]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/btrfs/inode.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
btrfs: make btrfs_inode_block_unlocked_dio take btrfs_inode
[mirror_ubuntu-bionic-kernel.git] / fs / btrfs / inode.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/compat.h>
34 #include <linux/bit_spinlock.h>
35 #include <linux/xattr.h>
36 #include <linux/posix_acl.h>
37 #include <linux/falloc.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/mount.h>
41 #include <linux/btrfs.h>
42 #include <linux/blkdev.h>
43 #include <linux/posix_acl_xattr.h>
44 #include <linux/uio.h>
45 #include "ctree.h"
46 #include "disk-io.h"
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "print-tree.h"
50 #include "ordered-data.h"
51 #include "xattr.h"
52 #include "tree-log.h"
53 #include "volumes.h"
54 #include "compression.h"
55 #include "locking.h"
56 #include "free-space-cache.h"
57 #include "inode-map.h"
58 #include "backref.h"
59 #include "hash.h"
60 #include "props.h"
61 #include "qgroup.h"
62 #include "dedupe.h"
63
64 struct btrfs_iget_args {
65 struct btrfs_key *location;
66 struct btrfs_root *root;
67 };
68
69 struct btrfs_dio_data {
70 u64 outstanding_extents;
71 u64 reserve;
72 u64 unsubmitted_oe_range_start;
73 u64 unsubmitted_oe_range_end;
74 int overwrite;
75 };
76
77 static const struct inode_operations btrfs_dir_inode_operations;
78 static const struct inode_operations btrfs_symlink_inode_operations;
79 static const struct inode_operations btrfs_dir_ro_inode_operations;
80 static const struct inode_operations btrfs_special_inode_operations;
81 static const struct inode_operations btrfs_file_inode_operations;
82 static const struct address_space_operations btrfs_aops;
83 static const struct address_space_operations btrfs_symlink_aops;
84 static const struct file_operations btrfs_dir_file_operations;
85 static const struct extent_io_ops btrfs_extent_io_ops;
86
87 static struct kmem_cache *btrfs_inode_cachep;
88 struct kmem_cache *btrfs_trans_handle_cachep;
89 struct kmem_cache *btrfs_transaction_cachep;
90 struct kmem_cache *btrfs_path_cachep;
91 struct kmem_cache *btrfs_free_space_cachep;
92
93 #define S_SHIFT 12
94 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
95 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
96 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
97 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
98 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
99 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
100 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
101 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
102 };
103
104 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
105 static int btrfs_truncate(struct inode *inode);
106 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
107 static noinline int cow_file_range(struct inode *inode,
108 struct page *locked_page,
109 u64 start, u64 end, u64 delalloc_end,
110 int *page_started, unsigned long *nr_written,
111 int unlock, struct btrfs_dedupe_hash *hash);
112 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
113 u64 orig_start, u64 block_start,
114 u64 block_len, u64 orig_block_len,
115 u64 ram_bytes, int compress_type,
116 int type);
117
118 static int btrfs_dirty_inode(struct inode *inode);
119
120 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
121 void btrfs_test_inode_set_ops(struct inode *inode)
122 {
123 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
124 }
125 #endif
126
127 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
128 struct inode *inode, struct inode *dir,
129 const struct qstr *qstr)
130 {
131 int err;
132
133 err = btrfs_init_acl(trans, inode, dir);
134 if (!err)
135 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
136 return err;
137 }
138
139 /*
140 * this does all the hard work for inserting an inline extent into
141 * the btree. The caller should have done a btrfs_drop_extents so that
142 * no overlapping inline items exist in the btree
143 */
144 static int insert_inline_extent(struct btrfs_trans_handle *trans,
145 struct btrfs_path *path, int extent_inserted,
146 struct btrfs_root *root, struct inode *inode,
147 u64 start, size_t size, size_t compressed_size,
148 int compress_type,
149 struct page **compressed_pages)
150 {
151 struct extent_buffer *leaf;
152 struct page *page = NULL;
153 char *kaddr;
154 unsigned long ptr;
155 struct btrfs_file_extent_item *ei;
156 int err = 0;
157 int ret;
158 size_t cur_size = size;
159 unsigned long offset;
160
161 if (compressed_size && compressed_pages)
162 cur_size = compressed_size;
163
164 inode_add_bytes(inode, size);
165
166 if (!extent_inserted) {
167 struct btrfs_key key;
168 size_t datasize;
169
170 key.objectid = btrfs_ino(BTRFS_I(inode));
171 key.offset = start;
172 key.type = BTRFS_EXTENT_DATA_KEY;
173
174 datasize = btrfs_file_extent_calc_inline_size(cur_size);
175 path->leave_spinning = 1;
176 ret = btrfs_insert_empty_item(trans, root, path, &key,
177 datasize);
178 if (ret) {
179 err = ret;
180 goto fail;
181 }
182 }
183 leaf = path->nodes[0];
184 ei = btrfs_item_ptr(leaf, path->slots[0],
185 struct btrfs_file_extent_item);
186 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
187 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
188 btrfs_set_file_extent_encryption(leaf, ei, 0);
189 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
190 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
191 ptr = btrfs_file_extent_inline_start(ei);
192
193 if (compress_type != BTRFS_COMPRESS_NONE) {
194 struct page *cpage;
195 int i = 0;
196 while (compressed_size > 0) {
197 cpage = compressed_pages[i];
198 cur_size = min_t(unsigned long, compressed_size,
199 PAGE_SIZE);
200
201 kaddr = kmap_atomic(cpage);
202 write_extent_buffer(leaf, kaddr, ptr, cur_size);
203 kunmap_atomic(kaddr);
204
205 i++;
206 ptr += cur_size;
207 compressed_size -= cur_size;
208 }
209 btrfs_set_file_extent_compression(leaf, ei,
210 compress_type);
211 } else {
212 page = find_get_page(inode->i_mapping,
213 start >> PAGE_SHIFT);
214 btrfs_set_file_extent_compression(leaf, ei, 0);
215 kaddr = kmap_atomic(page);
216 offset = start & (PAGE_SIZE - 1);
217 write_extent_buffer(leaf, kaddr + offset, ptr, size);
218 kunmap_atomic(kaddr);
219 put_page(page);
220 }
221 btrfs_mark_buffer_dirty(leaf);
222 btrfs_release_path(path);
223
224 /*
225 * we're an inline extent, so nobody can
226 * extend the file past i_size without locking
227 * a page we already have locked.
228 *
229 * We must do any isize and inode updates
230 * before we unlock the pages. Otherwise we
231 * could end up racing with unlink.
232 */
233 BTRFS_I(inode)->disk_i_size = inode->i_size;
234 ret = btrfs_update_inode(trans, root, inode);
235
236 return ret;
237 fail:
238 return err;
239 }
240
241
242 /*
243 * conditionally insert an inline extent into the file. This
244 * does the checks required to make sure the data is small enough
245 * to fit as an inline extent.
246 */
247 static noinline int cow_file_range_inline(struct btrfs_root *root,
248 struct inode *inode, u64 start,
249 u64 end, size_t compressed_size,
250 int compress_type,
251 struct page **compressed_pages)
252 {
253 struct btrfs_fs_info *fs_info = root->fs_info;
254 struct btrfs_trans_handle *trans;
255 u64 isize = i_size_read(inode);
256 u64 actual_end = min(end + 1, isize);
257 u64 inline_len = actual_end - start;
258 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
259 u64 data_len = inline_len;
260 int ret;
261 struct btrfs_path *path;
262 int extent_inserted = 0;
263 u32 extent_item_size;
264
265 if (compressed_size)
266 data_len = compressed_size;
267
268 if (start > 0 ||
269 actual_end > fs_info->sectorsize ||
270 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
271 (!compressed_size &&
272 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
273 end + 1 < isize ||
274 data_len > fs_info->max_inline) {
275 return 1;
276 }
277
278 path = btrfs_alloc_path();
279 if (!path)
280 return -ENOMEM;
281
282 trans = btrfs_join_transaction(root);
283 if (IS_ERR(trans)) {
284 btrfs_free_path(path);
285 return PTR_ERR(trans);
286 }
287 trans->block_rsv = &fs_info->delalloc_block_rsv;
288
289 if (compressed_size && compressed_pages)
290 extent_item_size = btrfs_file_extent_calc_inline_size(
291 compressed_size);
292 else
293 extent_item_size = btrfs_file_extent_calc_inline_size(
294 inline_len);
295
296 ret = __btrfs_drop_extents(trans, root, inode, path,
297 start, aligned_end, NULL,
298 1, 1, extent_item_size, &extent_inserted);
299 if (ret) {
300 btrfs_abort_transaction(trans, ret);
301 goto out;
302 }
303
304 if (isize > actual_end)
305 inline_len = min_t(u64, isize, actual_end);
306 ret = insert_inline_extent(trans, path, extent_inserted,
307 root, inode, start,
308 inline_len, compressed_size,
309 compress_type, compressed_pages);
310 if (ret && ret != -ENOSPC) {
311 btrfs_abort_transaction(trans, ret);
312 goto out;
313 } else if (ret == -ENOSPC) {
314 ret = 1;
315 goto out;
316 }
317
318 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
319 btrfs_delalloc_release_metadata(BTRFS_I(inode), end + 1 - start);
320 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
321 out:
322 /*
323 * Don't forget to free the reserved space, as for inlined extent
324 * it won't count as data extent, free them directly here.
325 * And at reserve time, it's always aligned to page size, so
326 * just free one page here.
327 */
328 btrfs_qgroup_free_data(inode, 0, PAGE_SIZE);
329 btrfs_free_path(path);
330 btrfs_end_transaction(trans);
331 return ret;
332 }
333
334 struct async_extent {
335 u64 start;
336 u64 ram_size;
337 u64 compressed_size;
338 struct page **pages;
339 unsigned long nr_pages;
340 int compress_type;
341 struct list_head list;
342 };
343
344 struct async_cow {
345 struct inode *inode;
346 struct btrfs_root *root;
347 struct page *locked_page;
348 u64 start;
349 u64 end;
350 struct list_head extents;
351 struct btrfs_work work;
352 };
353
354 static noinline int add_async_extent(struct async_cow *cow,
355 u64 start, u64 ram_size,
356 u64 compressed_size,
357 struct page **pages,
358 unsigned long nr_pages,
359 int compress_type)
360 {
361 struct async_extent *async_extent;
362
363 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
364 BUG_ON(!async_extent); /* -ENOMEM */
365 async_extent->start = start;
366 async_extent->ram_size = ram_size;
367 async_extent->compressed_size = compressed_size;
368 async_extent->pages = pages;
369 async_extent->nr_pages = nr_pages;
370 async_extent->compress_type = compress_type;
371 list_add_tail(&async_extent->list, &cow->extents);
372 return 0;
373 }
374
375 static inline int inode_need_compress(struct inode *inode)
376 {
377 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
378
379 /* force compress */
380 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
381 return 1;
382 /* bad compression ratios */
383 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
384 return 0;
385 if (btrfs_test_opt(fs_info, COMPRESS) ||
386 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
387 BTRFS_I(inode)->force_compress)
388 return 1;
389 return 0;
390 }
391
392 static inline void inode_should_defrag(struct btrfs_inode *inode,
393 u64 start, u64 end, u64 num_bytes, u64 small_write)
394 {
395 /* If this is a small write inside eof, kick off a defrag */
396 if (num_bytes < small_write &&
397 (start > 0 || end + 1 < inode->disk_i_size))
398 btrfs_add_inode_defrag(NULL, inode);
399 }
400
401 /*
402 * we create compressed extents in two phases. The first
403 * phase compresses a range of pages that have already been
404 * locked (both pages and state bits are locked).
405 *
406 * This is done inside an ordered work queue, and the compression
407 * is spread across many cpus. The actual IO submission is step
408 * two, and the ordered work queue takes care of making sure that
409 * happens in the same order things were put onto the queue by
410 * writepages and friends.
411 *
412 * If this code finds it can't get good compression, it puts an
413 * entry onto the work queue to write the uncompressed bytes. This
414 * makes sure that both compressed inodes and uncompressed inodes
415 * are written in the same order that the flusher thread sent them
416 * down.
417 */
418 static noinline void compress_file_range(struct inode *inode,
419 struct page *locked_page,
420 u64 start, u64 end,
421 struct async_cow *async_cow,
422 int *num_added)
423 {
424 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
425 struct btrfs_root *root = BTRFS_I(inode)->root;
426 u64 num_bytes;
427 u64 blocksize = fs_info->sectorsize;
428 u64 actual_end;
429 u64 isize = i_size_read(inode);
430 int ret = 0;
431 struct page **pages = NULL;
432 unsigned long nr_pages;
433 unsigned long nr_pages_ret = 0;
434 unsigned long total_compressed = 0;
435 unsigned long total_in = 0;
436 unsigned long max_compressed = SZ_128K;
437 unsigned long max_uncompressed = SZ_128K;
438 int i;
439 int will_compress;
440 int compress_type = fs_info->compress_type;
441 int redirty = 0;
442
443 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
444 SZ_16K);
445
446 actual_end = min_t(u64, isize, end + 1);
447 again:
448 will_compress = 0;
449 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
450 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_SIZE);
451
452 /*
453 * we don't want to send crud past the end of i_size through
454 * compression, that's just a waste of CPU time. So, if the
455 * end of the file is before the start of our current
456 * requested range of bytes, we bail out to the uncompressed
457 * cleanup code that can deal with all of this.
458 *
459 * It isn't really the fastest way to fix things, but this is a
460 * very uncommon corner.
461 */
462 if (actual_end <= start)
463 goto cleanup_and_bail_uncompressed;
464
465 total_compressed = actual_end - start;
466
467 /*
468 * skip compression for a small file range(<=blocksize) that
469 * isn't an inline extent, since it doesn't save disk space at all.
470 */
471 if (total_compressed <= blocksize &&
472 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
473 goto cleanup_and_bail_uncompressed;
474
475 /* we want to make sure that amount of ram required to uncompress
476 * an extent is reasonable, so we limit the total size in ram
477 * of a compressed extent to 128k. This is a crucial number
478 * because it also controls how easily we can spread reads across
479 * cpus for decompression.
480 *
481 * We also want to make sure the amount of IO required to do
482 * a random read is reasonably small, so we limit the size of
483 * a compressed extent to 128k.
484 */
485 total_compressed = min(total_compressed, max_uncompressed);
486 num_bytes = ALIGN(end - start + 1, blocksize);
487 num_bytes = max(blocksize, num_bytes);
488 total_in = 0;
489 ret = 0;
490
491 /*
492 * we do compression for mount -o compress and when the
493 * inode has not been flagged as nocompress. This flag can
494 * change at any time if we discover bad compression ratios.
495 */
496 if (inode_need_compress(inode)) {
497 WARN_ON(pages);
498 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
499 if (!pages) {
500 /* just bail out to the uncompressed code */
501 goto cont;
502 }
503
504 if (BTRFS_I(inode)->force_compress)
505 compress_type = BTRFS_I(inode)->force_compress;
506
507 /*
508 * we need to call clear_page_dirty_for_io on each
509 * page in the range. Otherwise applications with the file
510 * mmap'd can wander in and change the page contents while
511 * we are compressing them.
512 *
513 * If the compression fails for any reason, we set the pages
514 * dirty again later on.
515 */
516 extent_range_clear_dirty_for_io(inode, start, end);
517 redirty = 1;
518 ret = btrfs_compress_pages(compress_type,
519 inode->i_mapping, start,
520 total_compressed, pages,
521 nr_pages, &nr_pages_ret,
522 &total_in,
523 &total_compressed,
524 max_compressed);
525
526 if (!ret) {
527 unsigned long offset = total_compressed &
528 (PAGE_SIZE - 1);
529 struct page *page = pages[nr_pages_ret - 1];
530 char *kaddr;
531
532 /* zero the tail end of the last page, we might be
533 * sending it down to disk
534 */
535 if (offset) {
536 kaddr = kmap_atomic(page);
537 memset(kaddr + offset, 0,
538 PAGE_SIZE - offset);
539 kunmap_atomic(kaddr);
540 }
541 will_compress = 1;
542 }
543 }
544 cont:
545 if (start == 0) {
546 /* lets try to make an inline extent */
547 if (ret || total_in < (actual_end - start)) {
548 /* we didn't compress the entire range, try
549 * to make an uncompressed inline extent.
550 */
551 ret = cow_file_range_inline(root, inode, start, end,
552 0, BTRFS_COMPRESS_NONE, NULL);
553 } else {
554 /* try making a compressed inline extent */
555 ret = cow_file_range_inline(root, inode, start, end,
556 total_compressed,
557 compress_type, pages);
558 }
559 if (ret <= 0) {
560 unsigned long clear_flags = EXTENT_DELALLOC |
561 EXTENT_DEFRAG;
562 unsigned long page_error_op;
563
564 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
565 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
566
567 /*
568 * inline extent creation worked or returned error,
569 * we don't need to create any more async work items.
570 * Unlock and free up our temp pages.
571 */
572 extent_clear_unlock_delalloc(inode, start, end, end,
573 NULL, clear_flags,
574 PAGE_UNLOCK |
575 PAGE_CLEAR_DIRTY |
576 PAGE_SET_WRITEBACK |
577 page_error_op |
578 PAGE_END_WRITEBACK);
579 btrfs_free_reserved_data_space_noquota(inode, start,
580 end - start + 1);
581 goto free_pages_out;
582 }
583 }
584
585 if (will_compress) {
586 /*
587 * we aren't doing an inline extent round the compressed size
588 * up to a block size boundary so the allocator does sane
589 * things
590 */
591 total_compressed = ALIGN(total_compressed, blocksize);
592
593 /*
594 * one last check to make sure the compression is really a
595 * win, compare the page count read with the blocks on disk
596 */
597 total_in = ALIGN(total_in, PAGE_SIZE);
598 if (total_compressed >= total_in) {
599 will_compress = 0;
600 } else {
601 num_bytes = total_in;
602 *num_added += 1;
603
604 /*
605 * The async work queues will take care of doing actual
606 * allocation on disk for these compressed pages, and
607 * will submit them to the elevator.
608 */
609 add_async_extent(async_cow, start, num_bytes,
610 total_compressed, pages, nr_pages_ret,
611 compress_type);
612
613 if (start + num_bytes < end) {
614 start += num_bytes;
615 pages = NULL;
616 cond_resched();
617 goto again;
618 }
619 return;
620 }
621 }
622 if (pages) {
623 /*
624 * the compression code ran but failed to make things smaller,
625 * free any pages it allocated and our page pointer array
626 */
627 for (i = 0; i < nr_pages_ret; i++) {
628 WARN_ON(pages[i]->mapping);
629 put_page(pages[i]);
630 }
631 kfree(pages);
632 pages = NULL;
633 total_compressed = 0;
634 nr_pages_ret = 0;
635
636 /* flag the file so we don't compress in the future */
637 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
638 !(BTRFS_I(inode)->force_compress)) {
639 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
640 }
641 }
642 cleanup_and_bail_uncompressed:
643 /*
644 * No compression, but we still need to write the pages in the file
645 * we've been given so far. redirty the locked page if it corresponds
646 * to our extent and set things up for the async work queue to run
647 * cow_file_range to do the normal delalloc dance.
648 */
649 if (page_offset(locked_page) >= start &&
650 page_offset(locked_page) <= end)
651 __set_page_dirty_nobuffers(locked_page);
652 /* unlocked later on in the async handlers */
653
654 if (redirty)
655 extent_range_redirty_for_io(inode, start, end);
656 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
657 BTRFS_COMPRESS_NONE);
658 *num_added += 1;
659
660 return;
661
662 free_pages_out:
663 for (i = 0; i < nr_pages_ret; i++) {
664 WARN_ON(pages[i]->mapping);
665 put_page(pages[i]);
666 }
667 kfree(pages);
668 }
669
670 static void free_async_extent_pages(struct async_extent *async_extent)
671 {
672 int i;
673
674 if (!async_extent->pages)
675 return;
676
677 for (i = 0; i < async_extent->nr_pages; i++) {
678 WARN_ON(async_extent->pages[i]->mapping);
679 put_page(async_extent->pages[i]);
680 }
681 kfree(async_extent->pages);
682 async_extent->nr_pages = 0;
683 async_extent->pages = NULL;
684 }
685
686 /*
687 * phase two of compressed writeback. This is the ordered portion
688 * of the code, which only gets called in the order the work was
689 * queued. We walk all the async extents created by compress_file_range
690 * and send them down to the disk.
691 */
692 static noinline void submit_compressed_extents(struct inode *inode,
693 struct async_cow *async_cow)
694 {
695 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
696 struct async_extent *async_extent;
697 u64 alloc_hint = 0;
698 struct btrfs_key ins;
699 struct extent_map *em;
700 struct btrfs_root *root = BTRFS_I(inode)->root;
701 struct extent_io_tree *io_tree;
702 int ret = 0;
703
704 again:
705 while (!list_empty(&async_cow->extents)) {
706 async_extent = list_entry(async_cow->extents.next,
707 struct async_extent, list);
708 list_del(&async_extent->list);
709
710 io_tree = &BTRFS_I(inode)->io_tree;
711
712 retry:
713 /* did the compression code fall back to uncompressed IO? */
714 if (!async_extent->pages) {
715 int page_started = 0;
716 unsigned long nr_written = 0;
717
718 lock_extent(io_tree, async_extent->start,
719 async_extent->start +
720 async_extent->ram_size - 1);
721
722 /* allocate blocks */
723 ret = cow_file_range(inode, async_cow->locked_page,
724 async_extent->start,
725 async_extent->start +
726 async_extent->ram_size - 1,
727 async_extent->start +
728 async_extent->ram_size - 1,
729 &page_started, &nr_written, 0,
730 NULL);
731
732 /* JDM XXX */
733
734 /*
735 * if page_started, cow_file_range inserted an
736 * inline extent and took care of all the unlocking
737 * and IO for us. Otherwise, we need to submit
738 * all those pages down to the drive.
739 */
740 if (!page_started && !ret)
741 extent_write_locked_range(io_tree,
742 inode, async_extent->start,
743 async_extent->start +
744 async_extent->ram_size - 1,
745 btrfs_get_extent,
746 WB_SYNC_ALL);
747 else if (ret)
748 unlock_page(async_cow->locked_page);
749 kfree(async_extent);
750 cond_resched();
751 continue;
752 }
753
754 lock_extent(io_tree, async_extent->start,
755 async_extent->start + async_extent->ram_size - 1);
756
757 ret = btrfs_reserve_extent(root, async_extent->ram_size,
758 async_extent->compressed_size,
759 async_extent->compressed_size,
760 0, alloc_hint, &ins, 1, 1);
761 if (ret) {
762 free_async_extent_pages(async_extent);
763
764 if (ret == -ENOSPC) {
765 unlock_extent(io_tree, async_extent->start,
766 async_extent->start +
767 async_extent->ram_size - 1);
768
769 /*
770 * we need to redirty the pages if we decide to
771 * fallback to uncompressed IO, otherwise we
772 * will not submit these pages down to lower
773 * layers.
774 */
775 extent_range_redirty_for_io(inode,
776 async_extent->start,
777 async_extent->start +
778 async_extent->ram_size - 1);
779
780 goto retry;
781 }
782 goto out_free;
783 }
784 /*
785 * here we're doing allocation and writeback of the
786 * compressed pages
787 */
788 em = create_io_em(inode, async_extent->start,
789 async_extent->ram_size, /* len */
790 async_extent->start, /* orig_start */
791 ins.objectid, /* block_start */
792 ins.offset, /* block_len */
793 ins.offset, /* orig_block_len */
794 async_extent->ram_size, /* ram_bytes */
795 async_extent->compress_type,
796 BTRFS_ORDERED_COMPRESSED);
797 if (IS_ERR(em))
798 /* ret value is not necessary due to void function */
799 goto out_free_reserve;
800 free_extent_map(em);
801
802 ret = btrfs_add_ordered_extent_compress(inode,
803 async_extent->start,
804 ins.objectid,
805 async_extent->ram_size,
806 ins.offset,
807 BTRFS_ORDERED_COMPRESSED,
808 async_extent->compress_type);
809 if (ret) {
810 btrfs_drop_extent_cache(BTRFS_I(inode),
811 async_extent->start,
812 async_extent->start +
813 async_extent->ram_size - 1, 0);
814 goto out_free_reserve;
815 }
816 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
817
818 /*
819 * clear dirty, set writeback and unlock the pages.
820 */
821 extent_clear_unlock_delalloc(inode, async_extent->start,
822 async_extent->start +
823 async_extent->ram_size - 1,
824 async_extent->start +
825 async_extent->ram_size - 1,
826 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
827 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
828 PAGE_SET_WRITEBACK);
829 ret = btrfs_submit_compressed_write(inode,
830 async_extent->start,
831 async_extent->ram_size,
832 ins.objectid,
833 ins.offset, async_extent->pages,
834 async_extent->nr_pages);
835 if (ret) {
836 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
837 struct page *p = async_extent->pages[0];
838 const u64 start = async_extent->start;
839 const u64 end = start + async_extent->ram_size - 1;
840
841 p->mapping = inode->i_mapping;
842 tree->ops->writepage_end_io_hook(p, start, end,
843 NULL, 0);
844 p->mapping = NULL;
845 extent_clear_unlock_delalloc(inode, start, end, end,
846 NULL, 0,
847 PAGE_END_WRITEBACK |
848 PAGE_SET_ERROR);
849 free_async_extent_pages(async_extent);
850 }
851 alloc_hint = ins.objectid + ins.offset;
852 kfree(async_extent);
853 cond_resched();
854 }
855 return;
856 out_free_reserve:
857 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
858 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
859 out_free:
860 extent_clear_unlock_delalloc(inode, async_extent->start,
861 async_extent->start +
862 async_extent->ram_size - 1,
863 async_extent->start +
864 async_extent->ram_size - 1,
865 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
866 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
867 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
868 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
869 PAGE_SET_ERROR);
870 free_async_extent_pages(async_extent);
871 kfree(async_extent);
872 goto again;
873 }
874
875 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
876 u64 num_bytes)
877 {
878 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
879 struct extent_map *em;
880 u64 alloc_hint = 0;
881
882 read_lock(&em_tree->lock);
883 em = search_extent_mapping(em_tree, start, num_bytes);
884 if (em) {
885 /*
886 * if block start isn't an actual block number then find the
887 * first block in this inode and use that as a hint. If that
888 * block is also bogus then just don't worry about it.
889 */
890 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
891 free_extent_map(em);
892 em = search_extent_mapping(em_tree, 0, 0);
893 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
894 alloc_hint = em->block_start;
895 if (em)
896 free_extent_map(em);
897 } else {
898 alloc_hint = em->block_start;
899 free_extent_map(em);
900 }
901 }
902 read_unlock(&em_tree->lock);
903
904 return alloc_hint;
905 }
906
907 /*
908 * when extent_io.c finds a delayed allocation range in the file,
909 * the call backs end up in this code. The basic idea is to
910 * allocate extents on disk for the range, and create ordered data structs
911 * in ram to track those extents.
912 *
913 * locked_page is the page that writepage had locked already. We use
914 * it to make sure we don't do extra locks or unlocks.
915 *
916 * *page_started is set to one if we unlock locked_page and do everything
917 * required to start IO on it. It may be clean and already done with
918 * IO when we return.
919 */
920 static noinline int cow_file_range(struct inode *inode,
921 struct page *locked_page,
922 u64 start, u64 end, u64 delalloc_end,
923 int *page_started, unsigned long *nr_written,
924 int unlock, struct btrfs_dedupe_hash *hash)
925 {
926 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
927 struct btrfs_root *root = BTRFS_I(inode)->root;
928 u64 alloc_hint = 0;
929 u64 num_bytes;
930 unsigned long ram_size;
931 u64 disk_num_bytes;
932 u64 cur_alloc_size;
933 u64 blocksize = fs_info->sectorsize;
934 struct btrfs_key ins;
935 struct extent_map *em;
936 int ret = 0;
937
938 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
939 WARN_ON_ONCE(1);
940 ret = -EINVAL;
941 goto out_unlock;
942 }
943
944 num_bytes = ALIGN(end - start + 1, blocksize);
945 num_bytes = max(blocksize, num_bytes);
946 disk_num_bytes = num_bytes;
947
948 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
949
950 if (start == 0) {
951 /* lets try to make an inline extent */
952 ret = cow_file_range_inline(root, inode, start, end, 0,
953 BTRFS_COMPRESS_NONE, NULL);
954 if (ret == 0) {
955 extent_clear_unlock_delalloc(inode, start, end,
956 delalloc_end, NULL,
957 EXTENT_LOCKED | EXTENT_DELALLOC |
958 EXTENT_DEFRAG, PAGE_UNLOCK |
959 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
960 PAGE_END_WRITEBACK);
961 btrfs_free_reserved_data_space_noquota(inode, start,
962 end - start + 1);
963 *nr_written = *nr_written +
964 (end - start + PAGE_SIZE) / PAGE_SIZE;
965 *page_started = 1;
966 goto out;
967 } else if (ret < 0) {
968 goto out_unlock;
969 }
970 }
971
972 BUG_ON(disk_num_bytes >
973 btrfs_super_total_bytes(fs_info->super_copy));
974
975 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
976 btrfs_drop_extent_cache(BTRFS_I(inode), start,
977 start + num_bytes - 1, 0);
978
979 while (disk_num_bytes > 0) {
980 unsigned long op;
981
982 cur_alloc_size = disk_num_bytes;
983 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
984 fs_info->sectorsize, 0, alloc_hint,
985 &ins, 1, 1);
986 if (ret < 0)
987 goto out_unlock;
988
989 ram_size = ins.offset;
990 em = create_io_em(inode, start, ins.offset, /* len */
991 start, /* orig_start */
992 ins.objectid, /* block_start */
993 ins.offset, /* block_len */
994 ins.offset, /* orig_block_len */
995 ram_size, /* ram_bytes */
996 BTRFS_COMPRESS_NONE, /* compress_type */
997 BTRFS_ORDERED_REGULAR /* type */);
998 if (IS_ERR(em))
999 goto out_reserve;
1000 free_extent_map(em);
1001
1002 cur_alloc_size = ins.offset;
1003 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1004 ram_size, cur_alloc_size, 0);
1005 if (ret)
1006 goto out_drop_extent_cache;
1007
1008 if (root->root_key.objectid ==
1009 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1010 ret = btrfs_reloc_clone_csums(inode, start,
1011 cur_alloc_size);
1012 if (ret)
1013 goto out_drop_extent_cache;
1014 }
1015
1016 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1017
1018 if (disk_num_bytes < cur_alloc_size)
1019 break;
1020
1021 /* we're not doing compressed IO, don't unlock the first
1022 * page (which the caller expects to stay locked), don't
1023 * clear any dirty bits and don't set any writeback bits
1024 *
1025 * Do set the Private2 bit so we know this page was properly
1026 * setup for writepage
1027 */
1028 op = unlock ? PAGE_UNLOCK : 0;
1029 op |= PAGE_SET_PRIVATE2;
1030
1031 extent_clear_unlock_delalloc(inode, start,
1032 start + ram_size - 1,
1033 delalloc_end, locked_page,
1034 EXTENT_LOCKED | EXTENT_DELALLOC,
1035 op);
1036 disk_num_bytes -= cur_alloc_size;
1037 num_bytes -= cur_alloc_size;
1038 alloc_hint = ins.objectid + ins.offset;
1039 start += cur_alloc_size;
1040 }
1041 out:
1042 return ret;
1043
1044 out_drop_extent_cache:
1045 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1046 out_reserve:
1047 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1048 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1049 out_unlock:
1050 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1051 locked_page,
1052 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1053 EXTENT_DELALLOC | EXTENT_DEFRAG,
1054 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1055 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1056 goto out;
1057 }
1058
1059 /*
1060 * work queue call back to started compression on a file and pages
1061 */
1062 static noinline void async_cow_start(struct btrfs_work *work)
1063 {
1064 struct async_cow *async_cow;
1065 int num_added = 0;
1066 async_cow = container_of(work, struct async_cow, work);
1067
1068 compress_file_range(async_cow->inode, async_cow->locked_page,
1069 async_cow->start, async_cow->end, async_cow,
1070 &num_added);
1071 if (num_added == 0) {
1072 btrfs_add_delayed_iput(async_cow->inode);
1073 async_cow->inode = NULL;
1074 }
1075 }
1076
1077 /*
1078 * work queue call back to submit previously compressed pages
1079 */
1080 static noinline void async_cow_submit(struct btrfs_work *work)
1081 {
1082 struct btrfs_fs_info *fs_info;
1083 struct async_cow *async_cow;
1084 struct btrfs_root *root;
1085 unsigned long nr_pages;
1086
1087 async_cow = container_of(work, struct async_cow, work);
1088
1089 root = async_cow->root;
1090 fs_info = root->fs_info;
1091 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1092 PAGE_SHIFT;
1093
1094 /*
1095 * atomic_sub_return implies a barrier for waitqueue_active
1096 */
1097 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1098 5 * SZ_1M &&
1099 waitqueue_active(&fs_info->async_submit_wait))
1100 wake_up(&fs_info->async_submit_wait);
1101
1102 if (async_cow->inode)
1103 submit_compressed_extents(async_cow->inode, async_cow);
1104 }
1105
1106 static noinline void async_cow_free(struct btrfs_work *work)
1107 {
1108 struct async_cow *async_cow;
1109 async_cow = container_of(work, struct async_cow, work);
1110 if (async_cow->inode)
1111 btrfs_add_delayed_iput(async_cow->inode);
1112 kfree(async_cow);
1113 }
1114
1115 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1116 u64 start, u64 end, int *page_started,
1117 unsigned long *nr_written)
1118 {
1119 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1120 struct async_cow *async_cow;
1121 struct btrfs_root *root = BTRFS_I(inode)->root;
1122 unsigned long nr_pages;
1123 u64 cur_end;
1124
1125 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1126 1, 0, NULL, GFP_NOFS);
1127 while (start < end) {
1128 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1129 BUG_ON(!async_cow); /* -ENOMEM */
1130 async_cow->inode = igrab(inode);
1131 async_cow->root = root;
1132 async_cow->locked_page = locked_page;
1133 async_cow->start = start;
1134
1135 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1136 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1137 cur_end = end;
1138 else
1139 cur_end = min(end, start + SZ_512K - 1);
1140
1141 async_cow->end = cur_end;
1142 INIT_LIST_HEAD(&async_cow->extents);
1143
1144 btrfs_init_work(&async_cow->work,
1145 btrfs_delalloc_helper,
1146 async_cow_start, async_cow_submit,
1147 async_cow_free);
1148
1149 nr_pages = (cur_end - start + PAGE_SIZE) >>
1150 PAGE_SHIFT;
1151 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1152
1153 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1154
1155 while (atomic_read(&fs_info->async_submit_draining) &&
1156 atomic_read(&fs_info->async_delalloc_pages)) {
1157 wait_event(fs_info->async_submit_wait,
1158 (atomic_read(&fs_info->async_delalloc_pages) ==
1159 0));
1160 }
1161
1162 *nr_written += nr_pages;
1163 start = cur_end + 1;
1164 }
1165 *page_started = 1;
1166 return 0;
1167 }
1168
1169 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1170 u64 bytenr, u64 num_bytes)
1171 {
1172 int ret;
1173 struct btrfs_ordered_sum *sums;
1174 LIST_HEAD(list);
1175
1176 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1177 bytenr + num_bytes - 1, &list, 0);
1178 if (ret == 0 && list_empty(&list))
1179 return 0;
1180
1181 while (!list_empty(&list)) {
1182 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1183 list_del(&sums->list);
1184 kfree(sums);
1185 }
1186 return 1;
1187 }
1188
1189 /*
1190 * when nowcow writeback call back. This checks for snapshots or COW copies
1191 * of the extents that exist in the file, and COWs the file as required.
1192 *
1193 * If no cow copies or snapshots exist, we write directly to the existing
1194 * blocks on disk
1195 */
1196 static noinline int run_delalloc_nocow(struct inode *inode,
1197 struct page *locked_page,
1198 u64 start, u64 end, int *page_started, int force,
1199 unsigned long *nr_written)
1200 {
1201 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1202 struct btrfs_root *root = BTRFS_I(inode)->root;
1203 struct extent_buffer *leaf;
1204 struct btrfs_path *path;
1205 struct btrfs_file_extent_item *fi;
1206 struct btrfs_key found_key;
1207 struct extent_map *em;
1208 u64 cow_start;
1209 u64 cur_offset;
1210 u64 extent_end;
1211 u64 extent_offset;
1212 u64 disk_bytenr;
1213 u64 num_bytes;
1214 u64 disk_num_bytes;
1215 u64 ram_bytes;
1216 int extent_type;
1217 int ret, err;
1218 int type;
1219 int nocow;
1220 int check_prev = 1;
1221 bool nolock;
1222 u64 ino = btrfs_ino(BTRFS_I(inode));
1223
1224 path = btrfs_alloc_path();
1225 if (!path) {
1226 extent_clear_unlock_delalloc(inode, start, end, end,
1227 locked_page,
1228 EXTENT_LOCKED | EXTENT_DELALLOC |
1229 EXTENT_DO_ACCOUNTING |
1230 EXTENT_DEFRAG, PAGE_UNLOCK |
1231 PAGE_CLEAR_DIRTY |
1232 PAGE_SET_WRITEBACK |
1233 PAGE_END_WRITEBACK);
1234 return -ENOMEM;
1235 }
1236
1237 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1238
1239 cow_start = (u64)-1;
1240 cur_offset = start;
1241 while (1) {
1242 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1243 cur_offset, 0);
1244 if (ret < 0)
1245 goto error;
1246 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1247 leaf = path->nodes[0];
1248 btrfs_item_key_to_cpu(leaf, &found_key,
1249 path->slots[0] - 1);
1250 if (found_key.objectid == ino &&
1251 found_key.type == BTRFS_EXTENT_DATA_KEY)
1252 path->slots[0]--;
1253 }
1254 check_prev = 0;
1255 next_slot:
1256 leaf = path->nodes[0];
1257 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1258 ret = btrfs_next_leaf(root, path);
1259 if (ret < 0)
1260 goto error;
1261 if (ret > 0)
1262 break;
1263 leaf = path->nodes[0];
1264 }
1265
1266 nocow = 0;
1267 disk_bytenr = 0;
1268 num_bytes = 0;
1269 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1270
1271 if (found_key.objectid > ino)
1272 break;
1273 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1274 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1275 path->slots[0]++;
1276 goto next_slot;
1277 }
1278 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1279 found_key.offset > end)
1280 break;
1281
1282 if (found_key.offset > cur_offset) {
1283 extent_end = found_key.offset;
1284 extent_type = 0;
1285 goto out_check;
1286 }
1287
1288 fi = btrfs_item_ptr(leaf, path->slots[0],
1289 struct btrfs_file_extent_item);
1290 extent_type = btrfs_file_extent_type(leaf, fi);
1291
1292 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1293 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1294 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1295 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1296 extent_offset = btrfs_file_extent_offset(leaf, fi);
1297 extent_end = found_key.offset +
1298 btrfs_file_extent_num_bytes(leaf, fi);
1299 disk_num_bytes =
1300 btrfs_file_extent_disk_num_bytes(leaf, fi);
1301 if (extent_end <= start) {
1302 path->slots[0]++;
1303 goto next_slot;
1304 }
1305 if (disk_bytenr == 0)
1306 goto out_check;
1307 if (btrfs_file_extent_compression(leaf, fi) ||
1308 btrfs_file_extent_encryption(leaf, fi) ||
1309 btrfs_file_extent_other_encoding(leaf, fi))
1310 goto out_check;
1311 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1312 goto out_check;
1313 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1314 goto out_check;
1315 if (btrfs_cross_ref_exist(root, ino,
1316 found_key.offset -
1317 extent_offset, disk_bytenr))
1318 goto out_check;
1319 disk_bytenr += extent_offset;
1320 disk_bytenr += cur_offset - found_key.offset;
1321 num_bytes = min(end + 1, extent_end) - cur_offset;
1322 /*
1323 * if there are pending snapshots for this root,
1324 * we fall into common COW way.
1325 */
1326 if (!nolock) {
1327 err = btrfs_start_write_no_snapshoting(root);
1328 if (!err)
1329 goto out_check;
1330 }
1331 /*
1332 * force cow if csum exists in the range.
1333 * this ensure that csum for a given extent are
1334 * either valid or do not exist.
1335 */
1336 if (csum_exist_in_range(fs_info, disk_bytenr,
1337 num_bytes))
1338 goto out_check;
1339 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1340 goto out_check;
1341 nocow = 1;
1342 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1343 extent_end = found_key.offset +
1344 btrfs_file_extent_inline_len(leaf,
1345 path->slots[0], fi);
1346 extent_end = ALIGN(extent_end,
1347 fs_info->sectorsize);
1348 } else {
1349 BUG_ON(1);
1350 }
1351 out_check:
1352 if (extent_end <= start) {
1353 path->slots[0]++;
1354 if (!nolock && nocow)
1355 btrfs_end_write_no_snapshoting(root);
1356 if (nocow)
1357 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1358 goto next_slot;
1359 }
1360 if (!nocow) {
1361 if (cow_start == (u64)-1)
1362 cow_start = cur_offset;
1363 cur_offset = extent_end;
1364 if (cur_offset > end)
1365 break;
1366 path->slots[0]++;
1367 goto next_slot;
1368 }
1369
1370 btrfs_release_path(path);
1371 if (cow_start != (u64)-1) {
1372 ret = cow_file_range(inode, locked_page,
1373 cow_start, found_key.offset - 1,
1374 end, page_started, nr_written, 1,
1375 NULL);
1376 if (ret) {
1377 if (!nolock && nocow)
1378 btrfs_end_write_no_snapshoting(root);
1379 if (nocow)
1380 btrfs_dec_nocow_writers(fs_info,
1381 disk_bytenr);
1382 goto error;
1383 }
1384 cow_start = (u64)-1;
1385 }
1386
1387 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1388 u64 orig_start = found_key.offset - extent_offset;
1389
1390 em = create_io_em(inode, cur_offset, num_bytes,
1391 orig_start,
1392 disk_bytenr, /* block_start */
1393 num_bytes, /* block_len */
1394 disk_num_bytes, /* orig_block_len */
1395 ram_bytes, BTRFS_COMPRESS_NONE,
1396 BTRFS_ORDERED_PREALLOC);
1397 if (IS_ERR(em)) {
1398 if (!nolock && nocow)
1399 btrfs_end_write_no_snapshoting(root);
1400 if (nocow)
1401 btrfs_dec_nocow_writers(fs_info,
1402 disk_bytenr);
1403 ret = PTR_ERR(em);
1404 goto error;
1405 }
1406 free_extent_map(em);
1407 }
1408
1409 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1410 type = BTRFS_ORDERED_PREALLOC;
1411 } else {
1412 type = BTRFS_ORDERED_NOCOW;
1413 }
1414
1415 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1416 num_bytes, num_bytes, type);
1417 if (nocow)
1418 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1419 BUG_ON(ret); /* -ENOMEM */
1420
1421 if (root->root_key.objectid ==
1422 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1423 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1424 num_bytes);
1425 if (ret) {
1426 if (!nolock && nocow)
1427 btrfs_end_write_no_snapshoting(root);
1428 goto error;
1429 }
1430 }
1431
1432 extent_clear_unlock_delalloc(inode, cur_offset,
1433 cur_offset + num_bytes - 1, end,
1434 locked_page, EXTENT_LOCKED |
1435 EXTENT_DELALLOC |
1436 EXTENT_CLEAR_DATA_RESV,
1437 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1438
1439 if (!nolock && nocow)
1440 btrfs_end_write_no_snapshoting(root);
1441 cur_offset = extent_end;
1442 if (cur_offset > end)
1443 break;
1444 }
1445 btrfs_release_path(path);
1446
1447 if (cur_offset <= end && cow_start == (u64)-1) {
1448 cow_start = cur_offset;
1449 cur_offset = end;
1450 }
1451
1452 if (cow_start != (u64)-1) {
1453 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1454 page_started, nr_written, 1, NULL);
1455 if (ret)
1456 goto error;
1457 }
1458
1459 error:
1460 if (ret && cur_offset < end)
1461 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1462 locked_page, EXTENT_LOCKED |
1463 EXTENT_DELALLOC | EXTENT_DEFRAG |
1464 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1465 PAGE_CLEAR_DIRTY |
1466 PAGE_SET_WRITEBACK |
1467 PAGE_END_WRITEBACK);
1468 btrfs_free_path(path);
1469 return ret;
1470 }
1471
1472 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1473 {
1474
1475 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1476 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1477 return 0;
1478
1479 /*
1480 * @defrag_bytes is a hint value, no spinlock held here,
1481 * if is not zero, it means the file is defragging.
1482 * Force cow if given extent needs to be defragged.
1483 */
1484 if (BTRFS_I(inode)->defrag_bytes &&
1485 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1486 EXTENT_DEFRAG, 0, NULL))
1487 return 1;
1488
1489 return 0;
1490 }
1491
1492 /*
1493 * extent_io.c call back to do delayed allocation processing
1494 */
1495 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1496 u64 start, u64 end, int *page_started,
1497 unsigned long *nr_written)
1498 {
1499 int ret;
1500 int force_cow = need_force_cow(inode, start, end);
1501
1502 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1503 ret = run_delalloc_nocow(inode, locked_page, start, end,
1504 page_started, 1, nr_written);
1505 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1506 ret = run_delalloc_nocow(inode, locked_page, start, end,
1507 page_started, 0, nr_written);
1508 } else if (!inode_need_compress(inode)) {
1509 ret = cow_file_range(inode, locked_page, start, end, end,
1510 page_started, nr_written, 1, NULL);
1511 } else {
1512 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1513 &BTRFS_I(inode)->runtime_flags);
1514 ret = cow_file_range_async(inode, locked_page, start, end,
1515 page_started, nr_written);
1516 }
1517 return ret;
1518 }
1519
1520 static void btrfs_split_extent_hook(struct inode *inode,
1521 struct extent_state *orig, u64 split)
1522 {
1523 u64 size;
1524
1525 /* not delalloc, ignore it */
1526 if (!(orig->state & EXTENT_DELALLOC))
1527 return;
1528
1529 size = orig->end - orig->start + 1;
1530 if (size > BTRFS_MAX_EXTENT_SIZE) {
1531 u32 num_extents;
1532 u64 new_size;
1533
1534 /*
1535 * See the explanation in btrfs_merge_extent_hook, the same
1536 * applies here, just in reverse.
1537 */
1538 new_size = orig->end - split + 1;
1539 num_extents = count_max_extents(new_size);
1540 new_size = split - orig->start;
1541 num_extents += count_max_extents(new_size);
1542 if (count_max_extents(size) >= num_extents)
1543 return;
1544 }
1545
1546 spin_lock(&BTRFS_I(inode)->lock);
1547 BTRFS_I(inode)->outstanding_extents++;
1548 spin_unlock(&BTRFS_I(inode)->lock);
1549 }
1550
1551 /*
1552 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1553 * extents so we can keep track of new extents that are just merged onto old
1554 * extents, such as when we are doing sequential writes, so we can properly
1555 * account for the metadata space we'll need.
1556 */
1557 static void btrfs_merge_extent_hook(struct inode *inode,
1558 struct extent_state *new,
1559 struct extent_state *other)
1560 {
1561 u64 new_size, old_size;
1562 u32 num_extents;
1563
1564 /* not delalloc, ignore it */
1565 if (!(other->state & EXTENT_DELALLOC))
1566 return;
1567
1568 if (new->start > other->start)
1569 new_size = new->end - other->start + 1;
1570 else
1571 new_size = other->end - new->start + 1;
1572
1573 /* we're not bigger than the max, unreserve the space and go */
1574 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1575 spin_lock(&BTRFS_I(inode)->lock);
1576 BTRFS_I(inode)->outstanding_extents--;
1577 spin_unlock(&BTRFS_I(inode)->lock);
1578 return;
1579 }
1580
1581 /*
1582 * We have to add up either side to figure out how many extents were
1583 * accounted for before we merged into one big extent. If the number of
1584 * extents we accounted for is <= the amount we need for the new range
1585 * then we can return, otherwise drop. Think of it like this
1586 *
1587 * [ 4k][MAX_SIZE]
1588 *
1589 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1590 * need 2 outstanding extents, on one side we have 1 and the other side
1591 * we have 1 so they are == and we can return. But in this case
1592 *
1593 * [MAX_SIZE+4k][MAX_SIZE+4k]
1594 *
1595 * Each range on their own accounts for 2 extents, but merged together
1596 * they are only 3 extents worth of accounting, so we need to drop in
1597 * this case.
1598 */
1599 old_size = other->end - other->start + 1;
1600 num_extents = count_max_extents(old_size);
1601 old_size = new->end - new->start + 1;
1602 num_extents += count_max_extents(old_size);
1603 if (count_max_extents(new_size) >= num_extents)
1604 return;
1605
1606 spin_lock(&BTRFS_I(inode)->lock);
1607 BTRFS_I(inode)->outstanding_extents--;
1608 spin_unlock(&BTRFS_I(inode)->lock);
1609 }
1610
1611 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1612 struct inode *inode)
1613 {
1614 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1615
1616 spin_lock(&root->delalloc_lock);
1617 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1618 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1619 &root->delalloc_inodes);
1620 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1621 &BTRFS_I(inode)->runtime_flags);
1622 root->nr_delalloc_inodes++;
1623 if (root->nr_delalloc_inodes == 1) {
1624 spin_lock(&fs_info->delalloc_root_lock);
1625 BUG_ON(!list_empty(&root->delalloc_root));
1626 list_add_tail(&root->delalloc_root,
1627 &fs_info->delalloc_roots);
1628 spin_unlock(&fs_info->delalloc_root_lock);
1629 }
1630 }
1631 spin_unlock(&root->delalloc_lock);
1632 }
1633
1634 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1635 struct btrfs_inode *inode)
1636 {
1637 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1638
1639 spin_lock(&root->delalloc_lock);
1640 if (!list_empty(&inode->delalloc_inodes)) {
1641 list_del_init(&inode->delalloc_inodes);
1642 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1643 &inode->runtime_flags);
1644 root->nr_delalloc_inodes--;
1645 if (!root->nr_delalloc_inodes) {
1646 spin_lock(&fs_info->delalloc_root_lock);
1647 BUG_ON(list_empty(&root->delalloc_root));
1648 list_del_init(&root->delalloc_root);
1649 spin_unlock(&fs_info->delalloc_root_lock);
1650 }
1651 }
1652 spin_unlock(&root->delalloc_lock);
1653 }
1654
1655 /*
1656 * extent_io.c set_bit_hook, used to track delayed allocation
1657 * bytes in this file, and to maintain the list of inodes that
1658 * have pending delalloc work to be done.
1659 */
1660 static void btrfs_set_bit_hook(struct inode *inode,
1661 struct extent_state *state, unsigned *bits)
1662 {
1663
1664 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1665
1666 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1667 WARN_ON(1);
1668 /*
1669 * set_bit and clear bit hooks normally require _irqsave/restore
1670 * but in this case, we are only testing for the DELALLOC
1671 * bit, which is only set or cleared with irqs on
1672 */
1673 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1674 struct btrfs_root *root = BTRFS_I(inode)->root;
1675 u64 len = state->end + 1 - state->start;
1676 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1677
1678 if (*bits & EXTENT_FIRST_DELALLOC) {
1679 *bits &= ~EXTENT_FIRST_DELALLOC;
1680 } else {
1681 spin_lock(&BTRFS_I(inode)->lock);
1682 BTRFS_I(inode)->outstanding_extents++;
1683 spin_unlock(&BTRFS_I(inode)->lock);
1684 }
1685
1686 /* For sanity tests */
1687 if (btrfs_is_testing(fs_info))
1688 return;
1689
1690 __percpu_counter_add(&fs_info->delalloc_bytes, len,
1691 fs_info->delalloc_batch);
1692 spin_lock(&BTRFS_I(inode)->lock);
1693 BTRFS_I(inode)->delalloc_bytes += len;
1694 if (*bits & EXTENT_DEFRAG)
1695 BTRFS_I(inode)->defrag_bytes += len;
1696 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1697 &BTRFS_I(inode)->runtime_flags))
1698 btrfs_add_delalloc_inodes(root, inode);
1699 spin_unlock(&BTRFS_I(inode)->lock);
1700 }
1701 }
1702
1703 /*
1704 * extent_io.c clear_bit_hook, see set_bit_hook for why
1705 */
1706 static void btrfs_clear_bit_hook(struct btrfs_inode *inode,
1707 struct extent_state *state,
1708 unsigned *bits)
1709 {
1710 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1711 u64 len = state->end + 1 - state->start;
1712 u32 num_extents = count_max_extents(len);
1713
1714 spin_lock(&inode->lock);
1715 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1716 inode->defrag_bytes -= len;
1717 spin_unlock(&inode->lock);
1718
1719 /*
1720 * set_bit and clear bit hooks normally require _irqsave/restore
1721 * but in this case, we are only testing for the DELALLOC
1722 * bit, which is only set or cleared with irqs on
1723 */
1724 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1725 struct btrfs_root *root = inode->root;
1726 bool do_list = !btrfs_is_free_space_inode(inode);
1727
1728 if (*bits & EXTENT_FIRST_DELALLOC) {
1729 *bits &= ~EXTENT_FIRST_DELALLOC;
1730 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1731 spin_lock(&inode->lock);
1732 inode->outstanding_extents -= num_extents;
1733 spin_unlock(&inode->lock);
1734 }
1735
1736 /*
1737 * We don't reserve metadata space for space cache inodes so we
1738 * don't need to call dellalloc_release_metadata if there is an
1739 * error.
1740 */
1741 if (*bits & EXTENT_DO_ACCOUNTING &&
1742 root != fs_info->tree_root)
1743 btrfs_delalloc_release_metadata(inode, len);
1744
1745 /* For sanity tests. */
1746 if (btrfs_is_testing(fs_info))
1747 return;
1748
1749 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1750 && do_list && !(state->state & EXTENT_NORESERVE)
1751 && (*bits & (EXTENT_DO_ACCOUNTING |
1752 EXTENT_CLEAR_DATA_RESV)))
1753 btrfs_free_reserved_data_space_noquota(
1754 &inode->vfs_inode,
1755 state->start, len);
1756
1757 __percpu_counter_add(&fs_info->delalloc_bytes, -len,
1758 fs_info->delalloc_batch);
1759 spin_lock(&inode->lock);
1760 inode->delalloc_bytes -= len;
1761 if (do_list && inode->delalloc_bytes == 0 &&
1762 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1763 &inode->runtime_flags))
1764 btrfs_del_delalloc_inode(root, inode);
1765 spin_unlock(&inode->lock);
1766 }
1767 }
1768
1769 /*
1770 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1771 * we don't create bios that span stripes or chunks
1772 *
1773 * return 1 if page cannot be merged to bio
1774 * return 0 if page can be merged to bio
1775 * return error otherwise
1776 */
1777 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1778 size_t size, struct bio *bio,
1779 unsigned long bio_flags)
1780 {
1781 struct inode *inode = page->mapping->host;
1782 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1783 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1784 u64 length = 0;
1785 u64 map_length;
1786 int ret;
1787
1788 if (bio_flags & EXTENT_BIO_COMPRESSED)
1789 return 0;
1790
1791 length = bio->bi_iter.bi_size;
1792 map_length = length;
1793 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1794 NULL, 0);
1795 if (ret < 0)
1796 return ret;
1797 if (map_length < length + size)
1798 return 1;
1799 return 0;
1800 }
1801
1802 /*
1803 * in order to insert checksums into the metadata in large chunks,
1804 * we wait until bio submission time. All the pages in the bio are
1805 * checksummed and sums are attached onto the ordered extent record.
1806 *
1807 * At IO completion time the cums attached on the ordered extent record
1808 * are inserted into the btree
1809 */
1810 static int __btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
1811 int mirror_num, unsigned long bio_flags,
1812 u64 bio_offset)
1813 {
1814 int ret = 0;
1815
1816 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1817 BUG_ON(ret); /* -ENOMEM */
1818 return 0;
1819 }
1820
1821 /*
1822 * in order to insert checksums into the metadata in large chunks,
1823 * we wait until bio submission time. All the pages in the bio are
1824 * checksummed and sums are attached onto the ordered extent record.
1825 *
1826 * At IO completion time the cums attached on the ordered extent record
1827 * are inserted into the btree
1828 */
1829 static int __btrfs_submit_bio_done(struct inode *inode, struct bio *bio,
1830 int mirror_num, unsigned long bio_flags,
1831 u64 bio_offset)
1832 {
1833 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1834 int ret;
1835
1836 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1837 if (ret) {
1838 bio->bi_error = ret;
1839 bio_endio(bio);
1840 }
1841 return ret;
1842 }
1843
1844 /*
1845 * extent_io.c submission hook. This does the right thing for csum calculation
1846 * on write, or reading the csums from the tree before a read
1847 */
1848 static int btrfs_submit_bio_hook(struct inode *inode, struct bio *bio,
1849 int mirror_num, unsigned long bio_flags,
1850 u64 bio_offset)
1851 {
1852 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1853 struct btrfs_root *root = BTRFS_I(inode)->root;
1854 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1855 int ret = 0;
1856 int skip_sum;
1857 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1858
1859 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1860
1861 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1862 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1863
1864 if (bio_op(bio) != REQ_OP_WRITE) {
1865 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1866 if (ret)
1867 goto out;
1868
1869 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1870 ret = btrfs_submit_compressed_read(inode, bio,
1871 mirror_num,
1872 bio_flags);
1873 goto out;
1874 } else if (!skip_sum) {
1875 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
1876 if (ret)
1877 goto out;
1878 }
1879 goto mapit;
1880 } else if (async && !skip_sum) {
1881 /* csum items have already been cloned */
1882 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1883 goto mapit;
1884 /* we're doing a write, do the async checksumming */
1885 ret = btrfs_wq_submit_bio(fs_info, inode, bio, mirror_num,
1886 bio_flags, bio_offset,
1887 __btrfs_submit_bio_start,
1888 __btrfs_submit_bio_done);
1889 goto out;
1890 } else if (!skip_sum) {
1891 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1892 if (ret)
1893 goto out;
1894 }
1895
1896 mapit:
1897 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1898
1899 out:
1900 if (ret < 0) {
1901 bio->bi_error = ret;
1902 bio_endio(bio);
1903 }
1904 return ret;
1905 }
1906
1907 /*
1908 * given a list of ordered sums record them in the inode. This happens
1909 * at IO completion time based on sums calculated at bio submission time.
1910 */
1911 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1912 struct inode *inode, struct list_head *list)
1913 {
1914 struct btrfs_ordered_sum *sum;
1915
1916 list_for_each_entry(sum, list, list) {
1917 trans->adding_csums = 1;
1918 btrfs_csum_file_blocks(trans,
1919 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1920 trans->adding_csums = 0;
1921 }
1922 return 0;
1923 }
1924
1925 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1926 struct extent_state **cached_state, int dedupe)
1927 {
1928 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
1929 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1930 cached_state);
1931 }
1932
1933 /* see btrfs_writepage_start_hook for details on why this is required */
1934 struct btrfs_writepage_fixup {
1935 struct page *page;
1936 struct btrfs_work work;
1937 };
1938
1939 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1940 {
1941 struct btrfs_writepage_fixup *fixup;
1942 struct btrfs_ordered_extent *ordered;
1943 struct extent_state *cached_state = NULL;
1944 struct page *page;
1945 struct inode *inode;
1946 u64 page_start;
1947 u64 page_end;
1948 int ret;
1949
1950 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1951 page = fixup->page;
1952 again:
1953 lock_page(page);
1954 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1955 ClearPageChecked(page);
1956 goto out_page;
1957 }
1958
1959 inode = page->mapping->host;
1960 page_start = page_offset(page);
1961 page_end = page_offset(page) + PAGE_SIZE - 1;
1962
1963 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
1964 &cached_state);
1965
1966 /* already ordered? We're done */
1967 if (PagePrivate2(page))
1968 goto out;
1969
1970 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
1971 PAGE_SIZE);
1972 if (ordered) {
1973 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1974 page_end, &cached_state, GFP_NOFS);
1975 unlock_page(page);
1976 btrfs_start_ordered_extent(inode, ordered, 1);
1977 btrfs_put_ordered_extent(ordered);
1978 goto again;
1979 }
1980
1981 ret = btrfs_delalloc_reserve_space(inode, page_start,
1982 PAGE_SIZE);
1983 if (ret) {
1984 mapping_set_error(page->mapping, ret);
1985 end_extent_writepage(page, ret, page_start, page_end);
1986 ClearPageChecked(page);
1987 goto out;
1988 }
1989
1990 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state,
1991 0);
1992 ClearPageChecked(page);
1993 set_page_dirty(page);
1994 out:
1995 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1996 &cached_state, GFP_NOFS);
1997 out_page:
1998 unlock_page(page);
1999 put_page(page);
2000 kfree(fixup);
2001 }
2002
2003 /*
2004 * There are a few paths in the higher layers of the kernel that directly
2005 * set the page dirty bit without asking the filesystem if it is a
2006 * good idea. This causes problems because we want to make sure COW
2007 * properly happens and the data=ordered rules are followed.
2008 *
2009 * In our case any range that doesn't have the ORDERED bit set
2010 * hasn't been properly setup for IO. We kick off an async process
2011 * to fix it up. The async helper will wait for ordered extents, set
2012 * the delalloc bit and make it safe to write the page.
2013 */
2014 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2015 {
2016 struct inode *inode = page->mapping->host;
2017 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2018 struct btrfs_writepage_fixup *fixup;
2019
2020 /* this page is properly in the ordered list */
2021 if (TestClearPagePrivate2(page))
2022 return 0;
2023
2024 if (PageChecked(page))
2025 return -EAGAIN;
2026
2027 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2028 if (!fixup)
2029 return -EAGAIN;
2030
2031 SetPageChecked(page);
2032 get_page(page);
2033 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2034 btrfs_writepage_fixup_worker, NULL, NULL);
2035 fixup->page = page;
2036 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2037 return -EBUSY;
2038 }
2039
2040 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2041 struct inode *inode, u64 file_pos,
2042 u64 disk_bytenr, u64 disk_num_bytes,
2043 u64 num_bytes, u64 ram_bytes,
2044 u8 compression, u8 encryption,
2045 u16 other_encoding, int extent_type)
2046 {
2047 struct btrfs_root *root = BTRFS_I(inode)->root;
2048 struct btrfs_file_extent_item *fi;
2049 struct btrfs_path *path;
2050 struct extent_buffer *leaf;
2051 struct btrfs_key ins;
2052 int extent_inserted = 0;
2053 int ret;
2054
2055 path = btrfs_alloc_path();
2056 if (!path)
2057 return -ENOMEM;
2058
2059 /*
2060 * we may be replacing one extent in the tree with another.
2061 * The new extent is pinned in the extent map, and we don't want
2062 * to drop it from the cache until it is completely in the btree.
2063 *
2064 * So, tell btrfs_drop_extents to leave this extent in the cache.
2065 * the caller is expected to unpin it and allow it to be merged
2066 * with the others.
2067 */
2068 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2069 file_pos + num_bytes, NULL, 0,
2070 1, sizeof(*fi), &extent_inserted);
2071 if (ret)
2072 goto out;
2073
2074 if (!extent_inserted) {
2075 ins.objectid = btrfs_ino(BTRFS_I(inode));
2076 ins.offset = file_pos;
2077 ins.type = BTRFS_EXTENT_DATA_KEY;
2078
2079 path->leave_spinning = 1;
2080 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2081 sizeof(*fi));
2082 if (ret)
2083 goto out;
2084 }
2085 leaf = path->nodes[0];
2086 fi = btrfs_item_ptr(leaf, path->slots[0],
2087 struct btrfs_file_extent_item);
2088 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2089 btrfs_set_file_extent_type(leaf, fi, extent_type);
2090 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2091 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2092 btrfs_set_file_extent_offset(leaf, fi, 0);
2093 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2094 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2095 btrfs_set_file_extent_compression(leaf, fi, compression);
2096 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2097 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2098
2099 btrfs_mark_buffer_dirty(leaf);
2100 btrfs_release_path(path);
2101
2102 inode_add_bytes(inode, num_bytes);
2103
2104 ins.objectid = disk_bytenr;
2105 ins.offset = disk_num_bytes;
2106 ins.type = BTRFS_EXTENT_ITEM_KEY;
2107 ret = btrfs_alloc_reserved_file_extent(trans, root->root_key.objectid,
2108 btrfs_ino(BTRFS_I(inode)), file_pos, ram_bytes, &ins);
2109 /*
2110 * Release the reserved range from inode dirty range map, as it is
2111 * already moved into delayed_ref_head
2112 */
2113 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2114 out:
2115 btrfs_free_path(path);
2116
2117 return ret;
2118 }
2119
2120 /* snapshot-aware defrag */
2121 struct sa_defrag_extent_backref {
2122 struct rb_node node;
2123 struct old_sa_defrag_extent *old;
2124 u64 root_id;
2125 u64 inum;
2126 u64 file_pos;
2127 u64 extent_offset;
2128 u64 num_bytes;
2129 u64 generation;
2130 };
2131
2132 struct old_sa_defrag_extent {
2133 struct list_head list;
2134 struct new_sa_defrag_extent *new;
2135
2136 u64 extent_offset;
2137 u64 bytenr;
2138 u64 offset;
2139 u64 len;
2140 int count;
2141 };
2142
2143 struct new_sa_defrag_extent {
2144 struct rb_root root;
2145 struct list_head head;
2146 struct btrfs_path *path;
2147 struct inode *inode;
2148 u64 file_pos;
2149 u64 len;
2150 u64 bytenr;
2151 u64 disk_len;
2152 u8 compress_type;
2153 };
2154
2155 static int backref_comp(struct sa_defrag_extent_backref *b1,
2156 struct sa_defrag_extent_backref *b2)
2157 {
2158 if (b1->root_id < b2->root_id)
2159 return -1;
2160 else if (b1->root_id > b2->root_id)
2161 return 1;
2162
2163 if (b1->inum < b2->inum)
2164 return -1;
2165 else if (b1->inum > b2->inum)
2166 return 1;
2167
2168 if (b1->file_pos < b2->file_pos)
2169 return -1;
2170 else if (b1->file_pos > b2->file_pos)
2171 return 1;
2172
2173 /*
2174 * [------------------------------] ===> (a range of space)
2175 * |<--->| |<---->| =============> (fs/file tree A)
2176 * |<---------------------------->| ===> (fs/file tree B)
2177 *
2178 * A range of space can refer to two file extents in one tree while
2179 * refer to only one file extent in another tree.
2180 *
2181 * So we may process a disk offset more than one time(two extents in A)
2182 * and locate at the same extent(one extent in B), then insert two same
2183 * backrefs(both refer to the extent in B).
2184 */
2185 return 0;
2186 }
2187
2188 static void backref_insert(struct rb_root *root,
2189 struct sa_defrag_extent_backref *backref)
2190 {
2191 struct rb_node **p = &root->rb_node;
2192 struct rb_node *parent = NULL;
2193 struct sa_defrag_extent_backref *entry;
2194 int ret;
2195
2196 while (*p) {
2197 parent = *p;
2198 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2199
2200 ret = backref_comp(backref, entry);
2201 if (ret < 0)
2202 p = &(*p)->rb_left;
2203 else
2204 p = &(*p)->rb_right;
2205 }
2206
2207 rb_link_node(&backref->node, parent, p);
2208 rb_insert_color(&backref->node, root);
2209 }
2210
2211 /*
2212 * Note the backref might has changed, and in this case we just return 0.
2213 */
2214 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2215 void *ctx)
2216 {
2217 struct btrfs_file_extent_item *extent;
2218 struct old_sa_defrag_extent *old = ctx;
2219 struct new_sa_defrag_extent *new = old->new;
2220 struct btrfs_path *path = new->path;
2221 struct btrfs_key key;
2222 struct btrfs_root *root;
2223 struct sa_defrag_extent_backref *backref;
2224 struct extent_buffer *leaf;
2225 struct inode *inode = new->inode;
2226 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2227 int slot;
2228 int ret;
2229 u64 extent_offset;
2230 u64 num_bytes;
2231
2232 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2233 inum == btrfs_ino(BTRFS_I(inode)))
2234 return 0;
2235
2236 key.objectid = root_id;
2237 key.type = BTRFS_ROOT_ITEM_KEY;
2238 key.offset = (u64)-1;
2239
2240 root = btrfs_read_fs_root_no_name(fs_info, &key);
2241 if (IS_ERR(root)) {
2242 if (PTR_ERR(root) == -ENOENT)
2243 return 0;
2244 WARN_ON(1);
2245 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2246 inum, offset, root_id);
2247 return PTR_ERR(root);
2248 }
2249
2250 key.objectid = inum;
2251 key.type = BTRFS_EXTENT_DATA_KEY;
2252 if (offset > (u64)-1 << 32)
2253 key.offset = 0;
2254 else
2255 key.offset = offset;
2256
2257 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2258 if (WARN_ON(ret < 0))
2259 return ret;
2260 ret = 0;
2261
2262 while (1) {
2263 cond_resched();
2264
2265 leaf = path->nodes[0];
2266 slot = path->slots[0];
2267
2268 if (slot >= btrfs_header_nritems(leaf)) {
2269 ret = btrfs_next_leaf(root, path);
2270 if (ret < 0) {
2271 goto out;
2272 } else if (ret > 0) {
2273 ret = 0;
2274 goto out;
2275 }
2276 continue;
2277 }
2278
2279 path->slots[0]++;
2280
2281 btrfs_item_key_to_cpu(leaf, &key, slot);
2282
2283 if (key.objectid > inum)
2284 goto out;
2285
2286 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2287 continue;
2288
2289 extent = btrfs_item_ptr(leaf, slot,
2290 struct btrfs_file_extent_item);
2291
2292 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2293 continue;
2294
2295 /*
2296 * 'offset' refers to the exact key.offset,
2297 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2298 * (key.offset - extent_offset).
2299 */
2300 if (key.offset != offset)
2301 continue;
2302
2303 extent_offset = btrfs_file_extent_offset(leaf, extent);
2304 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2305
2306 if (extent_offset >= old->extent_offset + old->offset +
2307 old->len || extent_offset + num_bytes <=
2308 old->extent_offset + old->offset)
2309 continue;
2310 break;
2311 }
2312
2313 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2314 if (!backref) {
2315 ret = -ENOENT;
2316 goto out;
2317 }
2318
2319 backref->root_id = root_id;
2320 backref->inum = inum;
2321 backref->file_pos = offset;
2322 backref->num_bytes = num_bytes;
2323 backref->extent_offset = extent_offset;
2324 backref->generation = btrfs_file_extent_generation(leaf, extent);
2325 backref->old = old;
2326 backref_insert(&new->root, backref);
2327 old->count++;
2328 out:
2329 btrfs_release_path(path);
2330 WARN_ON(ret);
2331 return ret;
2332 }
2333
2334 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2335 struct new_sa_defrag_extent *new)
2336 {
2337 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2338 struct old_sa_defrag_extent *old, *tmp;
2339 int ret;
2340
2341 new->path = path;
2342
2343 list_for_each_entry_safe(old, tmp, &new->head, list) {
2344 ret = iterate_inodes_from_logical(old->bytenr +
2345 old->extent_offset, fs_info,
2346 path, record_one_backref,
2347 old);
2348 if (ret < 0 && ret != -ENOENT)
2349 return false;
2350
2351 /* no backref to be processed for this extent */
2352 if (!old->count) {
2353 list_del(&old->list);
2354 kfree(old);
2355 }
2356 }
2357
2358 if (list_empty(&new->head))
2359 return false;
2360
2361 return true;
2362 }
2363
2364 static int relink_is_mergable(struct extent_buffer *leaf,
2365 struct btrfs_file_extent_item *fi,
2366 struct new_sa_defrag_extent *new)
2367 {
2368 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2369 return 0;
2370
2371 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2372 return 0;
2373
2374 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2375 return 0;
2376
2377 if (btrfs_file_extent_encryption(leaf, fi) ||
2378 btrfs_file_extent_other_encoding(leaf, fi))
2379 return 0;
2380
2381 return 1;
2382 }
2383
2384 /*
2385 * Note the backref might has changed, and in this case we just return 0.
2386 */
2387 static noinline int relink_extent_backref(struct btrfs_path *path,
2388 struct sa_defrag_extent_backref *prev,
2389 struct sa_defrag_extent_backref *backref)
2390 {
2391 struct btrfs_file_extent_item *extent;
2392 struct btrfs_file_extent_item *item;
2393 struct btrfs_ordered_extent *ordered;
2394 struct btrfs_trans_handle *trans;
2395 struct btrfs_root *root;
2396 struct btrfs_key key;
2397 struct extent_buffer *leaf;
2398 struct old_sa_defrag_extent *old = backref->old;
2399 struct new_sa_defrag_extent *new = old->new;
2400 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2401 struct inode *inode;
2402 struct extent_state *cached = NULL;
2403 int ret = 0;
2404 u64 start;
2405 u64 len;
2406 u64 lock_start;
2407 u64 lock_end;
2408 bool merge = false;
2409 int index;
2410
2411 if (prev && prev->root_id == backref->root_id &&
2412 prev->inum == backref->inum &&
2413 prev->file_pos + prev->num_bytes == backref->file_pos)
2414 merge = true;
2415
2416 /* step 1: get root */
2417 key.objectid = backref->root_id;
2418 key.type = BTRFS_ROOT_ITEM_KEY;
2419 key.offset = (u64)-1;
2420
2421 index = srcu_read_lock(&fs_info->subvol_srcu);
2422
2423 root = btrfs_read_fs_root_no_name(fs_info, &key);
2424 if (IS_ERR(root)) {
2425 srcu_read_unlock(&fs_info->subvol_srcu, index);
2426 if (PTR_ERR(root) == -ENOENT)
2427 return 0;
2428 return PTR_ERR(root);
2429 }
2430
2431 if (btrfs_root_readonly(root)) {
2432 srcu_read_unlock(&fs_info->subvol_srcu, index);
2433 return 0;
2434 }
2435
2436 /* step 2: get inode */
2437 key.objectid = backref->inum;
2438 key.type = BTRFS_INODE_ITEM_KEY;
2439 key.offset = 0;
2440
2441 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2442 if (IS_ERR(inode)) {
2443 srcu_read_unlock(&fs_info->subvol_srcu, index);
2444 return 0;
2445 }
2446
2447 srcu_read_unlock(&fs_info->subvol_srcu, index);
2448
2449 /* step 3: relink backref */
2450 lock_start = backref->file_pos;
2451 lock_end = backref->file_pos + backref->num_bytes - 1;
2452 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2453 &cached);
2454
2455 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2456 if (ordered) {
2457 btrfs_put_ordered_extent(ordered);
2458 goto out_unlock;
2459 }
2460
2461 trans = btrfs_join_transaction(root);
2462 if (IS_ERR(trans)) {
2463 ret = PTR_ERR(trans);
2464 goto out_unlock;
2465 }
2466
2467 key.objectid = backref->inum;
2468 key.type = BTRFS_EXTENT_DATA_KEY;
2469 key.offset = backref->file_pos;
2470
2471 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2472 if (ret < 0) {
2473 goto out_free_path;
2474 } else if (ret > 0) {
2475 ret = 0;
2476 goto out_free_path;
2477 }
2478
2479 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2480 struct btrfs_file_extent_item);
2481
2482 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2483 backref->generation)
2484 goto out_free_path;
2485
2486 btrfs_release_path(path);
2487
2488 start = backref->file_pos;
2489 if (backref->extent_offset < old->extent_offset + old->offset)
2490 start += old->extent_offset + old->offset -
2491 backref->extent_offset;
2492
2493 len = min(backref->extent_offset + backref->num_bytes,
2494 old->extent_offset + old->offset + old->len);
2495 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2496
2497 ret = btrfs_drop_extents(trans, root, inode, start,
2498 start + len, 1);
2499 if (ret)
2500 goto out_free_path;
2501 again:
2502 key.objectid = btrfs_ino(BTRFS_I(inode));
2503 key.type = BTRFS_EXTENT_DATA_KEY;
2504 key.offset = start;
2505
2506 path->leave_spinning = 1;
2507 if (merge) {
2508 struct btrfs_file_extent_item *fi;
2509 u64 extent_len;
2510 struct btrfs_key found_key;
2511
2512 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2513 if (ret < 0)
2514 goto out_free_path;
2515
2516 path->slots[0]--;
2517 leaf = path->nodes[0];
2518 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2519
2520 fi = btrfs_item_ptr(leaf, path->slots[0],
2521 struct btrfs_file_extent_item);
2522 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2523
2524 if (extent_len + found_key.offset == start &&
2525 relink_is_mergable(leaf, fi, new)) {
2526 btrfs_set_file_extent_num_bytes(leaf, fi,
2527 extent_len + len);
2528 btrfs_mark_buffer_dirty(leaf);
2529 inode_add_bytes(inode, len);
2530
2531 ret = 1;
2532 goto out_free_path;
2533 } else {
2534 merge = false;
2535 btrfs_release_path(path);
2536 goto again;
2537 }
2538 }
2539
2540 ret = btrfs_insert_empty_item(trans, root, path, &key,
2541 sizeof(*extent));
2542 if (ret) {
2543 btrfs_abort_transaction(trans, ret);
2544 goto out_free_path;
2545 }
2546
2547 leaf = path->nodes[0];
2548 item = btrfs_item_ptr(leaf, path->slots[0],
2549 struct btrfs_file_extent_item);
2550 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2551 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2552 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2553 btrfs_set_file_extent_num_bytes(leaf, item, len);
2554 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2555 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2556 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2557 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2558 btrfs_set_file_extent_encryption(leaf, item, 0);
2559 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2560
2561 btrfs_mark_buffer_dirty(leaf);
2562 inode_add_bytes(inode, len);
2563 btrfs_release_path(path);
2564
2565 ret = btrfs_inc_extent_ref(trans, fs_info, new->bytenr,
2566 new->disk_len, 0,
2567 backref->root_id, backref->inum,
2568 new->file_pos); /* start - extent_offset */
2569 if (ret) {
2570 btrfs_abort_transaction(trans, ret);
2571 goto out_free_path;
2572 }
2573
2574 ret = 1;
2575 out_free_path:
2576 btrfs_release_path(path);
2577 path->leave_spinning = 0;
2578 btrfs_end_transaction(trans);
2579 out_unlock:
2580 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2581 &cached, GFP_NOFS);
2582 iput(inode);
2583 return ret;
2584 }
2585
2586 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2587 {
2588 struct old_sa_defrag_extent *old, *tmp;
2589
2590 if (!new)
2591 return;
2592
2593 list_for_each_entry_safe(old, tmp, &new->head, list) {
2594 kfree(old);
2595 }
2596 kfree(new);
2597 }
2598
2599 static void relink_file_extents(struct new_sa_defrag_extent *new)
2600 {
2601 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2602 struct btrfs_path *path;
2603 struct sa_defrag_extent_backref *backref;
2604 struct sa_defrag_extent_backref *prev = NULL;
2605 struct inode *inode;
2606 struct btrfs_root *root;
2607 struct rb_node *node;
2608 int ret;
2609
2610 inode = new->inode;
2611 root = BTRFS_I(inode)->root;
2612
2613 path = btrfs_alloc_path();
2614 if (!path)
2615 return;
2616
2617 if (!record_extent_backrefs(path, new)) {
2618 btrfs_free_path(path);
2619 goto out;
2620 }
2621 btrfs_release_path(path);
2622
2623 while (1) {
2624 node = rb_first(&new->root);
2625 if (!node)
2626 break;
2627 rb_erase(node, &new->root);
2628
2629 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2630
2631 ret = relink_extent_backref(path, prev, backref);
2632 WARN_ON(ret < 0);
2633
2634 kfree(prev);
2635
2636 if (ret == 1)
2637 prev = backref;
2638 else
2639 prev = NULL;
2640 cond_resched();
2641 }
2642 kfree(prev);
2643
2644 btrfs_free_path(path);
2645 out:
2646 free_sa_defrag_extent(new);
2647
2648 atomic_dec(&fs_info->defrag_running);
2649 wake_up(&fs_info->transaction_wait);
2650 }
2651
2652 static struct new_sa_defrag_extent *
2653 record_old_file_extents(struct inode *inode,
2654 struct btrfs_ordered_extent *ordered)
2655 {
2656 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2657 struct btrfs_root *root = BTRFS_I(inode)->root;
2658 struct btrfs_path *path;
2659 struct btrfs_key key;
2660 struct old_sa_defrag_extent *old;
2661 struct new_sa_defrag_extent *new;
2662 int ret;
2663
2664 new = kmalloc(sizeof(*new), GFP_NOFS);
2665 if (!new)
2666 return NULL;
2667
2668 new->inode = inode;
2669 new->file_pos = ordered->file_offset;
2670 new->len = ordered->len;
2671 new->bytenr = ordered->start;
2672 new->disk_len = ordered->disk_len;
2673 new->compress_type = ordered->compress_type;
2674 new->root = RB_ROOT;
2675 INIT_LIST_HEAD(&new->head);
2676
2677 path = btrfs_alloc_path();
2678 if (!path)
2679 goto out_kfree;
2680
2681 key.objectid = btrfs_ino(BTRFS_I(inode));
2682 key.type = BTRFS_EXTENT_DATA_KEY;
2683 key.offset = new->file_pos;
2684
2685 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2686 if (ret < 0)
2687 goto out_free_path;
2688 if (ret > 0 && path->slots[0] > 0)
2689 path->slots[0]--;
2690
2691 /* find out all the old extents for the file range */
2692 while (1) {
2693 struct btrfs_file_extent_item *extent;
2694 struct extent_buffer *l;
2695 int slot;
2696 u64 num_bytes;
2697 u64 offset;
2698 u64 end;
2699 u64 disk_bytenr;
2700 u64 extent_offset;
2701
2702 l = path->nodes[0];
2703 slot = path->slots[0];
2704
2705 if (slot >= btrfs_header_nritems(l)) {
2706 ret = btrfs_next_leaf(root, path);
2707 if (ret < 0)
2708 goto out_free_path;
2709 else if (ret > 0)
2710 break;
2711 continue;
2712 }
2713
2714 btrfs_item_key_to_cpu(l, &key, slot);
2715
2716 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2717 break;
2718 if (key.type != BTRFS_EXTENT_DATA_KEY)
2719 break;
2720 if (key.offset >= new->file_pos + new->len)
2721 break;
2722
2723 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2724
2725 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2726 if (key.offset + num_bytes < new->file_pos)
2727 goto next;
2728
2729 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2730 if (!disk_bytenr)
2731 goto next;
2732
2733 extent_offset = btrfs_file_extent_offset(l, extent);
2734
2735 old = kmalloc(sizeof(*old), GFP_NOFS);
2736 if (!old)
2737 goto out_free_path;
2738
2739 offset = max(new->file_pos, key.offset);
2740 end = min(new->file_pos + new->len, key.offset + num_bytes);
2741
2742 old->bytenr = disk_bytenr;
2743 old->extent_offset = extent_offset;
2744 old->offset = offset - key.offset;
2745 old->len = end - offset;
2746 old->new = new;
2747 old->count = 0;
2748 list_add_tail(&old->list, &new->head);
2749 next:
2750 path->slots[0]++;
2751 cond_resched();
2752 }
2753
2754 btrfs_free_path(path);
2755 atomic_inc(&fs_info->defrag_running);
2756
2757 return new;
2758
2759 out_free_path:
2760 btrfs_free_path(path);
2761 out_kfree:
2762 free_sa_defrag_extent(new);
2763 return NULL;
2764 }
2765
2766 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2767 u64 start, u64 len)
2768 {
2769 struct btrfs_block_group_cache *cache;
2770
2771 cache = btrfs_lookup_block_group(fs_info, start);
2772 ASSERT(cache);
2773
2774 spin_lock(&cache->lock);
2775 cache->delalloc_bytes -= len;
2776 spin_unlock(&cache->lock);
2777
2778 btrfs_put_block_group(cache);
2779 }
2780
2781 /* as ordered data IO finishes, this gets called so we can finish
2782 * an ordered extent if the range of bytes in the file it covers are
2783 * fully written.
2784 */
2785 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2786 {
2787 struct inode *inode = ordered_extent->inode;
2788 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2789 struct btrfs_root *root = BTRFS_I(inode)->root;
2790 struct btrfs_trans_handle *trans = NULL;
2791 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2792 struct extent_state *cached_state = NULL;
2793 struct new_sa_defrag_extent *new = NULL;
2794 int compress_type = 0;
2795 int ret = 0;
2796 u64 logical_len = ordered_extent->len;
2797 bool nolock;
2798 bool truncated = false;
2799
2800 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2801
2802 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2803 ret = -EIO;
2804 goto out;
2805 }
2806
2807 btrfs_free_io_failure_record(BTRFS_I(inode),
2808 ordered_extent->file_offset,
2809 ordered_extent->file_offset +
2810 ordered_extent->len - 1);
2811
2812 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2813 truncated = true;
2814 logical_len = ordered_extent->truncated_len;
2815 /* Truncated the entire extent, don't bother adding */
2816 if (!logical_len)
2817 goto out;
2818 }
2819
2820 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2821 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2822
2823 /*
2824 * For mwrite(mmap + memset to write) case, we still reserve
2825 * space for NOCOW range.
2826 * As NOCOW won't cause a new delayed ref, just free the space
2827 */
2828 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2829 ordered_extent->len);
2830 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2831 if (nolock)
2832 trans = btrfs_join_transaction_nolock(root);
2833 else
2834 trans = btrfs_join_transaction(root);
2835 if (IS_ERR(trans)) {
2836 ret = PTR_ERR(trans);
2837 trans = NULL;
2838 goto out;
2839 }
2840 trans->block_rsv = &fs_info->delalloc_block_rsv;
2841 ret = btrfs_update_inode_fallback(trans, root, inode);
2842 if (ret) /* -ENOMEM or corruption */
2843 btrfs_abort_transaction(trans, ret);
2844 goto out;
2845 }
2846
2847 lock_extent_bits(io_tree, ordered_extent->file_offset,
2848 ordered_extent->file_offset + ordered_extent->len - 1,
2849 &cached_state);
2850
2851 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2852 ordered_extent->file_offset + ordered_extent->len - 1,
2853 EXTENT_DEFRAG, 1, cached_state);
2854 if (ret) {
2855 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2856 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2857 /* the inode is shared */
2858 new = record_old_file_extents(inode, ordered_extent);
2859
2860 clear_extent_bit(io_tree, ordered_extent->file_offset,
2861 ordered_extent->file_offset + ordered_extent->len - 1,
2862 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2863 }
2864
2865 if (nolock)
2866 trans = btrfs_join_transaction_nolock(root);
2867 else
2868 trans = btrfs_join_transaction(root);
2869 if (IS_ERR(trans)) {
2870 ret = PTR_ERR(trans);
2871 trans = NULL;
2872 goto out_unlock;
2873 }
2874
2875 trans->block_rsv = &fs_info->delalloc_block_rsv;
2876
2877 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2878 compress_type = ordered_extent->compress_type;
2879 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2880 BUG_ON(compress_type);
2881 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
2882 ordered_extent->file_offset,
2883 ordered_extent->file_offset +
2884 logical_len);
2885 } else {
2886 BUG_ON(root == fs_info->tree_root);
2887 ret = insert_reserved_file_extent(trans, inode,
2888 ordered_extent->file_offset,
2889 ordered_extent->start,
2890 ordered_extent->disk_len,
2891 logical_len, logical_len,
2892 compress_type, 0, 0,
2893 BTRFS_FILE_EXTENT_REG);
2894 if (!ret)
2895 btrfs_release_delalloc_bytes(fs_info,
2896 ordered_extent->start,
2897 ordered_extent->disk_len);
2898 }
2899 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2900 ordered_extent->file_offset, ordered_extent->len,
2901 trans->transid);
2902 if (ret < 0) {
2903 btrfs_abort_transaction(trans, ret);
2904 goto out_unlock;
2905 }
2906
2907 add_pending_csums(trans, inode, &ordered_extent->list);
2908
2909 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2910 ret = btrfs_update_inode_fallback(trans, root, inode);
2911 if (ret) { /* -ENOMEM or corruption */
2912 btrfs_abort_transaction(trans, ret);
2913 goto out_unlock;
2914 }
2915 ret = 0;
2916 out_unlock:
2917 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2918 ordered_extent->file_offset +
2919 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2920 out:
2921 if (root != fs_info->tree_root)
2922 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2923 ordered_extent->len);
2924 if (trans)
2925 btrfs_end_transaction(trans);
2926
2927 if (ret || truncated) {
2928 u64 start, end;
2929
2930 if (truncated)
2931 start = ordered_extent->file_offset + logical_len;
2932 else
2933 start = ordered_extent->file_offset;
2934 end = ordered_extent->file_offset + ordered_extent->len - 1;
2935 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2936
2937 /* Drop the cache for the part of the extent we didn't write. */
2938 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
2939
2940 /*
2941 * If the ordered extent had an IOERR or something else went
2942 * wrong we need to return the space for this ordered extent
2943 * back to the allocator. We only free the extent in the
2944 * truncated case if we didn't write out the extent at all.
2945 */
2946 if ((ret || !logical_len) &&
2947 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2948 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2949 btrfs_free_reserved_extent(fs_info,
2950 ordered_extent->start,
2951 ordered_extent->disk_len, 1);
2952 }
2953
2954
2955 /*
2956 * This needs to be done to make sure anybody waiting knows we are done
2957 * updating everything for this ordered extent.
2958 */
2959 btrfs_remove_ordered_extent(inode, ordered_extent);
2960
2961 /* for snapshot-aware defrag */
2962 if (new) {
2963 if (ret) {
2964 free_sa_defrag_extent(new);
2965 atomic_dec(&fs_info->defrag_running);
2966 } else {
2967 relink_file_extents(new);
2968 }
2969 }
2970
2971 /* once for us */
2972 btrfs_put_ordered_extent(ordered_extent);
2973 /* once for the tree */
2974 btrfs_put_ordered_extent(ordered_extent);
2975
2976 return ret;
2977 }
2978
2979 static void finish_ordered_fn(struct btrfs_work *work)
2980 {
2981 struct btrfs_ordered_extent *ordered_extent;
2982 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2983 btrfs_finish_ordered_io(ordered_extent);
2984 }
2985
2986 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2987 struct extent_state *state, int uptodate)
2988 {
2989 struct inode *inode = page->mapping->host;
2990 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2991 struct btrfs_ordered_extent *ordered_extent = NULL;
2992 struct btrfs_workqueue *wq;
2993 btrfs_work_func_t func;
2994
2995 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2996
2997 ClearPagePrivate2(page);
2998 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2999 end - start + 1, uptodate))
3000 return 0;
3001
3002 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3003 wq = fs_info->endio_freespace_worker;
3004 func = btrfs_freespace_write_helper;
3005 } else {
3006 wq = fs_info->endio_write_workers;
3007 func = btrfs_endio_write_helper;
3008 }
3009
3010 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3011 NULL);
3012 btrfs_queue_work(wq, &ordered_extent->work);
3013
3014 return 0;
3015 }
3016
3017 static int __readpage_endio_check(struct inode *inode,
3018 struct btrfs_io_bio *io_bio,
3019 int icsum, struct page *page,
3020 int pgoff, u64 start, size_t len)
3021 {
3022 char *kaddr;
3023 u32 csum_expected;
3024 u32 csum = ~(u32)0;
3025
3026 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3027
3028 kaddr = kmap_atomic(page);
3029 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3030 btrfs_csum_final(csum, (u8 *)&csum);
3031 if (csum != csum_expected)
3032 goto zeroit;
3033
3034 kunmap_atomic(kaddr);
3035 return 0;
3036 zeroit:
3037 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3038 io_bio->mirror_num);
3039 memset(kaddr + pgoff, 1, len);
3040 flush_dcache_page(page);
3041 kunmap_atomic(kaddr);
3042 if (csum_expected == 0)
3043 return 0;
3044 return -EIO;
3045 }
3046
3047 /*
3048 * when reads are done, we need to check csums to verify the data is correct
3049 * if there's a match, we allow the bio to finish. If not, the code in
3050 * extent_io.c will try to find good copies for us.
3051 */
3052 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3053 u64 phy_offset, struct page *page,
3054 u64 start, u64 end, int mirror)
3055 {
3056 size_t offset = start - page_offset(page);
3057 struct inode *inode = page->mapping->host;
3058 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3059 struct btrfs_root *root = BTRFS_I(inode)->root;
3060
3061 if (PageChecked(page)) {
3062 ClearPageChecked(page);
3063 return 0;
3064 }
3065
3066 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3067 return 0;
3068
3069 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3070 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3071 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3072 return 0;
3073 }
3074
3075 phy_offset >>= inode->i_sb->s_blocksize_bits;
3076 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3077 start, (size_t)(end - start + 1));
3078 }
3079
3080 void btrfs_add_delayed_iput(struct inode *inode)
3081 {
3082 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3083 struct btrfs_inode *binode = BTRFS_I(inode);
3084
3085 if (atomic_add_unless(&inode->i_count, -1, 1))
3086 return;
3087
3088 spin_lock(&fs_info->delayed_iput_lock);
3089 if (binode->delayed_iput_count == 0) {
3090 ASSERT(list_empty(&binode->delayed_iput));
3091 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3092 } else {
3093 binode->delayed_iput_count++;
3094 }
3095 spin_unlock(&fs_info->delayed_iput_lock);
3096 }
3097
3098 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3099 {
3100
3101 spin_lock(&fs_info->delayed_iput_lock);
3102 while (!list_empty(&fs_info->delayed_iputs)) {
3103 struct btrfs_inode *inode;
3104
3105 inode = list_first_entry(&fs_info->delayed_iputs,
3106 struct btrfs_inode, delayed_iput);
3107 if (inode->delayed_iput_count) {
3108 inode->delayed_iput_count--;
3109 list_move_tail(&inode->delayed_iput,
3110 &fs_info->delayed_iputs);
3111 } else {
3112 list_del_init(&inode->delayed_iput);
3113 }
3114 spin_unlock(&fs_info->delayed_iput_lock);
3115 iput(&inode->vfs_inode);
3116 spin_lock(&fs_info->delayed_iput_lock);
3117 }
3118 spin_unlock(&fs_info->delayed_iput_lock);
3119 }
3120
3121 /*
3122 * This is called in transaction commit time. If there are no orphan
3123 * files in the subvolume, it removes orphan item and frees block_rsv
3124 * structure.
3125 */
3126 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3127 struct btrfs_root *root)
3128 {
3129 struct btrfs_fs_info *fs_info = root->fs_info;
3130 struct btrfs_block_rsv *block_rsv;
3131 int ret;
3132
3133 if (atomic_read(&root->orphan_inodes) ||
3134 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3135 return;
3136
3137 spin_lock(&root->orphan_lock);
3138 if (atomic_read(&root->orphan_inodes)) {
3139 spin_unlock(&root->orphan_lock);
3140 return;
3141 }
3142
3143 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3144 spin_unlock(&root->orphan_lock);
3145 return;
3146 }
3147
3148 block_rsv = root->orphan_block_rsv;
3149 root->orphan_block_rsv = NULL;
3150 spin_unlock(&root->orphan_lock);
3151
3152 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3153 btrfs_root_refs(&root->root_item) > 0) {
3154 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3155 root->root_key.objectid);
3156 if (ret)
3157 btrfs_abort_transaction(trans, ret);
3158 else
3159 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3160 &root->state);
3161 }
3162
3163 if (block_rsv) {
3164 WARN_ON(block_rsv->size > 0);
3165 btrfs_free_block_rsv(fs_info, block_rsv);
3166 }
3167 }
3168
3169 /*
3170 * This creates an orphan entry for the given inode in case something goes
3171 * wrong in the middle of an unlink/truncate.
3172 *
3173 * NOTE: caller of this function should reserve 5 units of metadata for
3174 * this function.
3175 */
3176 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3177 struct btrfs_inode *inode)
3178 {
3179 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3180 struct btrfs_root *root = inode->root;
3181 struct btrfs_block_rsv *block_rsv = NULL;
3182 int reserve = 0;
3183 int insert = 0;
3184 int ret;
3185
3186 if (!root->orphan_block_rsv) {
3187 block_rsv = btrfs_alloc_block_rsv(fs_info,
3188 BTRFS_BLOCK_RSV_TEMP);
3189 if (!block_rsv)
3190 return -ENOMEM;
3191 }
3192
3193 spin_lock(&root->orphan_lock);
3194 if (!root->orphan_block_rsv) {
3195 root->orphan_block_rsv = block_rsv;
3196 } else if (block_rsv) {
3197 btrfs_free_block_rsv(fs_info, block_rsv);
3198 block_rsv = NULL;
3199 }
3200
3201 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3202 &inode->runtime_flags)) {
3203 #if 0
3204 /*
3205 * For proper ENOSPC handling, we should do orphan
3206 * cleanup when mounting. But this introduces backward
3207 * compatibility issue.
3208 */
3209 if (!xchg(&root->orphan_item_inserted, 1))
3210 insert = 2;
3211 else
3212 insert = 1;
3213 #endif
3214 insert = 1;
3215 atomic_inc(&root->orphan_inodes);
3216 }
3217
3218 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3219 &inode->runtime_flags))
3220 reserve = 1;
3221 spin_unlock(&root->orphan_lock);
3222
3223 /* grab metadata reservation from transaction handle */
3224 if (reserve) {
3225 ret = btrfs_orphan_reserve_metadata(trans, inode);
3226 ASSERT(!ret);
3227 if (ret) {
3228 atomic_dec(&root->orphan_inodes);
3229 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3230 &inode->runtime_flags);
3231 if (insert)
3232 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3233 &inode->runtime_flags);
3234 return ret;
3235 }
3236 }
3237
3238 /* insert an orphan item to track this unlinked/truncated file */
3239 if (insert >= 1) {
3240 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3241 if (ret) {
3242 atomic_dec(&root->orphan_inodes);
3243 if (reserve) {
3244 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3245 &inode->runtime_flags);
3246 btrfs_orphan_release_metadata(inode);
3247 }
3248 if (ret != -EEXIST) {
3249 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3250 &inode->runtime_flags);
3251 btrfs_abort_transaction(trans, ret);
3252 return ret;
3253 }
3254 }
3255 ret = 0;
3256 }
3257
3258 /* insert an orphan item to track subvolume contains orphan files */
3259 if (insert >= 2) {
3260 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3261 root->root_key.objectid);
3262 if (ret && ret != -EEXIST) {
3263 btrfs_abort_transaction(trans, ret);
3264 return ret;
3265 }
3266 }
3267 return 0;
3268 }
3269
3270 /*
3271 * We have done the truncate/delete so we can go ahead and remove the orphan
3272 * item for this particular inode.
3273 */
3274 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3275 struct btrfs_inode *inode)
3276 {
3277 struct btrfs_root *root = inode->root;
3278 int delete_item = 0;
3279 int release_rsv = 0;
3280 int ret = 0;
3281
3282 spin_lock(&root->orphan_lock);
3283 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3284 &inode->runtime_flags))
3285 delete_item = 1;
3286
3287 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3288 &inode->runtime_flags))
3289 release_rsv = 1;
3290 spin_unlock(&root->orphan_lock);
3291
3292 if (delete_item) {
3293 atomic_dec(&root->orphan_inodes);
3294 if (trans)
3295 ret = btrfs_del_orphan_item(trans, root,
3296 btrfs_ino(inode));
3297 }
3298
3299 if (release_rsv)
3300 btrfs_orphan_release_metadata(inode);
3301
3302 return ret;
3303 }
3304
3305 /*
3306 * this cleans up any orphans that may be left on the list from the last use
3307 * of this root.
3308 */
3309 int btrfs_orphan_cleanup(struct btrfs_root *root)
3310 {
3311 struct btrfs_fs_info *fs_info = root->fs_info;
3312 struct btrfs_path *path;
3313 struct extent_buffer *leaf;
3314 struct btrfs_key key, found_key;
3315 struct btrfs_trans_handle *trans;
3316 struct inode *inode;
3317 u64 last_objectid = 0;
3318 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3319
3320 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3321 return 0;
3322
3323 path = btrfs_alloc_path();
3324 if (!path) {
3325 ret = -ENOMEM;
3326 goto out;
3327 }
3328 path->reada = READA_BACK;
3329
3330 key.objectid = BTRFS_ORPHAN_OBJECTID;
3331 key.type = BTRFS_ORPHAN_ITEM_KEY;
3332 key.offset = (u64)-1;
3333
3334 while (1) {
3335 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3336 if (ret < 0)
3337 goto out;
3338
3339 /*
3340 * if ret == 0 means we found what we were searching for, which
3341 * is weird, but possible, so only screw with path if we didn't
3342 * find the key and see if we have stuff that matches
3343 */
3344 if (ret > 0) {
3345 ret = 0;
3346 if (path->slots[0] == 0)
3347 break;
3348 path->slots[0]--;
3349 }
3350
3351 /* pull out the item */
3352 leaf = path->nodes[0];
3353 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3354
3355 /* make sure the item matches what we want */
3356 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3357 break;
3358 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3359 break;
3360
3361 /* release the path since we're done with it */
3362 btrfs_release_path(path);
3363
3364 /*
3365 * this is where we are basically btrfs_lookup, without the
3366 * crossing root thing. we store the inode number in the
3367 * offset of the orphan item.
3368 */
3369
3370 if (found_key.offset == last_objectid) {
3371 btrfs_err(fs_info,
3372 "Error removing orphan entry, stopping orphan cleanup");
3373 ret = -EINVAL;
3374 goto out;
3375 }
3376
3377 last_objectid = found_key.offset;
3378
3379 found_key.objectid = found_key.offset;
3380 found_key.type = BTRFS_INODE_ITEM_KEY;
3381 found_key.offset = 0;
3382 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3383 ret = PTR_ERR_OR_ZERO(inode);
3384 if (ret && ret != -ENOENT)
3385 goto out;
3386
3387 if (ret == -ENOENT && root == fs_info->tree_root) {
3388 struct btrfs_root *dead_root;
3389 struct btrfs_fs_info *fs_info = root->fs_info;
3390 int is_dead_root = 0;
3391
3392 /*
3393 * this is an orphan in the tree root. Currently these
3394 * could come from 2 sources:
3395 * a) a snapshot deletion in progress
3396 * b) a free space cache inode
3397 * We need to distinguish those two, as the snapshot
3398 * orphan must not get deleted.
3399 * find_dead_roots already ran before us, so if this
3400 * is a snapshot deletion, we should find the root
3401 * in the dead_roots list
3402 */
3403 spin_lock(&fs_info->trans_lock);
3404 list_for_each_entry(dead_root, &fs_info->dead_roots,
3405 root_list) {
3406 if (dead_root->root_key.objectid ==
3407 found_key.objectid) {
3408 is_dead_root = 1;
3409 break;
3410 }
3411 }
3412 spin_unlock(&fs_info->trans_lock);
3413 if (is_dead_root) {
3414 /* prevent this orphan from being found again */
3415 key.offset = found_key.objectid - 1;
3416 continue;
3417 }
3418 }
3419 /*
3420 * Inode is already gone but the orphan item is still there,
3421 * kill the orphan item.
3422 */
3423 if (ret == -ENOENT) {
3424 trans = btrfs_start_transaction(root, 1);
3425 if (IS_ERR(trans)) {
3426 ret = PTR_ERR(trans);
3427 goto out;
3428 }
3429 btrfs_debug(fs_info, "auto deleting %Lu",
3430 found_key.objectid);
3431 ret = btrfs_del_orphan_item(trans, root,
3432 found_key.objectid);
3433 btrfs_end_transaction(trans);
3434 if (ret)
3435 goto out;
3436 continue;
3437 }
3438
3439 /*
3440 * add this inode to the orphan list so btrfs_orphan_del does
3441 * the proper thing when we hit it
3442 */
3443 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3444 &BTRFS_I(inode)->runtime_flags);
3445 atomic_inc(&root->orphan_inodes);
3446
3447 /* if we have links, this was a truncate, lets do that */
3448 if (inode->i_nlink) {
3449 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3450 iput(inode);
3451 continue;
3452 }
3453 nr_truncate++;
3454
3455 /* 1 for the orphan item deletion. */
3456 trans = btrfs_start_transaction(root, 1);
3457 if (IS_ERR(trans)) {
3458 iput(inode);
3459 ret = PTR_ERR(trans);
3460 goto out;
3461 }
3462 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3463 btrfs_end_transaction(trans);
3464 if (ret) {
3465 iput(inode);
3466 goto out;
3467 }
3468
3469 ret = btrfs_truncate(inode);
3470 if (ret)
3471 btrfs_orphan_del(NULL, BTRFS_I(inode));
3472 } else {
3473 nr_unlink++;
3474 }
3475
3476 /* this will do delete_inode and everything for us */
3477 iput(inode);
3478 if (ret)
3479 goto out;
3480 }
3481 /* release the path since we're done with it */
3482 btrfs_release_path(path);
3483
3484 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3485
3486 if (root->orphan_block_rsv)
3487 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3488 (u64)-1);
3489
3490 if (root->orphan_block_rsv ||
3491 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3492 trans = btrfs_join_transaction(root);
3493 if (!IS_ERR(trans))
3494 btrfs_end_transaction(trans);
3495 }
3496
3497 if (nr_unlink)
3498 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3499 if (nr_truncate)
3500 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3501
3502 out:
3503 if (ret)
3504 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3505 btrfs_free_path(path);
3506 return ret;
3507 }
3508
3509 /*
3510 * very simple check to peek ahead in the leaf looking for xattrs. If we
3511 * don't find any xattrs, we know there can't be any acls.
3512 *
3513 * slot is the slot the inode is in, objectid is the objectid of the inode
3514 */
3515 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3516 int slot, u64 objectid,
3517 int *first_xattr_slot)
3518 {
3519 u32 nritems = btrfs_header_nritems(leaf);
3520 struct btrfs_key found_key;
3521 static u64 xattr_access = 0;
3522 static u64 xattr_default = 0;
3523 int scanned = 0;
3524
3525 if (!xattr_access) {
3526 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3527 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3528 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3529 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3530 }
3531
3532 slot++;
3533 *first_xattr_slot = -1;
3534 while (slot < nritems) {
3535 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3536
3537 /* we found a different objectid, there must not be acls */
3538 if (found_key.objectid != objectid)
3539 return 0;
3540
3541 /* we found an xattr, assume we've got an acl */
3542 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3543 if (*first_xattr_slot == -1)
3544 *first_xattr_slot = slot;
3545 if (found_key.offset == xattr_access ||
3546 found_key.offset == xattr_default)
3547 return 1;
3548 }
3549
3550 /*
3551 * we found a key greater than an xattr key, there can't
3552 * be any acls later on
3553 */
3554 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3555 return 0;
3556
3557 slot++;
3558 scanned++;
3559
3560 /*
3561 * it goes inode, inode backrefs, xattrs, extents,
3562 * so if there are a ton of hard links to an inode there can
3563 * be a lot of backrefs. Don't waste time searching too hard,
3564 * this is just an optimization
3565 */
3566 if (scanned >= 8)
3567 break;
3568 }
3569 /* we hit the end of the leaf before we found an xattr or
3570 * something larger than an xattr. We have to assume the inode
3571 * has acls
3572 */
3573 if (*first_xattr_slot == -1)
3574 *first_xattr_slot = slot;
3575 return 1;
3576 }
3577
3578 /*
3579 * read an inode from the btree into the in-memory inode
3580 */
3581 static int btrfs_read_locked_inode(struct inode *inode)
3582 {
3583 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3584 struct btrfs_path *path;
3585 struct extent_buffer *leaf;
3586 struct btrfs_inode_item *inode_item;
3587 struct btrfs_root *root = BTRFS_I(inode)->root;
3588 struct btrfs_key location;
3589 unsigned long ptr;
3590 int maybe_acls;
3591 u32 rdev;
3592 int ret;
3593 bool filled = false;
3594 int first_xattr_slot;
3595
3596 ret = btrfs_fill_inode(inode, &rdev);
3597 if (!ret)
3598 filled = true;
3599
3600 path = btrfs_alloc_path();
3601 if (!path) {
3602 ret = -ENOMEM;
3603 goto make_bad;
3604 }
3605
3606 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3607
3608 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3609 if (ret) {
3610 if (ret > 0)
3611 ret = -ENOENT;
3612 goto make_bad;
3613 }
3614
3615 leaf = path->nodes[0];
3616
3617 if (filled)
3618 goto cache_index;
3619
3620 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3621 struct btrfs_inode_item);
3622 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3623 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3624 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3625 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3626 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3627
3628 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3629 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3630
3631 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3632 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3633
3634 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3635 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3636
3637 BTRFS_I(inode)->i_otime.tv_sec =
3638 btrfs_timespec_sec(leaf, &inode_item->otime);
3639 BTRFS_I(inode)->i_otime.tv_nsec =
3640 btrfs_timespec_nsec(leaf, &inode_item->otime);
3641
3642 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3643 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3644 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3645
3646 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3647 inode->i_generation = BTRFS_I(inode)->generation;
3648 inode->i_rdev = 0;
3649 rdev = btrfs_inode_rdev(leaf, inode_item);
3650
3651 BTRFS_I(inode)->index_cnt = (u64)-1;
3652 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3653
3654 cache_index:
3655 /*
3656 * If we were modified in the current generation and evicted from memory
3657 * and then re-read we need to do a full sync since we don't have any
3658 * idea about which extents were modified before we were evicted from
3659 * cache.
3660 *
3661 * This is required for both inode re-read from disk and delayed inode
3662 * in delayed_nodes_tree.
3663 */
3664 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3665 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3666 &BTRFS_I(inode)->runtime_flags);
3667
3668 /*
3669 * We don't persist the id of the transaction where an unlink operation
3670 * against the inode was last made. So here we assume the inode might
3671 * have been evicted, and therefore the exact value of last_unlink_trans
3672 * lost, and set it to last_trans to avoid metadata inconsistencies
3673 * between the inode and its parent if the inode is fsync'ed and the log
3674 * replayed. For example, in the scenario:
3675 *
3676 * touch mydir/foo
3677 * ln mydir/foo mydir/bar
3678 * sync
3679 * unlink mydir/bar
3680 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3681 * xfs_io -c fsync mydir/foo
3682 * <power failure>
3683 * mount fs, triggers fsync log replay
3684 *
3685 * We must make sure that when we fsync our inode foo we also log its
3686 * parent inode, otherwise after log replay the parent still has the
3687 * dentry with the "bar" name but our inode foo has a link count of 1
3688 * and doesn't have an inode ref with the name "bar" anymore.
3689 *
3690 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3691 * but it guarantees correctness at the expense of occasional full
3692 * transaction commits on fsync if our inode is a directory, or if our
3693 * inode is not a directory, logging its parent unnecessarily.
3694 */
3695 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3696
3697 path->slots[0]++;
3698 if (inode->i_nlink != 1 ||
3699 path->slots[0] >= btrfs_header_nritems(leaf))
3700 goto cache_acl;
3701
3702 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3703 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3704 goto cache_acl;
3705
3706 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3707 if (location.type == BTRFS_INODE_REF_KEY) {
3708 struct btrfs_inode_ref *ref;
3709
3710 ref = (struct btrfs_inode_ref *)ptr;
3711 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3712 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3713 struct btrfs_inode_extref *extref;
3714
3715 extref = (struct btrfs_inode_extref *)ptr;
3716 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3717 extref);
3718 }
3719 cache_acl:
3720 /*
3721 * try to precache a NULL acl entry for files that don't have
3722 * any xattrs or acls
3723 */
3724 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3725 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3726 if (first_xattr_slot != -1) {
3727 path->slots[0] = first_xattr_slot;
3728 ret = btrfs_load_inode_props(inode, path);
3729 if (ret)
3730 btrfs_err(fs_info,
3731 "error loading props for ino %llu (root %llu): %d",
3732 btrfs_ino(BTRFS_I(inode)),
3733 root->root_key.objectid, ret);
3734 }
3735 btrfs_free_path(path);
3736
3737 if (!maybe_acls)
3738 cache_no_acl(inode);
3739
3740 switch (inode->i_mode & S_IFMT) {
3741 case S_IFREG:
3742 inode->i_mapping->a_ops = &btrfs_aops;
3743 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3744 inode->i_fop = &btrfs_file_operations;
3745 inode->i_op = &btrfs_file_inode_operations;
3746 break;
3747 case S_IFDIR:
3748 inode->i_fop = &btrfs_dir_file_operations;
3749 inode->i_op = &btrfs_dir_inode_operations;
3750 break;
3751 case S_IFLNK:
3752 inode->i_op = &btrfs_symlink_inode_operations;
3753 inode_nohighmem(inode);
3754 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3755 break;
3756 default:
3757 inode->i_op = &btrfs_special_inode_operations;
3758 init_special_inode(inode, inode->i_mode, rdev);
3759 break;
3760 }
3761
3762 btrfs_update_iflags(inode);
3763 return 0;
3764
3765 make_bad:
3766 btrfs_free_path(path);
3767 make_bad_inode(inode);
3768 return ret;
3769 }
3770
3771 /*
3772 * given a leaf and an inode, copy the inode fields into the leaf
3773 */
3774 static void fill_inode_item(struct btrfs_trans_handle *trans,
3775 struct extent_buffer *leaf,
3776 struct btrfs_inode_item *item,
3777 struct inode *inode)
3778 {
3779 struct btrfs_map_token token;
3780
3781 btrfs_init_map_token(&token);
3782
3783 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3784 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3785 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3786 &token);
3787 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3788 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3789
3790 btrfs_set_token_timespec_sec(leaf, &item->atime,
3791 inode->i_atime.tv_sec, &token);
3792 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3793 inode->i_atime.tv_nsec, &token);
3794
3795 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3796 inode->i_mtime.tv_sec, &token);
3797 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3798 inode->i_mtime.tv_nsec, &token);
3799
3800 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3801 inode->i_ctime.tv_sec, &token);
3802 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3803 inode->i_ctime.tv_nsec, &token);
3804
3805 btrfs_set_token_timespec_sec(leaf, &item->otime,
3806 BTRFS_I(inode)->i_otime.tv_sec, &token);
3807 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3808 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3809
3810 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3811 &token);
3812 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3813 &token);
3814 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3815 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3816 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3817 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3818 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3819 }
3820
3821 /*
3822 * copy everything in the in-memory inode into the btree.
3823 */
3824 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3825 struct btrfs_root *root, struct inode *inode)
3826 {
3827 struct btrfs_inode_item *inode_item;
3828 struct btrfs_path *path;
3829 struct extent_buffer *leaf;
3830 int ret;
3831
3832 path = btrfs_alloc_path();
3833 if (!path)
3834 return -ENOMEM;
3835
3836 path->leave_spinning = 1;
3837 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3838 1);
3839 if (ret) {
3840 if (ret > 0)
3841 ret = -ENOENT;
3842 goto failed;
3843 }
3844
3845 leaf = path->nodes[0];
3846 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3847 struct btrfs_inode_item);
3848
3849 fill_inode_item(trans, leaf, inode_item, inode);
3850 btrfs_mark_buffer_dirty(leaf);
3851 btrfs_set_inode_last_trans(trans, inode);
3852 ret = 0;
3853 failed:
3854 btrfs_free_path(path);
3855 return ret;
3856 }
3857
3858 /*
3859 * copy everything in the in-memory inode into the btree.
3860 */
3861 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3862 struct btrfs_root *root, struct inode *inode)
3863 {
3864 struct btrfs_fs_info *fs_info = root->fs_info;
3865 int ret;
3866
3867 /*
3868 * If the inode is a free space inode, we can deadlock during commit
3869 * if we put it into the delayed code.
3870 *
3871 * The data relocation inode should also be directly updated
3872 * without delay
3873 */
3874 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3875 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3876 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3877 btrfs_update_root_times(trans, root);
3878
3879 ret = btrfs_delayed_update_inode(trans, root, inode);
3880 if (!ret)
3881 btrfs_set_inode_last_trans(trans, inode);
3882 return ret;
3883 }
3884
3885 return btrfs_update_inode_item(trans, root, inode);
3886 }
3887
3888 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3889 struct btrfs_root *root,
3890 struct inode *inode)
3891 {
3892 int ret;
3893
3894 ret = btrfs_update_inode(trans, root, inode);
3895 if (ret == -ENOSPC)
3896 return btrfs_update_inode_item(trans, root, inode);
3897 return ret;
3898 }
3899
3900 /*
3901 * unlink helper that gets used here in inode.c and in the tree logging
3902 * recovery code. It remove a link in a directory with a given name, and
3903 * also drops the back refs in the inode to the directory
3904 */
3905 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3906 struct btrfs_root *root,
3907 struct btrfs_inode *dir,
3908 struct btrfs_inode *inode,
3909 const char *name, int name_len)
3910 {
3911 struct btrfs_fs_info *fs_info = root->fs_info;
3912 struct btrfs_path *path;
3913 int ret = 0;
3914 struct extent_buffer *leaf;
3915 struct btrfs_dir_item *di;
3916 struct btrfs_key key;
3917 u64 index;
3918 u64 ino = btrfs_ino(inode);
3919 u64 dir_ino = btrfs_ino(dir);
3920
3921 path = btrfs_alloc_path();
3922 if (!path) {
3923 ret = -ENOMEM;
3924 goto out;
3925 }
3926
3927 path->leave_spinning = 1;
3928 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3929 name, name_len, -1);
3930 if (IS_ERR(di)) {
3931 ret = PTR_ERR(di);
3932 goto err;
3933 }
3934 if (!di) {
3935 ret = -ENOENT;
3936 goto err;
3937 }
3938 leaf = path->nodes[0];
3939 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3940 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3941 if (ret)
3942 goto err;
3943 btrfs_release_path(path);
3944
3945 /*
3946 * If we don't have dir index, we have to get it by looking up
3947 * the inode ref, since we get the inode ref, remove it directly,
3948 * it is unnecessary to do delayed deletion.
3949 *
3950 * But if we have dir index, needn't search inode ref to get it.
3951 * Since the inode ref is close to the inode item, it is better
3952 * that we delay to delete it, and just do this deletion when
3953 * we update the inode item.
3954 */
3955 if (inode->dir_index) {
3956 ret = btrfs_delayed_delete_inode_ref(inode);
3957 if (!ret) {
3958 index = inode->dir_index;
3959 goto skip_backref;
3960 }
3961 }
3962
3963 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3964 dir_ino, &index);
3965 if (ret) {
3966 btrfs_info(fs_info,
3967 "failed to delete reference to %.*s, inode %llu parent %llu",
3968 name_len, name, ino, dir_ino);
3969 btrfs_abort_transaction(trans, ret);
3970 goto err;
3971 }
3972 skip_backref:
3973 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
3974 if (ret) {
3975 btrfs_abort_transaction(trans, ret);
3976 goto err;
3977 }
3978
3979 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3980 dir_ino);
3981 if (ret != 0 && ret != -ENOENT) {
3982 btrfs_abort_transaction(trans, ret);
3983 goto err;
3984 }
3985
3986 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3987 index);
3988 if (ret == -ENOENT)
3989 ret = 0;
3990 else if (ret)
3991 btrfs_abort_transaction(trans, ret);
3992 err:
3993 btrfs_free_path(path);
3994 if (ret)
3995 goto out;
3996
3997 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3998 inode_inc_iversion(&inode->vfs_inode);
3999 inode_inc_iversion(&dir->vfs_inode);
4000 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4001 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4002 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4003 out:
4004 return ret;
4005 }
4006
4007 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4008 struct btrfs_root *root,
4009 struct btrfs_inode *dir, struct btrfs_inode *inode,
4010 const char *name, int name_len)
4011 {
4012 int ret;
4013 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4014 if (!ret) {
4015 drop_nlink(&inode->vfs_inode);
4016 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4017 }
4018 return ret;
4019 }
4020
4021 /*
4022 * helper to start transaction for unlink and rmdir.
4023 *
4024 * unlink and rmdir are special in btrfs, they do not always free space, so
4025 * if we cannot make our reservations the normal way try and see if there is
4026 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4027 * allow the unlink to occur.
4028 */
4029 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4030 {
4031 struct btrfs_root *root = BTRFS_I(dir)->root;
4032
4033 /*
4034 * 1 for the possible orphan item
4035 * 1 for the dir item
4036 * 1 for the dir index
4037 * 1 for the inode ref
4038 * 1 for the inode
4039 */
4040 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4041 }
4042
4043 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4044 {
4045 struct btrfs_root *root = BTRFS_I(dir)->root;
4046 struct btrfs_trans_handle *trans;
4047 struct inode *inode = d_inode(dentry);
4048 int ret;
4049
4050 trans = __unlink_start_trans(dir);
4051 if (IS_ERR(trans))
4052 return PTR_ERR(trans);
4053
4054 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4055 0);
4056
4057 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4058 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4059 dentry->d_name.len);
4060 if (ret)
4061 goto out;
4062
4063 if (inode->i_nlink == 0) {
4064 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4065 if (ret)
4066 goto out;
4067 }
4068
4069 out:
4070 btrfs_end_transaction(trans);
4071 btrfs_btree_balance_dirty(root->fs_info);
4072 return ret;
4073 }
4074
4075 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4076 struct btrfs_root *root,
4077 struct inode *dir, u64 objectid,
4078 const char *name, int name_len)
4079 {
4080 struct btrfs_fs_info *fs_info = root->fs_info;
4081 struct btrfs_path *path;
4082 struct extent_buffer *leaf;
4083 struct btrfs_dir_item *di;
4084 struct btrfs_key key;
4085 u64 index;
4086 int ret;
4087 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4088
4089 path = btrfs_alloc_path();
4090 if (!path)
4091 return -ENOMEM;
4092
4093 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4094 name, name_len, -1);
4095 if (IS_ERR_OR_NULL(di)) {
4096 if (!di)
4097 ret = -ENOENT;
4098 else
4099 ret = PTR_ERR(di);
4100 goto out;
4101 }
4102
4103 leaf = path->nodes[0];
4104 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4105 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4106 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4107 if (ret) {
4108 btrfs_abort_transaction(trans, ret);
4109 goto out;
4110 }
4111 btrfs_release_path(path);
4112
4113 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4114 root->root_key.objectid, dir_ino,
4115 &index, name, name_len);
4116 if (ret < 0) {
4117 if (ret != -ENOENT) {
4118 btrfs_abort_transaction(trans, ret);
4119 goto out;
4120 }
4121 di = btrfs_search_dir_index_item(root, path, dir_ino,
4122 name, name_len);
4123 if (IS_ERR_OR_NULL(di)) {
4124 if (!di)
4125 ret = -ENOENT;
4126 else
4127 ret = PTR_ERR(di);
4128 btrfs_abort_transaction(trans, ret);
4129 goto out;
4130 }
4131
4132 leaf = path->nodes[0];
4133 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4134 btrfs_release_path(path);
4135 index = key.offset;
4136 }
4137 btrfs_release_path(path);
4138
4139 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4140 if (ret) {
4141 btrfs_abort_transaction(trans, ret);
4142 goto out;
4143 }
4144
4145 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4146 inode_inc_iversion(dir);
4147 dir->i_mtime = dir->i_ctime = current_time(dir);
4148 ret = btrfs_update_inode_fallback(trans, root, dir);
4149 if (ret)
4150 btrfs_abort_transaction(trans, ret);
4151 out:
4152 btrfs_free_path(path);
4153 return ret;
4154 }
4155
4156 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4157 {
4158 struct inode *inode = d_inode(dentry);
4159 int err = 0;
4160 struct btrfs_root *root = BTRFS_I(dir)->root;
4161 struct btrfs_trans_handle *trans;
4162 u64 last_unlink_trans;
4163
4164 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4165 return -ENOTEMPTY;
4166 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4167 return -EPERM;
4168
4169 trans = __unlink_start_trans(dir);
4170 if (IS_ERR(trans))
4171 return PTR_ERR(trans);
4172
4173 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4174 err = btrfs_unlink_subvol(trans, root, dir,
4175 BTRFS_I(inode)->location.objectid,
4176 dentry->d_name.name,
4177 dentry->d_name.len);
4178 goto out;
4179 }
4180
4181 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4182 if (err)
4183 goto out;
4184
4185 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4186
4187 /* now the directory is empty */
4188 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4189 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4190 dentry->d_name.len);
4191 if (!err) {
4192 btrfs_i_size_write(BTRFS_I(inode), 0);
4193 /*
4194 * Propagate the last_unlink_trans value of the deleted dir to
4195 * its parent directory. This is to prevent an unrecoverable
4196 * log tree in the case we do something like this:
4197 * 1) create dir foo
4198 * 2) create snapshot under dir foo
4199 * 3) delete the snapshot
4200 * 4) rmdir foo
4201 * 5) mkdir foo
4202 * 6) fsync foo or some file inside foo
4203 */
4204 if (last_unlink_trans >= trans->transid)
4205 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4206 }
4207 out:
4208 btrfs_end_transaction(trans);
4209 btrfs_btree_balance_dirty(root->fs_info);
4210
4211 return err;
4212 }
4213
4214 static int truncate_space_check(struct btrfs_trans_handle *trans,
4215 struct btrfs_root *root,
4216 u64 bytes_deleted)
4217 {
4218 struct btrfs_fs_info *fs_info = root->fs_info;
4219 int ret;
4220
4221 /*
4222 * This is only used to apply pressure to the enospc system, we don't
4223 * intend to use this reservation at all.
4224 */
4225 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4226 bytes_deleted *= fs_info->nodesize;
4227 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4228 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4229 if (!ret) {
4230 trace_btrfs_space_reservation(fs_info, "transaction",
4231 trans->transid,
4232 bytes_deleted, 1);
4233 trans->bytes_reserved += bytes_deleted;
4234 }
4235 return ret;
4236
4237 }
4238
4239 static int truncate_inline_extent(struct inode *inode,
4240 struct btrfs_path *path,
4241 struct btrfs_key *found_key,
4242 const u64 item_end,
4243 const u64 new_size)
4244 {
4245 struct extent_buffer *leaf = path->nodes[0];
4246 int slot = path->slots[0];
4247 struct btrfs_file_extent_item *fi;
4248 u32 size = (u32)(new_size - found_key->offset);
4249 struct btrfs_root *root = BTRFS_I(inode)->root;
4250
4251 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4252
4253 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4254 loff_t offset = new_size;
4255 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4256
4257 /*
4258 * Zero out the remaining of the last page of our inline extent,
4259 * instead of directly truncating our inline extent here - that
4260 * would be much more complex (decompressing all the data, then
4261 * compressing the truncated data, which might be bigger than
4262 * the size of the inline extent, resize the extent, etc).
4263 * We release the path because to get the page we might need to
4264 * read the extent item from disk (data not in the page cache).
4265 */
4266 btrfs_release_path(path);
4267 return btrfs_truncate_block(inode, offset, page_end - offset,
4268 0);
4269 }
4270
4271 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4272 size = btrfs_file_extent_calc_inline_size(size);
4273 btrfs_truncate_item(root->fs_info, path, size, 1);
4274
4275 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4276 inode_sub_bytes(inode, item_end + 1 - new_size);
4277
4278 return 0;
4279 }
4280
4281 /*
4282 * this can truncate away extent items, csum items and directory items.
4283 * It starts at a high offset and removes keys until it can't find
4284 * any higher than new_size
4285 *
4286 * csum items that cross the new i_size are truncated to the new size
4287 * as well.
4288 *
4289 * min_type is the minimum key type to truncate down to. If set to 0, this
4290 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4291 */
4292 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4293 struct btrfs_root *root,
4294 struct inode *inode,
4295 u64 new_size, u32 min_type)
4296 {
4297 struct btrfs_fs_info *fs_info = root->fs_info;
4298 struct btrfs_path *path;
4299 struct extent_buffer *leaf;
4300 struct btrfs_file_extent_item *fi;
4301 struct btrfs_key key;
4302 struct btrfs_key found_key;
4303 u64 extent_start = 0;
4304 u64 extent_num_bytes = 0;
4305 u64 extent_offset = 0;
4306 u64 item_end = 0;
4307 u64 last_size = new_size;
4308 u32 found_type = (u8)-1;
4309 int found_extent;
4310 int del_item;
4311 int pending_del_nr = 0;
4312 int pending_del_slot = 0;
4313 int extent_type = -1;
4314 int ret;
4315 int err = 0;
4316 u64 ino = btrfs_ino(BTRFS_I(inode));
4317 u64 bytes_deleted = 0;
4318 bool be_nice = 0;
4319 bool should_throttle = 0;
4320 bool should_end = 0;
4321
4322 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4323
4324 /*
4325 * for non-free space inodes and ref cows, we want to back off from
4326 * time to time
4327 */
4328 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4329 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4330 be_nice = 1;
4331
4332 path = btrfs_alloc_path();
4333 if (!path)
4334 return -ENOMEM;
4335 path->reada = READA_BACK;
4336
4337 /*
4338 * We want to drop from the next block forward in case this new size is
4339 * not block aligned since we will be keeping the last block of the
4340 * extent just the way it is.
4341 */
4342 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4343 root == fs_info->tree_root)
4344 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4345 fs_info->sectorsize),
4346 (u64)-1, 0);
4347
4348 /*
4349 * This function is also used to drop the items in the log tree before
4350 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4351 * it is used to drop the loged items. So we shouldn't kill the delayed
4352 * items.
4353 */
4354 if (min_type == 0 && root == BTRFS_I(inode)->root)
4355 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4356
4357 key.objectid = ino;
4358 key.offset = (u64)-1;
4359 key.type = (u8)-1;
4360
4361 search_again:
4362 /*
4363 * with a 16K leaf size and 128MB extents, you can actually queue
4364 * up a huge file in a single leaf. Most of the time that
4365 * bytes_deleted is > 0, it will be huge by the time we get here
4366 */
4367 if (be_nice && bytes_deleted > SZ_32M) {
4368 if (btrfs_should_end_transaction(trans)) {
4369 err = -EAGAIN;
4370 goto error;
4371 }
4372 }
4373
4374
4375 path->leave_spinning = 1;
4376 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4377 if (ret < 0) {
4378 err = ret;
4379 goto out;
4380 }
4381
4382 if (ret > 0) {
4383 /* there are no items in the tree for us to truncate, we're
4384 * done
4385 */
4386 if (path->slots[0] == 0)
4387 goto out;
4388 path->slots[0]--;
4389 }
4390
4391 while (1) {
4392 fi = NULL;
4393 leaf = path->nodes[0];
4394 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4395 found_type = found_key.type;
4396
4397 if (found_key.objectid != ino)
4398 break;
4399
4400 if (found_type < min_type)
4401 break;
4402
4403 item_end = found_key.offset;
4404 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4405 fi = btrfs_item_ptr(leaf, path->slots[0],
4406 struct btrfs_file_extent_item);
4407 extent_type = btrfs_file_extent_type(leaf, fi);
4408 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4409 item_end +=
4410 btrfs_file_extent_num_bytes(leaf, fi);
4411 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4412 item_end += btrfs_file_extent_inline_len(leaf,
4413 path->slots[0], fi);
4414 }
4415 item_end--;
4416 }
4417 if (found_type > min_type) {
4418 del_item = 1;
4419 } else {
4420 if (item_end < new_size) {
4421 /*
4422 * With NO_HOLES mode, for the following mapping
4423 *
4424 * [0-4k][hole][8k-12k]
4425 *
4426 * if truncating isize down to 6k, it ends up
4427 * isize being 8k.
4428 */
4429 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
4430 last_size = new_size;
4431 break;
4432 }
4433 if (found_key.offset >= new_size)
4434 del_item = 1;
4435 else
4436 del_item = 0;
4437 }
4438 found_extent = 0;
4439 /* FIXME, shrink the extent if the ref count is only 1 */
4440 if (found_type != BTRFS_EXTENT_DATA_KEY)
4441 goto delete;
4442
4443 if (del_item)
4444 last_size = found_key.offset;
4445 else
4446 last_size = new_size;
4447
4448 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4449 u64 num_dec;
4450 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4451 if (!del_item) {
4452 u64 orig_num_bytes =
4453 btrfs_file_extent_num_bytes(leaf, fi);
4454 extent_num_bytes = ALIGN(new_size -
4455 found_key.offset,
4456 fs_info->sectorsize);
4457 btrfs_set_file_extent_num_bytes(leaf, fi,
4458 extent_num_bytes);
4459 num_dec = (orig_num_bytes -
4460 extent_num_bytes);
4461 if (test_bit(BTRFS_ROOT_REF_COWS,
4462 &root->state) &&
4463 extent_start != 0)
4464 inode_sub_bytes(inode, num_dec);
4465 btrfs_mark_buffer_dirty(leaf);
4466 } else {
4467 extent_num_bytes =
4468 btrfs_file_extent_disk_num_bytes(leaf,
4469 fi);
4470 extent_offset = found_key.offset -
4471 btrfs_file_extent_offset(leaf, fi);
4472
4473 /* FIXME blocksize != 4096 */
4474 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4475 if (extent_start != 0) {
4476 found_extent = 1;
4477 if (test_bit(BTRFS_ROOT_REF_COWS,
4478 &root->state))
4479 inode_sub_bytes(inode, num_dec);
4480 }
4481 }
4482 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4483 /*
4484 * we can't truncate inline items that have had
4485 * special encodings
4486 */
4487 if (!del_item &&
4488 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4489 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4490
4491 /*
4492 * Need to release path in order to truncate a
4493 * compressed extent. So delete any accumulated
4494 * extent items so far.
4495 */
4496 if (btrfs_file_extent_compression(leaf, fi) !=
4497 BTRFS_COMPRESS_NONE && pending_del_nr) {
4498 err = btrfs_del_items(trans, root, path,
4499 pending_del_slot,
4500 pending_del_nr);
4501 if (err) {
4502 btrfs_abort_transaction(trans,
4503 err);
4504 goto error;
4505 }
4506 pending_del_nr = 0;
4507 }
4508
4509 err = truncate_inline_extent(inode, path,
4510 &found_key,
4511 item_end,
4512 new_size);
4513 if (err) {
4514 btrfs_abort_transaction(trans, err);
4515 goto error;
4516 }
4517 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4518 &root->state)) {
4519 inode_sub_bytes(inode, item_end + 1 - new_size);
4520 }
4521 }
4522 delete:
4523 if (del_item) {
4524 if (!pending_del_nr) {
4525 /* no pending yet, add ourselves */
4526 pending_del_slot = path->slots[0];
4527 pending_del_nr = 1;
4528 } else if (pending_del_nr &&
4529 path->slots[0] + 1 == pending_del_slot) {
4530 /* hop on the pending chunk */
4531 pending_del_nr++;
4532 pending_del_slot = path->slots[0];
4533 } else {
4534 BUG();
4535 }
4536 } else {
4537 break;
4538 }
4539 should_throttle = 0;
4540
4541 if (found_extent &&
4542 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4543 root == fs_info->tree_root)) {
4544 btrfs_set_path_blocking(path);
4545 bytes_deleted += extent_num_bytes;
4546 ret = btrfs_free_extent(trans, fs_info, extent_start,
4547 extent_num_bytes, 0,
4548 btrfs_header_owner(leaf),
4549 ino, extent_offset);
4550 BUG_ON(ret);
4551 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4552 btrfs_async_run_delayed_refs(fs_info,
4553 trans->delayed_ref_updates * 2,
4554 trans->transid, 0);
4555 if (be_nice) {
4556 if (truncate_space_check(trans, root,
4557 extent_num_bytes)) {
4558 should_end = 1;
4559 }
4560 if (btrfs_should_throttle_delayed_refs(trans,
4561 fs_info))
4562 should_throttle = 1;
4563 }
4564 }
4565
4566 if (found_type == BTRFS_INODE_ITEM_KEY)
4567 break;
4568
4569 if (path->slots[0] == 0 ||
4570 path->slots[0] != pending_del_slot ||
4571 should_throttle || should_end) {
4572 if (pending_del_nr) {
4573 ret = btrfs_del_items(trans, root, path,
4574 pending_del_slot,
4575 pending_del_nr);
4576 if (ret) {
4577 btrfs_abort_transaction(trans, ret);
4578 goto error;
4579 }
4580 pending_del_nr = 0;
4581 }
4582 btrfs_release_path(path);
4583 if (should_throttle) {
4584 unsigned long updates = trans->delayed_ref_updates;
4585 if (updates) {
4586 trans->delayed_ref_updates = 0;
4587 ret = btrfs_run_delayed_refs(trans,
4588 fs_info,
4589 updates * 2);
4590 if (ret && !err)
4591 err = ret;
4592 }
4593 }
4594 /*
4595 * if we failed to refill our space rsv, bail out
4596 * and let the transaction restart
4597 */
4598 if (should_end) {
4599 err = -EAGAIN;
4600 goto error;
4601 }
4602 goto search_again;
4603 } else {
4604 path->slots[0]--;
4605 }
4606 }
4607 out:
4608 if (pending_del_nr) {
4609 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4610 pending_del_nr);
4611 if (ret)
4612 btrfs_abort_transaction(trans, ret);
4613 }
4614 error:
4615 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4616 btrfs_ordered_update_i_size(inode, last_size, NULL);
4617
4618 btrfs_free_path(path);
4619
4620 if (err == 0) {
4621 /* only inline file may have last_size != new_size */
4622 if (new_size >= fs_info->sectorsize ||
4623 new_size > fs_info->max_inline)
4624 ASSERT(last_size == new_size);
4625 }
4626
4627 if (be_nice && bytes_deleted > SZ_32M) {
4628 unsigned long updates = trans->delayed_ref_updates;
4629 if (updates) {
4630 trans->delayed_ref_updates = 0;
4631 ret = btrfs_run_delayed_refs(trans, fs_info,
4632 updates * 2);
4633 if (ret && !err)
4634 err = ret;
4635 }
4636 }
4637 return err;
4638 }
4639
4640 /*
4641 * btrfs_truncate_block - read, zero a chunk and write a block
4642 * @inode - inode that we're zeroing
4643 * @from - the offset to start zeroing
4644 * @len - the length to zero, 0 to zero the entire range respective to the
4645 * offset
4646 * @front - zero up to the offset instead of from the offset on
4647 *
4648 * This will find the block for the "from" offset and cow the block and zero the
4649 * part we want to zero. This is used with truncate and hole punching.
4650 */
4651 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4652 int front)
4653 {
4654 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4655 struct address_space *mapping = inode->i_mapping;
4656 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4657 struct btrfs_ordered_extent *ordered;
4658 struct extent_state *cached_state = NULL;
4659 char *kaddr;
4660 u32 blocksize = fs_info->sectorsize;
4661 pgoff_t index = from >> PAGE_SHIFT;
4662 unsigned offset = from & (blocksize - 1);
4663 struct page *page;
4664 gfp_t mask = btrfs_alloc_write_mask(mapping);
4665 int ret = 0;
4666 u64 block_start;
4667 u64 block_end;
4668
4669 if ((offset & (blocksize - 1)) == 0 &&
4670 (!len || ((len & (blocksize - 1)) == 0)))
4671 goto out;
4672
4673 ret = btrfs_delalloc_reserve_space(inode,
4674 round_down(from, blocksize), blocksize);
4675 if (ret)
4676 goto out;
4677
4678 again:
4679 page = find_or_create_page(mapping, index, mask);
4680 if (!page) {
4681 btrfs_delalloc_release_space(inode,
4682 round_down(from, blocksize),
4683 blocksize);
4684 ret = -ENOMEM;
4685 goto out;
4686 }
4687
4688 block_start = round_down(from, blocksize);
4689 block_end = block_start + blocksize - 1;
4690
4691 if (!PageUptodate(page)) {
4692 ret = btrfs_readpage(NULL, page);
4693 lock_page(page);
4694 if (page->mapping != mapping) {
4695 unlock_page(page);
4696 put_page(page);
4697 goto again;
4698 }
4699 if (!PageUptodate(page)) {
4700 ret = -EIO;
4701 goto out_unlock;
4702 }
4703 }
4704 wait_on_page_writeback(page);
4705
4706 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4707 set_page_extent_mapped(page);
4708
4709 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4710 if (ordered) {
4711 unlock_extent_cached(io_tree, block_start, block_end,
4712 &cached_state, GFP_NOFS);
4713 unlock_page(page);
4714 put_page(page);
4715 btrfs_start_ordered_extent(inode, ordered, 1);
4716 btrfs_put_ordered_extent(ordered);
4717 goto again;
4718 }
4719
4720 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4721 EXTENT_DIRTY | EXTENT_DELALLOC |
4722 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4723 0, 0, &cached_state, GFP_NOFS);
4724
4725 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4726 &cached_state, 0);
4727 if (ret) {
4728 unlock_extent_cached(io_tree, block_start, block_end,
4729 &cached_state, GFP_NOFS);
4730 goto out_unlock;
4731 }
4732
4733 if (offset != blocksize) {
4734 if (!len)
4735 len = blocksize - offset;
4736 kaddr = kmap(page);
4737 if (front)
4738 memset(kaddr + (block_start - page_offset(page)),
4739 0, offset);
4740 else
4741 memset(kaddr + (block_start - page_offset(page)) + offset,
4742 0, len);
4743 flush_dcache_page(page);
4744 kunmap(page);
4745 }
4746 ClearPageChecked(page);
4747 set_page_dirty(page);
4748 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4749 GFP_NOFS);
4750
4751 out_unlock:
4752 if (ret)
4753 btrfs_delalloc_release_space(inode, block_start,
4754 blocksize);
4755 unlock_page(page);
4756 put_page(page);
4757 out:
4758 return ret;
4759 }
4760
4761 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4762 u64 offset, u64 len)
4763 {
4764 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4765 struct btrfs_trans_handle *trans;
4766 int ret;
4767
4768 /*
4769 * Still need to make sure the inode looks like it's been updated so
4770 * that any holes get logged if we fsync.
4771 */
4772 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4773 BTRFS_I(inode)->last_trans = fs_info->generation;
4774 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4775 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4776 return 0;
4777 }
4778
4779 /*
4780 * 1 - for the one we're dropping
4781 * 1 - for the one we're adding
4782 * 1 - for updating the inode.
4783 */
4784 trans = btrfs_start_transaction(root, 3);
4785 if (IS_ERR(trans))
4786 return PTR_ERR(trans);
4787
4788 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4789 if (ret) {
4790 btrfs_abort_transaction(trans, ret);
4791 btrfs_end_transaction(trans);
4792 return ret;
4793 }
4794
4795 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4796 offset, 0, 0, len, 0, len, 0, 0, 0);
4797 if (ret)
4798 btrfs_abort_transaction(trans, ret);
4799 else
4800 btrfs_update_inode(trans, root, inode);
4801 btrfs_end_transaction(trans);
4802 return ret;
4803 }
4804
4805 /*
4806 * This function puts in dummy file extents for the area we're creating a hole
4807 * for. So if we are truncating this file to a larger size we need to insert
4808 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4809 * the range between oldsize and size
4810 */
4811 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4812 {
4813 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4814 struct btrfs_root *root = BTRFS_I(inode)->root;
4815 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4816 struct extent_map *em = NULL;
4817 struct extent_state *cached_state = NULL;
4818 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4819 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4820 u64 block_end = ALIGN(size, fs_info->sectorsize);
4821 u64 last_byte;
4822 u64 cur_offset;
4823 u64 hole_size;
4824 int err = 0;
4825
4826 /*
4827 * If our size started in the middle of a block we need to zero out the
4828 * rest of the block before we expand the i_size, otherwise we could
4829 * expose stale data.
4830 */
4831 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4832 if (err)
4833 return err;
4834
4835 if (size <= hole_start)
4836 return 0;
4837
4838 while (1) {
4839 struct btrfs_ordered_extent *ordered;
4840
4841 lock_extent_bits(io_tree, hole_start, block_end - 1,
4842 &cached_state);
4843 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4844 block_end - hole_start);
4845 if (!ordered)
4846 break;
4847 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4848 &cached_state, GFP_NOFS);
4849 btrfs_start_ordered_extent(inode, ordered, 1);
4850 btrfs_put_ordered_extent(ordered);
4851 }
4852
4853 cur_offset = hole_start;
4854 while (1) {
4855 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4856 block_end - cur_offset, 0);
4857 if (IS_ERR(em)) {
4858 err = PTR_ERR(em);
4859 em = NULL;
4860 break;
4861 }
4862 last_byte = min(extent_map_end(em), block_end);
4863 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4864 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4865 struct extent_map *hole_em;
4866 hole_size = last_byte - cur_offset;
4867
4868 err = maybe_insert_hole(root, inode, cur_offset,
4869 hole_size);
4870 if (err)
4871 break;
4872 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4873 cur_offset + hole_size - 1, 0);
4874 hole_em = alloc_extent_map();
4875 if (!hole_em) {
4876 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4877 &BTRFS_I(inode)->runtime_flags);
4878 goto next;
4879 }
4880 hole_em->start = cur_offset;
4881 hole_em->len = hole_size;
4882 hole_em->orig_start = cur_offset;
4883
4884 hole_em->block_start = EXTENT_MAP_HOLE;
4885 hole_em->block_len = 0;
4886 hole_em->orig_block_len = 0;
4887 hole_em->ram_bytes = hole_size;
4888 hole_em->bdev = fs_info->fs_devices->latest_bdev;
4889 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4890 hole_em->generation = fs_info->generation;
4891
4892 while (1) {
4893 write_lock(&em_tree->lock);
4894 err = add_extent_mapping(em_tree, hole_em, 1);
4895 write_unlock(&em_tree->lock);
4896 if (err != -EEXIST)
4897 break;
4898 btrfs_drop_extent_cache(BTRFS_I(inode),
4899 cur_offset,
4900 cur_offset +
4901 hole_size - 1, 0);
4902 }
4903 free_extent_map(hole_em);
4904 }
4905 next:
4906 free_extent_map(em);
4907 em = NULL;
4908 cur_offset = last_byte;
4909 if (cur_offset >= block_end)
4910 break;
4911 }
4912 free_extent_map(em);
4913 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4914 GFP_NOFS);
4915 return err;
4916 }
4917
4918 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4919 {
4920 struct btrfs_root *root = BTRFS_I(inode)->root;
4921 struct btrfs_trans_handle *trans;
4922 loff_t oldsize = i_size_read(inode);
4923 loff_t newsize = attr->ia_size;
4924 int mask = attr->ia_valid;
4925 int ret;
4926
4927 /*
4928 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4929 * special case where we need to update the times despite not having
4930 * these flags set. For all other operations the VFS set these flags
4931 * explicitly if it wants a timestamp update.
4932 */
4933 if (newsize != oldsize) {
4934 inode_inc_iversion(inode);
4935 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4936 inode->i_ctime = inode->i_mtime =
4937 current_time(inode);
4938 }
4939
4940 if (newsize > oldsize) {
4941 /*
4942 * Don't do an expanding truncate while snapshoting is ongoing.
4943 * This is to ensure the snapshot captures a fully consistent
4944 * state of this file - if the snapshot captures this expanding
4945 * truncation, it must capture all writes that happened before
4946 * this truncation.
4947 */
4948 btrfs_wait_for_snapshot_creation(root);
4949 ret = btrfs_cont_expand(inode, oldsize, newsize);
4950 if (ret) {
4951 btrfs_end_write_no_snapshoting(root);
4952 return ret;
4953 }
4954
4955 trans = btrfs_start_transaction(root, 1);
4956 if (IS_ERR(trans)) {
4957 btrfs_end_write_no_snapshoting(root);
4958 return PTR_ERR(trans);
4959 }
4960
4961 i_size_write(inode, newsize);
4962 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4963 pagecache_isize_extended(inode, oldsize, newsize);
4964 ret = btrfs_update_inode(trans, root, inode);
4965 btrfs_end_write_no_snapshoting(root);
4966 btrfs_end_transaction(trans);
4967 } else {
4968
4969 /*
4970 * We're truncating a file that used to have good data down to
4971 * zero. Make sure it gets into the ordered flush list so that
4972 * any new writes get down to disk quickly.
4973 */
4974 if (newsize == 0)
4975 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4976 &BTRFS_I(inode)->runtime_flags);
4977
4978 /*
4979 * 1 for the orphan item we're going to add
4980 * 1 for the orphan item deletion.
4981 */
4982 trans = btrfs_start_transaction(root, 2);
4983 if (IS_ERR(trans))
4984 return PTR_ERR(trans);
4985
4986 /*
4987 * We need to do this in case we fail at _any_ point during the
4988 * actual truncate. Once we do the truncate_setsize we could
4989 * invalidate pages which forces any outstanding ordered io to
4990 * be instantly completed which will give us extents that need
4991 * to be truncated. If we fail to get an orphan inode down we
4992 * could have left over extents that were never meant to live,
4993 * so we need to guarantee from this point on that everything
4994 * will be consistent.
4995 */
4996 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4997 btrfs_end_transaction(trans);
4998 if (ret)
4999 return ret;
5000
5001 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5002 truncate_setsize(inode, newsize);
5003
5004 /* Disable nonlocked read DIO to avoid the end less truncate */
5005 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5006 inode_dio_wait(inode);
5007 btrfs_inode_resume_unlocked_dio(inode);
5008
5009 ret = btrfs_truncate(inode);
5010 if (ret && inode->i_nlink) {
5011 int err;
5012
5013 /* To get a stable disk_i_size */
5014 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5015 if (err) {
5016 btrfs_orphan_del(NULL, BTRFS_I(inode));
5017 return err;
5018 }
5019
5020 /*
5021 * failed to truncate, disk_i_size is only adjusted down
5022 * as we remove extents, so it should represent the true
5023 * size of the inode, so reset the in memory size and
5024 * delete our orphan entry.
5025 */
5026 trans = btrfs_join_transaction(root);
5027 if (IS_ERR(trans)) {
5028 btrfs_orphan_del(NULL, BTRFS_I(inode));
5029 return ret;
5030 }
5031 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5032 err = btrfs_orphan_del(trans, BTRFS_I(inode));
5033 if (err)
5034 btrfs_abort_transaction(trans, err);
5035 btrfs_end_transaction(trans);
5036 }
5037 }
5038
5039 return ret;
5040 }
5041
5042 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5043 {
5044 struct inode *inode = d_inode(dentry);
5045 struct btrfs_root *root = BTRFS_I(inode)->root;
5046 int err;
5047
5048 if (btrfs_root_readonly(root))
5049 return -EROFS;
5050
5051 err = setattr_prepare(dentry, attr);
5052 if (err)
5053 return err;
5054
5055 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5056 err = btrfs_setsize(inode, attr);
5057 if (err)
5058 return err;
5059 }
5060
5061 if (attr->ia_valid) {
5062 setattr_copy(inode, attr);
5063 inode_inc_iversion(inode);
5064 err = btrfs_dirty_inode(inode);
5065
5066 if (!err && attr->ia_valid & ATTR_MODE)
5067 err = posix_acl_chmod(inode, inode->i_mode);
5068 }
5069
5070 return err;
5071 }
5072
5073 /*
5074 * While truncating the inode pages during eviction, we get the VFS calling
5075 * btrfs_invalidatepage() against each page of the inode. This is slow because
5076 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5077 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5078 * extent_state structures over and over, wasting lots of time.
5079 *
5080 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5081 * those expensive operations on a per page basis and do only the ordered io
5082 * finishing, while we release here the extent_map and extent_state structures,
5083 * without the excessive merging and splitting.
5084 */
5085 static void evict_inode_truncate_pages(struct inode *inode)
5086 {
5087 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5088 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5089 struct rb_node *node;
5090
5091 ASSERT(inode->i_state & I_FREEING);
5092 truncate_inode_pages_final(&inode->i_data);
5093
5094 write_lock(&map_tree->lock);
5095 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5096 struct extent_map *em;
5097
5098 node = rb_first(&map_tree->map);
5099 em = rb_entry(node, struct extent_map, rb_node);
5100 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5101 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5102 remove_extent_mapping(map_tree, em);
5103 free_extent_map(em);
5104 if (need_resched()) {
5105 write_unlock(&map_tree->lock);
5106 cond_resched();
5107 write_lock(&map_tree->lock);
5108 }
5109 }
5110 write_unlock(&map_tree->lock);
5111
5112 /*
5113 * Keep looping until we have no more ranges in the io tree.
5114 * We can have ongoing bios started by readpages (called from readahead)
5115 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5116 * still in progress (unlocked the pages in the bio but did not yet
5117 * unlocked the ranges in the io tree). Therefore this means some
5118 * ranges can still be locked and eviction started because before
5119 * submitting those bios, which are executed by a separate task (work
5120 * queue kthread), inode references (inode->i_count) were not taken
5121 * (which would be dropped in the end io callback of each bio).
5122 * Therefore here we effectively end up waiting for those bios and
5123 * anyone else holding locked ranges without having bumped the inode's
5124 * reference count - if we don't do it, when they access the inode's
5125 * io_tree to unlock a range it may be too late, leading to an
5126 * use-after-free issue.
5127 */
5128 spin_lock(&io_tree->lock);
5129 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5130 struct extent_state *state;
5131 struct extent_state *cached_state = NULL;
5132 u64 start;
5133 u64 end;
5134
5135 node = rb_first(&io_tree->state);
5136 state = rb_entry(node, struct extent_state, rb_node);
5137 start = state->start;
5138 end = state->end;
5139 spin_unlock(&io_tree->lock);
5140
5141 lock_extent_bits(io_tree, start, end, &cached_state);
5142
5143 /*
5144 * If still has DELALLOC flag, the extent didn't reach disk,
5145 * and its reserved space won't be freed by delayed_ref.
5146 * So we need to free its reserved space here.
5147 * (Refer to comment in btrfs_invalidatepage, case 2)
5148 *
5149 * Note, end is the bytenr of last byte, so we need + 1 here.
5150 */
5151 if (state->state & EXTENT_DELALLOC)
5152 btrfs_qgroup_free_data(inode, start, end - start + 1);
5153
5154 clear_extent_bit(io_tree, start, end,
5155 EXTENT_LOCKED | EXTENT_DIRTY |
5156 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5157 EXTENT_DEFRAG, 1, 1,
5158 &cached_state, GFP_NOFS);
5159
5160 cond_resched();
5161 spin_lock(&io_tree->lock);
5162 }
5163 spin_unlock(&io_tree->lock);
5164 }
5165
5166 void btrfs_evict_inode(struct inode *inode)
5167 {
5168 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5169 struct btrfs_trans_handle *trans;
5170 struct btrfs_root *root = BTRFS_I(inode)->root;
5171 struct btrfs_block_rsv *rsv, *global_rsv;
5172 int steal_from_global = 0;
5173 u64 min_size;
5174 int ret;
5175
5176 trace_btrfs_inode_evict(inode);
5177
5178 if (!root) {
5179 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5180 return;
5181 }
5182
5183 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5184
5185 evict_inode_truncate_pages(inode);
5186
5187 if (inode->i_nlink &&
5188 ((btrfs_root_refs(&root->root_item) != 0 &&
5189 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5190 btrfs_is_free_space_inode(BTRFS_I(inode))))
5191 goto no_delete;
5192
5193 if (is_bad_inode(inode)) {
5194 btrfs_orphan_del(NULL, BTRFS_I(inode));
5195 goto no_delete;
5196 }
5197 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5198 if (!special_file(inode->i_mode))
5199 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5200
5201 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5202
5203 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5204 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5205 &BTRFS_I(inode)->runtime_flags));
5206 goto no_delete;
5207 }
5208
5209 if (inode->i_nlink > 0) {
5210 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5211 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5212 goto no_delete;
5213 }
5214
5215 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5216 if (ret) {
5217 btrfs_orphan_del(NULL, BTRFS_I(inode));
5218 goto no_delete;
5219 }
5220
5221 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5222 if (!rsv) {
5223 btrfs_orphan_del(NULL, BTRFS_I(inode));
5224 goto no_delete;
5225 }
5226 rsv->size = min_size;
5227 rsv->failfast = 1;
5228 global_rsv = &fs_info->global_block_rsv;
5229
5230 btrfs_i_size_write(BTRFS_I(inode), 0);
5231
5232 /*
5233 * This is a bit simpler than btrfs_truncate since we've already
5234 * reserved our space for our orphan item in the unlink, so we just
5235 * need to reserve some slack space in case we add bytes and update
5236 * inode item when doing the truncate.
5237 */
5238 while (1) {
5239 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5240 BTRFS_RESERVE_FLUSH_LIMIT);
5241
5242 /*
5243 * Try and steal from the global reserve since we will
5244 * likely not use this space anyway, we want to try as
5245 * hard as possible to get this to work.
5246 */
5247 if (ret)
5248 steal_from_global++;
5249 else
5250 steal_from_global = 0;
5251 ret = 0;
5252
5253 /*
5254 * steal_from_global == 0: we reserved stuff, hooray!
5255 * steal_from_global == 1: we didn't reserve stuff, boo!
5256 * steal_from_global == 2: we've committed, still not a lot of
5257 * room but maybe we'll have room in the global reserve this
5258 * time.
5259 * steal_from_global == 3: abandon all hope!
5260 */
5261 if (steal_from_global > 2) {
5262 btrfs_warn(fs_info,
5263 "Could not get space for a delete, will truncate on mount %d",
5264 ret);
5265 btrfs_orphan_del(NULL, BTRFS_I(inode));
5266 btrfs_free_block_rsv(fs_info, rsv);
5267 goto no_delete;
5268 }
5269
5270 trans = btrfs_join_transaction(root);
5271 if (IS_ERR(trans)) {
5272 btrfs_orphan_del(NULL, BTRFS_I(inode));
5273 btrfs_free_block_rsv(fs_info, rsv);
5274 goto no_delete;
5275 }
5276
5277 /*
5278 * We can't just steal from the global reserve, we need to make
5279 * sure there is room to do it, if not we need to commit and try
5280 * again.
5281 */
5282 if (steal_from_global) {
5283 if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5284 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5285 min_size, 0);
5286 else
5287 ret = -ENOSPC;
5288 }
5289
5290 /*
5291 * Couldn't steal from the global reserve, we have too much
5292 * pending stuff built up, commit the transaction and try it
5293 * again.
5294 */
5295 if (ret) {
5296 ret = btrfs_commit_transaction(trans);
5297 if (ret) {
5298 btrfs_orphan_del(NULL, BTRFS_I(inode));
5299 btrfs_free_block_rsv(fs_info, rsv);
5300 goto no_delete;
5301 }
5302 continue;
5303 } else {
5304 steal_from_global = 0;
5305 }
5306
5307 trans->block_rsv = rsv;
5308
5309 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5310 if (ret != -ENOSPC && ret != -EAGAIN)
5311 break;
5312
5313 trans->block_rsv = &fs_info->trans_block_rsv;
5314 btrfs_end_transaction(trans);
5315 trans = NULL;
5316 btrfs_btree_balance_dirty(fs_info);
5317 }
5318
5319 btrfs_free_block_rsv(fs_info, rsv);
5320
5321 /*
5322 * Errors here aren't a big deal, it just means we leave orphan items
5323 * in the tree. They will be cleaned up on the next mount.
5324 */
5325 if (ret == 0) {
5326 trans->block_rsv = root->orphan_block_rsv;
5327 btrfs_orphan_del(trans, BTRFS_I(inode));
5328 } else {
5329 btrfs_orphan_del(NULL, BTRFS_I(inode));
5330 }
5331
5332 trans->block_rsv = &fs_info->trans_block_rsv;
5333 if (!(root == fs_info->tree_root ||
5334 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5335 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5336
5337 btrfs_end_transaction(trans);
5338 btrfs_btree_balance_dirty(fs_info);
5339 no_delete:
5340 btrfs_remove_delayed_node(BTRFS_I(inode));
5341 clear_inode(inode);
5342 }
5343
5344 /*
5345 * this returns the key found in the dir entry in the location pointer.
5346 * If no dir entries were found, location->objectid is 0.
5347 */
5348 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5349 struct btrfs_key *location)
5350 {
5351 const char *name = dentry->d_name.name;
5352 int namelen = dentry->d_name.len;
5353 struct btrfs_dir_item *di;
5354 struct btrfs_path *path;
5355 struct btrfs_root *root = BTRFS_I(dir)->root;
5356 int ret = 0;
5357
5358 path = btrfs_alloc_path();
5359 if (!path)
5360 return -ENOMEM;
5361
5362 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5363 name, namelen, 0);
5364 if (IS_ERR(di))
5365 ret = PTR_ERR(di);
5366
5367 if (IS_ERR_OR_NULL(di))
5368 goto out_err;
5369
5370 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5371 out:
5372 btrfs_free_path(path);
5373 return ret;
5374 out_err:
5375 location->objectid = 0;
5376 goto out;
5377 }
5378
5379 /*
5380 * when we hit a tree root in a directory, the btrfs part of the inode
5381 * needs to be changed to reflect the root directory of the tree root. This
5382 * is kind of like crossing a mount point.
5383 */
5384 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5385 struct inode *dir,
5386 struct dentry *dentry,
5387 struct btrfs_key *location,
5388 struct btrfs_root **sub_root)
5389 {
5390 struct btrfs_path *path;
5391 struct btrfs_root *new_root;
5392 struct btrfs_root_ref *ref;
5393 struct extent_buffer *leaf;
5394 struct btrfs_key key;
5395 int ret;
5396 int err = 0;
5397
5398 path = btrfs_alloc_path();
5399 if (!path) {
5400 err = -ENOMEM;
5401 goto out;
5402 }
5403
5404 err = -ENOENT;
5405 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5406 key.type = BTRFS_ROOT_REF_KEY;
5407 key.offset = location->objectid;
5408
5409 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5410 if (ret) {
5411 if (ret < 0)
5412 err = ret;
5413 goto out;
5414 }
5415
5416 leaf = path->nodes[0];
5417 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5418 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5419 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5420 goto out;
5421
5422 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5423 (unsigned long)(ref + 1),
5424 dentry->d_name.len);
5425 if (ret)
5426 goto out;
5427
5428 btrfs_release_path(path);
5429
5430 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5431 if (IS_ERR(new_root)) {
5432 err = PTR_ERR(new_root);
5433 goto out;
5434 }
5435
5436 *sub_root = new_root;
5437 location->objectid = btrfs_root_dirid(&new_root->root_item);
5438 location->type = BTRFS_INODE_ITEM_KEY;
5439 location->offset = 0;
5440 err = 0;
5441 out:
5442 btrfs_free_path(path);
5443 return err;
5444 }
5445
5446 static void inode_tree_add(struct inode *inode)
5447 {
5448 struct btrfs_root *root = BTRFS_I(inode)->root;
5449 struct btrfs_inode *entry;
5450 struct rb_node **p;
5451 struct rb_node *parent;
5452 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5453 u64 ino = btrfs_ino(BTRFS_I(inode));
5454
5455 if (inode_unhashed(inode))
5456 return;
5457 parent = NULL;
5458 spin_lock(&root->inode_lock);
5459 p = &root->inode_tree.rb_node;
5460 while (*p) {
5461 parent = *p;
5462 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5463
5464 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5465 p = &parent->rb_left;
5466 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5467 p = &parent->rb_right;
5468 else {
5469 WARN_ON(!(entry->vfs_inode.i_state &
5470 (I_WILL_FREE | I_FREEING)));
5471 rb_replace_node(parent, new, &root->inode_tree);
5472 RB_CLEAR_NODE(parent);
5473 spin_unlock(&root->inode_lock);
5474 return;
5475 }
5476 }
5477 rb_link_node(new, parent, p);
5478 rb_insert_color(new, &root->inode_tree);
5479 spin_unlock(&root->inode_lock);
5480 }
5481
5482 static void inode_tree_del(struct inode *inode)
5483 {
5484 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5485 struct btrfs_root *root = BTRFS_I(inode)->root;
5486 int empty = 0;
5487
5488 spin_lock(&root->inode_lock);
5489 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5490 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5491 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5492 empty = RB_EMPTY_ROOT(&root->inode_tree);
5493 }
5494 spin_unlock(&root->inode_lock);
5495
5496 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5497 synchronize_srcu(&fs_info->subvol_srcu);
5498 spin_lock(&root->inode_lock);
5499 empty = RB_EMPTY_ROOT(&root->inode_tree);
5500 spin_unlock(&root->inode_lock);
5501 if (empty)
5502 btrfs_add_dead_root(root);
5503 }
5504 }
5505
5506 void btrfs_invalidate_inodes(struct btrfs_root *root)
5507 {
5508 struct btrfs_fs_info *fs_info = root->fs_info;
5509 struct rb_node *node;
5510 struct rb_node *prev;
5511 struct btrfs_inode *entry;
5512 struct inode *inode;
5513 u64 objectid = 0;
5514
5515 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5516 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5517
5518 spin_lock(&root->inode_lock);
5519 again:
5520 node = root->inode_tree.rb_node;
5521 prev = NULL;
5522 while (node) {
5523 prev = node;
5524 entry = rb_entry(node, struct btrfs_inode, rb_node);
5525
5526 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5527 node = node->rb_left;
5528 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5529 node = node->rb_right;
5530 else
5531 break;
5532 }
5533 if (!node) {
5534 while (prev) {
5535 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5536 if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5537 node = prev;
5538 break;
5539 }
5540 prev = rb_next(prev);
5541 }
5542 }
5543 while (node) {
5544 entry = rb_entry(node, struct btrfs_inode, rb_node);
5545 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5546 inode = igrab(&entry->vfs_inode);
5547 if (inode) {
5548 spin_unlock(&root->inode_lock);
5549 if (atomic_read(&inode->i_count) > 1)
5550 d_prune_aliases(inode);
5551 /*
5552 * btrfs_drop_inode will have it removed from
5553 * the inode cache when its usage count
5554 * hits zero.
5555 */
5556 iput(inode);
5557 cond_resched();
5558 spin_lock(&root->inode_lock);
5559 goto again;
5560 }
5561
5562 if (cond_resched_lock(&root->inode_lock))
5563 goto again;
5564
5565 node = rb_next(node);
5566 }
5567 spin_unlock(&root->inode_lock);
5568 }
5569
5570 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5571 {
5572 struct btrfs_iget_args *args = p;
5573 inode->i_ino = args->location->objectid;
5574 memcpy(&BTRFS_I(inode)->location, args->location,
5575 sizeof(*args->location));
5576 BTRFS_I(inode)->root = args->root;
5577 return 0;
5578 }
5579
5580 static int btrfs_find_actor(struct inode *inode, void *opaque)
5581 {
5582 struct btrfs_iget_args *args = opaque;
5583 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5584 args->root == BTRFS_I(inode)->root;
5585 }
5586
5587 static struct inode *btrfs_iget_locked(struct super_block *s,
5588 struct btrfs_key *location,
5589 struct btrfs_root *root)
5590 {
5591 struct inode *inode;
5592 struct btrfs_iget_args args;
5593 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5594
5595 args.location = location;
5596 args.root = root;
5597
5598 inode = iget5_locked(s, hashval, btrfs_find_actor,
5599 btrfs_init_locked_inode,
5600 (void *)&args);
5601 return inode;
5602 }
5603
5604 /* Get an inode object given its location and corresponding root.
5605 * Returns in *is_new if the inode was read from disk
5606 */
5607 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5608 struct btrfs_root *root, int *new)
5609 {
5610 struct inode *inode;
5611
5612 inode = btrfs_iget_locked(s, location, root);
5613 if (!inode)
5614 return ERR_PTR(-ENOMEM);
5615
5616 if (inode->i_state & I_NEW) {
5617 int ret;
5618
5619 ret = btrfs_read_locked_inode(inode);
5620 if (!is_bad_inode(inode)) {
5621 inode_tree_add(inode);
5622 unlock_new_inode(inode);
5623 if (new)
5624 *new = 1;
5625 } else {
5626 unlock_new_inode(inode);
5627 iput(inode);
5628 ASSERT(ret < 0);
5629 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5630 }
5631 }
5632
5633 return inode;
5634 }
5635
5636 static struct inode *new_simple_dir(struct super_block *s,
5637 struct btrfs_key *key,
5638 struct btrfs_root *root)
5639 {
5640 struct inode *inode = new_inode(s);
5641
5642 if (!inode)
5643 return ERR_PTR(-ENOMEM);
5644
5645 BTRFS_I(inode)->root = root;
5646 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5647 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5648
5649 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5650 inode->i_op = &btrfs_dir_ro_inode_operations;
5651 inode->i_opflags &= ~IOP_XATTR;
5652 inode->i_fop = &simple_dir_operations;
5653 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5654 inode->i_mtime = current_time(inode);
5655 inode->i_atime = inode->i_mtime;
5656 inode->i_ctime = inode->i_mtime;
5657 BTRFS_I(inode)->i_otime = inode->i_mtime;
5658
5659 return inode;
5660 }
5661
5662 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5663 {
5664 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5665 struct inode *inode;
5666 struct btrfs_root *root = BTRFS_I(dir)->root;
5667 struct btrfs_root *sub_root = root;
5668 struct btrfs_key location;
5669 int index;
5670 int ret = 0;
5671
5672 if (dentry->d_name.len > BTRFS_NAME_LEN)
5673 return ERR_PTR(-ENAMETOOLONG);
5674
5675 ret = btrfs_inode_by_name(dir, dentry, &location);
5676 if (ret < 0)
5677 return ERR_PTR(ret);
5678
5679 if (location.objectid == 0)
5680 return ERR_PTR(-ENOENT);
5681
5682 if (location.type == BTRFS_INODE_ITEM_KEY) {
5683 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5684 return inode;
5685 }
5686
5687 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5688
5689 index = srcu_read_lock(&fs_info->subvol_srcu);
5690 ret = fixup_tree_root_location(fs_info, dir, dentry,
5691 &location, &sub_root);
5692 if (ret < 0) {
5693 if (ret != -ENOENT)
5694 inode = ERR_PTR(ret);
5695 else
5696 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5697 } else {
5698 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5699 }
5700 srcu_read_unlock(&fs_info->subvol_srcu, index);
5701
5702 if (!IS_ERR(inode) && root != sub_root) {
5703 down_read(&fs_info->cleanup_work_sem);
5704 if (!(inode->i_sb->s_flags & MS_RDONLY))
5705 ret = btrfs_orphan_cleanup(sub_root);
5706 up_read(&fs_info->cleanup_work_sem);
5707 if (ret) {
5708 iput(inode);
5709 inode = ERR_PTR(ret);
5710 }
5711 }
5712
5713 return inode;
5714 }
5715
5716 static int btrfs_dentry_delete(const struct dentry *dentry)
5717 {
5718 struct btrfs_root *root;
5719 struct inode *inode = d_inode(dentry);
5720
5721 if (!inode && !IS_ROOT(dentry))
5722 inode = d_inode(dentry->d_parent);
5723
5724 if (inode) {
5725 root = BTRFS_I(inode)->root;
5726 if (btrfs_root_refs(&root->root_item) == 0)
5727 return 1;
5728
5729 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5730 return 1;
5731 }
5732 return 0;
5733 }
5734
5735 static void btrfs_dentry_release(struct dentry *dentry)
5736 {
5737 kfree(dentry->d_fsdata);
5738 }
5739
5740 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5741 unsigned int flags)
5742 {
5743 struct inode *inode;
5744
5745 inode = btrfs_lookup_dentry(dir, dentry);
5746 if (IS_ERR(inode)) {
5747 if (PTR_ERR(inode) == -ENOENT)
5748 inode = NULL;
5749 else
5750 return ERR_CAST(inode);
5751 }
5752
5753 return d_splice_alias(inode, dentry);
5754 }
5755
5756 unsigned char btrfs_filetype_table[] = {
5757 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5758 };
5759
5760 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5761 {
5762 struct inode *inode = file_inode(file);
5763 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5764 struct btrfs_root *root = BTRFS_I(inode)->root;
5765 struct btrfs_item *item;
5766 struct btrfs_dir_item *di;
5767 struct btrfs_key key;
5768 struct btrfs_key found_key;
5769 struct btrfs_path *path;
5770 struct list_head ins_list;
5771 struct list_head del_list;
5772 int ret;
5773 struct extent_buffer *leaf;
5774 int slot;
5775 unsigned char d_type;
5776 int over = 0;
5777 char tmp_name[32];
5778 char *name_ptr;
5779 int name_len;
5780 bool put = false;
5781 struct btrfs_key location;
5782
5783 if (!dir_emit_dots(file, ctx))
5784 return 0;
5785
5786 path = btrfs_alloc_path();
5787 if (!path)
5788 return -ENOMEM;
5789
5790 path->reada = READA_FORWARD;
5791
5792 INIT_LIST_HEAD(&ins_list);
5793 INIT_LIST_HEAD(&del_list);
5794 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5795
5796 key.type = BTRFS_DIR_INDEX_KEY;
5797 key.offset = ctx->pos;
5798 key.objectid = btrfs_ino(BTRFS_I(inode));
5799
5800 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5801 if (ret < 0)
5802 goto err;
5803
5804 while (1) {
5805 leaf = path->nodes[0];
5806 slot = path->slots[0];
5807 if (slot >= btrfs_header_nritems(leaf)) {
5808 ret = btrfs_next_leaf(root, path);
5809 if (ret < 0)
5810 goto err;
5811 else if (ret > 0)
5812 break;
5813 continue;
5814 }
5815
5816 item = btrfs_item_nr(slot);
5817 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5818
5819 if (found_key.objectid != key.objectid)
5820 break;
5821 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5822 break;
5823 if (found_key.offset < ctx->pos)
5824 goto next;
5825 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5826 goto next;
5827
5828 ctx->pos = found_key.offset;
5829
5830 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5831 if (verify_dir_item(fs_info, leaf, di))
5832 goto next;
5833
5834 name_len = btrfs_dir_name_len(leaf, di);
5835 if (name_len <= sizeof(tmp_name)) {
5836 name_ptr = tmp_name;
5837 } else {
5838 name_ptr = kmalloc(name_len, GFP_KERNEL);
5839 if (!name_ptr) {
5840 ret = -ENOMEM;
5841 goto err;
5842 }
5843 }
5844 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5845 name_len);
5846
5847 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5848 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5849
5850 over = !dir_emit(ctx, name_ptr, name_len, location.objectid,
5851 d_type);
5852
5853 if (name_ptr != tmp_name)
5854 kfree(name_ptr);
5855
5856 if (over)
5857 goto nopos;
5858 ctx->pos++;
5859 next:
5860 path->slots[0]++;
5861 }
5862
5863 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5864 if (ret)
5865 goto nopos;
5866
5867 /*
5868 * Stop new entries from being returned after we return the last
5869 * entry.
5870 *
5871 * New directory entries are assigned a strictly increasing
5872 * offset. This means that new entries created during readdir
5873 * are *guaranteed* to be seen in the future by that readdir.
5874 * This has broken buggy programs which operate on names as
5875 * they're returned by readdir. Until we re-use freed offsets
5876 * we have this hack to stop new entries from being returned
5877 * under the assumption that they'll never reach this huge
5878 * offset.
5879 *
5880 * This is being careful not to overflow 32bit loff_t unless the
5881 * last entry requires it because doing so has broken 32bit apps
5882 * in the past.
5883 */
5884 if (ctx->pos >= INT_MAX)
5885 ctx->pos = LLONG_MAX;
5886 else
5887 ctx->pos = INT_MAX;
5888 nopos:
5889 ret = 0;
5890 err:
5891 if (put)
5892 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5893 btrfs_free_path(path);
5894 return ret;
5895 }
5896
5897 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5898 {
5899 struct btrfs_root *root = BTRFS_I(inode)->root;
5900 struct btrfs_trans_handle *trans;
5901 int ret = 0;
5902 bool nolock = false;
5903
5904 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5905 return 0;
5906
5907 if (btrfs_fs_closing(root->fs_info) &&
5908 btrfs_is_free_space_inode(BTRFS_I(inode)))
5909 nolock = true;
5910
5911 if (wbc->sync_mode == WB_SYNC_ALL) {
5912 if (nolock)
5913 trans = btrfs_join_transaction_nolock(root);
5914 else
5915 trans = btrfs_join_transaction(root);
5916 if (IS_ERR(trans))
5917 return PTR_ERR(trans);
5918 ret = btrfs_commit_transaction(trans);
5919 }
5920 return ret;
5921 }
5922
5923 /*
5924 * This is somewhat expensive, updating the tree every time the
5925 * inode changes. But, it is most likely to find the inode in cache.
5926 * FIXME, needs more benchmarking...there are no reasons other than performance
5927 * to keep or drop this code.
5928 */
5929 static int btrfs_dirty_inode(struct inode *inode)
5930 {
5931 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5932 struct btrfs_root *root = BTRFS_I(inode)->root;
5933 struct btrfs_trans_handle *trans;
5934 int ret;
5935
5936 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5937 return 0;
5938
5939 trans = btrfs_join_transaction(root);
5940 if (IS_ERR(trans))
5941 return PTR_ERR(trans);
5942
5943 ret = btrfs_update_inode(trans, root, inode);
5944 if (ret && ret == -ENOSPC) {
5945 /* whoops, lets try again with the full transaction */
5946 btrfs_end_transaction(trans);
5947 trans = btrfs_start_transaction(root, 1);
5948 if (IS_ERR(trans))
5949 return PTR_ERR(trans);
5950
5951 ret = btrfs_update_inode(trans, root, inode);
5952 }
5953 btrfs_end_transaction(trans);
5954 if (BTRFS_I(inode)->delayed_node)
5955 btrfs_balance_delayed_items(fs_info);
5956
5957 return ret;
5958 }
5959
5960 /*
5961 * This is a copy of file_update_time. We need this so we can return error on
5962 * ENOSPC for updating the inode in the case of file write and mmap writes.
5963 */
5964 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5965 int flags)
5966 {
5967 struct btrfs_root *root = BTRFS_I(inode)->root;
5968
5969 if (btrfs_root_readonly(root))
5970 return -EROFS;
5971
5972 if (flags & S_VERSION)
5973 inode_inc_iversion(inode);
5974 if (flags & S_CTIME)
5975 inode->i_ctime = *now;
5976 if (flags & S_MTIME)
5977 inode->i_mtime = *now;
5978 if (flags & S_ATIME)
5979 inode->i_atime = *now;
5980 return btrfs_dirty_inode(inode);
5981 }
5982
5983 /*
5984 * find the highest existing sequence number in a directory
5985 * and then set the in-memory index_cnt variable to reflect
5986 * free sequence numbers
5987 */
5988 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5989 {
5990 struct btrfs_root *root = inode->root;
5991 struct btrfs_key key, found_key;
5992 struct btrfs_path *path;
5993 struct extent_buffer *leaf;
5994 int ret;
5995
5996 key.objectid = btrfs_ino(inode);
5997 key.type = BTRFS_DIR_INDEX_KEY;
5998 key.offset = (u64)-1;
5999
6000 path = btrfs_alloc_path();
6001 if (!path)
6002 return -ENOMEM;
6003
6004 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6005 if (ret < 0)
6006 goto out;
6007 /* FIXME: we should be able to handle this */
6008 if (ret == 0)
6009 goto out;
6010 ret = 0;
6011
6012 /*
6013 * MAGIC NUMBER EXPLANATION:
6014 * since we search a directory based on f_pos we have to start at 2
6015 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6016 * else has to start at 2
6017 */
6018 if (path->slots[0] == 0) {
6019 inode->index_cnt = 2;
6020 goto out;
6021 }
6022
6023 path->slots[0]--;
6024
6025 leaf = path->nodes[0];
6026 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6027
6028 if (found_key.objectid != btrfs_ino(inode) ||
6029 found_key.type != BTRFS_DIR_INDEX_KEY) {
6030 inode->index_cnt = 2;
6031 goto out;
6032 }
6033
6034 inode->index_cnt = found_key.offset + 1;
6035 out:
6036 btrfs_free_path(path);
6037 return ret;
6038 }
6039
6040 /*
6041 * helper to find a free sequence number in a given directory. This current
6042 * code is very simple, later versions will do smarter things in the btree
6043 */
6044 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6045 {
6046 int ret = 0;
6047
6048 if (dir->index_cnt == (u64)-1) {
6049 ret = btrfs_inode_delayed_dir_index_count(dir);
6050 if (ret) {
6051 ret = btrfs_set_inode_index_count(dir);
6052 if (ret)
6053 return ret;
6054 }
6055 }
6056
6057 *index = dir->index_cnt;
6058 dir->index_cnt++;
6059
6060 return ret;
6061 }
6062
6063 static int btrfs_insert_inode_locked(struct inode *inode)
6064 {
6065 struct btrfs_iget_args args;
6066 args.location = &BTRFS_I(inode)->location;
6067 args.root = BTRFS_I(inode)->root;
6068
6069 return insert_inode_locked4(inode,
6070 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6071 btrfs_find_actor, &args);
6072 }
6073
6074 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6075 struct btrfs_root *root,
6076 struct inode *dir,
6077 const char *name, int name_len,
6078 u64 ref_objectid, u64 objectid,
6079 umode_t mode, u64 *index)
6080 {
6081 struct btrfs_fs_info *fs_info = root->fs_info;
6082 struct inode *inode;
6083 struct btrfs_inode_item *inode_item;
6084 struct btrfs_key *location;
6085 struct btrfs_path *path;
6086 struct btrfs_inode_ref *ref;
6087 struct btrfs_key key[2];
6088 u32 sizes[2];
6089 int nitems = name ? 2 : 1;
6090 unsigned long ptr;
6091 int ret;
6092
6093 path = btrfs_alloc_path();
6094 if (!path)
6095 return ERR_PTR(-ENOMEM);
6096
6097 inode = new_inode(fs_info->sb);
6098 if (!inode) {
6099 btrfs_free_path(path);
6100 return ERR_PTR(-ENOMEM);
6101 }
6102
6103 /*
6104 * O_TMPFILE, set link count to 0, so that after this point,
6105 * we fill in an inode item with the correct link count.
6106 */
6107 if (!name)
6108 set_nlink(inode, 0);
6109
6110 /*
6111 * we have to initialize this early, so we can reclaim the inode
6112 * number if we fail afterwards in this function.
6113 */
6114 inode->i_ino = objectid;
6115
6116 if (dir && name) {
6117 trace_btrfs_inode_request(dir);
6118
6119 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6120 if (ret) {
6121 btrfs_free_path(path);
6122 iput(inode);
6123 return ERR_PTR(ret);
6124 }
6125 } else if (dir) {
6126 *index = 0;
6127 }
6128 /*
6129 * index_cnt is ignored for everything but a dir,
6130 * btrfs_get_inode_index_count has an explanation for the magic
6131 * number
6132 */
6133 BTRFS_I(inode)->index_cnt = 2;
6134 BTRFS_I(inode)->dir_index = *index;
6135 BTRFS_I(inode)->root = root;
6136 BTRFS_I(inode)->generation = trans->transid;
6137 inode->i_generation = BTRFS_I(inode)->generation;
6138
6139 /*
6140 * We could have gotten an inode number from somebody who was fsynced
6141 * and then removed in this same transaction, so let's just set full
6142 * sync since it will be a full sync anyway and this will blow away the
6143 * old info in the log.
6144 */
6145 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6146
6147 key[0].objectid = objectid;
6148 key[0].type = BTRFS_INODE_ITEM_KEY;
6149 key[0].offset = 0;
6150
6151 sizes[0] = sizeof(struct btrfs_inode_item);
6152
6153 if (name) {
6154 /*
6155 * Start new inodes with an inode_ref. This is slightly more
6156 * efficient for small numbers of hard links since they will
6157 * be packed into one item. Extended refs will kick in if we
6158 * add more hard links than can fit in the ref item.
6159 */
6160 key[1].objectid = objectid;
6161 key[1].type = BTRFS_INODE_REF_KEY;
6162 key[1].offset = ref_objectid;
6163
6164 sizes[1] = name_len + sizeof(*ref);
6165 }
6166
6167 location = &BTRFS_I(inode)->location;
6168 location->objectid = objectid;
6169 location->offset = 0;
6170 location->type = BTRFS_INODE_ITEM_KEY;
6171
6172 ret = btrfs_insert_inode_locked(inode);
6173 if (ret < 0)
6174 goto fail;
6175
6176 path->leave_spinning = 1;
6177 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6178 if (ret != 0)
6179 goto fail_unlock;
6180
6181 inode_init_owner(inode, dir, mode);
6182 inode_set_bytes(inode, 0);
6183
6184 inode->i_mtime = current_time(inode);
6185 inode->i_atime = inode->i_mtime;
6186 inode->i_ctime = inode->i_mtime;
6187 BTRFS_I(inode)->i_otime = inode->i_mtime;
6188
6189 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6190 struct btrfs_inode_item);
6191 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6192 sizeof(*inode_item));
6193 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6194
6195 if (name) {
6196 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6197 struct btrfs_inode_ref);
6198 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6199 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6200 ptr = (unsigned long)(ref + 1);
6201 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6202 }
6203
6204 btrfs_mark_buffer_dirty(path->nodes[0]);
6205 btrfs_free_path(path);
6206
6207 btrfs_inherit_iflags(inode, dir);
6208
6209 if (S_ISREG(mode)) {
6210 if (btrfs_test_opt(fs_info, NODATASUM))
6211 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6212 if (btrfs_test_opt(fs_info, NODATACOW))
6213 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6214 BTRFS_INODE_NODATASUM;
6215 }
6216
6217 inode_tree_add(inode);
6218
6219 trace_btrfs_inode_new(inode);
6220 btrfs_set_inode_last_trans(trans, inode);
6221
6222 btrfs_update_root_times(trans, root);
6223
6224 ret = btrfs_inode_inherit_props(trans, inode, dir);
6225 if (ret)
6226 btrfs_err(fs_info,
6227 "error inheriting props for ino %llu (root %llu): %d",
6228 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6229
6230 return inode;
6231
6232 fail_unlock:
6233 unlock_new_inode(inode);
6234 fail:
6235 if (dir && name)
6236 BTRFS_I(dir)->index_cnt--;
6237 btrfs_free_path(path);
6238 iput(inode);
6239 return ERR_PTR(ret);
6240 }
6241
6242 static inline u8 btrfs_inode_type(struct inode *inode)
6243 {
6244 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6245 }
6246
6247 /*
6248 * utility function to add 'inode' into 'parent_inode' with
6249 * a give name and a given sequence number.
6250 * if 'add_backref' is true, also insert a backref from the
6251 * inode to the parent directory.
6252 */
6253 int btrfs_add_link(struct btrfs_trans_handle *trans,
6254 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6255 const char *name, int name_len, int add_backref, u64 index)
6256 {
6257 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6258 int ret = 0;
6259 struct btrfs_key key;
6260 struct btrfs_root *root = parent_inode->root;
6261 u64 ino = btrfs_ino(inode);
6262 u64 parent_ino = btrfs_ino(parent_inode);
6263
6264 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6265 memcpy(&key, &inode->root->root_key, sizeof(key));
6266 } else {
6267 key.objectid = ino;
6268 key.type = BTRFS_INODE_ITEM_KEY;
6269 key.offset = 0;
6270 }
6271
6272 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6273 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6274 root->root_key.objectid, parent_ino,
6275 index, name, name_len);
6276 } else if (add_backref) {
6277 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6278 parent_ino, index);
6279 }
6280
6281 /* Nothing to clean up yet */
6282 if (ret)
6283 return ret;
6284
6285 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6286 parent_inode, &key,
6287 btrfs_inode_type(&inode->vfs_inode), index);
6288 if (ret == -EEXIST || ret == -EOVERFLOW)
6289 goto fail_dir_item;
6290 else if (ret) {
6291 btrfs_abort_transaction(trans, ret);
6292 return ret;
6293 }
6294
6295 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6296 name_len * 2);
6297 inode_inc_iversion(&parent_inode->vfs_inode);
6298 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6299 current_time(&parent_inode->vfs_inode);
6300 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6301 if (ret)
6302 btrfs_abort_transaction(trans, ret);
6303 return ret;
6304
6305 fail_dir_item:
6306 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6307 u64 local_index;
6308 int err;
6309 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6310 root->root_key.objectid, parent_ino,
6311 &local_index, name, name_len);
6312
6313 } else if (add_backref) {
6314 u64 local_index;
6315 int err;
6316
6317 err = btrfs_del_inode_ref(trans, root, name, name_len,
6318 ino, parent_ino, &local_index);
6319 }
6320 return ret;
6321 }
6322
6323 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6324 struct btrfs_inode *dir, struct dentry *dentry,
6325 struct btrfs_inode *inode, int backref, u64 index)
6326 {
6327 int err = btrfs_add_link(trans, dir, inode,
6328 dentry->d_name.name, dentry->d_name.len,
6329 backref, index);
6330 if (err > 0)
6331 err = -EEXIST;
6332 return err;
6333 }
6334
6335 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6336 umode_t mode, dev_t rdev)
6337 {
6338 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6339 struct btrfs_trans_handle *trans;
6340 struct btrfs_root *root = BTRFS_I(dir)->root;
6341 struct inode *inode = NULL;
6342 int err;
6343 int drop_inode = 0;
6344 u64 objectid;
6345 u64 index = 0;
6346
6347 /*
6348 * 2 for inode item and ref
6349 * 2 for dir items
6350 * 1 for xattr if selinux is on
6351 */
6352 trans = btrfs_start_transaction(root, 5);
6353 if (IS_ERR(trans))
6354 return PTR_ERR(trans);
6355
6356 err = btrfs_find_free_ino(root, &objectid);
6357 if (err)
6358 goto out_unlock;
6359
6360 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6361 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6362 mode, &index);
6363 if (IS_ERR(inode)) {
6364 err = PTR_ERR(inode);
6365 goto out_unlock;
6366 }
6367
6368 /*
6369 * If the active LSM wants to access the inode during
6370 * d_instantiate it needs these. Smack checks to see
6371 * if the filesystem supports xattrs by looking at the
6372 * ops vector.
6373 */
6374 inode->i_op = &btrfs_special_inode_operations;
6375 init_special_inode(inode, inode->i_mode, rdev);
6376
6377 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6378 if (err)
6379 goto out_unlock_inode;
6380
6381 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6382 0, index);
6383 if (err) {
6384 goto out_unlock_inode;
6385 } else {
6386 btrfs_update_inode(trans, root, inode);
6387 unlock_new_inode(inode);
6388 d_instantiate(dentry, inode);
6389 }
6390
6391 out_unlock:
6392 btrfs_end_transaction(trans);
6393 btrfs_balance_delayed_items(fs_info);
6394 btrfs_btree_balance_dirty(fs_info);
6395 if (drop_inode) {
6396 inode_dec_link_count(inode);
6397 iput(inode);
6398 }
6399 return err;
6400
6401 out_unlock_inode:
6402 drop_inode = 1;
6403 unlock_new_inode(inode);
6404 goto out_unlock;
6405
6406 }
6407
6408 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6409 umode_t mode, bool excl)
6410 {
6411 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6412 struct btrfs_trans_handle *trans;
6413 struct btrfs_root *root = BTRFS_I(dir)->root;
6414 struct inode *inode = NULL;
6415 int drop_inode_on_err = 0;
6416 int err;
6417 u64 objectid;
6418 u64 index = 0;
6419
6420 /*
6421 * 2 for inode item and ref
6422 * 2 for dir items
6423 * 1 for xattr if selinux is on
6424 */
6425 trans = btrfs_start_transaction(root, 5);
6426 if (IS_ERR(trans))
6427 return PTR_ERR(trans);
6428
6429 err = btrfs_find_free_ino(root, &objectid);
6430 if (err)
6431 goto out_unlock;
6432
6433 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6434 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6435 mode, &index);
6436 if (IS_ERR(inode)) {
6437 err = PTR_ERR(inode);
6438 goto out_unlock;
6439 }
6440 drop_inode_on_err = 1;
6441 /*
6442 * If the active LSM wants to access the inode during
6443 * d_instantiate it needs these. Smack checks to see
6444 * if the filesystem supports xattrs by looking at the
6445 * ops vector.
6446 */
6447 inode->i_fop = &btrfs_file_operations;
6448 inode->i_op = &btrfs_file_inode_operations;
6449 inode->i_mapping->a_ops = &btrfs_aops;
6450
6451 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6452 if (err)
6453 goto out_unlock_inode;
6454
6455 err = btrfs_update_inode(trans, root, inode);
6456 if (err)
6457 goto out_unlock_inode;
6458
6459 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6460 0, index);
6461 if (err)
6462 goto out_unlock_inode;
6463
6464 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6465 unlock_new_inode(inode);
6466 d_instantiate(dentry, inode);
6467
6468 out_unlock:
6469 btrfs_end_transaction(trans);
6470 if (err && drop_inode_on_err) {
6471 inode_dec_link_count(inode);
6472 iput(inode);
6473 }
6474 btrfs_balance_delayed_items(fs_info);
6475 btrfs_btree_balance_dirty(fs_info);
6476 return err;
6477
6478 out_unlock_inode:
6479 unlock_new_inode(inode);
6480 goto out_unlock;
6481
6482 }
6483
6484 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6485 struct dentry *dentry)
6486 {
6487 struct btrfs_trans_handle *trans = NULL;
6488 struct btrfs_root *root = BTRFS_I(dir)->root;
6489 struct inode *inode = d_inode(old_dentry);
6490 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6491 u64 index;
6492 int err;
6493 int drop_inode = 0;
6494
6495 /* do not allow sys_link's with other subvols of the same device */
6496 if (root->objectid != BTRFS_I(inode)->root->objectid)
6497 return -EXDEV;
6498
6499 if (inode->i_nlink >= BTRFS_LINK_MAX)
6500 return -EMLINK;
6501
6502 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6503 if (err)
6504 goto fail;
6505
6506 /*
6507 * 2 items for inode and inode ref
6508 * 2 items for dir items
6509 * 1 item for parent inode
6510 */
6511 trans = btrfs_start_transaction(root, 5);
6512 if (IS_ERR(trans)) {
6513 err = PTR_ERR(trans);
6514 trans = NULL;
6515 goto fail;
6516 }
6517
6518 /* There are several dir indexes for this inode, clear the cache. */
6519 BTRFS_I(inode)->dir_index = 0ULL;
6520 inc_nlink(inode);
6521 inode_inc_iversion(inode);
6522 inode->i_ctime = current_time(inode);
6523 ihold(inode);
6524 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6525
6526 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6527 1, index);
6528
6529 if (err) {
6530 drop_inode = 1;
6531 } else {
6532 struct dentry *parent = dentry->d_parent;
6533 err = btrfs_update_inode(trans, root, inode);
6534 if (err)
6535 goto fail;
6536 if (inode->i_nlink == 1) {
6537 /*
6538 * If new hard link count is 1, it's a file created
6539 * with open(2) O_TMPFILE flag.
6540 */
6541 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6542 if (err)
6543 goto fail;
6544 }
6545 d_instantiate(dentry, inode);
6546 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6547 }
6548
6549 btrfs_balance_delayed_items(fs_info);
6550 fail:
6551 if (trans)
6552 btrfs_end_transaction(trans);
6553 if (drop_inode) {
6554 inode_dec_link_count(inode);
6555 iput(inode);
6556 }
6557 btrfs_btree_balance_dirty(fs_info);
6558 return err;
6559 }
6560
6561 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6562 {
6563 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6564 struct inode *inode = NULL;
6565 struct btrfs_trans_handle *trans;
6566 struct btrfs_root *root = BTRFS_I(dir)->root;
6567 int err = 0;
6568 int drop_on_err = 0;
6569 u64 objectid = 0;
6570 u64 index = 0;
6571
6572 /*
6573 * 2 items for inode and ref
6574 * 2 items for dir items
6575 * 1 for xattr if selinux is on
6576 */
6577 trans = btrfs_start_transaction(root, 5);
6578 if (IS_ERR(trans))
6579 return PTR_ERR(trans);
6580
6581 err = btrfs_find_free_ino(root, &objectid);
6582 if (err)
6583 goto out_fail;
6584
6585 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6586 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6587 S_IFDIR | mode, &index);
6588 if (IS_ERR(inode)) {
6589 err = PTR_ERR(inode);
6590 goto out_fail;
6591 }
6592
6593 drop_on_err = 1;
6594 /* these must be set before we unlock the inode */
6595 inode->i_op = &btrfs_dir_inode_operations;
6596 inode->i_fop = &btrfs_dir_file_operations;
6597
6598 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6599 if (err)
6600 goto out_fail_inode;
6601
6602 btrfs_i_size_write(BTRFS_I(inode), 0);
6603 err = btrfs_update_inode(trans, root, inode);
6604 if (err)
6605 goto out_fail_inode;
6606
6607 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6608 dentry->d_name.name,
6609 dentry->d_name.len, 0, index);
6610 if (err)
6611 goto out_fail_inode;
6612
6613 d_instantiate(dentry, inode);
6614 /*
6615 * mkdir is special. We're unlocking after we call d_instantiate
6616 * to avoid a race with nfsd calling d_instantiate.
6617 */
6618 unlock_new_inode(inode);
6619 drop_on_err = 0;
6620
6621 out_fail:
6622 btrfs_end_transaction(trans);
6623 if (drop_on_err) {
6624 inode_dec_link_count(inode);
6625 iput(inode);
6626 }
6627 btrfs_balance_delayed_items(fs_info);
6628 btrfs_btree_balance_dirty(fs_info);
6629 return err;
6630
6631 out_fail_inode:
6632 unlock_new_inode(inode);
6633 goto out_fail;
6634 }
6635
6636 /* Find next extent map of a given extent map, caller needs to ensure locks */
6637 static struct extent_map *next_extent_map(struct extent_map *em)
6638 {
6639 struct rb_node *next;
6640
6641 next = rb_next(&em->rb_node);
6642 if (!next)
6643 return NULL;
6644 return container_of(next, struct extent_map, rb_node);
6645 }
6646
6647 static struct extent_map *prev_extent_map(struct extent_map *em)
6648 {
6649 struct rb_node *prev;
6650
6651 prev = rb_prev(&em->rb_node);
6652 if (!prev)
6653 return NULL;
6654 return container_of(prev, struct extent_map, rb_node);
6655 }
6656
6657 /* helper for btfs_get_extent. Given an existing extent in the tree,
6658 * the existing extent is the nearest extent to map_start,
6659 * and an extent that you want to insert, deal with overlap and insert
6660 * the best fitted new extent into the tree.
6661 */
6662 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6663 struct extent_map *existing,
6664 struct extent_map *em,
6665 u64 map_start)
6666 {
6667 struct extent_map *prev;
6668 struct extent_map *next;
6669 u64 start;
6670 u64 end;
6671 u64 start_diff;
6672
6673 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6674
6675 if (existing->start > map_start) {
6676 next = existing;
6677 prev = prev_extent_map(next);
6678 } else {
6679 prev = existing;
6680 next = next_extent_map(prev);
6681 }
6682
6683 start = prev ? extent_map_end(prev) : em->start;
6684 start = max_t(u64, start, em->start);
6685 end = next ? next->start : extent_map_end(em);
6686 end = min_t(u64, end, extent_map_end(em));
6687 start_diff = start - em->start;
6688 em->start = start;
6689 em->len = end - start;
6690 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6691 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6692 em->block_start += start_diff;
6693 em->block_len -= start_diff;
6694 }
6695 return add_extent_mapping(em_tree, em, 0);
6696 }
6697
6698 static noinline int uncompress_inline(struct btrfs_path *path,
6699 struct page *page,
6700 size_t pg_offset, u64 extent_offset,
6701 struct btrfs_file_extent_item *item)
6702 {
6703 int ret;
6704 struct extent_buffer *leaf = path->nodes[0];
6705 char *tmp;
6706 size_t max_size;
6707 unsigned long inline_size;
6708 unsigned long ptr;
6709 int compress_type;
6710
6711 WARN_ON(pg_offset != 0);
6712 compress_type = btrfs_file_extent_compression(leaf, item);
6713 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6714 inline_size = btrfs_file_extent_inline_item_len(leaf,
6715 btrfs_item_nr(path->slots[0]));
6716 tmp = kmalloc(inline_size, GFP_NOFS);
6717 if (!tmp)
6718 return -ENOMEM;
6719 ptr = btrfs_file_extent_inline_start(item);
6720
6721 read_extent_buffer(leaf, tmp, ptr, inline_size);
6722
6723 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6724 ret = btrfs_decompress(compress_type, tmp, page,
6725 extent_offset, inline_size, max_size);
6726 kfree(tmp);
6727 return ret;
6728 }
6729
6730 /*
6731 * a bit scary, this does extent mapping from logical file offset to the disk.
6732 * the ugly parts come from merging extents from the disk with the in-ram
6733 * representation. This gets more complex because of the data=ordered code,
6734 * where the in-ram extents might be locked pending data=ordered completion.
6735 *
6736 * This also copies inline extents directly into the page.
6737 */
6738
6739 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6740 struct page *page,
6741 size_t pg_offset, u64 start, u64 len,
6742 int create)
6743 {
6744 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6745 int ret;
6746 int err = 0;
6747 u64 extent_start = 0;
6748 u64 extent_end = 0;
6749 u64 objectid = btrfs_ino(inode);
6750 u32 found_type;
6751 struct btrfs_path *path = NULL;
6752 struct btrfs_root *root = inode->root;
6753 struct btrfs_file_extent_item *item;
6754 struct extent_buffer *leaf;
6755 struct btrfs_key found_key;
6756 struct extent_map *em = NULL;
6757 struct extent_map_tree *em_tree = &inode->extent_tree;
6758 struct extent_io_tree *io_tree = &inode->io_tree;
6759 struct btrfs_trans_handle *trans = NULL;
6760 const bool new_inline = !page || create;
6761
6762 again:
6763 read_lock(&em_tree->lock);
6764 em = lookup_extent_mapping(em_tree, start, len);
6765 if (em)
6766 em->bdev = fs_info->fs_devices->latest_bdev;
6767 read_unlock(&em_tree->lock);
6768
6769 if (em) {
6770 if (em->start > start || em->start + em->len <= start)
6771 free_extent_map(em);
6772 else if (em->block_start == EXTENT_MAP_INLINE && page)
6773 free_extent_map(em);
6774 else
6775 goto out;
6776 }
6777 em = alloc_extent_map();
6778 if (!em) {
6779 err = -ENOMEM;
6780 goto out;
6781 }
6782 em->bdev = fs_info->fs_devices->latest_bdev;
6783 em->start = EXTENT_MAP_HOLE;
6784 em->orig_start = EXTENT_MAP_HOLE;
6785 em->len = (u64)-1;
6786 em->block_len = (u64)-1;
6787
6788 if (!path) {
6789 path = btrfs_alloc_path();
6790 if (!path) {
6791 err = -ENOMEM;
6792 goto out;
6793 }
6794 /*
6795 * Chances are we'll be called again, so go ahead and do
6796 * readahead
6797 */
6798 path->reada = READA_FORWARD;
6799 }
6800
6801 ret = btrfs_lookup_file_extent(trans, root, path,
6802 objectid, start, trans != NULL);
6803 if (ret < 0) {
6804 err = ret;
6805 goto out;
6806 }
6807
6808 if (ret != 0) {
6809 if (path->slots[0] == 0)
6810 goto not_found;
6811 path->slots[0]--;
6812 }
6813
6814 leaf = path->nodes[0];
6815 item = btrfs_item_ptr(leaf, path->slots[0],
6816 struct btrfs_file_extent_item);
6817 /* are we inside the extent that was found? */
6818 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6819 found_type = found_key.type;
6820 if (found_key.objectid != objectid ||
6821 found_type != BTRFS_EXTENT_DATA_KEY) {
6822 /*
6823 * If we backup past the first extent we want to move forward
6824 * and see if there is an extent in front of us, otherwise we'll
6825 * say there is a hole for our whole search range which can
6826 * cause problems.
6827 */
6828 extent_end = start;
6829 goto next;
6830 }
6831
6832 found_type = btrfs_file_extent_type(leaf, item);
6833 extent_start = found_key.offset;
6834 if (found_type == BTRFS_FILE_EXTENT_REG ||
6835 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6836 extent_end = extent_start +
6837 btrfs_file_extent_num_bytes(leaf, item);
6838 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6839 size_t size;
6840 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6841 extent_end = ALIGN(extent_start + size,
6842 fs_info->sectorsize);
6843 }
6844 next:
6845 if (start >= extent_end) {
6846 path->slots[0]++;
6847 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6848 ret = btrfs_next_leaf(root, path);
6849 if (ret < 0) {
6850 err = ret;
6851 goto out;
6852 }
6853 if (ret > 0)
6854 goto not_found;
6855 leaf = path->nodes[0];
6856 }
6857 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6858 if (found_key.objectid != objectid ||
6859 found_key.type != BTRFS_EXTENT_DATA_KEY)
6860 goto not_found;
6861 if (start + len <= found_key.offset)
6862 goto not_found;
6863 if (start > found_key.offset)
6864 goto next;
6865 em->start = start;
6866 em->orig_start = start;
6867 em->len = found_key.offset - start;
6868 goto not_found_em;
6869 }
6870
6871 btrfs_extent_item_to_extent_map(inode, path, item,
6872 new_inline, em);
6873
6874 if (found_type == BTRFS_FILE_EXTENT_REG ||
6875 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6876 goto insert;
6877 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6878 unsigned long ptr;
6879 char *map;
6880 size_t size;
6881 size_t extent_offset;
6882 size_t copy_size;
6883
6884 if (new_inline)
6885 goto out;
6886
6887 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6888 extent_offset = page_offset(page) + pg_offset - extent_start;
6889 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6890 size - extent_offset);
6891 em->start = extent_start + extent_offset;
6892 em->len = ALIGN(copy_size, fs_info->sectorsize);
6893 em->orig_block_len = em->len;
6894 em->orig_start = em->start;
6895 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6896 if (create == 0 && !PageUptodate(page)) {
6897 if (btrfs_file_extent_compression(leaf, item) !=
6898 BTRFS_COMPRESS_NONE) {
6899 ret = uncompress_inline(path, page, pg_offset,
6900 extent_offset, item);
6901 if (ret) {
6902 err = ret;
6903 goto out;
6904 }
6905 } else {
6906 map = kmap(page);
6907 read_extent_buffer(leaf, map + pg_offset, ptr,
6908 copy_size);
6909 if (pg_offset + copy_size < PAGE_SIZE) {
6910 memset(map + pg_offset + copy_size, 0,
6911 PAGE_SIZE - pg_offset -
6912 copy_size);
6913 }
6914 kunmap(page);
6915 }
6916 flush_dcache_page(page);
6917 } else if (create && PageUptodate(page)) {
6918 BUG();
6919 if (!trans) {
6920 kunmap(page);
6921 free_extent_map(em);
6922 em = NULL;
6923
6924 btrfs_release_path(path);
6925 trans = btrfs_join_transaction(root);
6926
6927 if (IS_ERR(trans))
6928 return ERR_CAST(trans);
6929 goto again;
6930 }
6931 map = kmap(page);
6932 write_extent_buffer(leaf, map + pg_offset, ptr,
6933 copy_size);
6934 kunmap(page);
6935 btrfs_mark_buffer_dirty(leaf);
6936 }
6937 set_extent_uptodate(io_tree, em->start,
6938 extent_map_end(em) - 1, NULL, GFP_NOFS);
6939 goto insert;
6940 }
6941 not_found:
6942 em->start = start;
6943 em->orig_start = start;
6944 em->len = len;
6945 not_found_em:
6946 em->block_start = EXTENT_MAP_HOLE;
6947 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6948 insert:
6949 btrfs_release_path(path);
6950 if (em->start > start || extent_map_end(em) <= start) {
6951 btrfs_err(fs_info,
6952 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6953 em->start, em->len, start, len);
6954 err = -EIO;
6955 goto out;
6956 }
6957
6958 err = 0;
6959 write_lock(&em_tree->lock);
6960 ret = add_extent_mapping(em_tree, em, 0);
6961 /* it is possible that someone inserted the extent into the tree
6962 * while we had the lock dropped. It is also possible that
6963 * an overlapping map exists in the tree
6964 */
6965 if (ret == -EEXIST) {
6966 struct extent_map *existing;
6967
6968 ret = 0;
6969
6970 existing = search_extent_mapping(em_tree, start, len);
6971 /*
6972 * existing will always be non-NULL, since there must be
6973 * extent causing the -EEXIST.
6974 */
6975 if (existing->start == em->start &&
6976 extent_map_end(existing) >= extent_map_end(em) &&
6977 em->block_start == existing->block_start) {
6978 /*
6979 * The existing extent map already encompasses the
6980 * entire extent map we tried to add.
6981 */
6982 free_extent_map(em);
6983 em = existing;
6984 err = 0;
6985
6986 } else if (start >= extent_map_end(existing) ||
6987 start <= existing->start) {
6988 /*
6989 * The existing extent map is the one nearest to
6990 * the [start, start + len) range which overlaps
6991 */
6992 err = merge_extent_mapping(em_tree, existing,
6993 em, start);
6994 free_extent_map(existing);
6995 if (err) {
6996 free_extent_map(em);
6997 em = NULL;
6998 }
6999 } else {
7000 free_extent_map(em);
7001 em = existing;
7002 err = 0;
7003 }
7004 }
7005 write_unlock(&em_tree->lock);
7006 out:
7007
7008 trace_btrfs_get_extent(root, inode, em);
7009
7010 btrfs_free_path(path);
7011 if (trans) {
7012 ret = btrfs_end_transaction(trans);
7013 if (!err)
7014 err = ret;
7015 }
7016 if (err) {
7017 free_extent_map(em);
7018 return ERR_PTR(err);
7019 }
7020 BUG_ON(!em); /* Error is always set */
7021 return em;
7022 }
7023
7024 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7025 struct page *page,
7026 size_t pg_offset, u64 start, u64 len,
7027 int create)
7028 {
7029 struct extent_map *em;
7030 struct extent_map *hole_em = NULL;
7031 u64 range_start = start;
7032 u64 end;
7033 u64 found;
7034 u64 found_end;
7035 int err = 0;
7036
7037 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7038 if (IS_ERR(em))
7039 return em;
7040 if (em) {
7041 /*
7042 * if our em maps to
7043 * - a hole or
7044 * - a pre-alloc extent,
7045 * there might actually be delalloc bytes behind it.
7046 */
7047 if (em->block_start != EXTENT_MAP_HOLE &&
7048 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7049 return em;
7050 else
7051 hole_em = em;
7052 }
7053
7054 /* check to see if we've wrapped (len == -1 or similar) */
7055 end = start + len;
7056 if (end < start)
7057 end = (u64)-1;
7058 else
7059 end -= 1;
7060
7061 em = NULL;
7062
7063 /* ok, we didn't find anything, lets look for delalloc */
7064 found = count_range_bits(&inode->io_tree, &range_start,
7065 end, len, EXTENT_DELALLOC, 1);
7066 found_end = range_start + found;
7067 if (found_end < range_start)
7068 found_end = (u64)-1;
7069
7070 /*
7071 * we didn't find anything useful, return
7072 * the original results from get_extent()
7073 */
7074 if (range_start > end || found_end <= start) {
7075 em = hole_em;
7076 hole_em = NULL;
7077 goto out;
7078 }
7079
7080 /* adjust the range_start to make sure it doesn't
7081 * go backwards from the start they passed in
7082 */
7083 range_start = max(start, range_start);
7084 found = found_end - range_start;
7085
7086 if (found > 0) {
7087 u64 hole_start = start;
7088 u64 hole_len = len;
7089
7090 em = alloc_extent_map();
7091 if (!em) {
7092 err = -ENOMEM;
7093 goto out;
7094 }
7095 /*
7096 * when btrfs_get_extent can't find anything it
7097 * returns one huge hole
7098 *
7099 * make sure what it found really fits our range, and
7100 * adjust to make sure it is based on the start from
7101 * the caller
7102 */
7103 if (hole_em) {
7104 u64 calc_end = extent_map_end(hole_em);
7105
7106 if (calc_end <= start || (hole_em->start > end)) {
7107 free_extent_map(hole_em);
7108 hole_em = NULL;
7109 } else {
7110 hole_start = max(hole_em->start, start);
7111 hole_len = calc_end - hole_start;
7112 }
7113 }
7114 em->bdev = NULL;
7115 if (hole_em && range_start > hole_start) {
7116 /* our hole starts before our delalloc, so we
7117 * have to return just the parts of the hole
7118 * that go until the delalloc starts
7119 */
7120 em->len = min(hole_len,
7121 range_start - hole_start);
7122 em->start = hole_start;
7123 em->orig_start = hole_start;
7124 /*
7125 * don't adjust block start at all,
7126 * it is fixed at EXTENT_MAP_HOLE
7127 */
7128 em->block_start = hole_em->block_start;
7129 em->block_len = hole_len;
7130 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7131 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7132 } else {
7133 em->start = range_start;
7134 em->len = found;
7135 em->orig_start = range_start;
7136 em->block_start = EXTENT_MAP_DELALLOC;
7137 em->block_len = found;
7138 }
7139 } else if (hole_em) {
7140 return hole_em;
7141 }
7142 out:
7143
7144 free_extent_map(hole_em);
7145 if (err) {
7146 free_extent_map(em);
7147 return ERR_PTR(err);
7148 }
7149 return em;
7150 }
7151
7152 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7153 const u64 start,
7154 const u64 len,
7155 const u64 orig_start,
7156 const u64 block_start,
7157 const u64 block_len,
7158 const u64 orig_block_len,
7159 const u64 ram_bytes,
7160 const int type)
7161 {
7162 struct extent_map *em = NULL;
7163 int ret;
7164
7165 if (type != BTRFS_ORDERED_NOCOW) {
7166 em = create_io_em(inode, start, len, orig_start,
7167 block_start, block_len, orig_block_len,
7168 ram_bytes,
7169 BTRFS_COMPRESS_NONE, /* compress_type */
7170 type);
7171 if (IS_ERR(em))
7172 goto out;
7173 }
7174 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7175 len, block_len, type);
7176 if (ret) {
7177 if (em) {
7178 free_extent_map(em);
7179 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7180 start + len - 1, 0);
7181 }
7182 em = ERR_PTR(ret);
7183 }
7184 out:
7185
7186 return em;
7187 }
7188
7189 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7190 u64 start, u64 len)
7191 {
7192 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7193 struct btrfs_root *root = BTRFS_I(inode)->root;
7194 struct extent_map *em;
7195 struct btrfs_key ins;
7196 u64 alloc_hint;
7197 int ret;
7198
7199 alloc_hint = get_extent_allocation_hint(inode, start, len);
7200 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7201 0, alloc_hint, &ins, 1, 1);
7202 if (ret)
7203 return ERR_PTR(ret);
7204
7205 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7206 ins.objectid, ins.offset, ins.offset,
7207 ins.offset, BTRFS_ORDERED_REGULAR);
7208 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7209 if (IS_ERR(em))
7210 btrfs_free_reserved_extent(fs_info, ins.objectid,
7211 ins.offset, 1);
7212
7213 return em;
7214 }
7215
7216 /*
7217 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7218 * block must be cow'd
7219 */
7220 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7221 u64 *orig_start, u64 *orig_block_len,
7222 u64 *ram_bytes)
7223 {
7224 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7225 struct btrfs_path *path;
7226 int ret;
7227 struct extent_buffer *leaf;
7228 struct btrfs_root *root = BTRFS_I(inode)->root;
7229 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7230 struct btrfs_file_extent_item *fi;
7231 struct btrfs_key key;
7232 u64 disk_bytenr;
7233 u64 backref_offset;
7234 u64 extent_end;
7235 u64 num_bytes;
7236 int slot;
7237 int found_type;
7238 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7239
7240 path = btrfs_alloc_path();
7241 if (!path)
7242 return -ENOMEM;
7243
7244 ret = btrfs_lookup_file_extent(NULL, root, path,
7245 btrfs_ino(BTRFS_I(inode)), offset, 0);
7246 if (ret < 0)
7247 goto out;
7248
7249 slot = path->slots[0];
7250 if (ret == 1) {
7251 if (slot == 0) {
7252 /* can't find the item, must cow */
7253 ret = 0;
7254 goto out;
7255 }
7256 slot--;
7257 }
7258 ret = 0;
7259 leaf = path->nodes[0];
7260 btrfs_item_key_to_cpu(leaf, &key, slot);
7261 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7262 key.type != BTRFS_EXTENT_DATA_KEY) {
7263 /* not our file or wrong item type, must cow */
7264 goto out;
7265 }
7266
7267 if (key.offset > offset) {
7268 /* Wrong offset, must cow */
7269 goto out;
7270 }
7271
7272 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7273 found_type = btrfs_file_extent_type(leaf, fi);
7274 if (found_type != BTRFS_FILE_EXTENT_REG &&
7275 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7276 /* not a regular extent, must cow */
7277 goto out;
7278 }
7279
7280 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7281 goto out;
7282
7283 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7284 if (extent_end <= offset)
7285 goto out;
7286
7287 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7288 if (disk_bytenr == 0)
7289 goto out;
7290
7291 if (btrfs_file_extent_compression(leaf, fi) ||
7292 btrfs_file_extent_encryption(leaf, fi) ||
7293 btrfs_file_extent_other_encoding(leaf, fi))
7294 goto out;
7295
7296 backref_offset = btrfs_file_extent_offset(leaf, fi);
7297
7298 if (orig_start) {
7299 *orig_start = key.offset - backref_offset;
7300 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7301 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7302 }
7303
7304 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7305 goto out;
7306
7307 num_bytes = min(offset + *len, extent_end) - offset;
7308 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7309 u64 range_end;
7310
7311 range_end = round_up(offset + num_bytes,
7312 root->fs_info->sectorsize) - 1;
7313 ret = test_range_bit(io_tree, offset, range_end,
7314 EXTENT_DELALLOC, 0, NULL);
7315 if (ret) {
7316 ret = -EAGAIN;
7317 goto out;
7318 }
7319 }
7320
7321 btrfs_release_path(path);
7322
7323 /*
7324 * look for other files referencing this extent, if we
7325 * find any we must cow
7326 */
7327
7328 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7329 key.offset - backref_offset, disk_bytenr);
7330 if (ret) {
7331 ret = 0;
7332 goto out;
7333 }
7334
7335 /*
7336 * adjust disk_bytenr and num_bytes to cover just the bytes
7337 * in this extent we are about to write. If there
7338 * are any csums in that range we have to cow in order
7339 * to keep the csums correct
7340 */
7341 disk_bytenr += backref_offset;
7342 disk_bytenr += offset - key.offset;
7343 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7344 goto out;
7345 /*
7346 * all of the above have passed, it is safe to overwrite this extent
7347 * without cow
7348 */
7349 *len = num_bytes;
7350 ret = 1;
7351 out:
7352 btrfs_free_path(path);
7353 return ret;
7354 }
7355
7356 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7357 {
7358 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7359 int found = false;
7360 void **pagep = NULL;
7361 struct page *page = NULL;
7362 int start_idx;
7363 int end_idx;
7364
7365 start_idx = start >> PAGE_SHIFT;
7366
7367 /*
7368 * end is the last byte in the last page. end == start is legal
7369 */
7370 end_idx = end >> PAGE_SHIFT;
7371
7372 rcu_read_lock();
7373
7374 /* Most of the code in this while loop is lifted from
7375 * find_get_page. It's been modified to begin searching from a
7376 * page and return just the first page found in that range. If the
7377 * found idx is less than or equal to the end idx then we know that
7378 * a page exists. If no pages are found or if those pages are
7379 * outside of the range then we're fine (yay!) */
7380 while (page == NULL &&
7381 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7382 page = radix_tree_deref_slot(pagep);
7383 if (unlikely(!page))
7384 break;
7385
7386 if (radix_tree_exception(page)) {
7387 if (radix_tree_deref_retry(page)) {
7388 page = NULL;
7389 continue;
7390 }
7391 /*
7392 * Otherwise, shmem/tmpfs must be storing a swap entry
7393 * here as an exceptional entry: so return it without
7394 * attempting to raise page count.
7395 */
7396 page = NULL;
7397 break; /* TODO: Is this relevant for this use case? */
7398 }
7399
7400 if (!page_cache_get_speculative(page)) {
7401 page = NULL;
7402 continue;
7403 }
7404
7405 /*
7406 * Has the page moved?
7407 * This is part of the lockless pagecache protocol. See
7408 * include/linux/pagemap.h for details.
7409 */
7410 if (unlikely(page != *pagep)) {
7411 put_page(page);
7412 page = NULL;
7413 }
7414 }
7415
7416 if (page) {
7417 if (page->index <= end_idx)
7418 found = true;
7419 put_page(page);
7420 }
7421
7422 rcu_read_unlock();
7423 return found;
7424 }
7425
7426 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7427 struct extent_state **cached_state, int writing)
7428 {
7429 struct btrfs_ordered_extent *ordered;
7430 int ret = 0;
7431
7432 while (1) {
7433 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7434 cached_state);
7435 /*
7436 * We're concerned with the entire range that we're going to be
7437 * doing DIO to, so we need to make sure there's no ordered
7438 * extents in this range.
7439 */
7440 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7441 lockend - lockstart + 1);
7442
7443 /*
7444 * We need to make sure there are no buffered pages in this
7445 * range either, we could have raced between the invalidate in
7446 * generic_file_direct_write and locking the extent. The
7447 * invalidate needs to happen so that reads after a write do not
7448 * get stale data.
7449 */
7450 if (!ordered &&
7451 (!writing ||
7452 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7453 break;
7454
7455 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7456 cached_state, GFP_NOFS);
7457
7458 if (ordered) {
7459 /*
7460 * If we are doing a DIO read and the ordered extent we
7461 * found is for a buffered write, we can not wait for it
7462 * to complete and retry, because if we do so we can
7463 * deadlock with concurrent buffered writes on page
7464 * locks. This happens only if our DIO read covers more
7465 * than one extent map, if at this point has already
7466 * created an ordered extent for a previous extent map
7467 * and locked its range in the inode's io tree, and a
7468 * concurrent write against that previous extent map's
7469 * range and this range started (we unlock the ranges
7470 * in the io tree only when the bios complete and
7471 * buffered writes always lock pages before attempting
7472 * to lock range in the io tree).
7473 */
7474 if (writing ||
7475 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7476 btrfs_start_ordered_extent(inode, ordered, 1);
7477 else
7478 ret = -ENOTBLK;
7479 btrfs_put_ordered_extent(ordered);
7480 } else {
7481 /*
7482 * We could trigger writeback for this range (and wait
7483 * for it to complete) and then invalidate the pages for
7484 * this range (through invalidate_inode_pages2_range()),
7485 * but that can lead us to a deadlock with a concurrent
7486 * call to readpages() (a buffered read or a defrag call
7487 * triggered a readahead) on a page lock due to an
7488 * ordered dio extent we created before but did not have
7489 * yet a corresponding bio submitted (whence it can not
7490 * complete), which makes readpages() wait for that
7491 * ordered extent to complete while holding a lock on
7492 * that page.
7493 */
7494 ret = -ENOTBLK;
7495 }
7496
7497 if (ret)
7498 break;
7499
7500 cond_resched();
7501 }
7502
7503 return ret;
7504 }
7505
7506 /* The callers of this must take lock_extent() */
7507 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7508 u64 orig_start, u64 block_start,
7509 u64 block_len, u64 orig_block_len,
7510 u64 ram_bytes, int compress_type,
7511 int type)
7512 {
7513 struct extent_map_tree *em_tree;
7514 struct extent_map *em;
7515 struct btrfs_root *root = BTRFS_I(inode)->root;
7516 int ret;
7517
7518 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7519 type == BTRFS_ORDERED_COMPRESSED ||
7520 type == BTRFS_ORDERED_NOCOW ||
7521 type == BTRFS_ORDERED_REGULAR);
7522
7523 em_tree = &BTRFS_I(inode)->extent_tree;
7524 em = alloc_extent_map();
7525 if (!em)
7526 return ERR_PTR(-ENOMEM);
7527
7528 em->start = start;
7529 em->orig_start = orig_start;
7530 em->len = len;
7531 em->block_len = block_len;
7532 em->block_start = block_start;
7533 em->bdev = root->fs_info->fs_devices->latest_bdev;
7534 em->orig_block_len = orig_block_len;
7535 em->ram_bytes = ram_bytes;
7536 em->generation = -1;
7537 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7538 if (type == BTRFS_ORDERED_PREALLOC) {
7539 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7540 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7541 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7542 em->compress_type = compress_type;
7543 }
7544
7545 do {
7546 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7547 em->start + em->len - 1, 0);
7548 write_lock(&em_tree->lock);
7549 ret = add_extent_mapping(em_tree, em, 1);
7550 write_unlock(&em_tree->lock);
7551 /*
7552 * The caller has taken lock_extent(), who could race with us
7553 * to add em?
7554 */
7555 } while (ret == -EEXIST);
7556
7557 if (ret) {
7558 free_extent_map(em);
7559 return ERR_PTR(ret);
7560 }
7561
7562 /* em got 2 refs now, callers needs to do free_extent_map once. */
7563 return em;
7564 }
7565
7566 static void adjust_dio_outstanding_extents(struct inode *inode,
7567 struct btrfs_dio_data *dio_data,
7568 const u64 len)
7569 {
7570 unsigned num_extents = count_max_extents(len);
7571
7572 /*
7573 * If we have an outstanding_extents count still set then we're
7574 * within our reservation, otherwise we need to adjust our inode
7575 * counter appropriately.
7576 */
7577 if (dio_data->outstanding_extents >= num_extents) {
7578 dio_data->outstanding_extents -= num_extents;
7579 } else {
7580 /*
7581 * If dio write length has been split due to no large enough
7582 * contiguous space, we need to compensate our inode counter
7583 * appropriately.
7584 */
7585 u64 num_needed = num_extents - dio_data->outstanding_extents;
7586
7587 spin_lock(&BTRFS_I(inode)->lock);
7588 BTRFS_I(inode)->outstanding_extents += num_needed;
7589 spin_unlock(&BTRFS_I(inode)->lock);
7590 }
7591 }
7592
7593 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7594 struct buffer_head *bh_result, int create)
7595 {
7596 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7597 struct extent_map *em;
7598 struct extent_state *cached_state = NULL;
7599 struct btrfs_dio_data *dio_data = NULL;
7600 u64 start = iblock << inode->i_blkbits;
7601 u64 lockstart, lockend;
7602 u64 len = bh_result->b_size;
7603 int unlock_bits = EXTENT_LOCKED;
7604 int ret = 0;
7605
7606 if (create)
7607 unlock_bits |= EXTENT_DIRTY;
7608 else
7609 len = min_t(u64, len, fs_info->sectorsize);
7610
7611 lockstart = start;
7612 lockend = start + len - 1;
7613
7614 if (current->journal_info) {
7615 /*
7616 * Need to pull our outstanding extents and set journal_info to NULL so
7617 * that anything that needs to check if there's a transaction doesn't get
7618 * confused.
7619 */
7620 dio_data = current->journal_info;
7621 current->journal_info = NULL;
7622 }
7623
7624 /*
7625 * If this errors out it's because we couldn't invalidate pagecache for
7626 * this range and we need to fallback to buffered.
7627 */
7628 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7629 create)) {
7630 ret = -ENOTBLK;
7631 goto err;
7632 }
7633
7634 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7635 if (IS_ERR(em)) {
7636 ret = PTR_ERR(em);
7637 goto unlock_err;
7638 }
7639
7640 /*
7641 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7642 * io. INLINE is special, and we could probably kludge it in here, but
7643 * it's still buffered so for safety lets just fall back to the generic
7644 * buffered path.
7645 *
7646 * For COMPRESSED we _have_ to read the entire extent in so we can
7647 * decompress it, so there will be buffering required no matter what we
7648 * do, so go ahead and fallback to buffered.
7649 *
7650 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7651 * to buffered IO. Don't blame me, this is the price we pay for using
7652 * the generic code.
7653 */
7654 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7655 em->block_start == EXTENT_MAP_INLINE) {
7656 free_extent_map(em);
7657 ret = -ENOTBLK;
7658 goto unlock_err;
7659 }
7660
7661 /* Just a good old fashioned hole, return */
7662 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7663 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7664 free_extent_map(em);
7665 goto unlock_err;
7666 }
7667
7668 /*
7669 * We don't allocate a new extent in the following cases
7670 *
7671 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7672 * existing extent.
7673 * 2) The extent is marked as PREALLOC. We're good to go here and can
7674 * just use the extent.
7675 *
7676 */
7677 if (!create) {
7678 len = min(len, em->len - (start - em->start));
7679 lockstart = start + len;
7680 goto unlock;
7681 }
7682
7683 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7684 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7685 em->block_start != EXTENT_MAP_HOLE)) {
7686 int type;
7687 u64 block_start, orig_start, orig_block_len, ram_bytes;
7688
7689 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7690 type = BTRFS_ORDERED_PREALLOC;
7691 else
7692 type = BTRFS_ORDERED_NOCOW;
7693 len = min(len, em->len - (start - em->start));
7694 block_start = em->block_start + (start - em->start);
7695
7696 if (can_nocow_extent(inode, start, &len, &orig_start,
7697 &orig_block_len, &ram_bytes) == 1 &&
7698 btrfs_inc_nocow_writers(fs_info, block_start)) {
7699 struct extent_map *em2;
7700
7701 em2 = btrfs_create_dio_extent(inode, start, len,
7702 orig_start, block_start,
7703 len, orig_block_len,
7704 ram_bytes, type);
7705 btrfs_dec_nocow_writers(fs_info, block_start);
7706 if (type == BTRFS_ORDERED_PREALLOC) {
7707 free_extent_map(em);
7708 em = em2;
7709 }
7710 if (em2 && IS_ERR(em2)) {
7711 ret = PTR_ERR(em2);
7712 goto unlock_err;
7713 }
7714 /*
7715 * For inode marked NODATACOW or extent marked PREALLOC,
7716 * use the existing or preallocated extent, so does not
7717 * need to adjust btrfs_space_info's bytes_may_use.
7718 */
7719 btrfs_free_reserved_data_space_noquota(inode,
7720 start, len);
7721 goto unlock;
7722 }
7723 }
7724
7725 /*
7726 * this will cow the extent, reset the len in case we changed
7727 * it above
7728 */
7729 len = bh_result->b_size;
7730 free_extent_map(em);
7731 em = btrfs_new_extent_direct(inode, start, len);
7732 if (IS_ERR(em)) {
7733 ret = PTR_ERR(em);
7734 goto unlock_err;
7735 }
7736 len = min(len, em->len - (start - em->start));
7737 unlock:
7738 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7739 inode->i_blkbits;
7740 bh_result->b_size = len;
7741 bh_result->b_bdev = em->bdev;
7742 set_buffer_mapped(bh_result);
7743 if (create) {
7744 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7745 set_buffer_new(bh_result);
7746
7747 /*
7748 * Need to update the i_size under the extent lock so buffered
7749 * readers will get the updated i_size when we unlock.
7750 */
7751 if (!dio_data->overwrite && start + len > i_size_read(inode))
7752 i_size_write(inode, start + len);
7753
7754 adjust_dio_outstanding_extents(inode, dio_data, len);
7755 WARN_ON(dio_data->reserve < len);
7756 dio_data->reserve -= len;
7757 dio_data->unsubmitted_oe_range_end = start + len;
7758 current->journal_info = dio_data;
7759 }
7760
7761 /*
7762 * In the case of write we need to clear and unlock the entire range,
7763 * in the case of read we need to unlock only the end area that we
7764 * aren't using if there is any left over space.
7765 */
7766 if (lockstart < lockend) {
7767 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7768 lockend, unlock_bits, 1, 0,
7769 &cached_state, GFP_NOFS);
7770 } else {
7771 free_extent_state(cached_state);
7772 }
7773
7774 free_extent_map(em);
7775
7776 return 0;
7777
7778 unlock_err:
7779 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7780 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7781 err:
7782 if (dio_data)
7783 current->journal_info = dio_data;
7784 /*
7785 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7786 * write less data then expected, so that we don't underflow our inode's
7787 * outstanding extents counter.
7788 */
7789 if (create && dio_data)
7790 adjust_dio_outstanding_extents(inode, dio_data, len);
7791
7792 return ret;
7793 }
7794
7795 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7796 int mirror_num)
7797 {
7798 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7799 int ret;
7800
7801 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7802
7803 bio_get(bio);
7804
7805 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7806 if (ret)
7807 goto err;
7808
7809 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7810 err:
7811 bio_put(bio);
7812 return ret;
7813 }
7814
7815 static int btrfs_check_dio_repairable(struct inode *inode,
7816 struct bio *failed_bio,
7817 struct io_failure_record *failrec,
7818 int failed_mirror)
7819 {
7820 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7821 int num_copies;
7822
7823 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7824 if (num_copies == 1) {
7825 /*
7826 * we only have a single copy of the data, so don't bother with
7827 * all the retry and error correction code that follows. no
7828 * matter what the error is, it is very likely to persist.
7829 */
7830 btrfs_debug(fs_info,
7831 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7832 num_copies, failrec->this_mirror, failed_mirror);
7833 return 0;
7834 }
7835
7836 failrec->failed_mirror = failed_mirror;
7837 failrec->this_mirror++;
7838 if (failrec->this_mirror == failed_mirror)
7839 failrec->this_mirror++;
7840
7841 if (failrec->this_mirror > num_copies) {
7842 btrfs_debug(fs_info,
7843 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7844 num_copies, failrec->this_mirror, failed_mirror);
7845 return 0;
7846 }
7847
7848 return 1;
7849 }
7850
7851 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7852 struct page *page, unsigned int pgoff,
7853 u64 start, u64 end, int failed_mirror,
7854 bio_end_io_t *repair_endio, void *repair_arg)
7855 {
7856 struct io_failure_record *failrec;
7857 struct bio *bio;
7858 int isector;
7859 int read_mode = 0;
7860 int ret;
7861
7862 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7863
7864 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7865 if (ret)
7866 return ret;
7867
7868 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7869 failed_mirror);
7870 if (!ret) {
7871 free_io_failure(BTRFS_I(inode), failrec);
7872 return -EIO;
7873 }
7874
7875 if ((failed_bio->bi_vcnt > 1)
7876 || (failed_bio->bi_io_vec->bv_len
7877 > btrfs_inode_sectorsize(inode)))
7878 read_mode |= REQ_FAILFAST_DEV;
7879
7880 isector = start - btrfs_io_bio(failed_bio)->logical;
7881 isector >>= inode->i_sb->s_blocksize_bits;
7882 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7883 pgoff, isector, repair_endio, repair_arg);
7884 if (!bio) {
7885 free_io_failure(BTRFS_I(inode), failrec);
7886 return -EIO;
7887 }
7888 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
7889
7890 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7891 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7892 read_mode, failrec->this_mirror, failrec->in_validation);
7893
7894 ret = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7895 if (ret) {
7896 free_io_failure(BTRFS_I(inode), failrec);
7897 bio_put(bio);
7898 }
7899
7900 return ret;
7901 }
7902
7903 struct btrfs_retry_complete {
7904 struct completion done;
7905 struct inode *inode;
7906 u64 start;
7907 int uptodate;
7908 };
7909
7910 static void btrfs_retry_endio_nocsum(struct bio *bio)
7911 {
7912 struct btrfs_retry_complete *done = bio->bi_private;
7913 struct inode *inode;
7914 struct bio_vec *bvec;
7915 int i;
7916
7917 if (bio->bi_error)
7918 goto end;
7919
7920 ASSERT(bio->bi_vcnt == 1);
7921 inode = bio->bi_io_vec->bv_page->mapping->host;
7922 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
7923
7924 done->uptodate = 1;
7925 bio_for_each_segment_all(bvec, bio, i)
7926 clean_io_failure(BTRFS_I(done->inode), done->start, bvec->bv_page, 0);
7927 end:
7928 complete(&done->done);
7929 bio_put(bio);
7930 }
7931
7932 static int __btrfs_correct_data_nocsum(struct inode *inode,
7933 struct btrfs_io_bio *io_bio)
7934 {
7935 struct btrfs_fs_info *fs_info;
7936 struct bio_vec *bvec;
7937 struct btrfs_retry_complete done;
7938 u64 start;
7939 unsigned int pgoff;
7940 u32 sectorsize;
7941 int nr_sectors;
7942 int i;
7943 int ret;
7944
7945 fs_info = BTRFS_I(inode)->root->fs_info;
7946 sectorsize = fs_info->sectorsize;
7947
7948 start = io_bio->logical;
7949 done.inode = inode;
7950
7951 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7952 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7953 pgoff = bvec->bv_offset;
7954
7955 next_block_or_try_again:
7956 done.uptodate = 0;
7957 done.start = start;
7958 init_completion(&done.done);
7959
7960 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7961 pgoff, start, start + sectorsize - 1,
7962 io_bio->mirror_num,
7963 btrfs_retry_endio_nocsum, &done);
7964 if (ret)
7965 return ret;
7966
7967 wait_for_completion(&done.done);
7968
7969 if (!done.uptodate) {
7970 /* We might have another mirror, so try again */
7971 goto next_block_or_try_again;
7972 }
7973
7974 start += sectorsize;
7975
7976 if (nr_sectors--) {
7977 pgoff += sectorsize;
7978 goto next_block_or_try_again;
7979 }
7980 }
7981
7982 return 0;
7983 }
7984
7985 static void btrfs_retry_endio(struct bio *bio)
7986 {
7987 struct btrfs_retry_complete *done = bio->bi_private;
7988 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7989 struct inode *inode;
7990 struct bio_vec *bvec;
7991 u64 start;
7992 int uptodate;
7993 int ret;
7994 int i;
7995
7996 if (bio->bi_error)
7997 goto end;
7998
7999 uptodate = 1;
8000
8001 start = done->start;
8002
8003 ASSERT(bio->bi_vcnt == 1);
8004 inode = bio->bi_io_vec->bv_page->mapping->host;
8005 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
8006
8007 bio_for_each_segment_all(bvec, bio, i) {
8008 ret = __readpage_endio_check(done->inode, io_bio, i,
8009 bvec->bv_page, bvec->bv_offset,
8010 done->start, bvec->bv_len);
8011 if (!ret)
8012 clean_io_failure(BTRFS_I(done->inode), done->start,
8013 bvec->bv_page, bvec->bv_offset);
8014 else
8015 uptodate = 0;
8016 }
8017
8018 done->uptodate = uptodate;
8019 end:
8020 complete(&done->done);
8021 bio_put(bio);
8022 }
8023
8024 static int __btrfs_subio_endio_read(struct inode *inode,
8025 struct btrfs_io_bio *io_bio, int err)
8026 {
8027 struct btrfs_fs_info *fs_info;
8028 struct bio_vec *bvec;
8029 struct btrfs_retry_complete done;
8030 u64 start;
8031 u64 offset = 0;
8032 u32 sectorsize;
8033 int nr_sectors;
8034 unsigned int pgoff;
8035 int csum_pos;
8036 int i;
8037 int ret;
8038
8039 fs_info = BTRFS_I(inode)->root->fs_info;
8040 sectorsize = fs_info->sectorsize;
8041
8042 err = 0;
8043 start = io_bio->logical;
8044 done.inode = inode;
8045
8046 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
8047 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8048
8049 pgoff = bvec->bv_offset;
8050 next_block:
8051 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8052 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8053 bvec->bv_page, pgoff, start,
8054 sectorsize);
8055 if (likely(!ret))
8056 goto next;
8057 try_again:
8058 done.uptodate = 0;
8059 done.start = start;
8060 init_completion(&done.done);
8061
8062 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8063 pgoff, start, start + sectorsize - 1,
8064 io_bio->mirror_num,
8065 btrfs_retry_endio, &done);
8066 if (ret) {
8067 err = ret;
8068 goto next;
8069 }
8070
8071 wait_for_completion(&done.done);
8072
8073 if (!done.uptodate) {
8074 /* We might have another mirror, so try again */
8075 goto try_again;
8076 }
8077 next:
8078 offset += sectorsize;
8079 start += sectorsize;
8080
8081 ASSERT(nr_sectors);
8082
8083 if (--nr_sectors) {
8084 pgoff += sectorsize;
8085 goto next_block;
8086 }
8087 }
8088
8089 return err;
8090 }
8091
8092 static int btrfs_subio_endio_read(struct inode *inode,
8093 struct btrfs_io_bio *io_bio, int err)
8094 {
8095 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8096
8097 if (skip_csum) {
8098 if (unlikely(err))
8099 return __btrfs_correct_data_nocsum(inode, io_bio);
8100 else
8101 return 0;
8102 } else {
8103 return __btrfs_subio_endio_read(inode, io_bio, err);
8104 }
8105 }
8106
8107 static void btrfs_endio_direct_read(struct bio *bio)
8108 {
8109 struct btrfs_dio_private *dip = bio->bi_private;
8110 struct inode *inode = dip->inode;
8111 struct bio *dio_bio;
8112 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8113 int err = bio->bi_error;
8114
8115 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8116 err = btrfs_subio_endio_read(inode, io_bio, err);
8117
8118 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8119 dip->logical_offset + dip->bytes - 1);
8120 dio_bio = dip->dio_bio;
8121
8122 kfree(dip);
8123
8124 dio_bio->bi_error = bio->bi_error;
8125 dio_end_io(dio_bio, bio->bi_error);
8126
8127 if (io_bio->end_io)
8128 io_bio->end_io(io_bio, err);
8129 bio_put(bio);
8130 }
8131
8132 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8133 const u64 offset,
8134 const u64 bytes,
8135 const int uptodate)
8136 {
8137 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8138 struct btrfs_ordered_extent *ordered = NULL;
8139 u64 ordered_offset = offset;
8140 u64 ordered_bytes = bytes;
8141 int ret;
8142
8143 again:
8144 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8145 &ordered_offset,
8146 ordered_bytes,
8147 uptodate);
8148 if (!ret)
8149 goto out_test;
8150
8151 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8152 finish_ordered_fn, NULL, NULL);
8153 btrfs_queue_work(fs_info->endio_write_workers, &ordered->work);
8154 out_test:
8155 /*
8156 * our bio might span multiple ordered extents. If we haven't
8157 * completed the accounting for the whole dio, go back and try again
8158 */
8159 if (ordered_offset < offset + bytes) {
8160 ordered_bytes = offset + bytes - ordered_offset;
8161 ordered = NULL;
8162 goto again;
8163 }
8164 }
8165
8166 static void btrfs_endio_direct_write(struct bio *bio)
8167 {
8168 struct btrfs_dio_private *dip = bio->bi_private;
8169 struct bio *dio_bio = dip->dio_bio;
8170
8171 btrfs_endio_direct_write_update_ordered(dip->inode,
8172 dip->logical_offset,
8173 dip->bytes,
8174 !bio->bi_error);
8175
8176 kfree(dip);
8177
8178 dio_bio->bi_error = bio->bi_error;
8179 dio_end_io(dio_bio, bio->bi_error);
8180 bio_put(bio);
8181 }
8182
8183 static int __btrfs_submit_bio_start_direct_io(struct inode *inode,
8184 struct bio *bio, int mirror_num,
8185 unsigned long bio_flags, u64 offset)
8186 {
8187 int ret;
8188 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8189 BUG_ON(ret); /* -ENOMEM */
8190 return 0;
8191 }
8192
8193 static void btrfs_end_dio_bio(struct bio *bio)
8194 {
8195 struct btrfs_dio_private *dip = bio->bi_private;
8196 int err = bio->bi_error;
8197
8198 if (err)
8199 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8200 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8201 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8202 bio->bi_opf,
8203 (unsigned long long)bio->bi_iter.bi_sector,
8204 bio->bi_iter.bi_size, err);
8205
8206 if (dip->subio_endio)
8207 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8208
8209 if (err) {
8210 dip->errors = 1;
8211
8212 /*
8213 * before atomic variable goto zero, we must make sure
8214 * dip->errors is perceived to be set.
8215 */
8216 smp_mb__before_atomic();
8217 }
8218
8219 /* if there are more bios still pending for this dio, just exit */
8220 if (!atomic_dec_and_test(&dip->pending_bios))
8221 goto out;
8222
8223 if (dip->errors) {
8224 bio_io_error(dip->orig_bio);
8225 } else {
8226 dip->dio_bio->bi_error = 0;
8227 bio_endio(dip->orig_bio);
8228 }
8229 out:
8230 bio_put(bio);
8231 }
8232
8233 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8234 u64 first_sector, gfp_t gfp_flags)
8235 {
8236 struct bio *bio;
8237 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8238 if (bio)
8239 bio_associate_current(bio);
8240 return bio;
8241 }
8242
8243 static inline int btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8244 struct btrfs_dio_private *dip,
8245 struct bio *bio,
8246 u64 file_offset)
8247 {
8248 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8249 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8250 int ret;
8251
8252 /*
8253 * We load all the csum data we need when we submit
8254 * the first bio to reduce the csum tree search and
8255 * contention.
8256 */
8257 if (dip->logical_offset == file_offset) {
8258 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8259 file_offset);
8260 if (ret)
8261 return ret;
8262 }
8263
8264 if (bio == dip->orig_bio)
8265 return 0;
8266
8267 file_offset -= dip->logical_offset;
8268 file_offset >>= inode->i_sb->s_blocksize_bits;
8269 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8270
8271 return 0;
8272 }
8273
8274 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8275 u64 file_offset, int skip_sum,
8276 int async_submit)
8277 {
8278 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8279 struct btrfs_dio_private *dip = bio->bi_private;
8280 bool write = bio_op(bio) == REQ_OP_WRITE;
8281 int ret;
8282
8283 if (async_submit)
8284 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8285
8286 bio_get(bio);
8287
8288 if (!write) {
8289 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8290 if (ret)
8291 goto err;
8292 }
8293
8294 if (skip_sum)
8295 goto map;
8296
8297 if (write && async_submit) {
8298 ret = btrfs_wq_submit_bio(fs_info, inode, bio, 0, 0,
8299 file_offset,
8300 __btrfs_submit_bio_start_direct_io,
8301 __btrfs_submit_bio_done);
8302 goto err;
8303 } else if (write) {
8304 /*
8305 * If we aren't doing async submit, calculate the csum of the
8306 * bio now.
8307 */
8308 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8309 if (ret)
8310 goto err;
8311 } else {
8312 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8313 file_offset);
8314 if (ret)
8315 goto err;
8316 }
8317 map:
8318 ret = btrfs_map_bio(fs_info, bio, 0, async_submit);
8319 err:
8320 bio_put(bio);
8321 return ret;
8322 }
8323
8324 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip,
8325 int skip_sum)
8326 {
8327 struct inode *inode = dip->inode;
8328 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8329 struct btrfs_root *root = BTRFS_I(inode)->root;
8330 struct bio *bio;
8331 struct bio *orig_bio = dip->orig_bio;
8332 struct bio_vec *bvec;
8333 u64 start_sector = orig_bio->bi_iter.bi_sector;
8334 u64 file_offset = dip->logical_offset;
8335 u64 submit_len = 0;
8336 u64 map_length;
8337 u32 blocksize = fs_info->sectorsize;
8338 int async_submit = 0;
8339 int nr_sectors;
8340 int ret;
8341 int i, j;
8342
8343 map_length = orig_bio->bi_iter.bi_size;
8344 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8345 &map_length, NULL, 0);
8346 if (ret)
8347 return -EIO;
8348
8349 if (map_length >= orig_bio->bi_iter.bi_size) {
8350 bio = orig_bio;
8351 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8352 goto submit;
8353 }
8354
8355 /* async crcs make it difficult to collect full stripe writes. */
8356 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8357 async_submit = 0;
8358 else
8359 async_submit = 1;
8360
8361 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8362 if (!bio)
8363 return -ENOMEM;
8364
8365 bio->bi_opf = orig_bio->bi_opf;
8366 bio->bi_private = dip;
8367 bio->bi_end_io = btrfs_end_dio_bio;
8368 btrfs_io_bio(bio)->logical = file_offset;
8369 atomic_inc(&dip->pending_bios);
8370
8371 bio_for_each_segment_all(bvec, orig_bio, j) {
8372 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8373 i = 0;
8374 next_block:
8375 if (unlikely(map_length < submit_len + blocksize ||
8376 bio_add_page(bio, bvec->bv_page, blocksize,
8377 bvec->bv_offset + (i * blocksize)) < blocksize)) {
8378 /*
8379 * inc the count before we submit the bio so
8380 * we know the end IO handler won't happen before
8381 * we inc the count. Otherwise, the dip might get freed
8382 * before we're done setting it up
8383 */
8384 atomic_inc(&dip->pending_bios);
8385 ret = __btrfs_submit_dio_bio(bio, inode,
8386 file_offset, skip_sum,
8387 async_submit);
8388 if (ret) {
8389 bio_put(bio);
8390 atomic_dec(&dip->pending_bios);
8391 goto out_err;
8392 }
8393
8394 start_sector += submit_len >> 9;
8395 file_offset += submit_len;
8396
8397 submit_len = 0;
8398
8399 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8400 start_sector, GFP_NOFS);
8401 if (!bio)
8402 goto out_err;
8403 bio->bi_opf = orig_bio->bi_opf;
8404 bio->bi_private = dip;
8405 bio->bi_end_io = btrfs_end_dio_bio;
8406 btrfs_io_bio(bio)->logical = file_offset;
8407
8408 map_length = orig_bio->bi_iter.bi_size;
8409 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8410 start_sector << 9,
8411 &map_length, NULL, 0);
8412 if (ret) {
8413 bio_put(bio);
8414 goto out_err;
8415 }
8416
8417 goto next_block;
8418 } else {
8419 submit_len += blocksize;
8420 if (--nr_sectors) {
8421 i++;
8422 goto next_block;
8423 }
8424 }
8425 }
8426
8427 submit:
8428 ret = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8429 async_submit);
8430 if (!ret)
8431 return 0;
8432
8433 bio_put(bio);
8434 out_err:
8435 dip->errors = 1;
8436 /*
8437 * before atomic variable goto zero, we must
8438 * make sure dip->errors is perceived to be set.
8439 */
8440 smp_mb__before_atomic();
8441 if (atomic_dec_and_test(&dip->pending_bios))
8442 bio_io_error(dip->orig_bio);
8443
8444 /* bio_end_io() will handle error, so we needn't return it */
8445 return 0;
8446 }
8447
8448 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8449 loff_t file_offset)
8450 {
8451 struct btrfs_dio_private *dip = NULL;
8452 struct bio *io_bio = NULL;
8453 struct btrfs_io_bio *btrfs_bio;
8454 int skip_sum;
8455 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8456 int ret = 0;
8457
8458 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8459
8460 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8461 if (!io_bio) {
8462 ret = -ENOMEM;
8463 goto free_ordered;
8464 }
8465
8466 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8467 if (!dip) {
8468 ret = -ENOMEM;
8469 goto free_ordered;
8470 }
8471
8472 dip->private = dio_bio->bi_private;
8473 dip->inode = inode;
8474 dip->logical_offset = file_offset;
8475 dip->bytes = dio_bio->bi_iter.bi_size;
8476 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8477 io_bio->bi_private = dip;
8478 dip->orig_bio = io_bio;
8479 dip->dio_bio = dio_bio;
8480 atomic_set(&dip->pending_bios, 0);
8481 btrfs_bio = btrfs_io_bio(io_bio);
8482 btrfs_bio->logical = file_offset;
8483
8484 if (write) {
8485 io_bio->bi_end_io = btrfs_endio_direct_write;
8486 } else {
8487 io_bio->bi_end_io = btrfs_endio_direct_read;
8488 dip->subio_endio = btrfs_subio_endio_read;
8489 }
8490
8491 /*
8492 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8493 * even if we fail to submit a bio, because in such case we do the
8494 * corresponding error handling below and it must not be done a second
8495 * time by btrfs_direct_IO().
8496 */
8497 if (write) {
8498 struct btrfs_dio_data *dio_data = current->journal_info;
8499
8500 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8501 dip->bytes;
8502 dio_data->unsubmitted_oe_range_start =
8503 dio_data->unsubmitted_oe_range_end;
8504 }
8505
8506 ret = btrfs_submit_direct_hook(dip, skip_sum);
8507 if (!ret)
8508 return;
8509
8510 if (btrfs_bio->end_io)
8511 btrfs_bio->end_io(btrfs_bio, ret);
8512
8513 free_ordered:
8514 /*
8515 * If we arrived here it means either we failed to submit the dip
8516 * or we either failed to clone the dio_bio or failed to allocate the
8517 * dip. If we cloned the dio_bio and allocated the dip, we can just
8518 * call bio_endio against our io_bio so that we get proper resource
8519 * cleanup if we fail to submit the dip, otherwise, we must do the
8520 * same as btrfs_endio_direct_[write|read] because we can't call these
8521 * callbacks - they require an allocated dip and a clone of dio_bio.
8522 */
8523 if (io_bio && dip) {
8524 io_bio->bi_error = -EIO;
8525 bio_endio(io_bio);
8526 /*
8527 * The end io callbacks free our dip, do the final put on io_bio
8528 * and all the cleanup and final put for dio_bio (through
8529 * dio_end_io()).
8530 */
8531 dip = NULL;
8532 io_bio = NULL;
8533 } else {
8534 if (write)
8535 btrfs_endio_direct_write_update_ordered(inode,
8536 file_offset,
8537 dio_bio->bi_iter.bi_size,
8538 0);
8539 else
8540 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8541 file_offset + dio_bio->bi_iter.bi_size - 1);
8542
8543 dio_bio->bi_error = -EIO;
8544 /*
8545 * Releases and cleans up our dio_bio, no need to bio_put()
8546 * nor bio_endio()/bio_io_error() against dio_bio.
8547 */
8548 dio_end_io(dio_bio, ret);
8549 }
8550 if (io_bio)
8551 bio_put(io_bio);
8552 kfree(dip);
8553 }
8554
8555 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8556 struct kiocb *iocb,
8557 const struct iov_iter *iter, loff_t offset)
8558 {
8559 int seg;
8560 int i;
8561 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8562 ssize_t retval = -EINVAL;
8563
8564 if (offset & blocksize_mask)
8565 goto out;
8566
8567 if (iov_iter_alignment(iter) & blocksize_mask)
8568 goto out;
8569
8570 /* If this is a write we don't need to check anymore */
8571 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8572 return 0;
8573 /*
8574 * Check to make sure we don't have duplicate iov_base's in this
8575 * iovec, if so return EINVAL, otherwise we'll get csum errors
8576 * when reading back.
8577 */
8578 for (seg = 0; seg < iter->nr_segs; seg++) {
8579 for (i = seg + 1; i < iter->nr_segs; i++) {
8580 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8581 goto out;
8582 }
8583 }
8584 retval = 0;
8585 out:
8586 return retval;
8587 }
8588
8589 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8590 {
8591 struct file *file = iocb->ki_filp;
8592 struct inode *inode = file->f_mapping->host;
8593 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8594 struct btrfs_dio_data dio_data = { 0 };
8595 loff_t offset = iocb->ki_pos;
8596 size_t count = 0;
8597 int flags = 0;
8598 bool wakeup = true;
8599 bool relock = false;
8600 ssize_t ret;
8601
8602 if (check_direct_IO(fs_info, iocb, iter, offset))
8603 return 0;
8604
8605 inode_dio_begin(inode);
8606 smp_mb__after_atomic();
8607
8608 /*
8609 * The generic stuff only does filemap_write_and_wait_range, which
8610 * isn't enough if we've written compressed pages to this area, so
8611 * we need to flush the dirty pages again to make absolutely sure
8612 * that any outstanding dirty pages are on disk.
8613 */
8614 count = iov_iter_count(iter);
8615 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8616 &BTRFS_I(inode)->runtime_flags))
8617 filemap_fdatawrite_range(inode->i_mapping, offset,
8618 offset + count - 1);
8619
8620 if (iov_iter_rw(iter) == WRITE) {
8621 /*
8622 * If the write DIO is beyond the EOF, we need update
8623 * the isize, but it is protected by i_mutex. So we can
8624 * not unlock the i_mutex at this case.
8625 */
8626 if (offset + count <= inode->i_size) {
8627 dio_data.overwrite = 1;
8628 inode_unlock(inode);
8629 relock = true;
8630 }
8631 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8632 if (ret)
8633 goto out;
8634 dio_data.outstanding_extents = count_max_extents(count);
8635
8636 /*
8637 * We need to know how many extents we reserved so that we can
8638 * do the accounting properly if we go over the number we
8639 * originally calculated. Abuse current->journal_info for this.
8640 */
8641 dio_data.reserve = round_up(count,
8642 fs_info->sectorsize);
8643 dio_data.unsubmitted_oe_range_start = (u64)offset;
8644 dio_data.unsubmitted_oe_range_end = (u64)offset;
8645 current->journal_info = &dio_data;
8646 down_read(&BTRFS_I(inode)->dio_sem);
8647 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8648 &BTRFS_I(inode)->runtime_flags)) {
8649 inode_dio_end(inode);
8650 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8651 wakeup = false;
8652 }
8653
8654 ret = __blockdev_direct_IO(iocb, inode,
8655 fs_info->fs_devices->latest_bdev,
8656 iter, btrfs_get_blocks_direct, NULL,
8657 btrfs_submit_direct, flags);
8658 if (iov_iter_rw(iter) == WRITE) {
8659 up_read(&BTRFS_I(inode)->dio_sem);
8660 current->journal_info = NULL;
8661 if (ret < 0 && ret != -EIOCBQUEUED) {
8662 if (dio_data.reserve)
8663 btrfs_delalloc_release_space(inode, offset,
8664 dio_data.reserve);
8665 /*
8666 * On error we might have left some ordered extents
8667 * without submitting corresponding bios for them, so
8668 * cleanup them up to avoid other tasks getting them
8669 * and waiting for them to complete forever.
8670 */
8671 if (dio_data.unsubmitted_oe_range_start <
8672 dio_data.unsubmitted_oe_range_end)
8673 btrfs_endio_direct_write_update_ordered(inode,
8674 dio_data.unsubmitted_oe_range_start,
8675 dio_data.unsubmitted_oe_range_end -
8676 dio_data.unsubmitted_oe_range_start,
8677 0);
8678 } else if (ret >= 0 && (size_t)ret < count)
8679 btrfs_delalloc_release_space(inode, offset,
8680 count - (size_t)ret);
8681 }
8682 out:
8683 if (wakeup)
8684 inode_dio_end(inode);
8685 if (relock)
8686 inode_lock(inode);
8687
8688 return ret;
8689 }
8690
8691 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8692
8693 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8694 __u64 start, __u64 len)
8695 {
8696 int ret;
8697
8698 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8699 if (ret)
8700 return ret;
8701
8702 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8703 }
8704
8705 int btrfs_readpage(struct file *file, struct page *page)
8706 {
8707 struct extent_io_tree *tree;
8708 tree = &BTRFS_I(page->mapping->host)->io_tree;
8709 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8710 }
8711
8712 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8713 {
8714 struct extent_io_tree *tree;
8715 struct inode *inode = page->mapping->host;
8716 int ret;
8717
8718 if (current->flags & PF_MEMALLOC) {
8719 redirty_page_for_writepage(wbc, page);
8720 unlock_page(page);
8721 return 0;
8722 }
8723
8724 /*
8725 * If we are under memory pressure we will call this directly from the
8726 * VM, we need to make sure we have the inode referenced for the ordered
8727 * extent. If not just return like we didn't do anything.
8728 */
8729 if (!igrab(inode)) {
8730 redirty_page_for_writepage(wbc, page);
8731 return AOP_WRITEPAGE_ACTIVATE;
8732 }
8733 tree = &BTRFS_I(page->mapping->host)->io_tree;
8734 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8735 btrfs_add_delayed_iput(inode);
8736 return ret;
8737 }
8738
8739 static int btrfs_writepages(struct address_space *mapping,
8740 struct writeback_control *wbc)
8741 {
8742 struct extent_io_tree *tree;
8743
8744 tree = &BTRFS_I(mapping->host)->io_tree;
8745 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8746 }
8747
8748 static int
8749 btrfs_readpages(struct file *file, struct address_space *mapping,
8750 struct list_head *pages, unsigned nr_pages)
8751 {
8752 struct extent_io_tree *tree;
8753 tree = &BTRFS_I(mapping->host)->io_tree;
8754 return extent_readpages(tree, mapping, pages, nr_pages,
8755 btrfs_get_extent);
8756 }
8757 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8758 {
8759 struct extent_io_tree *tree;
8760 struct extent_map_tree *map;
8761 int ret;
8762
8763 tree = &BTRFS_I(page->mapping->host)->io_tree;
8764 map = &BTRFS_I(page->mapping->host)->extent_tree;
8765 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8766 if (ret == 1) {
8767 ClearPagePrivate(page);
8768 set_page_private(page, 0);
8769 put_page(page);
8770 }
8771 return ret;
8772 }
8773
8774 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8775 {
8776 if (PageWriteback(page) || PageDirty(page))
8777 return 0;
8778 return __btrfs_releasepage(page, gfp_flags);
8779 }
8780
8781 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8782 unsigned int length)
8783 {
8784 struct inode *inode = page->mapping->host;
8785 struct extent_io_tree *tree;
8786 struct btrfs_ordered_extent *ordered;
8787 struct extent_state *cached_state = NULL;
8788 u64 page_start = page_offset(page);
8789 u64 page_end = page_start + PAGE_SIZE - 1;
8790 u64 start;
8791 u64 end;
8792 int inode_evicting = inode->i_state & I_FREEING;
8793
8794 /*
8795 * we have the page locked, so new writeback can't start,
8796 * and the dirty bit won't be cleared while we are here.
8797 *
8798 * Wait for IO on this page so that we can safely clear
8799 * the PagePrivate2 bit and do ordered accounting
8800 */
8801 wait_on_page_writeback(page);
8802
8803 tree = &BTRFS_I(inode)->io_tree;
8804 if (offset) {
8805 btrfs_releasepage(page, GFP_NOFS);
8806 return;
8807 }
8808
8809 if (!inode_evicting)
8810 lock_extent_bits(tree, page_start, page_end, &cached_state);
8811 again:
8812 start = page_start;
8813 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8814 page_end - start + 1);
8815 if (ordered) {
8816 end = min(page_end, ordered->file_offset + ordered->len - 1);
8817 /*
8818 * IO on this page will never be started, so we need
8819 * to account for any ordered extents now
8820 */
8821 if (!inode_evicting)
8822 clear_extent_bit(tree, start, end,
8823 EXTENT_DIRTY | EXTENT_DELALLOC |
8824 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8825 EXTENT_DEFRAG, 1, 0, &cached_state,
8826 GFP_NOFS);
8827 /*
8828 * whoever cleared the private bit is responsible
8829 * for the finish_ordered_io
8830 */
8831 if (TestClearPagePrivate2(page)) {
8832 struct btrfs_ordered_inode_tree *tree;
8833 u64 new_len;
8834
8835 tree = &BTRFS_I(inode)->ordered_tree;
8836
8837 spin_lock_irq(&tree->lock);
8838 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8839 new_len = start - ordered->file_offset;
8840 if (new_len < ordered->truncated_len)
8841 ordered->truncated_len = new_len;
8842 spin_unlock_irq(&tree->lock);
8843
8844 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8845 start,
8846 end - start + 1, 1))
8847 btrfs_finish_ordered_io(ordered);
8848 }
8849 btrfs_put_ordered_extent(ordered);
8850 if (!inode_evicting) {
8851 cached_state = NULL;
8852 lock_extent_bits(tree, start, end,
8853 &cached_state);
8854 }
8855
8856 start = end + 1;
8857 if (start < page_end)
8858 goto again;
8859 }
8860
8861 /*
8862 * Qgroup reserved space handler
8863 * Page here will be either
8864 * 1) Already written to disk
8865 * In this case, its reserved space is released from data rsv map
8866 * and will be freed by delayed_ref handler finally.
8867 * So even we call qgroup_free_data(), it won't decrease reserved
8868 * space.
8869 * 2) Not written to disk
8870 * This means the reserved space should be freed here. However,
8871 * if a truncate invalidates the page (by clearing PageDirty)
8872 * and the page is accounted for while allocating extent
8873 * in btrfs_check_data_free_space() we let delayed_ref to
8874 * free the entire extent.
8875 */
8876 if (PageDirty(page))
8877 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
8878 if (!inode_evicting) {
8879 clear_extent_bit(tree, page_start, page_end,
8880 EXTENT_LOCKED | EXTENT_DIRTY |
8881 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8882 EXTENT_DEFRAG, 1, 1,
8883 &cached_state, GFP_NOFS);
8884
8885 __btrfs_releasepage(page, GFP_NOFS);
8886 }
8887
8888 ClearPageChecked(page);
8889 if (PagePrivate(page)) {
8890 ClearPagePrivate(page);
8891 set_page_private(page, 0);
8892 put_page(page);
8893 }
8894 }
8895
8896 /*
8897 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8898 * called from a page fault handler when a page is first dirtied. Hence we must
8899 * be careful to check for EOF conditions here. We set the page up correctly
8900 * for a written page which means we get ENOSPC checking when writing into
8901 * holes and correct delalloc and unwritten extent mapping on filesystems that
8902 * support these features.
8903 *
8904 * We are not allowed to take the i_mutex here so we have to play games to
8905 * protect against truncate races as the page could now be beyond EOF. Because
8906 * vmtruncate() writes the inode size before removing pages, once we have the
8907 * page lock we can determine safely if the page is beyond EOF. If it is not
8908 * beyond EOF, then the page is guaranteed safe against truncation until we
8909 * unlock the page.
8910 */
8911 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8912 {
8913 struct page *page = vmf->page;
8914 struct inode *inode = file_inode(vma->vm_file);
8915 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8916 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8917 struct btrfs_ordered_extent *ordered;
8918 struct extent_state *cached_state = NULL;
8919 char *kaddr;
8920 unsigned long zero_start;
8921 loff_t size;
8922 int ret;
8923 int reserved = 0;
8924 u64 reserved_space;
8925 u64 page_start;
8926 u64 page_end;
8927 u64 end;
8928
8929 reserved_space = PAGE_SIZE;
8930
8931 sb_start_pagefault(inode->i_sb);
8932 page_start = page_offset(page);
8933 page_end = page_start + PAGE_SIZE - 1;
8934 end = page_end;
8935
8936 /*
8937 * Reserving delalloc space after obtaining the page lock can lead to
8938 * deadlock. For example, if a dirty page is locked by this function
8939 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8940 * dirty page write out, then the btrfs_writepage() function could
8941 * end up waiting indefinitely to get a lock on the page currently
8942 * being processed by btrfs_page_mkwrite() function.
8943 */
8944 ret = btrfs_delalloc_reserve_space(inode, page_start,
8945 reserved_space);
8946 if (!ret) {
8947 ret = file_update_time(vma->vm_file);
8948 reserved = 1;
8949 }
8950 if (ret) {
8951 if (ret == -ENOMEM)
8952 ret = VM_FAULT_OOM;
8953 else /* -ENOSPC, -EIO, etc */
8954 ret = VM_FAULT_SIGBUS;
8955 if (reserved)
8956 goto out;
8957 goto out_noreserve;
8958 }
8959
8960 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8961 again:
8962 lock_page(page);
8963 size = i_size_read(inode);
8964
8965 if ((page->mapping != inode->i_mapping) ||
8966 (page_start >= size)) {
8967 /* page got truncated out from underneath us */
8968 goto out_unlock;
8969 }
8970 wait_on_page_writeback(page);
8971
8972 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8973 set_page_extent_mapped(page);
8974
8975 /*
8976 * we can't set the delalloc bits if there are pending ordered
8977 * extents. Drop our locks and wait for them to finish
8978 */
8979 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8980 PAGE_SIZE);
8981 if (ordered) {
8982 unlock_extent_cached(io_tree, page_start, page_end,
8983 &cached_state, GFP_NOFS);
8984 unlock_page(page);
8985 btrfs_start_ordered_extent(inode, ordered, 1);
8986 btrfs_put_ordered_extent(ordered);
8987 goto again;
8988 }
8989
8990 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8991 reserved_space = round_up(size - page_start,
8992 fs_info->sectorsize);
8993 if (reserved_space < PAGE_SIZE) {
8994 end = page_start + reserved_space - 1;
8995 spin_lock(&BTRFS_I(inode)->lock);
8996 BTRFS_I(inode)->outstanding_extents++;
8997 spin_unlock(&BTRFS_I(inode)->lock);
8998 btrfs_delalloc_release_space(inode, page_start,
8999 PAGE_SIZE - reserved_space);
9000 }
9001 }
9002
9003 /*
9004 * page_mkwrite gets called when the page is firstly dirtied after it's
9005 * faulted in, but write(2) could also dirty a page and set delalloc
9006 * bits, thus in this case for space account reason, we still need to
9007 * clear any delalloc bits within this page range since we have to
9008 * reserve data&meta space before lock_page() (see above comments).
9009 */
9010 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9011 EXTENT_DIRTY | EXTENT_DELALLOC |
9012 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9013 0, 0, &cached_state, GFP_NOFS);
9014
9015 ret = btrfs_set_extent_delalloc(inode, page_start, end,
9016 &cached_state, 0);
9017 if (ret) {
9018 unlock_extent_cached(io_tree, page_start, page_end,
9019 &cached_state, GFP_NOFS);
9020 ret = VM_FAULT_SIGBUS;
9021 goto out_unlock;
9022 }
9023 ret = 0;
9024
9025 /* page is wholly or partially inside EOF */
9026 if (page_start + PAGE_SIZE > size)
9027 zero_start = size & ~PAGE_MASK;
9028 else
9029 zero_start = PAGE_SIZE;
9030
9031 if (zero_start != PAGE_SIZE) {
9032 kaddr = kmap(page);
9033 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9034 flush_dcache_page(page);
9035 kunmap(page);
9036 }
9037 ClearPageChecked(page);
9038 set_page_dirty(page);
9039 SetPageUptodate(page);
9040
9041 BTRFS_I(inode)->last_trans = fs_info->generation;
9042 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9043 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9044
9045 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9046
9047 out_unlock:
9048 if (!ret) {
9049 sb_end_pagefault(inode->i_sb);
9050 return VM_FAULT_LOCKED;
9051 }
9052 unlock_page(page);
9053 out:
9054 btrfs_delalloc_release_space(inode, page_start, reserved_space);
9055 out_noreserve:
9056 sb_end_pagefault(inode->i_sb);
9057 return ret;
9058 }
9059
9060 static int btrfs_truncate(struct inode *inode)
9061 {
9062 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9063 struct btrfs_root *root = BTRFS_I(inode)->root;
9064 struct btrfs_block_rsv *rsv;
9065 int ret = 0;
9066 int err = 0;
9067 struct btrfs_trans_handle *trans;
9068 u64 mask = fs_info->sectorsize - 1;
9069 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9070
9071 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9072 (u64)-1);
9073 if (ret)
9074 return ret;
9075
9076 /*
9077 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9078 * 3 things going on here
9079 *
9080 * 1) We need to reserve space for our orphan item and the space to
9081 * delete our orphan item. Lord knows we don't want to have a dangling
9082 * orphan item because we didn't reserve space to remove it.
9083 *
9084 * 2) We need to reserve space to update our inode.
9085 *
9086 * 3) We need to have something to cache all the space that is going to
9087 * be free'd up by the truncate operation, but also have some slack
9088 * space reserved in case it uses space during the truncate (thank you
9089 * very much snapshotting).
9090 *
9091 * And we need these to all be separate. The fact is we can use a lot of
9092 * space doing the truncate, and we have no earthly idea how much space
9093 * we will use, so we need the truncate reservation to be separate so it
9094 * doesn't end up using space reserved for updating the inode or
9095 * removing the orphan item. We also need to be able to stop the
9096 * transaction and start a new one, which means we need to be able to
9097 * update the inode several times, and we have no idea of knowing how
9098 * many times that will be, so we can't just reserve 1 item for the
9099 * entirety of the operation, so that has to be done separately as well.
9100 * Then there is the orphan item, which does indeed need to be held on
9101 * to for the whole operation, and we need nobody to touch this reserved
9102 * space except the orphan code.
9103 *
9104 * So that leaves us with
9105 *
9106 * 1) root->orphan_block_rsv - for the orphan deletion.
9107 * 2) rsv - for the truncate reservation, which we will steal from the
9108 * transaction reservation.
9109 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9110 * updating the inode.
9111 */
9112 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9113 if (!rsv)
9114 return -ENOMEM;
9115 rsv->size = min_size;
9116 rsv->failfast = 1;
9117
9118 /*
9119 * 1 for the truncate slack space
9120 * 1 for updating the inode.
9121 */
9122 trans = btrfs_start_transaction(root, 2);
9123 if (IS_ERR(trans)) {
9124 err = PTR_ERR(trans);
9125 goto out;
9126 }
9127
9128 /* Migrate the slack space for the truncate to our reserve */
9129 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9130 min_size, 0);
9131 BUG_ON(ret);
9132
9133 /*
9134 * So if we truncate and then write and fsync we normally would just
9135 * write the extents that changed, which is a problem if we need to
9136 * first truncate that entire inode. So set this flag so we write out
9137 * all of the extents in the inode to the sync log so we're completely
9138 * safe.
9139 */
9140 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9141 trans->block_rsv = rsv;
9142
9143 while (1) {
9144 ret = btrfs_truncate_inode_items(trans, root, inode,
9145 inode->i_size,
9146 BTRFS_EXTENT_DATA_KEY);
9147 if (ret != -ENOSPC && ret != -EAGAIN) {
9148 err = ret;
9149 break;
9150 }
9151
9152 trans->block_rsv = &fs_info->trans_block_rsv;
9153 ret = btrfs_update_inode(trans, root, inode);
9154 if (ret) {
9155 err = ret;
9156 break;
9157 }
9158
9159 btrfs_end_transaction(trans);
9160 btrfs_btree_balance_dirty(fs_info);
9161
9162 trans = btrfs_start_transaction(root, 2);
9163 if (IS_ERR(trans)) {
9164 ret = err = PTR_ERR(trans);
9165 trans = NULL;
9166 break;
9167 }
9168
9169 btrfs_block_rsv_release(fs_info, rsv, -1);
9170 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9171 rsv, min_size, 0);
9172 BUG_ON(ret); /* shouldn't happen */
9173 trans->block_rsv = rsv;
9174 }
9175
9176 if (ret == 0 && inode->i_nlink > 0) {
9177 trans->block_rsv = root->orphan_block_rsv;
9178 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9179 if (ret)
9180 err = ret;
9181 }
9182
9183 if (trans) {
9184 trans->block_rsv = &fs_info->trans_block_rsv;
9185 ret = btrfs_update_inode(trans, root, inode);
9186 if (ret && !err)
9187 err = ret;
9188
9189 ret = btrfs_end_transaction(trans);
9190 btrfs_btree_balance_dirty(fs_info);
9191 }
9192 out:
9193 btrfs_free_block_rsv(fs_info, rsv);
9194
9195 if (ret && !err)
9196 err = ret;
9197
9198 return err;
9199 }
9200
9201 /*
9202 * create a new subvolume directory/inode (helper for the ioctl).
9203 */
9204 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9205 struct btrfs_root *new_root,
9206 struct btrfs_root *parent_root,
9207 u64 new_dirid)
9208 {
9209 struct inode *inode;
9210 int err;
9211 u64 index = 0;
9212
9213 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9214 new_dirid, new_dirid,
9215 S_IFDIR | (~current_umask() & S_IRWXUGO),
9216 &index);
9217 if (IS_ERR(inode))
9218 return PTR_ERR(inode);
9219 inode->i_op = &btrfs_dir_inode_operations;
9220 inode->i_fop = &btrfs_dir_file_operations;
9221
9222 set_nlink(inode, 1);
9223 btrfs_i_size_write(BTRFS_I(inode), 0);
9224 unlock_new_inode(inode);
9225
9226 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9227 if (err)
9228 btrfs_err(new_root->fs_info,
9229 "error inheriting subvolume %llu properties: %d",
9230 new_root->root_key.objectid, err);
9231
9232 err = btrfs_update_inode(trans, new_root, inode);
9233
9234 iput(inode);
9235 return err;
9236 }
9237
9238 struct inode *btrfs_alloc_inode(struct super_block *sb)
9239 {
9240 struct btrfs_inode *ei;
9241 struct inode *inode;
9242
9243 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9244 if (!ei)
9245 return NULL;
9246
9247 ei->root = NULL;
9248 ei->generation = 0;
9249 ei->last_trans = 0;
9250 ei->last_sub_trans = 0;
9251 ei->logged_trans = 0;
9252 ei->delalloc_bytes = 0;
9253 ei->defrag_bytes = 0;
9254 ei->disk_i_size = 0;
9255 ei->flags = 0;
9256 ei->csum_bytes = 0;
9257 ei->index_cnt = (u64)-1;
9258 ei->dir_index = 0;
9259 ei->last_unlink_trans = 0;
9260 ei->last_log_commit = 0;
9261 ei->delayed_iput_count = 0;
9262
9263 spin_lock_init(&ei->lock);
9264 ei->outstanding_extents = 0;
9265 ei->reserved_extents = 0;
9266
9267 ei->runtime_flags = 0;
9268 ei->force_compress = BTRFS_COMPRESS_NONE;
9269
9270 ei->delayed_node = NULL;
9271
9272 ei->i_otime.tv_sec = 0;
9273 ei->i_otime.tv_nsec = 0;
9274
9275 inode = &ei->vfs_inode;
9276 extent_map_tree_init(&ei->extent_tree);
9277 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9278 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9279 ei->io_tree.track_uptodate = 1;
9280 ei->io_failure_tree.track_uptodate = 1;
9281 atomic_set(&ei->sync_writers, 0);
9282 mutex_init(&ei->log_mutex);
9283 mutex_init(&ei->delalloc_mutex);
9284 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9285 INIT_LIST_HEAD(&ei->delalloc_inodes);
9286 INIT_LIST_HEAD(&ei->delayed_iput);
9287 RB_CLEAR_NODE(&ei->rb_node);
9288 init_rwsem(&ei->dio_sem);
9289
9290 return inode;
9291 }
9292
9293 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9294 void btrfs_test_destroy_inode(struct inode *inode)
9295 {
9296 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9297 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9298 }
9299 #endif
9300
9301 static void btrfs_i_callback(struct rcu_head *head)
9302 {
9303 struct inode *inode = container_of(head, struct inode, i_rcu);
9304 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9305 }
9306
9307 void btrfs_destroy_inode(struct inode *inode)
9308 {
9309 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9310 struct btrfs_ordered_extent *ordered;
9311 struct btrfs_root *root = BTRFS_I(inode)->root;
9312
9313 WARN_ON(!hlist_empty(&inode->i_dentry));
9314 WARN_ON(inode->i_data.nrpages);
9315 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9316 WARN_ON(BTRFS_I(inode)->reserved_extents);
9317 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9318 WARN_ON(BTRFS_I(inode)->csum_bytes);
9319 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9320
9321 /*
9322 * This can happen where we create an inode, but somebody else also
9323 * created the same inode and we need to destroy the one we already
9324 * created.
9325 */
9326 if (!root)
9327 goto free;
9328
9329 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9330 &BTRFS_I(inode)->runtime_flags)) {
9331 btrfs_info(fs_info, "inode %llu still on the orphan list",
9332 btrfs_ino(BTRFS_I(inode)));
9333 atomic_dec(&root->orphan_inodes);
9334 }
9335
9336 while (1) {
9337 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9338 if (!ordered)
9339 break;
9340 else {
9341 btrfs_err(fs_info,
9342 "found ordered extent %llu %llu on inode cleanup",
9343 ordered->file_offset, ordered->len);
9344 btrfs_remove_ordered_extent(inode, ordered);
9345 btrfs_put_ordered_extent(ordered);
9346 btrfs_put_ordered_extent(ordered);
9347 }
9348 }
9349 btrfs_qgroup_check_reserved_leak(inode);
9350 inode_tree_del(inode);
9351 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9352 free:
9353 call_rcu(&inode->i_rcu, btrfs_i_callback);
9354 }
9355
9356 int btrfs_drop_inode(struct inode *inode)
9357 {
9358 struct btrfs_root *root = BTRFS_I(inode)->root;
9359
9360 if (root == NULL)
9361 return 1;
9362
9363 /* the snap/subvol tree is on deleting */
9364 if (btrfs_root_refs(&root->root_item) == 0)
9365 return 1;
9366 else
9367 return generic_drop_inode(inode);
9368 }
9369
9370 static void init_once(void *foo)
9371 {
9372 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9373
9374 inode_init_once(&ei->vfs_inode);
9375 }
9376
9377 void btrfs_destroy_cachep(void)
9378 {
9379 /*
9380 * Make sure all delayed rcu free inodes are flushed before we
9381 * destroy cache.
9382 */
9383 rcu_barrier();
9384 kmem_cache_destroy(btrfs_inode_cachep);
9385 kmem_cache_destroy(btrfs_trans_handle_cachep);
9386 kmem_cache_destroy(btrfs_transaction_cachep);
9387 kmem_cache_destroy(btrfs_path_cachep);
9388 kmem_cache_destroy(btrfs_free_space_cachep);
9389 }
9390
9391 int btrfs_init_cachep(void)
9392 {
9393 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9394 sizeof(struct btrfs_inode), 0,
9395 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9396 init_once);
9397 if (!btrfs_inode_cachep)
9398 goto fail;
9399
9400 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9401 sizeof(struct btrfs_trans_handle), 0,
9402 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9403 if (!btrfs_trans_handle_cachep)
9404 goto fail;
9405
9406 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9407 sizeof(struct btrfs_transaction), 0,
9408 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9409 if (!btrfs_transaction_cachep)
9410 goto fail;
9411
9412 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9413 sizeof(struct btrfs_path), 0,
9414 SLAB_MEM_SPREAD, NULL);
9415 if (!btrfs_path_cachep)
9416 goto fail;
9417
9418 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9419 sizeof(struct btrfs_free_space), 0,
9420 SLAB_MEM_SPREAD, NULL);
9421 if (!btrfs_free_space_cachep)
9422 goto fail;
9423
9424 return 0;
9425 fail:
9426 btrfs_destroy_cachep();
9427 return -ENOMEM;
9428 }
9429
9430 static int btrfs_getattr(struct vfsmount *mnt,
9431 struct dentry *dentry, struct kstat *stat)
9432 {
9433 u64 delalloc_bytes;
9434 struct inode *inode = d_inode(dentry);
9435 u32 blocksize = inode->i_sb->s_blocksize;
9436
9437 generic_fillattr(inode, stat);
9438 stat->dev = BTRFS_I(inode)->root->anon_dev;
9439
9440 spin_lock(&BTRFS_I(inode)->lock);
9441 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9442 spin_unlock(&BTRFS_I(inode)->lock);
9443 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9444 ALIGN(delalloc_bytes, blocksize)) >> 9;
9445 return 0;
9446 }
9447
9448 static int btrfs_rename_exchange(struct inode *old_dir,
9449 struct dentry *old_dentry,
9450 struct inode *new_dir,
9451 struct dentry *new_dentry)
9452 {
9453 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9454 struct btrfs_trans_handle *trans;
9455 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9456 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9457 struct inode *new_inode = new_dentry->d_inode;
9458 struct inode *old_inode = old_dentry->d_inode;
9459 struct timespec ctime = current_time(old_inode);
9460 struct dentry *parent;
9461 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9462 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9463 u64 old_idx = 0;
9464 u64 new_idx = 0;
9465 u64 root_objectid;
9466 int ret;
9467 bool root_log_pinned = false;
9468 bool dest_log_pinned = false;
9469
9470 /* we only allow rename subvolume link between subvolumes */
9471 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9472 return -EXDEV;
9473
9474 /* close the race window with snapshot create/destroy ioctl */
9475 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9476 down_read(&fs_info->subvol_sem);
9477 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9478 down_read(&fs_info->subvol_sem);
9479
9480 /*
9481 * We want to reserve the absolute worst case amount of items. So if
9482 * both inodes are subvols and we need to unlink them then that would
9483 * require 4 item modifications, but if they are both normal inodes it
9484 * would require 5 item modifications, so we'll assume their normal
9485 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9486 * should cover the worst case number of items we'll modify.
9487 */
9488 trans = btrfs_start_transaction(root, 12);
9489 if (IS_ERR(trans)) {
9490 ret = PTR_ERR(trans);
9491 goto out_notrans;
9492 }
9493
9494 /*
9495 * We need to find a free sequence number both in the source and
9496 * in the destination directory for the exchange.
9497 */
9498 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9499 if (ret)
9500 goto out_fail;
9501 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9502 if (ret)
9503 goto out_fail;
9504
9505 BTRFS_I(old_inode)->dir_index = 0ULL;
9506 BTRFS_I(new_inode)->dir_index = 0ULL;
9507
9508 /* Reference for the source. */
9509 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9510 /* force full log commit if subvolume involved. */
9511 btrfs_set_log_full_commit(fs_info, trans);
9512 } else {
9513 btrfs_pin_log_trans(root);
9514 root_log_pinned = true;
9515 ret = btrfs_insert_inode_ref(trans, dest,
9516 new_dentry->d_name.name,
9517 new_dentry->d_name.len,
9518 old_ino,
9519 btrfs_ino(BTRFS_I(new_dir)),
9520 old_idx);
9521 if (ret)
9522 goto out_fail;
9523 }
9524
9525 /* And now for the dest. */
9526 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9527 /* force full log commit if subvolume involved. */
9528 btrfs_set_log_full_commit(fs_info, trans);
9529 } else {
9530 btrfs_pin_log_trans(dest);
9531 dest_log_pinned = true;
9532 ret = btrfs_insert_inode_ref(trans, root,
9533 old_dentry->d_name.name,
9534 old_dentry->d_name.len,
9535 new_ino,
9536 btrfs_ino(BTRFS_I(old_dir)),
9537 new_idx);
9538 if (ret)
9539 goto out_fail;
9540 }
9541
9542 /* Update inode version and ctime/mtime. */
9543 inode_inc_iversion(old_dir);
9544 inode_inc_iversion(new_dir);
9545 inode_inc_iversion(old_inode);
9546 inode_inc_iversion(new_inode);
9547 old_dir->i_ctime = old_dir->i_mtime = ctime;
9548 new_dir->i_ctime = new_dir->i_mtime = ctime;
9549 old_inode->i_ctime = ctime;
9550 new_inode->i_ctime = ctime;
9551
9552 if (old_dentry->d_parent != new_dentry->d_parent) {
9553 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9554 BTRFS_I(old_inode), 1);
9555 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9556 BTRFS_I(new_inode), 1);
9557 }
9558
9559 /* src is a subvolume */
9560 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9561 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9562 ret = btrfs_unlink_subvol(trans, root, old_dir,
9563 root_objectid,
9564 old_dentry->d_name.name,
9565 old_dentry->d_name.len);
9566 } else { /* src is an inode */
9567 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9568 BTRFS_I(old_dentry->d_inode),
9569 old_dentry->d_name.name,
9570 old_dentry->d_name.len);
9571 if (!ret)
9572 ret = btrfs_update_inode(trans, root, old_inode);
9573 }
9574 if (ret) {
9575 btrfs_abort_transaction(trans, ret);
9576 goto out_fail;
9577 }
9578
9579 /* dest is a subvolume */
9580 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9581 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9582 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9583 root_objectid,
9584 new_dentry->d_name.name,
9585 new_dentry->d_name.len);
9586 } else { /* dest is an inode */
9587 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9588 BTRFS_I(new_dentry->d_inode),
9589 new_dentry->d_name.name,
9590 new_dentry->d_name.len);
9591 if (!ret)
9592 ret = btrfs_update_inode(trans, dest, new_inode);
9593 }
9594 if (ret) {
9595 btrfs_abort_transaction(trans, ret);
9596 goto out_fail;
9597 }
9598
9599 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9600 new_dentry->d_name.name,
9601 new_dentry->d_name.len, 0, old_idx);
9602 if (ret) {
9603 btrfs_abort_transaction(trans, ret);
9604 goto out_fail;
9605 }
9606
9607 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9608 old_dentry->d_name.name,
9609 old_dentry->d_name.len, 0, new_idx);
9610 if (ret) {
9611 btrfs_abort_transaction(trans, ret);
9612 goto out_fail;
9613 }
9614
9615 if (old_inode->i_nlink == 1)
9616 BTRFS_I(old_inode)->dir_index = old_idx;
9617 if (new_inode->i_nlink == 1)
9618 BTRFS_I(new_inode)->dir_index = new_idx;
9619
9620 if (root_log_pinned) {
9621 parent = new_dentry->d_parent;
9622 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9623 parent);
9624 btrfs_end_log_trans(root);
9625 root_log_pinned = false;
9626 }
9627 if (dest_log_pinned) {
9628 parent = old_dentry->d_parent;
9629 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9630 parent);
9631 btrfs_end_log_trans(dest);
9632 dest_log_pinned = false;
9633 }
9634 out_fail:
9635 /*
9636 * If we have pinned a log and an error happened, we unpin tasks
9637 * trying to sync the log and force them to fallback to a transaction
9638 * commit if the log currently contains any of the inodes involved in
9639 * this rename operation (to ensure we do not persist a log with an
9640 * inconsistent state for any of these inodes or leading to any
9641 * inconsistencies when replayed). If the transaction was aborted, the
9642 * abortion reason is propagated to userspace when attempting to commit
9643 * the transaction. If the log does not contain any of these inodes, we
9644 * allow the tasks to sync it.
9645 */
9646 if (ret && (root_log_pinned || dest_log_pinned)) {
9647 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9648 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9649 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9650 (new_inode &&
9651 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9652 btrfs_set_log_full_commit(fs_info, trans);
9653
9654 if (root_log_pinned) {
9655 btrfs_end_log_trans(root);
9656 root_log_pinned = false;
9657 }
9658 if (dest_log_pinned) {
9659 btrfs_end_log_trans(dest);
9660 dest_log_pinned = false;
9661 }
9662 }
9663 ret = btrfs_end_transaction(trans);
9664 out_notrans:
9665 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9666 up_read(&fs_info->subvol_sem);
9667 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9668 up_read(&fs_info->subvol_sem);
9669
9670 return ret;
9671 }
9672
9673 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9674 struct btrfs_root *root,
9675 struct inode *dir,
9676 struct dentry *dentry)
9677 {
9678 int ret;
9679 struct inode *inode;
9680 u64 objectid;
9681 u64 index;
9682
9683 ret = btrfs_find_free_ino(root, &objectid);
9684 if (ret)
9685 return ret;
9686
9687 inode = btrfs_new_inode(trans, root, dir,
9688 dentry->d_name.name,
9689 dentry->d_name.len,
9690 btrfs_ino(BTRFS_I(dir)),
9691 objectid,
9692 S_IFCHR | WHITEOUT_MODE,
9693 &index);
9694
9695 if (IS_ERR(inode)) {
9696 ret = PTR_ERR(inode);
9697 return ret;
9698 }
9699
9700 inode->i_op = &btrfs_special_inode_operations;
9701 init_special_inode(inode, inode->i_mode,
9702 WHITEOUT_DEV);
9703
9704 ret = btrfs_init_inode_security(trans, inode, dir,
9705 &dentry->d_name);
9706 if (ret)
9707 goto out;
9708
9709 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9710 BTRFS_I(inode), 0, index);
9711 if (ret)
9712 goto out;
9713
9714 ret = btrfs_update_inode(trans, root, inode);
9715 out:
9716 unlock_new_inode(inode);
9717 if (ret)
9718 inode_dec_link_count(inode);
9719 iput(inode);
9720
9721 return ret;
9722 }
9723
9724 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9725 struct inode *new_dir, struct dentry *new_dentry,
9726 unsigned int flags)
9727 {
9728 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9729 struct btrfs_trans_handle *trans;
9730 unsigned int trans_num_items;
9731 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9732 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9733 struct inode *new_inode = d_inode(new_dentry);
9734 struct inode *old_inode = d_inode(old_dentry);
9735 u64 index = 0;
9736 u64 root_objectid;
9737 int ret;
9738 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9739 bool log_pinned = false;
9740
9741 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9742 return -EPERM;
9743
9744 /* we only allow rename subvolume link between subvolumes */
9745 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9746 return -EXDEV;
9747
9748 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9749 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9750 return -ENOTEMPTY;
9751
9752 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9753 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9754 return -ENOTEMPTY;
9755
9756
9757 /* check for collisions, even if the name isn't there */
9758 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9759 new_dentry->d_name.name,
9760 new_dentry->d_name.len);
9761
9762 if (ret) {
9763 if (ret == -EEXIST) {
9764 /* we shouldn't get
9765 * eexist without a new_inode */
9766 if (WARN_ON(!new_inode)) {
9767 return ret;
9768 }
9769 } else {
9770 /* maybe -EOVERFLOW */
9771 return ret;
9772 }
9773 }
9774 ret = 0;
9775
9776 /*
9777 * we're using rename to replace one file with another. Start IO on it
9778 * now so we don't add too much work to the end of the transaction
9779 */
9780 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9781 filemap_flush(old_inode->i_mapping);
9782
9783 /* close the racy window with snapshot create/destroy ioctl */
9784 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9785 down_read(&fs_info->subvol_sem);
9786 /*
9787 * We want to reserve the absolute worst case amount of items. So if
9788 * both inodes are subvols and we need to unlink them then that would
9789 * require 4 item modifications, but if they are both normal inodes it
9790 * would require 5 item modifications, so we'll assume they are normal
9791 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9792 * should cover the worst case number of items we'll modify.
9793 * If our rename has the whiteout flag, we need more 5 units for the
9794 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9795 * when selinux is enabled).
9796 */
9797 trans_num_items = 11;
9798 if (flags & RENAME_WHITEOUT)
9799 trans_num_items += 5;
9800 trans = btrfs_start_transaction(root, trans_num_items);
9801 if (IS_ERR(trans)) {
9802 ret = PTR_ERR(trans);
9803 goto out_notrans;
9804 }
9805
9806 if (dest != root)
9807 btrfs_record_root_in_trans(trans, dest);
9808
9809 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9810 if (ret)
9811 goto out_fail;
9812
9813 BTRFS_I(old_inode)->dir_index = 0ULL;
9814 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9815 /* force full log commit if subvolume involved. */
9816 btrfs_set_log_full_commit(fs_info, trans);
9817 } else {
9818 btrfs_pin_log_trans(root);
9819 log_pinned = true;
9820 ret = btrfs_insert_inode_ref(trans, dest,
9821 new_dentry->d_name.name,
9822 new_dentry->d_name.len,
9823 old_ino,
9824 btrfs_ino(BTRFS_I(new_dir)), index);
9825 if (ret)
9826 goto out_fail;
9827 }
9828
9829 inode_inc_iversion(old_dir);
9830 inode_inc_iversion(new_dir);
9831 inode_inc_iversion(old_inode);
9832 old_dir->i_ctime = old_dir->i_mtime =
9833 new_dir->i_ctime = new_dir->i_mtime =
9834 old_inode->i_ctime = current_time(old_dir);
9835
9836 if (old_dentry->d_parent != new_dentry->d_parent)
9837 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9838 BTRFS_I(old_inode), 1);
9839
9840 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9841 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9842 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9843 old_dentry->d_name.name,
9844 old_dentry->d_name.len);
9845 } else {
9846 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9847 BTRFS_I(d_inode(old_dentry)),
9848 old_dentry->d_name.name,
9849 old_dentry->d_name.len);
9850 if (!ret)
9851 ret = btrfs_update_inode(trans, root, old_inode);
9852 }
9853 if (ret) {
9854 btrfs_abort_transaction(trans, ret);
9855 goto out_fail;
9856 }
9857
9858 if (new_inode) {
9859 inode_inc_iversion(new_inode);
9860 new_inode->i_ctime = current_time(new_inode);
9861 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9862 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9863 root_objectid = BTRFS_I(new_inode)->location.objectid;
9864 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9865 root_objectid,
9866 new_dentry->d_name.name,
9867 new_dentry->d_name.len);
9868 BUG_ON(new_inode->i_nlink == 0);
9869 } else {
9870 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9871 BTRFS_I(d_inode(new_dentry)),
9872 new_dentry->d_name.name,
9873 new_dentry->d_name.len);
9874 }
9875 if (!ret && new_inode->i_nlink == 0)
9876 ret = btrfs_orphan_add(trans,
9877 BTRFS_I(d_inode(new_dentry)));
9878 if (ret) {
9879 btrfs_abort_transaction(trans, ret);
9880 goto out_fail;
9881 }
9882 }
9883
9884 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9885 new_dentry->d_name.name,
9886 new_dentry->d_name.len, 0, index);
9887 if (ret) {
9888 btrfs_abort_transaction(trans, ret);
9889 goto out_fail;
9890 }
9891
9892 if (old_inode->i_nlink == 1)
9893 BTRFS_I(old_inode)->dir_index = index;
9894
9895 if (log_pinned) {
9896 struct dentry *parent = new_dentry->d_parent;
9897
9898 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9899 parent);
9900 btrfs_end_log_trans(root);
9901 log_pinned = false;
9902 }
9903
9904 if (flags & RENAME_WHITEOUT) {
9905 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9906 old_dentry);
9907
9908 if (ret) {
9909 btrfs_abort_transaction(trans, ret);
9910 goto out_fail;
9911 }
9912 }
9913 out_fail:
9914 /*
9915 * If we have pinned the log and an error happened, we unpin tasks
9916 * trying to sync the log and force them to fallback to a transaction
9917 * commit if the log currently contains any of the inodes involved in
9918 * this rename operation (to ensure we do not persist a log with an
9919 * inconsistent state for any of these inodes or leading to any
9920 * inconsistencies when replayed). If the transaction was aborted, the
9921 * abortion reason is propagated to userspace when attempting to commit
9922 * the transaction. If the log does not contain any of these inodes, we
9923 * allow the tasks to sync it.
9924 */
9925 if (ret && log_pinned) {
9926 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9927 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9928 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9929 (new_inode &&
9930 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9931 btrfs_set_log_full_commit(fs_info, trans);
9932
9933 btrfs_end_log_trans(root);
9934 log_pinned = false;
9935 }
9936 btrfs_end_transaction(trans);
9937 out_notrans:
9938 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9939 up_read(&fs_info->subvol_sem);
9940
9941 return ret;
9942 }
9943
9944 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9945 struct inode *new_dir, struct dentry *new_dentry,
9946 unsigned int flags)
9947 {
9948 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9949 return -EINVAL;
9950
9951 if (flags & RENAME_EXCHANGE)
9952 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9953 new_dentry);
9954
9955 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9956 }
9957
9958 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9959 {
9960 struct btrfs_delalloc_work *delalloc_work;
9961 struct inode *inode;
9962
9963 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9964 work);
9965 inode = delalloc_work->inode;
9966 filemap_flush(inode->i_mapping);
9967 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9968 &BTRFS_I(inode)->runtime_flags))
9969 filemap_flush(inode->i_mapping);
9970
9971 if (delalloc_work->delay_iput)
9972 btrfs_add_delayed_iput(inode);
9973 else
9974 iput(inode);
9975 complete(&delalloc_work->completion);
9976 }
9977
9978 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9979 int delay_iput)
9980 {
9981 struct btrfs_delalloc_work *work;
9982
9983 work = kmalloc(sizeof(*work), GFP_NOFS);
9984 if (!work)
9985 return NULL;
9986
9987 init_completion(&work->completion);
9988 INIT_LIST_HEAD(&work->list);
9989 work->inode = inode;
9990 work->delay_iput = delay_iput;
9991 WARN_ON_ONCE(!inode);
9992 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9993 btrfs_run_delalloc_work, NULL, NULL);
9994
9995 return work;
9996 }
9997
9998 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9999 {
10000 wait_for_completion(&work->completion);
10001 kfree(work);
10002 }
10003
10004 /*
10005 * some fairly slow code that needs optimization. This walks the list
10006 * of all the inodes with pending delalloc and forces them to disk.
10007 */
10008 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10009 int nr)
10010 {
10011 struct btrfs_inode *binode;
10012 struct inode *inode;
10013 struct btrfs_delalloc_work *work, *next;
10014 struct list_head works;
10015 struct list_head splice;
10016 int ret = 0;
10017
10018 INIT_LIST_HEAD(&works);
10019 INIT_LIST_HEAD(&splice);
10020
10021 mutex_lock(&root->delalloc_mutex);
10022 spin_lock(&root->delalloc_lock);
10023 list_splice_init(&root->delalloc_inodes, &splice);
10024 while (!list_empty(&splice)) {
10025 binode = list_entry(splice.next, struct btrfs_inode,
10026 delalloc_inodes);
10027
10028 list_move_tail(&binode->delalloc_inodes,
10029 &root->delalloc_inodes);
10030 inode = igrab(&binode->vfs_inode);
10031 if (!inode) {
10032 cond_resched_lock(&root->delalloc_lock);
10033 continue;
10034 }
10035 spin_unlock(&root->delalloc_lock);
10036
10037 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10038 if (!work) {
10039 if (delay_iput)
10040 btrfs_add_delayed_iput(inode);
10041 else
10042 iput(inode);
10043 ret = -ENOMEM;
10044 goto out;
10045 }
10046 list_add_tail(&work->list, &works);
10047 btrfs_queue_work(root->fs_info->flush_workers,
10048 &work->work);
10049 ret++;
10050 if (nr != -1 && ret >= nr)
10051 goto out;
10052 cond_resched();
10053 spin_lock(&root->delalloc_lock);
10054 }
10055 spin_unlock(&root->delalloc_lock);
10056
10057 out:
10058 list_for_each_entry_safe(work, next, &works, list) {
10059 list_del_init(&work->list);
10060 btrfs_wait_and_free_delalloc_work(work);
10061 }
10062
10063 if (!list_empty_careful(&splice)) {
10064 spin_lock(&root->delalloc_lock);
10065 list_splice_tail(&splice, &root->delalloc_inodes);
10066 spin_unlock(&root->delalloc_lock);
10067 }
10068 mutex_unlock(&root->delalloc_mutex);
10069 return ret;
10070 }
10071
10072 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10073 {
10074 struct btrfs_fs_info *fs_info = root->fs_info;
10075 int ret;
10076
10077 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10078 return -EROFS;
10079
10080 ret = __start_delalloc_inodes(root, delay_iput, -1);
10081 if (ret > 0)
10082 ret = 0;
10083 /*
10084 * the filemap_flush will queue IO into the worker threads, but
10085 * we have to make sure the IO is actually started and that
10086 * ordered extents get created before we return
10087 */
10088 atomic_inc(&fs_info->async_submit_draining);
10089 while (atomic_read(&fs_info->nr_async_submits) ||
10090 atomic_read(&fs_info->async_delalloc_pages)) {
10091 wait_event(fs_info->async_submit_wait,
10092 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10093 atomic_read(&fs_info->async_delalloc_pages) == 0));
10094 }
10095 atomic_dec(&fs_info->async_submit_draining);
10096 return ret;
10097 }
10098
10099 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10100 int nr)
10101 {
10102 struct btrfs_root *root;
10103 struct list_head splice;
10104 int ret;
10105
10106 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10107 return -EROFS;
10108
10109 INIT_LIST_HEAD(&splice);
10110
10111 mutex_lock(&fs_info->delalloc_root_mutex);
10112 spin_lock(&fs_info->delalloc_root_lock);
10113 list_splice_init(&fs_info->delalloc_roots, &splice);
10114 while (!list_empty(&splice) && nr) {
10115 root = list_first_entry(&splice, struct btrfs_root,
10116 delalloc_root);
10117 root = btrfs_grab_fs_root(root);
10118 BUG_ON(!root);
10119 list_move_tail(&root->delalloc_root,
10120 &fs_info->delalloc_roots);
10121 spin_unlock(&fs_info->delalloc_root_lock);
10122
10123 ret = __start_delalloc_inodes(root, delay_iput, nr);
10124 btrfs_put_fs_root(root);
10125 if (ret < 0)
10126 goto out;
10127
10128 if (nr != -1) {
10129 nr -= ret;
10130 WARN_ON(nr < 0);
10131 }
10132 spin_lock(&fs_info->delalloc_root_lock);
10133 }
10134 spin_unlock(&fs_info->delalloc_root_lock);
10135
10136 ret = 0;
10137 atomic_inc(&fs_info->async_submit_draining);
10138 while (atomic_read(&fs_info->nr_async_submits) ||
10139 atomic_read(&fs_info->async_delalloc_pages)) {
10140 wait_event(fs_info->async_submit_wait,
10141 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10142 atomic_read(&fs_info->async_delalloc_pages) == 0));
10143 }
10144 atomic_dec(&fs_info->async_submit_draining);
10145 out:
10146 if (!list_empty_careful(&splice)) {
10147 spin_lock(&fs_info->delalloc_root_lock);
10148 list_splice_tail(&splice, &fs_info->delalloc_roots);
10149 spin_unlock(&fs_info->delalloc_root_lock);
10150 }
10151 mutex_unlock(&fs_info->delalloc_root_mutex);
10152 return ret;
10153 }
10154
10155 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10156 const char *symname)
10157 {
10158 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10159 struct btrfs_trans_handle *trans;
10160 struct btrfs_root *root = BTRFS_I(dir)->root;
10161 struct btrfs_path *path;
10162 struct btrfs_key key;
10163 struct inode *inode = NULL;
10164 int err;
10165 int drop_inode = 0;
10166 u64 objectid;
10167 u64 index = 0;
10168 int name_len;
10169 int datasize;
10170 unsigned long ptr;
10171 struct btrfs_file_extent_item *ei;
10172 struct extent_buffer *leaf;
10173
10174 name_len = strlen(symname);
10175 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10176 return -ENAMETOOLONG;
10177
10178 /*
10179 * 2 items for inode item and ref
10180 * 2 items for dir items
10181 * 1 item for updating parent inode item
10182 * 1 item for the inline extent item
10183 * 1 item for xattr if selinux is on
10184 */
10185 trans = btrfs_start_transaction(root, 7);
10186 if (IS_ERR(trans))
10187 return PTR_ERR(trans);
10188
10189 err = btrfs_find_free_ino(root, &objectid);
10190 if (err)
10191 goto out_unlock;
10192
10193 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10194 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10195 objectid, S_IFLNK|S_IRWXUGO, &index);
10196 if (IS_ERR(inode)) {
10197 err = PTR_ERR(inode);
10198 goto out_unlock;
10199 }
10200
10201 /*
10202 * If the active LSM wants to access the inode during
10203 * d_instantiate it needs these. Smack checks to see
10204 * if the filesystem supports xattrs by looking at the
10205 * ops vector.
10206 */
10207 inode->i_fop = &btrfs_file_operations;
10208 inode->i_op = &btrfs_file_inode_operations;
10209 inode->i_mapping->a_ops = &btrfs_aops;
10210 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10211
10212 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10213 if (err)
10214 goto out_unlock_inode;
10215
10216 path = btrfs_alloc_path();
10217 if (!path) {
10218 err = -ENOMEM;
10219 goto out_unlock_inode;
10220 }
10221 key.objectid = btrfs_ino(BTRFS_I(inode));
10222 key.offset = 0;
10223 key.type = BTRFS_EXTENT_DATA_KEY;
10224 datasize = btrfs_file_extent_calc_inline_size(name_len);
10225 err = btrfs_insert_empty_item(trans, root, path, &key,
10226 datasize);
10227 if (err) {
10228 btrfs_free_path(path);
10229 goto out_unlock_inode;
10230 }
10231 leaf = path->nodes[0];
10232 ei = btrfs_item_ptr(leaf, path->slots[0],
10233 struct btrfs_file_extent_item);
10234 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10235 btrfs_set_file_extent_type(leaf, ei,
10236 BTRFS_FILE_EXTENT_INLINE);
10237 btrfs_set_file_extent_encryption(leaf, ei, 0);
10238 btrfs_set_file_extent_compression(leaf, ei, 0);
10239 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10240 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10241
10242 ptr = btrfs_file_extent_inline_start(ei);
10243 write_extent_buffer(leaf, symname, ptr, name_len);
10244 btrfs_mark_buffer_dirty(leaf);
10245 btrfs_free_path(path);
10246
10247 inode->i_op = &btrfs_symlink_inode_operations;
10248 inode_nohighmem(inode);
10249 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10250 inode_set_bytes(inode, name_len);
10251 btrfs_i_size_write(BTRFS_I(inode), name_len);
10252 err = btrfs_update_inode(trans, root, inode);
10253 /*
10254 * Last step, add directory indexes for our symlink inode. This is the
10255 * last step to avoid extra cleanup of these indexes if an error happens
10256 * elsewhere above.
10257 */
10258 if (!err)
10259 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10260 BTRFS_I(inode), 0, index);
10261 if (err) {
10262 drop_inode = 1;
10263 goto out_unlock_inode;
10264 }
10265
10266 unlock_new_inode(inode);
10267 d_instantiate(dentry, inode);
10268
10269 out_unlock:
10270 btrfs_end_transaction(trans);
10271 if (drop_inode) {
10272 inode_dec_link_count(inode);
10273 iput(inode);
10274 }
10275 btrfs_btree_balance_dirty(fs_info);
10276 return err;
10277
10278 out_unlock_inode:
10279 drop_inode = 1;
10280 unlock_new_inode(inode);
10281 goto out_unlock;
10282 }
10283
10284 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10285 u64 start, u64 num_bytes, u64 min_size,
10286 loff_t actual_len, u64 *alloc_hint,
10287 struct btrfs_trans_handle *trans)
10288 {
10289 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10290 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10291 struct extent_map *em;
10292 struct btrfs_root *root = BTRFS_I(inode)->root;
10293 struct btrfs_key ins;
10294 u64 cur_offset = start;
10295 u64 i_size;
10296 u64 cur_bytes;
10297 u64 last_alloc = (u64)-1;
10298 int ret = 0;
10299 bool own_trans = true;
10300 u64 end = start + num_bytes - 1;
10301
10302 if (trans)
10303 own_trans = false;
10304 while (num_bytes > 0) {
10305 if (own_trans) {
10306 trans = btrfs_start_transaction(root, 3);
10307 if (IS_ERR(trans)) {
10308 ret = PTR_ERR(trans);
10309 break;
10310 }
10311 }
10312
10313 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10314 cur_bytes = max(cur_bytes, min_size);
10315 /*
10316 * If we are severely fragmented we could end up with really
10317 * small allocations, so if the allocator is returning small
10318 * chunks lets make its job easier by only searching for those
10319 * sized chunks.
10320 */
10321 cur_bytes = min(cur_bytes, last_alloc);
10322 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10323 min_size, 0, *alloc_hint, &ins, 1, 0);
10324 if (ret) {
10325 if (own_trans)
10326 btrfs_end_transaction(trans);
10327 break;
10328 }
10329 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10330
10331 last_alloc = ins.offset;
10332 ret = insert_reserved_file_extent(trans, inode,
10333 cur_offset, ins.objectid,
10334 ins.offset, ins.offset,
10335 ins.offset, 0, 0, 0,
10336 BTRFS_FILE_EXTENT_PREALLOC);
10337 if (ret) {
10338 btrfs_free_reserved_extent(fs_info, ins.objectid,
10339 ins.offset, 0);
10340 btrfs_abort_transaction(trans, ret);
10341 if (own_trans)
10342 btrfs_end_transaction(trans);
10343 break;
10344 }
10345
10346 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10347 cur_offset + ins.offset -1, 0);
10348
10349 em = alloc_extent_map();
10350 if (!em) {
10351 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10352 &BTRFS_I(inode)->runtime_flags);
10353 goto next;
10354 }
10355
10356 em->start = cur_offset;
10357 em->orig_start = cur_offset;
10358 em->len = ins.offset;
10359 em->block_start = ins.objectid;
10360 em->block_len = ins.offset;
10361 em->orig_block_len = ins.offset;
10362 em->ram_bytes = ins.offset;
10363 em->bdev = fs_info->fs_devices->latest_bdev;
10364 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10365 em->generation = trans->transid;
10366
10367 while (1) {
10368 write_lock(&em_tree->lock);
10369 ret = add_extent_mapping(em_tree, em, 1);
10370 write_unlock(&em_tree->lock);
10371 if (ret != -EEXIST)
10372 break;
10373 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10374 cur_offset + ins.offset - 1,
10375 0);
10376 }
10377 free_extent_map(em);
10378 next:
10379 num_bytes -= ins.offset;
10380 cur_offset += ins.offset;
10381 *alloc_hint = ins.objectid + ins.offset;
10382
10383 inode_inc_iversion(inode);
10384 inode->i_ctime = current_time(inode);
10385 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10386 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10387 (actual_len > inode->i_size) &&
10388 (cur_offset > inode->i_size)) {
10389 if (cur_offset > actual_len)
10390 i_size = actual_len;
10391 else
10392 i_size = cur_offset;
10393 i_size_write(inode, i_size);
10394 btrfs_ordered_update_i_size(inode, i_size, NULL);
10395 }
10396
10397 ret = btrfs_update_inode(trans, root, inode);
10398
10399 if (ret) {
10400 btrfs_abort_transaction(trans, ret);
10401 if (own_trans)
10402 btrfs_end_transaction(trans);
10403 break;
10404 }
10405
10406 if (own_trans)
10407 btrfs_end_transaction(trans);
10408 }
10409 if (cur_offset < end)
10410 btrfs_free_reserved_data_space(inode, cur_offset,
10411 end - cur_offset + 1);
10412 return ret;
10413 }
10414
10415 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10416 u64 start, u64 num_bytes, u64 min_size,
10417 loff_t actual_len, u64 *alloc_hint)
10418 {
10419 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10420 min_size, actual_len, alloc_hint,
10421 NULL);
10422 }
10423
10424 int btrfs_prealloc_file_range_trans(struct inode *inode,
10425 struct btrfs_trans_handle *trans, int mode,
10426 u64 start, u64 num_bytes, u64 min_size,
10427 loff_t actual_len, u64 *alloc_hint)
10428 {
10429 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10430 min_size, actual_len, alloc_hint, trans);
10431 }
10432
10433 static int btrfs_set_page_dirty(struct page *page)
10434 {
10435 return __set_page_dirty_nobuffers(page);
10436 }
10437
10438 static int btrfs_permission(struct inode *inode, int mask)
10439 {
10440 struct btrfs_root *root = BTRFS_I(inode)->root;
10441 umode_t mode = inode->i_mode;
10442
10443 if (mask & MAY_WRITE &&
10444 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10445 if (btrfs_root_readonly(root))
10446 return -EROFS;
10447 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10448 return -EACCES;
10449 }
10450 return generic_permission(inode, mask);
10451 }
10452
10453 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10454 {
10455 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10456 struct btrfs_trans_handle *trans;
10457 struct btrfs_root *root = BTRFS_I(dir)->root;
10458 struct inode *inode = NULL;
10459 u64 objectid;
10460 u64 index;
10461 int ret = 0;
10462
10463 /*
10464 * 5 units required for adding orphan entry
10465 */
10466 trans = btrfs_start_transaction(root, 5);
10467 if (IS_ERR(trans))
10468 return PTR_ERR(trans);
10469
10470 ret = btrfs_find_free_ino(root, &objectid);
10471 if (ret)
10472 goto out;
10473
10474 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10475 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10476 if (IS_ERR(inode)) {
10477 ret = PTR_ERR(inode);
10478 inode = NULL;
10479 goto out;
10480 }
10481
10482 inode->i_fop = &btrfs_file_operations;
10483 inode->i_op = &btrfs_file_inode_operations;
10484
10485 inode->i_mapping->a_ops = &btrfs_aops;
10486 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10487
10488 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10489 if (ret)
10490 goto out_inode;
10491
10492 ret = btrfs_update_inode(trans, root, inode);
10493 if (ret)
10494 goto out_inode;
10495 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10496 if (ret)
10497 goto out_inode;
10498
10499 /*
10500 * We set number of links to 0 in btrfs_new_inode(), and here we set
10501 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10502 * through:
10503 *
10504 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10505 */
10506 set_nlink(inode, 1);
10507 unlock_new_inode(inode);
10508 d_tmpfile(dentry, inode);
10509 mark_inode_dirty(inode);
10510
10511 out:
10512 btrfs_end_transaction(trans);
10513 if (ret)
10514 iput(inode);
10515 btrfs_balance_delayed_items(fs_info);
10516 btrfs_btree_balance_dirty(fs_info);
10517 return ret;
10518
10519 out_inode:
10520 unlock_new_inode(inode);
10521 goto out;
10522
10523 }
10524
10525 static const struct inode_operations btrfs_dir_inode_operations = {
10526 .getattr = btrfs_getattr,
10527 .lookup = btrfs_lookup,
10528 .create = btrfs_create,
10529 .unlink = btrfs_unlink,
10530 .link = btrfs_link,
10531 .mkdir = btrfs_mkdir,
10532 .rmdir = btrfs_rmdir,
10533 .rename = btrfs_rename2,
10534 .symlink = btrfs_symlink,
10535 .setattr = btrfs_setattr,
10536 .mknod = btrfs_mknod,
10537 .listxattr = btrfs_listxattr,
10538 .permission = btrfs_permission,
10539 .get_acl = btrfs_get_acl,
10540 .set_acl = btrfs_set_acl,
10541 .update_time = btrfs_update_time,
10542 .tmpfile = btrfs_tmpfile,
10543 };
10544 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10545 .lookup = btrfs_lookup,
10546 .permission = btrfs_permission,
10547 .update_time = btrfs_update_time,
10548 };
10549
10550 static const struct file_operations btrfs_dir_file_operations = {
10551 .llseek = generic_file_llseek,
10552 .read = generic_read_dir,
10553 .iterate_shared = btrfs_real_readdir,
10554 .unlocked_ioctl = btrfs_ioctl,
10555 #ifdef CONFIG_COMPAT
10556 .compat_ioctl = btrfs_compat_ioctl,
10557 #endif
10558 .release = btrfs_release_file,
10559 .fsync = btrfs_sync_file,
10560 };
10561
10562 static const struct extent_io_ops btrfs_extent_io_ops = {
10563 .fill_delalloc = run_delalloc_range,
10564 .submit_bio_hook = btrfs_submit_bio_hook,
10565 .merge_bio_hook = btrfs_merge_bio_hook,
10566 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10567 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10568 .writepage_start_hook = btrfs_writepage_start_hook,
10569 .set_bit_hook = btrfs_set_bit_hook,
10570 .clear_bit_hook = btrfs_clear_bit_hook,
10571 .merge_extent_hook = btrfs_merge_extent_hook,
10572 .split_extent_hook = btrfs_split_extent_hook,
10573 };
10574
10575 /*
10576 * btrfs doesn't support the bmap operation because swapfiles
10577 * use bmap to make a mapping of extents in the file. They assume
10578 * these extents won't change over the life of the file and they
10579 * use the bmap result to do IO directly to the drive.
10580 *
10581 * the btrfs bmap call would return logical addresses that aren't
10582 * suitable for IO and they also will change frequently as COW
10583 * operations happen. So, swapfile + btrfs == corruption.
10584 *
10585 * For now we're avoiding this by dropping bmap.
10586 */
10587 static const struct address_space_operations btrfs_aops = {
10588 .readpage = btrfs_readpage,
10589 .writepage = btrfs_writepage,
10590 .writepages = btrfs_writepages,
10591 .readpages = btrfs_readpages,
10592 .direct_IO = btrfs_direct_IO,
10593 .invalidatepage = btrfs_invalidatepage,
10594 .releasepage = btrfs_releasepage,
10595 .set_page_dirty = btrfs_set_page_dirty,
10596 .error_remove_page = generic_error_remove_page,
10597 };
10598
10599 static const struct address_space_operations btrfs_symlink_aops = {
10600 .readpage = btrfs_readpage,
10601 .writepage = btrfs_writepage,
10602 .invalidatepage = btrfs_invalidatepage,
10603 .releasepage = btrfs_releasepage,
10604 };
10605
10606 static const struct inode_operations btrfs_file_inode_operations = {
10607 .getattr = btrfs_getattr,
10608 .setattr = btrfs_setattr,
10609 .listxattr = btrfs_listxattr,
10610 .permission = btrfs_permission,
10611 .fiemap = btrfs_fiemap,
10612 .get_acl = btrfs_get_acl,
10613 .set_acl = btrfs_set_acl,
10614 .update_time = btrfs_update_time,
10615 };
10616 static const struct inode_operations btrfs_special_inode_operations = {
10617 .getattr = btrfs_getattr,
10618 .setattr = btrfs_setattr,
10619 .permission = btrfs_permission,
10620 .listxattr = btrfs_listxattr,
10621 .get_acl = btrfs_get_acl,
10622 .set_acl = btrfs_set_acl,
10623 .update_time = btrfs_update_time,
10624 };
10625 static const struct inode_operations btrfs_symlink_inode_operations = {
10626 .get_link = page_get_link,
10627 .getattr = btrfs_getattr,
10628 .setattr = btrfs_setattr,
10629 .permission = btrfs_permission,
10630 .listxattr = btrfs_listxattr,
10631 .update_time = btrfs_update_time,
10632 };
10633
10634 const struct dentry_operations btrfs_dentry_operations = {
10635 .d_delete = btrfs_dentry_delete,
10636 .d_release = btrfs_dentry_release,
10637 };
ubuntu-bionic-kernel mirror