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/file.h>
10 #include <linux/fs.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/writeback.h>
18 #include <linux/compat.h>
19 #include <linux/xattr.h>
20 #include <linux/posix_acl.h>
21 #include <linux/falloc.h>
22 #include <linux/slab.h>
23 #include <linux/ratelimit.h>
24 #include <linux/btrfs.h>
25 #include <linux/blkdev.h>
26 #include <linux/posix_acl_xattr.h>
27 #include <linux/uio.h>
28 #include <linux/magic.h>
29 #include <linux/iversion.h>
30 #include <linux/swap.h>
31 #include <linux/migrate.h>
32 #include <linux/sched/mm.h>
33 #include <linux/iomap.h>
34 #include <asm/unaligned.h>
35 #include <linux/fsverity.h>
36 #include "misc.h"
37 #include "ctree.h"
38 #include "disk-io.h"
39 #include "transaction.h"
40 #include "btrfs_inode.h"
41 #include "print-tree.h"
42 #include "ordered-data.h"
43 #include "xattr.h"
44 #include "tree-log.h"
45 #include "volumes.h"
46 #include "compression.h"
47 #include "locking.h"
48 #include "free-space-cache.h"
49 #include "props.h"
50 #include "qgroup.h"
51 #include "delalloc-space.h"
52 #include "block-group.h"
53 #include "space-info.h"
54 #include "zoned.h"
55 #include "subpage.h"
56
57 struct btrfs_iget_args {
58 u64 ino;
59 struct btrfs_root *root;
60 };
61
62 struct btrfs_dio_data {
63 u64 reserve;
64 loff_t length;
65 ssize_t submitted;
66 struct extent_changeset *data_reserved;
67 };
68
69 static const struct inode_operations btrfs_dir_inode_operations;
70 static const struct inode_operations btrfs_symlink_inode_operations;
71 static const struct inode_operations btrfs_special_inode_operations;
72 static const struct inode_operations btrfs_file_inode_operations;
73 static const struct address_space_operations btrfs_aops;
74 static const struct file_operations btrfs_dir_file_operations;
75
76 static struct kmem_cache *btrfs_inode_cachep;
77 struct kmem_cache *btrfs_trans_handle_cachep;
78 struct kmem_cache *btrfs_path_cachep;
79 struct kmem_cache *btrfs_free_space_cachep;
80 struct kmem_cache *btrfs_free_space_bitmap_cachep;
81
82 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
83 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
84 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
85 static noinline int cow_file_range(struct btrfs_inode *inode,
86 struct page *locked_page,
87 u64 start, u64 end, int *page_started,
88 unsigned long *nr_written, int unlock);
89 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
90 u64 len, u64 orig_start, u64 block_start,
91 u64 block_len, u64 orig_block_len,
92 u64 ram_bytes, int compress_type,
93 int type);
94
95 static void __endio_write_update_ordered(struct btrfs_inode *inode,
96 const u64 offset, const u64 bytes,
97 const bool uptodate);
98
99 /*
100 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
101 *
102 * ilock_flags can have the following bit set:
103 *
104 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
105 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
106 * return -EAGAIN
107 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
108 */
btrfs_inode_lock(struct inode * inode,unsigned int ilock_flags)109 int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags)
110 {
111 if (ilock_flags & BTRFS_ILOCK_SHARED) {
112 if (ilock_flags & BTRFS_ILOCK_TRY) {
113 if (!inode_trylock_shared(inode))
114 return -EAGAIN;
115 else
116 return 0;
117 }
118 inode_lock_shared(inode);
119 } else {
120 if (ilock_flags & BTRFS_ILOCK_TRY) {
121 if (!inode_trylock(inode))
122 return -EAGAIN;
123 else
124 return 0;
125 }
126 inode_lock(inode);
127 }
128 if (ilock_flags & BTRFS_ILOCK_MMAP)
129 down_write(&BTRFS_I(inode)->i_mmap_lock);
130 return 0;
131 }
132
133 /*
134 * btrfs_inode_unlock - unock inode i_rwsem
135 *
136 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
137 * to decide whether the lock acquired is shared or exclusive.
138 */
btrfs_inode_unlock(struct inode * inode,unsigned int ilock_flags)139 void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags)
140 {
141 if (ilock_flags & BTRFS_ILOCK_MMAP)
142 up_write(&BTRFS_I(inode)->i_mmap_lock);
143 if (ilock_flags & BTRFS_ILOCK_SHARED)
144 inode_unlock_shared(inode);
145 else
146 inode_unlock(inode);
147 }
148
149 /*
150 * Cleanup all submitted ordered extents in specified range to handle errors
151 * from the btrfs_run_delalloc_range() callback.
152 *
153 * NOTE: caller must ensure that when an error happens, it can not call
154 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
155 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
156 * to be released, which we want to happen only when finishing the ordered
157 * extent (btrfs_finish_ordered_io()).
158 */
btrfs_cleanup_ordered_extents(struct btrfs_inode * inode,struct page * locked_page,u64 offset,u64 bytes)159 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
160 struct page *locked_page,
161 u64 offset, u64 bytes)
162 {
163 unsigned long index = offset >> PAGE_SHIFT;
164 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
165 u64 page_start = page_offset(locked_page);
166 u64 page_end = page_start + PAGE_SIZE - 1;
167
168 struct page *page;
169
170 while (index <= end_index) {
171 /*
172 * For locked page, we will call end_extent_writepage() on it
173 * in run_delalloc_range() for the error handling. That
174 * end_extent_writepage() function will call
175 * btrfs_mark_ordered_io_finished() to clear page Ordered and
176 * run the ordered extent accounting.
177 *
178 * Here we can't just clear the Ordered bit, or
179 * btrfs_mark_ordered_io_finished() would skip the accounting
180 * for the page range, and the ordered extent will never finish.
181 */
182 if (index == (page_offset(locked_page) >> PAGE_SHIFT)) {
183 index++;
184 continue;
185 }
186 page = find_get_page(inode->vfs_inode.i_mapping, index);
187 index++;
188 if (!page)
189 continue;
190
191 /*
192 * Here we just clear all Ordered bits for every page in the
193 * range, then __endio_write_update_ordered() will handle
194 * the ordered extent accounting for the range.
195 */
196 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
197 offset, bytes);
198 put_page(page);
199 }
200
201 /* The locked page covers the full range, nothing needs to be done */
202 if (bytes + offset <= page_offset(locked_page) + PAGE_SIZE)
203 return;
204 /*
205 * In case this page belongs to the delalloc range being instantiated
206 * then skip it, since the first page of a range is going to be
207 * properly cleaned up by the caller of run_delalloc_range
208 */
209 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
210 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
211 offset = page_offset(locked_page) + PAGE_SIZE;
212 }
213
214 return __endio_write_update_ordered(inode, offset, bytes, false);
215 }
216
217 static int btrfs_dirty_inode(struct inode *inode);
218
btrfs_init_inode_security(struct btrfs_trans_handle * trans,struct inode * inode,struct inode * dir,const struct qstr * qstr)219 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
220 struct inode *inode, struct inode *dir,
221 const struct qstr *qstr)
222 {
223 int err;
224
225 err = btrfs_init_acl(trans, inode, dir);
226 if (!err)
227 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
228 return err;
229 }
230
231 /*
232 * this does all the hard work for inserting an inline extent into
233 * the btree. The caller should have done a btrfs_drop_extents so that
234 * no overlapping inline items exist in the btree
235 */
insert_inline_extent(struct btrfs_trans_handle * trans,struct btrfs_path * path,bool extent_inserted,struct btrfs_root * root,struct inode * inode,u64 start,size_t size,size_t compressed_size,int compress_type,struct page ** compressed_pages)236 static int insert_inline_extent(struct btrfs_trans_handle *trans,
237 struct btrfs_path *path, bool extent_inserted,
238 struct btrfs_root *root, struct inode *inode,
239 u64 start, size_t size, size_t compressed_size,
240 int compress_type,
241 struct page **compressed_pages)
242 {
243 struct extent_buffer *leaf;
244 struct page *page = NULL;
245 char *kaddr;
246 unsigned long ptr;
247 struct btrfs_file_extent_item *ei;
248 int ret;
249 size_t cur_size = size;
250 unsigned long offset;
251
252 ASSERT((compressed_size > 0 && compressed_pages) ||
253 (compressed_size == 0 && !compressed_pages));
254
255 if (compressed_size && compressed_pages)
256 cur_size = compressed_size;
257
258 if (!extent_inserted) {
259 struct btrfs_key key;
260 size_t datasize;
261
262 key.objectid = btrfs_ino(BTRFS_I(inode));
263 key.offset = start;
264 key.type = BTRFS_EXTENT_DATA_KEY;
265
266 datasize = btrfs_file_extent_calc_inline_size(cur_size);
267 ret = btrfs_insert_empty_item(trans, root, path, &key,
268 datasize);
269 if (ret)
270 goto fail;
271 }
272 leaf = path->nodes[0];
273 ei = btrfs_item_ptr(leaf, path->slots[0],
274 struct btrfs_file_extent_item);
275 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
276 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
277 btrfs_set_file_extent_encryption(leaf, ei, 0);
278 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
279 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
280 ptr = btrfs_file_extent_inline_start(ei);
281
282 if (compress_type != BTRFS_COMPRESS_NONE) {
283 struct page *cpage;
284 int i = 0;
285 while (compressed_size > 0) {
286 cpage = compressed_pages[i];
287 cur_size = min_t(unsigned long, compressed_size,
288 PAGE_SIZE);
289
290 kaddr = kmap_atomic(cpage);
291 write_extent_buffer(leaf, kaddr, ptr, cur_size);
292 kunmap_atomic(kaddr);
293
294 i++;
295 ptr += cur_size;
296 compressed_size -= cur_size;
297 }
298 btrfs_set_file_extent_compression(leaf, ei,
299 compress_type);
300 } else {
301 page = find_get_page(inode->i_mapping,
302 start >> PAGE_SHIFT);
303 btrfs_set_file_extent_compression(leaf, ei, 0);
304 kaddr = kmap_atomic(page);
305 offset = offset_in_page(start);
306 write_extent_buffer(leaf, kaddr + offset, ptr, size);
307 kunmap_atomic(kaddr);
308 put_page(page);
309 }
310 btrfs_mark_buffer_dirty(leaf);
311 btrfs_release_path(path);
312
313 /*
314 * We align size to sectorsize for inline extents just for simplicity
315 * sake.
316 */
317 size = ALIGN(size, root->fs_info->sectorsize);
318 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
319 if (ret)
320 goto fail;
321
322 /*
323 * we're an inline extent, so nobody can
324 * extend the file past i_size without locking
325 * a page we already have locked.
326 *
327 * We must do any isize and inode updates
328 * before we unlock the pages. Otherwise we
329 * could end up racing with unlink.
330 */
331 BTRFS_I(inode)->disk_i_size = inode->i_size;
332 fail:
333 return ret;
334 }
335
336
337 /*
338 * conditionally insert an inline extent into the file. This
339 * does the checks required to make sure the data is small enough
340 * to fit as an inline extent.
341 */
cow_file_range_inline(struct btrfs_inode * inode,u64 start,u64 end,size_t compressed_size,int compress_type,struct page ** compressed_pages)342 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start,
343 u64 end, size_t compressed_size,
344 int compress_type,
345 struct page **compressed_pages)
346 {
347 struct btrfs_drop_extents_args drop_args = { 0 };
348 struct btrfs_root *root = inode->root;
349 struct btrfs_fs_info *fs_info = root->fs_info;
350 struct btrfs_trans_handle *trans;
351 u64 isize = i_size_read(&inode->vfs_inode);
352 u64 actual_end = min(end + 1, isize);
353 u64 inline_len = actual_end - start;
354 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
355 u64 data_len = inline_len;
356 int ret;
357 struct btrfs_path *path;
358
359 if (compressed_size)
360 data_len = compressed_size;
361
362 if (start > 0 ||
363 actual_end > fs_info->sectorsize ||
364 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
365 (!compressed_size &&
366 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
367 end + 1 < isize ||
368 data_len > fs_info->max_inline) {
369 return 1;
370 }
371
372 path = btrfs_alloc_path();
373 if (!path)
374 return -ENOMEM;
375
376 trans = btrfs_join_transaction(root);
377 if (IS_ERR(trans)) {
378 btrfs_free_path(path);
379 return PTR_ERR(trans);
380 }
381 trans->block_rsv = &inode->block_rsv;
382
383 drop_args.path = path;
384 drop_args.start = start;
385 drop_args.end = aligned_end;
386 drop_args.drop_cache = true;
387 drop_args.replace_extent = true;
388
389 if (compressed_size && compressed_pages)
390 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
391 compressed_size);
392 else
393 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
394 inline_len);
395
396 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
397 if (ret) {
398 btrfs_abort_transaction(trans, ret);
399 goto out;
400 }
401
402 if (isize > actual_end)
403 inline_len = min_t(u64, isize, actual_end);
404 ret = insert_inline_extent(trans, path, drop_args.extent_inserted,
405 root, &inode->vfs_inode, start,
406 inline_len, compressed_size,
407 compress_type, compressed_pages);
408 if (ret && ret != -ENOSPC) {
409 btrfs_abort_transaction(trans, ret);
410 goto out;
411 } else if (ret == -ENOSPC) {
412 ret = 1;
413 goto out;
414 }
415
416 btrfs_update_inode_bytes(inode, inline_len, drop_args.bytes_found);
417 ret = btrfs_update_inode(trans, root, inode);
418 if (ret && ret != -ENOSPC) {
419 btrfs_abort_transaction(trans, ret);
420 goto out;
421 } else if (ret == -ENOSPC) {
422 ret = 1;
423 goto out;
424 }
425
426 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
427 out:
428 /*
429 * Don't forget to free the reserved space, as for inlined extent
430 * it won't count as data extent, free them directly here.
431 * And at reserve time, it's always aligned to page size, so
432 * just free one page here.
433 */
434 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
435 btrfs_free_path(path);
436 btrfs_end_transaction(trans);
437 return ret;
438 }
439
440 struct async_extent {
441 u64 start;
442 u64 ram_size;
443 u64 compressed_size;
444 struct page **pages;
445 unsigned long nr_pages;
446 int compress_type;
447 struct list_head list;
448 };
449
450 struct async_chunk {
451 struct inode *inode;
452 struct page *locked_page;
453 u64 start;
454 u64 end;
455 unsigned int write_flags;
456 struct list_head extents;
457 struct cgroup_subsys_state *blkcg_css;
458 struct btrfs_work work;
459 atomic_t *pending;
460 };
461
462 struct async_cow {
463 /* Number of chunks in flight; must be first in the structure */
464 atomic_t num_chunks;
465 struct async_chunk chunks[];
466 };
467
add_async_extent(struct async_chunk * cow,u64 start,u64 ram_size,u64 compressed_size,struct page ** pages,unsigned long nr_pages,int compress_type)468 static noinline int add_async_extent(struct async_chunk *cow,
469 u64 start, u64 ram_size,
470 u64 compressed_size,
471 struct page **pages,
472 unsigned long nr_pages,
473 int compress_type)
474 {
475 struct async_extent *async_extent;
476
477 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
478 BUG_ON(!async_extent); /* -ENOMEM */
479 async_extent->start = start;
480 async_extent->ram_size = ram_size;
481 async_extent->compressed_size = compressed_size;
482 async_extent->pages = pages;
483 async_extent->nr_pages = nr_pages;
484 async_extent->compress_type = compress_type;
485 list_add_tail(&async_extent->list, &cow->extents);
486 return 0;
487 }
488
489 /*
490 * Check if the inode has flags compatible with compression
491 */
inode_can_compress(struct btrfs_inode * inode)492 static inline bool inode_can_compress(struct btrfs_inode *inode)
493 {
494 /* Subpage doesn't support compression yet */
495 if (inode->root->fs_info->sectorsize < PAGE_SIZE)
496 return false;
497 if (inode->flags & BTRFS_INODE_NODATACOW ||
498 inode->flags & BTRFS_INODE_NODATASUM)
499 return false;
500 return true;
501 }
502
503 /*
504 * Check if the inode needs to be submitted to compression, based on mount
505 * options, defragmentation, properties or heuristics.
506 */
inode_need_compress(struct btrfs_inode * inode,u64 start,u64 end)507 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
508 u64 end)
509 {
510 struct btrfs_fs_info *fs_info = inode->root->fs_info;
511
512 if (!inode_can_compress(inode)) {
513 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
514 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
515 btrfs_ino(inode));
516 return 0;
517 }
518 /* force compress */
519 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
520 return 1;
521 /* defrag ioctl */
522 if (inode->defrag_compress)
523 return 1;
524 /* bad compression ratios */
525 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
526 return 0;
527 if (btrfs_test_opt(fs_info, COMPRESS) ||
528 inode->flags & BTRFS_INODE_COMPRESS ||
529 inode->prop_compress)
530 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
531 return 0;
532 }
533
inode_should_defrag(struct btrfs_inode * inode,u64 start,u64 end,u64 num_bytes,u64 small_write)534 static inline void inode_should_defrag(struct btrfs_inode *inode,
535 u64 start, u64 end, u64 num_bytes, u64 small_write)
536 {
537 /* If this is a small write inside eof, kick off a defrag */
538 if (num_bytes < small_write &&
539 (start > 0 || end + 1 < inode->disk_i_size))
540 btrfs_add_inode_defrag(NULL, inode);
541 }
542
543 /*
544 * we create compressed extents in two phases. The first
545 * phase compresses a range of pages that have already been
546 * locked (both pages and state bits are locked).
547 *
548 * This is done inside an ordered work queue, and the compression
549 * is spread across many cpus. The actual IO submission is step
550 * two, and the ordered work queue takes care of making sure that
551 * happens in the same order things were put onto the queue by
552 * writepages and friends.
553 *
554 * If this code finds it can't get good compression, it puts an
555 * entry onto the work queue to write the uncompressed bytes. This
556 * makes sure that both compressed inodes and uncompressed inodes
557 * are written in the same order that the flusher thread sent them
558 * down.
559 */
compress_file_range(struct async_chunk * async_chunk)560 static noinline int compress_file_range(struct async_chunk *async_chunk)
561 {
562 struct inode *inode = async_chunk->inode;
563 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
564 u64 blocksize = fs_info->sectorsize;
565 u64 start = async_chunk->start;
566 u64 end = async_chunk->end;
567 u64 actual_end;
568 u64 i_size;
569 int ret = 0;
570 struct page **pages = NULL;
571 unsigned long nr_pages;
572 unsigned long total_compressed = 0;
573 unsigned long total_in = 0;
574 int i;
575 int will_compress;
576 int compress_type = fs_info->compress_type;
577 int compressed_extents = 0;
578 int redirty = 0;
579
580 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
581 SZ_16K);
582
583 /*
584 * We need to save i_size before now because it could change in between
585 * us evaluating the size and assigning it. This is because we lock and
586 * unlock the page in truncate and fallocate, and then modify the i_size
587 * later on.
588 *
589 * The barriers are to emulate READ_ONCE, remove that once i_size_read
590 * does that for us.
591 */
592 barrier();
593 i_size = i_size_read(inode);
594 barrier();
595 actual_end = min_t(u64, i_size, end + 1);
596 again:
597 will_compress = 0;
598 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
599 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
600 nr_pages = min_t(unsigned long, nr_pages,
601 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
602
603 /*
604 * we don't want to send crud past the end of i_size through
605 * compression, that's just a waste of CPU time. So, if the
606 * end of the file is before the start of our current
607 * requested range of bytes, we bail out to the uncompressed
608 * cleanup code that can deal with all of this.
609 *
610 * It isn't really the fastest way to fix things, but this is a
611 * very uncommon corner.
612 */
613 if (actual_end <= start)
614 goto cleanup_and_bail_uncompressed;
615
616 total_compressed = actual_end - start;
617
618 /*
619 * skip compression for a small file range(<=blocksize) that
620 * isn't an inline extent, since it doesn't save disk space at all.
621 */
622 if (total_compressed <= blocksize &&
623 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
624 goto cleanup_and_bail_uncompressed;
625
626 total_compressed = min_t(unsigned long, total_compressed,
627 BTRFS_MAX_UNCOMPRESSED);
628 total_in = 0;
629 ret = 0;
630
631 /*
632 * we do compression for mount -o compress and when the
633 * inode has not been flagged as nocompress. This flag can
634 * change at any time if we discover bad compression ratios.
635 */
636 if (inode_need_compress(BTRFS_I(inode), start, end)) {
637 WARN_ON(pages);
638 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
639 if (!pages) {
640 /* just bail out to the uncompressed code */
641 nr_pages = 0;
642 goto cont;
643 }
644
645 if (BTRFS_I(inode)->defrag_compress)
646 compress_type = BTRFS_I(inode)->defrag_compress;
647 else if (BTRFS_I(inode)->prop_compress)
648 compress_type = BTRFS_I(inode)->prop_compress;
649
650 /*
651 * we need to call clear_page_dirty_for_io on each
652 * page in the range. Otherwise applications with the file
653 * mmap'd can wander in and change the page contents while
654 * we are compressing them.
655 *
656 * If the compression fails for any reason, we set the pages
657 * dirty again later on.
658 *
659 * Note that the remaining part is redirtied, the start pointer
660 * has moved, the end is the original one.
661 */
662 if (!redirty) {
663 extent_range_clear_dirty_for_io(inode, start, end);
664 redirty = 1;
665 }
666
667 /* Compression level is applied here and only here */
668 ret = btrfs_compress_pages(
669 compress_type | (fs_info->compress_level << 4),
670 inode->i_mapping, start,
671 pages,
672 &nr_pages,
673 &total_in,
674 &total_compressed);
675
676 if (!ret) {
677 unsigned long offset = offset_in_page(total_compressed);
678 struct page *page = pages[nr_pages - 1];
679
680 /* zero the tail end of the last page, we might be
681 * sending it down to disk
682 */
683 if (offset)
684 memzero_page(page, offset, PAGE_SIZE - offset);
685 will_compress = 1;
686 }
687 }
688 cont:
689 /*
690 * Check cow_file_range() for why we don't even try to create inline
691 * extent for subpage case.
692 */
693 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
694 /* lets try to make an inline extent */
695 if (ret || total_in < actual_end) {
696 /* we didn't compress the entire range, try
697 * to make an uncompressed inline extent.
698 */
699 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
700 0, BTRFS_COMPRESS_NONE,
701 NULL);
702 } else {
703 /* try making a compressed inline extent */
704 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
705 total_compressed,
706 compress_type, pages);
707 }
708 if (ret <= 0) {
709 unsigned long clear_flags = EXTENT_DELALLOC |
710 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
711 EXTENT_DO_ACCOUNTING;
712 unsigned long page_error_op;
713
714 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
715
716 /*
717 * inline extent creation worked or returned error,
718 * we don't need to create any more async work items.
719 * Unlock and free up our temp pages.
720 *
721 * We use DO_ACCOUNTING here because we need the
722 * delalloc_release_metadata to be done _after_ we drop
723 * our outstanding extent for clearing delalloc for this
724 * range.
725 */
726 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
727 NULL,
728 clear_flags,
729 PAGE_UNLOCK |
730 PAGE_START_WRITEBACK |
731 page_error_op |
732 PAGE_END_WRITEBACK);
733
734 /*
735 * Ensure we only free the compressed pages if we have
736 * them allocated, as we can still reach here with
737 * inode_need_compress() == false.
738 */
739 if (pages) {
740 for (i = 0; i < nr_pages; i++) {
741 WARN_ON(pages[i]->mapping);
742 put_page(pages[i]);
743 }
744 kfree(pages);
745 }
746 return 0;
747 }
748 }
749
750 if (will_compress) {
751 /*
752 * we aren't doing an inline extent round the compressed size
753 * up to a block size boundary so the allocator does sane
754 * things
755 */
756 total_compressed = ALIGN(total_compressed, blocksize);
757
758 /*
759 * one last check to make sure the compression is really a
760 * win, compare the page count read with the blocks on disk,
761 * compression must free at least one sector size
762 */
763 total_in = ALIGN(total_in, PAGE_SIZE);
764 if (total_compressed + blocksize <= total_in) {
765 compressed_extents++;
766
767 /*
768 * The async work queues will take care of doing actual
769 * allocation on disk for these compressed pages, and
770 * will submit them to the elevator.
771 */
772 add_async_extent(async_chunk, start, total_in,
773 total_compressed, pages, nr_pages,
774 compress_type);
775
776 if (start + total_in < end) {
777 start += total_in;
778 pages = NULL;
779 cond_resched();
780 goto again;
781 }
782 return compressed_extents;
783 }
784 }
785 if (pages) {
786 /*
787 * the compression code ran but failed to make things smaller,
788 * free any pages it allocated and our page pointer array
789 */
790 for (i = 0; i < nr_pages; i++) {
791 WARN_ON(pages[i]->mapping);
792 put_page(pages[i]);
793 }
794 kfree(pages);
795 pages = NULL;
796 total_compressed = 0;
797 nr_pages = 0;
798
799 /* flag the file so we don't compress in the future */
800 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
801 !(BTRFS_I(inode)->prop_compress)) {
802 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
803 }
804 }
805 cleanup_and_bail_uncompressed:
806 /*
807 * No compression, but we still need to write the pages in the file
808 * we've been given so far. redirty the locked page if it corresponds
809 * to our extent and set things up for the async work queue to run
810 * cow_file_range to do the normal delalloc dance.
811 */
812 if (async_chunk->locked_page &&
813 (page_offset(async_chunk->locked_page) >= start &&
814 page_offset(async_chunk->locked_page)) <= end) {
815 __set_page_dirty_nobuffers(async_chunk->locked_page);
816 /* unlocked later on in the async handlers */
817 }
818
819 if (redirty)
820 extent_range_redirty_for_io(inode, start, end);
821 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
822 BTRFS_COMPRESS_NONE);
823 compressed_extents++;
824
825 return compressed_extents;
826 }
827
free_async_extent_pages(struct async_extent * async_extent)828 static void free_async_extent_pages(struct async_extent *async_extent)
829 {
830 int i;
831
832 if (!async_extent->pages)
833 return;
834
835 for (i = 0; i < async_extent->nr_pages; i++) {
836 WARN_ON(async_extent->pages[i]->mapping);
837 put_page(async_extent->pages[i]);
838 }
839 kfree(async_extent->pages);
840 async_extent->nr_pages = 0;
841 async_extent->pages = NULL;
842 }
843
844 /*
845 * phase two of compressed writeback. This is the ordered portion
846 * of the code, which only gets called in the order the work was
847 * queued. We walk all the async extents created by compress_file_range
848 * and send them down to the disk.
849 */
submit_compressed_extents(struct async_chunk * async_chunk)850 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
851 {
852 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
853 struct btrfs_fs_info *fs_info = inode->root->fs_info;
854 struct async_extent *async_extent;
855 u64 alloc_hint = 0;
856 struct btrfs_key ins;
857 struct extent_map *em;
858 struct btrfs_root *root = inode->root;
859 struct extent_io_tree *io_tree = &inode->io_tree;
860 int ret = 0;
861
862 again:
863 while (!list_empty(&async_chunk->extents)) {
864 async_extent = list_entry(async_chunk->extents.next,
865 struct async_extent, list);
866 list_del(&async_extent->list);
867
868 retry:
869 lock_extent(io_tree, async_extent->start,
870 async_extent->start + async_extent->ram_size - 1);
871 /* did the compression code fall back to uncompressed IO? */
872 if (!async_extent->pages) {
873 int page_started = 0;
874 unsigned long nr_written = 0;
875
876 /* allocate blocks */
877 ret = cow_file_range(inode, async_chunk->locked_page,
878 async_extent->start,
879 async_extent->start +
880 async_extent->ram_size - 1,
881 &page_started, &nr_written, 0);
882
883 /* JDM XXX */
884
885 /*
886 * if page_started, cow_file_range inserted an
887 * inline extent and took care of all the unlocking
888 * and IO for us. Otherwise, we need to submit
889 * all those pages down to the drive.
890 */
891 if (!page_started && !ret)
892 extent_write_locked_range(&inode->vfs_inode,
893 async_extent->start,
894 async_extent->start +
895 async_extent->ram_size - 1,
896 WB_SYNC_ALL);
897 else if (ret && async_chunk->locked_page)
898 unlock_page(async_chunk->locked_page);
899 kfree(async_extent);
900 cond_resched();
901 continue;
902 }
903
904 ret = btrfs_reserve_extent(root, async_extent->ram_size,
905 async_extent->compressed_size,
906 async_extent->compressed_size,
907 0, alloc_hint, &ins, 1, 1);
908 if (ret) {
909 free_async_extent_pages(async_extent);
910
911 if (ret == -ENOSPC) {
912 unlock_extent(io_tree, async_extent->start,
913 async_extent->start +
914 async_extent->ram_size - 1);
915
916 /*
917 * we need to redirty the pages if we decide to
918 * fallback to uncompressed IO, otherwise we
919 * will not submit these pages down to lower
920 * layers.
921 */
922 extent_range_redirty_for_io(&inode->vfs_inode,
923 async_extent->start,
924 async_extent->start +
925 async_extent->ram_size - 1);
926
927 goto retry;
928 }
929 goto out_free;
930 }
931 /*
932 * here we're doing allocation and writeback of the
933 * compressed pages
934 */
935 em = create_io_em(inode, async_extent->start,
936 async_extent->ram_size, /* len */
937 async_extent->start, /* orig_start */
938 ins.objectid, /* block_start */
939 ins.offset, /* block_len */
940 ins.offset, /* orig_block_len */
941 async_extent->ram_size, /* ram_bytes */
942 async_extent->compress_type,
943 BTRFS_ORDERED_COMPRESSED);
944 if (IS_ERR(em))
945 /* ret value is not necessary due to void function */
946 goto out_free_reserve;
947 free_extent_map(em);
948
949 ret = btrfs_add_ordered_extent_compress(inode,
950 async_extent->start,
951 ins.objectid,
952 async_extent->ram_size,
953 ins.offset,
954 async_extent->compress_type);
955 if (ret) {
956 btrfs_drop_extent_cache(inode, async_extent->start,
957 async_extent->start +
958 async_extent->ram_size - 1, 0);
959 goto out_free_reserve;
960 }
961 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
962
963 /*
964 * clear dirty, set writeback and unlock the pages.
965 */
966 extent_clear_unlock_delalloc(inode, async_extent->start,
967 async_extent->start +
968 async_extent->ram_size - 1,
969 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
970 PAGE_UNLOCK | PAGE_START_WRITEBACK);
971 if (btrfs_submit_compressed_write(inode, async_extent->start,
972 async_extent->ram_size,
973 ins.objectid,
974 ins.offset, async_extent->pages,
975 async_extent->nr_pages,
976 async_chunk->write_flags,
977 async_chunk->blkcg_css)) {
978 struct page *p = async_extent->pages[0];
979 const u64 start = async_extent->start;
980 const u64 end = start + async_extent->ram_size - 1;
981
982 p->mapping = inode->vfs_inode.i_mapping;
983 btrfs_writepage_endio_finish_ordered(inode, p, start,
984 end, false);
985
986 p->mapping = NULL;
987 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
988 PAGE_END_WRITEBACK |
989 PAGE_SET_ERROR);
990 free_async_extent_pages(async_extent);
991 }
992 alloc_hint = ins.objectid + ins.offset;
993 kfree(async_extent);
994 cond_resched();
995 }
996 return;
997 out_free_reserve:
998 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
999 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1000 out_free:
1001 extent_clear_unlock_delalloc(inode, async_extent->start,
1002 async_extent->start +
1003 async_extent->ram_size - 1,
1004 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1005 EXTENT_DELALLOC_NEW |
1006 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1007 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1008 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1009 free_async_extent_pages(async_extent);
1010 kfree(async_extent);
1011 goto again;
1012 }
1013
get_extent_allocation_hint(struct btrfs_inode * inode,u64 start,u64 num_bytes)1014 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1015 u64 num_bytes)
1016 {
1017 struct extent_map_tree *em_tree = &inode->extent_tree;
1018 struct extent_map *em;
1019 u64 alloc_hint = 0;
1020
1021 read_lock(&em_tree->lock);
1022 em = search_extent_mapping(em_tree, start, num_bytes);
1023 if (em) {
1024 /*
1025 * if block start isn't an actual block number then find the
1026 * first block in this inode and use that as a hint. If that
1027 * block is also bogus then just don't worry about it.
1028 */
1029 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1030 free_extent_map(em);
1031 em = search_extent_mapping(em_tree, 0, 0);
1032 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1033 alloc_hint = em->block_start;
1034 if (em)
1035 free_extent_map(em);
1036 } else {
1037 alloc_hint = em->block_start;
1038 free_extent_map(em);
1039 }
1040 }
1041 read_unlock(&em_tree->lock);
1042
1043 return alloc_hint;
1044 }
1045
1046 /*
1047 * when extent_io.c finds a delayed allocation range in the file,
1048 * the call backs end up in this code. The basic idea is to
1049 * allocate extents on disk for the range, and create ordered data structs
1050 * in ram to track those extents.
1051 *
1052 * locked_page is the page that writepage had locked already. We use
1053 * it to make sure we don't do extra locks or unlocks.
1054 *
1055 * *page_started is set to one if we unlock locked_page and do everything
1056 * required to start IO on it. It may be clean and already done with
1057 * IO when we return.
1058 */
cow_file_range(struct btrfs_inode * inode,struct page * locked_page,u64 start,u64 end,int * page_started,unsigned long * nr_written,int unlock)1059 static noinline int cow_file_range(struct btrfs_inode *inode,
1060 struct page *locked_page,
1061 u64 start, u64 end, int *page_started,
1062 unsigned long *nr_written, int unlock)
1063 {
1064 struct btrfs_root *root = inode->root;
1065 struct btrfs_fs_info *fs_info = root->fs_info;
1066 u64 alloc_hint = 0;
1067 u64 num_bytes;
1068 unsigned long ram_size;
1069 u64 cur_alloc_size = 0;
1070 u64 min_alloc_size;
1071 u64 blocksize = fs_info->sectorsize;
1072 struct btrfs_key ins;
1073 struct extent_map *em;
1074 unsigned clear_bits;
1075 unsigned long page_ops;
1076 bool extent_reserved = false;
1077 int ret = 0;
1078
1079 if (btrfs_is_free_space_inode(inode)) {
1080 WARN_ON_ONCE(1);
1081 ret = -EINVAL;
1082 goto out_unlock;
1083 }
1084
1085 num_bytes = ALIGN(end - start + 1, blocksize);
1086 num_bytes = max(blocksize, num_bytes);
1087 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1088
1089 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1090
1091 /*
1092 * Due to the page size limit, for subpage we can only trigger the
1093 * writeback for the dirty sectors of page, that means data writeback
1094 * is doing more writeback than what we want.
1095 *
1096 * This is especially unexpected for some call sites like fallocate,
1097 * where we only increase i_size after everything is done.
1098 * This means we can trigger inline extent even if we didn't want to.
1099 * So here we skip inline extent creation completely.
1100 */
1101 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
1102 /* lets try to make an inline extent */
1103 ret = cow_file_range_inline(inode, start, end, 0,
1104 BTRFS_COMPRESS_NONE, NULL);
1105 if (ret == 0) {
1106 /*
1107 * We use DO_ACCOUNTING here because we need the
1108 * delalloc_release_metadata to be run _after_ we drop
1109 * our outstanding extent for clearing delalloc for this
1110 * range.
1111 */
1112 extent_clear_unlock_delalloc(inode, start, end,
1113 locked_page,
1114 EXTENT_LOCKED | EXTENT_DELALLOC |
1115 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1116 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1117 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1118 *nr_written = *nr_written +
1119 (end - start + PAGE_SIZE) / PAGE_SIZE;
1120 *page_started = 1;
1121 /*
1122 * locked_page is locked by the caller of
1123 * writepage_delalloc(), not locked by
1124 * __process_pages_contig().
1125 *
1126 * We can't let __process_pages_contig() to unlock it,
1127 * as it doesn't have any subpage::writers recorded.
1128 *
1129 * Here we manually unlock the page, since the caller
1130 * can't use page_started to determine if it's an
1131 * inline extent or a compressed extent.
1132 */
1133 unlock_page(locked_page);
1134 goto out;
1135 } else if (ret < 0) {
1136 goto out_unlock;
1137 }
1138 }
1139
1140 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1141 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1142
1143 /*
1144 * Relocation relies on the relocated extents to have exactly the same
1145 * size as the original extents. Normally writeback for relocation data
1146 * extents follows a NOCOW path because relocation preallocates the
1147 * extents. However, due to an operation such as scrub turning a block
1148 * group to RO mode, it may fallback to COW mode, so we must make sure
1149 * an extent allocated during COW has exactly the requested size and can
1150 * not be split into smaller extents, otherwise relocation breaks and
1151 * fails during the stage where it updates the bytenr of file extent
1152 * items.
1153 */
1154 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1155 min_alloc_size = num_bytes;
1156 else
1157 min_alloc_size = fs_info->sectorsize;
1158
1159 while (num_bytes > 0) {
1160 cur_alloc_size = num_bytes;
1161 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1162 min_alloc_size, 0, alloc_hint,
1163 &ins, 1, 1);
1164 if (ret < 0)
1165 goto out_unlock;
1166 cur_alloc_size = ins.offset;
1167 extent_reserved = true;
1168
1169 ram_size = ins.offset;
1170 em = create_io_em(inode, start, ins.offset, /* len */
1171 start, /* orig_start */
1172 ins.objectid, /* block_start */
1173 ins.offset, /* block_len */
1174 ins.offset, /* orig_block_len */
1175 ram_size, /* ram_bytes */
1176 BTRFS_COMPRESS_NONE, /* compress_type */
1177 BTRFS_ORDERED_REGULAR /* type */);
1178 if (IS_ERR(em)) {
1179 ret = PTR_ERR(em);
1180 goto out_reserve;
1181 }
1182 free_extent_map(em);
1183
1184 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1185 ram_size, cur_alloc_size,
1186 BTRFS_ORDERED_REGULAR);
1187 if (ret)
1188 goto out_drop_extent_cache;
1189
1190 if (root->root_key.objectid ==
1191 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1192 ret = btrfs_reloc_clone_csums(inode, start,
1193 cur_alloc_size);
1194 /*
1195 * Only drop cache here, and process as normal.
1196 *
1197 * We must not allow extent_clear_unlock_delalloc()
1198 * at out_unlock label to free meta of this ordered
1199 * extent, as its meta should be freed by
1200 * btrfs_finish_ordered_io().
1201 *
1202 * So we must continue until @start is increased to
1203 * skip current ordered extent.
1204 */
1205 if (ret)
1206 btrfs_drop_extent_cache(inode, start,
1207 start + ram_size - 1, 0);
1208 }
1209
1210 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1211
1212 /*
1213 * We're not doing compressed IO, don't unlock the first page
1214 * (which the caller expects to stay locked), don't clear any
1215 * dirty bits and don't set any writeback bits
1216 *
1217 * Do set the Ordered (Private2) bit so we know this page was
1218 * properly setup for writepage.
1219 */
1220 page_ops = unlock ? PAGE_UNLOCK : 0;
1221 page_ops |= PAGE_SET_ORDERED;
1222
1223 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1224 locked_page,
1225 EXTENT_LOCKED | EXTENT_DELALLOC,
1226 page_ops);
1227 if (num_bytes < cur_alloc_size)
1228 num_bytes = 0;
1229 else
1230 num_bytes -= cur_alloc_size;
1231 alloc_hint = ins.objectid + ins.offset;
1232 start += cur_alloc_size;
1233 extent_reserved = false;
1234
1235 /*
1236 * btrfs_reloc_clone_csums() error, since start is increased
1237 * extent_clear_unlock_delalloc() at out_unlock label won't
1238 * free metadata of current ordered extent, we're OK to exit.
1239 */
1240 if (ret)
1241 goto out_unlock;
1242 }
1243 out:
1244 return ret;
1245
1246 out_drop_extent_cache:
1247 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1248 out_reserve:
1249 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1250 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1251 out_unlock:
1252 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1253 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1254 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1255 /*
1256 * If we reserved an extent for our delalloc range (or a subrange) and
1257 * failed to create the respective ordered extent, then it means that
1258 * when we reserved the extent we decremented the extent's size from
1259 * the data space_info's bytes_may_use counter and incremented the
1260 * space_info's bytes_reserved counter by the same amount. We must make
1261 * sure extent_clear_unlock_delalloc() does not try to decrement again
1262 * the data space_info's bytes_may_use counter, therefore we do not pass
1263 * it the flag EXTENT_CLEAR_DATA_RESV.
1264 */
1265 if (extent_reserved) {
1266 extent_clear_unlock_delalloc(inode, start,
1267 start + cur_alloc_size - 1,
1268 locked_page,
1269 clear_bits,
1270 page_ops);
1271 start += cur_alloc_size;
1272 if (start >= end)
1273 goto out;
1274 }
1275 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1276 clear_bits | EXTENT_CLEAR_DATA_RESV,
1277 page_ops);
1278 goto out;
1279 }
1280
1281 /*
1282 * work queue call back to started compression on a file and pages
1283 */
async_cow_start(struct btrfs_work * work)1284 static noinline void async_cow_start(struct btrfs_work *work)
1285 {
1286 struct async_chunk *async_chunk;
1287 int compressed_extents;
1288
1289 async_chunk = container_of(work, struct async_chunk, work);
1290
1291 compressed_extents = compress_file_range(async_chunk);
1292 if (compressed_extents == 0) {
1293 btrfs_add_delayed_iput(async_chunk->inode);
1294 async_chunk->inode = NULL;
1295 }
1296 }
1297
1298 /*
1299 * work queue call back to submit previously compressed pages
1300 */
async_cow_submit(struct btrfs_work * work)1301 static noinline void async_cow_submit(struct btrfs_work *work)
1302 {
1303 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1304 work);
1305 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1306 unsigned long nr_pages;
1307
1308 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1309 PAGE_SHIFT;
1310
1311 /*
1312 * ->inode could be NULL if async_chunk_start has failed to compress,
1313 * in which case we don't have anything to submit, yet we need to
1314 * always adjust ->async_delalloc_pages as its paired with the init
1315 * happening in cow_file_range_async
1316 */
1317 if (async_chunk->inode)
1318 submit_compressed_extents(async_chunk);
1319
1320 /* atomic_sub_return implies a barrier */
1321 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1322 5 * SZ_1M)
1323 cond_wake_up_nomb(&fs_info->async_submit_wait);
1324 }
1325
async_cow_free(struct btrfs_work * work)1326 static noinline void async_cow_free(struct btrfs_work *work)
1327 {
1328 struct async_chunk *async_chunk;
1329
1330 async_chunk = container_of(work, struct async_chunk, work);
1331 if (async_chunk->inode)
1332 btrfs_add_delayed_iput(async_chunk->inode);
1333 if (async_chunk->blkcg_css)
1334 css_put(async_chunk->blkcg_css);
1335 /*
1336 * Since the pointer to 'pending' is at the beginning of the array of
1337 * async_chunk's, freeing it ensures the whole array has been freed.
1338 */
1339 if (atomic_dec_and_test(async_chunk->pending))
1340 kvfree(async_chunk->pending);
1341 }
1342
cow_file_range_async(struct btrfs_inode * inode,struct writeback_control * wbc,struct page * locked_page,u64 start,u64 end,int * page_started,unsigned long * nr_written)1343 static int cow_file_range_async(struct btrfs_inode *inode,
1344 struct writeback_control *wbc,
1345 struct page *locked_page,
1346 u64 start, u64 end, int *page_started,
1347 unsigned long *nr_written)
1348 {
1349 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1350 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1351 struct async_cow *ctx;
1352 struct async_chunk *async_chunk;
1353 unsigned long nr_pages;
1354 u64 cur_end;
1355 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1356 int i;
1357 bool should_compress;
1358 unsigned nofs_flag;
1359 const unsigned int write_flags = wbc_to_write_flags(wbc);
1360
1361 unlock_extent(&inode->io_tree, start, end);
1362
1363 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1364 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1365 num_chunks = 1;
1366 should_compress = false;
1367 } else {
1368 should_compress = true;
1369 }
1370
1371 nofs_flag = memalloc_nofs_save();
1372 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1373 memalloc_nofs_restore(nofs_flag);
1374
1375 if (!ctx) {
1376 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1377 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1378 EXTENT_DO_ACCOUNTING;
1379 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1380 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1381
1382 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1383 clear_bits, page_ops);
1384 return -ENOMEM;
1385 }
1386
1387 async_chunk = ctx->chunks;
1388 atomic_set(&ctx->num_chunks, num_chunks);
1389
1390 for (i = 0; i < num_chunks; i++) {
1391 if (should_compress)
1392 cur_end = min(end, start + SZ_512K - 1);
1393 else
1394 cur_end = end;
1395
1396 /*
1397 * igrab is called higher up in the call chain, take only the
1398 * lightweight reference for the callback lifetime
1399 */
1400 ihold(&inode->vfs_inode);
1401 async_chunk[i].pending = &ctx->num_chunks;
1402 async_chunk[i].inode = &inode->vfs_inode;
1403 async_chunk[i].start = start;
1404 async_chunk[i].end = cur_end;
1405 async_chunk[i].write_flags = write_flags;
1406 INIT_LIST_HEAD(&async_chunk[i].extents);
1407
1408 /*
1409 * The locked_page comes all the way from writepage and its
1410 * the original page we were actually given. As we spread
1411 * this large delalloc region across multiple async_chunk
1412 * structs, only the first struct needs a pointer to locked_page
1413 *
1414 * This way we don't need racey decisions about who is supposed
1415 * to unlock it.
1416 */
1417 if (locked_page) {
1418 /*
1419 * Depending on the compressibility, the pages might or
1420 * might not go through async. We want all of them to
1421 * be accounted against wbc once. Let's do it here
1422 * before the paths diverge. wbc accounting is used
1423 * only for foreign writeback detection and doesn't
1424 * need full accuracy. Just account the whole thing
1425 * against the first page.
1426 */
1427 wbc_account_cgroup_owner(wbc, locked_page,
1428 cur_end - start);
1429 async_chunk[i].locked_page = locked_page;
1430 locked_page = NULL;
1431 } else {
1432 async_chunk[i].locked_page = NULL;
1433 }
1434
1435 if (blkcg_css != blkcg_root_css) {
1436 css_get(blkcg_css);
1437 async_chunk[i].blkcg_css = blkcg_css;
1438 } else {
1439 async_chunk[i].blkcg_css = NULL;
1440 }
1441
1442 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1443 async_cow_submit, async_cow_free);
1444
1445 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1446 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1447
1448 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1449
1450 *nr_written += nr_pages;
1451 start = cur_end + 1;
1452 }
1453 *page_started = 1;
1454 return 0;
1455 }
1456
run_delalloc_zoned(struct btrfs_inode * inode,struct page * locked_page,u64 start,u64 end,int * page_started,unsigned long * nr_written)1457 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1458 struct page *locked_page, u64 start,
1459 u64 end, int *page_started,
1460 unsigned long *nr_written)
1461 {
1462 int ret;
1463
1464 ret = cow_file_range(inode, locked_page, start, end, page_started,
1465 nr_written, 0);
1466 if (ret)
1467 return ret;
1468
1469 if (*page_started)
1470 return 0;
1471
1472 __set_page_dirty_nobuffers(locked_page);
1473 account_page_redirty(locked_page);
1474 extent_write_locked_range(&inode->vfs_inode, start, end, WB_SYNC_ALL);
1475 *page_started = 1;
1476
1477 return 0;
1478 }
1479
csum_exist_in_range(struct btrfs_fs_info * fs_info,u64 bytenr,u64 num_bytes)1480 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1481 u64 bytenr, u64 num_bytes)
1482 {
1483 int ret;
1484 struct btrfs_ordered_sum *sums;
1485 LIST_HEAD(list);
1486
1487 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1488 bytenr + num_bytes - 1, &list, 0);
1489 if (ret == 0 && list_empty(&list))
1490 return 0;
1491
1492 while (!list_empty(&list)) {
1493 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1494 list_del(&sums->list);
1495 kfree(sums);
1496 }
1497 if (ret < 0)
1498 return ret;
1499 return 1;
1500 }
1501
fallback_to_cow(struct btrfs_inode * inode,struct page * locked_page,const u64 start,const u64 end,int * page_started,unsigned long * nr_written)1502 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1503 const u64 start, const u64 end,
1504 int *page_started, unsigned long *nr_written)
1505 {
1506 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1507 const bool is_reloc_ino = (inode->root->root_key.objectid ==
1508 BTRFS_DATA_RELOC_TREE_OBJECTID);
1509 const u64 range_bytes = end + 1 - start;
1510 struct extent_io_tree *io_tree = &inode->io_tree;
1511 u64 range_start = start;
1512 u64 count;
1513
1514 /*
1515 * If EXTENT_NORESERVE is set it means that when the buffered write was
1516 * made we had not enough available data space and therefore we did not
1517 * reserve data space for it, since we though we could do NOCOW for the
1518 * respective file range (either there is prealloc extent or the inode
1519 * has the NOCOW bit set).
1520 *
1521 * However when we need to fallback to COW mode (because for example the
1522 * block group for the corresponding extent was turned to RO mode by a
1523 * scrub or relocation) we need to do the following:
1524 *
1525 * 1) We increment the bytes_may_use counter of the data space info.
1526 * If COW succeeds, it allocates a new data extent and after doing
1527 * that it decrements the space info's bytes_may_use counter and
1528 * increments its bytes_reserved counter by the same amount (we do
1529 * this at btrfs_add_reserved_bytes()). So we need to increment the
1530 * bytes_may_use counter to compensate (when space is reserved at
1531 * buffered write time, the bytes_may_use counter is incremented);
1532 *
1533 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1534 * that if the COW path fails for any reason, it decrements (through
1535 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1536 * data space info, which we incremented in the step above.
1537 *
1538 * If we need to fallback to cow and the inode corresponds to a free
1539 * space cache inode or an inode of the data relocation tree, we must
1540 * also increment bytes_may_use of the data space_info for the same
1541 * reason. Space caches and relocated data extents always get a prealloc
1542 * extent for them, however scrub or balance may have set the block
1543 * group that contains that extent to RO mode and therefore force COW
1544 * when starting writeback.
1545 */
1546 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1547 EXTENT_NORESERVE, 0);
1548 if (count > 0 || is_space_ino || is_reloc_ino) {
1549 u64 bytes = count;
1550 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1551 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1552
1553 if (is_space_ino || is_reloc_ino)
1554 bytes = range_bytes;
1555
1556 spin_lock(&sinfo->lock);
1557 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1558 spin_unlock(&sinfo->lock);
1559
1560 if (count > 0)
1561 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1562 0, 0, NULL);
1563 }
1564
1565 return cow_file_range(inode, locked_page, start, end, page_started,
1566 nr_written, 1);
1567 }
1568
1569 /*
1570 * when nowcow writeback call back. This checks for snapshots or COW copies
1571 * of the extents that exist in the file, and COWs the file as required.
1572 *
1573 * If no cow copies or snapshots exist, we write directly to the existing
1574 * blocks on disk
1575 */
run_delalloc_nocow(struct btrfs_inode * inode,struct page * locked_page,const u64 start,const u64 end,int * page_started,unsigned long * nr_written)1576 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1577 struct page *locked_page,
1578 const u64 start, const u64 end,
1579 int *page_started,
1580 unsigned long *nr_written)
1581 {
1582 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1583 struct btrfs_root *root = inode->root;
1584 struct btrfs_path *path;
1585 u64 cow_start = (u64)-1;
1586 u64 cur_offset = start;
1587 int ret;
1588 bool check_prev = true;
1589 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1590 u64 ino = btrfs_ino(inode);
1591 bool nocow = false;
1592 u64 disk_bytenr = 0;
1593 const bool force = inode->flags & BTRFS_INODE_NODATACOW;
1594
1595 path = btrfs_alloc_path();
1596 if (!path) {
1597 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1598 EXTENT_LOCKED | EXTENT_DELALLOC |
1599 EXTENT_DO_ACCOUNTING |
1600 EXTENT_DEFRAG, PAGE_UNLOCK |
1601 PAGE_START_WRITEBACK |
1602 PAGE_END_WRITEBACK);
1603 return -ENOMEM;
1604 }
1605
1606 while (1) {
1607 struct btrfs_key found_key;
1608 struct btrfs_file_extent_item *fi;
1609 struct extent_buffer *leaf;
1610 u64 extent_end;
1611 u64 extent_offset;
1612 u64 num_bytes = 0;
1613 u64 disk_num_bytes;
1614 u64 ram_bytes;
1615 int extent_type;
1616
1617 nocow = false;
1618
1619 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1620 cur_offset, 0);
1621 if (ret < 0)
1622 goto error;
1623
1624 /*
1625 * If there is no extent for our range when doing the initial
1626 * search, then go back to the previous slot as it will be the
1627 * one containing the search offset
1628 */
1629 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1630 leaf = path->nodes[0];
1631 btrfs_item_key_to_cpu(leaf, &found_key,
1632 path->slots[0] - 1);
1633 if (found_key.objectid == ino &&
1634 found_key.type == BTRFS_EXTENT_DATA_KEY)
1635 path->slots[0]--;
1636 }
1637 check_prev = false;
1638 next_slot:
1639 /* Go to next leaf if we have exhausted the current one */
1640 leaf = path->nodes[0];
1641 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1642 ret = btrfs_next_leaf(root, path);
1643 if (ret < 0) {
1644 if (cow_start != (u64)-1)
1645 cur_offset = cow_start;
1646 goto error;
1647 }
1648 if (ret > 0)
1649 break;
1650 leaf = path->nodes[0];
1651 }
1652
1653 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1654
1655 /* Didn't find anything for our INO */
1656 if (found_key.objectid > ino)
1657 break;
1658 /*
1659 * Keep searching until we find an EXTENT_ITEM or there are no
1660 * more extents for this inode
1661 */
1662 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1663 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1664 path->slots[0]++;
1665 goto next_slot;
1666 }
1667
1668 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1669 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1670 found_key.offset > end)
1671 break;
1672
1673 /*
1674 * If the found extent starts after requested offset, then
1675 * adjust extent_end to be right before this extent begins
1676 */
1677 if (found_key.offset > cur_offset) {
1678 extent_end = found_key.offset;
1679 extent_type = 0;
1680 goto out_check;
1681 }
1682
1683 /*
1684 * Found extent which begins before our range and potentially
1685 * intersect it
1686 */
1687 fi = btrfs_item_ptr(leaf, path->slots[0],
1688 struct btrfs_file_extent_item);
1689 extent_type = btrfs_file_extent_type(leaf, fi);
1690
1691 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1692 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1693 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1694 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1695 extent_offset = btrfs_file_extent_offset(leaf, fi);
1696 extent_end = found_key.offset +
1697 btrfs_file_extent_num_bytes(leaf, fi);
1698 disk_num_bytes =
1699 btrfs_file_extent_disk_num_bytes(leaf, fi);
1700 /*
1701 * If the extent we got ends before our current offset,
1702 * skip to the next extent.
1703 */
1704 if (extent_end <= cur_offset) {
1705 path->slots[0]++;
1706 goto next_slot;
1707 }
1708 /* Skip holes */
1709 if (disk_bytenr == 0)
1710 goto out_check;
1711 /* Skip compressed/encrypted/encoded extents */
1712 if (btrfs_file_extent_compression(leaf, fi) ||
1713 btrfs_file_extent_encryption(leaf, fi) ||
1714 btrfs_file_extent_other_encoding(leaf, fi))
1715 goto out_check;
1716 /*
1717 * If extent is created before the last volume's snapshot
1718 * this implies the extent is shared, hence we can't do
1719 * nocow. This is the same check as in
1720 * btrfs_cross_ref_exist but without calling
1721 * btrfs_search_slot.
1722 */
1723 if (!freespace_inode &&
1724 btrfs_file_extent_generation(leaf, fi) <=
1725 btrfs_root_last_snapshot(&root->root_item))
1726 goto out_check;
1727 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1728 goto out_check;
1729
1730 /*
1731 * The following checks can be expensive, as they need to
1732 * take other locks and do btree or rbtree searches, so
1733 * release the path to avoid blocking other tasks for too
1734 * long.
1735 */
1736 btrfs_release_path(path);
1737
1738 ret = btrfs_cross_ref_exist(root, ino,
1739 found_key.offset -
1740 extent_offset, disk_bytenr, false);
1741 if (ret) {
1742 /*
1743 * ret could be -EIO if the above fails to read
1744 * metadata.
1745 */
1746 if (ret < 0) {
1747 if (cow_start != (u64)-1)
1748 cur_offset = cow_start;
1749 goto error;
1750 }
1751
1752 WARN_ON_ONCE(freespace_inode);
1753 goto out_check;
1754 }
1755 disk_bytenr += extent_offset;
1756 disk_bytenr += cur_offset - found_key.offset;
1757 num_bytes = min(end + 1, extent_end) - cur_offset;
1758 /*
1759 * If there are pending snapshots for this root, we
1760 * fall into common COW way
1761 */
1762 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1763 goto out_check;
1764 /*
1765 * force cow if csum exists in the range.
1766 * this ensure that csum for a given extent are
1767 * either valid or do not exist.
1768 */
1769 ret = csum_exist_in_range(fs_info, disk_bytenr,
1770 num_bytes);
1771 if (ret) {
1772 /*
1773 * ret could be -EIO if the above fails to read
1774 * metadata.
1775 */
1776 if (ret < 0) {
1777 if (cow_start != (u64)-1)
1778 cur_offset = cow_start;
1779 goto error;
1780 }
1781 WARN_ON_ONCE(freespace_inode);
1782 goto out_check;
1783 }
1784 /* If the extent's block group is RO, we must COW */
1785 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1786 goto out_check;
1787 nocow = true;
1788 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1789 extent_end = found_key.offset + ram_bytes;
1790 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1791 /* Skip extents outside of our requested range */
1792 if (extent_end <= start) {
1793 path->slots[0]++;
1794 goto next_slot;
1795 }
1796 } else {
1797 /* If this triggers then we have a memory corruption */
1798 BUG();
1799 }
1800 out_check:
1801 /*
1802 * If nocow is false then record the beginning of the range
1803 * that needs to be COWed
1804 */
1805 if (!nocow) {
1806 if (cow_start == (u64)-1)
1807 cow_start = cur_offset;
1808 cur_offset = extent_end;
1809 if (cur_offset > end)
1810 break;
1811 if (!path->nodes[0])
1812 continue;
1813 path->slots[0]++;
1814 goto next_slot;
1815 }
1816
1817 /*
1818 * COW range from cow_start to found_key.offset - 1. As the key
1819 * will contain the beginning of the first extent that can be
1820 * NOCOW, following one which needs to be COW'ed
1821 */
1822 if (cow_start != (u64)-1) {
1823 ret = fallback_to_cow(inode, locked_page,
1824 cow_start, found_key.offset - 1,
1825 page_started, nr_written);
1826 if (ret)
1827 goto error;
1828 cow_start = (u64)-1;
1829 }
1830
1831 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1832 u64 orig_start = found_key.offset - extent_offset;
1833 struct extent_map *em;
1834
1835 em = create_io_em(inode, cur_offset, num_bytes,
1836 orig_start,
1837 disk_bytenr, /* block_start */
1838 num_bytes, /* block_len */
1839 disk_num_bytes, /* orig_block_len */
1840 ram_bytes, BTRFS_COMPRESS_NONE,
1841 BTRFS_ORDERED_PREALLOC);
1842 if (IS_ERR(em)) {
1843 ret = PTR_ERR(em);
1844 goto error;
1845 }
1846 free_extent_map(em);
1847 ret = btrfs_add_ordered_extent(inode, cur_offset,
1848 disk_bytenr, num_bytes,
1849 num_bytes,
1850 BTRFS_ORDERED_PREALLOC);
1851 if (ret) {
1852 btrfs_drop_extent_cache(inode, cur_offset,
1853 cur_offset + num_bytes - 1,
1854 0);
1855 goto error;
1856 }
1857 } else {
1858 ret = btrfs_add_ordered_extent(inode, cur_offset,
1859 disk_bytenr, num_bytes,
1860 num_bytes,
1861 BTRFS_ORDERED_NOCOW);
1862 if (ret)
1863 goto error;
1864 }
1865
1866 if (nocow)
1867 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1868 nocow = false;
1869
1870 if (root->root_key.objectid ==
1871 BTRFS_DATA_RELOC_TREE_OBJECTID)
1872 /*
1873 * Error handled later, as we must prevent
1874 * extent_clear_unlock_delalloc() in error handler
1875 * from freeing metadata of created ordered extent.
1876 */
1877 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1878 num_bytes);
1879
1880 extent_clear_unlock_delalloc(inode, cur_offset,
1881 cur_offset + num_bytes - 1,
1882 locked_page, EXTENT_LOCKED |
1883 EXTENT_DELALLOC |
1884 EXTENT_CLEAR_DATA_RESV,
1885 PAGE_UNLOCK | PAGE_SET_ORDERED);
1886
1887 cur_offset = extent_end;
1888
1889 /*
1890 * btrfs_reloc_clone_csums() error, now we're OK to call error
1891 * handler, as metadata for created ordered extent will only
1892 * be freed by btrfs_finish_ordered_io().
1893 */
1894 if (ret)
1895 goto error;
1896 if (cur_offset > end)
1897 break;
1898 }
1899 btrfs_release_path(path);
1900
1901 if (cur_offset <= end && cow_start == (u64)-1)
1902 cow_start = cur_offset;
1903
1904 if (cow_start != (u64)-1) {
1905 cur_offset = end;
1906 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1907 page_started, nr_written);
1908 if (ret)
1909 goto error;
1910 }
1911
1912 error:
1913 if (nocow)
1914 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1915
1916 if (ret && cur_offset < end)
1917 extent_clear_unlock_delalloc(inode, cur_offset, end,
1918 locked_page, EXTENT_LOCKED |
1919 EXTENT_DELALLOC | EXTENT_DEFRAG |
1920 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1921 PAGE_START_WRITEBACK |
1922 PAGE_END_WRITEBACK);
1923 btrfs_free_path(path);
1924 return ret;
1925 }
1926
should_nocow(struct btrfs_inode * inode,u64 start,u64 end)1927 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
1928 {
1929 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
1930 if (inode->defrag_bytes &&
1931 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
1932 0, NULL))
1933 return false;
1934 return true;
1935 }
1936 return false;
1937 }
1938
1939 /*
1940 * Function to process delayed allocation (create CoW) for ranges which are
1941 * being touched for the first time.
1942 */
btrfs_run_delalloc_range(struct btrfs_inode * inode,struct page * locked_page,u64 start,u64 end,int * page_started,unsigned long * nr_written,struct writeback_control * wbc)1943 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1944 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1945 struct writeback_control *wbc)
1946 {
1947 int ret;
1948 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
1949
1950 if (should_nocow(inode, start, end)) {
1951 ASSERT(!zoned);
1952 ret = run_delalloc_nocow(inode, locked_page, start, end,
1953 page_started, nr_written);
1954 } else if (!inode_can_compress(inode) ||
1955 !inode_need_compress(inode, start, end)) {
1956 if (zoned)
1957 ret = run_delalloc_zoned(inode, locked_page, start, end,
1958 page_started, nr_written);
1959 else
1960 ret = cow_file_range(inode, locked_page, start, end,
1961 page_started, nr_written, 1);
1962 } else {
1963 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1964 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
1965 page_started, nr_written);
1966 }
1967 ASSERT(ret <= 0);
1968 if (ret)
1969 btrfs_cleanup_ordered_extents(inode, locked_page, start,
1970 end - start + 1);
1971 return ret;
1972 }
1973
btrfs_split_delalloc_extent(struct inode * inode,struct extent_state * orig,u64 split)1974 void btrfs_split_delalloc_extent(struct inode *inode,
1975 struct extent_state *orig, u64 split)
1976 {
1977 u64 size;
1978
1979 /* not delalloc, ignore it */
1980 if (!(orig->state & EXTENT_DELALLOC))
1981 return;
1982
1983 size = orig->end - orig->start + 1;
1984 if (size > BTRFS_MAX_EXTENT_SIZE) {
1985 u32 num_extents;
1986 u64 new_size;
1987
1988 /*
1989 * See the explanation in btrfs_merge_delalloc_extent, the same
1990 * applies here, just in reverse.
1991 */
1992 new_size = orig->end - split + 1;
1993 num_extents = count_max_extents(new_size);
1994 new_size = split - orig->start;
1995 num_extents += count_max_extents(new_size);
1996 if (count_max_extents(size) >= num_extents)
1997 return;
1998 }
1999
2000 spin_lock(&BTRFS_I(inode)->lock);
2001 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
2002 spin_unlock(&BTRFS_I(inode)->lock);
2003 }
2004
2005 /*
2006 * Handle merged delayed allocation extents so we can keep track of new extents
2007 * that are just merged onto old extents, such as when we are doing sequential
2008 * writes, so we can properly account for the metadata space we'll need.
2009 */
btrfs_merge_delalloc_extent(struct inode * inode,struct extent_state * new,struct extent_state * other)2010 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
2011 struct extent_state *other)
2012 {
2013 u64 new_size, old_size;
2014 u32 num_extents;
2015
2016 /* not delalloc, ignore it */
2017 if (!(other->state & EXTENT_DELALLOC))
2018 return;
2019
2020 if (new->start > other->start)
2021 new_size = new->end - other->start + 1;
2022 else
2023 new_size = other->end - new->start + 1;
2024
2025 /* we're not bigger than the max, unreserve the space and go */
2026 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
2027 spin_lock(&BTRFS_I(inode)->lock);
2028 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2029 spin_unlock(&BTRFS_I(inode)->lock);
2030 return;
2031 }
2032
2033 /*
2034 * We have to add up either side to figure out how many extents were
2035 * accounted for before we merged into one big extent. If the number of
2036 * extents we accounted for is <= the amount we need for the new range
2037 * then we can return, otherwise drop. Think of it like this
2038 *
2039 * [ 4k][MAX_SIZE]
2040 *
2041 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2042 * need 2 outstanding extents, on one side we have 1 and the other side
2043 * we have 1 so they are == and we can return. But in this case
2044 *
2045 * [MAX_SIZE+4k][MAX_SIZE+4k]
2046 *
2047 * Each range on their own accounts for 2 extents, but merged together
2048 * they are only 3 extents worth of accounting, so we need to drop in
2049 * this case.
2050 */
2051 old_size = other->end - other->start + 1;
2052 num_extents = count_max_extents(old_size);
2053 old_size = new->end - new->start + 1;
2054 num_extents += count_max_extents(old_size);
2055 if (count_max_extents(new_size) >= num_extents)
2056 return;
2057
2058 spin_lock(&BTRFS_I(inode)->lock);
2059 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2060 spin_unlock(&BTRFS_I(inode)->lock);
2061 }
2062
btrfs_add_delalloc_inodes(struct btrfs_root * root,struct inode * inode)2063 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2064 struct inode *inode)
2065 {
2066 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2067
2068 spin_lock(&root->delalloc_lock);
2069 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
2070 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2071 &root->delalloc_inodes);
2072 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2073 &BTRFS_I(inode)->runtime_flags);
2074 root->nr_delalloc_inodes++;
2075 if (root->nr_delalloc_inodes == 1) {
2076 spin_lock(&fs_info->delalloc_root_lock);
2077 BUG_ON(!list_empty(&root->delalloc_root));
2078 list_add_tail(&root->delalloc_root,
2079 &fs_info->delalloc_roots);
2080 spin_unlock(&fs_info->delalloc_root_lock);
2081 }
2082 }
2083 spin_unlock(&root->delalloc_lock);
2084 }
2085
2086
__btrfs_del_delalloc_inode(struct btrfs_root * root,struct btrfs_inode * inode)2087 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2088 struct btrfs_inode *inode)
2089 {
2090 struct btrfs_fs_info *fs_info = root->fs_info;
2091
2092 if (!list_empty(&inode->delalloc_inodes)) {
2093 list_del_init(&inode->delalloc_inodes);
2094 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2095 &inode->runtime_flags);
2096 root->nr_delalloc_inodes--;
2097 if (!root->nr_delalloc_inodes) {
2098 ASSERT(list_empty(&root->delalloc_inodes));
2099 spin_lock(&fs_info->delalloc_root_lock);
2100 BUG_ON(list_empty(&root->delalloc_root));
2101 list_del_init(&root->delalloc_root);
2102 spin_unlock(&fs_info->delalloc_root_lock);
2103 }
2104 }
2105 }
2106
btrfs_del_delalloc_inode(struct btrfs_root * root,struct btrfs_inode * inode)2107 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2108 struct btrfs_inode *inode)
2109 {
2110 spin_lock(&root->delalloc_lock);
2111 __btrfs_del_delalloc_inode(root, inode);
2112 spin_unlock(&root->delalloc_lock);
2113 }
2114
2115 /*
2116 * Properly track delayed allocation bytes in the inode and to maintain the
2117 * list of inodes that have pending delalloc work to be done.
2118 */
btrfs_set_delalloc_extent(struct inode * inode,struct extent_state * state,unsigned * bits)2119 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2120 unsigned *bits)
2121 {
2122 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2123
2124 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2125 WARN_ON(1);
2126 /*
2127 * set_bit and clear bit hooks normally require _irqsave/restore
2128 * but in this case, we are only testing for the DELALLOC
2129 * bit, which is only set or cleared with irqs on
2130 */
2131 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2132 struct btrfs_root *root = BTRFS_I(inode)->root;
2133 u64 len = state->end + 1 - state->start;
2134 u32 num_extents = count_max_extents(len);
2135 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2136
2137 spin_lock(&BTRFS_I(inode)->lock);
2138 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2139 spin_unlock(&BTRFS_I(inode)->lock);
2140
2141 /* For sanity tests */
2142 if (btrfs_is_testing(fs_info))
2143 return;
2144
2145 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2146 fs_info->delalloc_batch);
2147 spin_lock(&BTRFS_I(inode)->lock);
2148 BTRFS_I(inode)->delalloc_bytes += len;
2149 if (*bits & EXTENT_DEFRAG)
2150 BTRFS_I(inode)->defrag_bytes += len;
2151 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2152 &BTRFS_I(inode)->runtime_flags))
2153 btrfs_add_delalloc_inodes(root, inode);
2154 spin_unlock(&BTRFS_I(inode)->lock);
2155 }
2156
2157 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2158 (*bits & EXTENT_DELALLOC_NEW)) {
2159 spin_lock(&BTRFS_I(inode)->lock);
2160 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2161 state->start;
2162 spin_unlock(&BTRFS_I(inode)->lock);
2163 }
2164 }
2165
2166 /*
2167 * Once a range is no longer delalloc this function ensures that proper
2168 * accounting happens.
2169 */
btrfs_clear_delalloc_extent(struct inode * vfs_inode,struct extent_state * state,unsigned * bits)2170 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2171 struct extent_state *state, unsigned *bits)
2172 {
2173 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2174 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2175 u64 len = state->end + 1 - state->start;
2176 u32 num_extents = count_max_extents(len);
2177
2178 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2179 spin_lock(&inode->lock);
2180 inode->defrag_bytes -= len;
2181 spin_unlock(&inode->lock);
2182 }
2183
2184 /*
2185 * set_bit and clear bit hooks normally require _irqsave/restore
2186 * but in this case, we are only testing for the DELALLOC
2187 * bit, which is only set or cleared with irqs on
2188 */
2189 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2190 struct btrfs_root *root = inode->root;
2191 bool do_list = !btrfs_is_free_space_inode(inode);
2192
2193 spin_lock(&inode->lock);
2194 btrfs_mod_outstanding_extents(inode, -num_extents);
2195 spin_unlock(&inode->lock);
2196
2197 /*
2198 * We don't reserve metadata space for space cache inodes so we
2199 * don't need to call delalloc_release_metadata if there is an
2200 * error.
2201 */
2202 if (*bits & EXTENT_CLEAR_META_RESV &&
2203 root != fs_info->tree_root)
2204 btrfs_delalloc_release_metadata(inode, len, false);
2205
2206 /* For sanity tests. */
2207 if (btrfs_is_testing(fs_info))
2208 return;
2209
2210 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2211 do_list && !(state->state & EXTENT_NORESERVE) &&
2212 (*bits & EXTENT_CLEAR_DATA_RESV))
2213 btrfs_free_reserved_data_space_noquota(fs_info, len);
2214
2215 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2216 fs_info->delalloc_batch);
2217 spin_lock(&inode->lock);
2218 inode->delalloc_bytes -= len;
2219 if (do_list && inode->delalloc_bytes == 0 &&
2220 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2221 &inode->runtime_flags))
2222 btrfs_del_delalloc_inode(root, inode);
2223 spin_unlock(&inode->lock);
2224 }
2225
2226 if ((state->state & EXTENT_DELALLOC_NEW) &&
2227 (*bits & EXTENT_DELALLOC_NEW)) {
2228 spin_lock(&inode->lock);
2229 ASSERT(inode->new_delalloc_bytes >= len);
2230 inode->new_delalloc_bytes -= len;
2231 if (*bits & EXTENT_ADD_INODE_BYTES)
2232 inode_add_bytes(&inode->vfs_inode, len);
2233 spin_unlock(&inode->lock);
2234 }
2235 }
2236
2237 /*
2238 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
2239 * in a chunk's stripe. This function ensures that bios do not span a
2240 * stripe/chunk
2241 *
2242 * @page - The page we are about to add to the bio
2243 * @size - size we want to add to the bio
2244 * @bio - bio we want to ensure is smaller than a stripe
2245 * @bio_flags - flags of the bio
2246 *
2247 * return 1 if page cannot be added to the bio
2248 * return 0 if page can be added to the bio
2249 * return error otherwise
2250 */
btrfs_bio_fits_in_stripe(struct page * page,size_t size,struct bio * bio,unsigned long bio_flags)2251 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
2252 unsigned long bio_flags)
2253 {
2254 struct inode *inode = page->mapping->host;
2255 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2256 u64 logical = bio->bi_iter.bi_sector << 9;
2257 u32 bio_len = bio->bi_iter.bi_size;
2258 struct extent_map *em;
2259 int ret = 0;
2260 struct btrfs_io_geometry geom;
2261
2262 if (bio_flags & EXTENT_BIO_COMPRESSED)
2263 return 0;
2264
2265 em = btrfs_get_chunk_map(fs_info, logical, fs_info->sectorsize);
2266 if (IS_ERR(em))
2267 return PTR_ERR(em);
2268 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio), logical, &geom);
2269 if (ret < 0)
2270 goto out;
2271
2272 if (geom.len < bio_len + size)
2273 ret = 1;
2274 out:
2275 free_extent_map(em);
2276 return ret;
2277 }
2278
2279 /*
2280 * in order to insert checksums into the metadata in large chunks,
2281 * we wait until bio submission time. All the pages in the bio are
2282 * checksummed and sums are attached onto the ordered extent record.
2283 *
2284 * At IO completion time the cums attached on the ordered extent record
2285 * are inserted into the btree
2286 */
btrfs_submit_bio_start(struct inode * inode,struct bio * bio,u64 dio_file_offset)2287 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2288 u64 dio_file_offset)
2289 {
2290 return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2291 }
2292
2293 /*
2294 * Split an extent_map at [start, start + len]
2295 *
2296 * This function is intended to be used only for extract_ordered_extent().
2297 */
split_zoned_em(struct btrfs_inode * inode,u64 start,u64 len,u64 pre,u64 post)2298 static int split_zoned_em(struct btrfs_inode *inode, u64 start, u64 len,
2299 u64 pre, u64 post)
2300 {
2301 struct extent_map_tree *em_tree = &inode->extent_tree;
2302 struct extent_map *em;
2303 struct extent_map *split_pre = NULL;
2304 struct extent_map *split_mid = NULL;
2305 struct extent_map *split_post = NULL;
2306 int ret = 0;
2307 unsigned long flags;
2308
2309 /* Sanity check */
2310 if (pre == 0 && post == 0)
2311 return 0;
2312
2313 split_pre = alloc_extent_map();
2314 if (pre)
2315 split_mid = alloc_extent_map();
2316 if (post)
2317 split_post = alloc_extent_map();
2318 if (!split_pre || (pre && !split_mid) || (post && !split_post)) {
2319 ret = -ENOMEM;
2320 goto out;
2321 }
2322
2323 ASSERT(pre + post < len);
2324
2325 lock_extent(&inode->io_tree, start, start + len - 1);
2326 write_lock(&em_tree->lock);
2327 em = lookup_extent_mapping(em_tree, start, len);
2328 if (!em) {
2329 ret = -EIO;
2330 goto out_unlock;
2331 }
2332
2333 ASSERT(em->len == len);
2334 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2335 ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE);
2336 ASSERT(test_bit(EXTENT_FLAG_PINNED, &em->flags));
2337 ASSERT(!test_bit(EXTENT_FLAG_LOGGING, &em->flags));
2338 ASSERT(!list_empty(&em->list));
2339
2340 flags = em->flags;
2341 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
2342
2343 /* First, replace the em with a new extent_map starting from * em->start */
2344 split_pre->start = em->start;
2345 split_pre->len = (pre ? pre : em->len - post);
2346 split_pre->orig_start = split_pre->start;
2347 split_pre->block_start = em->block_start;
2348 split_pre->block_len = split_pre->len;
2349 split_pre->orig_block_len = split_pre->block_len;
2350 split_pre->ram_bytes = split_pre->len;
2351 split_pre->flags = flags;
2352 split_pre->compress_type = em->compress_type;
2353 split_pre->generation = em->generation;
2354
2355 replace_extent_mapping(em_tree, em, split_pre, 1);
2356
2357 /*
2358 * Now we only have an extent_map at:
2359 * [em->start, em->start + pre] if pre != 0
2360 * [em->start, em->start + em->len - post] if pre == 0
2361 */
2362
2363 if (pre) {
2364 /* Insert the middle extent_map */
2365 split_mid->start = em->start + pre;
2366 split_mid->len = em->len - pre - post;
2367 split_mid->orig_start = split_mid->start;
2368 split_mid->block_start = em->block_start + pre;
2369 split_mid->block_len = split_mid->len;
2370 split_mid->orig_block_len = split_mid->block_len;
2371 split_mid->ram_bytes = split_mid->len;
2372 split_mid->flags = flags;
2373 split_mid->compress_type = em->compress_type;
2374 split_mid->generation = em->generation;
2375 add_extent_mapping(em_tree, split_mid, 1);
2376 }
2377
2378 if (post) {
2379 split_post->start = em->start + em->len - post;
2380 split_post->len = post;
2381 split_post->orig_start = split_post->start;
2382 split_post->block_start = em->block_start + em->len - post;
2383 split_post->block_len = split_post->len;
2384 split_post->orig_block_len = split_post->block_len;
2385 split_post->ram_bytes = split_post->len;
2386 split_post->flags = flags;
2387 split_post->compress_type = em->compress_type;
2388 split_post->generation = em->generation;
2389 add_extent_mapping(em_tree, split_post, 1);
2390 }
2391
2392 /* Once for us */
2393 free_extent_map(em);
2394 /* Once for the tree */
2395 free_extent_map(em);
2396
2397 out_unlock:
2398 write_unlock(&em_tree->lock);
2399 unlock_extent(&inode->io_tree, start, start + len - 1);
2400 out:
2401 free_extent_map(split_pre);
2402 free_extent_map(split_mid);
2403 free_extent_map(split_post);
2404
2405 return ret;
2406 }
2407
extract_ordered_extent(struct btrfs_inode * inode,struct bio * bio,loff_t file_offset)2408 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2409 struct bio *bio, loff_t file_offset)
2410 {
2411 struct btrfs_ordered_extent *ordered;
2412 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2413 u64 file_len;
2414 u64 len = bio->bi_iter.bi_size;
2415 u64 end = start + len;
2416 u64 ordered_end;
2417 u64 pre, post;
2418 int ret = 0;
2419
2420 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2421 if (WARN_ON_ONCE(!ordered))
2422 return BLK_STS_IOERR;
2423
2424 /* No need to split */
2425 if (ordered->disk_num_bytes == len)
2426 goto out;
2427
2428 /* We cannot split once end_bio'd ordered extent */
2429 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2430 ret = -EINVAL;
2431 goto out;
2432 }
2433
2434 /* We cannot split a compressed ordered extent */
2435 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2436 ret = -EINVAL;
2437 goto out;
2438 }
2439
2440 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2441 /* bio must be in one ordered extent */
2442 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2443 ret = -EINVAL;
2444 goto out;
2445 }
2446
2447 /* Checksum list should be empty */
2448 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2449 ret = -EINVAL;
2450 goto out;
2451 }
2452
2453 file_len = ordered->num_bytes;
2454 pre = start - ordered->disk_bytenr;
2455 post = ordered_end - end;
2456
2457 ret = btrfs_split_ordered_extent(ordered, pre, post);
2458 if (ret)
2459 goto out;
2460 ret = split_zoned_em(inode, file_offset, file_len, pre, post);
2461
2462 out:
2463 btrfs_put_ordered_extent(ordered);
2464
2465 return errno_to_blk_status(ret);
2466 }
2467
2468 /*
2469 * extent_io.c submission hook. This does the right thing for csum calculation
2470 * on write, or reading the csums from the tree before a read.
2471 *
2472 * Rules about async/sync submit,
2473 * a) read: sync submit
2474 *
2475 * b) write without checksum: sync submit
2476 *
2477 * c) write with checksum:
2478 * c-1) if bio is issued by fsync: sync submit
2479 * (sync_writers != 0)
2480 *
2481 * c-2) if root is reloc root: sync submit
2482 * (only in case of buffered IO)
2483 *
2484 * c-3) otherwise: async submit
2485 */
btrfs_submit_data_bio(struct inode * inode,struct bio * bio,int mirror_num,unsigned long bio_flags)2486 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2487 int mirror_num, unsigned long bio_flags)
2488
2489 {
2490 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2491 struct btrfs_root *root = BTRFS_I(inode)->root;
2492 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2493 blk_status_t ret = 0;
2494 int skip_sum;
2495 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2496
2497 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2498 !fs_info->csum_root;
2499
2500 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2501 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2502
2503 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2504 struct page *page = bio_first_bvec_all(bio)->bv_page;
2505 loff_t file_offset = page_offset(page);
2506
2507 ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset);
2508 if (ret)
2509 goto out;
2510 }
2511
2512 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
2513 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2514 if (ret)
2515 goto out;
2516
2517 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2518 ret = btrfs_submit_compressed_read(inode, bio,
2519 mirror_num,
2520 bio_flags);
2521 goto out;
2522 } else {
2523 /*
2524 * Lookup bio sums does extra checks around whether we
2525 * need to csum or not, which is why we ignore skip_sum
2526 * here.
2527 */
2528 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2529 if (ret)
2530 goto out;
2531 }
2532 goto mapit;
2533 } else if (async && !skip_sum) {
2534 /* csum items have already been cloned */
2535 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2536 goto mapit;
2537 /* we're doing a write, do the async checksumming */
2538 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags,
2539 0, btrfs_submit_bio_start);
2540 goto out;
2541 } else if (!skip_sum) {
2542 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2543 if (ret)
2544 goto out;
2545 }
2546
2547 mapit:
2548 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2549
2550 out:
2551 if (ret) {
2552 bio->bi_status = ret;
2553 bio_endio(bio);
2554 }
2555 return ret;
2556 }
2557
2558 /*
2559 * given a list of ordered sums record them in the inode. This happens
2560 * at IO completion time based on sums calculated at bio submission time.
2561 */
add_pending_csums(struct btrfs_trans_handle * trans,struct list_head * list)2562 static int add_pending_csums(struct btrfs_trans_handle *trans,
2563 struct list_head *list)
2564 {
2565 struct btrfs_ordered_sum *sum;
2566 int ret;
2567
2568 list_for_each_entry(sum, list, list) {
2569 trans->adding_csums = true;
2570 ret = btrfs_csum_file_blocks(trans, trans->fs_info->csum_root, sum);
2571 trans->adding_csums = false;
2572 if (ret)
2573 return ret;
2574 }
2575 return 0;
2576 }
2577
btrfs_find_new_delalloc_bytes(struct btrfs_inode * inode,const u64 start,const u64 len,struct extent_state ** cached_state)2578 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2579 const u64 start,
2580 const u64 len,
2581 struct extent_state **cached_state)
2582 {
2583 u64 search_start = start;
2584 const u64 end = start + len - 1;
2585
2586 while (search_start < end) {
2587 const u64 search_len = end - search_start + 1;
2588 struct extent_map *em;
2589 u64 em_len;
2590 int ret = 0;
2591
2592 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2593 if (IS_ERR(em))
2594 return PTR_ERR(em);
2595
2596 if (em->block_start != EXTENT_MAP_HOLE)
2597 goto next;
2598
2599 em_len = em->len;
2600 if (em->start < search_start)
2601 em_len -= search_start - em->start;
2602 if (em_len > search_len)
2603 em_len = search_len;
2604
2605 ret = set_extent_bit(&inode->io_tree, search_start,
2606 search_start + em_len - 1,
2607 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2608 GFP_NOFS, NULL);
2609 next:
2610 search_start = extent_map_end(em);
2611 free_extent_map(em);
2612 if (ret)
2613 return ret;
2614 }
2615 return 0;
2616 }
2617
btrfs_set_extent_delalloc(struct btrfs_inode * inode,u64 start,u64 end,unsigned int extra_bits,struct extent_state ** cached_state)2618 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2619 unsigned int extra_bits,
2620 struct extent_state **cached_state)
2621 {
2622 WARN_ON(PAGE_ALIGNED(end));
2623
2624 if (start >= i_size_read(&inode->vfs_inode) &&
2625 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2626 /*
2627 * There can't be any extents following eof in this case so just
2628 * set the delalloc new bit for the range directly.
2629 */
2630 extra_bits |= EXTENT_DELALLOC_NEW;
2631 } else {
2632 int ret;
2633
2634 ret = btrfs_find_new_delalloc_bytes(inode, start,
2635 end + 1 - start,
2636 cached_state);
2637 if (ret)
2638 return ret;
2639 }
2640
2641 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2642 cached_state);
2643 }
2644
2645 /* see btrfs_writepage_start_hook for details on why this is required */
2646 struct btrfs_writepage_fixup {
2647 struct page *page;
2648 struct inode *inode;
2649 struct btrfs_work work;
2650 };
2651
btrfs_writepage_fixup_worker(struct btrfs_work * work)2652 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2653 {
2654 struct btrfs_writepage_fixup *fixup;
2655 struct btrfs_ordered_extent *ordered;
2656 struct extent_state *cached_state = NULL;
2657 struct extent_changeset *data_reserved = NULL;
2658 struct page *page;
2659 struct btrfs_inode *inode;
2660 u64 page_start;
2661 u64 page_end;
2662 int ret = 0;
2663 bool free_delalloc_space = true;
2664
2665 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2666 page = fixup->page;
2667 inode = BTRFS_I(fixup->inode);
2668 page_start = page_offset(page);
2669 page_end = page_offset(page) + PAGE_SIZE - 1;
2670
2671 /*
2672 * This is similar to page_mkwrite, we need to reserve the space before
2673 * we take the page lock.
2674 */
2675 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2676 PAGE_SIZE);
2677 again:
2678 lock_page(page);
2679
2680 /*
2681 * Before we queued this fixup, we took a reference on the page.
2682 * page->mapping may go NULL, but it shouldn't be moved to a different
2683 * address space.
2684 */
2685 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2686 /*
2687 * Unfortunately this is a little tricky, either
2688 *
2689 * 1) We got here and our page had already been dealt with and
2690 * we reserved our space, thus ret == 0, so we need to just
2691 * drop our space reservation and bail. This can happen the
2692 * first time we come into the fixup worker, or could happen
2693 * while waiting for the ordered extent.
2694 * 2) Our page was already dealt with, but we happened to get an
2695 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2696 * this case we obviously don't have anything to release, but
2697 * because the page was already dealt with we don't want to
2698 * mark the page with an error, so make sure we're resetting
2699 * ret to 0. This is why we have this check _before_ the ret
2700 * check, because we do not want to have a surprise ENOSPC
2701 * when the page was already properly dealt with.
2702 */
2703 if (!ret) {
2704 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2705 btrfs_delalloc_release_space(inode, data_reserved,
2706 page_start, PAGE_SIZE,
2707 true);
2708 }
2709 ret = 0;
2710 goto out_page;
2711 }
2712
2713 /*
2714 * We can't mess with the page state unless it is locked, so now that
2715 * it is locked bail if we failed to make our space reservation.
2716 */
2717 if (ret)
2718 goto out_page;
2719
2720 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2721
2722 /* already ordered? We're done */
2723 if (PageOrdered(page))
2724 goto out_reserved;
2725
2726 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2727 if (ordered) {
2728 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2729 &cached_state);
2730 unlock_page(page);
2731 btrfs_start_ordered_extent(ordered, 1);
2732 btrfs_put_ordered_extent(ordered);
2733 goto again;
2734 }
2735
2736 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2737 &cached_state);
2738 if (ret)
2739 goto out_reserved;
2740
2741 /*
2742 * Everything went as planned, we're now the owner of a dirty page with
2743 * delayed allocation bits set and space reserved for our COW
2744 * destination.
2745 *
2746 * The page was dirty when we started, nothing should have cleaned it.
2747 */
2748 BUG_ON(!PageDirty(page));
2749 free_delalloc_space = false;
2750 out_reserved:
2751 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2752 if (free_delalloc_space)
2753 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2754 PAGE_SIZE, true);
2755 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2756 &cached_state);
2757 out_page:
2758 if (ret) {
2759 /*
2760 * We hit ENOSPC or other errors. Update the mapping and page
2761 * to reflect the errors and clean the page.
2762 */
2763 mapping_set_error(page->mapping, ret);
2764 end_extent_writepage(page, ret, page_start, page_end);
2765 clear_page_dirty_for_io(page);
2766 SetPageError(page);
2767 }
2768 ClearPageChecked(page);
2769 unlock_page(page);
2770 put_page(page);
2771 kfree(fixup);
2772 extent_changeset_free(data_reserved);
2773 /*
2774 * As a precaution, do a delayed iput in case it would be the last iput
2775 * that could need flushing space. Recursing back to fixup worker would
2776 * deadlock.
2777 */
2778 btrfs_add_delayed_iput(&inode->vfs_inode);
2779 }
2780
2781 /*
2782 * There are a few paths in the higher layers of the kernel that directly
2783 * set the page dirty bit without asking the filesystem if it is a
2784 * good idea. This causes problems because we want to make sure COW
2785 * properly happens and the data=ordered rules are followed.
2786 *
2787 * In our case any range that doesn't have the ORDERED bit set
2788 * hasn't been properly setup for IO. We kick off an async process
2789 * to fix it up. The async helper will wait for ordered extents, set
2790 * the delalloc bit and make it safe to write the page.
2791 */
btrfs_writepage_cow_fixup(struct page * page)2792 int btrfs_writepage_cow_fixup(struct page *page)
2793 {
2794 struct inode *inode = page->mapping->host;
2795 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2796 struct btrfs_writepage_fixup *fixup;
2797
2798 /* This page has ordered extent covering it already */
2799 if (PageOrdered(page))
2800 return 0;
2801
2802 /*
2803 * PageChecked is set below when we create a fixup worker for this page,
2804 * don't try to create another one if we're already PageChecked()
2805 *
2806 * The extent_io writepage code will redirty the page if we send back
2807 * EAGAIN.
2808 */
2809 if (PageChecked(page))
2810 return -EAGAIN;
2811
2812 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2813 if (!fixup)
2814 return -EAGAIN;
2815
2816 /*
2817 * We are already holding a reference to this inode from
2818 * write_cache_pages. We need to hold it because the space reservation
2819 * takes place outside of the page lock, and we can't trust
2820 * page->mapping outside of the page lock.
2821 */
2822 ihold(inode);
2823 SetPageChecked(page);
2824 get_page(page);
2825 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2826 fixup->page = page;
2827 fixup->inode = inode;
2828 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2829
2830 return -EAGAIN;
2831 }
2832
insert_reserved_file_extent(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,u64 file_pos,struct btrfs_file_extent_item * stack_fi,const bool update_inode_bytes,u64 qgroup_reserved)2833 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2834 struct btrfs_inode *inode, u64 file_pos,
2835 struct btrfs_file_extent_item *stack_fi,
2836 const bool update_inode_bytes,
2837 u64 qgroup_reserved)
2838 {
2839 struct btrfs_root *root = inode->root;
2840 const u64 sectorsize = root->fs_info->sectorsize;
2841 struct btrfs_path *path;
2842 struct extent_buffer *leaf;
2843 struct btrfs_key ins;
2844 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2845 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2846 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2847 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2848 struct btrfs_drop_extents_args drop_args = { 0 };
2849 int ret;
2850
2851 path = btrfs_alloc_path();
2852 if (!path)
2853 return -ENOMEM;
2854
2855 /*
2856 * we may be replacing one extent in the tree with another.
2857 * The new extent is pinned in the extent map, and we don't want
2858 * to drop it from the cache until it is completely in the btree.
2859 *
2860 * So, tell btrfs_drop_extents to leave this extent in the cache.
2861 * the caller is expected to unpin it and allow it to be merged
2862 * with the others.
2863 */
2864 drop_args.path = path;
2865 drop_args.start = file_pos;
2866 drop_args.end = file_pos + num_bytes;
2867 drop_args.replace_extent = true;
2868 drop_args.extent_item_size = sizeof(*stack_fi);
2869 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2870 if (ret)
2871 goto out;
2872
2873 if (!drop_args.extent_inserted) {
2874 ins.objectid = btrfs_ino(inode);
2875 ins.offset = file_pos;
2876 ins.type = BTRFS_EXTENT_DATA_KEY;
2877
2878 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2879 sizeof(*stack_fi));
2880 if (ret)
2881 goto out;
2882 }
2883 leaf = path->nodes[0];
2884 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2885 write_extent_buffer(leaf, stack_fi,
2886 btrfs_item_ptr_offset(leaf, path->slots[0]),
2887 sizeof(struct btrfs_file_extent_item));
2888
2889 btrfs_mark_buffer_dirty(leaf);
2890 btrfs_release_path(path);
2891
2892 /*
2893 * If we dropped an inline extent here, we know the range where it is
2894 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2895 * number of bytes only for that range containing the inline extent.
2896 * The remaining of the range will be processed when clearning the
2897 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2898 */
2899 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2900 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2901
2902 inline_size = drop_args.bytes_found - inline_size;
2903 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2904 drop_args.bytes_found -= inline_size;
2905 num_bytes -= sectorsize;
2906 }
2907
2908 if (update_inode_bytes)
2909 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2910
2911 ins.objectid = disk_bytenr;
2912 ins.offset = disk_num_bytes;
2913 ins.type = BTRFS_EXTENT_ITEM_KEY;
2914
2915 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2916 if (ret)
2917 goto out;
2918
2919 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2920 file_pos, qgroup_reserved, &ins);
2921 out:
2922 btrfs_free_path(path);
2923
2924 return ret;
2925 }
2926
btrfs_release_delalloc_bytes(struct btrfs_fs_info * fs_info,u64 start,u64 len)2927 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2928 u64 start, u64 len)
2929 {
2930 struct btrfs_block_group *cache;
2931
2932 cache = btrfs_lookup_block_group(fs_info, start);
2933 ASSERT(cache);
2934
2935 spin_lock(&cache->lock);
2936 cache->delalloc_bytes -= len;
2937 spin_unlock(&cache->lock);
2938
2939 btrfs_put_block_group(cache);
2940 }
2941
insert_ordered_extent_file_extent(struct btrfs_trans_handle * trans,struct btrfs_ordered_extent * oe)2942 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2943 struct btrfs_ordered_extent *oe)
2944 {
2945 struct btrfs_file_extent_item stack_fi;
2946 u64 logical_len;
2947 bool update_inode_bytes;
2948
2949 memset(&stack_fi, 0, sizeof(stack_fi));
2950 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2951 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2952 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2953 oe->disk_num_bytes);
2954 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2955 logical_len = oe->truncated_len;
2956 else
2957 logical_len = oe->num_bytes;
2958 btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len);
2959 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len);
2960 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2961 /* Encryption and other encoding is reserved and all 0 */
2962
2963 /*
2964 * For delalloc, when completing an ordered extent we update the inode's
2965 * bytes when clearing the range in the inode's io tree, so pass false
2966 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2967 * except if the ordered extent was truncated.
2968 */
2969 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
2970 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
2971
2972 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
2973 oe->file_offset, &stack_fi,
2974 update_inode_bytes, oe->qgroup_rsv);
2975 }
2976
2977 /*
2978 * As ordered data IO finishes, this gets called so we can finish
2979 * an ordered extent if the range of bytes in the file it covers are
2980 * fully written.
2981 */
btrfs_finish_ordered_io(struct btrfs_ordered_extent * ordered_extent)2982 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2983 {
2984 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
2985 struct btrfs_root *root = inode->root;
2986 struct btrfs_fs_info *fs_info = root->fs_info;
2987 struct btrfs_trans_handle *trans = NULL;
2988 struct extent_io_tree *io_tree = &inode->io_tree;
2989 struct extent_state *cached_state = NULL;
2990 u64 start, end;
2991 int compress_type = 0;
2992 int ret = 0;
2993 u64 logical_len = ordered_extent->num_bytes;
2994 bool freespace_inode;
2995 bool truncated = false;
2996 bool clear_reserved_extent = true;
2997 unsigned int clear_bits = EXTENT_DEFRAG;
2998
2999 start = ordered_extent->file_offset;
3000 end = start + ordered_extent->num_bytes - 1;
3001
3002 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3003 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3004 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
3005 clear_bits |= EXTENT_DELALLOC_NEW;
3006
3007 freespace_inode = btrfs_is_free_space_inode(inode);
3008
3009 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3010 ret = -EIO;
3011 goto out;
3012 }
3013
3014 if (ordered_extent->bdev)
3015 btrfs_rewrite_logical_zoned(ordered_extent);
3016
3017 btrfs_free_io_failure_record(inode, start, end);
3018
3019 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3020 truncated = true;
3021 logical_len = ordered_extent->truncated_len;
3022 /* Truncated the entire extent, don't bother adding */
3023 if (!logical_len)
3024 goto out;
3025 }
3026
3027 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3028 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3029
3030 btrfs_inode_safe_disk_i_size_write(inode, 0);
3031 if (freespace_inode)
3032 trans = btrfs_join_transaction_spacecache(root);
3033 else
3034 trans = btrfs_join_transaction(root);
3035 if (IS_ERR(trans)) {
3036 ret = PTR_ERR(trans);
3037 trans = NULL;
3038 goto out;
3039 }
3040 trans->block_rsv = &inode->block_rsv;
3041 ret = btrfs_update_inode_fallback(trans, root, inode);
3042 if (ret) /* -ENOMEM or corruption */
3043 btrfs_abort_transaction(trans, ret);
3044 goto out;
3045 }
3046
3047 clear_bits |= EXTENT_LOCKED;
3048 lock_extent_bits(io_tree, start, end, &cached_state);
3049
3050 if (freespace_inode)
3051 trans = btrfs_join_transaction_spacecache(root);
3052 else
3053 trans = btrfs_join_transaction(root);
3054 if (IS_ERR(trans)) {
3055 ret = PTR_ERR(trans);
3056 trans = NULL;
3057 goto out;
3058 }
3059
3060 trans->block_rsv = &inode->block_rsv;
3061
3062 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3063 compress_type = ordered_extent->compress_type;
3064 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3065 BUG_ON(compress_type);
3066 ret = btrfs_mark_extent_written(trans, inode,
3067 ordered_extent->file_offset,
3068 ordered_extent->file_offset +
3069 logical_len);
3070 } else {
3071 BUG_ON(root == fs_info->tree_root);
3072 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3073 if (!ret) {
3074 clear_reserved_extent = false;
3075 btrfs_release_delalloc_bytes(fs_info,
3076 ordered_extent->disk_bytenr,
3077 ordered_extent->disk_num_bytes);
3078 }
3079 }
3080 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3081 ordered_extent->num_bytes, trans->transid);
3082 if (ret < 0) {
3083 btrfs_abort_transaction(trans, ret);
3084 goto out;
3085 }
3086
3087 ret = add_pending_csums(trans, &ordered_extent->list);
3088 if (ret) {
3089 btrfs_abort_transaction(trans, ret);
3090 goto out;
3091 }
3092
3093 /*
3094 * If this is a new delalloc range, clear its new delalloc flag to
3095 * update the inode's number of bytes. This needs to be done first
3096 * before updating the inode item.
3097 */
3098 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3099 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3100 clear_extent_bit(&inode->io_tree, start, end,
3101 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3102 0, 0, &cached_state);
3103
3104 btrfs_inode_safe_disk_i_size_write(inode, 0);
3105 ret = btrfs_update_inode_fallback(trans, root, inode);
3106 if (ret) { /* -ENOMEM or corruption */
3107 btrfs_abort_transaction(trans, ret);
3108 goto out;
3109 }
3110 ret = 0;
3111 out:
3112 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3113 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
3114 &cached_state);
3115
3116 if (trans)
3117 btrfs_end_transaction(trans);
3118
3119 if (ret || truncated) {
3120 u64 unwritten_start = start;
3121
3122 /*
3123 * If we failed to finish this ordered extent for any reason we
3124 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3125 * extent, and mark the inode with the error if it wasn't
3126 * already set. Any error during writeback would have already
3127 * set the mapping error, so we need to set it if we're the ones
3128 * marking this ordered extent as failed.
3129 */
3130 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3131 &ordered_extent->flags))
3132 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3133
3134 if (truncated)
3135 unwritten_start += logical_len;
3136 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3137
3138 /* Drop the cache for the part of the extent we didn't write. */
3139 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
3140
3141 /*
3142 * If the ordered extent had an IOERR or something else went
3143 * wrong we need to return the space for this ordered extent
3144 * back to the allocator. We only free the extent in the
3145 * truncated case if we didn't write out the extent at all.
3146 *
3147 * If we made it past insert_reserved_file_extent before we
3148 * errored out then we don't need to do this as the accounting
3149 * has already been done.
3150 */
3151 if ((ret || !logical_len) &&
3152 clear_reserved_extent &&
3153 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3154 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3155 /*
3156 * Discard the range before returning it back to the
3157 * free space pool
3158 */
3159 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3160 btrfs_discard_extent(fs_info,
3161 ordered_extent->disk_bytenr,
3162 ordered_extent->disk_num_bytes,
3163 NULL);
3164 btrfs_free_reserved_extent(fs_info,
3165 ordered_extent->disk_bytenr,
3166 ordered_extent->disk_num_bytes, 1);
3167 }
3168 }
3169
3170 /*
3171 * This needs to be done to make sure anybody waiting knows we are done
3172 * updating everything for this ordered extent.
3173 */
3174 btrfs_remove_ordered_extent(inode, ordered_extent);
3175
3176 /* once for us */
3177 btrfs_put_ordered_extent(ordered_extent);
3178 /* once for the tree */
3179 btrfs_put_ordered_extent(ordered_extent);
3180
3181 return ret;
3182 }
3183
finish_ordered_fn(struct btrfs_work * work)3184 static void finish_ordered_fn(struct btrfs_work *work)
3185 {
3186 struct btrfs_ordered_extent *ordered_extent;
3187 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3188 btrfs_finish_ordered_io(ordered_extent);
3189 }
3190
btrfs_writepage_endio_finish_ordered(struct btrfs_inode * inode,struct page * page,u64 start,u64 end,bool uptodate)3191 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3192 struct page *page, u64 start,
3193 u64 end, bool uptodate)
3194 {
3195 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3196
3197 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start,
3198 finish_ordered_fn, uptodate);
3199 }
3200
3201 /*
3202 * check_data_csum - verify checksum of one sector of uncompressed data
3203 * @inode: inode
3204 * @io_bio: btrfs_io_bio which contains the csum
3205 * @bio_offset: offset to the beginning of the bio (in bytes)
3206 * @page: page where is the data to be verified
3207 * @pgoff: offset inside the page
3208 * @start: logical offset in the file
3209 *
3210 * The length of such check is always one sector size.
3211 */
check_data_csum(struct inode * inode,struct btrfs_io_bio * io_bio,u32 bio_offset,struct page * page,u32 pgoff,u64 start)3212 static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio,
3213 u32 bio_offset, struct page *page, u32 pgoff,
3214 u64 start)
3215 {
3216 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3217 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3218 char *kaddr;
3219 u32 len = fs_info->sectorsize;
3220 const u32 csum_size = fs_info->csum_size;
3221 unsigned int offset_sectors;
3222 u8 *csum_expected;
3223 u8 csum[BTRFS_CSUM_SIZE];
3224
3225 ASSERT(pgoff + len <= PAGE_SIZE);
3226
3227 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
3228 csum_expected = ((u8 *)io_bio->csum) + offset_sectors * csum_size;
3229
3230 kaddr = kmap_atomic(page);
3231 shash->tfm = fs_info->csum_shash;
3232
3233 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
3234
3235 if (memcmp(csum, csum_expected, csum_size))
3236 goto zeroit;
3237
3238 kunmap_atomic(kaddr);
3239 return 0;
3240 zeroit:
3241 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3242 io_bio->mirror_num);
3243 if (io_bio->device)
3244 btrfs_dev_stat_inc_and_print(io_bio->device,
3245 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3246 memset(kaddr + pgoff, 1, len);
3247 flush_dcache_page(page);
3248 kunmap_atomic(kaddr);
3249 return -EIO;
3250 }
3251
3252 /*
3253 * When reads are done, we need to check csums to verify the data is correct.
3254 * if there's a match, we allow the bio to finish. If not, the code in
3255 * extent_io.c will try to find good copies for us.
3256 *
3257 * @bio_offset: offset to the beginning of the bio (in bytes)
3258 * @start: file offset of the range start
3259 * @end: file offset of the range end (inclusive)
3260 *
3261 * Return a bitmap where bit set means a csum mismatch, and bit not set means
3262 * csum match.
3263 */
btrfs_verify_data_csum(struct btrfs_io_bio * io_bio,u32 bio_offset,struct page * page,u64 start,u64 end)3264 unsigned int btrfs_verify_data_csum(struct btrfs_io_bio *io_bio, u32 bio_offset,
3265 struct page *page, u64 start, u64 end)
3266 {
3267 struct inode *inode = page->mapping->host;
3268 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3269 struct btrfs_root *root = BTRFS_I(inode)->root;
3270 const u32 sectorsize = root->fs_info->sectorsize;
3271 u32 pg_off;
3272 unsigned int result = 0;
3273
3274 if (PageChecked(page)) {
3275 ClearPageChecked(page);
3276 return 0;
3277 }
3278
3279 /*
3280 * For subpage case, above PageChecked is not safe as it's not subpage
3281 * compatible.
3282 * But for now only cow fixup and compressed read utilize PageChecked
3283 * flag, while in this context we can easily use io_bio->csum to
3284 * determine if we really need to do csum verification.
3285 *
3286 * So for now, just exit if io_bio->csum is NULL, as it means it's
3287 * compressed read, and its compressed data csum has already been
3288 * verified.
3289 */
3290 if (io_bio->csum == NULL)
3291 return 0;
3292
3293 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3294 return 0;
3295
3296 if (!root->fs_info->csum_root)
3297 return 0;
3298
3299 ASSERT(page_offset(page) <= start &&
3300 end <= page_offset(page) + PAGE_SIZE - 1);
3301 for (pg_off = offset_in_page(start);
3302 pg_off < offset_in_page(end);
3303 pg_off += sectorsize, bio_offset += sectorsize) {
3304 u64 file_offset = pg_off + page_offset(page);
3305 int ret;
3306
3307 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3308 test_range_bit(io_tree, file_offset,
3309 file_offset + sectorsize - 1,
3310 EXTENT_NODATASUM, 1, NULL)) {
3311 /* Skip the range without csum for data reloc inode */
3312 clear_extent_bits(io_tree, file_offset,
3313 file_offset + sectorsize - 1,
3314 EXTENT_NODATASUM);
3315 continue;
3316 }
3317 ret = check_data_csum(inode, io_bio, bio_offset, page, pg_off,
3318 page_offset(page) + pg_off);
3319 if (ret < 0) {
3320 const int nr_bit = (pg_off - offset_in_page(start)) >>
3321 root->fs_info->sectorsize_bits;
3322
3323 result |= (1U << nr_bit);
3324 }
3325 }
3326 return result;
3327 }
3328
3329 /*
3330 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3331 *
3332 * @inode: The inode we want to perform iput on
3333 *
3334 * This function uses the generic vfs_inode::i_count to track whether we should
3335 * just decrement it (in case it's > 1) or if this is the last iput then link
3336 * the inode to the delayed iput machinery. Delayed iputs are processed at
3337 * transaction commit time/superblock commit/cleaner kthread.
3338 */
btrfs_add_delayed_iput(struct inode * inode)3339 void btrfs_add_delayed_iput(struct inode *inode)
3340 {
3341 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3342 struct btrfs_inode *binode = BTRFS_I(inode);
3343
3344 if (atomic_add_unless(&inode->i_count, -1, 1))
3345 return;
3346
3347 atomic_inc(&fs_info->nr_delayed_iputs);
3348 spin_lock(&fs_info->delayed_iput_lock);
3349 ASSERT(list_empty(&binode->delayed_iput));
3350 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3351 spin_unlock(&fs_info->delayed_iput_lock);
3352 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3353 wake_up_process(fs_info->cleaner_kthread);
3354 }
3355
run_delayed_iput_locked(struct btrfs_fs_info * fs_info,struct btrfs_inode * inode)3356 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3357 struct btrfs_inode *inode)
3358 {
3359 list_del_init(&inode->delayed_iput);
3360 spin_unlock(&fs_info->delayed_iput_lock);
3361 iput(&inode->vfs_inode);
3362 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3363 wake_up(&fs_info->delayed_iputs_wait);
3364 spin_lock(&fs_info->delayed_iput_lock);
3365 }
3366
btrfs_run_delayed_iput(struct btrfs_fs_info * fs_info,struct btrfs_inode * inode)3367 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3368 struct btrfs_inode *inode)
3369 {
3370 if (!list_empty(&inode->delayed_iput)) {
3371 spin_lock(&fs_info->delayed_iput_lock);
3372 if (!list_empty(&inode->delayed_iput))
3373 run_delayed_iput_locked(fs_info, inode);
3374 spin_unlock(&fs_info->delayed_iput_lock);
3375 }
3376 }
3377
btrfs_run_delayed_iputs(struct btrfs_fs_info * fs_info)3378 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3379 {
3380
3381 spin_lock(&fs_info->delayed_iput_lock);
3382 while (!list_empty(&fs_info->delayed_iputs)) {
3383 struct btrfs_inode *inode;
3384
3385 inode = list_first_entry(&fs_info->delayed_iputs,
3386 struct btrfs_inode, delayed_iput);
3387 run_delayed_iput_locked(fs_info, inode);
3388 cond_resched_lock(&fs_info->delayed_iput_lock);
3389 }
3390 spin_unlock(&fs_info->delayed_iput_lock);
3391 }
3392
3393 /**
3394 * Wait for flushing all delayed iputs
3395 *
3396 * @fs_info: the filesystem
3397 *
3398 * This will wait on any delayed iputs that are currently running with KILLABLE
3399 * set. Once they are all done running we will return, unless we are killed in
3400 * which case we return EINTR. This helps in user operations like fallocate etc
3401 * that might get blocked on the iputs.
3402 *
3403 * Return EINTR if we were killed, 0 if nothing's pending
3404 */
btrfs_wait_on_delayed_iputs(struct btrfs_fs_info * fs_info)3405 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3406 {
3407 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3408 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3409 if (ret)
3410 return -EINTR;
3411 return 0;
3412 }
3413
3414 /*
3415 * This creates an orphan entry for the given inode in case something goes wrong
3416 * in the middle of an unlink.
3417 */
btrfs_orphan_add(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)3418 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3419 struct btrfs_inode *inode)
3420 {
3421 int ret;
3422
3423 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3424 if (ret && ret != -EEXIST) {
3425 btrfs_abort_transaction(trans, ret);
3426 return ret;
3427 }
3428
3429 return 0;
3430 }
3431
3432 /*
3433 * We have done the delete so we can go ahead and remove the orphan item for
3434 * this particular inode.
3435 */
btrfs_orphan_del(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)3436 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3437 struct btrfs_inode *inode)
3438 {
3439 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3440 }
3441
3442 /*
3443 * this cleans up any orphans that may be left on the list from the last use
3444 * of this root.
3445 */
btrfs_orphan_cleanup(struct btrfs_root * root)3446 int btrfs_orphan_cleanup(struct btrfs_root *root)
3447 {
3448 struct btrfs_fs_info *fs_info = root->fs_info;
3449 struct btrfs_path *path;
3450 struct extent_buffer *leaf;
3451 struct btrfs_key key, found_key;
3452 struct btrfs_trans_handle *trans;
3453 struct inode *inode;
3454 u64 last_objectid = 0;
3455 int ret = 0, nr_unlink = 0;
3456
3457 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3458 return 0;
3459
3460 path = btrfs_alloc_path();
3461 if (!path) {
3462 ret = -ENOMEM;
3463 goto out;
3464 }
3465 path->reada = READA_BACK;
3466
3467 key.objectid = BTRFS_ORPHAN_OBJECTID;
3468 key.type = BTRFS_ORPHAN_ITEM_KEY;
3469 key.offset = (u64)-1;
3470
3471 while (1) {
3472 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3473 if (ret < 0)
3474 goto out;
3475
3476 /*
3477 * if ret == 0 means we found what we were searching for, which
3478 * is weird, but possible, so only screw with path if we didn't
3479 * find the key and see if we have stuff that matches
3480 */
3481 if (ret > 0) {
3482 ret = 0;
3483 if (path->slots[0] == 0)
3484 break;
3485 path->slots[0]--;
3486 }
3487
3488 /* pull out the item */
3489 leaf = path->nodes[0];
3490 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3491
3492 /* make sure the item matches what we want */
3493 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3494 break;
3495 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3496 break;
3497
3498 /* release the path since we're done with it */
3499 btrfs_release_path(path);
3500
3501 /*
3502 * this is where we are basically btrfs_lookup, without the
3503 * crossing root thing. we store the inode number in the
3504 * offset of the orphan item.
3505 */
3506
3507 if (found_key.offset == last_objectid) {
3508 btrfs_err(fs_info,
3509 "Error removing orphan entry, stopping orphan cleanup");
3510 ret = -EINVAL;
3511 goto out;
3512 }
3513
3514 last_objectid = found_key.offset;
3515
3516 found_key.objectid = found_key.offset;
3517 found_key.type = BTRFS_INODE_ITEM_KEY;
3518 found_key.offset = 0;
3519 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3520 ret = PTR_ERR_OR_ZERO(inode);
3521 if (ret && ret != -ENOENT)
3522 goto out;
3523
3524 if (ret == -ENOENT && root == fs_info->tree_root) {
3525 struct btrfs_root *dead_root;
3526 int is_dead_root = 0;
3527
3528 /*
3529 * This is an orphan in the tree root. Currently these
3530 * could come from 2 sources:
3531 * a) a root (snapshot/subvolume) deletion in progress
3532 * b) a free space cache inode
3533 * We need to distinguish those two, as the orphan item
3534 * for a root must not get deleted before the deletion
3535 * of the snapshot/subvolume's tree completes.
3536 *
3537 * btrfs_find_orphan_roots() ran before us, which has
3538 * found all deleted roots and loaded them into
3539 * fs_info->fs_roots_radix. So here we can find if an
3540 * orphan item corresponds to a deleted root by looking
3541 * up the root from that radix tree.
3542 */
3543
3544 spin_lock(&fs_info->fs_roots_radix_lock);
3545 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3546 (unsigned long)found_key.objectid);
3547 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3548 is_dead_root = 1;
3549 spin_unlock(&fs_info->fs_roots_radix_lock);
3550
3551 if (is_dead_root) {
3552 /* prevent this orphan from being found again */
3553 key.offset = found_key.objectid - 1;
3554 continue;
3555 }
3556
3557 }
3558
3559 /*
3560 * If we have an inode with links, there are a couple of
3561 * possibilities:
3562 *
3563 * 1. We were halfway through creating fsverity metadata for the
3564 * file. In that case, the orphan item represents incomplete
3565 * fsverity metadata which must be cleaned up with
3566 * btrfs_drop_verity_items and deleting the orphan item.
3567
3568 * 2. Old kernels (before v3.12) used to create an
3569 * orphan item for truncate indicating that there were possibly
3570 * extent items past i_size that needed to be deleted. In v3.12,
3571 * truncate was changed to update i_size in sync with the extent
3572 * items, but the (useless) orphan item was still created. Since
3573 * v4.18, we don't create the orphan item for truncate at all.
3574 *
3575 * So, this item could mean that we need to do a truncate, but
3576 * only if this filesystem was last used on a pre-v3.12 kernel
3577 * and was not cleanly unmounted. The odds of that are quite
3578 * slim, and it's a pain to do the truncate now, so just delete
3579 * the orphan item.
3580 *
3581 * It's also possible that this orphan item was supposed to be
3582 * deleted but wasn't. The inode number may have been reused,
3583 * but either way, we can delete the orphan item.
3584 */
3585 if (ret == -ENOENT || inode->i_nlink) {
3586 if (!ret) {
3587 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3588 iput(inode);
3589 if (ret)
3590 goto out;
3591 }
3592 trans = btrfs_start_transaction(root, 1);
3593 if (IS_ERR(trans)) {
3594 ret = PTR_ERR(trans);
3595 goto out;
3596 }
3597 btrfs_debug(fs_info, "auto deleting %Lu",
3598 found_key.objectid);
3599 ret = btrfs_del_orphan_item(trans, root,
3600 found_key.objectid);
3601 btrfs_end_transaction(trans);
3602 if (ret)
3603 goto out;
3604 continue;
3605 }
3606
3607 nr_unlink++;
3608
3609 /* this will do delete_inode and everything for us */
3610 iput(inode);
3611 }
3612 /* release the path since we're done with it */
3613 btrfs_release_path(path);
3614
3615 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3616
3617 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3618 trans = btrfs_join_transaction(root);
3619 if (!IS_ERR(trans))
3620 btrfs_end_transaction(trans);
3621 }
3622
3623 if (nr_unlink)
3624 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3625
3626 out:
3627 if (ret)
3628 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3629 btrfs_free_path(path);
3630 return ret;
3631 }
3632
3633 /*
3634 * very simple check to peek ahead in the leaf looking for xattrs. If we
3635 * don't find any xattrs, we know there can't be any acls.
3636 *
3637 * slot is the slot the inode is in, objectid is the objectid of the inode
3638 */
acls_after_inode_item(struct extent_buffer * leaf,int slot,u64 objectid,int * first_xattr_slot)3639 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3640 int slot, u64 objectid,
3641 int *first_xattr_slot)
3642 {
3643 u32 nritems = btrfs_header_nritems(leaf);
3644 struct btrfs_key found_key;
3645 static u64 xattr_access = 0;
3646 static u64 xattr_default = 0;
3647 int scanned = 0;
3648
3649 if (!xattr_access) {
3650 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3651 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3652 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3653 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3654 }
3655
3656 slot++;
3657 *first_xattr_slot = -1;
3658 while (slot < nritems) {
3659 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3660
3661 /* we found a different objectid, there must not be acls */
3662 if (found_key.objectid != objectid)
3663 return 0;
3664
3665 /* we found an xattr, assume we've got an acl */
3666 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3667 if (*first_xattr_slot == -1)
3668 *first_xattr_slot = slot;
3669 if (found_key.offset == xattr_access ||
3670 found_key.offset == xattr_default)
3671 return 1;
3672 }
3673
3674 /*
3675 * we found a key greater than an xattr key, there can't
3676 * be any acls later on
3677 */
3678 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3679 return 0;
3680
3681 slot++;
3682 scanned++;
3683
3684 /*
3685 * it goes inode, inode backrefs, xattrs, extents,
3686 * so if there are a ton of hard links to an inode there can
3687 * be a lot of backrefs. Don't waste time searching too hard,
3688 * this is just an optimization
3689 */
3690 if (scanned >= 8)
3691 break;
3692 }
3693 /* we hit the end of the leaf before we found an xattr or
3694 * something larger than an xattr. We have to assume the inode
3695 * has acls
3696 */
3697 if (*first_xattr_slot == -1)
3698 *first_xattr_slot = slot;
3699 return 1;
3700 }
3701
3702 /*
3703 * read an inode from the btree into the in-memory inode
3704 */
btrfs_read_locked_inode(struct inode * inode,struct btrfs_path * in_path)3705 static int btrfs_read_locked_inode(struct inode *inode,
3706 struct btrfs_path *in_path)
3707 {
3708 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3709 struct btrfs_path *path = in_path;
3710 struct extent_buffer *leaf;
3711 struct btrfs_inode_item *inode_item;
3712 struct btrfs_root *root = BTRFS_I(inode)->root;
3713 struct btrfs_key location;
3714 unsigned long ptr;
3715 int maybe_acls;
3716 u32 rdev;
3717 int ret;
3718 bool filled = false;
3719 int first_xattr_slot;
3720
3721 ret = btrfs_fill_inode(inode, &rdev);
3722 if (!ret)
3723 filled = true;
3724
3725 if (!path) {
3726 path = btrfs_alloc_path();
3727 if (!path)
3728 return -ENOMEM;
3729 }
3730
3731 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3732
3733 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3734 if (ret) {
3735 if (path != in_path)
3736 btrfs_free_path(path);
3737 return ret;
3738 }
3739
3740 leaf = path->nodes[0];
3741
3742 if (filled)
3743 goto cache_index;
3744
3745 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3746 struct btrfs_inode_item);
3747 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3748 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3749 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3750 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3751 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3752 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3753 round_up(i_size_read(inode), fs_info->sectorsize));
3754
3755 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3756 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3757
3758 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3759 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3760
3761 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3762 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3763
3764 BTRFS_I(inode)->i_otime.tv_sec =
3765 btrfs_timespec_sec(leaf, &inode_item->otime);
3766 BTRFS_I(inode)->i_otime.tv_nsec =
3767 btrfs_timespec_nsec(leaf, &inode_item->otime);
3768
3769 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3770 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3771 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3772
3773 inode_set_iversion_queried(inode,
3774 btrfs_inode_sequence(leaf, inode_item));
3775 inode->i_generation = BTRFS_I(inode)->generation;
3776 inode->i_rdev = 0;
3777 rdev = btrfs_inode_rdev(leaf, inode_item);
3778
3779 BTRFS_I(inode)->index_cnt = (u64)-1;
3780 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3781 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3782
3783 cache_index:
3784 /*
3785 * If we were modified in the current generation and evicted from memory
3786 * and then re-read we need to do a full sync since we don't have any
3787 * idea about which extents were modified before we were evicted from
3788 * cache.
3789 *
3790 * This is required for both inode re-read from disk and delayed inode
3791 * in delayed_nodes_tree.
3792 */
3793 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3794 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3795 &BTRFS_I(inode)->runtime_flags);
3796
3797 /*
3798 * We don't persist the id of the transaction where an unlink operation
3799 * against the inode was last made. So here we assume the inode might
3800 * have been evicted, and therefore the exact value of last_unlink_trans
3801 * lost, and set it to last_trans to avoid metadata inconsistencies
3802 * between the inode and its parent if the inode is fsync'ed and the log
3803 * replayed. For example, in the scenario:
3804 *
3805 * touch mydir/foo
3806 * ln mydir/foo mydir/bar
3807 * sync
3808 * unlink mydir/bar
3809 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3810 * xfs_io -c fsync mydir/foo
3811 * <power failure>
3812 * mount fs, triggers fsync log replay
3813 *
3814 * We must make sure that when we fsync our inode foo we also log its
3815 * parent inode, otherwise after log replay the parent still has the
3816 * dentry with the "bar" name but our inode foo has a link count of 1
3817 * and doesn't have an inode ref with the name "bar" anymore.
3818 *
3819 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3820 * but it guarantees correctness at the expense of occasional full
3821 * transaction commits on fsync if our inode is a directory, or if our
3822 * inode is not a directory, logging its parent unnecessarily.
3823 */
3824 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3825
3826 /*
3827 * Same logic as for last_unlink_trans. We don't persist the generation
3828 * of the last transaction where this inode was used for a reflink
3829 * operation, so after eviction and reloading the inode we must be
3830 * pessimistic and assume the last transaction that modified the inode.
3831 */
3832 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3833
3834 path->slots[0]++;
3835 if (inode->i_nlink != 1 ||
3836 path->slots[0] >= btrfs_header_nritems(leaf))
3837 goto cache_acl;
3838
3839 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3840 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3841 goto cache_acl;
3842
3843 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3844 if (location.type == BTRFS_INODE_REF_KEY) {
3845 struct btrfs_inode_ref *ref;
3846
3847 ref = (struct btrfs_inode_ref *)ptr;
3848 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3849 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3850 struct btrfs_inode_extref *extref;
3851
3852 extref = (struct btrfs_inode_extref *)ptr;
3853 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3854 extref);
3855 }
3856 cache_acl:
3857 /*
3858 * try to precache a NULL acl entry for files that don't have
3859 * any xattrs or acls
3860 */
3861 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3862 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3863 if (first_xattr_slot != -1) {
3864 path->slots[0] = first_xattr_slot;
3865 ret = btrfs_load_inode_props(inode, path);
3866 if (ret)
3867 btrfs_err(fs_info,
3868 "error loading props for ino %llu (root %llu): %d",
3869 btrfs_ino(BTRFS_I(inode)),
3870 root->root_key.objectid, ret);
3871 }
3872 if (path != in_path)
3873 btrfs_free_path(path);
3874
3875 if (!maybe_acls)
3876 cache_no_acl(inode);
3877
3878 switch (inode->i_mode & S_IFMT) {
3879 case S_IFREG:
3880 inode->i_mapping->a_ops = &btrfs_aops;
3881 inode->i_fop = &btrfs_file_operations;
3882 inode->i_op = &btrfs_file_inode_operations;
3883 break;
3884 case S_IFDIR:
3885 inode->i_fop = &btrfs_dir_file_operations;
3886 inode->i_op = &btrfs_dir_inode_operations;
3887 break;
3888 case S_IFLNK:
3889 inode->i_op = &btrfs_symlink_inode_operations;
3890 inode_nohighmem(inode);
3891 inode->i_mapping->a_ops = &btrfs_aops;
3892 break;
3893 default:
3894 inode->i_op = &btrfs_special_inode_operations;
3895 init_special_inode(inode, inode->i_mode, rdev);
3896 break;
3897 }
3898
3899 btrfs_sync_inode_flags_to_i_flags(inode);
3900 return 0;
3901 }
3902
3903 /*
3904 * given a leaf and an inode, copy the inode fields into the leaf
3905 */
fill_inode_item(struct btrfs_trans_handle * trans,struct extent_buffer * leaf,struct btrfs_inode_item * item,struct inode * inode)3906 static void fill_inode_item(struct btrfs_trans_handle *trans,
3907 struct extent_buffer *leaf,
3908 struct btrfs_inode_item *item,
3909 struct inode *inode)
3910 {
3911 struct btrfs_map_token token;
3912 u64 flags;
3913
3914 btrfs_init_map_token(&token, leaf);
3915
3916 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3917 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3918 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3919 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3920 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3921
3922 btrfs_set_token_timespec_sec(&token, &item->atime,
3923 inode->i_atime.tv_sec);
3924 btrfs_set_token_timespec_nsec(&token, &item->atime,
3925 inode->i_atime.tv_nsec);
3926
3927 btrfs_set_token_timespec_sec(&token, &item->mtime,
3928 inode->i_mtime.tv_sec);
3929 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3930 inode->i_mtime.tv_nsec);
3931
3932 btrfs_set_token_timespec_sec(&token, &item->ctime,
3933 inode->i_ctime.tv_sec);
3934 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3935 inode->i_ctime.tv_nsec);
3936
3937 btrfs_set_token_timespec_sec(&token, &item->otime,
3938 BTRFS_I(inode)->i_otime.tv_sec);
3939 btrfs_set_token_timespec_nsec(&token, &item->otime,
3940 BTRFS_I(inode)->i_otime.tv_nsec);
3941
3942 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3943 btrfs_set_token_inode_generation(&token, item,
3944 BTRFS_I(inode)->generation);
3945 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3946 btrfs_set_token_inode_transid(&token, item, trans->transid);
3947 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3948 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3949 BTRFS_I(inode)->ro_flags);
3950 btrfs_set_token_inode_flags(&token, item, flags);
3951 btrfs_set_token_inode_block_group(&token, item, 0);
3952 }
3953
3954 /*
3955 * copy everything in the in-memory inode into the btree.
3956 */
btrfs_update_inode_item(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode)3957 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3958 struct btrfs_root *root,
3959 struct btrfs_inode *inode)
3960 {
3961 struct btrfs_inode_item *inode_item;
3962 struct btrfs_path *path;
3963 struct extent_buffer *leaf;
3964 int ret;
3965
3966 path = btrfs_alloc_path();
3967 if (!path)
3968 return -ENOMEM;
3969
3970 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
3971 if (ret) {
3972 if (ret > 0)
3973 ret = -ENOENT;
3974 goto failed;
3975 }
3976
3977 leaf = path->nodes[0];
3978 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3979 struct btrfs_inode_item);
3980
3981 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
3982 btrfs_mark_buffer_dirty(leaf);
3983 btrfs_set_inode_last_trans(trans, inode);
3984 ret = 0;
3985 failed:
3986 btrfs_free_path(path);
3987 return ret;
3988 }
3989
3990 /*
3991 * copy everything in the in-memory inode into the btree.
3992 */
btrfs_update_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode)3993 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3994 struct btrfs_root *root,
3995 struct btrfs_inode *inode)
3996 {
3997 struct btrfs_fs_info *fs_info = root->fs_info;
3998 int ret;
3999
4000 /*
4001 * If the inode is a free space inode, we can deadlock during commit
4002 * if we put it into the delayed code.
4003 *
4004 * The data relocation inode should also be directly updated
4005 * without delay
4006 */
4007 if (!btrfs_is_free_space_inode(inode)
4008 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
4009 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4010 btrfs_update_root_times(trans, root);
4011
4012 ret = btrfs_delayed_update_inode(trans, root, inode);
4013 if (!ret)
4014 btrfs_set_inode_last_trans(trans, inode);
4015 return ret;
4016 }
4017
4018 return btrfs_update_inode_item(trans, root, inode);
4019 }
4020
btrfs_update_inode_fallback(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode)4021 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4022 struct btrfs_root *root, struct btrfs_inode *inode)
4023 {
4024 int ret;
4025
4026 ret = btrfs_update_inode(trans, root, inode);
4027 if (ret == -ENOSPC)
4028 return btrfs_update_inode_item(trans, root, inode);
4029 return ret;
4030 }
4031
4032 /*
4033 * unlink helper that gets used here in inode.c and in the tree logging
4034 * recovery code. It remove a link in a directory with a given name, and
4035 * also drops the back refs in the inode to the directory
4036 */
__btrfs_unlink_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * dir,struct btrfs_inode * inode,const char * name,int name_len)4037 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4038 struct btrfs_root *root,
4039 struct btrfs_inode *dir,
4040 struct btrfs_inode *inode,
4041 const char *name, int name_len)
4042 {
4043 struct btrfs_fs_info *fs_info = root->fs_info;
4044 struct btrfs_path *path;
4045 int ret = 0;
4046 struct btrfs_dir_item *di;
4047 u64 index;
4048 u64 ino = btrfs_ino(inode);
4049 u64 dir_ino = btrfs_ino(dir);
4050
4051 path = btrfs_alloc_path();
4052 if (!path) {
4053 ret = -ENOMEM;
4054 goto out;
4055 }
4056
4057 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4058 name, name_len, -1);
4059 if (IS_ERR_OR_NULL(di)) {
4060 ret = di ? PTR_ERR(di) : -ENOENT;
4061 goto err;
4062 }
4063 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4064 if (ret)
4065 goto err;
4066 btrfs_release_path(path);
4067
4068 /*
4069 * If we don't have dir index, we have to get it by looking up
4070 * the inode ref, since we get the inode ref, remove it directly,
4071 * it is unnecessary to do delayed deletion.
4072 *
4073 * But if we have dir index, needn't search inode ref to get it.
4074 * Since the inode ref is close to the inode item, it is better
4075 * that we delay to delete it, and just do this deletion when
4076 * we update the inode item.
4077 */
4078 if (inode->dir_index) {
4079 ret = btrfs_delayed_delete_inode_ref(inode);
4080 if (!ret) {
4081 index = inode->dir_index;
4082 goto skip_backref;
4083 }
4084 }
4085
4086 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4087 dir_ino, &index);
4088 if (ret) {
4089 btrfs_info(fs_info,
4090 "failed to delete reference to %.*s, inode %llu parent %llu",
4091 name_len, name, ino, dir_ino);
4092 btrfs_abort_transaction(trans, ret);
4093 goto err;
4094 }
4095 skip_backref:
4096 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4097 if (ret) {
4098 btrfs_abort_transaction(trans, ret);
4099 goto err;
4100 }
4101
4102 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4103 dir_ino);
4104 if (ret != 0 && ret != -ENOENT) {
4105 btrfs_abort_transaction(trans, ret);
4106 goto err;
4107 }
4108
4109 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4110 index);
4111 if (ret == -ENOENT)
4112 ret = 0;
4113 else if (ret)
4114 btrfs_abort_transaction(trans, ret);
4115
4116 /*
4117 * If we have a pending delayed iput we could end up with the final iput
4118 * being run in btrfs-cleaner context. If we have enough of these built
4119 * up we can end up burning a lot of time in btrfs-cleaner without any
4120 * way to throttle the unlinks. Since we're currently holding a ref on
4121 * the inode we can run the delayed iput here without any issues as the
4122 * final iput won't be done until after we drop the ref we're currently
4123 * holding.
4124 */
4125 btrfs_run_delayed_iput(fs_info, inode);
4126 err:
4127 btrfs_free_path(path);
4128 if (ret)
4129 goto out;
4130
4131 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4132 inode_inc_iversion(&inode->vfs_inode);
4133 inode_inc_iversion(&dir->vfs_inode);
4134 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4135 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4136 ret = btrfs_update_inode(trans, root, dir);
4137 out:
4138 return ret;
4139 }
4140
btrfs_unlink_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * dir,struct btrfs_inode * inode,const char * name,int name_len)4141 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4142 struct btrfs_root *root,
4143 struct btrfs_inode *dir, struct btrfs_inode *inode,
4144 const char *name, int name_len)
4145 {
4146 int ret;
4147 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4148 if (!ret) {
4149 drop_nlink(&inode->vfs_inode);
4150 ret = btrfs_update_inode(trans, root, inode);
4151 }
4152 return ret;
4153 }
4154
4155 /*
4156 * helper to start transaction for unlink and rmdir.
4157 *
4158 * unlink and rmdir are special in btrfs, they do not always free space, so
4159 * if we cannot make our reservations the normal way try and see if there is
4160 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4161 * allow the unlink to occur.
4162 */
__unlink_start_trans(struct inode * dir)4163 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4164 {
4165 struct btrfs_root *root = BTRFS_I(dir)->root;
4166
4167 /*
4168 * 1 for the possible orphan item
4169 * 1 for the dir item
4170 * 1 for the dir index
4171 * 1 for the inode ref
4172 * 1 for the inode
4173 */
4174 return btrfs_start_transaction_fallback_global_rsv(root, 5);
4175 }
4176
btrfs_unlink(struct inode * dir,struct dentry * dentry)4177 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4178 {
4179 struct btrfs_root *root = BTRFS_I(dir)->root;
4180 struct btrfs_trans_handle *trans;
4181 struct inode *inode = d_inode(dentry);
4182 int ret;
4183
4184 trans = __unlink_start_trans(dir);
4185 if (IS_ERR(trans))
4186 return PTR_ERR(trans);
4187
4188 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4189 0);
4190
4191 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4192 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4193 dentry->d_name.len);
4194 if (ret)
4195 goto out;
4196
4197 if (inode->i_nlink == 0) {
4198 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4199 if (ret)
4200 goto out;
4201 }
4202
4203 out:
4204 btrfs_end_transaction(trans);
4205 btrfs_btree_balance_dirty(root->fs_info);
4206 return ret;
4207 }
4208
btrfs_unlink_subvol(struct btrfs_trans_handle * trans,struct inode * dir,struct dentry * dentry)4209 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4210 struct inode *dir, struct dentry *dentry)
4211 {
4212 struct btrfs_root *root = BTRFS_I(dir)->root;
4213 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4214 struct btrfs_path *path;
4215 struct extent_buffer *leaf;
4216 struct btrfs_dir_item *di;
4217 struct btrfs_key key;
4218 const char *name = dentry->d_name.name;
4219 int name_len = dentry->d_name.len;
4220 u64 index;
4221 int ret;
4222 u64 objectid;
4223 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4224
4225 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4226 objectid = inode->root->root_key.objectid;
4227 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4228 objectid = inode->location.objectid;
4229 } else {
4230 WARN_ON(1);
4231 return -EINVAL;
4232 }
4233
4234 path = btrfs_alloc_path();
4235 if (!path)
4236 return -ENOMEM;
4237
4238 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4239 name, name_len, -1);
4240 if (IS_ERR_OR_NULL(di)) {
4241 ret = di ? PTR_ERR(di) : -ENOENT;
4242 goto out;
4243 }
4244
4245 leaf = path->nodes[0];
4246 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4247 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4248 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4249 if (ret) {
4250 btrfs_abort_transaction(trans, ret);
4251 goto out;
4252 }
4253 btrfs_release_path(path);
4254
4255 /*
4256 * This is a placeholder inode for a subvolume we didn't have a
4257 * reference to at the time of the snapshot creation. In the meantime
4258 * we could have renamed the real subvol link into our snapshot, so
4259 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4260 * Instead simply lookup the dir_index_item for this entry so we can
4261 * remove it. Otherwise we know we have a ref to the root and we can
4262 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4263 */
4264 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4265 di = btrfs_search_dir_index_item(root, path, dir_ino,
4266 name, name_len);
4267 if (IS_ERR_OR_NULL(di)) {
4268 if (!di)
4269 ret = -ENOENT;
4270 else
4271 ret = PTR_ERR(di);
4272 btrfs_abort_transaction(trans, ret);
4273 goto out;
4274 }
4275
4276 leaf = path->nodes[0];
4277 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4278 index = key.offset;
4279 btrfs_release_path(path);
4280 } else {
4281 ret = btrfs_del_root_ref(trans, objectid,
4282 root->root_key.objectid, dir_ino,
4283 &index, name, name_len);
4284 if (ret) {
4285 btrfs_abort_transaction(trans, ret);
4286 goto out;
4287 }
4288 }
4289
4290 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4291 if (ret) {
4292 btrfs_abort_transaction(trans, ret);
4293 goto out;
4294 }
4295
4296 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4297 inode_inc_iversion(dir);
4298 dir->i_mtime = dir->i_ctime = current_time(dir);
4299 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4300 if (ret)
4301 btrfs_abort_transaction(trans, ret);
4302 out:
4303 btrfs_free_path(path);
4304 return ret;
4305 }
4306
4307 /*
4308 * Helper to check if the subvolume references other subvolumes or if it's
4309 * default.
4310 */
may_destroy_subvol(struct btrfs_root * root)4311 static noinline int may_destroy_subvol(struct btrfs_root *root)
4312 {
4313 struct btrfs_fs_info *fs_info = root->fs_info;
4314 struct btrfs_path *path;
4315 struct btrfs_dir_item *di;
4316 struct btrfs_key key;
4317 u64 dir_id;
4318 int ret;
4319
4320 path = btrfs_alloc_path();
4321 if (!path)
4322 return -ENOMEM;
4323
4324 /* Make sure this root isn't set as the default subvol */
4325 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4326 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4327 dir_id, "default", 7, 0);
4328 if (di && !IS_ERR(di)) {
4329 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4330 if (key.objectid == root->root_key.objectid) {
4331 ret = -EPERM;
4332 btrfs_err(fs_info,
4333 "deleting default subvolume %llu is not allowed",
4334 key.objectid);
4335 goto out;
4336 }
4337 btrfs_release_path(path);
4338 }
4339
4340 key.objectid = root->root_key.objectid;
4341 key.type = BTRFS_ROOT_REF_KEY;
4342 key.offset = (u64)-1;
4343
4344 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4345 if (ret < 0)
4346 goto out;
4347 BUG_ON(ret == 0);
4348
4349 ret = 0;
4350 if (path->slots[0] > 0) {
4351 path->slots[0]--;
4352 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4353 if (key.objectid == root->root_key.objectid &&
4354 key.type == BTRFS_ROOT_REF_KEY)
4355 ret = -ENOTEMPTY;
4356 }
4357 out:
4358 btrfs_free_path(path);
4359 return ret;
4360 }
4361
4362 /* Delete all dentries for inodes belonging to the root */
btrfs_prune_dentries(struct btrfs_root * root)4363 static void btrfs_prune_dentries(struct btrfs_root *root)
4364 {
4365 struct btrfs_fs_info *fs_info = root->fs_info;
4366 struct rb_node *node;
4367 struct rb_node *prev;
4368 struct btrfs_inode *entry;
4369 struct inode *inode;
4370 u64 objectid = 0;
4371
4372 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4373 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4374
4375 spin_lock(&root->inode_lock);
4376 again:
4377 node = root->inode_tree.rb_node;
4378 prev = NULL;
4379 while (node) {
4380 prev = node;
4381 entry = rb_entry(node, struct btrfs_inode, rb_node);
4382
4383 if (objectid < btrfs_ino(entry))
4384 node = node->rb_left;
4385 else if (objectid > btrfs_ino(entry))
4386 node = node->rb_right;
4387 else
4388 break;
4389 }
4390 if (!node) {
4391 while (prev) {
4392 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4393 if (objectid <= btrfs_ino(entry)) {
4394 node = prev;
4395 break;
4396 }
4397 prev = rb_next(prev);
4398 }
4399 }
4400 while (node) {
4401 entry = rb_entry(node, struct btrfs_inode, rb_node);
4402 objectid = btrfs_ino(entry) + 1;
4403 inode = igrab(&entry->vfs_inode);
4404 if (inode) {
4405 spin_unlock(&root->inode_lock);
4406 if (atomic_read(&inode->i_count) > 1)
4407 d_prune_aliases(inode);
4408 /*
4409 * btrfs_drop_inode will have it removed from the inode
4410 * cache when its usage count hits zero.
4411 */
4412 iput(inode);
4413 cond_resched();
4414 spin_lock(&root->inode_lock);
4415 goto again;
4416 }
4417
4418 if (cond_resched_lock(&root->inode_lock))
4419 goto again;
4420
4421 node = rb_next(node);
4422 }
4423 spin_unlock(&root->inode_lock);
4424 }
4425
btrfs_delete_subvolume(struct inode * dir,struct dentry * dentry)4426 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4427 {
4428 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4429 struct btrfs_root *root = BTRFS_I(dir)->root;
4430 struct inode *inode = d_inode(dentry);
4431 struct btrfs_root *dest = BTRFS_I(inode)->root;
4432 struct btrfs_trans_handle *trans;
4433 struct btrfs_block_rsv block_rsv;
4434 u64 root_flags;
4435 int ret;
4436
4437 /*
4438 * Don't allow to delete a subvolume with send in progress. This is
4439 * inside the inode lock so the error handling that has to drop the bit
4440 * again is not run concurrently.
4441 */
4442 spin_lock(&dest->root_item_lock);
4443 if (dest->send_in_progress) {
4444 spin_unlock(&dest->root_item_lock);
4445 btrfs_warn(fs_info,
4446 "attempt to delete subvolume %llu during send",
4447 dest->root_key.objectid);
4448 return -EPERM;
4449 }
4450 root_flags = btrfs_root_flags(&dest->root_item);
4451 btrfs_set_root_flags(&dest->root_item,
4452 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4453 spin_unlock(&dest->root_item_lock);
4454
4455 down_write(&fs_info->subvol_sem);
4456
4457 ret = may_destroy_subvol(dest);
4458 if (ret)
4459 goto out_up_write;
4460
4461 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4462 /*
4463 * One for dir inode,
4464 * two for dir entries,
4465 * two for root ref/backref.
4466 */
4467 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4468 if (ret)
4469 goto out_up_write;
4470
4471 trans = btrfs_start_transaction(root, 0);
4472 if (IS_ERR(trans)) {
4473 ret = PTR_ERR(trans);
4474 goto out_release;
4475 }
4476 trans->block_rsv = &block_rsv;
4477 trans->bytes_reserved = block_rsv.size;
4478
4479 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4480
4481 ret = btrfs_unlink_subvol(trans, dir, dentry);
4482 if (ret) {
4483 btrfs_abort_transaction(trans, ret);
4484 goto out_end_trans;
4485 }
4486
4487 ret = btrfs_record_root_in_trans(trans, dest);
4488 if (ret) {
4489 btrfs_abort_transaction(trans, ret);
4490 goto out_end_trans;
4491 }
4492
4493 memset(&dest->root_item.drop_progress, 0,
4494 sizeof(dest->root_item.drop_progress));
4495 btrfs_set_root_drop_level(&dest->root_item, 0);
4496 btrfs_set_root_refs(&dest->root_item, 0);
4497
4498 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4499 ret = btrfs_insert_orphan_item(trans,
4500 fs_info->tree_root,
4501 dest->root_key.objectid);
4502 if (ret) {
4503 btrfs_abort_transaction(trans, ret);
4504 goto out_end_trans;
4505 }
4506 }
4507
4508 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4509 BTRFS_UUID_KEY_SUBVOL,
4510 dest->root_key.objectid);
4511 if (ret && ret != -ENOENT) {
4512 btrfs_abort_transaction(trans, ret);
4513 goto out_end_trans;
4514 }
4515 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4516 ret = btrfs_uuid_tree_remove(trans,
4517 dest->root_item.received_uuid,
4518 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4519 dest->root_key.objectid);
4520 if (ret && ret != -ENOENT) {
4521 btrfs_abort_transaction(trans, ret);
4522 goto out_end_trans;
4523 }
4524 }
4525
4526 free_anon_bdev(dest->anon_dev);
4527 dest->anon_dev = 0;
4528 out_end_trans:
4529 trans->block_rsv = NULL;
4530 trans->bytes_reserved = 0;
4531 ret = btrfs_end_transaction(trans);
4532 inode->i_flags |= S_DEAD;
4533 out_release:
4534 btrfs_subvolume_release_metadata(root, &block_rsv);
4535 out_up_write:
4536 up_write(&fs_info->subvol_sem);
4537 if (ret) {
4538 spin_lock(&dest->root_item_lock);
4539 root_flags = btrfs_root_flags(&dest->root_item);
4540 btrfs_set_root_flags(&dest->root_item,
4541 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4542 spin_unlock(&dest->root_item_lock);
4543 } else {
4544 d_invalidate(dentry);
4545 btrfs_prune_dentries(dest);
4546 ASSERT(dest->send_in_progress == 0);
4547 }
4548
4549 return ret;
4550 }
4551
btrfs_rmdir(struct inode * dir,struct dentry * dentry)4552 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4553 {
4554 struct inode *inode = d_inode(dentry);
4555 int err = 0;
4556 struct btrfs_root *root = BTRFS_I(dir)->root;
4557 struct btrfs_trans_handle *trans;
4558 u64 last_unlink_trans;
4559
4560 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4561 return -ENOTEMPTY;
4562 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4563 return btrfs_delete_subvolume(dir, dentry);
4564
4565 trans = __unlink_start_trans(dir);
4566 if (IS_ERR(trans))
4567 return PTR_ERR(trans);
4568
4569 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4570 err = btrfs_unlink_subvol(trans, dir, dentry);
4571 goto out;
4572 }
4573
4574 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4575 if (err)
4576 goto out;
4577
4578 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4579
4580 /* now the directory is empty */
4581 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4582 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4583 dentry->d_name.len);
4584 if (!err) {
4585 btrfs_i_size_write(BTRFS_I(inode), 0);
4586 /*
4587 * Propagate the last_unlink_trans value of the deleted dir to
4588 * its parent directory. This is to prevent an unrecoverable
4589 * log tree in the case we do something like this:
4590 * 1) create dir foo
4591 * 2) create snapshot under dir foo
4592 * 3) delete the snapshot
4593 * 4) rmdir foo
4594 * 5) mkdir foo
4595 * 6) fsync foo or some file inside foo
4596 */
4597 if (last_unlink_trans >= trans->transid)
4598 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4599 }
4600 out:
4601 btrfs_end_transaction(trans);
4602 btrfs_btree_balance_dirty(root->fs_info);
4603
4604 return err;
4605 }
4606
4607 /*
4608 * Return this if we need to call truncate_block for the last bit of the
4609 * truncate.
4610 */
4611 #define NEED_TRUNCATE_BLOCK 1
4612
4613 /*
4614 * Remove inode items from a given root.
4615 *
4616 * @trans: A transaction handle.
4617 * @root: The root from which to remove items.
4618 * @inode: The inode whose items we want to remove.
4619 * @new_size: The new i_size for the inode. This is only applicable when
4620 * @min_type is BTRFS_EXTENT_DATA_KEY, must be 0 otherwise.
4621 * @min_type: The minimum key type to remove. All keys with a type
4622 * greater than this value are removed and all keys with
4623 * this type are removed only if their offset is >= @new_size.
4624 * @extents_found: Output parameter that will contain the number of file
4625 * extent items that were removed or adjusted to the new
4626 * inode i_size. The caller is responsible for initializing
4627 * the counter. Also, it can be NULL if the caller does not
4628 * need this counter.
4629 *
4630 * Remove all keys associated with the inode from the given root that have a key
4631 * with a type greater than or equals to @min_type. When @min_type has a value of
4632 * BTRFS_EXTENT_DATA_KEY, only remove file extent items that have an offset value
4633 * greater than or equals to @new_size. If a file extent item that starts before
4634 * @new_size and ends after it is found, its length is adjusted.
4635 *
4636 * Returns: 0 on success, < 0 on error and NEED_TRUNCATE_BLOCK when @min_type is
4637 * BTRFS_EXTENT_DATA_KEY and the caller must truncate the last block.
4638 */
btrfs_truncate_inode_items(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode,u64 new_size,u32 min_type,u64 * extents_found)4639 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4640 struct btrfs_root *root,
4641 struct btrfs_inode *inode,
4642 u64 new_size, u32 min_type,
4643 u64 *extents_found)
4644 {
4645 struct btrfs_fs_info *fs_info = root->fs_info;
4646 struct btrfs_path *path;
4647 struct extent_buffer *leaf;
4648 struct btrfs_file_extent_item *fi;
4649 struct btrfs_key key;
4650 struct btrfs_key found_key;
4651 u64 extent_start = 0;
4652 u64 extent_num_bytes = 0;
4653 u64 extent_offset = 0;
4654 u64 item_end = 0;
4655 u64 last_size = new_size;
4656 u32 found_type = (u8)-1;
4657 int found_extent;
4658 int del_item;
4659 int pending_del_nr = 0;
4660 int pending_del_slot = 0;
4661 int extent_type = -1;
4662 int ret;
4663 u64 ino = btrfs_ino(inode);
4664 u64 bytes_deleted = 0;
4665 bool be_nice = false;
4666 bool should_throttle = false;
4667 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
4668 struct extent_state *cached_state = NULL;
4669
4670 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4671
4672 /*
4673 * For non-free space inodes and non-shareable roots, we want to back
4674 * off from time to time. This means all inodes in subvolume roots,
4675 * reloc roots, and data reloc roots.
4676 */
4677 if (!btrfs_is_free_space_inode(inode) &&
4678 test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4679 be_nice = true;
4680
4681 path = btrfs_alloc_path();
4682 if (!path)
4683 return -ENOMEM;
4684 path->reada = READA_BACK;
4685
4686 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4687 lock_extent_bits(&inode->io_tree, lock_start, (u64)-1,
4688 &cached_state);
4689
4690 /*
4691 * We want to drop from the next block forward in case this
4692 * new size is not block aligned since we will be keeping the
4693 * last block of the extent just the way it is.
4694 */
4695 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4696 fs_info->sectorsize),
4697 (u64)-1, 0);
4698 }
4699
4700 /*
4701 * This function is also used to drop the items in the log tree before
4702 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4703 * it is used to drop the logged items. So we shouldn't kill the delayed
4704 * items.
4705 */
4706 if (min_type == 0 && root == inode->root)
4707 btrfs_kill_delayed_inode_items(inode);
4708
4709 key.objectid = ino;
4710 key.offset = (u64)-1;
4711 key.type = (u8)-1;
4712
4713 search_again:
4714 /*
4715 * with a 16K leaf size and 128MB extents, you can actually queue
4716 * up a huge file in a single leaf. Most of the time that
4717 * bytes_deleted is > 0, it will be huge by the time we get here
4718 */
4719 if (be_nice && bytes_deleted > SZ_32M &&
4720 btrfs_should_end_transaction(trans)) {
4721 ret = -EAGAIN;
4722 goto out;
4723 }
4724
4725 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4726 if (ret < 0)
4727 goto out;
4728
4729 if (ret > 0) {
4730 ret = 0;
4731 /* there are no items in the tree for us to truncate, we're
4732 * done
4733 */
4734 if (path->slots[0] == 0)
4735 goto out;
4736 path->slots[0]--;
4737 }
4738
4739 while (1) {
4740 u64 clear_start = 0, clear_len = 0;
4741
4742 fi = NULL;
4743 leaf = path->nodes[0];
4744 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4745 found_type = found_key.type;
4746
4747 if (found_key.objectid != ino)
4748 break;
4749
4750 if (found_type < min_type)
4751 break;
4752
4753 item_end = found_key.offset;
4754 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4755 fi = btrfs_item_ptr(leaf, path->slots[0],
4756 struct btrfs_file_extent_item);
4757 extent_type = btrfs_file_extent_type(leaf, fi);
4758 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4759 item_end +=
4760 btrfs_file_extent_num_bytes(leaf, fi);
4761
4762 trace_btrfs_truncate_show_fi_regular(
4763 inode, leaf, fi, found_key.offset);
4764 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4765 item_end += btrfs_file_extent_ram_bytes(leaf,
4766 fi);
4767
4768 trace_btrfs_truncate_show_fi_inline(
4769 inode, leaf, fi, path->slots[0],
4770 found_key.offset);
4771 }
4772 item_end--;
4773 }
4774 if (found_type > min_type) {
4775 del_item = 1;
4776 } else {
4777 if (item_end < new_size)
4778 break;
4779 if (found_key.offset >= new_size)
4780 del_item = 1;
4781 else
4782 del_item = 0;
4783 }
4784 found_extent = 0;
4785 /* FIXME, shrink the extent if the ref count is only 1 */
4786 if (found_type != BTRFS_EXTENT_DATA_KEY)
4787 goto delete;
4788
4789 if (extents_found != NULL)
4790 (*extents_found)++;
4791
4792 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4793 u64 num_dec;
4794
4795 clear_start = found_key.offset;
4796 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4797 if (!del_item) {
4798 u64 orig_num_bytes =
4799 btrfs_file_extent_num_bytes(leaf, fi);
4800 extent_num_bytes = ALIGN(new_size -
4801 found_key.offset,
4802 fs_info->sectorsize);
4803 clear_start = ALIGN(new_size, fs_info->sectorsize);
4804 btrfs_set_file_extent_num_bytes(leaf, fi,
4805 extent_num_bytes);
4806 num_dec = (orig_num_bytes -
4807 extent_num_bytes);
4808 if (test_bit(BTRFS_ROOT_SHAREABLE,
4809 &root->state) &&
4810 extent_start != 0)
4811 inode_sub_bytes(&inode->vfs_inode,
4812 num_dec);
4813 btrfs_mark_buffer_dirty(leaf);
4814 } else {
4815 extent_num_bytes =
4816 btrfs_file_extent_disk_num_bytes(leaf,
4817 fi);
4818 extent_offset = found_key.offset -
4819 btrfs_file_extent_offset(leaf, fi);
4820
4821 /* FIXME blocksize != 4096 */
4822 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4823 if (extent_start != 0) {
4824 found_extent = 1;
4825 if (test_bit(BTRFS_ROOT_SHAREABLE,
4826 &root->state))
4827 inode_sub_bytes(&inode->vfs_inode,
4828 num_dec);
4829 }
4830 }
4831 clear_len = num_dec;
4832 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4833 /*
4834 * we can't truncate inline items that have had
4835 * special encodings
4836 */
4837 if (!del_item &&
4838 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4839 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4840 btrfs_file_extent_compression(leaf, fi) == 0) {
4841 u32 size = (u32)(new_size - found_key.offset);
4842
4843 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4844 size = btrfs_file_extent_calc_inline_size(size);
4845 btrfs_truncate_item(path, size, 1);
4846 } else if (!del_item) {
4847 /*
4848 * We have to bail so the last_size is set to
4849 * just before this extent.
4850 */
4851 ret = NEED_TRUNCATE_BLOCK;
4852 break;
4853 } else {
4854 /*
4855 * Inline extents are special, we just treat
4856 * them as a full sector worth in the file
4857 * extent tree just for simplicity sake.
4858 */
4859 clear_len = fs_info->sectorsize;
4860 }
4861
4862 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4863 inode_sub_bytes(&inode->vfs_inode,
4864 item_end + 1 - new_size);
4865 }
4866 delete:
4867 /*
4868 * We use btrfs_truncate_inode_items() to clean up log trees for
4869 * multiple fsyncs, and in this case we don't want to clear the
4870 * file extent range because it's just the log.
4871 */
4872 if (root == inode->root) {
4873 ret = btrfs_inode_clear_file_extent_range(inode,
4874 clear_start, clear_len);
4875 if (ret) {
4876 btrfs_abort_transaction(trans, ret);
4877 break;
4878 }
4879 }
4880
4881 if (del_item)
4882 last_size = found_key.offset;
4883 else
4884 last_size = new_size;
4885 if (del_item) {
4886 if (!pending_del_nr) {
4887 /* no pending yet, add ourselves */
4888 pending_del_slot = path->slots[0];
4889 pending_del_nr = 1;
4890 } else if (pending_del_nr &&
4891 path->slots[0] + 1 == pending_del_slot) {
4892 /* hop on the pending chunk */
4893 pending_del_nr++;
4894 pending_del_slot = path->slots[0];
4895 } else {
4896 BUG();
4897 }
4898 } else {
4899 break;
4900 }
4901 should_throttle = false;
4902
4903 if (found_extent &&
4904 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4905 struct btrfs_ref ref = { 0 };
4906
4907 bytes_deleted += extent_num_bytes;
4908
4909 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
4910 extent_start, extent_num_bytes, 0);
4911 ref.real_root = root->root_key.objectid;
4912 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
4913 ino, extent_offset);
4914 ret = btrfs_free_extent(trans, &ref);
4915 if (ret) {
4916 btrfs_abort_transaction(trans, ret);
4917 break;
4918 }
4919 if (be_nice) {
4920 if (btrfs_should_throttle_delayed_refs(trans))
4921 should_throttle = true;
4922 }
4923 }
4924
4925 if (found_type == BTRFS_INODE_ITEM_KEY)
4926 break;
4927
4928 if (path->slots[0] == 0 ||
4929 path->slots[0] != pending_del_slot ||
4930 should_throttle) {
4931 if (pending_del_nr) {
4932 ret = btrfs_del_items(trans, root, path,
4933 pending_del_slot,
4934 pending_del_nr);
4935 if (ret) {
4936 btrfs_abort_transaction(trans, ret);
4937 break;
4938 }
4939 pending_del_nr = 0;
4940 }
4941 btrfs_release_path(path);
4942
4943 /*
4944 * We can generate a lot of delayed refs, so we need to
4945 * throttle every once and a while and make sure we're
4946 * adding enough space to keep up with the work we are
4947 * generating. Since we hold a transaction here we
4948 * can't flush, and we don't want to FLUSH_LIMIT because
4949 * we could have generated too many delayed refs to
4950 * actually allocate, so just bail if we're short and
4951 * let the normal reservation dance happen higher up.
4952 */
4953 if (should_throttle) {
4954 ret = btrfs_delayed_refs_rsv_refill(fs_info,
4955 BTRFS_RESERVE_NO_FLUSH);
4956 if (ret) {
4957 ret = -EAGAIN;
4958 break;
4959 }
4960 }
4961 goto search_again;
4962 } else {
4963 path->slots[0]--;
4964 }
4965 }
4966 out:
4967 if (ret >= 0 && pending_del_nr) {
4968 int err;
4969
4970 err = btrfs_del_items(trans, root, path, pending_del_slot,
4971 pending_del_nr);
4972 if (err) {
4973 btrfs_abort_transaction(trans, err);
4974 ret = err;
4975 }
4976 }
4977 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4978 ASSERT(last_size >= new_size);
4979 if (!ret && last_size > new_size)
4980 last_size = new_size;
4981 btrfs_inode_safe_disk_i_size_write(inode, last_size);
4982 unlock_extent_cached(&inode->io_tree, lock_start, (u64)-1,
4983 &cached_state);
4984 }
4985
4986 btrfs_free_path(path);
4987 return ret;
4988 }
4989
4990 /*
4991 * btrfs_truncate_block - read, zero a chunk and write a block
4992 * @inode - inode that we're zeroing
4993 * @from - the offset to start zeroing
4994 * @len - the length to zero, 0 to zero the entire range respective to the
4995 * offset
4996 * @front - zero up to the offset instead of from the offset on
4997 *
4998 * This will find the block for the "from" offset and cow the block and zero the
4999 * part we want to zero. This is used with truncate and hole punching.
5000 */
btrfs_truncate_block(struct btrfs_inode * inode,loff_t from,loff_t len,int front)5001 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
5002 int front)
5003 {
5004 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5005 struct address_space *mapping = inode->vfs_inode.i_mapping;
5006 struct extent_io_tree *io_tree = &inode->io_tree;
5007 struct btrfs_ordered_extent *ordered;
5008 struct extent_state *cached_state = NULL;
5009 struct extent_changeset *data_reserved = NULL;
5010 bool only_release_metadata = false;
5011 u32 blocksize = fs_info->sectorsize;
5012 pgoff_t index = from >> PAGE_SHIFT;
5013 unsigned offset = from & (blocksize - 1);
5014 struct page *page;
5015 gfp_t mask = btrfs_alloc_write_mask(mapping);
5016 size_t write_bytes = blocksize;
5017 int ret = 0;
5018 u64 block_start;
5019 u64 block_end;
5020
5021 if (IS_ALIGNED(offset, blocksize) &&
5022 (!len || IS_ALIGNED(len, blocksize)))
5023 goto out;
5024
5025 block_start = round_down(from, blocksize);
5026 block_end = block_start + blocksize - 1;
5027
5028 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
5029 blocksize);
5030 if (ret < 0) {
5031 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
5032 /* For nocow case, no need to reserve data space */
5033 only_release_metadata = true;
5034 } else {
5035 goto out;
5036 }
5037 }
5038 ret = btrfs_delalloc_reserve_metadata(inode, blocksize);
5039 if (ret < 0) {
5040 if (!only_release_metadata)
5041 btrfs_free_reserved_data_space(inode, data_reserved,
5042 block_start, blocksize);
5043 goto out;
5044 }
5045 again:
5046 page = find_or_create_page(mapping, index, mask);
5047 if (!page) {
5048 btrfs_delalloc_release_space(inode, data_reserved, block_start,
5049 blocksize, true);
5050 btrfs_delalloc_release_extents(inode, blocksize);
5051 ret = -ENOMEM;
5052 goto out;
5053 }
5054 ret = set_page_extent_mapped(page);
5055 if (ret < 0)
5056 goto out_unlock;
5057
5058 if (!PageUptodate(page)) {
5059 ret = btrfs_readpage(NULL, page);
5060 lock_page(page);
5061 if (page->mapping != mapping) {
5062 unlock_page(page);
5063 put_page(page);
5064 goto again;
5065 }
5066 if (!PageUptodate(page)) {
5067 ret = -EIO;
5068 goto out_unlock;
5069 }
5070 }
5071 wait_on_page_writeback(page);
5072
5073 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
5074
5075 ordered = btrfs_lookup_ordered_extent(inode, block_start);
5076 if (ordered) {
5077 unlock_extent_cached(io_tree, block_start, block_end,
5078 &cached_state);
5079 unlock_page(page);
5080 put_page(page);
5081 btrfs_start_ordered_extent(ordered, 1);
5082 btrfs_put_ordered_extent(ordered);
5083 goto again;
5084 }
5085
5086 clear_extent_bit(&inode->io_tree, block_start, block_end,
5087 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
5088 0, 0, &cached_state);
5089
5090 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
5091 &cached_state);
5092 if (ret) {
5093 unlock_extent_cached(io_tree, block_start, block_end,
5094 &cached_state);
5095 goto out_unlock;
5096 }
5097
5098 if (offset != blocksize) {
5099 if (!len)
5100 len = blocksize - offset;
5101 if (front)
5102 memzero_page(page, (block_start - page_offset(page)),
5103 offset);
5104 else
5105 memzero_page(page, (block_start - page_offset(page)) + offset,
5106 len);
5107 flush_dcache_page(page);
5108 }
5109 ClearPageChecked(page);
5110 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
5111 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
5112
5113 if (only_release_metadata)
5114 set_extent_bit(&inode->io_tree, block_start, block_end,
5115 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
5116
5117 out_unlock:
5118 if (ret) {
5119 if (only_release_metadata)
5120 btrfs_delalloc_release_metadata(inode, blocksize, true);
5121 else
5122 btrfs_delalloc_release_space(inode, data_reserved,
5123 block_start, blocksize, true);
5124 }
5125 btrfs_delalloc_release_extents(inode, blocksize);
5126 unlock_page(page);
5127 put_page(page);
5128 out:
5129 if (only_release_metadata)
5130 btrfs_check_nocow_unlock(inode);
5131 extent_changeset_free(data_reserved);
5132 return ret;
5133 }
5134
maybe_insert_hole(struct btrfs_root * root,struct btrfs_inode * inode,u64 offset,u64 len)5135 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
5136 u64 offset, u64 len)
5137 {
5138 struct btrfs_fs_info *fs_info = root->fs_info;
5139 struct btrfs_trans_handle *trans;
5140 struct btrfs_drop_extents_args drop_args = { 0 };
5141 int ret;
5142
5143 /*
5144 * If NO_HOLES is enabled, we don't need to do anything.
5145 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
5146 * or btrfs_update_inode() will be called, which guarantee that the next
5147 * fsync will know this inode was changed and needs to be logged.
5148 */
5149 if (btrfs_fs_incompat(fs_info, NO_HOLES))
5150 return 0;
5151
5152 /*
5153 * 1 - for the one we're dropping
5154 * 1 - for the one we're adding
5155 * 1 - for updating the inode.
5156 */
5157 trans = btrfs_start_transaction(root, 3);
5158 if (IS_ERR(trans))
5159 return PTR_ERR(trans);
5160
5161 drop_args.start = offset;
5162 drop_args.end = offset + len;
5163 drop_args.drop_cache = true;
5164
5165 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
5166 if (ret) {
5167 btrfs_abort_transaction(trans, ret);
5168 btrfs_end_transaction(trans);
5169 return ret;
5170 }
5171
5172 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
5173 offset, 0, 0, len, 0, len, 0, 0, 0);
5174 if (ret) {
5175 btrfs_abort_transaction(trans, ret);
5176 } else {
5177 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
5178 btrfs_update_inode(trans, root, inode);
5179 }
5180 btrfs_end_transaction(trans);
5181 return ret;
5182 }
5183
5184 /*
5185 * This function puts in dummy file extents for the area we're creating a hole
5186 * for. So if we are truncating this file to a larger size we need to insert
5187 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5188 * the range between oldsize and size
5189 */
btrfs_cont_expand(struct btrfs_inode * inode,loff_t oldsize,loff_t size)5190 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
5191 {
5192 struct btrfs_root *root = inode->root;
5193 struct btrfs_fs_info *fs_info = root->fs_info;
5194 struct extent_io_tree *io_tree = &inode->io_tree;
5195 struct extent_map *em = NULL;
5196 struct extent_state *cached_state = NULL;
5197 struct extent_map_tree *em_tree = &inode->extent_tree;
5198 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
5199 u64 block_end = ALIGN(size, fs_info->sectorsize);
5200 u64 last_byte;
5201 u64 cur_offset;
5202 u64 hole_size;
5203 int err = 0;
5204
5205 /*
5206 * If our size started in the middle of a block we need to zero out the
5207 * rest of the block before we expand the i_size, otherwise we could
5208 * expose stale data.
5209 */
5210 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5211 if (err)
5212 return err;
5213
5214 if (size <= hole_start)
5215 return 0;
5216
5217 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
5218 &cached_state);
5219 cur_offset = hole_start;
5220 while (1) {
5221 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5222 block_end - cur_offset);
5223 if (IS_ERR(em)) {
5224 err = PTR_ERR(em);
5225 em = NULL;
5226 break;
5227 }
5228 last_byte = min(extent_map_end(em), block_end);
5229 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5230 hole_size = last_byte - cur_offset;
5231
5232 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5233 struct extent_map *hole_em;
5234
5235 err = maybe_insert_hole(root, inode, cur_offset,
5236 hole_size);
5237 if (err)
5238 break;
5239
5240 err = btrfs_inode_set_file_extent_range(inode,
5241 cur_offset, hole_size);
5242 if (err)
5243 break;
5244
5245 btrfs_drop_extent_cache(inode, cur_offset,
5246 cur_offset + hole_size - 1, 0);
5247 hole_em = alloc_extent_map();
5248 if (!hole_em) {
5249 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5250 &inode->runtime_flags);
5251 goto next;
5252 }
5253 hole_em->start = cur_offset;
5254 hole_em->len = hole_size;
5255 hole_em->orig_start = cur_offset;
5256
5257 hole_em->block_start = EXTENT_MAP_HOLE;
5258 hole_em->block_len = 0;
5259 hole_em->orig_block_len = 0;
5260 hole_em->ram_bytes = hole_size;
5261 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5262 hole_em->generation = fs_info->generation;
5263
5264 while (1) {
5265 write_lock(&em_tree->lock);
5266 err = add_extent_mapping(em_tree, hole_em, 1);
5267 write_unlock(&em_tree->lock);
5268 if (err != -EEXIST)
5269 break;
5270 btrfs_drop_extent_cache(inode, cur_offset,
5271 cur_offset +
5272 hole_size - 1, 0);
5273 }
5274 free_extent_map(hole_em);
5275 } else {
5276 err = btrfs_inode_set_file_extent_range(inode,
5277 cur_offset, hole_size);
5278 if (err)
5279 break;
5280 }
5281 next:
5282 free_extent_map(em);
5283 em = NULL;
5284 cur_offset = last_byte;
5285 if (cur_offset >= block_end)
5286 break;
5287 }
5288 free_extent_map(em);
5289 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5290 return err;
5291 }
5292
btrfs_setsize(struct inode * inode,struct iattr * attr)5293 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5294 {
5295 struct btrfs_root *root = BTRFS_I(inode)->root;
5296 struct btrfs_trans_handle *trans;
5297 loff_t oldsize = i_size_read(inode);
5298 loff_t newsize = attr->ia_size;
5299 int mask = attr->ia_valid;
5300 int ret;
5301
5302 /*
5303 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5304 * special case where we need to update the times despite not having
5305 * these flags set. For all other operations the VFS set these flags
5306 * explicitly if it wants a timestamp update.
5307 */
5308 if (newsize != oldsize) {
5309 inode_inc_iversion(inode);
5310 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5311 inode->i_ctime = inode->i_mtime =
5312 current_time(inode);
5313 }
5314
5315 if (newsize > oldsize) {
5316 /*
5317 * Don't do an expanding truncate while snapshotting is ongoing.
5318 * This is to ensure the snapshot captures a fully consistent
5319 * state of this file - if the snapshot captures this expanding
5320 * truncation, it must capture all writes that happened before
5321 * this truncation.
5322 */
5323 btrfs_drew_write_lock(&root->snapshot_lock);
5324 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5325 if (ret) {
5326 btrfs_drew_write_unlock(&root->snapshot_lock);
5327 return ret;
5328 }
5329
5330 trans = btrfs_start_transaction(root, 1);
5331 if (IS_ERR(trans)) {
5332 btrfs_drew_write_unlock(&root->snapshot_lock);
5333 return PTR_ERR(trans);
5334 }
5335
5336 i_size_write(inode, newsize);
5337 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5338 pagecache_isize_extended(inode, oldsize, newsize);
5339 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5340 btrfs_drew_write_unlock(&root->snapshot_lock);
5341 btrfs_end_transaction(trans);
5342 } else {
5343 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5344
5345 if (btrfs_is_zoned(fs_info)) {
5346 ret = btrfs_wait_ordered_range(inode,
5347 ALIGN(newsize, fs_info->sectorsize),
5348 (u64)-1);
5349 if (ret)
5350 return ret;
5351 }
5352
5353 /*
5354 * We're truncating a file that used to have good data down to
5355 * zero. Make sure any new writes to the file get on disk
5356 * on close.
5357 */
5358 if (newsize == 0)
5359 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5360 &BTRFS_I(inode)->runtime_flags);
5361
5362 truncate_setsize(inode, newsize);
5363
5364 inode_dio_wait(inode);
5365
5366 ret = btrfs_truncate(inode, newsize == oldsize);
5367 if (ret && inode->i_nlink) {
5368 int err;
5369
5370 /*
5371 * Truncate failed, so fix up the in-memory size. We
5372 * adjusted disk_i_size down as we removed extents, so
5373 * wait for disk_i_size to be stable and then update the
5374 * in-memory size to match.
5375 */
5376 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5377 if (err)
5378 return err;
5379 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5380 }
5381 }
5382
5383 return ret;
5384 }
5385
btrfs_setattr(struct user_namespace * mnt_userns,struct dentry * dentry,struct iattr * attr)5386 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5387 struct iattr *attr)
5388 {
5389 struct inode *inode = d_inode(dentry);
5390 struct btrfs_root *root = BTRFS_I(inode)->root;
5391 int err;
5392
5393 if (btrfs_root_readonly(root))
5394 return -EROFS;
5395
5396 err = setattr_prepare(mnt_userns, dentry, attr);
5397 if (err)
5398 return err;
5399
5400 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5401 err = btrfs_setsize(inode, attr);
5402 if (err)
5403 return err;
5404 }
5405
5406 if (attr->ia_valid) {
5407 setattr_copy(mnt_userns, inode, attr);
5408 inode_inc_iversion(inode);
5409 err = btrfs_dirty_inode(inode);
5410
5411 if (!err && attr->ia_valid & ATTR_MODE)
5412 err = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5413 }
5414
5415 return err;
5416 }
5417
5418 /*
5419 * While truncating the inode pages during eviction, we get the VFS calling
5420 * btrfs_invalidatepage() against each page of the inode. This is slow because
5421 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5422 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5423 * extent_state structures over and over, wasting lots of time.
5424 *
5425 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5426 * those expensive operations on a per page basis and do only the ordered io
5427 * finishing, while we release here the extent_map and extent_state structures,
5428 * without the excessive merging and splitting.
5429 */
evict_inode_truncate_pages(struct inode * inode)5430 static void evict_inode_truncate_pages(struct inode *inode)
5431 {
5432 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5433 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5434 struct rb_node *node;
5435
5436 ASSERT(inode->i_state & I_FREEING);
5437 truncate_inode_pages_final(&inode->i_data);
5438
5439 write_lock(&map_tree->lock);
5440 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5441 struct extent_map *em;
5442
5443 node = rb_first_cached(&map_tree->map);
5444 em = rb_entry(node, struct extent_map, rb_node);
5445 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5446 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5447 remove_extent_mapping(map_tree, em);
5448 free_extent_map(em);
5449 if (need_resched()) {
5450 write_unlock(&map_tree->lock);
5451 cond_resched();
5452 write_lock(&map_tree->lock);
5453 }
5454 }
5455 write_unlock(&map_tree->lock);
5456
5457 /*
5458 * Keep looping until we have no more ranges in the io tree.
5459 * We can have ongoing bios started by readahead that have
5460 * their endio callback (extent_io.c:end_bio_extent_readpage)
5461 * still in progress (unlocked the pages in the bio but did not yet
5462 * unlocked the ranges in the io tree). Therefore this means some
5463 * ranges can still be locked and eviction started because before
5464 * submitting those bios, which are executed by a separate task (work
5465 * queue kthread), inode references (inode->i_count) were not taken
5466 * (which would be dropped in the end io callback of each bio).
5467 * Therefore here we effectively end up waiting for those bios and
5468 * anyone else holding locked ranges without having bumped the inode's
5469 * reference count - if we don't do it, when they access the inode's
5470 * io_tree to unlock a range it may be too late, leading to an
5471 * use-after-free issue.
5472 */
5473 spin_lock(&io_tree->lock);
5474 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5475 struct extent_state *state;
5476 struct extent_state *cached_state = NULL;
5477 u64 start;
5478 u64 end;
5479 unsigned state_flags;
5480
5481 node = rb_first(&io_tree->state);
5482 state = rb_entry(node, struct extent_state, rb_node);
5483 start = state->start;
5484 end = state->end;
5485 state_flags = state->state;
5486 spin_unlock(&io_tree->lock);
5487
5488 lock_extent_bits(io_tree, start, end, &cached_state);
5489
5490 /*
5491 * If still has DELALLOC flag, the extent didn't reach disk,
5492 * and its reserved space won't be freed by delayed_ref.
5493 * So we need to free its reserved space here.
5494 * (Refer to comment in btrfs_invalidatepage, case 2)
5495 *
5496 * Note, end is the bytenr of last byte, so we need + 1 here.
5497 */
5498 if (state_flags & EXTENT_DELALLOC)
5499 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5500 end - start + 1);
5501
5502 clear_extent_bit(io_tree, start, end,
5503 EXTENT_LOCKED | EXTENT_DELALLOC |
5504 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5505 &cached_state);
5506
5507 cond_resched();
5508 spin_lock(&io_tree->lock);
5509 }
5510 spin_unlock(&io_tree->lock);
5511 }
5512
evict_refill_and_join(struct btrfs_root * root,struct btrfs_block_rsv * rsv)5513 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5514 struct btrfs_block_rsv *rsv)
5515 {
5516 struct btrfs_fs_info *fs_info = root->fs_info;
5517 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5518 struct btrfs_trans_handle *trans;
5519 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5520 int ret;
5521
5522 /*
5523 * Eviction should be taking place at some place safe because of our
5524 * delayed iputs. However the normal flushing code will run delayed
5525 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5526 *
5527 * We reserve the delayed_refs_extra here again because we can't use
5528 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5529 * above. We reserve our extra bit here because we generate a ton of
5530 * delayed refs activity by truncating.
5531 *
5532 * If we cannot make our reservation we'll attempt to steal from the
5533 * global reserve, because we really want to be able to free up space.
5534 */
5535 ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra,
5536 BTRFS_RESERVE_FLUSH_EVICT);
5537 if (ret) {
5538 /*
5539 * Try to steal from the global reserve if there is space for
5540 * it.
5541 */
5542 if (btrfs_check_space_for_delayed_refs(fs_info) ||
5543 btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) {
5544 btrfs_warn(fs_info,
5545 "could not allocate space for delete; will truncate on mount");
5546 return ERR_PTR(-ENOSPC);
5547 }
5548 delayed_refs_extra = 0;
5549 }
5550
5551 trans = btrfs_join_transaction(root);
5552 if (IS_ERR(trans))
5553 return trans;
5554
5555 if (delayed_refs_extra) {
5556 trans->block_rsv = &fs_info->trans_block_rsv;
5557 trans->bytes_reserved = delayed_refs_extra;
5558 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5559 delayed_refs_extra, 1);
5560 }
5561 return trans;
5562 }
5563
btrfs_evict_inode(struct inode * inode)5564 void btrfs_evict_inode(struct inode *inode)
5565 {
5566 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5567 struct btrfs_trans_handle *trans;
5568 struct btrfs_root *root = BTRFS_I(inode)->root;
5569 struct btrfs_block_rsv *rsv;
5570 int ret;
5571
5572 trace_btrfs_inode_evict(inode);
5573
5574 if (!root) {
5575 fsverity_cleanup_inode(inode);
5576 clear_inode(inode);
5577 return;
5578 }
5579
5580 evict_inode_truncate_pages(inode);
5581
5582 if (inode->i_nlink &&
5583 ((btrfs_root_refs(&root->root_item) != 0 &&
5584 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5585 btrfs_is_free_space_inode(BTRFS_I(inode))))
5586 goto no_delete;
5587
5588 if (is_bad_inode(inode))
5589 goto no_delete;
5590
5591 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5592
5593 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5594 goto no_delete;
5595
5596 if (inode->i_nlink > 0) {
5597 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5598 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5599 goto no_delete;
5600 }
5601
5602 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5603 if (ret)
5604 goto no_delete;
5605
5606 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5607 if (!rsv)
5608 goto no_delete;
5609 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5610 rsv->failfast = 1;
5611
5612 btrfs_i_size_write(BTRFS_I(inode), 0);
5613
5614 while (1) {
5615 trans = evict_refill_and_join(root, rsv);
5616 if (IS_ERR(trans))
5617 goto free_rsv;
5618
5619 trans->block_rsv = rsv;
5620
5621 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
5622 0, 0, NULL);
5623 trans->block_rsv = &fs_info->trans_block_rsv;
5624 btrfs_end_transaction(trans);
5625 btrfs_btree_balance_dirty(fs_info);
5626 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5627 goto free_rsv;
5628 else if (!ret)
5629 break;
5630 }
5631
5632 /*
5633 * Errors here aren't a big deal, it just means we leave orphan items in
5634 * the tree. They will be cleaned up on the next mount. If the inode
5635 * number gets reused, cleanup deletes the orphan item without doing
5636 * anything, and unlink reuses the existing orphan item.
5637 *
5638 * If it turns out that we are dropping too many of these, we might want
5639 * to add a mechanism for retrying these after a commit.
5640 */
5641 trans = evict_refill_and_join(root, rsv);
5642 if (!IS_ERR(trans)) {
5643 trans->block_rsv = rsv;
5644 btrfs_orphan_del(trans, BTRFS_I(inode));
5645 trans->block_rsv = &fs_info->trans_block_rsv;
5646 btrfs_end_transaction(trans);
5647 }
5648
5649 free_rsv:
5650 btrfs_free_block_rsv(fs_info, rsv);
5651 no_delete:
5652 /*
5653 * If we didn't successfully delete, the orphan item will still be in
5654 * the tree and we'll retry on the next mount. Again, we might also want
5655 * to retry these periodically in the future.
5656 */
5657 btrfs_remove_delayed_node(BTRFS_I(inode));
5658 fsverity_cleanup_inode(inode);
5659 clear_inode(inode);
5660 }
5661
5662 /*
5663 * Return the key found in the dir entry in the location pointer, fill @type
5664 * with BTRFS_FT_*, and return 0.
5665 *
5666 * If no dir entries were found, returns -ENOENT.
5667 * If found a corrupted location in dir entry, returns -EUCLEAN.
5668 */
btrfs_inode_by_name(struct inode * dir,struct dentry * dentry,struct btrfs_key * location,u8 * type)5669 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5670 struct btrfs_key *location, u8 *type)
5671 {
5672 const char *name = dentry->d_name.name;
5673 int namelen = dentry->d_name.len;
5674 struct btrfs_dir_item *di;
5675 struct btrfs_path *path;
5676 struct btrfs_root *root = BTRFS_I(dir)->root;
5677 int ret = 0;
5678
5679 path = btrfs_alloc_path();
5680 if (!path)
5681 return -ENOMEM;
5682
5683 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5684 name, namelen, 0);
5685 if (IS_ERR_OR_NULL(di)) {
5686 ret = di ? PTR_ERR(di) : -ENOENT;
5687 goto out;
5688 }
5689
5690 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5691 if (location->type != BTRFS_INODE_ITEM_KEY &&
5692 location->type != BTRFS_ROOT_ITEM_KEY) {
5693 ret = -EUCLEAN;
5694 btrfs_warn(root->fs_info,
5695 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5696 __func__, name, btrfs_ino(BTRFS_I(dir)),
5697 location->objectid, location->type, location->offset);
5698 }
5699 if (!ret)
5700 *type = btrfs_dir_type(path->nodes[0], di);
5701 out:
5702 btrfs_free_path(path);
5703 return ret;
5704 }
5705
5706 /*
5707 * when we hit a tree root in a directory, the btrfs part of the inode
5708 * needs to be changed to reflect the root directory of the tree root. This
5709 * is kind of like crossing a mount point.
5710 */
fixup_tree_root_location(struct btrfs_fs_info * fs_info,struct inode * dir,struct dentry * dentry,struct btrfs_key * location,struct btrfs_root ** sub_root)5711 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5712 struct inode *dir,
5713 struct dentry *dentry,
5714 struct btrfs_key *location,
5715 struct btrfs_root **sub_root)
5716 {
5717 struct btrfs_path *path;
5718 struct btrfs_root *new_root;
5719 struct btrfs_root_ref *ref;
5720 struct extent_buffer *leaf;
5721 struct btrfs_key key;
5722 int ret;
5723 int err = 0;
5724
5725 path = btrfs_alloc_path();
5726 if (!path) {
5727 err = -ENOMEM;
5728 goto out;
5729 }
5730
5731 err = -ENOENT;
5732 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5733 key.type = BTRFS_ROOT_REF_KEY;
5734 key.offset = location->objectid;
5735
5736 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5737 if (ret) {
5738 if (ret < 0)
5739 err = ret;
5740 goto out;
5741 }
5742
5743 leaf = path->nodes[0];
5744 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5745 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5746 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5747 goto out;
5748
5749 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5750 (unsigned long)(ref + 1),
5751 dentry->d_name.len);
5752 if (ret)
5753 goto out;
5754
5755 btrfs_release_path(path);
5756
5757 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5758 if (IS_ERR(new_root)) {
5759 err = PTR_ERR(new_root);
5760 goto out;
5761 }
5762
5763 *sub_root = new_root;
5764 location->objectid = btrfs_root_dirid(&new_root->root_item);
5765 location->type = BTRFS_INODE_ITEM_KEY;
5766 location->offset = 0;
5767 err = 0;
5768 out:
5769 btrfs_free_path(path);
5770 return err;
5771 }
5772
inode_tree_add(struct inode * inode)5773 static void inode_tree_add(struct inode *inode)
5774 {
5775 struct btrfs_root *root = BTRFS_I(inode)->root;
5776 struct btrfs_inode *entry;
5777 struct rb_node **p;
5778 struct rb_node *parent;
5779 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5780 u64 ino = btrfs_ino(BTRFS_I(inode));
5781
5782 if (inode_unhashed(inode))
5783 return;
5784 parent = NULL;
5785 spin_lock(&root->inode_lock);
5786 p = &root->inode_tree.rb_node;
5787 while (*p) {
5788 parent = *p;
5789 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5790
5791 if (ino < btrfs_ino(entry))
5792 p = &parent->rb_left;
5793 else if (ino > btrfs_ino(entry))
5794 p = &parent->rb_right;
5795 else {
5796 WARN_ON(!(entry->vfs_inode.i_state &
5797 (I_WILL_FREE | I_FREEING)));
5798 rb_replace_node(parent, new, &root->inode_tree);
5799 RB_CLEAR_NODE(parent);
5800 spin_unlock(&root->inode_lock);
5801 return;
5802 }
5803 }
5804 rb_link_node(new, parent, p);
5805 rb_insert_color(new, &root->inode_tree);
5806 spin_unlock(&root->inode_lock);
5807 }
5808
inode_tree_del(struct btrfs_inode * inode)5809 static void inode_tree_del(struct btrfs_inode *inode)
5810 {
5811 struct btrfs_root *root = inode->root;
5812 int empty = 0;
5813
5814 spin_lock(&root->inode_lock);
5815 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5816 rb_erase(&inode->rb_node, &root->inode_tree);
5817 RB_CLEAR_NODE(&inode->rb_node);
5818 empty = RB_EMPTY_ROOT(&root->inode_tree);
5819 }
5820 spin_unlock(&root->inode_lock);
5821
5822 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5823 spin_lock(&root->inode_lock);
5824 empty = RB_EMPTY_ROOT(&root->inode_tree);
5825 spin_unlock(&root->inode_lock);
5826 if (empty)
5827 btrfs_add_dead_root(root);
5828 }
5829 }
5830
5831
btrfs_init_locked_inode(struct inode * inode,void * p)5832 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5833 {
5834 struct btrfs_iget_args *args = p;
5835
5836 inode->i_ino = args->ino;
5837 BTRFS_I(inode)->location.objectid = args->ino;
5838 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5839 BTRFS_I(inode)->location.offset = 0;
5840 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5841 BUG_ON(args->root && !BTRFS_I(inode)->root);
5842 return 0;
5843 }
5844
btrfs_find_actor(struct inode * inode,void * opaque)5845 static int btrfs_find_actor(struct inode *inode, void *opaque)
5846 {
5847 struct btrfs_iget_args *args = opaque;
5848
5849 return args->ino == BTRFS_I(inode)->location.objectid &&
5850 args->root == BTRFS_I(inode)->root;
5851 }
5852
btrfs_iget_locked(struct super_block * s,u64 ino,struct btrfs_root * root)5853 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5854 struct btrfs_root *root)
5855 {
5856 struct inode *inode;
5857 struct btrfs_iget_args args;
5858 unsigned long hashval = btrfs_inode_hash(ino, root);
5859
5860 args.ino = ino;
5861 args.root = root;
5862
5863 inode = iget5_locked(s, hashval, btrfs_find_actor,
5864 btrfs_init_locked_inode,
5865 (void *)&args);
5866 return inode;
5867 }
5868
5869 /*
5870 * Get an inode object given its inode number and corresponding root.
5871 * Path can be preallocated to prevent recursing back to iget through
5872 * allocator. NULL is also valid but may require an additional allocation
5873 * later.
5874 */
btrfs_iget_path(struct super_block * s,u64 ino,struct btrfs_root * root,struct btrfs_path * path)5875 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5876 struct btrfs_root *root, struct btrfs_path *path)
5877 {
5878 struct inode *inode;
5879
5880 inode = btrfs_iget_locked(s, ino, root);
5881 if (!inode)
5882 return ERR_PTR(-ENOMEM);
5883
5884 if (inode->i_state & I_NEW) {
5885 int ret;
5886
5887 ret = btrfs_read_locked_inode(inode, path);
5888 if (!ret) {
5889 inode_tree_add(inode);
5890 unlock_new_inode(inode);
5891 } else {
5892 iget_failed(inode);
5893 /*
5894 * ret > 0 can come from btrfs_search_slot called by
5895 * btrfs_read_locked_inode, this means the inode item
5896 * was not found.
5897 */
5898 if (ret > 0)
5899 ret = -ENOENT;
5900 inode = ERR_PTR(ret);
5901 }
5902 }
5903
5904 return inode;
5905 }
5906
btrfs_iget(struct super_block * s,u64 ino,struct btrfs_root * root)5907 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5908 {
5909 return btrfs_iget_path(s, ino, root, NULL);
5910 }
5911
new_simple_dir(struct super_block * s,struct btrfs_key * key,struct btrfs_root * root)5912 static struct inode *new_simple_dir(struct super_block *s,
5913 struct btrfs_key *key,
5914 struct btrfs_root *root)
5915 {
5916 struct inode *inode = new_inode(s);
5917
5918 if (!inode)
5919 return ERR_PTR(-ENOMEM);
5920
5921 BTRFS_I(inode)->root = btrfs_grab_root(root);
5922 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5923 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5924
5925 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5926 /*
5927 * We only need lookup, the rest is read-only and there's no inode
5928 * associated with the dentry
5929 */
5930 inode->i_op = &simple_dir_inode_operations;
5931 inode->i_opflags &= ~IOP_XATTR;
5932 inode->i_fop = &simple_dir_operations;
5933 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5934 inode->i_mtime = current_time(inode);
5935 inode->i_atime = inode->i_mtime;
5936 inode->i_ctime = inode->i_mtime;
5937 BTRFS_I(inode)->i_otime = inode->i_mtime;
5938
5939 return inode;
5940 }
5941
btrfs_inode_type(struct inode * inode)5942 static inline u8 btrfs_inode_type(struct inode *inode)
5943 {
5944 /*
5945 * Compile-time asserts that generic FT_* types still match
5946 * BTRFS_FT_* types
5947 */
5948 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5949 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5950 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5951 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5952 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5953 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5954 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5955 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5956
5957 return fs_umode_to_ftype(inode->i_mode);
5958 }
5959
btrfs_lookup_dentry(struct inode * dir,struct dentry * dentry)5960 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5961 {
5962 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5963 struct inode *inode;
5964 struct btrfs_root *root = BTRFS_I(dir)->root;
5965 struct btrfs_root *sub_root = root;
5966 struct btrfs_key location;
5967 u8 di_type = 0;
5968 int ret = 0;
5969
5970 if (dentry->d_name.len > BTRFS_NAME_LEN)
5971 return ERR_PTR(-ENAMETOOLONG);
5972
5973 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5974 if (ret < 0)
5975 return ERR_PTR(ret);
5976
5977 if (location.type == BTRFS_INODE_ITEM_KEY) {
5978 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5979 if (IS_ERR(inode))
5980 return inode;
5981
5982 /* Do extra check against inode mode with di_type */
5983 if (btrfs_inode_type(inode) != di_type) {
5984 btrfs_crit(fs_info,
5985 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5986 inode->i_mode, btrfs_inode_type(inode),
5987 di_type);
5988 iput(inode);
5989 return ERR_PTR(-EUCLEAN);
5990 }
5991 return inode;
5992 }
5993
5994 ret = fixup_tree_root_location(fs_info, dir, dentry,
5995 &location, &sub_root);
5996 if (ret < 0) {
5997 if (ret != -ENOENT)
5998 inode = ERR_PTR(ret);
5999 else
6000 inode = new_simple_dir(dir->i_sb, &location, sub_root);
6001 } else {
6002 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
6003 }
6004 if (root != sub_root)
6005 btrfs_put_root(sub_root);
6006
6007 if (!IS_ERR(inode) && root != sub_root) {
6008 down_read(&fs_info->cleanup_work_sem);
6009 if (!sb_rdonly(inode->i_sb))
6010 ret = btrfs_orphan_cleanup(sub_root);
6011 up_read(&fs_info->cleanup_work_sem);
6012 if (ret) {
6013 iput(inode);
6014 inode = ERR_PTR(ret);
6015 }
6016 }
6017
6018 return inode;
6019 }
6020
btrfs_dentry_delete(const struct dentry * dentry)6021 static int btrfs_dentry_delete(const struct dentry *dentry)
6022 {
6023 struct btrfs_root *root;
6024 struct inode *inode = d_inode(dentry);
6025
6026 if (!inode && !IS_ROOT(dentry))
6027 inode = d_inode(dentry->d_parent);
6028
6029 if (inode) {
6030 root = BTRFS_I(inode)->root;
6031 if (btrfs_root_refs(&root->root_item) == 0)
6032 return 1;
6033
6034 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6035 return 1;
6036 }
6037 return 0;
6038 }
6039
btrfs_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)6040 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
6041 unsigned int flags)
6042 {
6043 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
6044
6045 if (inode == ERR_PTR(-ENOENT))
6046 inode = NULL;
6047 return d_splice_alias(inode, dentry);
6048 }
6049
6050 /*
6051 * All this infrastructure exists because dir_emit can fault, and we are holding
6052 * the tree lock when doing readdir. For now just allocate a buffer and copy
6053 * our information into that, and then dir_emit from the buffer. This is
6054 * similar to what NFS does, only we don't keep the buffer around in pagecache
6055 * because I'm afraid I'll mess that up. Long term we need to make filldir do
6056 * copy_to_user_inatomic so we don't have to worry about page faulting under the
6057 * tree lock.
6058 */
btrfs_opendir(struct inode * inode,struct file * file)6059 static int btrfs_opendir(struct inode *inode, struct file *file)
6060 {
6061 struct btrfs_file_private *private;
6062
6063 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
6064 if (!private)
6065 return -ENOMEM;
6066 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
6067 if (!private->filldir_buf) {
6068 kfree(private);
6069 return -ENOMEM;
6070 }
6071 file->private_data = private;
6072 return 0;
6073 }
6074
6075 struct dir_entry {
6076 u64 ino;
6077 u64 offset;
6078 unsigned type;
6079 int name_len;
6080 };
6081
btrfs_filldir(void * addr,int entries,struct dir_context * ctx)6082 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
6083 {
6084 while (entries--) {
6085 struct dir_entry *entry = addr;
6086 char *name = (char *)(entry + 1);
6087
6088 ctx->pos = get_unaligned(&entry->offset);
6089 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
6090 get_unaligned(&entry->ino),
6091 get_unaligned(&entry->type)))
6092 return 1;
6093 addr += sizeof(struct dir_entry) +
6094 get_unaligned(&entry->name_len);
6095 ctx->pos++;
6096 }
6097 return 0;
6098 }
6099
btrfs_real_readdir(struct file * file,struct dir_context * ctx)6100 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
6101 {
6102 struct inode *inode = file_inode(file);
6103 struct btrfs_root *root = BTRFS_I(inode)->root;
6104 struct btrfs_file_private *private = file->private_data;
6105 struct btrfs_dir_item *di;
6106 struct btrfs_key key;
6107 struct btrfs_key found_key;
6108 struct btrfs_path *path;
6109 void *addr;
6110 struct list_head ins_list;
6111 struct list_head del_list;
6112 int ret;
6113 struct extent_buffer *leaf;
6114 int slot;
6115 char *name_ptr;
6116 int name_len;
6117 int entries = 0;
6118 int total_len = 0;
6119 bool put = false;
6120 struct btrfs_key location;
6121
6122 if (!dir_emit_dots(file, ctx))
6123 return 0;
6124
6125 path = btrfs_alloc_path();
6126 if (!path)
6127 return -ENOMEM;
6128
6129 addr = private->filldir_buf;
6130 path->reada = READA_FORWARD;
6131
6132 INIT_LIST_HEAD(&ins_list);
6133 INIT_LIST_HEAD(&del_list);
6134 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
6135
6136 again:
6137 key.type = BTRFS_DIR_INDEX_KEY;
6138 key.offset = ctx->pos;
6139 key.objectid = btrfs_ino(BTRFS_I(inode));
6140
6141 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6142 if (ret < 0)
6143 goto err;
6144
6145 while (1) {
6146 struct dir_entry *entry;
6147
6148 leaf = path->nodes[0];
6149 slot = path->slots[0];
6150 if (slot >= btrfs_header_nritems(leaf)) {
6151 ret = btrfs_next_leaf(root, path);
6152 if (ret < 0)
6153 goto err;
6154 else if (ret > 0)
6155 break;
6156 continue;
6157 }
6158
6159 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6160
6161 if (found_key.objectid != key.objectid)
6162 break;
6163 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6164 break;
6165 if (found_key.offset < ctx->pos)
6166 goto next;
6167 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6168 goto next;
6169 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
6170 name_len = btrfs_dir_name_len(leaf, di);
6171 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6172 PAGE_SIZE) {
6173 btrfs_release_path(path);
6174 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6175 if (ret)
6176 goto nopos;
6177 addr = private->filldir_buf;
6178 entries = 0;
6179 total_len = 0;
6180 goto again;
6181 }
6182
6183 entry = addr;
6184 put_unaligned(name_len, &entry->name_len);
6185 name_ptr = (char *)(entry + 1);
6186 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6187 name_len);
6188 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
6189 &entry->type);
6190 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6191 put_unaligned(location.objectid, &entry->ino);
6192 put_unaligned(found_key.offset, &entry->offset);
6193 entries++;
6194 addr += sizeof(struct dir_entry) + name_len;
6195 total_len += sizeof(struct dir_entry) + name_len;
6196 next:
6197 path->slots[0]++;
6198 }
6199 btrfs_release_path(path);
6200
6201 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6202 if (ret)
6203 goto nopos;
6204
6205 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6206 if (ret)
6207 goto nopos;
6208
6209 /*
6210 * Stop new entries from being returned after we return the last
6211 * entry.
6212 *
6213 * New directory entries are assigned a strictly increasing
6214 * offset. This means that new entries created during readdir
6215 * are *guaranteed* to be seen in the future by that readdir.
6216 * This has broken buggy programs which operate on names as
6217 * they're returned by readdir. Until we re-use freed offsets
6218 * we have this hack to stop new entries from being returned
6219 * under the assumption that they'll never reach this huge
6220 * offset.
6221 *
6222 * This is being careful not to overflow 32bit loff_t unless the
6223 * last entry requires it because doing so has broken 32bit apps
6224 * in the past.
6225 */
6226 if (ctx->pos >= INT_MAX)
6227 ctx->pos = LLONG_MAX;
6228 else
6229 ctx->pos = INT_MAX;
6230 nopos:
6231 ret = 0;
6232 err:
6233 if (put)
6234 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6235 btrfs_free_path(path);
6236 return ret;
6237 }
6238
6239 /*
6240 * This is somewhat expensive, updating the tree every time the
6241 * inode changes. But, it is most likely to find the inode in cache.
6242 * FIXME, needs more benchmarking...there are no reasons other than performance
6243 * to keep or drop this code.
6244 */
btrfs_dirty_inode(struct inode * inode)6245 static int btrfs_dirty_inode(struct inode *inode)
6246 {
6247 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6248 struct btrfs_root *root = BTRFS_I(inode)->root;
6249 struct btrfs_trans_handle *trans;
6250 int ret;
6251
6252 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6253 return 0;
6254
6255 trans = btrfs_join_transaction(root);
6256 if (IS_ERR(trans))
6257 return PTR_ERR(trans);
6258
6259 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6260 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6261 /* whoops, lets try again with the full transaction */
6262 btrfs_end_transaction(trans);
6263 trans = btrfs_start_transaction(root, 1);
6264 if (IS_ERR(trans))
6265 return PTR_ERR(trans);
6266
6267 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6268 }
6269 btrfs_end_transaction(trans);
6270 if (BTRFS_I(inode)->delayed_node)
6271 btrfs_balance_delayed_items(fs_info);
6272
6273 return ret;
6274 }
6275
6276 /*
6277 * This is a copy of file_update_time. We need this so we can return error on
6278 * ENOSPC for updating the inode in the case of file write and mmap writes.
6279 */
btrfs_update_time(struct inode * inode,struct timespec64 * now,int flags)6280 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6281 int flags)
6282 {
6283 struct btrfs_root *root = BTRFS_I(inode)->root;
6284 bool dirty = flags & ~S_VERSION;
6285
6286 if (btrfs_root_readonly(root))
6287 return -EROFS;
6288
6289 if (flags & S_VERSION)
6290 dirty |= inode_maybe_inc_iversion(inode, dirty);
6291 if (flags & S_CTIME)
6292 inode->i_ctime = *now;
6293 if (flags & S_MTIME)
6294 inode->i_mtime = *now;
6295 if (flags & S_ATIME)
6296 inode->i_atime = *now;
6297 return dirty ? btrfs_dirty_inode(inode) : 0;
6298 }
6299
6300 /*
6301 * find the highest existing sequence number in a directory
6302 * and then set the in-memory index_cnt variable to reflect
6303 * free sequence numbers
6304 */
btrfs_set_inode_index_count(struct btrfs_inode * inode)6305 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6306 {
6307 struct btrfs_root *root = inode->root;
6308 struct btrfs_key key, found_key;
6309 struct btrfs_path *path;
6310 struct extent_buffer *leaf;
6311 int ret;
6312
6313 key.objectid = btrfs_ino(inode);
6314 key.type = BTRFS_DIR_INDEX_KEY;
6315 key.offset = (u64)-1;
6316
6317 path = btrfs_alloc_path();
6318 if (!path)
6319 return -ENOMEM;
6320
6321 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6322 if (ret < 0)
6323 goto out;
6324 /* FIXME: we should be able to handle this */
6325 if (ret == 0)
6326 goto out;
6327 ret = 0;
6328
6329 /*
6330 * MAGIC NUMBER EXPLANATION:
6331 * since we search a directory based on f_pos we have to start at 2
6332 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6333 * else has to start at 2
6334 */
6335 if (path->slots[0] == 0) {
6336 inode->index_cnt = 2;
6337 goto out;
6338 }
6339
6340 path->slots[0]--;
6341
6342 leaf = path->nodes[0];
6343 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6344
6345 if (found_key.objectid != btrfs_ino(inode) ||
6346 found_key.type != BTRFS_DIR_INDEX_KEY) {
6347 inode->index_cnt = 2;
6348 goto out;
6349 }
6350
6351 inode->index_cnt = found_key.offset + 1;
6352 out:
6353 btrfs_free_path(path);
6354 return ret;
6355 }
6356
6357 /*
6358 * helper to find a free sequence number in a given directory. This current
6359 * code is very simple, later versions will do smarter things in the btree
6360 */
btrfs_set_inode_index(struct btrfs_inode * dir,u64 * index)6361 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6362 {
6363 int ret = 0;
6364
6365 if (dir->index_cnt == (u64)-1) {
6366 ret = btrfs_inode_delayed_dir_index_count(dir);
6367 if (ret) {
6368 ret = btrfs_set_inode_index_count(dir);
6369 if (ret)
6370 return ret;
6371 }
6372 }
6373
6374 *index = dir->index_cnt;
6375 dir->index_cnt++;
6376
6377 return ret;
6378 }
6379
btrfs_insert_inode_locked(struct inode * inode)6380 static int btrfs_insert_inode_locked(struct inode *inode)
6381 {
6382 struct btrfs_iget_args args;
6383
6384 args.ino = BTRFS_I(inode)->location.objectid;
6385 args.root = BTRFS_I(inode)->root;
6386
6387 return insert_inode_locked4(inode,
6388 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6389 btrfs_find_actor, &args);
6390 }
6391
6392 /*
6393 * Inherit flags from the parent inode.
6394 *
6395 * Currently only the compression flags and the cow flags are inherited.
6396 */
btrfs_inherit_iflags(struct inode * inode,struct inode * dir)6397 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6398 {
6399 unsigned int flags;
6400
6401 if (!dir)
6402 return;
6403
6404 flags = BTRFS_I(dir)->flags;
6405
6406 if (flags & BTRFS_INODE_NOCOMPRESS) {
6407 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6408 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6409 } else if (flags & BTRFS_INODE_COMPRESS) {
6410 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6411 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6412 }
6413
6414 if (flags & BTRFS_INODE_NODATACOW) {
6415 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6416 if (S_ISREG(inode->i_mode))
6417 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6418 }
6419
6420 btrfs_sync_inode_flags_to_i_flags(inode);
6421 }
6422
btrfs_new_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct user_namespace * mnt_userns,struct inode * dir,const char * name,int name_len,u64 ref_objectid,u64 objectid,umode_t mode,u64 * index)6423 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6424 struct btrfs_root *root,
6425 struct user_namespace *mnt_userns,
6426 struct inode *dir,
6427 const char *name, int name_len,
6428 u64 ref_objectid, u64 objectid,
6429 umode_t mode, u64 *index)
6430 {
6431 struct btrfs_fs_info *fs_info = root->fs_info;
6432 struct inode *inode;
6433 struct btrfs_inode_item *inode_item;
6434 struct btrfs_key *location;
6435 struct btrfs_path *path;
6436 struct btrfs_inode_ref *ref;
6437 struct btrfs_key key[2];
6438 u32 sizes[2];
6439 int nitems = name ? 2 : 1;
6440 unsigned long ptr;
6441 unsigned int nofs_flag;
6442 int ret;
6443
6444 path = btrfs_alloc_path();
6445 if (!path)
6446 return ERR_PTR(-ENOMEM);
6447
6448 nofs_flag = memalloc_nofs_save();
6449 inode = new_inode(fs_info->sb);
6450 memalloc_nofs_restore(nofs_flag);
6451 if (!inode) {
6452 btrfs_free_path(path);
6453 return ERR_PTR(-ENOMEM);
6454 }
6455
6456 /*
6457 * O_TMPFILE, set link count to 0, so that after this point,
6458 * we fill in an inode item with the correct link count.
6459 */
6460 if (!name)
6461 set_nlink(inode, 0);
6462
6463 /*
6464 * we have to initialize this early, so we can reclaim the inode
6465 * number if we fail afterwards in this function.
6466 */
6467 inode->i_ino = objectid;
6468
6469 if (dir && name) {
6470 trace_btrfs_inode_request(dir);
6471
6472 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6473 if (ret) {
6474 btrfs_free_path(path);
6475 iput(inode);
6476 return ERR_PTR(ret);
6477 }
6478 } else if (dir) {
6479 *index = 0;
6480 }
6481 /*
6482 * index_cnt is ignored for everything but a dir,
6483 * btrfs_set_inode_index_count has an explanation for the magic
6484 * number
6485 */
6486 BTRFS_I(inode)->index_cnt = 2;
6487 BTRFS_I(inode)->dir_index = *index;
6488 BTRFS_I(inode)->root = btrfs_grab_root(root);
6489 BTRFS_I(inode)->generation = trans->transid;
6490 inode->i_generation = BTRFS_I(inode)->generation;
6491
6492 /*
6493 * We could have gotten an inode number from somebody who was fsynced
6494 * and then removed in this same transaction, so let's just set full
6495 * sync since it will be a full sync anyway and this will blow away the
6496 * old info in the log.
6497 */
6498 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6499
6500 key[0].objectid = objectid;
6501 key[0].type = BTRFS_INODE_ITEM_KEY;
6502 key[0].offset = 0;
6503
6504 sizes[0] = sizeof(struct btrfs_inode_item);
6505
6506 if (name) {
6507 /*
6508 * Start new inodes with an inode_ref. This is slightly more
6509 * efficient for small numbers of hard links since they will
6510 * be packed into one item. Extended refs will kick in if we
6511 * add more hard links than can fit in the ref item.
6512 */
6513 key[1].objectid = objectid;
6514 key[1].type = BTRFS_INODE_REF_KEY;
6515 key[1].offset = ref_objectid;
6516
6517 sizes[1] = name_len + sizeof(*ref);
6518 }
6519
6520 location = &BTRFS_I(inode)->location;
6521 location->objectid = objectid;
6522 location->offset = 0;
6523 location->type = BTRFS_INODE_ITEM_KEY;
6524
6525 ret = btrfs_insert_inode_locked(inode);
6526 if (ret < 0) {
6527 iput(inode);
6528 goto fail;
6529 }
6530
6531 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6532 if (ret != 0)
6533 goto fail_unlock;
6534
6535 inode_init_owner(mnt_userns, inode, dir, mode);
6536 inode_set_bytes(inode, 0);
6537
6538 inode->i_mtime = current_time(inode);
6539 inode->i_atime = inode->i_mtime;
6540 inode->i_ctime = inode->i_mtime;
6541 BTRFS_I(inode)->i_otime = inode->i_mtime;
6542
6543 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6544 struct btrfs_inode_item);
6545 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6546 sizeof(*inode_item));
6547 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6548
6549 if (name) {
6550 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6551 struct btrfs_inode_ref);
6552 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6553 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6554 ptr = (unsigned long)(ref + 1);
6555 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6556 }
6557
6558 btrfs_mark_buffer_dirty(path->nodes[0]);
6559 btrfs_free_path(path);
6560
6561 btrfs_inherit_iflags(inode, dir);
6562
6563 if (S_ISREG(mode)) {
6564 if (btrfs_test_opt(fs_info, NODATASUM))
6565 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6566 if (btrfs_test_opt(fs_info, NODATACOW))
6567 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6568 BTRFS_INODE_NODATASUM;
6569 }
6570
6571 inode_tree_add(inode);
6572
6573 trace_btrfs_inode_new(inode);
6574 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6575
6576 btrfs_update_root_times(trans, root);
6577
6578 ret = btrfs_inode_inherit_props(trans, inode, dir);
6579 if (ret)
6580 btrfs_err(fs_info,
6581 "error inheriting props for ino %llu (root %llu): %d",
6582 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6583
6584 return inode;
6585
6586 fail_unlock:
6587 discard_new_inode(inode);
6588 fail:
6589 if (dir && name)
6590 BTRFS_I(dir)->index_cnt--;
6591 btrfs_free_path(path);
6592 return ERR_PTR(ret);
6593 }
6594
6595 /*
6596 * utility function to add 'inode' into 'parent_inode' with
6597 * a give name and a given sequence number.
6598 * if 'add_backref' is true, also insert a backref from the
6599 * inode to the parent directory.
6600 */
btrfs_add_link(struct btrfs_trans_handle * trans,struct btrfs_inode * parent_inode,struct btrfs_inode * inode,const char * name,int name_len,int add_backref,u64 index)6601 int btrfs_add_link(struct btrfs_trans_handle *trans,
6602 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6603 const char *name, int name_len, int add_backref, u64 index)
6604 {
6605 int ret = 0;
6606 struct btrfs_key key;
6607 struct btrfs_root *root = parent_inode->root;
6608 u64 ino = btrfs_ino(inode);
6609 u64 parent_ino = btrfs_ino(parent_inode);
6610
6611 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6612 memcpy(&key, &inode->root->root_key, sizeof(key));
6613 } else {
6614 key.objectid = ino;
6615 key.type = BTRFS_INODE_ITEM_KEY;
6616 key.offset = 0;
6617 }
6618
6619 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6620 ret = btrfs_add_root_ref(trans, key.objectid,
6621 root->root_key.objectid, parent_ino,
6622 index, name, name_len);
6623 } else if (add_backref) {
6624 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6625 parent_ino, index);
6626 }
6627
6628 /* Nothing to clean up yet */
6629 if (ret)
6630 return ret;
6631
6632 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6633 btrfs_inode_type(&inode->vfs_inode), index);
6634 if (ret == -EEXIST || ret == -EOVERFLOW)
6635 goto fail_dir_item;
6636 else if (ret) {
6637 btrfs_abort_transaction(trans, ret);
6638 return ret;
6639 }
6640
6641 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6642 name_len * 2);
6643 inode_inc_iversion(&parent_inode->vfs_inode);
6644 /*
6645 * If we are replaying a log tree, we do not want to update the mtime
6646 * and ctime of the parent directory with the current time, since the
6647 * log replay procedure is responsible for setting them to their correct
6648 * values (the ones it had when the fsync was done).
6649 */
6650 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6651 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6652
6653 parent_inode->vfs_inode.i_mtime = now;
6654 parent_inode->vfs_inode.i_ctime = now;
6655 }
6656 ret = btrfs_update_inode(trans, root, parent_inode);
6657 if (ret)
6658 btrfs_abort_transaction(trans, ret);
6659 return ret;
6660
6661 fail_dir_item:
6662 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6663 u64 local_index;
6664 int err;
6665 err = btrfs_del_root_ref(trans, key.objectid,
6666 root->root_key.objectid, parent_ino,
6667 &local_index, name, name_len);
6668 if (err)
6669 btrfs_abort_transaction(trans, err);
6670 } else if (add_backref) {
6671 u64 local_index;
6672 int err;
6673
6674 err = btrfs_del_inode_ref(trans, root, name, name_len,
6675 ino, parent_ino, &local_index);
6676 if (err)
6677 btrfs_abort_transaction(trans, err);
6678 }
6679
6680 /* Return the original error code */
6681 return ret;
6682 }
6683
btrfs_add_nondir(struct btrfs_trans_handle * trans,struct btrfs_inode * dir,struct dentry * dentry,struct btrfs_inode * inode,int backref,u64 index)6684 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6685 struct btrfs_inode *dir, struct dentry *dentry,
6686 struct btrfs_inode *inode, int backref, u64 index)
6687 {
6688 int err = btrfs_add_link(trans, dir, inode,
6689 dentry->d_name.name, dentry->d_name.len,
6690 backref, index);
6691 if (err > 0)
6692 err = -EEXIST;
6693 return err;
6694 }
6695
btrfs_mknod(struct user_namespace * mnt_userns,struct inode * dir,struct dentry * dentry,umode_t mode,dev_t rdev)6696 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6697 struct dentry *dentry, umode_t mode, dev_t rdev)
6698 {
6699 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6700 struct btrfs_trans_handle *trans;
6701 struct btrfs_root *root = BTRFS_I(dir)->root;
6702 struct inode *inode = NULL;
6703 int err;
6704 u64 objectid;
6705 u64 index = 0;
6706
6707 /*
6708 * 2 for inode item and ref
6709 * 2 for dir items
6710 * 1 for xattr if selinux is on
6711 */
6712 trans = btrfs_start_transaction(root, 5);
6713 if (IS_ERR(trans))
6714 return PTR_ERR(trans);
6715
6716 err = btrfs_get_free_objectid(root, &objectid);
6717 if (err)
6718 goto out_unlock;
6719
6720 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
6721 dentry->d_name.name, dentry->d_name.len,
6722 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
6723 if (IS_ERR(inode)) {
6724 err = PTR_ERR(inode);
6725 inode = NULL;
6726 goto out_unlock;
6727 }
6728
6729 /*
6730 * If the active LSM wants to access the inode during
6731 * d_instantiate it needs these. Smack checks to see
6732 * if the filesystem supports xattrs by looking at the
6733 * ops vector.
6734 */
6735 inode->i_op = &btrfs_special_inode_operations;
6736 init_special_inode(inode, inode->i_mode, rdev);
6737
6738 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6739 if (err)
6740 goto out_unlock;
6741
6742 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6743 0, index);
6744 if (err)
6745 goto out_unlock;
6746
6747 btrfs_update_inode(trans, root, BTRFS_I(inode));
6748 d_instantiate_new(dentry, inode);
6749
6750 out_unlock:
6751 btrfs_end_transaction(trans);
6752 btrfs_btree_balance_dirty(fs_info);
6753 if (err && inode) {
6754 inode_dec_link_count(inode);
6755 discard_new_inode(inode);
6756 }
6757 return err;
6758 }
6759
btrfs_create(struct user_namespace * mnt_userns,struct inode * dir,struct dentry * dentry,umode_t mode,bool excl)6760 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6761 struct dentry *dentry, umode_t mode, bool excl)
6762 {
6763 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6764 struct btrfs_trans_handle *trans;
6765 struct btrfs_root *root = BTRFS_I(dir)->root;
6766 struct inode *inode = NULL;
6767 int err;
6768 u64 objectid;
6769 u64 index = 0;
6770
6771 /*
6772 * 2 for inode item and ref
6773 * 2 for dir items
6774 * 1 for xattr if selinux is on
6775 */
6776 trans = btrfs_start_transaction(root, 5);
6777 if (IS_ERR(trans))
6778 return PTR_ERR(trans);
6779
6780 err = btrfs_get_free_objectid(root, &objectid);
6781 if (err)
6782 goto out_unlock;
6783
6784 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
6785 dentry->d_name.name, dentry->d_name.len,
6786 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
6787 if (IS_ERR(inode)) {
6788 err = PTR_ERR(inode);
6789 inode = NULL;
6790 goto out_unlock;
6791 }
6792 /*
6793 * If the active LSM wants to access the inode during
6794 * d_instantiate it needs these. Smack checks to see
6795 * if the filesystem supports xattrs by looking at the
6796 * ops vector.
6797 */
6798 inode->i_fop = &btrfs_file_operations;
6799 inode->i_op = &btrfs_file_inode_operations;
6800 inode->i_mapping->a_ops = &btrfs_aops;
6801
6802 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6803 if (err)
6804 goto out_unlock;
6805
6806 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6807 if (err)
6808 goto out_unlock;
6809
6810 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6811 0, index);
6812 if (err)
6813 goto out_unlock;
6814
6815 d_instantiate_new(dentry, inode);
6816
6817 out_unlock:
6818 btrfs_end_transaction(trans);
6819 if (err && inode) {
6820 inode_dec_link_count(inode);
6821 discard_new_inode(inode);
6822 }
6823 btrfs_btree_balance_dirty(fs_info);
6824 return err;
6825 }
6826
btrfs_link(struct dentry * old_dentry,struct inode * dir,struct dentry * dentry)6827 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6828 struct dentry *dentry)
6829 {
6830 struct btrfs_trans_handle *trans = NULL;
6831 struct btrfs_root *root = BTRFS_I(dir)->root;
6832 struct inode *inode = d_inode(old_dentry);
6833 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6834 u64 index;
6835 int err;
6836 int drop_inode = 0;
6837
6838 /* do not allow sys_link's with other subvols of the same device */
6839 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6840 return -EXDEV;
6841
6842 if (inode->i_nlink >= BTRFS_LINK_MAX)
6843 return -EMLINK;
6844
6845 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6846 if (err)
6847 goto fail;
6848
6849 /*
6850 * 2 items for inode and inode ref
6851 * 2 items for dir items
6852 * 1 item for parent inode
6853 * 1 item for orphan item deletion if O_TMPFILE
6854 */
6855 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6856 if (IS_ERR(trans)) {
6857 err = PTR_ERR(trans);
6858 trans = NULL;
6859 goto fail;
6860 }
6861
6862 /* There are several dir indexes for this inode, clear the cache. */
6863 BTRFS_I(inode)->dir_index = 0ULL;
6864 inc_nlink(inode);
6865 inode_inc_iversion(inode);
6866 inode->i_ctime = current_time(inode);
6867 ihold(inode);
6868 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6869
6870 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6871 1, index);
6872
6873 if (err) {
6874 drop_inode = 1;
6875 } else {
6876 struct dentry *parent = dentry->d_parent;
6877
6878 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6879 if (err)
6880 goto fail;
6881 if (inode->i_nlink == 1) {
6882 /*
6883 * If new hard link count is 1, it's a file created
6884 * with open(2) O_TMPFILE flag.
6885 */
6886 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6887 if (err)
6888 goto fail;
6889 }
6890 d_instantiate(dentry, inode);
6891 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6892 }
6893
6894 fail:
6895 if (trans)
6896 btrfs_end_transaction(trans);
6897 if (drop_inode) {
6898 inode_dec_link_count(inode);
6899 iput(inode);
6900 }
6901 btrfs_btree_balance_dirty(fs_info);
6902 return err;
6903 }
6904
btrfs_mkdir(struct user_namespace * mnt_userns,struct inode * dir,struct dentry * dentry,umode_t mode)6905 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6906 struct dentry *dentry, umode_t mode)
6907 {
6908 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6909 struct inode *inode = NULL;
6910 struct btrfs_trans_handle *trans;
6911 struct btrfs_root *root = BTRFS_I(dir)->root;
6912 int err = 0;
6913 u64 objectid = 0;
6914 u64 index = 0;
6915
6916 /*
6917 * 2 items for inode and ref
6918 * 2 items for dir items
6919 * 1 for xattr if selinux is on
6920 */
6921 trans = btrfs_start_transaction(root, 5);
6922 if (IS_ERR(trans))
6923 return PTR_ERR(trans);
6924
6925 err = btrfs_get_free_objectid(root, &objectid);
6926 if (err)
6927 goto out_fail;
6928
6929 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
6930 dentry->d_name.name, dentry->d_name.len,
6931 btrfs_ino(BTRFS_I(dir)), objectid,
6932 S_IFDIR | mode, &index);
6933 if (IS_ERR(inode)) {
6934 err = PTR_ERR(inode);
6935 inode = NULL;
6936 goto out_fail;
6937 }
6938
6939 /* these must be set before we unlock the inode */
6940 inode->i_op = &btrfs_dir_inode_operations;
6941 inode->i_fop = &btrfs_dir_file_operations;
6942
6943 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6944 if (err)
6945 goto out_fail;
6946
6947 btrfs_i_size_write(BTRFS_I(inode), 0);
6948 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6949 if (err)
6950 goto out_fail;
6951
6952 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6953 dentry->d_name.name,
6954 dentry->d_name.len, 0, index);
6955 if (err)
6956 goto out_fail;
6957
6958 d_instantiate_new(dentry, inode);
6959
6960 out_fail:
6961 btrfs_end_transaction(trans);
6962 if (err && inode) {
6963 inode_dec_link_count(inode);
6964 discard_new_inode(inode);
6965 }
6966 btrfs_btree_balance_dirty(fs_info);
6967 return err;
6968 }
6969
uncompress_inline(struct btrfs_path * path,struct page * page,size_t pg_offset,u64 extent_offset,struct btrfs_file_extent_item * item)6970 static noinline int uncompress_inline(struct btrfs_path *path,
6971 struct page *page,
6972 size_t pg_offset, u64 extent_offset,
6973 struct btrfs_file_extent_item *item)
6974 {
6975 int ret;
6976 struct extent_buffer *leaf = path->nodes[0];
6977 char *tmp;
6978 size_t max_size;
6979 unsigned long inline_size;
6980 unsigned long ptr;
6981 int compress_type;
6982
6983 WARN_ON(pg_offset != 0);
6984 compress_type = btrfs_file_extent_compression(leaf, item);
6985 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6986 inline_size = btrfs_file_extent_inline_item_len(leaf,
6987 btrfs_item_nr(path->slots[0]));
6988 tmp = kmalloc(inline_size, GFP_NOFS);
6989 if (!tmp)
6990 return -ENOMEM;
6991 ptr = btrfs_file_extent_inline_start(item);
6992
6993 read_extent_buffer(leaf, tmp, ptr, inline_size);
6994
6995 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6996 ret = btrfs_decompress(compress_type, tmp, page,
6997 extent_offset, inline_size, max_size);
6998
6999 /*
7000 * decompression code contains a memset to fill in any space between the end
7001 * of the uncompressed data and the end of max_size in case the decompressed
7002 * data ends up shorter than ram_bytes. That doesn't cover the hole between
7003 * the end of an inline extent and the beginning of the next block, so we
7004 * cover that region here.
7005 */
7006
7007 if (max_size + pg_offset < PAGE_SIZE)
7008 memzero_page(page, pg_offset + max_size,
7009 PAGE_SIZE - max_size - pg_offset);
7010 kfree(tmp);
7011 return ret;
7012 }
7013
7014 /**
7015 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
7016 * @inode: file to search in
7017 * @page: page to read extent data into if the extent is inline
7018 * @pg_offset: offset into @page to copy to
7019 * @start: file offset
7020 * @len: length of range starting at @start
7021 *
7022 * This returns the first &struct extent_map which overlaps with the given
7023 * range, reading it from the B-tree and caching it if necessary. Note that
7024 * there may be more extents which overlap the given range after the returned
7025 * extent_map.
7026 *
7027 * If @page is not NULL and the extent is inline, this also reads the extent
7028 * data directly into the page and marks the extent up to date in the io_tree.
7029 *
7030 * Return: ERR_PTR on error, non-NULL extent_map on success.
7031 */
btrfs_get_extent(struct btrfs_inode * inode,struct page * page,size_t pg_offset,u64 start,u64 len)7032 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
7033 struct page *page, size_t pg_offset,
7034 u64 start, u64 len)
7035 {
7036 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7037 int ret = 0;
7038 u64 extent_start = 0;
7039 u64 extent_end = 0;
7040 u64 objectid = btrfs_ino(inode);
7041 int extent_type = -1;
7042 struct btrfs_path *path = NULL;
7043 struct btrfs_root *root = inode->root;
7044 struct btrfs_file_extent_item *item;
7045 struct extent_buffer *leaf;
7046 struct btrfs_key found_key;
7047 struct extent_map *em = NULL;
7048 struct extent_map_tree *em_tree = &inode->extent_tree;
7049 struct extent_io_tree *io_tree = &inode->io_tree;
7050
7051 read_lock(&em_tree->lock);
7052 em = lookup_extent_mapping(em_tree, start, len);
7053 read_unlock(&em_tree->lock);
7054
7055 if (em) {
7056 if (em->start > start || em->start + em->len <= start)
7057 free_extent_map(em);
7058 else if (em->block_start == EXTENT_MAP_INLINE && page)
7059 free_extent_map(em);
7060 else
7061 goto out;
7062 }
7063 em = alloc_extent_map();
7064 if (!em) {
7065 ret = -ENOMEM;
7066 goto out;
7067 }
7068 em->start = EXTENT_MAP_HOLE;
7069 em->orig_start = EXTENT_MAP_HOLE;
7070 em->len = (u64)-1;
7071 em->block_len = (u64)-1;
7072
7073 path = btrfs_alloc_path();
7074 if (!path) {
7075 ret = -ENOMEM;
7076 goto out;
7077 }
7078
7079 /* Chances are we'll be called again, so go ahead and do readahead */
7080 path->reada = READA_FORWARD;
7081
7082 /*
7083 * The same explanation in load_free_space_cache applies here as well,
7084 * we only read when we're loading the free space cache, and at that
7085 * point the commit_root has everything we need.
7086 */
7087 if (btrfs_is_free_space_inode(inode)) {
7088 path->search_commit_root = 1;
7089 path->skip_locking = 1;
7090 }
7091
7092 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
7093 if (ret < 0) {
7094 goto out;
7095 } else if (ret > 0) {
7096 if (path->slots[0] == 0)
7097 goto not_found;
7098 path->slots[0]--;
7099 ret = 0;
7100 }
7101
7102 leaf = path->nodes[0];
7103 item = btrfs_item_ptr(leaf, path->slots[0],
7104 struct btrfs_file_extent_item);
7105 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7106 if (found_key.objectid != objectid ||
7107 found_key.type != BTRFS_EXTENT_DATA_KEY) {
7108 /*
7109 * If we backup past the first extent we want to move forward
7110 * and see if there is an extent in front of us, otherwise we'll
7111 * say there is a hole for our whole search range which can
7112 * cause problems.
7113 */
7114 extent_end = start;
7115 goto next;
7116 }
7117
7118 extent_type = btrfs_file_extent_type(leaf, item);
7119 extent_start = found_key.offset;
7120 extent_end = btrfs_file_extent_end(path);
7121 if (extent_type == BTRFS_FILE_EXTENT_REG ||
7122 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
7123 /* Only regular file could have regular/prealloc extent */
7124 if (!S_ISREG(inode->vfs_inode.i_mode)) {
7125 ret = -EUCLEAN;
7126 btrfs_crit(fs_info,
7127 "regular/prealloc extent found for non-regular inode %llu",
7128 btrfs_ino(inode));
7129 goto out;
7130 }
7131 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
7132 extent_start);
7133 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
7134 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
7135 path->slots[0],
7136 extent_start);
7137 }
7138 next:
7139 if (start >= extent_end) {
7140 path->slots[0]++;
7141 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7142 ret = btrfs_next_leaf(root, path);
7143 if (ret < 0)
7144 goto out;
7145 else if (ret > 0)
7146 goto not_found;
7147
7148 leaf = path->nodes[0];
7149 }
7150 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7151 if (found_key.objectid != objectid ||
7152 found_key.type != BTRFS_EXTENT_DATA_KEY)
7153 goto not_found;
7154 if (start + len <= found_key.offset)
7155 goto not_found;
7156 if (start > found_key.offset)
7157 goto next;
7158
7159 /* New extent overlaps with existing one */
7160 em->start = start;
7161 em->orig_start = start;
7162 em->len = found_key.offset - start;
7163 em->block_start = EXTENT_MAP_HOLE;
7164 goto insert;
7165 }
7166
7167 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
7168
7169 if (extent_type == BTRFS_FILE_EXTENT_REG ||
7170 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
7171 goto insert;
7172 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
7173 unsigned long ptr;
7174 char *map;
7175 size_t size;
7176 size_t extent_offset;
7177 size_t copy_size;
7178
7179 if (!page)
7180 goto out;
7181
7182 size = btrfs_file_extent_ram_bytes(leaf, item);
7183 extent_offset = page_offset(page) + pg_offset - extent_start;
7184 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7185 size - extent_offset);
7186 em->start = extent_start + extent_offset;
7187 em->len = ALIGN(copy_size, fs_info->sectorsize);
7188 em->orig_block_len = em->len;
7189 em->orig_start = em->start;
7190 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7191
7192 if (!PageUptodate(page)) {
7193 if (btrfs_file_extent_compression(leaf, item) !=
7194 BTRFS_COMPRESS_NONE) {
7195 ret = uncompress_inline(path, page, pg_offset,
7196 extent_offset, item);
7197 if (ret)
7198 goto out;
7199 } else {
7200 map = kmap_local_page(page);
7201 read_extent_buffer(leaf, map + pg_offset, ptr,
7202 copy_size);
7203 if (pg_offset + copy_size < PAGE_SIZE) {
7204 memset(map + pg_offset + copy_size, 0,
7205 PAGE_SIZE - pg_offset -
7206 copy_size);
7207 }
7208 kunmap_local(map);
7209 }
7210 flush_dcache_page(page);
7211 }
7212 set_extent_uptodate(io_tree, em->start,
7213 extent_map_end(em) - 1, NULL, GFP_NOFS);
7214 goto insert;
7215 }
7216 not_found:
7217 em->start = start;
7218 em->orig_start = start;
7219 em->len = len;
7220 em->block_start = EXTENT_MAP_HOLE;
7221 insert:
7222 ret = 0;
7223 btrfs_release_path(path);
7224 if (em->start > start || extent_map_end(em) <= start) {
7225 btrfs_err(fs_info,
7226 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7227 em->start, em->len, start, len);
7228 ret = -EIO;
7229 goto out;
7230 }
7231
7232 write_lock(&em_tree->lock);
7233 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7234 write_unlock(&em_tree->lock);
7235 out:
7236 btrfs_free_path(path);
7237
7238 trace_btrfs_get_extent(root, inode, em);
7239
7240 if (ret) {
7241 free_extent_map(em);
7242 return ERR_PTR(ret);
7243 }
7244 return em;
7245 }
7246
btrfs_get_extent_fiemap(struct btrfs_inode * inode,u64 start,u64 len)7247 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7248 u64 start, u64 len)
7249 {
7250 struct extent_map *em;
7251 struct extent_map *hole_em = NULL;
7252 u64 delalloc_start = start;
7253 u64 end;
7254 u64 delalloc_len;
7255 u64 delalloc_end;
7256 int err = 0;
7257
7258 em = btrfs_get_extent(inode, NULL, 0, start, len);
7259 if (IS_ERR(em))
7260 return em;
7261 /*
7262 * If our em maps to:
7263 * - a hole or
7264 * - a pre-alloc extent,
7265 * there might actually be delalloc bytes behind it.
7266 */
7267 if (em->block_start != EXTENT_MAP_HOLE &&
7268 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7269 return em;
7270 else
7271 hole_em = em;
7272
7273 /* check to see if we've wrapped (len == -1 or similar) */
7274 end = start + len;
7275 if (end < start)
7276 end = (u64)-1;
7277 else
7278 end -= 1;
7279
7280 em = NULL;
7281
7282 /* ok, we didn't find anything, lets look for delalloc */
7283 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
7284 end, len, EXTENT_DELALLOC, 1);
7285 delalloc_end = delalloc_start + delalloc_len;
7286 if (delalloc_end < delalloc_start)
7287 delalloc_end = (u64)-1;
7288
7289 /*
7290 * We didn't find anything useful, return the original results from
7291 * get_extent()
7292 */
7293 if (delalloc_start > end || delalloc_end <= start) {
7294 em = hole_em;
7295 hole_em = NULL;
7296 goto out;
7297 }
7298
7299 /*
7300 * Adjust the delalloc_start to make sure it doesn't go backwards from
7301 * the start they passed in
7302 */
7303 delalloc_start = max(start, delalloc_start);
7304 delalloc_len = delalloc_end - delalloc_start;
7305
7306 if (delalloc_len > 0) {
7307 u64 hole_start;
7308 u64 hole_len;
7309 const u64 hole_end = extent_map_end(hole_em);
7310
7311 em = alloc_extent_map();
7312 if (!em) {
7313 err = -ENOMEM;
7314 goto out;
7315 }
7316
7317 ASSERT(hole_em);
7318 /*
7319 * When btrfs_get_extent can't find anything it returns one
7320 * huge hole
7321 *
7322 * Make sure what it found really fits our range, and adjust to
7323 * make sure it is based on the start from the caller
7324 */
7325 if (hole_end <= start || hole_em->start > end) {
7326 free_extent_map(hole_em);
7327 hole_em = NULL;
7328 } else {
7329 hole_start = max(hole_em->start, start);
7330 hole_len = hole_end - hole_start;
7331 }
7332
7333 if (hole_em && delalloc_start > hole_start) {
7334 /*
7335 * Our hole starts before our delalloc, so we have to
7336 * return just the parts of the hole that go until the
7337 * delalloc starts
7338 */
7339 em->len = min(hole_len, delalloc_start - hole_start);
7340 em->start = hole_start;
7341 em->orig_start = hole_start;
7342 /*
7343 * Don't adjust block start at all, it is fixed at
7344 * EXTENT_MAP_HOLE
7345 */
7346 em->block_start = hole_em->block_start;
7347 em->block_len = hole_len;
7348 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7349 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7350 } else {
7351 /*
7352 * Hole is out of passed range or it starts after
7353 * delalloc range
7354 */
7355 em->start = delalloc_start;
7356 em->len = delalloc_len;
7357 em->orig_start = delalloc_start;
7358 em->block_start = EXTENT_MAP_DELALLOC;
7359 em->block_len = delalloc_len;
7360 }
7361 } else {
7362 return hole_em;
7363 }
7364 out:
7365
7366 free_extent_map(hole_em);
7367 if (err) {
7368 free_extent_map(em);
7369 return ERR_PTR(err);
7370 }
7371 return em;
7372 }
7373
btrfs_create_dio_extent(struct btrfs_inode * inode,const u64 start,const u64 len,const u64 orig_start,const u64 block_start,const u64 block_len,const u64 orig_block_len,const u64 ram_bytes,const int type)7374 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7375 const u64 start,
7376 const u64 len,
7377 const u64 orig_start,
7378 const u64 block_start,
7379 const u64 block_len,
7380 const u64 orig_block_len,
7381 const u64 ram_bytes,
7382 const int type)
7383 {
7384 struct extent_map *em = NULL;
7385 int ret;
7386
7387 if (type != BTRFS_ORDERED_NOCOW) {
7388 em = create_io_em(inode, start, len, orig_start, block_start,
7389 block_len, orig_block_len, ram_bytes,
7390 BTRFS_COMPRESS_NONE, /* compress_type */
7391 type);
7392 if (IS_ERR(em))
7393 goto out;
7394 }
7395 ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len,
7396 block_len, type);
7397 if (ret) {
7398 if (em) {
7399 free_extent_map(em);
7400 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7401 }
7402 em = ERR_PTR(ret);
7403 }
7404 out:
7405
7406 return em;
7407 }
7408
btrfs_new_extent_direct(struct btrfs_inode * inode,u64 start,u64 len)7409 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7410 u64 start, u64 len)
7411 {
7412 struct btrfs_root *root = inode->root;
7413 struct btrfs_fs_info *fs_info = root->fs_info;
7414 struct extent_map *em;
7415 struct btrfs_key ins;
7416 u64 alloc_hint;
7417 int ret;
7418
7419 alloc_hint = get_extent_allocation_hint(inode, start, len);
7420 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7421 0, alloc_hint, &ins, 1, 1);
7422 if (ret)
7423 return ERR_PTR(ret);
7424
7425 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7426 ins.objectid, ins.offset, ins.offset,
7427 ins.offset, BTRFS_ORDERED_REGULAR);
7428 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7429 if (IS_ERR(em))
7430 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7431 1);
7432
7433 return em;
7434 }
7435
btrfs_extent_readonly(struct btrfs_fs_info * fs_info,u64 bytenr)7436 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7437 {
7438 struct btrfs_block_group *block_group;
7439 bool readonly = false;
7440
7441 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7442 if (!block_group || block_group->ro)
7443 readonly = true;
7444 if (block_group)
7445 btrfs_put_block_group(block_group);
7446 return readonly;
7447 }
7448
7449 /*
7450 * Check if we can do nocow write into the range [@offset, @offset + @len)
7451 *
7452 * @offset: File offset
7453 * @len: The length to write, will be updated to the nocow writeable
7454 * range
7455 * @orig_start: (optional) Return the original file offset of the file extent
7456 * @orig_len: (optional) Return the original on-disk length of the file extent
7457 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7458 * @strict: if true, omit optimizations that might force us into unnecessary
7459 * cow. e.g., don't trust generation number.
7460 *
7461 * Return:
7462 * >0 and update @len if we can do nocow write
7463 * 0 if we can't do nocow write
7464 * <0 if error happened
7465 *
7466 * NOTE: This only checks the file extents, caller is responsible to wait for
7467 * any ordered extents.
7468 */
can_nocow_extent(struct inode * inode,u64 offset,u64 * len,u64 * orig_start,u64 * orig_block_len,u64 * ram_bytes,bool strict)7469 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7470 u64 *orig_start, u64 *orig_block_len,
7471 u64 *ram_bytes, bool strict)
7472 {
7473 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7474 struct btrfs_path *path;
7475 int ret;
7476 struct extent_buffer *leaf;
7477 struct btrfs_root *root = BTRFS_I(inode)->root;
7478 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7479 struct btrfs_file_extent_item *fi;
7480 struct btrfs_key key;
7481 u64 disk_bytenr;
7482 u64 backref_offset;
7483 u64 extent_end;
7484 u64 num_bytes;
7485 int slot;
7486 int found_type;
7487 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7488
7489 path = btrfs_alloc_path();
7490 if (!path)
7491 return -ENOMEM;
7492
7493 ret = btrfs_lookup_file_extent(NULL, root, path,
7494 btrfs_ino(BTRFS_I(inode)), offset, 0);
7495 if (ret < 0)
7496 goto out;
7497
7498 slot = path->slots[0];
7499 if (ret == 1) {
7500 if (slot == 0) {
7501 /* can't find the item, must cow */
7502 ret = 0;
7503 goto out;
7504 }
7505 slot--;
7506 }
7507 ret = 0;
7508 leaf = path->nodes[0];
7509 btrfs_item_key_to_cpu(leaf, &key, slot);
7510 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7511 key.type != BTRFS_EXTENT_DATA_KEY) {
7512 /* not our file or wrong item type, must cow */
7513 goto out;
7514 }
7515
7516 if (key.offset > offset) {
7517 /* Wrong offset, must cow */
7518 goto out;
7519 }
7520
7521 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7522 found_type = btrfs_file_extent_type(leaf, fi);
7523 if (found_type != BTRFS_FILE_EXTENT_REG &&
7524 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7525 /* not a regular extent, must cow */
7526 goto out;
7527 }
7528
7529 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7530 goto out;
7531
7532 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7533 if (extent_end <= offset)
7534 goto out;
7535
7536 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7537 if (disk_bytenr == 0)
7538 goto out;
7539
7540 if (btrfs_file_extent_compression(leaf, fi) ||
7541 btrfs_file_extent_encryption(leaf, fi) ||
7542 btrfs_file_extent_other_encoding(leaf, fi))
7543 goto out;
7544
7545 /*
7546 * Do the same check as in btrfs_cross_ref_exist but without the
7547 * unnecessary search.
7548 */
7549 if (!strict &&
7550 (btrfs_file_extent_generation(leaf, fi) <=
7551 btrfs_root_last_snapshot(&root->root_item)))
7552 goto out;
7553
7554 backref_offset = btrfs_file_extent_offset(leaf, fi);
7555
7556 if (orig_start) {
7557 *orig_start = key.offset - backref_offset;
7558 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7559 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7560 }
7561
7562 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7563 goto out;
7564
7565 num_bytes = min(offset + *len, extent_end) - offset;
7566 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7567 u64 range_end;
7568
7569 range_end = round_up(offset + num_bytes,
7570 root->fs_info->sectorsize) - 1;
7571 ret = test_range_bit(io_tree, offset, range_end,
7572 EXTENT_DELALLOC, 0, NULL);
7573 if (ret) {
7574 ret = -EAGAIN;
7575 goto out;
7576 }
7577 }
7578
7579 btrfs_release_path(path);
7580
7581 /*
7582 * look for other files referencing this extent, if we
7583 * find any we must cow
7584 */
7585
7586 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7587 key.offset - backref_offset, disk_bytenr,
7588 strict);
7589 if (ret) {
7590 ret = 0;
7591 goto out;
7592 }
7593
7594 /*
7595 * adjust disk_bytenr and num_bytes to cover just the bytes
7596 * in this extent we are about to write. If there
7597 * are any csums in that range we have to cow in order
7598 * to keep the csums correct
7599 */
7600 disk_bytenr += backref_offset;
7601 disk_bytenr += offset - key.offset;
7602 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7603 goto out;
7604 /*
7605 * all of the above have passed, it is safe to overwrite this extent
7606 * without cow
7607 */
7608 *len = num_bytes;
7609 ret = 1;
7610 out:
7611 btrfs_free_path(path);
7612 return ret;
7613 }
7614
lock_extent_direct(struct inode * inode,u64 lockstart,u64 lockend,struct extent_state ** cached_state,bool writing)7615 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7616 struct extent_state **cached_state, bool writing)
7617 {
7618 struct btrfs_ordered_extent *ordered;
7619 int ret = 0;
7620
7621 while (1) {
7622 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7623 cached_state);
7624 /*
7625 * We're concerned with the entire range that we're going to be
7626 * doing DIO to, so we need to make sure there's no ordered
7627 * extents in this range.
7628 */
7629 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7630 lockend - lockstart + 1);
7631
7632 /*
7633 * We need to make sure there are no buffered pages in this
7634 * range either, we could have raced between the invalidate in
7635 * generic_file_direct_write and locking the extent. The
7636 * invalidate needs to happen so that reads after a write do not
7637 * get stale data.
7638 */
7639 if (!ordered &&
7640 (!writing || !filemap_range_has_page(inode->i_mapping,
7641 lockstart, lockend)))
7642 break;
7643
7644 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7645 cached_state);
7646
7647 if (ordered) {
7648 /*
7649 * If we are doing a DIO read and the ordered extent we
7650 * found is for a buffered write, we can not wait for it
7651 * to complete and retry, because if we do so we can
7652 * deadlock with concurrent buffered writes on page
7653 * locks. This happens only if our DIO read covers more
7654 * than one extent map, if at this point has already
7655 * created an ordered extent for a previous extent map
7656 * and locked its range in the inode's io tree, and a
7657 * concurrent write against that previous extent map's
7658 * range and this range started (we unlock the ranges
7659 * in the io tree only when the bios complete and
7660 * buffered writes always lock pages before attempting
7661 * to lock range in the io tree).
7662 */
7663 if (writing ||
7664 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7665 btrfs_start_ordered_extent(ordered, 1);
7666 else
7667 ret = -ENOTBLK;
7668 btrfs_put_ordered_extent(ordered);
7669 } else {
7670 /*
7671 * We could trigger writeback for this range (and wait
7672 * for it to complete) and then invalidate the pages for
7673 * this range (through invalidate_inode_pages2_range()),
7674 * but that can lead us to a deadlock with a concurrent
7675 * call to readahead (a buffered read or a defrag call
7676 * triggered a readahead) on a page lock due to an
7677 * ordered dio extent we created before but did not have
7678 * yet a corresponding bio submitted (whence it can not
7679 * complete), which makes readahead wait for that
7680 * ordered extent to complete while holding a lock on
7681 * that page.
7682 */
7683 ret = -ENOTBLK;
7684 }
7685
7686 if (ret)
7687 break;
7688
7689 cond_resched();
7690 }
7691
7692 return ret;
7693 }
7694
7695 /* The callers of this must take lock_extent() */
create_io_em(struct btrfs_inode * inode,u64 start,u64 len,u64 orig_start,u64 block_start,u64 block_len,u64 orig_block_len,u64 ram_bytes,int compress_type,int type)7696 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7697 u64 len, u64 orig_start, u64 block_start,
7698 u64 block_len, u64 orig_block_len,
7699 u64 ram_bytes, int compress_type,
7700 int type)
7701 {
7702 struct extent_map_tree *em_tree;
7703 struct extent_map *em;
7704 int ret;
7705
7706 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7707 type == BTRFS_ORDERED_COMPRESSED ||
7708 type == BTRFS_ORDERED_NOCOW ||
7709 type == BTRFS_ORDERED_REGULAR);
7710
7711 em_tree = &inode->extent_tree;
7712 em = alloc_extent_map();
7713 if (!em)
7714 return ERR_PTR(-ENOMEM);
7715
7716 em->start = start;
7717 em->orig_start = orig_start;
7718 em->len = len;
7719 em->block_len = block_len;
7720 em->block_start = block_start;
7721 em->orig_block_len = orig_block_len;
7722 em->ram_bytes = ram_bytes;
7723 em->generation = -1;
7724 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7725 if (type == BTRFS_ORDERED_PREALLOC) {
7726 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7727 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7728 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7729 em->compress_type = compress_type;
7730 }
7731
7732 do {
7733 btrfs_drop_extent_cache(inode, em->start,
7734 em->start + em->len - 1, 0);
7735 write_lock(&em_tree->lock);
7736 ret = add_extent_mapping(em_tree, em, 1);
7737 write_unlock(&em_tree->lock);
7738 /*
7739 * The caller has taken lock_extent(), who could race with us
7740 * to add em?
7741 */
7742 } while (ret == -EEXIST);
7743
7744 if (ret) {
7745 free_extent_map(em);
7746 return ERR_PTR(ret);
7747 }
7748
7749 /* em got 2 refs now, callers needs to do free_extent_map once. */
7750 return em;
7751 }
7752
7753
btrfs_get_blocks_direct_write(struct extent_map ** map,struct inode * inode,struct btrfs_dio_data * dio_data,u64 start,u64 len)7754 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7755 struct inode *inode,
7756 struct btrfs_dio_data *dio_data,
7757 u64 start, u64 len)
7758 {
7759 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7760 struct extent_map *em = *map;
7761 int ret = 0;
7762
7763 /*
7764 * We don't allocate a new extent in the following cases
7765 *
7766 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7767 * existing extent.
7768 * 2) The extent is marked as PREALLOC. We're good to go here and can
7769 * just use the extent.
7770 *
7771 */
7772 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7773 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7774 em->block_start != EXTENT_MAP_HOLE)) {
7775 int type;
7776 u64 block_start, orig_start, orig_block_len, ram_bytes;
7777
7778 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7779 type = BTRFS_ORDERED_PREALLOC;
7780 else
7781 type = BTRFS_ORDERED_NOCOW;
7782 len = min(len, em->len - (start - em->start));
7783 block_start = em->block_start + (start - em->start);
7784
7785 if (can_nocow_extent(inode, start, &len, &orig_start,
7786 &orig_block_len, &ram_bytes, false) == 1 &&
7787 btrfs_inc_nocow_writers(fs_info, block_start)) {
7788 struct extent_map *em2;
7789
7790 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7791 orig_start, block_start,
7792 len, orig_block_len,
7793 ram_bytes, type);
7794 btrfs_dec_nocow_writers(fs_info, block_start);
7795 if (type == BTRFS_ORDERED_PREALLOC) {
7796 free_extent_map(em);
7797 *map = em = em2;
7798 }
7799
7800 if (em2 && IS_ERR(em2)) {
7801 ret = PTR_ERR(em2);
7802 goto out;
7803 }
7804 /*
7805 * For inode marked NODATACOW or extent marked PREALLOC,
7806 * use the existing or preallocated extent, so does not
7807 * need to adjust btrfs_space_info's bytes_may_use.
7808 */
7809 btrfs_free_reserved_data_space_noquota(fs_info, len);
7810 goto skip_cow;
7811 }
7812 }
7813
7814 /* this will cow the extent */
7815 free_extent_map(em);
7816 *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7817 if (IS_ERR(em)) {
7818 ret = PTR_ERR(em);
7819 goto out;
7820 }
7821
7822 len = min(len, em->len - (start - em->start));
7823
7824 skip_cow:
7825 /*
7826 * Need to update the i_size under the extent lock so buffered
7827 * readers will get the updated i_size when we unlock.
7828 */
7829 if (start + len > i_size_read(inode))
7830 i_size_write(inode, start + len);
7831
7832 dio_data->reserve -= len;
7833 out:
7834 return ret;
7835 }
7836
btrfs_dio_iomap_begin(struct inode * inode,loff_t start,loff_t length,unsigned int flags,struct iomap * iomap,struct iomap * srcmap)7837 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7838 loff_t length, unsigned int flags, struct iomap *iomap,
7839 struct iomap *srcmap)
7840 {
7841 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7842 struct extent_map *em;
7843 struct extent_state *cached_state = NULL;
7844 struct btrfs_dio_data *dio_data = NULL;
7845 u64 lockstart, lockend;
7846 const bool write = !!(flags & IOMAP_WRITE);
7847 int ret = 0;
7848 u64 len = length;
7849 bool unlock_extents = false;
7850
7851 if (!write)
7852 len = min_t(u64, len, fs_info->sectorsize);
7853
7854 lockstart = start;
7855 lockend = start + len - 1;
7856
7857 /*
7858 * The generic stuff only does filemap_write_and_wait_range, which
7859 * isn't enough if we've written compressed pages to this area, so we
7860 * need to flush the dirty pages again to make absolutely sure that any
7861 * outstanding dirty pages are on disk.
7862 */
7863 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7864 &BTRFS_I(inode)->runtime_flags)) {
7865 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7866 start + length - 1);
7867 if (ret)
7868 return ret;
7869 }
7870
7871 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7872 if (!dio_data)
7873 return -ENOMEM;
7874
7875 dio_data->length = length;
7876 if (write) {
7877 dio_data->reserve = round_up(length, fs_info->sectorsize);
7878 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7879 &dio_data->data_reserved,
7880 start, dio_data->reserve);
7881 if (ret) {
7882 extent_changeset_free(dio_data->data_reserved);
7883 kfree(dio_data);
7884 return ret;
7885 }
7886 }
7887 iomap->private = dio_data;
7888
7889
7890 /*
7891 * If this errors out it's because we couldn't invalidate pagecache for
7892 * this range and we need to fallback to buffered.
7893 */
7894 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7895 ret = -ENOTBLK;
7896 goto err;
7897 }
7898
7899 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7900 if (IS_ERR(em)) {
7901 ret = PTR_ERR(em);
7902 goto unlock_err;
7903 }
7904
7905 /*
7906 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7907 * io. INLINE is special, and we could probably kludge it in here, but
7908 * it's still buffered so for safety lets just fall back to the generic
7909 * buffered path.
7910 *
7911 * For COMPRESSED we _have_ to read the entire extent in so we can
7912 * decompress it, so there will be buffering required no matter what we
7913 * do, so go ahead and fallback to buffered.
7914 *
7915 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7916 * to buffered IO. Don't blame me, this is the price we pay for using
7917 * the generic code.
7918 */
7919 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7920 em->block_start == EXTENT_MAP_INLINE) {
7921 free_extent_map(em);
7922 ret = -ENOTBLK;
7923 goto unlock_err;
7924 }
7925
7926 len = min(len, em->len - (start - em->start));
7927 if (write) {
7928 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7929 start, len);
7930 if (ret < 0)
7931 goto unlock_err;
7932 unlock_extents = true;
7933 /* Recalc len in case the new em is smaller than requested */
7934 len = min(len, em->len - (start - em->start));
7935 } else {
7936 /*
7937 * We need to unlock only the end area that we aren't using.
7938 * The rest is going to be unlocked by the endio routine.
7939 */
7940 lockstart = start + len;
7941 if (lockstart < lockend)
7942 unlock_extents = true;
7943 }
7944
7945 if (unlock_extents)
7946 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7947 lockstart, lockend, &cached_state);
7948 else
7949 free_extent_state(cached_state);
7950
7951 /*
7952 * Translate extent map information to iomap.
7953 * We trim the extents (and move the addr) even though iomap code does
7954 * that, since we have locked only the parts we are performing I/O in.
7955 */
7956 if ((em->block_start == EXTENT_MAP_HOLE) ||
7957 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7958 iomap->addr = IOMAP_NULL_ADDR;
7959 iomap->type = IOMAP_HOLE;
7960 } else {
7961 iomap->addr = em->block_start + (start - em->start);
7962 iomap->type = IOMAP_MAPPED;
7963 }
7964 iomap->offset = start;
7965 iomap->bdev = fs_info->fs_devices->latest_bdev;
7966 iomap->length = len;
7967
7968 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7969 iomap->flags |= IOMAP_F_ZONE_APPEND;
7970
7971 free_extent_map(em);
7972
7973 return 0;
7974
7975 unlock_err:
7976 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7977 &cached_state);
7978 err:
7979 if (dio_data) {
7980 btrfs_delalloc_release_space(BTRFS_I(inode),
7981 dio_data->data_reserved, start,
7982 dio_data->reserve, true);
7983 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->reserve);
7984 extent_changeset_free(dio_data->data_reserved);
7985 kfree(dio_data);
7986 }
7987 return ret;
7988 }
7989
btrfs_dio_iomap_end(struct inode * inode,loff_t pos,loff_t length,ssize_t written,unsigned int flags,struct iomap * iomap)7990 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7991 ssize_t written, unsigned int flags, struct iomap *iomap)
7992 {
7993 int ret = 0;
7994 struct btrfs_dio_data *dio_data = iomap->private;
7995 size_t submitted = dio_data->submitted;
7996 const bool write = !!(flags & IOMAP_WRITE);
7997
7998 if (!write && (iomap->type == IOMAP_HOLE)) {
7999 /* If reading from a hole, unlock and return */
8000 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
8001 goto out;
8002 }
8003
8004 if (submitted < length) {
8005 pos += submitted;
8006 length -= submitted;
8007 if (write)
8008 __endio_write_update_ordered(BTRFS_I(inode), pos,
8009 length, false);
8010 else
8011 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
8012 pos + length - 1);
8013 ret = -ENOTBLK;
8014 }
8015
8016 if (write) {
8017 if (dio_data->reserve)
8018 btrfs_delalloc_release_space(BTRFS_I(inode),
8019 dio_data->data_reserved, pos,
8020 dio_data->reserve, true);
8021 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->length);
8022 extent_changeset_free(dio_data->data_reserved);
8023 }
8024 out:
8025 kfree(dio_data);
8026 iomap->private = NULL;
8027
8028 return ret;
8029 }
8030
btrfs_dio_private_put(struct btrfs_dio_private * dip)8031 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
8032 {
8033 /*
8034 * This implies a barrier so that stores to dio_bio->bi_status before
8035 * this and loads of dio_bio->bi_status after this are fully ordered.
8036 */
8037 if (!refcount_dec_and_test(&dip->refs))
8038 return;
8039
8040 if (btrfs_op(dip->dio_bio) == BTRFS_MAP_WRITE) {
8041 __endio_write_update_ordered(BTRFS_I(dip->inode),
8042 dip->logical_offset,
8043 dip->bytes,
8044 !dip->dio_bio->bi_status);
8045 } else {
8046 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
8047 dip->logical_offset,
8048 dip->logical_offset + dip->bytes - 1);
8049 }
8050
8051 bio_endio(dip->dio_bio);
8052 kfree(dip);
8053 }
8054
submit_dio_repair_bio(struct inode * inode,struct bio * bio,int mirror_num,unsigned long bio_flags)8055 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
8056 int mirror_num,
8057 unsigned long bio_flags)
8058 {
8059 struct btrfs_dio_private *dip = bio->bi_private;
8060 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8061 blk_status_t ret;
8062
8063 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
8064
8065 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8066 if (ret)
8067 return ret;
8068
8069 refcount_inc(&dip->refs);
8070 ret = btrfs_map_bio(fs_info, bio, mirror_num);
8071 if (ret)
8072 refcount_dec(&dip->refs);
8073 return ret;
8074 }
8075
btrfs_check_read_dio_bio(struct inode * inode,struct btrfs_io_bio * io_bio,const bool uptodate)8076 static blk_status_t btrfs_check_read_dio_bio(struct inode *inode,
8077 struct btrfs_io_bio *io_bio,
8078 const bool uptodate)
8079 {
8080 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
8081 const u32 sectorsize = fs_info->sectorsize;
8082 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
8083 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8084 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
8085 struct bio_vec bvec;
8086 struct bvec_iter iter;
8087 u64 start = io_bio->logical;
8088 u32 bio_offset = 0;
8089 blk_status_t err = BLK_STS_OK;
8090
8091 __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) {
8092 unsigned int i, nr_sectors, pgoff;
8093
8094 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8095 pgoff = bvec.bv_offset;
8096 for (i = 0; i < nr_sectors; i++) {
8097 ASSERT(pgoff < PAGE_SIZE);
8098 if (uptodate &&
8099 (!csum || !check_data_csum(inode, io_bio,
8100 bio_offset, bvec.bv_page,
8101 pgoff, start))) {
8102 clean_io_failure(fs_info, failure_tree, io_tree,
8103 start, bvec.bv_page,
8104 btrfs_ino(BTRFS_I(inode)),
8105 pgoff);
8106 } else {
8107 int ret;
8108
8109 ASSERT((start - io_bio->logical) < UINT_MAX);
8110 ret = btrfs_repair_one_sector(inode,
8111 &io_bio->bio,
8112 start - io_bio->logical,
8113 bvec.bv_page, pgoff,
8114 start, io_bio->mirror_num,
8115 submit_dio_repair_bio);
8116 if (ret)
8117 err = errno_to_blk_status(ret);
8118 }
8119 start += sectorsize;
8120 ASSERT(bio_offset + sectorsize > bio_offset);
8121 bio_offset += sectorsize;
8122 pgoff += sectorsize;
8123 }
8124 }
8125 return err;
8126 }
8127
__endio_write_update_ordered(struct btrfs_inode * inode,const u64 offset,const u64 bytes,const bool uptodate)8128 static void __endio_write_update_ordered(struct btrfs_inode *inode,
8129 const u64 offset, const u64 bytes,
8130 const bool uptodate)
8131 {
8132 btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes,
8133 finish_ordered_fn, uptodate);
8134 }
8135
btrfs_submit_bio_start_direct_io(struct inode * inode,struct bio * bio,u64 dio_file_offset)8136 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
8137 struct bio *bio,
8138 u64 dio_file_offset)
8139 {
8140 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, 1);
8141 }
8142
btrfs_end_dio_bio(struct bio * bio)8143 static void btrfs_end_dio_bio(struct bio *bio)
8144 {
8145 struct btrfs_dio_private *dip = bio->bi_private;
8146 blk_status_t err = bio->bi_status;
8147
8148 if (err)
8149 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8150 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8151 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8152 bio->bi_opf, bio->bi_iter.bi_sector,
8153 bio->bi_iter.bi_size, err);
8154
8155 if (bio_op(bio) == REQ_OP_READ) {
8156 err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio),
8157 !err);
8158 }
8159
8160 if (err)
8161 dip->dio_bio->bi_status = err;
8162
8163 btrfs_record_physical_zoned(dip->inode, dip->logical_offset, bio);
8164
8165 bio_put(bio);
8166 btrfs_dio_private_put(dip);
8167 }
8168
btrfs_submit_dio_bio(struct bio * bio,struct inode * inode,u64 file_offset,int async_submit)8169 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8170 struct inode *inode, u64 file_offset, int async_submit)
8171 {
8172 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8173 struct btrfs_dio_private *dip = bio->bi_private;
8174 bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
8175 blk_status_t ret;
8176
8177 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8178 if (async_submit)
8179 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8180
8181 if (!write) {
8182 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8183 if (ret)
8184 goto err;
8185 }
8186
8187 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8188 goto map;
8189
8190 if (write && async_submit) {
8191 ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset,
8192 btrfs_submit_bio_start_direct_io);
8193 goto err;
8194 } else if (write) {
8195 /*
8196 * If we aren't doing async submit, calculate the csum of the
8197 * bio now.
8198 */
8199 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1);
8200 if (ret)
8201 goto err;
8202 } else {
8203 u64 csum_offset;
8204
8205 csum_offset = file_offset - dip->logical_offset;
8206 csum_offset >>= fs_info->sectorsize_bits;
8207 csum_offset *= fs_info->csum_size;
8208 btrfs_io_bio(bio)->csum = dip->csums + csum_offset;
8209 }
8210 map:
8211 ret = btrfs_map_bio(fs_info, bio, 0);
8212 err:
8213 return ret;
8214 }
8215
8216 /*
8217 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
8218 * or ordered extents whether or not we submit any bios.
8219 */
btrfs_create_dio_private(struct bio * dio_bio,struct inode * inode,loff_t file_offset)8220 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
8221 struct inode *inode,
8222 loff_t file_offset)
8223 {
8224 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8225 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
8226 size_t dip_size;
8227 struct btrfs_dio_private *dip;
8228
8229 dip_size = sizeof(*dip);
8230 if (!write && csum) {
8231 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8232 size_t nblocks;
8233
8234 nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
8235 dip_size += fs_info->csum_size * nblocks;
8236 }
8237
8238 dip = kzalloc(dip_size, GFP_NOFS);
8239 if (!dip)
8240 return NULL;
8241
8242 dip->inode = inode;
8243 dip->logical_offset = file_offset;
8244 dip->bytes = dio_bio->bi_iter.bi_size;
8245 dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9;
8246 dip->dio_bio = dio_bio;
8247 refcount_set(&dip->refs, 1);
8248 return dip;
8249 }
8250
btrfs_submit_direct(const struct iomap_iter * iter,struct bio * dio_bio,loff_t file_offset)8251 static blk_qc_t btrfs_submit_direct(const struct iomap_iter *iter,
8252 struct bio *dio_bio, loff_t file_offset)
8253 {
8254 struct inode *inode = iter->inode;
8255 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8256 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8257 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
8258 BTRFS_BLOCK_GROUP_RAID56_MASK);
8259 struct btrfs_dio_private *dip;
8260 struct bio *bio;
8261 u64 start_sector;
8262 int async_submit = 0;
8263 u64 submit_len;
8264 u64 clone_offset = 0;
8265 u64 clone_len;
8266 u64 logical;
8267 int ret;
8268 blk_status_t status;
8269 struct btrfs_io_geometry geom;
8270 struct btrfs_dio_data *dio_data = iter->iomap.private;
8271 struct extent_map *em = NULL;
8272
8273 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
8274 if (!dip) {
8275 if (!write) {
8276 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8277 file_offset + dio_bio->bi_iter.bi_size - 1);
8278 }
8279 dio_bio->bi_status = BLK_STS_RESOURCE;
8280 bio_endio(dio_bio);
8281 return BLK_QC_T_NONE;
8282 }
8283
8284 if (!write) {
8285 /*
8286 * Load the csums up front to reduce csum tree searches and
8287 * contention when submitting bios.
8288 *
8289 * If we have csums disabled this will do nothing.
8290 */
8291 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
8292 if (status != BLK_STS_OK)
8293 goto out_err;
8294 }
8295
8296 start_sector = dio_bio->bi_iter.bi_sector;
8297 submit_len = dio_bio->bi_iter.bi_size;
8298
8299 do {
8300 logical = start_sector << 9;
8301 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
8302 if (IS_ERR(em)) {
8303 status = errno_to_blk_status(PTR_ERR(em));
8304 em = NULL;
8305 goto out_err_em;
8306 }
8307 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
8308 logical, &geom);
8309 if (ret) {
8310 status = errno_to_blk_status(ret);
8311 goto out_err_em;
8312 }
8313
8314 clone_len = min(submit_len, geom.len);
8315 ASSERT(clone_len <= UINT_MAX);
8316
8317 /*
8318 * This will never fail as it's passing GPF_NOFS and
8319 * the allocation is backed by btrfs_bioset.
8320 */
8321 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
8322 bio->bi_private = dip;
8323 bio->bi_end_io = btrfs_end_dio_bio;
8324 btrfs_io_bio(bio)->logical = file_offset;
8325
8326 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
8327 status = extract_ordered_extent(BTRFS_I(inode), bio,
8328 file_offset);
8329 if (status) {
8330 bio_put(bio);
8331 goto out_err;
8332 }
8333 }
8334
8335 ASSERT(submit_len >= clone_len);
8336 submit_len -= clone_len;
8337
8338 /*
8339 * Increase the count before we submit the bio so we know
8340 * the end IO handler won't happen before we increase the
8341 * count. Otherwise, the dip might get freed before we're
8342 * done setting it up.
8343 *
8344 * We transfer the initial reference to the last bio, so we
8345 * don't need to increment the reference count for the last one.
8346 */
8347 if (submit_len > 0) {
8348 refcount_inc(&dip->refs);
8349 /*
8350 * If we are submitting more than one bio, submit them
8351 * all asynchronously. The exception is RAID 5 or 6, as
8352 * asynchronous checksums make it difficult to collect
8353 * full stripe writes.
8354 */
8355 if (!raid56)
8356 async_submit = 1;
8357 }
8358
8359 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8360 async_submit);
8361 if (status) {
8362 bio_put(bio);
8363 if (submit_len > 0)
8364 refcount_dec(&dip->refs);
8365 goto out_err_em;
8366 }
8367
8368 dio_data->submitted += clone_len;
8369 clone_offset += clone_len;
8370 start_sector += clone_len >> 9;
8371 file_offset += clone_len;
8372
8373 free_extent_map(em);
8374 } while (submit_len > 0);
8375 return BLK_QC_T_NONE;
8376
8377 out_err_em:
8378 free_extent_map(em);
8379 out_err:
8380 dip->dio_bio->bi_status = status;
8381 btrfs_dio_private_put(dip);
8382
8383 return BLK_QC_T_NONE;
8384 }
8385
8386 const struct iomap_ops btrfs_dio_iomap_ops = {
8387 .iomap_begin = btrfs_dio_iomap_begin,
8388 .iomap_end = btrfs_dio_iomap_end,
8389 };
8390
8391 const struct iomap_dio_ops btrfs_dio_ops = {
8392 .submit_io = btrfs_submit_direct,
8393 };
8394
btrfs_fiemap(struct inode * inode,struct fiemap_extent_info * fieinfo,u64 start,u64 len)8395 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8396 u64 start, u64 len)
8397 {
8398 int ret;
8399
8400 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8401 if (ret)
8402 return ret;
8403
8404 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8405 }
8406
btrfs_readpage(struct file * file,struct page * page)8407 int btrfs_readpage(struct file *file, struct page *page)
8408 {
8409 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8410 u64 start = page_offset(page);
8411 u64 end = start + PAGE_SIZE - 1;
8412 struct btrfs_bio_ctrl bio_ctrl = { 0 };
8413 int ret;
8414
8415 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8416
8417 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL);
8418 if (bio_ctrl.bio)
8419 ret = submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.bio_flags);
8420 return ret;
8421 }
8422
btrfs_writepage(struct page * page,struct writeback_control * wbc)8423 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8424 {
8425 struct inode *inode = page->mapping->host;
8426 int ret;
8427
8428 if (current->flags & PF_MEMALLOC) {
8429 redirty_page_for_writepage(wbc, page);
8430 unlock_page(page);
8431 return 0;
8432 }
8433
8434 /*
8435 * If we are under memory pressure we will call this directly from the
8436 * VM, we need to make sure we have the inode referenced for the ordered
8437 * extent. If not just return like we didn't do anything.
8438 */
8439 if (!igrab(inode)) {
8440 redirty_page_for_writepage(wbc, page);
8441 return AOP_WRITEPAGE_ACTIVATE;
8442 }
8443 ret = extent_write_full_page(page, wbc);
8444 btrfs_add_delayed_iput(inode);
8445 return ret;
8446 }
8447
btrfs_writepages(struct address_space * mapping,struct writeback_control * wbc)8448 static int btrfs_writepages(struct address_space *mapping,
8449 struct writeback_control *wbc)
8450 {
8451 return extent_writepages(mapping, wbc);
8452 }
8453
btrfs_readahead(struct readahead_control * rac)8454 static void btrfs_readahead(struct readahead_control *rac)
8455 {
8456 extent_readahead(rac);
8457 }
8458
8459 /*
8460 * For releasepage() and invalidatepage() we have a race window where
8461 * end_page_writeback() is called but the subpage spinlock is not yet released.
8462 * If we continue to release/invalidate the page, we could cause use-after-free
8463 * for subpage spinlock. So this function is to spin and wait for subpage
8464 * spinlock.
8465 */
wait_subpage_spinlock(struct page * page)8466 static void wait_subpage_spinlock(struct page *page)
8467 {
8468 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
8469 struct btrfs_subpage *subpage;
8470
8471 if (fs_info->sectorsize == PAGE_SIZE)
8472 return;
8473
8474 ASSERT(PagePrivate(page) && page->private);
8475 subpage = (struct btrfs_subpage *)page->private;
8476
8477 /*
8478 * This may look insane as we just acquire the spinlock and release it,
8479 * without doing anything. But we just want to make sure no one is
8480 * still holding the subpage spinlock.
8481 * And since the page is not dirty nor writeback, and we have page
8482 * locked, the only possible way to hold a spinlock is from the endio
8483 * function to clear page writeback.
8484 *
8485 * Here we just acquire the spinlock so that all existing callers
8486 * should exit and we're safe to release/invalidate the page.
8487 */
8488 spin_lock_irq(&subpage->lock);
8489 spin_unlock_irq(&subpage->lock);
8490 }
8491
__btrfs_releasepage(struct page * page,gfp_t gfp_flags)8492 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8493 {
8494 int ret = try_release_extent_mapping(page, gfp_flags);
8495
8496 if (ret == 1) {
8497 wait_subpage_spinlock(page);
8498 clear_page_extent_mapped(page);
8499 }
8500 return ret;
8501 }
8502
btrfs_releasepage(struct page * page,gfp_t gfp_flags)8503 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8504 {
8505 if (PageWriteback(page) || PageDirty(page))
8506 return 0;
8507 return __btrfs_releasepage(page, gfp_flags);
8508 }
8509
8510 #ifdef CONFIG_MIGRATION
btrfs_migratepage(struct address_space * mapping,struct page * newpage,struct page * page,enum migrate_mode mode)8511 static int btrfs_migratepage(struct address_space *mapping,
8512 struct page *newpage, struct page *page,
8513 enum migrate_mode mode)
8514 {
8515 int ret;
8516
8517 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8518 if (ret != MIGRATEPAGE_SUCCESS)
8519 return ret;
8520
8521 if (page_has_private(page))
8522 attach_page_private(newpage, detach_page_private(page));
8523
8524 if (PageOrdered(page)) {
8525 ClearPageOrdered(page);
8526 SetPageOrdered(newpage);
8527 }
8528
8529 if (mode != MIGRATE_SYNC_NO_COPY)
8530 migrate_page_copy(newpage, page);
8531 else
8532 migrate_page_states(newpage, page);
8533 return MIGRATEPAGE_SUCCESS;
8534 }
8535 #endif
8536
btrfs_invalidatepage(struct page * page,unsigned int offset,unsigned int length)8537 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8538 unsigned int length)
8539 {
8540 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8541 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8542 struct extent_io_tree *tree = &inode->io_tree;
8543 struct extent_state *cached_state = NULL;
8544 u64 page_start = page_offset(page);
8545 u64 page_end = page_start + PAGE_SIZE - 1;
8546 u64 cur;
8547 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8548
8549 /*
8550 * We have page locked so no new ordered extent can be created on this
8551 * page, nor bio can be submitted for this page.
8552 *
8553 * But already submitted bio can still be finished on this page.
8554 * Furthermore, endio function won't skip page which has Ordered
8555 * (Private2) already cleared, so it's possible for endio and
8556 * invalidatepage to do the same ordered extent accounting twice
8557 * on one page.
8558 *
8559 * So here we wait for any submitted bios to finish, so that we won't
8560 * do double ordered extent accounting on the same page.
8561 */
8562 wait_on_page_writeback(page);
8563 wait_subpage_spinlock(page);
8564
8565 /*
8566 * For subpage case, we have call sites like
8567 * btrfs_punch_hole_lock_range() which passes range not aligned to
8568 * sectorsize.
8569 * If the range doesn't cover the full page, we don't need to and
8570 * shouldn't clear page extent mapped, as page->private can still
8571 * record subpage dirty bits for other part of the range.
8572 *
8573 * For cases that can invalidate the full even the range doesn't
8574 * cover the full page, like invalidating the last page, we're
8575 * still safe to wait for ordered extent to finish.
8576 */
8577 if (!(offset == 0 && length == PAGE_SIZE)) {
8578 btrfs_releasepage(page, GFP_NOFS);
8579 return;
8580 }
8581
8582 if (!inode_evicting)
8583 lock_extent_bits(tree, page_start, page_end, &cached_state);
8584
8585 cur = page_start;
8586 while (cur < page_end) {
8587 struct btrfs_ordered_extent *ordered;
8588 bool delete_states;
8589 u64 range_end;
8590 u32 range_len;
8591
8592 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8593 page_end + 1 - cur);
8594 if (!ordered) {
8595 range_end = page_end;
8596 /*
8597 * No ordered extent covering this range, we are safe
8598 * to delete all extent states in the range.
8599 */
8600 delete_states = true;
8601 goto next;
8602 }
8603 if (ordered->file_offset > cur) {
8604 /*
8605 * There is a range between [cur, oe->file_offset) not
8606 * covered by any ordered extent.
8607 * We are safe to delete all extent states, and handle
8608 * the ordered extent in the next iteration.
8609 */
8610 range_end = ordered->file_offset - 1;
8611 delete_states = true;
8612 goto next;
8613 }
8614
8615 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8616 page_end);
8617 ASSERT(range_end + 1 - cur < U32_MAX);
8618 range_len = range_end + 1 - cur;
8619 if (!btrfs_page_test_ordered(fs_info, page, cur, range_len)) {
8620 /*
8621 * If Ordered (Private2) is cleared, it means endio has
8622 * already been executed for the range.
8623 * We can't delete the extent states as
8624 * btrfs_finish_ordered_io() may still use some of them.
8625 */
8626 delete_states = false;
8627 goto next;
8628 }
8629 btrfs_page_clear_ordered(fs_info, page, cur, range_len);
8630
8631 /*
8632 * IO on this page will never be started, so we need to account
8633 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8634 * here, must leave that up for the ordered extent completion.
8635 *
8636 * This will also unlock the range for incoming
8637 * btrfs_finish_ordered_io().
8638 */
8639 if (!inode_evicting)
8640 clear_extent_bit(tree, cur, range_end,
8641 EXTENT_DELALLOC |
8642 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8643 EXTENT_DEFRAG, 1, 0, &cached_state);
8644
8645 spin_lock_irq(&inode->ordered_tree.lock);
8646 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8647 ordered->truncated_len = min(ordered->truncated_len,
8648 cur - ordered->file_offset);
8649 spin_unlock_irq(&inode->ordered_tree.lock);
8650
8651 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8652 cur, range_end + 1 - cur)) {
8653 btrfs_finish_ordered_io(ordered);
8654 /*
8655 * The ordered extent has finished, now we're again
8656 * safe to delete all extent states of the range.
8657 */
8658 delete_states = true;
8659 } else {
8660 /*
8661 * btrfs_finish_ordered_io() will get executed by endio
8662 * of other pages, thus we can't delete extent states
8663 * anymore
8664 */
8665 delete_states = false;
8666 }
8667 next:
8668 if (ordered)
8669 btrfs_put_ordered_extent(ordered);
8670 /*
8671 * Qgroup reserved space handler
8672 * Sector(s) here will be either:
8673 *
8674 * 1) Already written to disk or bio already finished
8675 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8676 * Qgroup will be handled by its qgroup_record then.
8677 * btrfs_qgroup_free_data() call will do nothing here.
8678 *
8679 * 2) Not written to disk yet
8680 * Then btrfs_qgroup_free_data() call will clear the
8681 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8682 * reserved data space.
8683 * Since the IO will never happen for this page.
8684 */
8685 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8686 if (!inode_evicting) {
8687 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8688 EXTENT_DELALLOC | EXTENT_UPTODATE |
8689 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8690 delete_states, &cached_state);
8691 }
8692 cur = range_end + 1;
8693 }
8694 /*
8695 * We have iterated through all ordered extents of the page, the page
8696 * should not have Ordered (Private2) anymore, or the above iteration
8697 * did something wrong.
8698 */
8699 ASSERT(!PageOrdered(page));
8700 if (!inode_evicting)
8701 __btrfs_releasepage(page, GFP_NOFS);
8702 ClearPageChecked(page);
8703 clear_page_extent_mapped(page);
8704 }
8705
8706 /*
8707 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8708 * called from a page fault handler when a page is first dirtied. Hence we must
8709 * be careful to check for EOF conditions here. We set the page up correctly
8710 * for a written page which means we get ENOSPC checking when writing into
8711 * holes and correct delalloc and unwritten extent mapping on filesystems that
8712 * support these features.
8713 *
8714 * We are not allowed to take the i_mutex here so we have to play games to
8715 * protect against truncate races as the page could now be beyond EOF. Because
8716 * truncate_setsize() writes the inode size before removing pages, once we have
8717 * the page lock we can determine safely if the page is beyond EOF. If it is not
8718 * beyond EOF, then the page is guaranteed safe against truncation until we
8719 * unlock the page.
8720 */
btrfs_page_mkwrite(struct vm_fault * vmf)8721 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8722 {
8723 struct page *page = vmf->page;
8724 struct inode *inode = file_inode(vmf->vma->vm_file);
8725 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8726 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8727 struct btrfs_ordered_extent *ordered;
8728 struct extent_state *cached_state = NULL;
8729 struct extent_changeset *data_reserved = NULL;
8730 unsigned long zero_start;
8731 loff_t size;
8732 vm_fault_t ret;
8733 int ret2;
8734 int reserved = 0;
8735 u64 reserved_space;
8736 u64 page_start;
8737 u64 page_end;
8738 u64 end;
8739
8740 reserved_space = PAGE_SIZE;
8741
8742 sb_start_pagefault(inode->i_sb);
8743 page_start = page_offset(page);
8744 page_end = page_start + PAGE_SIZE - 1;
8745 end = page_end;
8746
8747 /*
8748 * Reserving delalloc space after obtaining the page lock can lead to
8749 * deadlock. For example, if a dirty page is locked by this function
8750 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8751 * dirty page write out, then the btrfs_writepage() function could
8752 * end up waiting indefinitely to get a lock on the page currently
8753 * being processed by btrfs_page_mkwrite() function.
8754 */
8755 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8756 page_start, reserved_space);
8757 if (!ret2) {
8758 ret2 = file_update_time(vmf->vma->vm_file);
8759 reserved = 1;
8760 }
8761 if (ret2) {
8762 ret = vmf_error(ret2);
8763 if (reserved)
8764 goto out;
8765 goto out_noreserve;
8766 }
8767
8768 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8769 again:
8770 down_read(&BTRFS_I(inode)->i_mmap_lock);
8771 lock_page(page);
8772 size = i_size_read(inode);
8773
8774 if ((page->mapping != inode->i_mapping) ||
8775 (page_start >= size)) {
8776 /* page got truncated out from underneath us */
8777 goto out_unlock;
8778 }
8779 wait_on_page_writeback(page);
8780
8781 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8782 ret2 = set_page_extent_mapped(page);
8783 if (ret2 < 0) {
8784 ret = vmf_error(ret2);
8785 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8786 goto out_unlock;
8787 }
8788
8789 /*
8790 * we can't set the delalloc bits if there are pending ordered
8791 * extents. Drop our locks and wait for them to finish
8792 */
8793 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8794 PAGE_SIZE);
8795 if (ordered) {
8796 unlock_extent_cached(io_tree, page_start, page_end,
8797 &cached_state);
8798 unlock_page(page);
8799 up_read(&BTRFS_I(inode)->i_mmap_lock);
8800 btrfs_start_ordered_extent(ordered, 1);
8801 btrfs_put_ordered_extent(ordered);
8802 goto again;
8803 }
8804
8805 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8806 reserved_space = round_up(size - page_start,
8807 fs_info->sectorsize);
8808 if (reserved_space < PAGE_SIZE) {
8809 end = page_start + reserved_space - 1;
8810 btrfs_delalloc_release_space(BTRFS_I(inode),
8811 data_reserved, page_start,
8812 PAGE_SIZE - reserved_space, true);
8813 }
8814 }
8815
8816 /*
8817 * page_mkwrite gets called when the page is firstly dirtied after it's
8818 * faulted in, but write(2) could also dirty a page and set delalloc
8819 * bits, thus in this case for space account reason, we still need to
8820 * clear any delalloc bits within this page range since we have to
8821 * reserve data&meta space before lock_page() (see above comments).
8822 */
8823 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8824 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8825 EXTENT_DEFRAG, 0, 0, &cached_state);
8826
8827 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8828 &cached_state);
8829 if (ret2) {
8830 unlock_extent_cached(io_tree, page_start, page_end,
8831 &cached_state);
8832 ret = VM_FAULT_SIGBUS;
8833 goto out_unlock;
8834 }
8835
8836 /* page is wholly or partially inside EOF */
8837 if (page_start + PAGE_SIZE > size)
8838 zero_start = offset_in_page(size);
8839 else
8840 zero_start = PAGE_SIZE;
8841
8842 if (zero_start != PAGE_SIZE) {
8843 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8844 flush_dcache_page(page);
8845 }
8846 ClearPageChecked(page);
8847 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8848 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8849
8850 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8851
8852 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8853 up_read(&BTRFS_I(inode)->i_mmap_lock);
8854
8855 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8856 sb_end_pagefault(inode->i_sb);
8857 extent_changeset_free(data_reserved);
8858 return VM_FAULT_LOCKED;
8859
8860 out_unlock:
8861 unlock_page(page);
8862 up_read(&BTRFS_I(inode)->i_mmap_lock);
8863 out:
8864 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8865 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8866 reserved_space, (ret != 0));
8867 out_noreserve:
8868 sb_end_pagefault(inode->i_sb);
8869 extent_changeset_free(data_reserved);
8870 return ret;
8871 }
8872
btrfs_truncate(struct inode * inode,bool skip_writeback)8873 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8874 {
8875 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8876 struct btrfs_root *root = BTRFS_I(inode)->root;
8877 struct btrfs_block_rsv *rsv;
8878 int ret;
8879 struct btrfs_trans_handle *trans;
8880 u64 mask = fs_info->sectorsize - 1;
8881 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8882 u64 extents_found = 0;
8883
8884 if (!skip_writeback) {
8885 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8886 (u64)-1);
8887 if (ret)
8888 return ret;
8889 }
8890
8891 /*
8892 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8893 * things going on here:
8894 *
8895 * 1) We need to reserve space to update our inode.
8896 *
8897 * 2) We need to have something to cache all the space that is going to
8898 * be free'd up by the truncate operation, but also have some slack
8899 * space reserved in case it uses space during the truncate (thank you
8900 * very much snapshotting).
8901 *
8902 * And we need these to be separate. The fact is we can use a lot of
8903 * space doing the truncate, and we have no earthly idea how much space
8904 * we will use, so we need the truncate reservation to be separate so it
8905 * doesn't end up using space reserved for updating the inode. We also
8906 * need to be able to stop the transaction and start a new one, which
8907 * means we need to be able to update the inode several times, and we
8908 * have no idea of knowing how many times that will be, so we can't just
8909 * reserve 1 item for the entirety of the operation, so that has to be
8910 * done separately as well.
8911 *
8912 * So that leaves us with
8913 *
8914 * 1) rsv - for the truncate reservation, which we will steal from the
8915 * transaction reservation.
8916 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8917 * updating the inode.
8918 */
8919 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8920 if (!rsv)
8921 return -ENOMEM;
8922 rsv->size = min_size;
8923 rsv->failfast = 1;
8924
8925 /*
8926 * 1 for the truncate slack space
8927 * 1 for updating the inode.
8928 */
8929 trans = btrfs_start_transaction(root, 2);
8930 if (IS_ERR(trans)) {
8931 ret = PTR_ERR(trans);
8932 goto out;
8933 }
8934
8935 /* Migrate the slack space for the truncate to our reserve */
8936 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8937 min_size, false);
8938 BUG_ON(ret);
8939
8940 trans->block_rsv = rsv;
8941
8942 while (1) {
8943 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
8944 inode->i_size,
8945 BTRFS_EXTENT_DATA_KEY,
8946 &extents_found);
8947 trans->block_rsv = &fs_info->trans_block_rsv;
8948 if (ret != -ENOSPC && ret != -EAGAIN)
8949 break;
8950
8951 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8952 if (ret)
8953 break;
8954
8955 btrfs_end_transaction(trans);
8956 btrfs_btree_balance_dirty(fs_info);
8957
8958 trans = btrfs_start_transaction(root, 2);
8959 if (IS_ERR(trans)) {
8960 ret = PTR_ERR(trans);
8961 trans = NULL;
8962 break;
8963 }
8964
8965 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8966 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8967 rsv, min_size, false);
8968 BUG_ON(ret); /* shouldn't happen */
8969 trans->block_rsv = rsv;
8970 }
8971
8972 /*
8973 * We can't call btrfs_truncate_block inside a trans handle as we could
8974 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
8975 * we've truncated everything except the last little bit, and can do
8976 * btrfs_truncate_block and then update the disk_i_size.
8977 */
8978 if (ret == NEED_TRUNCATE_BLOCK) {
8979 btrfs_end_transaction(trans);
8980 btrfs_btree_balance_dirty(fs_info);
8981
8982 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8983 if (ret)
8984 goto out;
8985 trans = btrfs_start_transaction(root, 1);
8986 if (IS_ERR(trans)) {
8987 ret = PTR_ERR(trans);
8988 goto out;
8989 }
8990 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8991 }
8992
8993 if (trans) {
8994 int ret2;
8995
8996 trans->block_rsv = &fs_info->trans_block_rsv;
8997 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8998 if (ret2 && !ret)
8999 ret = ret2;
9000
9001 ret2 = btrfs_end_transaction(trans);
9002 if (ret2 && !ret)
9003 ret = ret2;
9004 btrfs_btree_balance_dirty(fs_info);
9005 }
9006 out:
9007 btrfs_free_block_rsv(fs_info, rsv);
9008 /*
9009 * So if we truncate and then write and fsync we normally would just
9010 * write the extents that changed, which is a problem if we need to
9011 * first truncate that entire inode. So set this flag so we write out
9012 * all of the extents in the inode to the sync log so we're completely
9013 * safe.
9014 *
9015 * If no extents were dropped or trimmed we don't need to force the next
9016 * fsync to truncate all the inode's items from the log and re-log them
9017 * all. This means the truncate operation did not change the file size,
9018 * or changed it to a smaller size but there was only an implicit hole
9019 * between the old i_size and the new i_size, and there were no prealloc
9020 * extents beyond i_size to drop.
9021 */
9022 if (extents_found > 0)
9023 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9024
9025 return ret;
9026 }
9027
9028 /*
9029 * create a new subvolume directory/inode (helper for the ioctl).
9030 */
btrfs_create_subvol_root(struct btrfs_trans_handle * trans,struct btrfs_root * new_root,struct btrfs_root * parent_root,struct user_namespace * mnt_userns)9031 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9032 struct btrfs_root *new_root,
9033 struct btrfs_root *parent_root,
9034 struct user_namespace *mnt_userns)
9035 {
9036 struct inode *inode;
9037 int err;
9038 u64 index = 0;
9039 u64 ino;
9040
9041 err = btrfs_get_free_objectid(new_root, &ino);
9042 if (err < 0)
9043 return err;
9044
9045 inode = btrfs_new_inode(trans, new_root, mnt_userns, NULL, "..", 2,
9046 ino, ino,
9047 S_IFDIR | (~current_umask() & S_IRWXUGO),
9048 &index);
9049 if (IS_ERR(inode))
9050 return PTR_ERR(inode);
9051 inode->i_op = &btrfs_dir_inode_operations;
9052 inode->i_fop = &btrfs_dir_file_operations;
9053
9054 set_nlink(inode, 1);
9055 btrfs_i_size_write(BTRFS_I(inode), 0);
9056 unlock_new_inode(inode);
9057
9058 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9059 if (err)
9060 btrfs_err(new_root->fs_info,
9061 "error inheriting subvolume %llu properties: %d",
9062 new_root->root_key.objectid, err);
9063
9064 err = btrfs_update_inode(trans, new_root, BTRFS_I(inode));
9065
9066 iput(inode);
9067 return err;
9068 }
9069
btrfs_alloc_inode(struct super_block * sb)9070 struct inode *btrfs_alloc_inode(struct super_block *sb)
9071 {
9072 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9073 struct btrfs_inode *ei;
9074 struct inode *inode;
9075
9076 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9077 if (!ei)
9078 return NULL;
9079
9080 ei->root = NULL;
9081 ei->generation = 0;
9082 ei->last_trans = 0;
9083 ei->last_sub_trans = 0;
9084 ei->logged_trans = 0;
9085 ei->delalloc_bytes = 0;
9086 ei->new_delalloc_bytes = 0;
9087 ei->defrag_bytes = 0;
9088 ei->disk_i_size = 0;
9089 ei->flags = 0;
9090 ei->ro_flags = 0;
9091 ei->csum_bytes = 0;
9092 ei->index_cnt = (u64)-1;
9093 ei->dir_index = 0;
9094 ei->last_unlink_trans = 0;
9095 ei->last_reflink_trans = 0;
9096 ei->last_log_commit = 0;
9097
9098 spin_lock_init(&ei->lock);
9099 ei->outstanding_extents = 0;
9100 if (sb->s_magic != BTRFS_TEST_MAGIC)
9101 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9102 BTRFS_BLOCK_RSV_DELALLOC);
9103 ei->runtime_flags = 0;
9104 ei->prop_compress = BTRFS_COMPRESS_NONE;
9105 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9106
9107 ei->delayed_node = NULL;
9108
9109 ei->i_otime.tv_sec = 0;
9110 ei->i_otime.tv_nsec = 0;
9111
9112 inode = &ei->vfs_inode;
9113 extent_map_tree_init(&ei->extent_tree);
9114 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
9115 extent_io_tree_init(fs_info, &ei->io_failure_tree,
9116 IO_TREE_INODE_IO_FAILURE, inode);
9117 extent_io_tree_init(fs_info, &ei->file_extent_tree,
9118 IO_TREE_INODE_FILE_EXTENT, inode);
9119 ei->io_tree.track_uptodate = true;
9120 ei->io_failure_tree.track_uptodate = true;
9121 atomic_set(&ei->sync_writers, 0);
9122 mutex_init(&ei->log_mutex);
9123 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9124 INIT_LIST_HEAD(&ei->delalloc_inodes);
9125 INIT_LIST_HEAD(&ei->delayed_iput);
9126 RB_CLEAR_NODE(&ei->rb_node);
9127 init_rwsem(&ei->i_mmap_lock);
9128
9129 return inode;
9130 }
9131
9132 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
btrfs_test_destroy_inode(struct inode * inode)9133 void btrfs_test_destroy_inode(struct inode *inode)
9134 {
9135 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9136 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9137 }
9138 #endif
9139
btrfs_free_inode(struct inode * inode)9140 void btrfs_free_inode(struct inode *inode)
9141 {
9142 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9143 }
9144
btrfs_destroy_inode(struct inode * vfs_inode)9145 void btrfs_destroy_inode(struct inode *vfs_inode)
9146 {
9147 struct btrfs_ordered_extent *ordered;
9148 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
9149 struct btrfs_root *root = inode->root;
9150
9151 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
9152 WARN_ON(vfs_inode->i_data.nrpages);
9153 WARN_ON(inode->block_rsv.reserved);
9154 WARN_ON(inode->block_rsv.size);
9155 WARN_ON(inode->outstanding_extents);
9156 WARN_ON(inode->delalloc_bytes);
9157 WARN_ON(inode->new_delalloc_bytes);
9158 WARN_ON(inode->csum_bytes);
9159 WARN_ON(inode->defrag_bytes);
9160
9161 /*
9162 * This can happen where we create an inode, but somebody else also
9163 * created the same inode and we need to destroy the one we already
9164 * created.
9165 */
9166 if (!root)
9167 return;
9168
9169 while (1) {
9170 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9171 if (!ordered)
9172 break;
9173 else {
9174 btrfs_err(root->fs_info,
9175 "found ordered extent %llu %llu on inode cleanup",
9176 ordered->file_offset, ordered->num_bytes);
9177 btrfs_remove_ordered_extent(inode, ordered);
9178 btrfs_put_ordered_extent(ordered);
9179 btrfs_put_ordered_extent(ordered);
9180 }
9181 }
9182 btrfs_qgroup_check_reserved_leak(inode);
9183 inode_tree_del(inode);
9184 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9185 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
9186 btrfs_put_root(inode->root);
9187 }
9188
btrfs_drop_inode(struct inode * inode)9189 int btrfs_drop_inode(struct inode *inode)
9190 {
9191 struct btrfs_root *root = BTRFS_I(inode)->root;
9192
9193 if (root == NULL)
9194 return 1;
9195
9196 /* the snap/subvol tree is on deleting */
9197 if (btrfs_root_refs(&root->root_item) == 0)
9198 return 1;
9199 else
9200 return generic_drop_inode(inode);
9201 }
9202
init_once(void * foo)9203 static void init_once(void *foo)
9204 {
9205 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9206
9207 inode_init_once(&ei->vfs_inode);
9208 }
9209
btrfs_destroy_cachep(void)9210 void __cold btrfs_destroy_cachep(void)
9211 {
9212 /*
9213 * Make sure all delayed rcu free inodes are flushed before we
9214 * destroy cache.
9215 */
9216 rcu_barrier();
9217 kmem_cache_destroy(btrfs_inode_cachep);
9218 kmem_cache_destroy(btrfs_trans_handle_cachep);
9219 kmem_cache_destroy(btrfs_path_cachep);
9220 kmem_cache_destroy(btrfs_free_space_cachep);
9221 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
9222 }
9223
btrfs_init_cachep(void)9224 int __init btrfs_init_cachep(void)
9225 {
9226 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9227 sizeof(struct btrfs_inode), 0,
9228 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9229 init_once);
9230 if (!btrfs_inode_cachep)
9231 goto fail;
9232
9233 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9234 sizeof(struct btrfs_trans_handle), 0,
9235 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9236 if (!btrfs_trans_handle_cachep)
9237 goto fail;
9238
9239 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9240 sizeof(struct btrfs_path), 0,
9241 SLAB_MEM_SPREAD, NULL);
9242 if (!btrfs_path_cachep)
9243 goto fail;
9244
9245 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9246 sizeof(struct btrfs_free_space), 0,
9247 SLAB_MEM_SPREAD, NULL);
9248 if (!btrfs_free_space_cachep)
9249 goto fail;
9250
9251 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
9252 PAGE_SIZE, PAGE_SIZE,
9253 SLAB_MEM_SPREAD, NULL);
9254 if (!btrfs_free_space_bitmap_cachep)
9255 goto fail;
9256
9257 return 0;
9258 fail:
9259 btrfs_destroy_cachep();
9260 return -ENOMEM;
9261 }
9262
btrfs_getattr(struct user_namespace * mnt_userns,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int flags)9263 static int btrfs_getattr(struct user_namespace *mnt_userns,
9264 const struct path *path, struct kstat *stat,
9265 u32 request_mask, unsigned int flags)
9266 {
9267 u64 delalloc_bytes;
9268 u64 inode_bytes;
9269 struct inode *inode = d_inode(path->dentry);
9270 u32 blocksize = inode->i_sb->s_blocksize;
9271 u32 bi_flags = BTRFS_I(inode)->flags;
9272 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
9273
9274 stat->result_mask |= STATX_BTIME;
9275 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9276 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9277 if (bi_flags & BTRFS_INODE_APPEND)
9278 stat->attributes |= STATX_ATTR_APPEND;
9279 if (bi_flags & BTRFS_INODE_COMPRESS)
9280 stat->attributes |= STATX_ATTR_COMPRESSED;
9281 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9282 stat->attributes |= STATX_ATTR_IMMUTABLE;
9283 if (bi_flags & BTRFS_INODE_NODUMP)
9284 stat->attributes |= STATX_ATTR_NODUMP;
9285 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
9286 stat->attributes |= STATX_ATTR_VERITY;
9287
9288 stat->attributes_mask |= (STATX_ATTR_APPEND |
9289 STATX_ATTR_COMPRESSED |
9290 STATX_ATTR_IMMUTABLE |
9291 STATX_ATTR_NODUMP);
9292
9293 generic_fillattr(mnt_userns, inode, stat);
9294 stat->dev = BTRFS_I(inode)->root->anon_dev;
9295
9296 spin_lock(&BTRFS_I(inode)->lock);
9297 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9298 inode_bytes = inode_get_bytes(inode);
9299 spin_unlock(&BTRFS_I(inode)->lock);
9300 stat->blocks = (ALIGN(inode_bytes, blocksize) +
9301 ALIGN(delalloc_bytes, blocksize)) >> 9;
9302 return 0;
9303 }
9304
btrfs_rename_exchange(struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry)9305 static int btrfs_rename_exchange(struct inode *old_dir,
9306 struct dentry *old_dentry,
9307 struct inode *new_dir,
9308 struct dentry *new_dentry)
9309 {
9310 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9311 struct btrfs_trans_handle *trans;
9312 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9313 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9314 struct inode *new_inode = new_dentry->d_inode;
9315 struct inode *old_inode = old_dentry->d_inode;
9316 struct timespec64 ctime = current_time(old_inode);
9317 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9318 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9319 u64 old_idx = 0;
9320 u64 new_idx = 0;
9321 int ret;
9322 int ret2;
9323 bool root_log_pinned = false;
9324 bool dest_log_pinned = false;
9325 bool need_abort = false;
9326
9327 /*
9328 * For non-subvolumes allow exchange only within one subvolume, in the
9329 * same inode namespace. Two subvolumes (represented as directory) can
9330 * be exchanged as they're a logical link and have a fixed inode number.
9331 */
9332 if (root != dest &&
9333 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
9334 new_ino != BTRFS_FIRST_FREE_OBJECTID))
9335 return -EXDEV;
9336
9337 /* close the race window with snapshot create/destroy ioctl */
9338 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9339 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9340 down_read(&fs_info->subvol_sem);
9341
9342 /*
9343 * We want to reserve the absolute worst case amount of items. So if
9344 * both inodes are subvols and we need to unlink them then that would
9345 * require 4 item modifications, but if they are both normal inodes it
9346 * would require 5 item modifications, so we'll assume their normal
9347 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9348 * should cover the worst case number of items we'll modify.
9349 */
9350 trans = btrfs_start_transaction(root, 12);
9351 if (IS_ERR(trans)) {
9352 ret = PTR_ERR(trans);
9353 goto out_notrans;
9354 }
9355
9356 if (dest != root) {
9357 ret = btrfs_record_root_in_trans(trans, dest);
9358 if (ret)
9359 goto out_fail;
9360 }
9361
9362 /*
9363 * We need to find a free sequence number both in the source and
9364 * in the destination directory for the exchange.
9365 */
9366 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9367 if (ret)
9368 goto out_fail;
9369 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9370 if (ret)
9371 goto out_fail;
9372
9373 BTRFS_I(old_inode)->dir_index = 0ULL;
9374 BTRFS_I(new_inode)->dir_index = 0ULL;
9375
9376 /* Reference for the source. */
9377 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9378 /* force full log commit if subvolume involved. */
9379 btrfs_set_log_full_commit(trans);
9380 } else {
9381 ret = btrfs_insert_inode_ref(trans, dest,
9382 new_dentry->d_name.name,
9383 new_dentry->d_name.len,
9384 old_ino,
9385 btrfs_ino(BTRFS_I(new_dir)),
9386 old_idx);
9387 if (ret)
9388 goto out_fail;
9389 need_abort = true;
9390 }
9391
9392 /* And now for the dest. */
9393 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9394 /* force full log commit if subvolume involved. */
9395 btrfs_set_log_full_commit(trans);
9396 } else {
9397 ret = btrfs_insert_inode_ref(trans, root,
9398 old_dentry->d_name.name,
9399 old_dentry->d_name.len,
9400 new_ino,
9401 btrfs_ino(BTRFS_I(old_dir)),
9402 new_idx);
9403 if (ret) {
9404 if (need_abort)
9405 btrfs_abort_transaction(trans, ret);
9406 goto out_fail;
9407 }
9408 }
9409
9410 /* Update inode version and ctime/mtime. */
9411 inode_inc_iversion(old_dir);
9412 inode_inc_iversion(new_dir);
9413 inode_inc_iversion(old_inode);
9414 inode_inc_iversion(new_inode);
9415 old_dir->i_ctime = old_dir->i_mtime = ctime;
9416 new_dir->i_ctime = new_dir->i_mtime = ctime;
9417 old_inode->i_ctime = ctime;
9418 new_inode->i_ctime = ctime;
9419
9420 if (old_dentry->d_parent != new_dentry->d_parent) {
9421 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9422 BTRFS_I(old_inode), 1);
9423 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9424 BTRFS_I(new_inode), 1);
9425 }
9426
9427 /*
9428 * Now pin the logs of the roots. We do it to ensure that no other task
9429 * can sync the logs while we are in progress with the rename, because
9430 * that could result in an inconsistency in case any of the inodes that
9431 * are part of this rename operation were logged before.
9432 *
9433 * We pin the logs even if at this precise moment none of the inodes was
9434 * logged before. This is because right after we checked for that, some
9435 * other task fsyncing some other inode not involved with this rename
9436 * operation could log that one of our inodes exists.
9437 *
9438 * We don't need to pin the logs before the above calls to
9439 * btrfs_insert_inode_ref(), since those don't ever need to change a log.
9440 */
9441 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9442 btrfs_pin_log_trans(root);
9443 root_log_pinned = true;
9444 }
9445 if (new_ino != BTRFS_FIRST_FREE_OBJECTID) {
9446 btrfs_pin_log_trans(dest);
9447 dest_log_pinned = true;
9448 }
9449
9450 /* src is a subvolume */
9451 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9452 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9453 } else { /* src is an inode */
9454 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9455 BTRFS_I(old_dentry->d_inode),
9456 old_dentry->d_name.name,
9457 old_dentry->d_name.len);
9458 if (!ret)
9459 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9460 }
9461 if (ret) {
9462 btrfs_abort_transaction(trans, ret);
9463 goto out_fail;
9464 }
9465
9466 /* dest is a subvolume */
9467 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9468 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9469 } else { /* dest is an inode */
9470 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9471 BTRFS_I(new_dentry->d_inode),
9472 new_dentry->d_name.name,
9473 new_dentry->d_name.len);
9474 if (!ret)
9475 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9476 }
9477 if (ret) {
9478 btrfs_abort_transaction(trans, ret);
9479 goto out_fail;
9480 }
9481
9482 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9483 new_dentry->d_name.name,
9484 new_dentry->d_name.len, 0, old_idx);
9485 if (ret) {
9486 btrfs_abort_transaction(trans, ret);
9487 goto out_fail;
9488 }
9489
9490 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9491 old_dentry->d_name.name,
9492 old_dentry->d_name.len, 0, new_idx);
9493 if (ret) {
9494 btrfs_abort_transaction(trans, ret);
9495 goto out_fail;
9496 }
9497
9498 if (old_inode->i_nlink == 1)
9499 BTRFS_I(old_inode)->dir_index = old_idx;
9500 if (new_inode->i_nlink == 1)
9501 BTRFS_I(new_inode)->dir_index = new_idx;
9502
9503 if (root_log_pinned) {
9504 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9505 new_dentry->d_parent);
9506 btrfs_end_log_trans(root);
9507 root_log_pinned = false;
9508 }
9509 if (dest_log_pinned) {
9510 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9511 old_dentry->d_parent);
9512 btrfs_end_log_trans(dest);
9513 dest_log_pinned = false;
9514 }
9515 out_fail:
9516 /*
9517 * If we have pinned a log and an error happened, we unpin tasks
9518 * trying to sync the log and force them to fallback to a transaction
9519 * commit if the log currently contains any of the inodes involved in
9520 * this rename operation (to ensure we do not persist a log with an
9521 * inconsistent state for any of these inodes or leading to any
9522 * inconsistencies when replayed). If the transaction was aborted, the
9523 * abortion reason is propagated to userspace when attempting to commit
9524 * the transaction. If the log does not contain any of these inodes, we
9525 * allow the tasks to sync it.
9526 */
9527 if (ret && (root_log_pinned || dest_log_pinned)) {
9528 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9529 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9530 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9531 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))
9532 btrfs_set_log_full_commit(trans);
9533
9534 if (root_log_pinned) {
9535 btrfs_end_log_trans(root);
9536 root_log_pinned = false;
9537 }
9538 if (dest_log_pinned) {
9539 btrfs_end_log_trans(dest);
9540 dest_log_pinned = false;
9541 }
9542 }
9543 ret2 = btrfs_end_transaction(trans);
9544 ret = ret ? ret : ret2;
9545 out_notrans:
9546 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9547 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9548 up_read(&fs_info->subvol_sem);
9549
9550 return ret;
9551 }
9552
btrfs_whiteout_for_rename(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct user_namespace * mnt_userns,struct inode * dir,struct dentry * dentry)9553 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9554 struct btrfs_root *root,
9555 struct user_namespace *mnt_userns,
9556 struct inode *dir,
9557 struct dentry *dentry)
9558 {
9559 int ret;
9560 struct inode *inode;
9561 u64 objectid;
9562 u64 index;
9563
9564 ret = btrfs_get_free_objectid(root, &objectid);
9565 if (ret)
9566 return ret;
9567
9568 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
9569 dentry->d_name.name,
9570 dentry->d_name.len,
9571 btrfs_ino(BTRFS_I(dir)),
9572 objectid,
9573 S_IFCHR | WHITEOUT_MODE,
9574 &index);
9575
9576 if (IS_ERR(inode)) {
9577 ret = PTR_ERR(inode);
9578 return ret;
9579 }
9580
9581 inode->i_op = &btrfs_special_inode_operations;
9582 init_special_inode(inode, inode->i_mode,
9583 WHITEOUT_DEV);
9584
9585 ret = btrfs_init_inode_security(trans, inode, dir,
9586 &dentry->d_name);
9587 if (ret)
9588 goto out;
9589
9590 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9591 BTRFS_I(inode), 0, index);
9592 if (ret)
9593 goto out;
9594
9595 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9596 out:
9597 unlock_new_inode(inode);
9598 if (ret)
9599 inode_dec_link_count(inode);
9600 iput(inode);
9601
9602 return ret;
9603 }
9604
btrfs_rename(struct user_namespace * mnt_userns,struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry,unsigned int flags)9605 static int btrfs_rename(struct user_namespace *mnt_userns,
9606 struct inode *old_dir, struct dentry *old_dentry,
9607 struct inode *new_dir, struct dentry *new_dentry,
9608 unsigned int flags)
9609 {
9610 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9611 struct btrfs_trans_handle *trans;
9612 unsigned int trans_num_items;
9613 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9614 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9615 struct inode *new_inode = d_inode(new_dentry);
9616 struct inode *old_inode = d_inode(old_dentry);
9617 u64 index = 0;
9618 int ret;
9619 int ret2;
9620 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9621 bool log_pinned = false;
9622
9623 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9624 return -EPERM;
9625
9626 /* we only allow rename subvolume link between subvolumes */
9627 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9628 return -EXDEV;
9629
9630 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9631 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9632 return -ENOTEMPTY;
9633
9634 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9635 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9636 return -ENOTEMPTY;
9637
9638
9639 /* check for collisions, even if the name isn't there */
9640 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9641 new_dentry->d_name.name,
9642 new_dentry->d_name.len);
9643
9644 if (ret) {
9645 if (ret == -EEXIST) {
9646 /* we shouldn't get
9647 * eexist without a new_inode */
9648 if (WARN_ON(!new_inode)) {
9649 return ret;
9650 }
9651 } else {
9652 /* maybe -EOVERFLOW */
9653 return ret;
9654 }
9655 }
9656 ret = 0;
9657
9658 /*
9659 * we're using rename to replace one file with another. Start IO on it
9660 * now so we don't add too much work to the end of the transaction
9661 */
9662 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9663 filemap_flush(old_inode->i_mapping);
9664
9665 /* close the racy window with snapshot create/destroy ioctl */
9666 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9667 down_read(&fs_info->subvol_sem);
9668 /*
9669 * We want to reserve the absolute worst case amount of items. So if
9670 * both inodes are subvols and we need to unlink them then that would
9671 * require 4 item modifications, but if they are both normal inodes it
9672 * would require 5 item modifications, so we'll assume they are normal
9673 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9674 * should cover the worst case number of items we'll modify.
9675 * If our rename has the whiteout flag, we need more 5 units for the
9676 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9677 * when selinux is enabled).
9678 */
9679 trans_num_items = 11;
9680 if (flags & RENAME_WHITEOUT)
9681 trans_num_items += 5;
9682 trans = btrfs_start_transaction(root, trans_num_items);
9683 if (IS_ERR(trans)) {
9684 ret = PTR_ERR(trans);
9685 goto out_notrans;
9686 }
9687
9688 if (dest != root) {
9689 ret = btrfs_record_root_in_trans(trans, dest);
9690 if (ret)
9691 goto out_fail;
9692 }
9693
9694 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9695 if (ret)
9696 goto out_fail;
9697
9698 BTRFS_I(old_inode)->dir_index = 0ULL;
9699 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9700 /* force full log commit if subvolume involved. */
9701 btrfs_set_log_full_commit(trans);
9702 } else {
9703 ret = btrfs_insert_inode_ref(trans, dest,
9704 new_dentry->d_name.name,
9705 new_dentry->d_name.len,
9706 old_ino,
9707 btrfs_ino(BTRFS_I(new_dir)), index);
9708 if (ret)
9709 goto out_fail;
9710 }
9711
9712 inode_inc_iversion(old_dir);
9713 inode_inc_iversion(new_dir);
9714 inode_inc_iversion(old_inode);
9715 old_dir->i_ctime = old_dir->i_mtime =
9716 new_dir->i_ctime = new_dir->i_mtime =
9717 old_inode->i_ctime = current_time(old_dir);
9718
9719 if (old_dentry->d_parent != new_dentry->d_parent)
9720 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9721 BTRFS_I(old_inode), 1);
9722
9723 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9724 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9725 } else {
9726 /*
9727 * Now pin the log. We do it to ensure that no other task can
9728 * sync the log while we are in progress with the rename, as
9729 * that could result in an inconsistency in case any of the
9730 * inodes that are part of this rename operation were logged
9731 * before.
9732 *
9733 * We pin the log even if at this precise moment none of the
9734 * inodes was logged before. This is because right after we
9735 * checked for that, some other task fsyncing some other inode
9736 * not involved with this rename operation could log that one of
9737 * our inodes exists.
9738 *
9739 * We don't need to pin the logs before the above call to
9740 * btrfs_insert_inode_ref(), since that does not need to change
9741 * a log.
9742 */
9743 btrfs_pin_log_trans(root);
9744 log_pinned = true;
9745 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9746 BTRFS_I(d_inode(old_dentry)),
9747 old_dentry->d_name.name,
9748 old_dentry->d_name.len);
9749 if (!ret)
9750 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9751 }
9752 if (ret) {
9753 btrfs_abort_transaction(trans, ret);
9754 goto out_fail;
9755 }
9756
9757 if (new_inode) {
9758 inode_inc_iversion(new_inode);
9759 new_inode->i_ctime = current_time(new_inode);
9760 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9761 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9762 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9763 BUG_ON(new_inode->i_nlink == 0);
9764 } else {
9765 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9766 BTRFS_I(d_inode(new_dentry)),
9767 new_dentry->d_name.name,
9768 new_dentry->d_name.len);
9769 }
9770 if (!ret && new_inode->i_nlink == 0)
9771 ret = btrfs_orphan_add(trans,
9772 BTRFS_I(d_inode(new_dentry)));
9773 if (ret) {
9774 btrfs_abort_transaction(trans, ret);
9775 goto out_fail;
9776 }
9777 }
9778
9779 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9780 new_dentry->d_name.name,
9781 new_dentry->d_name.len, 0, index);
9782 if (ret) {
9783 btrfs_abort_transaction(trans, ret);
9784 goto out_fail;
9785 }
9786
9787 if (old_inode->i_nlink == 1)
9788 BTRFS_I(old_inode)->dir_index = index;
9789
9790 if (log_pinned) {
9791 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9792 new_dentry->d_parent);
9793 btrfs_end_log_trans(root);
9794 log_pinned = false;
9795 }
9796
9797 if (flags & RENAME_WHITEOUT) {
9798 ret = btrfs_whiteout_for_rename(trans, root, mnt_userns,
9799 old_dir, old_dentry);
9800
9801 if (ret) {
9802 btrfs_abort_transaction(trans, ret);
9803 goto out_fail;
9804 }
9805 }
9806 out_fail:
9807 /*
9808 * If we have pinned the log and an error happened, we unpin tasks
9809 * trying to sync the log and force them to fallback to a transaction
9810 * commit if the log currently contains any of the inodes involved in
9811 * this rename operation (to ensure we do not persist a log with an
9812 * inconsistent state for any of these inodes or leading to any
9813 * inconsistencies when replayed). If the transaction was aborted, the
9814 * abortion reason is propagated to userspace when attempting to commit
9815 * the transaction. If the log does not contain any of these inodes, we
9816 * allow the tasks to sync it.
9817 */
9818 if (ret && log_pinned) {
9819 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9820 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9821 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9822 (new_inode &&
9823 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9824 btrfs_set_log_full_commit(trans);
9825
9826 btrfs_end_log_trans(root);
9827 log_pinned = false;
9828 }
9829 ret2 = btrfs_end_transaction(trans);
9830 ret = ret ? ret : ret2;
9831 out_notrans:
9832 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9833 up_read(&fs_info->subvol_sem);
9834
9835 return ret;
9836 }
9837
btrfs_rename2(struct user_namespace * mnt_userns,struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry,unsigned int flags)9838 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9839 struct dentry *old_dentry, struct inode *new_dir,
9840 struct dentry *new_dentry, unsigned int flags)
9841 {
9842 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9843 return -EINVAL;
9844
9845 if (flags & RENAME_EXCHANGE)
9846 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9847 new_dentry);
9848
9849 return btrfs_rename(mnt_userns, old_dir, old_dentry, new_dir,
9850 new_dentry, flags);
9851 }
9852
9853 struct btrfs_delalloc_work {
9854 struct inode *inode;
9855 struct completion completion;
9856 struct list_head list;
9857 struct btrfs_work work;
9858 };
9859
btrfs_run_delalloc_work(struct btrfs_work * work)9860 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9861 {
9862 struct btrfs_delalloc_work *delalloc_work;
9863 struct inode *inode;
9864
9865 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9866 work);
9867 inode = delalloc_work->inode;
9868 filemap_flush(inode->i_mapping);
9869 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9870 &BTRFS_I(inode)->runtime_flags))
9871 filemap_flush(inode->i_mapping);
9872
9873 iput(inode);
9874 complete(&delalloc_work->completion);
9875 }
9876
btrfs_alloc_delalloc_work(struct inode * inode)9877 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9878 {
9879 struct btrfs_delalloc_work *work;
9880
9881 work = kmalloc(sizeof(*work), GFP_NOFS);
9882 if (!work)
9883 return NULL;
9884
9885 init_completion(&work->completion);
9886 INIT_LIST_HEAD(&work->list);
9887 work->inode = inode;
9888 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9889
9890 return work;
9891 }
9892
9893 /*
9894 * some fairly slow code that needs optimization. This walks the list
9895 * of all the inodes with pending delalloc and forces them to disk.
9896 */
start_delalloc_inodes(struct btrfs_root * root,struct writeback_control * wbc,bool snapshot,bool in_reclaim_context)9897 static int start_delalloc_inodes(struct btrfs_root *root,
9898 struct writeback_control *wbc, bool snapshot,
9899 bool in_reclaim_context)
9900 {
9901 struct btrfs_inode *binode;
9902 struct inode *inode;
9903 struct btrfs_delalloc_work *work, *next;
9904 struct list_head works;
9905 struct list_head splice;
9906 int ret = 0;
9907 bool full_flush = wbc->nr_to_write == LONG_MAX;
9908
9909 INIT_LIST_HEAD(&works);
9910 INIT_LIST_HEAD(&splice);
9911
9912 mutex_lock(&root->delalloc_mutex);
9913 spin_lock(&root->delalloc_lock);
9914 list_splice_init(&root->delalloc_inodes, &splice);
9915 while (!list_empty(&splice)) {
9916 binode = list_entry(splice.next, struct btrfs_inode,
9917 delalloc_inodes);
9918
9919 list_move_tail(&binode->delalloc_inodes,
9920 &root->delalloc_inodes);
9921
9922 if (in_reclaim_context &&
9923 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9924 continue;
9925
9926 inode = igrab(&binode->vfs_inode);
9927 if (!inode) {
9928 cond_resched_lock(&root->delalloc_lock);
9929 continue;
9930 }
9931 spin_unlock(&root->delalloc_lock);
9932
9933 if (snapshot)
9934 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9935 &binode->runtime_flags);
9936 if (full_flush) {
9937 work = btrfs_alloc_delalloc_work(inode);
9938 if (!work) {
9939 iput(inode);
9940 ret = -ENOMEM;
9941 goto out;
9942 }
9943 list_add_tail(&work->list, &works);
9944 btrfs_queue_work(root->fs_info->flush_workers,
9945 &work->work);
9946 } else {
9947 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9948 btrfs_add_delayed_iput(inode);
9949 if (ret || wbc->nr_to_write <= 0)
9950 goto out;
9951 }
9952 cond_resched();
9953 spin_lock(&root->delalloc_lock);
9954 }
9955 spin_unlock(&root->delalloc_lock);
9956
9957 out:
9958 list_for_each_entry_safe(work, next, &works, list) {
9959 list_del_init(&work->list);
9960 wait_for_completion(&work->completion);
9961 kfree(work);
9962 }
9963
9964 if (!list_empty(&splice)) {
9965 spin_lock(&root->delalloc_lock);
9966 list_splice_tail(&splice, &root->delalloc_inodes);
9967 spin_unlock(&root->delalloc_lock);
9968 }
9969 mutex_unlock(&root->delalloc_mutex);
9970 return ret;
9971 }
9972
btrfs_start_delalloc_snapshot(struct btrfs_root * root,bool in_reclaim_context)9973 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9974 {
9975 struct writeback_control wbc = {
9976 .nr_to_write = LONG_MAX,
9977 .sync_mode = WB_SYNC_NONE,
9978 .range_start = 0,
9979 .range_end = LLONG_MAX,
9980 };
9981 struct btrfs_fs_info *fs_info = root->fs_info;
9982
9983 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9984 return -EROFS;
9985
9986 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9987 }
9988
btrfs_start_delalloc_roots(struct btrfs_fs_info * fs_info,long nr,bool in_reclaim_context)9989 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9990 bool in_reclaim_context)
9991 {
9992 struct writeback_control wbc = {
9993 .nr_to_write = nr,
9994 .sync_mode = WB_SYNC_NONE,
9995 .range_start = 0,
9996 .range_end = LLONG_MAX,
9997 };
9998 struct btrfs_root *root;
9999 struct list_head splice;
10000 int ret;
10001
10002 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10003 return -EROFS;
10004
10005 INIT_LIST_HEAD(&splice);
10006
10007 mutex_lock(&fs_info->delalloc_root_mutex);
10008 spin_lock(&fs_info->delalloc_root_lock);
10009 list_splice_init(&fs_info->delalloc_roots, &splice);
10010 while (!list_empty(&splice)) {
10011 /*
10012 * Reset nr_to_write here so we know that we're doing a full
10013 * flush.
10014 */
10015 if (nr == LONG_MAX)
10016 wbc.nr_to_write = LONG_MAX;
10017
10018 root = list_first_entry(&splice, struct btrfs_root,
10019 delalloc_root);
10020 root = btrfs_grab_root(root);
10021 BUG_ON(!root);
10022 list_move_tail(&root->delalloc_root,
10023 &fs_info->delalloc_roots);
10024 spin_unlock(&fs_info->delalloc_root_lock);
10025
10026 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
10027 btrfs_put_root(root);
10028 if (ret < 0 || wbc.nr_to_write <= 0)
10029 goto out;
10030 spin_lock(&fs_info->delalloc_root_lock);
10031 }
10032 spin_unlock(&fs_info->delalloc_root_lock);
10033
10034 ret = 0;
10035 out:
10036 if (!list_empty(&splice)) {
10037 spin_lock(&fs_info->delalloc_root_lock);
10038 list_splice_tail(&splice, &fs_info->delalloc_roots);
10039 spin_unlock(&fs_info->delalloc_root_lock);
10040 }
10041 mutex_unlock(&fs_info->delalloc_root_mutex);
10042 return ret;
10043 }
10044
btrfs_symlink(struct user_namespace * mnt_userns,struct inode * dir,struct dentry * dentry,const char * symname)10045 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
10046 struct dentry *dentry, const char *symname)
10047 {
10048 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10049 struct btrfs_trans_handle *trans;
10050 struct btrfs_root *root = BTRFS_I(dir)->root;
10051 struct btrfs_path *path;
10052 struct btrfs_key key;
10053 struct inode *inode = NULL;
10054 int err;
10055 u64 objectid;
10056 u64 index = 0;
10057 int name_len;
10058 int datasize;
10059 unsigned long ptr;
10060 struct btrfs_file_extent_item *ei;
10061 struct extent_buffer *leaf;
10062
10063 name_len = strlen(symname);
10064 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10065 return -ENAMETOOLONG;
10066
10067 /*
10068 * 2 items for inode item and ref
10069 * 2 items for dir items
10070 * 1 item for updating parent inode item
10071 * 1 item for the inline extent item
10072 * 1 item for xattr if selinux is on
10073 */
10074 trans = btrfs_start_transaction(root, 7);
10075 if (IS_ERR(trans))
10076 return PTR_ERR(trans);
10077
10078 err = btrfs_get_free_objectid(root, &objectid);
10079 if (err)
10080 goto out_unlock;
10081
10082 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
10083 dentry->d_name.name, dentry->d_name.len,
10084 btrfs_ino(BTRFS_I(dir)), objectid,
10085 S_IFLNK | S_IRWXUGO, &index);
10086 if (IS_ERR(inode)) {
10087 err = PTR_ERR(inode);
10088 inode = NULL;
10089 goto out_unlock;
10090 }
10091
10092 /*
10093 * If the active LSM wants to access the inode during
10094 * d_instantiate it needs these. Smack checks to see
10095 * if the filesystem supports xattrs by looking at the
10096 * ops vector.
10097 */
10098 inode->i_fop = &btrfs_file_operations;
10099 inode->i_op = &btrfs_file_inode_operations;
10100 inode->i_mapping->a_ops = &btrfs_aops;
10101
10102 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10103 if (err)
10104 goto out_unlock;
10105
10106 path = btrfs_alloc_path();
10107 if (!path) {
10108 err = -ENOMEM;
10109 goto out_unlock;
10110 }
10111 key.objectid = btrfs_ino(BTRFS_I(inode));
10112 key.offset = 0;
10113 key.type = BTRFS_EXTENT_DATA_KEY;
10114 datasize = btrfs_file_extent_calc_inline_size(name_len);
10115 err = btrfs_insert_empty_item(trans, root, path, &key,
10116 datasize);
10117 if (err) {
10118 btrfs_free_path(path);
10119 goto out_unlock;
10120 }
10121 leaf = path->nodes[0];
10122 ei = btrfs_item_ptr(leaf, path->slots[0],
10123 struct btrfs_file_extent_item);
10124 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10125 btrfs_set_file_extent_type(leaf, ei,
10126 BTRFS_FILE_EXTENT_INLINE);
10127 btrfs_set_file_extent_encryption(leaf, ei, 0);
10128 btrfs_set_file_extent_compression(leaf, ei, 0);
10129 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10130 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10131
10132 ptr = btrfs_file_extent_inline_start(ei);
10133 write_extent_buffer(leaf, symname, ptr, name_len);
10134 btrfs_mark_buffer_dirty(leaf);
10135 btrfs_free_path(path);
10136
10137 inode->i_op = &btrfs_symlink_inode_operations;
10138 inode_nohighmem(inode);
10139 inode_set_bytes(inode, name_len);
10140 btrfs_i_size_write(BTRFS_I(inode), name_len);
10141 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
10142 /*
10143 * Last step, add directory indexes for our symlink inode. This is the
10144 * last step to avoid extra cleanup of these indexes if an error happens
10145 * elsewhere above.
10146 */
10147 if (!err)
10148 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10149 BTRFS_I(inode), 0, index);
10150 if (err)
10151 goto out_unlock;
10152
10153 d_instantiate_new(dentry, inode);
10154
10155 out_unlock:
10156 btrfs_end_transaction(trans);
10157 if (err && inode) {
10158 inode_dec_link_count(inode);
10159 discard_new_inode(inode);
10160 }
10161 btrfs_btree_balance_dirty(fs_info);
10162 return err;
10163 }
10164
insert_prealloc_file_extent(struct btrfs_trans_handle * trans_in,struct btrfs_inode * inode,struct btrfs_key * ins,u64 file_offset)10165 static struct btrfs_trans_handle *insert_prealloc_file_extent(
10166 struct btrfs_trans_handle *trans_in,
10167 struct btrfs_inode *inode,
10168 struct btrfs_key *ins,
10169 u64 file_offset)
10170 {
10171 struct btrfs_file_extent_item stack_fi;
10172 struct btrfs_replace_extent_info extent_info;
10173 struct btrfs_trans_handle *trans = trans_in;
10174 struct btrfs_path *path;
10175 u64 start = ins->objectid;
10176 u64 len = ins->offset;
10177 int qgroup_released;
10178 int ret;
10179
10180 memset(&stack_fi, 0, sizeof(stack_fi));
10181
10182 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
10183 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
10184 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
10185 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
10186 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
10187 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
10188 /* Encryption and other encoding is reserved and all 0 */
10189
10190 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
10191 if (qgroup_released < 0)
10192 return ERR_PTR(qgroup_released);
10193
10194 if (trans) {
10195 ret = insert_reserved_file_extent(trans, inode,
10196 file_offset, &stack_fi,
10197 true, qgroup_released);
10198 if (ret)
10199 goto free_qgroup;
10200 return trans;
10201 }
10202
10203 extent_info.disk_offset = start;
10204 extent_info.disk_len = len;
10205 extent_info.data_offset = 0;
10206 extent_info.data_len = len;
10207 extent_info.file_offset = file_offset;
10208 extent_info.extent_buf = (char *)&stack_fi;
10209 extent_info.is_new_extent = true;
10210 extent_info.qgroup_reserved = qgroup_released;
10211 extent_info.insertions = 0;
10212
10213 path = btrfs_alloc_path();
10214 if (!path) {
10215 ret = -ENOMEM;
10216 goto free_qgroup;
10217 }
10218
10219 ret = btrfs_replace_file_extents(inode, path, file_offset,
10220 file_offset + len - 1, &extent_info,
10221 &trans);
10222 btrfs_free_path(path);
10223 if (ret)
10224 goto free_qgroup;
10225 return trans;
10226
10227 free_qgroup:
10228 /*
10229 * We have released qgroup data range at the beginning of the function,
10230 * and normally qgroup_released bytes will be freed when committing
10231 * transaction.
10232 * But if we error out early, we have to free what we have released
10233 * or we leak qgroup data reservation.
10234 */
10235 btrfs_qgroup_free_refroot(inode->root->fs_info,
10236 inode->root->root_key.objectid, qgroup_released,
10237 BTRFS_QGROUP_RSV_DATA);
10238 return ERR_PTR(ret);
10239 }
10240
__btrfs_prealloc_file_range(struct inode * inode,int mode,u64 start,u64 num_bytes,u64 min_size,loff_t actual_len,u64 * alloc_hint,struct btrfs_trans_handle * trans)10241 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10242 u64 start, u64 num_bytes, u64 min_size,
10243 loff_t actual_len, u64 *alloc_hint,
10244 struct btrfs_trans_handle *trans)
10245 {
10246 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10247 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10248 struct extent_map *em;
10249 struct btrfs_root *root = BTRFS_I(inode)->root;
10250 struct btrfs_key ins;
10251 u64 cur_offset = start;
10252 u64 clear_offset = start;
10253 u64 i_size;
10254 u64 cur_bytes;
10255 u64 last_alloc = (u64)-1;
10256 int ret = 0;
10257 bool own_trans = true;
10258 u64 end = start + num_bytes - 1;
10259
10260 if (trans)
10261 own_trans = false;
10262 while (num_bytes > 0) {
10263 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10264 cur_bytes = max(cur_bytes, min_size);
10265 /*
10266 * If we are severely fragmented we could end up with really
10267 * small allocations, so if the allocator is returning small
10268 * chunks lets make its job easier by only searching for those
10269 * sized chunks.
10270 */
10271 cur_bytes = min(cur_bytes, last_alloc);
10272 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10273 min_size, 0, *alloc_hint, &ins, 1, 0);
10274 if (ret)
10275 break;
10276
10277 /*
10278 * We've reserved this space, and thus converted it from
10279 * ->bytes_may_use to ->bytes_reserved. Any error that happens
10280 * from here on out we will only need to clear our reservation
10281 * for the remaining unreserved area, so advance our
10282 * clear_offset by our extent size.
10283 */
10284 clear_offset += ins.offset;
10285
10286 last_alloc = ins.offset;
10287 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
10288 &ins, cur_offset);
10289 /*
10290 * Now that we inserted the prealloc extent we can finally
10291 * decrement the number of reservations in the block group.
10292 * If we did it before, we could race with relocation and have
10293 * relocation miss the reserved extent, making it fail later.
10294 */
10295 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10296 if (IS_ERR(trans)) {
10297 ret = PTR_ERR(trans);
10298 btrfs_free_reserved_extent(fs_info, ins.objectid,
10299 ins.offset, 0);
10300 break;
10301 }
10302
10303 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10304 cur_offset + ins.offset -1, 0);
10305
10306 em = alloc_extent_map();
10307 if (!em) {
10308 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10309 &BTRFS_I(inode)->runtime_flags);
10310 goto next;
10311 }
10312
10313 em->start = cur_offset;
10314 em->orig_start = cur_offset;
10315 em->len = ins.offset;
10316 em->block_start = ins.objectid;
10317 em->block_len = ins.offset;
10318 em->orig_block_len = ins.offset;
10319 em->ram_bytes = ins.offset;
10320 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10321 em->generation = trans->transid;
10322
10323 while (1) {
10324 write_lock(&em_tree->lock);
10325 ret = add_extent_mapping(em_tree, em, 1);
10326 write_unlock(&em_tree->lock);
10327 if (ret != -EEXIST)
10328 break;
10329 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10330 cur_offset + ins.offset - 1,
10331 0);
10332 }
10333 free_extent_map(em);
10334 next:
10335 num_bytes -= ins.offset;
10336 cur_offset += ins.offset;
10337 *alloc_hint = ins.objectid + ins.offset;
10338
10339 inode_inc_iversion(inode);
10340 inode->i_ctime = current_time(inode);
10341 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10342 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10343 (actual_len > inode->i_size) &&
10344 (cur_offset > inode->i_size)) {
10345 if (cur_offset > actual_len)
10346 i_size = actual_len;
10347 else
10348 i_size = cur_offset;
10349 i_size_write(inode, i_size);
10350 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
10351 }
10352
10353 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10354
10355 if (ret) {
10356 btrfs_abort_transaction(trans, ret);
10357 if (own_trans)
10358 btrfs_end_transaction(trans);
10359 break;
10360 }
10361
10362 if (own_trans) {
10363 btrfs_end_transaction(trans);
10364 trans = NULL;
10365 }
10366 }
10367 if (clear_offset < end)
10368 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
10369 end - clear_offset + 1);
10370 return ret;
10371 }
10372
btrfs_prealloc_file_range(struct inode * inode,int mode,u64 start,u64 num_bytes,u64 min_size,loff_t actual_len,u64 * alloc_hint)10373 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10374 u64 start, u64 num_bytes, u64 min_size,
10375 loff_t actual_len, u64 *alloc_hint)
10376 {
10377 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10378 min_size, actual_len, alloc_hint,
10379 NULL);
10380 }
10381
btrfs_prealloc_file_range_trans(struct inode * inode,struct btrfs_trans_handle * trans,int mode,u64 start,u64 num_bytes,u64 min_size,loff_t actual_len,u64 * alloc_hint)10382 int btrfs_prealloc_file_range_trans(struct inode *inode,
10383 struct btrfs_trans_handle *trans, int mode,
10384 u64 start, u64 num_bytes, u64 min_size,
10385 loff_t actual_len, u64 *alloc_hint)
10386 {
10387 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10388 min_size, actual_len, alloc_hint, trans);
10389 }
10390
btrfs_set_page_dirty(struct page * page)10391 static int btrfs_set_page_dirty(struct page *page)
10392 {
10393 return __set_page_dirty_nobuffers(page);
10394 }
10395
btrfs_permission(struct user_namespace * mnt_userns,struct inode * inode,int mask)10396 static int btrfs_permission(struct user_namespace *mnt_userns,
10397 struct inode *inode, int mask)
10398 {
10399 struct btrfs_root *root = BTRFS_I(inode)->root;
10400 umode_t mode = inode->i_mode;
10401
10402 if (mask & MAY_WRITE &&
10403 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10404 if (btrfs_root_readonly(root))
10405 return -EROFS;
10406 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10407 return -EACCES;
10408 }
10409 return generic_permission(mnt_userns, inode, mask);
10410 }
10411
btrfs_tmpfile(struct user_namespace * mnt_userns,struct inode * dir,struct dentry * dentry,umode_t mode)10412 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
10413 struct dentry *dentry, umode_t mode)
10414 {
10415 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10416 struct btrfs_trans_handle *trans;
10417 struct btrfs_root *root = BTRFS_I(dir)->root;
10418 struct inode *inode = NULL;
10419 u64 objectid;
10420 u64 index;
10421 int ret = 0;
10422
10423 /*
10424 * 5 units required for adding orphan entry
10425 */
10426 trans = btrfs_start_transaction(root, 5);
10427 if (IS_ERR(trans))
10428 return PTR_ERR(trans);
10429
10430 ret = btrfs_get_free_objectid(root, &objectid);
10431 if (ret)
10432 goto out;
10433
10434 inode = btrfs_new_inode(trans, root, mnt_userns, dir, NULL, 0,
10435 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10436 if (IS_ERR(inode)) {
10437 ret = PTR_ERR(inode);
10438 inode = NULL;
10439 goto out;
10440 }
10441
10442 inode->i_fop = &btrfs_file_operations;
10443 inode->i_op = &btrfs_file_inode_operations;
10444
10445 inode->i_mapping->a_ops = &btrfs_aops;
10446
10447 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10448 if (ret)
10449 goto out;
10450
10451 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10452 if (ret)
10453 goto out;
10454 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10455 if (ret)
10456 goto out;
10457
10458 /*
10459 * We set number of links to 0 in btrfs_new_inode(), and here we set
10460 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10461 * through:
10462 *
10463 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10464 */
10465 set_nlink(inode, 1);
10466 d_tmpfile(dentry, inode);
10467 unlock_new_inode(inode);
10468 mark_inode_dirty(inode);
10469 out:
10470 btrfs_end_transaction(trans);
10471 if (ret && inode)
10472 discard_new_inode(inode);
10473 btrfs_btree_balance_dirty(fs_info);
10474 return ret;
10475 }
10476
btrfs_set_range_writeback(struct btrfs_inode * inode,u64 start,u64 end)10477 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
10478 {
10479 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10480 unsigned long index = start >> PAGE_SHIFT;
10481 unsigned long end_index = end >> PAGE_SHIFT;
10482 struct page *page;
10483 u32 len;
10484
10485 ASSERT(end + 1 - start <= U32_MAX);
10486 len = end + 1 - start;
10487 while (index <= end_index) {
10488 page = find_get_page(inode->vfs_inode.i_mapping, index);
10489 ASSERT(page); /* Pages should be in the extent_io_tree */
10490
10491 btrfs_page_set_writeback(fs_info, page, start, len);
10492 put_page(page);
10493 index++;
10494 }
10495 }
10496
10497 #ifdef CONFIG_SWAP
10498 /*
10499 * Add an entry indicating a block group or device which is pinned by a
10500 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10501 * negative errno on failure.
10502 */
btrfs_add_swapfile_pin(struct inode * inode,void * ptr,bool is_block_group)10503 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10504 bool is_block_group)
10505 {
10506 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10507 struct btrfs_swapfile_pin *sp, *entry;
10508 struct rb_node **p;
10509 struct rb_node *parent = NULL;
10510
10511 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10512 if (!sp)
10513 return -ENOMEM;
10514 sp->ptr = ptr;
10515 sp->inode = inode;
10516 sp->is_block_group = is_block_group;
10517 sp->bg_extent_count = 1;
10518
10519 spin_lock(&fs_info->swapfile_pins_lock);
10520 p = &fs_info->swapfile_pins.rb_node;
10521 while (*p) {
10522 parent = *p;
10523 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10524 if (sp->ptr < entry->ptr ||
10525 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10526 p = &(*p)->rb_left;
10527 } else if (sp->ptr > entry->ptr ||
10528 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10529 p = &(*p)->rb_right;
10530 } else {
10531 if (is_block_group)
10532 entry->bg_extent_count++;
10533 spin_unlock(&fs_info->swapfile_pins_lock);
10534 kfree(sp);
10535 return 1;
10536 }
10537 }
10538 rb_link_node(&sp->node, parent, p);
10539 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10540 spin_unlock(&fs_info->swapfile_pins_lock);
10541 return 0;
10542 }
10543
10544 /* Free all of the entries pinned by this swapfile. */
btrfs_free_swapfile_pins(struct inode * inode)10545 static void btrfs_free_swapfile_pins(struct inode *inode)
10546 {
10547 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10548 struct btrfs_swapfile_pin *sp;
10549 struct rb_node *node, *next;
10550
10551 spin_lock(&fs_info->swapfile_pins_lock);
10552 node = rb_first(&fs_info->swapfile_pins);
10553 while (node) {
10554 next = rb_next(node);
10555 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10556 if (sp->inode == inode) {
10557 rb_erase(&sp->node, &fs_info->swapfile_pins);
10558 if (sp->is_block_group) {
10559 btrfs_dec_block_group_swap_extents(sp->ptr,
10560 sp->bg_extent_count);
10561 btrfs_put_block_group(sp->ptr);
10562 }
10563 kfree(sp);
10564 }
10565 node = next;
10566 }
10567 spin_unlock(&fs_info->swapfile_pins_lock);
10568 }
10569
10570 struct btrfs_swap_info {
10571 u64 start;
10572 u64 block_start;
10573 u64 block_len;
10574 u64 lowest_ppage;
10575 u64 highest_ppage;
10576 unsigned long nr_pages;
10577 int nr_extents;
10578 };
10579
btrfs_add_swap_extent(struct swap_info_struct * sis,struct btrfs_swap_info * bsi)10580 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10581 struct btrfs_swap_info *bsi)
10582 {
10583 unsigned long nr_pages;
10584 u64 first_ppage, first_ppage_reported, next_ppage;
10585 int ret;
10586
10587 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10588 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10589 PAGE_SIZE) >> PAGE_SHIFT;
10590
10591 if (first_ppage >= next_ppage)
10592 return 0;
10593 nr_pages = next_ppage - first_ppage;
10594
10595 first_ppage_reported = first_ppage;
10596 if (bsi->start == 0)
10597 first_ppage_reported++;
10598 if (bsi->lowest_ppage > first_ppage_reported)
10599 bsi->lowest_ppage = first_ppage_reported;
10600 if (bsi->highest_ppage < (next_ppage - 1))
10601 bsi->highest_ppage = next_ppage - 1;
10602
10603 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10604 if (ret < 0)
10605 return ret;
10606 bsi->nr_extents += ret;
10607 bsi->nr_pages += nr_pages;
10608 return 0;
10609 }
10610
btrfs_swap_deactivate(struct file * file)10611 static void btrfs_swap_deactivate(struct file *file)
10612 {
10613 struct inode *inode = file_inode(file);
10614
10615 btrfs_free_swapfile_pins(inode);
10616 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10617 }
10618
btrfs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)10619 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10620 sector_t *span)
10621 {
10622 struct inode *inode = file_inode(file);
10623 struct btrfs_root *root = BTRFS_I(inode)->root;
10624 struct btrfs_fs_info *fs_info = root->fs_info;
10625 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10626 struct extent_state *cached_state = NULL;
10627 struct extent_map *em = NULL;
10628 struct btrfs_device *device = NULL;
10629 struct btrfs_swap_info bsi = {
10630 .lowest_ppage = (sector_t)-1ULL,
10631 };
10632 int ret = 0;
10633 u64 isize;
10634 u64 start;
10635
10636 /*
10637 * If the swap file was just created, make sure delalloc is done. If the
10638 * file changes again after this, the user is doing something stupid and
10639 * we don't really care.
10640 */
10641 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10642 if (ret)
10643 return ret;
10644
10645 /*
10646 * The inode is locked, so these flags won't change after we check them.
10647 */
10648 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10649 btrfs_warn(fs_info, "swapfile must not be compressed");
10650 return -EINVAL;
10651 }
10652 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10653 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10654 return -EINVAL;
10655 }
10656 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10657 btrfs_warn(fs_info, "swapfile must not be checksummed");
10658 return -EINVAL;
10659 }
10660
10661 /*
10662 * Balance or device remove/replace/resize can move stuff around from
10663 * under us. The exclop protection makes sure they aren't running/won't
10664 * run concurrently while we are mapping the swap extents, and
10665 * fs_info->swapfile_pins prevents them from running while the swap
10666 * file is active and moving the extents. Note that this also prevents
10667 * a concurrent device add which isn't actually necessary, but it's not
10668 * really worth the trouble to allow it.
10669 */
10670 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10671 btrfs_warn(fs_info,
10672 "cannot activate swapfile while exclusive operation is running");
10673 return -EBUSY;
10674 }
10675
10676 /*
10677 * Prevent snapshot creation while we are activating the swap file.
10678 * We do not want to race with snapshot creation. If snapshot creation
10679 * already started before we bumped nr_swapfiles from 0 to 1 and
10680 * completes before the first write into the swap file after it is
10681 * activated, than that write would fallback to COW.
10682 */
10683 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10684 btrfs_exclop_finish(fs_info);
10685 btrfs_warn(fs_info,
10686 "cannot activate swapfile because snapshot creation is in progress");
10687 return -EINVAL;
10688 }
10689 /*
10690 * Snapshots can create extents which require COW even if NODATACOW is
10691 * set. We use this counter to prevent snapshots. We must increment it
10692 * before walking the extents because we don't want a concurrent
10693 * snapshot to run after we've already checked the extents.
10694 */
10695 atomic_inc(&root->nr_swapfiles);
10696
10697 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10698
10699 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10700 start = 0;
10701 while (start < isize) {
10702 u64 logical_block_start, physical_block_start;
10703 struct btrfs_block_group *bg;
10704 u64 len = isize - start;
10705
10706 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10707 if (IS_ERR(em)) {
10708 ret = PTR_ERR(em);
10709 goto out;
10710 }
10711
10712 if (em->block_start == EXTENT_MAP_HOLE) {
10713 btrfs_warn(fs_info, "swapfile must not have holes");
10714 ret = -EINVAL;
10715 goto out;
10716 }
10717 if (em->block_start == EXTENT_MAP_INLINE) {
10718 /*
10719 * It's unlikely we'll ever actually find ourselves
10720 * here, as a file small enough to fit inline won't be
10721 * big enough to store more than the swap header, but in
10722 * case something changes in the future, let's catch it
10723 * here rather than later.
10724 */
10725 btrfs_warn(fs_info, "swapfile must not be inline");
10726 ret = -EINVAL;
10727 goto out;
10728 }
10729 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10730 btrfs_warn(fs_info, "swapfile must not be compressed");
10731 ret = -EINVAL;
10732 goto out;
10733 }
10734
10735 logical_block_start = em->block_start + (start - em->start);
10736 len = min(len, em->len - (start - em->start));
10737 free_extent_map(em);
10738 em = NULL;
10739
10740 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
10741 if (ret < 0) {
10742 goto out;
10743 } else if (ret) {
10744 ret = 0;
10745 } else {
10746 btrfs_warn(fs_info,
10747 "swapfile must not be copy-on-write");
10748 ret = -EINVAL;
10749 goto out;
10750 }
10751
10752 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10753 if (IS_ERR(em)) {
10754 ret = PTR_ERR(em);
10755 goto out;
10756 }
10757
10758 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10759 btrfs_warn(fs_info,
10760 "swapfile must have single data profile");
10761 ret = -EINVAL;
10762 goto out;
10763 }
10764
10765 if (device == NULL) {
10766 device = em->map_lookup->stripes[0].dev;
10767 ret = btrfs_add_swapfile_pin(inode, device, false);
10768 if (ret == 1)
10769 ret = 0;
10770 else if (ret)
10771 goto out;
10772 } else if (device != em->map_lookup->stripes[0].dev) {
10773 btrfs_warn(fs_info, "swapfile must be on one device");
10774 ret = -EINVAL;
10775 goto out;
10776 }
10777
10778 physical_block_start = (em->map_lookup->stripes[0].physical +
10779 (logical_block_start - em->start));
10780 len = min(len, em->len - (logical_block_start - em->start));
10781 free_extent_map(em);
10782 em = NULL;
10783
10784 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10785 if (!bg) {
10786 btrfs_warn(fs_info,
10787 "could not find block group containing swapfile");
10788 ret = -EINVAL;
10789 goto out;
10790 }
10791
10792 if (!btrfs_inc_block_group_swap_extents(bg)) {
10793 btrfs_warn(fs_info,
10794 "block group for swapfile at %llu is read-only%s",
10795 bg->start,
10796 atomic_read(&fs_info->scrubs_running) ?
10797 " (scrub running)" : "");
10798 btrfs_put_block_group(bg);
10799 ret = -EINVAL;
10800 goto out;
10801 }
10802
10803 ret = btrfs_add_swapfile_pin(inode, bg, true);
10804 if (ret) {
10805 btrfs_put_block_group(bg);
10806 if (ret == 1)
10807 ret = 0;
10808 else
10809 goto out;
10810 }
10811
10812 if (bsi.block_len &&
10813 bsi.block_start + bsi.block_len == physical_block_start) {
10814 bsi.block_len += len;
10815 } else {
10816 if (bsi.block_len) {
10817 ret = btrfs_add_swap_extent(sis, &bsi);
10818 if (ret)
10819 goto out;
10820 }
10821 bsi.start = start;
10822 bsi.block_start = physical_block_start;
10823 bsi.block_len = len;
10824 }
10825
10826 start += len;
10827 }
10828
10829 if (bsi.block_len)
10830 ret = btrfs_add_swap_extent(sis, &bsi);
10831
10832 out:
10833 if (!IS_ERR_OR_NULL(em))
10834 free_extent_map(em);
10835
10836 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10837
10838 if (ret)
10839 btrfs_swap_deactivate(file);
10840
10841 btrfs_drew_write_unlock(&root->snapshot_lock);
10842
10843 btrfs_exclop_finish(fs_info);
10844
10845 if (ret)
10846 return ret;
10847
10848 if (device)
10849 sis->bdev = device->bdev;
10850 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10851 sis->max = bsi.nr_pages;
10852 sis->pages = bsi.nr_pages - 1;
10853 sis->highest_bit = bsi.nr_pages - 1;
10854 return bsi.nr_extents;
10855 }
10856 #else
btrfs_swap_deactivate(struct file * file)10857 static void btrfs_swap_deactivate(struct file *file)
10858 {
10859 }
10860
btrfs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)10861 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10862 sector_t *span)
10863 {
10864 return -EOPNOTSUPP;
10865 }
10866 #endif
10867
10868 /*
10869 * Update the number of bytes used in the VFS' inode. When we replace extents in
10870 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10871 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10872 * always get a correct value.
10873 */
btrfs_update_inode_bytes(struct btrfs_inode * inode,const u64 add_bytes,const u64 del_bytes)10874 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10875 const u64 add_bytes,
10876 const u64 del_bytes)
10877 {
10878 if (add_bytes == del_bytes)
10879 return;
10880
10881 spin_lock(&inode->lock);
10882 if (del_bytes > 0)
10883 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10884 if (add_bytes > 0)
10885 inode_add_bytes(&inode->vfs_inode, add_bytes);
10886 spin_unlock(&inode->lock);
10887 }
10888
10889 static const struct inode_operations btrfs_dir_inode_operations = {
10890 .getattr = btrfs_getattr,
10891 .lookup = btrfs_lookup,
10892 .create = btrfs_create,
10893 .unlink = btrfs_unlink,
10894 .link = btrfs_link,
10895 .mkdir = btrfs_mkdir,
10896 .rmdir = btrfs_rmdir,
10897 .rename = btrfs_rename2,
10898 .symlink = btrfs_symlink,
10899 .setattr = btrfs_setattr,
10900 .mknod = btrfs_mknod,
10901 .listxattr = btrfs_listxattr,
10902 .permission = btrfs_permission,
10903 .get_acl = btrfs_get_acl,
10904 .set_acl = btrfs_set_acl,
10905 .update_time = btrfs_update_time,
10906 .tmpfile = btrfs_tmpfile,
10907 .fileattr_get = btrfs_fileattr_get,
10908 .fileattr_set = btrfs_fileattr_set,
10909 };
10910
10911 static const struct file_operations btrfs_dir_file_operations = {
10912 .llseek = generic_file_llseek,
10913 .read = generic_read_dir,
10914 .iterate_shared = btrfs_real_readdir,
10915 .open = btrfs_opendir,
10916 .unlocked_ioctl = btrfs_ioctl,
10917 #ifdef CONFIG_COMPAT
10918 .compat_ioctl = btrfs_compat_ioctl,
10919 #endif
10920 .release = btrfs_release_file,
10921 .fsync = btrfs_sync_file,
10922 };
10923
10924 /*
10925 * btrfs doesn't support the bmap operation because swapfiles
10926 * use bmap to make a mapping of extents in the file. They assume
10927 * these extents won't change over the life of the file and they
10928 * use the bmap result to do IO directly to the drive.
10929 *
10930 * the btrfs bmap call would return logical addresses that aren't
10931 * suitable for IO and they also will change frequently as COW
10932 * operations happen. So, swapfile + btrfs == corruption.
10933 *
10934 * For now we're avoiding this by dropping bmap.
10935 */
10936 static const struct address_space_operations btrfs_aops = {
10937 .readpage = btrfs_readpage,
10938 .writepage = btrfs_writepage,
10939 .writepages = btrfs_writepages,
10940 .readahead = btrfs_readahead,
10941 .direct_IO = noop_direct_IO,
10942 .invalidatepage = btrfs_invalidatepage,
10943 .releasepage = btrfs_releasepage,
10944 #ifdef CONFIG_MIGRATION
10945 .migratepage = btrfs_migratepage,
10946 #endif
10947 .set_page_dirty = btrfs_set_page_dirty,
10948 .error_remove_page = generic_error_remove_page,
10949 .swap_activate = btrfs_swap_activate,
10950 .swap_deactivate = btrfs_swap_deactivate,
10951 };
10952
10953 static const struct inode_operations btrfs_file_inode_operations = {
10954 .getattr = btrfs_getattr,
10955 .setattr = btrfs_setattr,
10956 .listxattr = btrfs_listxattr,
10957 .permission = btrfs_permission,
10958 .fiemap = btrfs_fiemap,
10959 .get_acl = btrfs_get_acl,
10960 .set_acl = btrfs_set_acl,
10961 .update_time = btrfs_update_time,
10962 .fileattr_get = btrfs_fileattr_get,
10963 .fileattr_set = btrfs_fileattr_set,
10964 };
10965 static const struct inode_operations btrfs_special_inode_operations = {
10966 .getattr = btrfs_getattr,
10967 .setattr = btrfs_setattr,
10968 .permission = btrfs_permission,
10969 .listxattr = btrfs_listxattr,
10970 .get_acl = btrfs_get_acl,
10971 .set_acl = btrfs_set_acl,
10972 .update_time = btrfs_update_time,
10973 };
10974 static const struct inode_operations btrfs_symlink_inode_operations = {
10975 .get_link = page_get_link,
10976 .getattr = btrfs_getattr,
10977 .setattr = btrfs_setattr,
10978 .permission = btrfs_permission,
10979 .listxattr = btrfs_listxattr,
10980 .update_time = btrfs_update_time,
10981 };
10982
10983 const struct dentry_operations btrfs_dentry_operations = {
10984 .d_delete = btrfs_dentry_delete,
10985 };
10986