10796 0 : goto out;
10797 : }
10798 0 : if (em->block_start == EXTENT_MAP_INLINE) {
10799 : /*
10800 : * It's unlikely we'll ever actually find ourselves
10801 : * here, as a file small enough to fit inline won't be
10802 : * big enough to store more than the swap header, but in
10803 : * case something changes in the future, let's catch it
10804 : * here rather than later.
10805 : */
10806 0 : btrfs_warn(fs_info, "swapfile must not be inline");
10807 0 : ret = -EINVAL;
10808 0 : goto out;
10809 : }
10810 0 : if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10811 0 : btrfs_warn(fs_info, "swapfile must not be compressed");
10812 0 : ret = -EINVAL;
10813 0 : goto out;
10814 : }
10815 :
10816 0 : logical_block_start = em->block_start + (start - em->start);
10817 0 : len = min(len, em->len - (start - em->start));
10818 0 : free_extent_map(em);
10819 0 : em = NULL;
10820 :
10821 0 : ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10822 0 : if (ret < 0) {
10823 0 : goto out;
10824 0 : } else if (ret) {
10825 0 : ret = 0;
10826 : } else {
10827 0 : btrfs_warn(fs_info,
10828 : "swapfile must not be copy-on-write");
10829 0 : ret = -EINVAL;
10830 0 : goto out;
10831 : }
10832 :
10833 0 : em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10834 0 : if (IS_ERR(em)) {
10835 0 : ret = PTR_ERR(em);
10836 0 : goto out;
10837 : }
10838 :
10839 0 : if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10840 0 : btrfs_warn(fs_info,
10841 : "swapfile must have single data profile");
10842 0 : ret = -EINVAL;
10843 0 : goto out;
10844 : }
10845 :
10846 0 : if (device == NULL) {
10847 0 : device = em->map_lookup->stripes[0].dev;
10848 0 : ret = btrfs_add_swapfile_pin(inode, device, false);
10849 0 : if (ret == 1)
10850 : ret = 0;
10851 0 : else if (ret)
10852 0 : goto out;
10853 0 : } else if (device != em->map_lookup->stripes[0].dev) {
10854 0 : btrfs_warn(fs_info, "swapfile must be on one device");
10855 0 : ret = -EINVAL;
10856 0 : goto out;
10857 : }
10858 :
10859 0 : physical_block_start = (em->map_lookup->stripes[0].physical +
10860 0 : (logical_block_start - em->start));
10861 0 : len = min(len, em->len - (logical_block_start - em->start));
10862 0 : free_extent_map(em);
10863 0 : em = NULL;
10864 :
10865 0 : bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10866 0 : if (!bg) {
10867 0 : btrfs_warn(fs_info,
10868 : "could not find block group containing swapfile");
10869 0 : ret = -EINVAL;
10870 0 : goto out;
10871 : }
10872 :
10873 0 : if (!btrfs_inc_block_group_swap_extents(bg)) {
10874 0 : btrfs_warn(fs_info,
10875 : "block group for swapfile at %llu is read-only%s",
10876 : bg->start,
10877 : atomic_read(&fs_info->scrubs_running) ?
10878 : " (scrub running)" : "");
10879 0 : btrfs_put_block_group(bg);
10880 0 : ret = -EINVAL;
10881 0 : goto out;
10882 : }
10883 :
10884 0 : ret = btrfs_add_swapfile_pin(inode, bg, true);
10885 0 : if (ret) {
10886 0 : btrfs_put_block_group(bg);
10887 0 : if (ret == 1)
10888 : ret = 0;
10889 : else
10890 0 : goto out;
10891 : }
10892 :
10893 0 : if (bsi.block_len &&
10894 0 : bsi.block_start + bsi.block_len == physical_block_start) {
10895 0 : bsi.block_len += len;
10896 : } else {
10897 0 : if (bsi.block_len) {
10898 0 : ret = btrfs_add_swap_extent(sis, &bsi);
10899 0 : if (ret)
10900 0 : goto out;
10901 : }
10902 0 : bsi.start = start;
10903 0 : bsi.block_start = physical_block_start;
10904 0 : bsi.block_len = len;
10905 : }
10906 :
10907 0 : start += len;
10908 : }
10909 :
10910 0 : if (bsi.block_len)
10911 0 : ret = btrfs_add_swap_extent(sis, &bsi);
10912 :
10913 0 : out:
10914 0 : if (!IS_ERR_OR_NULL(em))
10915 0 : free_extent_map(em);
10916 :
10917 0 : unlock_extent(io_tree, 0, isize - 1, &cached_state);
10918 :
10919 0 : if (ret)
10920 0 : btrfs_swap_deactivate(file);
10921 :
10922 0 : btrfs_drew_write_unlock(&root->snapshot_lock);
10923 :
10924 0 : btrfs_exclop_finish(fs_info);
10925 :
10926 0 : if (ret)
10927 : return ret;
10928 :
10929 0 : if (device)
10930 0 : sis->bdev = device->bdev;
10931 0 : *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10932 0 : sis->max = bsi.nr_pages;
10933 0 : sis->pages = bsi.nr_pages - 1;
10934 0 : sis->highest_bit = bsi.nr_pages - 1;
10935 0 : return bsi.nr_extents;
10936 : }
10937 : #else
10938 : static void btrfs_swap_deactivate(struct file *file)
10939 : {
10940 : }
10941 :
10942 : static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10943 : sector_t *span)
10944 : {
10945 : return -EOPNOTSUPP;
10946 : }
10947 : #endif
10948 :
10949 : /*
10950 : * Update the number of bytes used in the VFS' inode. When we replace extents in
10951 : * a range (clone, dedupe, fallocate's zero range), we must update the number of
10952 : * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10953 : * always get a correct value.
10954 : */
10955 0 : void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10956 : const u64 add_bytes,
10957 : const u64 del_bytes)
10958 : {
10959 0 : if (add_bytes == del_bytes)
10960 : return;
10961 :
10962 0 : spin_lock(&inode->lock);
10963 0 : if (del_bytes > 0)
10964 0 : inode_sub_bytes(&inode->vfs_inode, del_bytes);
10965 0 : if (add_bytes > 0)
10966 0 : inode_add_bytes(&inode->vfs_inode, add_bytes);
10967 0 : spin_unlock(&inode->lock);
10968 : }
10969 :
10970 : /*
10971 : * Verify that there are no ordered extents for a given file range.
10972 : *
10973 : * @inode: The target inode.
10974 : * @start: Start offset of the file range, should be sector size aligned.
10975 : * @end: End offset (inclusive) of the file range, its value +1 should be
10976 : * sector size aligned.
10977 : *
10978 : * This should typically be used for cases where we locked an inode's VFS lock in
10979 : * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10980 : * we have flushed all delalloc in the range, we have waited for all ordered
10981 : * extents in the range to complete and finally we have locked the file range in
10982 : * the inode's io_tree.
10983 : */
10984 0 : void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10985 : {
10986 0 : struct btrfs_root *root = inode->root;
10987 0 : struct btrfs_ordered_extent *ordered;
10988 :
10989 0 : if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10990 0 : return;
10991 :
10992 : ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10993 : if (ordered) {
10994 : btrfs_err(root->fs_info,
10995 : "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10996 : start, end, btrfs_ino(inode), root->root_key.objectid,
10997 : ordered->file_offset,
10998 : ordered->file_offset + ordered->num_bytes - 1);
10999 : btrfs_put_ordered_extent(ordered);
11000 : }
11001 :
11002 0 : ASSERT(ordered == NULL);
11003 : }
11004 :
11005 : static const struct inode_operations btrfs_dir_inode_operations = {
11006 : .getattr = btrfs_getattr,
11007 : .lookup = btrfs_lookup,
11008 : .create = btrfs_create,
11009 : .unlink = btrfs_unlink,
11010 : .link = btrfs_link,
11011 : .mkdir = btrfs_mkdir,
11012 : .rmdir = btrfs_rmdir,
11013 : .rename = btrfs_rename2,
11014 : .symlink = btrfs_symlink,
11015 : .setattr = btrfs_setattr,
11016 : .mknod = btrfs_mknod,
11017 : .listxattr = btrfs_listxattr,
11018 : .permission = btrfs_permission,
11019 : .get_inode_acl = btrfs_get_acl,
11020 : .set_acl = btrfs_set_acl,
11021 : .update_time = btrfs_update_time,
11022 : .tmpfile = btrfs_tmpfile,
11023 : .fileattr_get = btrfs_fileattr_get,
11024 : .fileattr_set = btrfs_fileattr_set,
11025 : };
11026 :
11027 : static const struct file_operations btrfs_dir_file_operations = {
11028 : .llseek = generic_file_llseek,
11029 : .read = generic_read_dir,
11030 : .iterate_shared = btrfs_real_readdir,
11031 : .open = btrfs_opendir,
11032 : .unlocked_ioctl = btrfs_ioctl,
11033 : #ifdef CONFIG_COMPAT
11034 : .compat_ioctl = btrfs_compat_ioctl,
11035 : #endif
11036 : .release = btrfs_release_file,
11037 : .fsync = btrfs_sync_file,
11038 : };
11039 :
11040 : /*
11041 : * btrfs doesn't support the bmap operation because swapfiles
11042 : * use bmap to make a mapping of extents in the file. They assume
11043 : * these extents won't change over the life of the file and they
11044 : * use the bmap result to do IO directly to the drive.
11045 : *
11046 : * the btrfs bmap call would return logical addresses that aren't
11047 : * suitable for IO and they also will change frequently as COW
11048 : * operations happen. So, swapfile + btrfs == corruption.
11049 : *
11050 : * For now we're avoiding this by dropping bmap.
11051 : */
11052 : static const struct address_space_operations btrfs_aops = {
11053 : .read_folio = btrfs_read_folio,
11054 : .writepages = btrfs_writepages,
11055 : .readahead = btrfs_readahead,
11056 : .invalidate_folio = btrfs_invalidate_folio,
11057 : .release_folio = btrfs_release_folio,
11058 : .migrate_folio = btrfs_migrate_folio,
11059 : .dirty_folio = filemap_dirty_folio,
11060 : .error_remove_page = generic_error_remove_page,
11061 : .swap_activate = btrfs_swap_activate,
11062 : .swap_deactivate = btrfs_swap_deactivate,
11063 : };
11064 :
11065 : static const struct inode_operations btrfs_file_inode_operations = {
11066 : .getattr = btrfs_getattr,
11067 : .setattr = btrfs_setattr,
11068 : .listxattr = btrfs_listxattr,
11069 : .permission = btrfs_permission,
11070 : .fiemap = btrfs_fiemap,
11071 : .get_inode_acl = btrfs_get_acl,
11072 : .set_acl = btrfs_set_acl,
11073 : .update_time = btrfs_update_time,
11074 : .fileattr_get = btrfs_fileattr_get,
11075 : .fileattr_set = btrfs_fileattr_set,
11076 : };
11077 : static const struct inode_operations btrfs_special_inode_operations = {
11078 : .getattr = btrfs_getattr,
11079 : .setattr = btrfs_setattr,
11080 : .permission = btrfs_permission,
11081 : .listxattr = btrfs_listxattr,
11082 : .get_inode_acl = btrfs_get_acl,
11083 : .set_acl = btrfs_set_acl,
11084 : .update_time = btrfs_update_time,
11085 : };
11086 : static const struct inode_operations btrfs_symlink_inode_operations = {
11087 : .get_link = page_get_link,
11088 : .getattr = btrfs_getattr,
11089 : .setattr = btrfs_setattr,
11090 : .permission = btrfs_permission,
11091 : .listxattr = btrfs_listxattr,
11092 : .update_time = btrfs_update_time,
11093 : };
11094 :
11095 : const struct dentry_operations btrfs_dentry_operations = {
11096 : .d_delete = btrfs_dentry_delete,
11097 : };
|
