]> 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_commit_inode_delayed_inode 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 };
75
76 static const struct inode_operations btrfs_dir_inode_operations;
77 static const struct inode_operations btrfs_symlink_inode_operations;
78 static const struct inode_operations btrfs_dir_ro_inode_operations;
79 static const struct inode_operations btrfs_special_inode_operations;
80 static const struct inode_operations btrfs_file_inode_operations;
81 static const struct address_space_operations btrfs_aops;
82 static const struct address_space_operations btrfs_symlink_aops;
83 static const struct file_operations btrfs_dir_file_operations;
84 static const struct extent_io_ops btrfs_extent_io_ops;
85
86 static struct kmem_cache *btrfs_inode_cachep;
87 struct kmem_cache *btrfs_trans_handle_cachep;
88 struct kmem_cache *btrfs_transaction_cachep;
89 struct kmem_cache *btrfs_path_cachep;
90 struct kmem_cache *btrfs_free_space_cachep;
91
92 #define S_SHIFT 12
93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
94 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
95 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
96 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
97 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
98 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
99 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
100 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
101 };
102
103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
104 static int btrfs_truncate(struct inode *inode);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
106 static noinline int cow_file_range(struct inode *inode,
107 struct page *locked_page,
108 u64 start, u64 end, u64 delalloc_end,
109 int *page_started, unsigned long *nr_written,
110 int unlock, struct btrfs_dedupe_hash *hash);
111 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
112 u64 len, u64 orig_start,
113 u64 block_start, u64 block_len,
114 u64 orig_block_len, u64 ram_bytes,
115 int type);
116
117 static int btrfs_dirty_inode(struct inode *inode);
118
119 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
120 void btrfs_test_inode_set_ops(struct inode *inode)
121 {
122 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
123 }
124 #endif
125
126 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
127 struct inode *inode, struct inode *dir,
128 const struct qstr *qstr)
129 {
130 int err;
131
132 err = btrfs_init_acl(trans, inode, dir);
133 if (!err)
134 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
135 return err;
136 }
137
138 /*
139 * this does all the hard work for inserting an inline extent into
140 * the btree. The caller should have done a btrfs_drop_extents so that
141 * no overlapping inline items exist in the btree
142 */
143 static int insert_inline_extent(struct btrfs_trans_handle *trans,
144 struct btrfs_path *path, int extent_inserted,
145 struct btrfs_root *root, struct inode *inode,
146 u64 start, size_t size, size_t compressed_size,
147 int compress_type,
148 struct page **compressed_pages)
149 {
150 struct extent_buffer *leaf;
151 struct page *page = NULL;
152 char *kaddr;
153 unsigned long ptr;
154 struct btrfs_file_extent_item *ei;
155 int err = 0;
156 int ret;
157 size_t cur_size = size;
158 unsigned long offset;
159
160 if (compressed_size && compressed_pages)
161 cur_size = compressed_size;
162
163 inode_add_bytes(inode, size);
164
165 if (!extent_inserted) {
166 struct btrfs_key key;
167 size_t datasize;
168
169 key.objectid = btrfs_ino(BTRFS_I(inode));
170 key.offset = start;
171 key.type = BTRFS_EXTENT_DATA_KEY;
172
173 datasize = btrfs_file_extent_calc_inline_size(cur_size);
174 path->leave_spinning = 1;
175 ret = btrfs_insert_empty_item(trans, root, path, &key,
176 datasize);
177 if (ret) {
178 err = ret;
179 goto fail;
180 }
181 }
182 leaf = path->nodes[0];
183 ei = btrfs_item_ptr(leaf, path->slots[0],
184 struct btrfs_file_extent_item);
185 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
186 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
187 btrfs_set_file_extent_encryption(leaf, ei, 0);
188 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
189 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
190 ptr = btrfs_file_extent_inline_start(ei);
191
192 if (compress_type != BTRFS_COMPRESS_NONE) {
193 struct page *cpage;
194 int i = 0;
195 while (compressed_size > 0) {
196 cpage = compressed_pages[i];
197 cur_size = min_t(unsigned long, compressed_size,
198 PAGE_SIZE);
199
200 kaddr = kmap_atomic(cpage);
201 write_extent_buffer(leaf, kaddr, ptr, cur_size);
202 kunmap_atomic(kaddr);
203
204 i++;
205 ptr += cur_size;
206 compressed_size -= cur_size;
207 }
208 btrfs_set_file_extent_compression(leaf, ei,
209 compress_type);
210 } else {
211 page = find_get_page(inode->i_mapping,
212 start >> PAGE_SHIFT);
213 btrfs_set_file_extent_compression(leaf, ei, 0);
214 kaddr = kmap_atomic(page);
215 offset = start & (PAGE_SIZE - 1);
216 write_extent_buffer(leaf, kaddr + offset, ptr, size);
217 kunmap_atomic(kaddr);
218 put_page(page);
219 }
220 btrfs_mark_buffer_dirty(leaf);
221 btrfs_release_path(path);
222
223 /*
224 * we're an inline extent, so nobody can
225 * extend the file past i_size without locking
226 * a page we already have locked.
227 *
228 * We must do any isize and inode updates
229 * before we unlock the pages. Otherwise we
230 * could end up racing with unlink.
231 */
232 BTRFS_I(inode)->disk_i_size = inode->i_size;
233 ret = btrfs_update_inode(trans, root, inode);
234
235 return ret;
236 fail:
237 return err;
238 }
239
240
241 /*
242 * conditionally insert an inline extent into the file. This
243 * does the checks required to make sure the data is small enough
244 * to fit as an inline extent.
245 */
246 static noinline int cow_file_range_inline(struct btrfs_root *root,
247 struct inode *inode, u64 start,
248 u64 end, size_t compressed_size,
249 int compress_type,
250 struct page **compressed_pages)
251 {
252 struct btrfs_fs_info *fs_info = root->fs_info;
253 struct btrfs_trans_handle *trans;
254 u64 isize = i_size_read(inode);
255 u64 actual_end = min(end + 1, isize);
256 u64 inline_len = actual_end - start;
257 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
258 u64 data_len = inline_len;
259 int ret;
260 struct btrfs_path *path;
261 int extent_inserted = 0;
262 u32 extent_item_size;
263
264 if (compressed_size)
265 data_len = compressed_size;
266
267 if (start > 0 ||
268 actual_end > fs_info->sectorsize ||
269 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
270 (!compressed_size &&
271 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
272 end + 1 < isize ||
273 data_len > fs_info->max_inline) {
274 return 1;
275 }
276
277 path = btrfs_alloc_path();
278 if (!path)
279 return -ENOMEM;
280
281 trans = btrfs_join_transaction(root);
282 if (IS_ERR(trans)) {
283 btrfs_free_path(path);
284 return PTR_ERR(trans);
285 }
286 trans->block_rsv = &fs_info->delalloc_block_rsv;
287
288 if (compressed_size && compressed_pages)
289 extent_item_size = btrfs_file_extent_calc_inline_size(
290 compressed_size);
291 else
292 extent_item_size = btrfs_file_extent_calc_inline_size(
293 inline_len);
294
295 ret = __btrfs_drop_extents(trans, root, inode, path,
296 start, aligned_end, NULL,
297 1, 1, extent_item_size, &extent_inserted);
298 if (ret) {
299 btrfs_abort_transaction(trans, ret);
300 goto out;
301 }
302
303 if (isize > actual_end)
304 inline_len = min_t(u64, isize, actual_end);
305 ret = insert_inline_extent(trans, path, extent_inserted,
306 root, inode, start,
307 inline_len, compressed_size,
308 compress_type, compressed_pages);
309 if (ret && ret != -ENOSPC) {
310 btrfs_abort_transaction(trans, ret);
311 goto out;
312 } else if (ret == -ENOSPC) {
313 ret = 1;
314 goto out;
315 }
316
317 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
318 btrfs_delalloc_release_metadata(inode, end + 1 - start);
319 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
320 out:
321 /*
322 * Don't forget to free the reserved space, as for inlined extent
323 * it won't count as data extent, free them directly here.
324 * And at reserve time, it's always aligned to page size, so
325 * just free one page here.
326 */
327 btrfs_qgroup_free_data(inode, 0, PAGE_SIZE);
328 btrfs_free_path(path);
329 btrfs_end_transaction(trans);
330 return ret;
331 }
332
333 struct async_extent {
334 u64 start;
335 u64 ram_size;
336 u64 compressed_size;
337 struct page **pages;
338 unsigned long nr_pages;
339 int compress_type;
340 struct list_head list;
341 };
342
343 struct async_cow {
344 struct inode *inode;
345 struct btrfs_root *root;
346 struct page *locked_page;
347 u64 start;
348 u64 end;
349 struct list_head extents;
350 struct btrfs_work work;
351 };
352
353 static noinline int add_async_extent(struct async_cow *cow,
354 u64 start, u64 ram_size,
355 u64 compressed_size,
356 struct page **pages,
357 unsigned long nr_pages,
358 int compress_type)
359 {
360 struct async_extent *async_extent;
361
362 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
363 BUG_ON(!async_extent); /* -ENOMEM */
364 async_extent->start = start;
365 async_extent->ram_size = ram_size;
366 async_extent->compressed_size = compressed_size;
367 async_extent->pages = pages;
368 async_extent->nr_pages = nr_pages;
369 async_extent->compress_type = compress_type;
370 list_add_tail(&async_extent->list, &cow->extents);
371 return 0;
372 }
373
374 static inline int inode_need_compress(struct inode *inode)
375 {
376 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
377
378 /* force compress */
379 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
380 return 1;
381 /* bad compression ratios */
382 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
383 return 0;
384 if (btrfs_test_opt(fs_info, COMPRESS) ||
385 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
386 BTRFS_I(inode)->force_compress)
387 return 1;
388 return 0;
389 }
390
391 static inline void inode_should_defrag(struct inode *inode,
392 u64 start, u64 end, u64 num_bytes, u64 small_write)
393 {
394 /* If this is a small write inside eof, kick off a defrag */
395 if (num_bytes < small_write &&
396 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
397 btrfs_add_inode_defrag(NULL, inode);
398 }
399
400 /*
401 * we create compressed extents in two phases. The first
402 * phase compresses a range of pages that have already been
403 * locked (both pages and state bits are locked).
404 *
405 * This is done inside an ordered work queue, and the compression
406 * is spread across many cpus. The actual IO submission is step
407 * two, and the ordered work queue takes care of making sure that
408 * happens in the same order things were put onto the queue by
409 * writepages and friends.
410 *
411 * If this code finds it can't get good compression, it puts an
412 * entry onto the work queue to write the uncompressed bytes. This
413 * makes sure that both compressed inodes and uncompressed inodes
414 * are written in the same order that the flusher thread sent them
415 * down.
416 */
417 static noinline void compress_file_range(struct inode *inode,
418 struct page *locked_page,
419 u64 start, u64 end,
420 struct async_cow *async_cow,
421 int *num_added)
422 {
423 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
424 struct btrfs_root *root = BTRFS_I(inode)->root;
425 u64 num_bytes;
426 u64 blocksize = fs_info->sectorsize;
427 u64 actual_end;
428 u64 isize = i_size_read(inode);
429 int ret = 0;
430 struct page **pages = NULL;
431 unsigned long nr_pages;
432 unsigned long nr_pages_ret = 0;
433 unsigned long total_compressed = 0;
434 unsigned long total_in = 0;
435 unsigned long max_compressed = SZ_128K;
436 unsigned long max_uncompressed = SZ_128K;
437 int i;
438 int will_compress;
439 int compress_type = fs_info->compress_type;
440 int redirty = 0;
441
442 inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
443
444 actual_end = min_t(u64, isize, end + 1);
445 again:
446 will_compress = 0;
447 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
448 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_SIZE);
449
450 /*
451 * we don't want to send crud past the end of i_size through
452 * compression, that's just a waste of CPU time. So, if the
453 * end of the file is before the start of our current
454 * requested range of bytes, we bail out to the uncompressed
455 * cleanup code that can deal with all of this.
456 *
457 * It isn't really the fastest way to fix things, but this is a
458 * very uncommon corner.
459 */
460 if (actual_end <= start)
461 goto cleanup_and_bail_uncompressed;
462
463 total_compressed = actual_end - start;
464
465 /*
466 * skip compression for a small file range(<=blocksize) that
467 * isn't an inline extent, since it doesn't save disk space at all.
468 */
469 if (total_compressed <= blocksize &&
470 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
471 goto cleanup_and_bail_uncompressed;
472
473 /* we want to make sure that amount of ram required to uncompress
474 * an extent is reasonable, so we limit the total size in ram
475 * of a compressed extent to 128k. This is a crucial number
476 * because it also controls how easily we can spread reads across
477 * cpus for decompression.
478 *
479 * We also want to make sure the amount of IO required to do
480 * a random read is reasonably small, so we limit the size of
481 * a compressed extent to 128k.
482 */
483 total_compressed = min(total_compressed, max_uncompressed);
484 num_bytes = ALIGN(end - start + 1, blocksize);
485 num_bytes = max(blocksize, num_bytes);
486 total_in = 0;
487 ret = 0;
488
489 /*
490 * we do compression for mount -o compress and when the
491 * inode has not been flagged as nocompress. This flag can
492 * change at any time if we discover bad compression ratios.
493 */
494 if (inode_need_compress(inode)) {
495 WARN_ON(pages);
496 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
497 if (!pages) {
498 /* just bail out to the uncompressed code */
499 goto cont;
500 }
501
502 if (BTRFS_I(inode)->force_compress)
503 compress_type = BTRFS_I(inode)->force_compress;
504
505 /*
506 * we need to call clear_page_dirty_for_io on each
507 * page in the range. Otherwise applications with the file
508 * mmap'd can wander in and change the page contents while
509 * we are compressing them.
510 *
511 * If the compression fails for any reason, we set the pages
512 * dirty again later on.
513 */
514 extent_range_clear_dirty_for_io(inode, start, end);
515 redirty = 1;
516 ret = btrfs_compress_pages(compress_type,
517 inode->i_mapping, start,
518 total_compressed, pages,
519 nr_pages, &nr_pages_ret,
520 &total_in,
521 &total_compressed,
522 max_compressed);
523
524 if (!ret) {
525 unsigned long offset = total_compressed &
526 (PAGE_SIZE - 1);
527 struct page *page = pages[nr_pages_ret - 1];
528 char *kaddr;
529
530 /* zero the tail end of the last page, we might be
531 * sending it down to disk
532 */
533 if (offset) {
534 kaddr = kmap_atomic(page);
535 memset(kaddr + offset, 0,
536 PAGE_SIZE - offset);
537 kunmap_atomic(kaddr);
538 }
539 will_compress = 1;
540 }
541 }
542 cont:
543 if (start == 0) {
544 /* lets try to make an inline extent */
545 if (ret || total_in < (actual_end - start)) {
546 /* we didn't compress the entire range, try
547 * to make an uncompressed inline extent.
548 */
549 ret = cow_file_range_inline(root, inode, start, end,
550 0, BTRFS_COMPRESS_NONE, NULL);
551 } else {
552 /* try making a compressed inline extent */
553 ret = cow_file_range_inline(root, inode, start, end,
554 total_compressed,
555 compress_type, pages);
556 }
557 if (ret <= 0) {
558 unsigned long clear_flags = EXTENT_DELALLOC |
559 EXTENT_DEFRAG;
560 unsigned long page_error_op;
561
562 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
563 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
564
565 /*
566 * inline extent creation worked or returned error,
567 * we don't need to create any more async work items.
568 * Unlock and free up our temp pages.
569 */
570 extent_clear_unlock_delalloc(inode, start, end, end,
571 NULL, clear_flags,
572 PAGE_UNLOCK |
573 PAGE_CLEAR_DIRTY |
574 PAGE_SET_WRITEBACK |
575 page_error_op |
576 PAGE_END_WRITEBACK);
577 btrfs_free_reserved_data_space_noquota(inode, start,
578 end - start + 1);
579 goto free_pages_out;
580 }
581 }
582
583 if (will_compress) {
584 /*
585 * we aren't doing an inline extent round the compressed size
586 * up to a block size boundary so the allocator does sane
587 * things
588 */
589 total_compressed = ALIGN(total_compressed, blocksize);
590
591 /*
592 * one last check to make sure the compression is really a
593 * win, compare the page count read with the blocks on disk
594 */
595 total_in = ALIGN(total_in, PAGE_SIZE);
596 if (total_compressed >= total_in) {
597 will_compress = 0;
598 } else {
599 num_bytes = total_in;
600 *num_added += 1;
601
602 /*
603 * The async work queues will take care of doing actual
604 * allocation on disk for these compressed pages, and
605 * will submit them to the elevator.
606 */
607 add_async_extent(async_cow, start, num_bytes,
608 total_compressed, pages, nr_pages_ret,
609 compress_type);
610
611 if (start + num_bytes < end) {
612 start += num_bytes;
613 pages = NULL;
614 cond_resched();
615 goto again;
616 }
617 return;
618 }
619 }
620 if (pages) {
621 /*
622 * the compression code ran but failed to make things smaller,
623 * free any pages it allocated and our page pointer array
624 */
625 for (i = 0; i < nr_pages_ret; i++) {
626 WARN_ON(pages[i]->mapping);
627 put_page(pages[i]);
628 }
629 kfree(pages);
630 pages = NULL;
631 total_compressed = 0;
632 nr_pages_ret = 0;
633
634 /* flag the file so we don't compress in the future */
635 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
636 !(BTRFS_I(inode)->force_compress)) {
637 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
638 }
639 }
640 cleanup_and_bail_uncompressed:
641 /*
642 * No compression, but we still need to write the pages in the file
643 * we've been given so far. redirty the locked page if it corresponds
644 * to our extent and set things up for the async work queue to run
645 * cow_file_range to do the normal delalloc dance.
646 */
647 if (page_offset(locked_page) >= start &&
648 page_offset(locked_page) <= end)
649 __set_page_dirty_nobuffers(locked_page);
650 /* unlocked later on in the async handlers */
651
652 if (redirty)
653 extent_range_redirty_for_io(inode, start, end);
654 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
655 BTRFS_COMPRESS_NONE);
656 *num_added += 1;
657
658 return;
659
660 free_pages_out:
661 for (i = 0; i < nr_pages_ret; i++) {
662 WARN_ON(pages[i]->mapping);
663 put_page(pages[i]);
664 }
665 kfree(pages);
666 }
667
668 static void free_async_extent_pages(struct async_extent *async_extent)
669 {
670 int i;
671
672 if (!async_extent->pages)
673 return;
674
675 for (i = 0; i < async_extent->nr_pages; i++) {
676 WARN_ON(async_extent->pages[i]->mapping);
677 put_page(async_extent->pages[i]);
678 }
679 kfree(async_extent->pages);
680 async_extent->nr_pages = 0;
681 async_extent->pages = NULL;
682 }
683
684 /*
685 * phase two of compressed writeback. This is the ordered portion
686 * of the code, which only gets called in the order the work was
687 * queued. We walk all the async extents created by compress_file_range
688 * and send them down to the disk.
689 */
690 static noinline void submit_compressed_extents(struct inode *inode,
691 struct async_cow *async_cow)
692 {
693 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
694 struct async_extent *async_extent;
695 u64 alloc_hint = 0;
696 struct btrfs_key ins;
697 struct extent_map *em;
698 struct btrfs_root *root = BTRFS_I(inode)->root;
699 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
700 struct extent_io_tree *io_tree;
701 int ret = 0;
702
703 again:
704 while (!list_empty(&async_cow->extents)) {
705 async_extent = list_entry(async_cow->extents.next,
706 struct async_extent, list);
707 list_del(&async_extent->list);
708
709 io_tree = &BTRFS_I(inode)->io_tree;
710
711 retry:
712 /* did the compression code fall back to uncompressed IO? */
713 if (!async_extent->pages) {
714 int page_started = 0;
715 unsigned long nr_written = 0;
716
717 lock_extent(io_tree, async_extent->start,
718 async_extent->start +
719 async_extent->ram_size - 1);
720
721 /* allocate blocks */
722 ret = cow_file_range(inode, async_cow->locked_page,
723 async_extent->start,
724 async_extent->start +
725 async_extent->ram_size - 1,
726 async_extent->start +
727 async_extent->ram_size - 1,
728 &page_started, &nr_written, 0,
729 NULL);
730
731 /* JDM XXX */
732
733 /*
734 * if page_started, cow_file_range inserted an
735 * inline extent and took care of all the unlocking
736 * and IO for us. Otherwise, we need to submit
737 * all those pages down to the drive.
738 */
739 if (!page_started && !ret)
740 extent_write_locked_range(io_tree,
741 inode, async_extent->start,
742 async_extent->start +
743 async_extent->ram_size - 1,
744 btrfs_get_extent,
745 WB_SYNC_ALL);
746 else if (ret)
747 unlock_page(async_cow->locked_page);
748 kfree(async_extent);
749 cond_resched();
750 continue;
751 }
752
753 lock_extent(io_tree, async_extent->start,
754 async_extent->start + async_extent->ram_size - 1);
755
756 ret = btrfs_reserve_extent(root, async_extent->ram_size,
757 async_extent->compressed_size,
758 async_extent->compressed_size,
759 0, alloc_hint, &ins, 1, 1);
760 if (ret) {
761 free_async_extent_pages(async_extent);
762
763 if (ret == -ENOSPC) {
764 unlock_extent(io_tree, async_extent->start,
765 async_extent->start +
766 async_extent->ram_size - 1);
767
768 /*
769 * we need to redirty the pages if we decide to
770 * fallback to uncompressed IO, otherwise we
771 * will not submit these pages down to lower
772 * layers.
773 */
774 extent_range_redirty_for_io(inode,
775 async_extent->start,
776 async_extent->start +
777 async_extent->ram_size - 1);
778
779 goto retry;
780 }
781 goto out_free;
782 }
783 /*
784 * here we're doing allocation and writeback of the
785 * compressed pages
786 */
787 btrfs_drop_extent_cache(inode, async_extent->start,
788 async_extent->start +
789 async_extent->ram_size - 1, 0);
790
791 em = alloc_extent_map();
792 if (!em) {
793 ret = -ENOMEM;
794 goto out_free_reserve;
795 }
796 em->start = async_extent->start;
797 em->len = async_extent->ram_size;
798 em->orig_start = em->start;
799 em->mod_start = em->start;
800 em->mod_len = em->len;
801
802 em->block_start = ins.objectid;
803 em->block_len = ins.offset;
804 em->orig_block_len = ins.offset;
805 em->ram_bytes = async_extent->ram_size;
806 em->bdev = fs_info->fs_devices->latest_bdev;
807 em->compress_type = async_extent->compress_type;
808 set_bit(EXTENT_FLAG_PINNED, &em->flags);
809 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
810 em->generation = -1;
811
812 while (1) {
813 write_lock(&em_tree->lock);
814 ret = add_extent_mapping(em_tree, em, 1);
815 write_unlock(&em_tree->lock);
816 if (ret != -EEXIST) {
817 free_extent_map(em);
818 break;
819 }
820 btrfs_drop_extent_cache(inode, async_extent->start,
821 async_extent->start +
822 async_extent->ram_size - 1, 0);
823 }
824
825 if (ret)
826 goto out_free_reserve;
827
828 ret = btrfs_add_ordered_extent_compress(inode,
829 async_extent->start,
830 ins.objectid,
831 async_extent->ram_size,
832 ins.offset,
833 BTRFS_ORDERED_COMPRESSED,
834 async_extent->compress_type);
835 if (ret) {
836 btrfs_drop_extent_cache(inode, async_extent->start,
837 async_extent->start +
838 async_extent->ram_size - 1, 0);
839 goto out_free_reserve;
840 }
841 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
842
843 /*
844 * clear dirty, set writeback and unlock the pages.
845 */
846 extent_clear_unlock_delalloc(inode, async_extent->start,
847 async_extent->start +
848 async_extent->ram_size - 1,
849 async_extent->start +
850 async_extent->ram_size - 1,
851 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
852 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
853 PAGE_SET_WRITEBACK);
854 ret = btrfs_submit_compressed_write(inode,
855 async_extent->start,
856 async_extent->ram_size,
857 ins.objectid,
858 ins.offset, async_extent->pages,
859 async_extent->nr_pages);
860 if (ret) {
861 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
862 struct page *p = async_extent->pages[0];
863 const u64 start = async_extent->start;
864 const u64 end = start + async_extent->ram_size - 1;
865
866 p->mapping = inode->i_mapping;
867 tree->ops->writepage_end_io_hook(p, start, end,
868 NULL, 0);
869 p->mapping = NULL;
870 extent_clear_unlock_delalloc(inode, start, end, end,
871 NULL, 0,
872 PAGE_END_WRITEBACK |
873 PAGE_SET_ERROR);
874 free_async_extent_pages(async_extent);
875 }
876 alloc_hint = ins.objectid + ins.offset;
877 kfree(async_extent);
878 cond_resched();
879 }
880 return;
881 out_free_reserve:
882 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
883 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
884 out_free:
885 extent_clear_unlock_delalloc(inode, async_extent->start,
886 async_extent->start +
887 async_extent->ram_size - 1,
888 async_extent->start +
889 async_extent->ram_size - 1,
890 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
891 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
892 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
893 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
894 PAGE_SET_ERROR);
895 free_async_extent_pages(async_extent);
896 kfree(async_extent);
897 goto again;
898 }
899
900 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
901 u64 num_bytes)
902 {
903 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
904 struct extent_map *em;
905 u64 alloc_hint = 0;
906
907 read_lock(&em_tree->lock);
908 em = search_extent_mapping(em_tree, start, num_bytes);
909 if (em) {
910 /*
911 * if block start isn't an actual block number then find the
912 * first block in this inode and use that as a hint. If that
913 * block is also bogus then just don't worry about it.
914 */
915 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
916 free_extent_map(em);
917 em = search_extent_mapping(em_tree, 0, 0);
918 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
919 alloc_hint = em->block_start;
920 if (em)
921 free_extent_map(em);
922 } else {
923 alloc_hint = em->block_start;
924 free_extent_map(em);
925 }
926 }
927 read_unlock(&em_tree->lock);
928
929 return alloc_hint;
930 }
931
932 /*
933 * when extent_io.c finds a delayed allocation range in the file,
934 * the call backs end up in this code. The basic idea is to
935 * allocate extents on disk for the range, and create ordered data structs
936 * in ram to track those extents.
937 *
938 * locked_page is the page that writepage had locked already. We use
939 * it to make sure we don't do extra locks or unlocks.
940 *
941 * *page_started is set to one if we unlock locked_page and do everything
942 * required to start IO on it. It may be clean and already done with
943 * IO when we return.
944 */
945 static noinline int cow_file_range(struct inode *inode,
946 struct page *locked_page,
947 u64 start, u64 end, u64 delalloc_end,
948 int *page_started, unsigned long *nr_written,
949 int unlock, struct btrfs_dedupe_hash *hash)
950 {
951 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
952 struct btrfs_root *root = BTRFS_I(inode)->root;
953 u64 alloc_hint = 0;
954 u64 num_bytes;
955 unsigned long ram_size;
956 u64 disk_num_bytes;
957 u64 cur_alloc_size;
958 u64 blocksize = fs_info->sectorsize;
959 struct btrfs_key ins;
960 struct extent_map *em;
961 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
962 int ret = 0;
963
964 if (btrfs_is_free_space_inode(inode)) {
965 WARN_ON_ONCE(1);
966 ret = -EINVAL;
967 goto out_unlock;
968 }
969
970 num_bytes = ALIGN(end - start + 1, blocksize);
971 num_bytes = max(blocksize, num_bytes);
972 disk_num_bytes = num_bytes;
973
974 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
975
976 if (start == 0) {
977 /* lets try to make an inline extent */
978 ret = cow_file_range_inline(root, inode, start, end, 0,
979 BTRFS_COMPRESS_NONE, NULL);
980 if (ret == 0) {
981 extent_clear_unlock_delalloc(inode, start, end,
982 delalloc_end, NULL,
983 EXTENT_LOCKED | EXTENT_DELALLOC |
984 EXTENT_DEFRAG, PAGE_UNLOCK |
985 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
986 PAGE_END_WRITEBACK);
987 btrfs_free_reserved_data_space_noquota(inode, start,
988 end - start + 1);
989 *nr_written = *nr_written +
990 (end - start + PAGE_SIZE) / PAGE_SIZE;
991 *page_started = 1;
992 goto out;
993 } else if (ret < 0) {
994 goto out_unlock;
995 }
996 }
997
998 BUG_ON(disk_num_bytes >
999 btrfs_super_total_bytes(fs_info->super_copy));
1000
1001 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1002 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1003
1004 while (disk_num_bytes > 0) {
1005 unsigned long op;
1006
1007 cur_alloc_size = disk_num_bytes;
1008 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1009 fs_info->sectorsize, 0, alloc_hint,
1010 &ins, 1, 1);
1011 if (ret < 0)
1012 goto out_unlock;
1013
1014 em = alloc_extent_map();
1015 if (!em) {
1016 ret = -ENOMEM;
1017 goto out_reserve;
1018 }
1019 em->start = start;
1020 em->orig_start = em->start;
1021 ram_size = ins.offset;
1022 em->len = ins.offset;
1023 em->mod_start = em->start;
1024 em->mod_len = em->len;
1025
1026 em->block_start = ins.objectid;
1027 em->block_len = ins.offset;
1028 em->orig_block_len = ins.offset;
1029 em->ram_bytes = ram_size;
1030 em->bdev = fs_info->fs_devices->latest_bdev;
1031 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1032 em->generation = -1;
1033
1034 while (1) {
1035 write_lock(&em_tree->lock);
1036 ret = add_extent_mapping(em_tree, em, 1);
1037 write_unlock(&em_tree->lock);
1038 if (ret != -EEXIST) {
1039 free_extent_map(em);
1040 break;
1041 }
1042 btrfs_drop_extent_cache(inode, start,
1043 start + ram_size - 1, 0);
1044 }
1045 if (ret)
1046 goto out_reserve;
1047
1048 cur_alloc_size = ins.offset;
1049 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1050 ram_size, cur_alloc_size, 0);
1051 if (ret)
1052 goto out_drop_extent_cache;
1053
1054 if (root->root_key.objectid ==
1055 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1056 ret = btrfs_reloc_clone_csums(inode, start,
1057 cur_alloc_size);
1058 if (ret)
1059 goto out_drop_extent_cache;
1060 }
1061
1062 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1063
1064 if (disk_num_bytes < cur_alloc_size)
1065 break;
1066
1067 /* we're not doing compressed IO, don't unlock the first
1068 * page (which the caller expects to stay locked), don't
1069 * clear any dirty bits and don't set any writeback bits
1070 *
1071 * Do set the Private2 bit so we know this page was properly
1072 * setup for writepage
1073 */
1074 op = unlock ? PAGE_UNLOCK : 0;
1075 op |= PAGE_SET_PRIVATE2;
1076
1077 extent_clear_unlock_delalloc(inode, start,
1078 start + ram_size - 1,
1079 delalloc_end, locked_page,
1080 EXTENT_LOCKED | EXTENT_DELALLOC,
1081 op);
1082 disk_num_bytes -= cur_alloc_size;
1083 num_bytes -= cur_alloc_size;
1084 alloc_hint = ins.objectid + ins.offset;
1085 start += cur_alloc_size;
1086 }
1087 out:
1088 return ret;
1089
1090 out_drop_extent_cache:
1091 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1092 out_reserve:
1093 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1094 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1095 out_unlock:
1096 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1097 locked_page,
1098 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1099 EXTENT_DELALLOC | EXTENT_DEFRAG,
1100 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1101 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1102 goto out;
1103 }
1104
1105 /*
1106 * work queue call back to started compression on a file and pages
1107 */
1108 static noinline void async_cow_start(struct btrfs_work *work)
1109 {
1110 struct async_cow *async_cow;
1111 int num_added = 0;
1112 async_cow = container_of(work, struct async_cow, work);
1113
1114 compress_file_range(async_cow->inode, async_cow->locked_page,
1115 async_cow->start, async_cow->end, async_cow,
1116 &num_added);
1117 if (num_added == 0) {
1118 btrfs_add_delayed_iput(async_cow->inode);
1119 async_cow->inode = NULL;
1120 }
1121 }
1122
1123 /*
1124 * work queue call back to submit previously compressed pages
1125 */
1126 static noinline void async_cow_submit(struct btrfs_work *work)
1127 {
1128 struct btrfs_fs_info *fs_info;
1129 struct async_cow *async_cow;
1130 struct btrfs_root *root;
1131 unsigned long nr_pages;
1132
1133 async_cow = container_of(work, struct async_cow, work);
1134
1135 root = async_cow->root;
1136 fs_info = root->fs_info;
1137 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1138 PAGE_SHIFT;
1139
1140 /*
1141 * atomic_sub_return implies a barrier for waitqueue_active
1142 */
1143 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1144 5 * SZ_1M &&
1145 waitqueue_active(&fs_info->async_submit_wait))
1146 wake_up(&fs_info->async_submit_wait);
1147
1148 if (async_cow->inode)
1149 submit_compressed_extents(async_cow->inode, async_cow);
1150 }
1151
1152 static noinline void async_cow_free(struct btrfs_work *work)
1153 {
1154 struct async_cow *async_cow;
1155 async_cow = container_of(work, struct async_cow, work);
1156 if (async_cow->inode)
1157 btrfs_add_delayed_iput(async_cow->inode);
1158 kfree(async_cow);
1159 }
1160
1161 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1162 u64 start, u64 end, int *page_started,
1163 unsigned long *nr_written)
1164 {
1165 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1166 struct async_cow *async_cow;
1167 struct btrfs_root *root = BTRFS_I(inode)->root;
1168 unsigned long nr_pages;
1169 u64 cur_end;
1170
1171 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1172 1, 0, NULL, GFP_NOFS);
1173 while (start < end) {
1174 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1175 BUG_ON(!async_cow); /* -ENOMEM */
1176 async_cow->inode = igrab(inode);
1177 async_cow->root = root;
1178 async_cow->locked_page = locked_page;
1179 async_cow->start = start;
1180
1181 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1182 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1183 cur_end = end;
1184 else
1185 cur_end = min(end, start + SZ_512K - 1);
1186
1187 async_cow->end = cur_end;
1188 INIT_LIST_HEAD(&async_cow->extents);
1189
1190 btrfs_init_work(&async_cow->work,
1191 btrfs_delalloc_helper,
1192 async_cow_start, async_cow_submit,
1193 async_cow_free);
1194
1195 nr_pages = (cur_end - start + PAGE_SIZE) >>
1196 PAGE_SHIFT;
1197 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1198
1199 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1200
1201 while (atomic_read(&fs_info->async_submit_draining) &&
1202 atomic_read(&fs_info->async_delalloc_pages)) {
1203 wait_event(fs_info->async_submit_wait,
1204 (atomic_read(&fs_info->async_delalloc_pages) ==
1205 0));
1206 }
1207
1208 *nr_written += nr_pages;
1209 start = cur_end + 1;
1210 }
1211 *page_started = 1;
1212 return 0;
1213 }
1214
1215 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1216 u64 bytenr, u64 num_bytes)
1217 {
1218 int ret;
1219 struct btrfs_ordered_sum *sums;
1220 LIST_HEAD(list);
1221
1222 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1223 bytenr + num_bytes - 1, &list, 0);
1224 if (ret == 0 && list_empty(&list))
1225 return 0;
1226
1227 while (!list_empty(&list)) {
1228 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1229 list_del(&sums->list);
1230 kfree(sums);
1231 }
1232 return 1;
1233 }
1234
1235 /*
1236 * when nowcow writeback call back. This checks for snapshots or COW copies
1237 * of the extents that exist in the file, and COWs the file as required.
1238 *
1239 * If no cow copies or snapshots exist, we write directly to the existing
1240 * blocks on disk
1241 */
1242 static noinline int run_delalloc_nocow(struct inode *inode,
1243 struct page *locked_page,
1244 u64 start, u64 end, int *page_started, int force,
1245 unsigned long *nr_written)
1246 {
1247 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1248 struct btrfs_root *root = BTRFS_I(inode)->root;
1249 struct btrfs_trans_handle *trans;
1250 struct extent_buffer *leaf;
1251 struct btrfs_path *path;
1252 struct btrfs_file_extent_item *fi;
1253 struct btrfs_key found_key;
1254 u64 cow_start;
1255 u64 cur_offset;
1256 u64 extent_end;
1257 u64 extent_offset;
1258 u64 disk_bytenr;
1259 u64 num_bytes;
1260 u64 disk_num_bytes;
1261 u64 ram_bytes;
1262 int extent_type;
1263 int ret, err;
1264 int type;
1265 int nocow;
1266 int check_prev = 1;
1267 bool nolock;
1268 u64 ino = btrfs_ino(BTRFS_I(inode));
1269
1270 path = btrfs_alloc_path();
1271 if (!path) {
1272 extent_clear_unlock_delalloc(inode, start, end, end,
1273 locked_page,
1274 EXTENT_LOCKED | EXTENT_DELALLOC |
1275 EXTENT_DO_ACCOUNTING |
1276 EXTENT_DEFRAG, PAGE_UNLOCK |
1277 PAGE_CLEAR_DIRTY |
1278 PAGE_SET_WRITEBACK |
1279 PAGE_END_WRITEBACK);
1280 return -ENOMEM;
1281 }
1282
1283 nolock = btrfs_is_free_space_inode(inode);
1284
1285 if (nolock)
1286 trans = btrfs_join_transaction_nolock(root);
1287 else
1288 trans = btrfs_join_transaction(root);
1289
1290 if (IS_ERR(trans)) {
1291 extent_clear_unlock_delalloc(inode, start, end, end,
1292 locked_page,
1293 EXTENT_LOCKED | EXTENT_DELALLOC |
1294 EXTENT_DO_ACCOUNTING |
1295 EXTENT_DEFRAG, PAGE_UNLOCK |
1296 PAGE_CLEAR_DIRTY |
1297 PAGE_SET_WRITEBACK |
1298 PAGE_END_WRITEBACK);
1299 btrfs_free_path(path);
1300 return PTR_ERR(trans);
1301 }
1302
1303 trans->block_rsv = &fs_info->delalloc_block_rsv;
1304
1305 cow_start = (u64)-1;
1306 cur_offset = start;
1307 while (1) {
1308 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1309 cur_offset, 0);
1310 if (ret < 0)
1311 goto error;
1312 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1313 leaf = path->nodes[0];
1314 btrfs_item_key_to_cpu(leaf, &found_key,
1315 path->slots[0] - 1);
1316 if (found_key.objectid == ino &&
1317 found_key.type == BTRFS_EXTENT_DATA_KEY)
1318 path->slots[0]--;
1319 }
1320 check_prev = 0;
1321 next_slot:
1322 leaf = path->nodes[0];
1323 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1324 ret = btrfs_next_leaf(root, path);
1325 if (ret < 0)
1326 goto error;
1327 if (ret > 0)
1328 break;
1329 leaf = path->nodes[0];
1330 }
1331
1332 nocow = 0;
1333 disk_bytenr = 0;
1334 num_bytes = 0;
1335 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1336
1337 if (found_key.objectid > ino)
1338 break;
1339 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1340 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1341 path->slots[0]++;
1342 goto next_slot;
1343 }
1344 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1345 found_key.offset > end)
1346 break;
1347
1348 if (found_key.offset > cur_offset) {
1349 extent_end = found_key.offset;
1350 extent_type = 0;
1351 goto out_check;
1352 }
1353
1354 fi = btrfs_item_ptr(leaf, path->slots[0],
1355 struct btrfs_file_extent_item);
1356 extent_type = btrfs_file_extent_type(leaf, fi);
1357
1358 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1359 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1360 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1361 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1362 extent_offset = btrfs_file_extent_offset(leaf, fi);
1363 extent_end = found_key.offset +
1364 btrfs_file_extent_num_bytes(leaf, fi);
1365 disk_num_bytes =
1366 btrfs_file_extent_disk_num_bytes(leaf, fi);
1367 if (extent_end <= start) {
1368 path->slots[0]++;
1369 goto next_slot;
1370 }
1371 if (disk_bytenr == 0)
1372 goto out_check;
1373 if (btrfs_file_extent_compression(leaf, fi) ||
1374 btrfs_file_extent_encryption(leaf, fi) ||
1375 btrfs_file_extent_other_encoding(leaf, fi))
1376 goto out_check;
1377 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1378 goto out_check;
1379 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1380 goto out_check;
1381 if (btrfs_cross_ref_exist(trans, root, ino,
1382 found_key.offset -
1383 extent_offset, disk_bytenr))
1384 goto out_check;
1385 disk_bytenr += extent_offset;
1386 disk_bytenr += cur_offset - found_key.offset;
1387 num_bytes = min(end + 1, extent_end) - cur_offset;
1388 /*
1389 * if there are pending snapshots for this root,
1390 * we fall into common COW way.
1391 */
1392 if (!nolock) {
1393 err = btrfs_start_write_no_snapshoting(root);
1394 if (!err)
1395 goto out_check;
1396 }
1397 /*
1398 * force cow if csum exists in the range.
1399 * this ensure that csum for a given extent are
1400 * either valid or do not exist.
1401 */
1402 if (csum_exist_in_range(fs_info, disk_bytenr,
1403 num_bytes))
1404 goto out_check;
1405 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1406 goto out_check;
1407 nocow = 1;
1408 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1409 extent_end = found_key.offset +
1410 btrfs_file_extent_inline_len(leaf,
1411 path->slots[0], fi);
1412 extent_end = ALIGN(extent_end,
1413 fs_info->sectorsize);
1414 } else {
1415 BUG_ON(1);
1416 }
1417 out_check:
1418 if (extent_end <= start) {
1419 path->slots[0]++;
1420 if (!nolock && nocow)
1421 btrfs_end_write_no_snapshoting(root);
1422 if (nocow)
1423 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1424 goto next_slot;
1425 }
1426 if (!nocow) {
1427 if (cow_start == (u64)-1)
1428 cow_start = cur_offset;
1429 cur_offset = extent_end;
1430 if (cur_offset > end)
1431 break;
1432 path->slots[0]++;
1433 goto next_slot;
1434 }
1435
1436 btrfs_release_path(path);
1437 if (cow_start != (u64)-1) {
1438 ret = cow_file_range(inode, locked_page,
1439 cow_start, found_key.offset - 1,
1440 end, page_started, nr_written, 1,
1441 NULL);
1442 if (ret) {
1443 if (!nolock && nocow)
1444 btrfs_end_write_no_snapshoting(root);
1445 if (nocow)
1446 btrfs_dec_nocow_writers(fs_info,
1447 disk_bytenr);
1448 goto error;
1449 }
1450 cow_start = (u64)-1;
1451 }
1452
1453 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1454 struct extent_map *em;
1455 struct extent_map_tree *em_tree;
1456 em_tree = &BTRFS_I(inode)->extent_tree;
1457 em = alloc_extent_map();
1458 BUG_ON(!em); /* -ENOMEM */
1459 em->start = cur_offset;
1460 em->orig_start = found_key.offset - extent_offset;
1461 em->len = num_bytes;
1462 em->block_len = num_bytes;
1463 em->block_start = disk_bytenr;
1464 em->orig_block_len = disk_num_bytes;
1465 em->ram_bytes = ram_bytes;
1466 em->bdev = fs_info->fs_devices->latest_bdev;
1467 em->mod_start = em->start;
1468 em->mod_len = em->len;
1469 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1470 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1471 em->generation = -1;
1472 while (1) {
1473 write_lock(&em_tree->lock);
1474 ret = add_extent_mapping(em_tree, em, 1);
1475 write_unlock(&em_tree->lock);
1476 if (ret != -EEXIST) {
1477 free_extent_map(em);
1478 break;
1479 }
1480 btrfs_drop_extent_cache(inode, em->start,
1481 em->start + em->len - 1, 0);
1482 }
1483 type = BTRFS_ORDERED_PREALLOC;
1484 } else {
1485 type = BTRFS_ORDERED_NOCOW;
1486 }
1487
1488 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1489 num_bytes, num_bytes, type);
1490 if (nocow)
1491 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1492 BUG_ON(ret); /* -ENOMEM */
1493
1494 if (root->root_key.objectid ==
1495 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1496 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1497 num_bytes);
1498 if (ret) {
1499 if (!nolock && nocow)
1500 btrfs_end_write_no_snapshoting(root);
1501 goto error;
1502 }
1503 }
1504
1505 extent_clear_unlock_delalloc(inode, cur_offset,
1506 cur_offset + num_bytes - 1, end,
1507 locked_page, EXTENT_LOCKED |
1508 EXTENT_DELALLOC |
1509 EXTENT_CLEAR_DATA_RESV,
1510 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1511
1512 if (!nolock && nocow)
1513 btrfs_end_write_no_snapshoting(root);
1514 cur_offset = extent_end;
1515 if (cur_offset > end)
1516 break;
1517 }
1518 btrfs_release_path(path);
1519
1520 if (cur_offset <= end && cow_start == (u64)-1) {
1521 cow_start = cur_offset;
1522 cur_offset = end;
1523 }
1524
1525 if (cow_start != (u64)-1) {
1526 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1527 page_started, nr_written, 1, NULL);
1528 if (ret)
1529 goto error;
1530 }
1531
1532 error:
1533 err = btrfs_end_transaction(trans);
1534 if (!ret)
1535 ret = err;
1536
1537 if (ret && cur_offset < end)
1538 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1539 locked_page, EXTENT_LOCKED |
1540 EXTENT_DELALLOC | EXTENT_DEFRAG |
1541 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1542 PAGE_CLEAR_DIRTY |
1543 PAGE_SET_WRITEBACK |
1544 PAGE_END_WRITEBACK);
1545 btrfs_free_path(path);
1546 return ret;
1547 }
1548
1549 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1550 {
1551
1552 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1553 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1554 return 0;
1555
1556 /*
1557 * @defrag_bytes is a hint value, no spinlock held here,
1558 * if is not zero, it means the file is defragging.
1559 * Force cow if given extent needs to be defragged.
1560 */
1561 if (BTRFS_I(inode)->defrag_bytes &&
1562 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1563 EXTENT_DEFRAG, 0, NULL))
1564 return 1;
1565
1566 return 0;
1567 }
1568
1569 /*
1570 * extent_io.c call back to do delayed allocation processing
1571 */
1572 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1573 u64 start, u64 end, int *page_started,
1574 unsigned long *nr_written)
1575 {
1576 int ret;
1577 int force_cow = need_force_cow(inode, start, end);
1578
1579 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1580 ret = run_delalloc_nocow(inode, locked_page, start, end,
1581 page_started, 1, nr_written);
1582 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1583 ret = run_delalloc_nocow(inode, locked_page, start, end,
1584 page_started, 0, nr_written);
1585 } else if (!inode_need_compress(inode)) {
1586 ret = cow_file_range(inode, locked_page, start, end, end,
1587 page_started, nr_written, 1, NULL);
1588 } else {
1589 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1590 &BTRFS_I(inode)->runtime_flags);
1591 ret = cow_file_range_async(inode, locked_page, start, end,
1592 page_started, nr_written);
1593 }
1594 return ret;
1595 }
1596
1597 static void btrfs_split_extent_hook(struct inode *inode,
1598 struct extent_state *orig, u64 split)
1599 {
1600 u64 size;
1601
1602 /* not delalloc, ignore it */
1603 if (!(orig->state & EXTENT_DELALLOC))
1604 return;
1605
1606 size = orig->end - orig->start + 1;
1607 if (size > BTRFS_MAX_EXTENT_SIZE) {
1608 u32 num_extents;
1609 u64 new_size;
1610
1611 /*
1612 * See the explanation in btrfs_merge_extent_hook, the same
1613 * applies here, just in reverse.
1614 */
1615 new_size = orig->end - split + 1;
1616 num_extents = count_max_extents(new_size);
1617 new_size = split - orig->start;
1618 num_extents += count_max_extents(new_size);
1619 if (count_max_extents(size) >= num_extents)
1620 return;
1621 }
1622
1623 spin_lock(&BTRFS_I(inode)->lock);
1624 BTRFS_I(inode)->outstanding_extents++;
1625 spin_unlock(&BTRFS_I(inode)->lock);
1626 }
1627
1628 /*
1629 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1630 * extents so we can keep track of new extents that are just merged onto old
1631 * extents, such as when we are doing sequential writes, so we can properly
1632 * account for the metadata space we'll need.
1633 */
1634 static void btrfs_merge_extent_hook(struct inode *inode,
1635 struct extent_state *new,
1636 struct extent_state *other)
1637 {
1638 u64 new_size, old_size;
1639 u32 num_extents;
1640
1641 /* not delalloc, ignore it */
1642 if (!(other->state & EXTENT_DELALLOC))
1643 return;
1644
1645 if (new->start > other->start)
1646 new_size = new->end - other->start + 1;
1647 else
1648 new_size = other->end - new->start + 1;
1649
1650 /* we're not bigger than the max, unreserve the space and go */
1651 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1652 spin_lock(&BTRFS_I(inode)->lock);
1653 BTRFS_I(inode)->outstanding_extents--;
1654 spin_unlock(&BTRFS_I(inode)->lock);
1655 return;
1656 }
1657
1658 /*
1659 * We have to add up either side to figure out how many extents were
1660 * accounted for before we merged into one big extent. If the number of
1661 * extents we accounted for is <= the amount we need for the new range
1662 * then we can return, otherwise drop. Think of it like this
1663 *
1664 * [ 4k][MAX_SIZE]
1665 *
1666 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1667 * need 2 outstanding extents, on one side we have 1 and the other side
1668 * we have 1 so they are == and we can return. But in this case
1669 *
1670 * [MAX_SIZE+4k][MAX_SIZE+4k]
1671 *
1672 * Each range on their own accounts for 2 extents, but merged together
1673 * they are only 3 extents worth of accounting, so we need to drop in
1674 * this case.
1675 */
1676 old_size = other->end - other->start + 1;
1677 num_extents = count_max_extents(old_size);
1678 old_size = new->end - new->start + 1;
1679 num_extents += count_max_extents(old_size);
1680 if (count_max_extents(new_size) >= num_extents)
1681 return;
1682
1683 spin_lock(&BTRFS_I(inode)->lock);
1684 BTRFS_I(inode)->outstanding_extents--;
1685 spin_unlock(&BTRFS_I(inode)->lock);
1686 }
1687
1688 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1689 struct inode *inode)
1690 {
1691 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1692
1693 spin_lock(&root->delalloc_lock);
1694 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1695 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1696 &root->delalloc_inodes);
1697 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1698 &BTRFS_I(inode)->runtime_flags);
1699 root->nr_delalloc_inodes++;
1700 if (root->nr_delalloc_inodes == 1) {
1701 spin_lock(&fs_info->delalloc_root_lock);
1702 BUG_ON(!list_empty(&root->delalloc_root));
1703 list_add_tail(&root->delalloc_root,
1704 &fs_info->delalloc_roots);
1705 spin_unlock(&fs_info->delalloc_root_lock);
1706 }
1707 }
1708 spin_unlock(&root->delalloc_lock);
1709 }
1710
1711 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1712 struct inode *inode)
1713 {
1714 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1715
1716 spin_lock(&root->delalloc_lock);
1717 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1718 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1719 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1720 &BTRFS_I(inode)->runtime_flags);
1721 root->nr_delalloc_inodes--;
1722 if (!root->nr_delalloc_inodes) {
1723 spin_lock(&fs_info->delalloc_root_lock);
1724 BUG_ON(list_empty(&root->delalloc_root));
1725 list_del_init(&root->delalloc_root);
1726 spin_unlock(&fs_info->delalloc_root_lock);
1727 }
1728 }
1729 spin_unlock(&root->delalloc_lock);
1730 }
1731
1732 /*
1733 * extent_io.c set_bit_hook, used to track delayed allocation
1734 * bytes in this file, and to maintain the list of inodes that
1735 * have pending delalloc work to be done.
1736 */
1737 static void btrfs_set_bit_hook(struct inode *inode,
1738 struct extent_state *state, unsigned *bits)
1739 {
1740
1741 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1742
1743 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1744 WARN_ON(1);
1745 /*
1746 * set_bit and clear bit hooks normally require _irqsave/restore
1747 * but in this case, we are only testing for the DELALLOC
1748 * bit, which is only set or cleared with irqs on
1749 */
1750 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1751 struct btrfs_root *root = BTRFS_I(inode)->root;
1752 u64 len = state->end + 1 - state->start;
1753 bool do_list = !btrfs_is_free_space_inode(inode);
1754
1755 if (*bits & EXTENT_FIRST_DELALLOC) {
1756 *bits &= ~EXTENT_FIRST_DELALLOC;
1757 } else {
1758 spin_lock(&BTRFS_I(inode)->lock);
1759 BTRFS_I(inode)->outstanding_extents++;
1760 spin_unlock(&BTRFS_I(inode)->lock);
1761 }
1762
1763 /* For sanity tests */
1764 if (btrfs_is_testing(fs_info))
1765 return;
1766
1767 __percpu_counter_add(&fs_info->delalloc_bytes, len,
1768 fs_info->delalloc_batch);
1769 spin_lock(&BTRFS_I(inode)->lock);
1770 BTRFS_I(inode)->delalloc_bytes += len;
1771 if (*bits & EXTENT_DEFRAG)
1772 BTRFS_I(inode)->defrag_bytes += len;
1773 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1774 &BTRFS_I(inode)->runtime_flags))
1775 btrfs_add_delalloc_inodes(root, inode);
1776 spin_unlock(&BTRFS_I(inode)->lock);
1777 }
1778 }
1779
1780 /*
1781 * extent_io.c clear_bit_hook, see set_bit_hook for why
1782 */
1783 static void btrfs_clear_bit_hook(struct inode *inode,
1784 struct extent_state *state,
1785 unsigned *bits)
1786 {
1787 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1788 u64 len = state->end + 1 - state->start;
1789 u32 num_extents = count_max_extents(len);
1790
1791 spin_lock(&BTRFS_I(inode)->lock);
1792 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1793 BTRFS_I(inode)->defrag_bytes -= len;
1794 spin_unlock(&BTRFS_I(inode)->lock);
1795
1796 /*
1797 * set_bit and clear bit hooks normally require _irqsave/restore
1798 * but in this case, we are only testing for the DELALLOC
1799 * bit, which is only set or cleared with irqs on
1800 */
1801 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1802 struct btrfs_root *root = BTRFS_I(inode)->root;
1803 bool do_list = !btrfs_is_free_space_inode(inode);
1804
1805 if (*bits & EXTENT_FIRST_DELALLOC) {
1806 *bits &= ~EXTENT_FIRST_DELALLOC;
1807 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1808 spin_lock(&BTRFS_I(inode)->lock);
1809 BTRFS_I(inode)->outstanding_extents -= num_extents;
1810 spin_unlock(&BTRFS_I(inode)->lock);
1811 }
1812
1813 /*
1814 * We don't reserve metadata space for space cache inodes so we
1815 * don't need to call dellalloc_release_metadata if there is an
1816 * error.
1817 */
1818 if (*bits & EXTENT_DO_ACCOUNTING &&
1819 root != fs_info->tree_root)
1820 btrfs_delalloc_release_metadata(inode, len);
1821
1822 /* For sanity tests. */
1823 if (btrfs_is_testing(fs_info))
1824 return;
1825
1826 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1827 && do_list && !(state->state & EXTENT_NORESERVE)
1828 && (*bits & (EXTENT_DO_ACCOUNTING |
1829 EXTENT_CLEAR_DATA_RESV)))
1830 btrfs_free_reserved_data_space_noquota(inode,
1831 state->start, len);
1832
1833 __percpu_counter_add(&fs_info->delalloc_bytes, -len,
1834 fs_info->delalloc_batch);
1835 spin_lock(&BTRFS_I(inode)->lock);
1836 BTRFS_I(inode)->delalloc_bytes -= len;
1837 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1838 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1839 &BTRFS_I(inode)->runtime_flags))
1840 btrfs_del_delalloc_inode(root, inode);
1841 spin_unlock(&BTRFS_I(inode)->lock);
1842 }
1843 }
1844
1845 /*
1846 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1847 * we don't create bios that span stripes or chunks
1848 *
1849 * return 1 if page cannot be merged to bio
1850 * return 0 if page can be merged to bio
1851 * return error otherwise
1852 */
1853 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1854 size_t size, struct bio *bio,
1855 unsigned long bio_flags)
1856 {
1857 struct inode *inode = page->mapping->host;
1858 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1859 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1860 u64 length = 0;
1861 u64 map_length;
1862 int ret;
1863
1864 if (bio_flags & EXTENT_BIO_COMPRESSED)
1865 return 0;
1866
1867 length = bio->bi_iter.bi_size;
1868 map_length = length;
1869 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1870 NULL, 0);
1871 if (ret < 0)
1872 return ret;
1873 if (map_length < length + size)
1874 return 1;
1875 return 0;
1876 }
1877
1878 /*
1879 * in order to insert checksums into the metadata in large chunks,
1880 * we wait until bio submission time. All the pages in the bio are
1881 * checksummed and sums are attached onto the ordered extent record.
1882 *
1883 * At IO completion time the cums attached on the ordered extent record
1884 * are inserted into the btree
1885 */
1886 static int __btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
1887 int mirror_num, unsigned long bio_flags,
1888 u64 bio_offset)
1889 {
1890 int ret = 0;
1891
1892 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1893 BUG_ON(ret); /* -ENOMEM */
1894 return 0;
1895 }
1896
1897 /*
1898 * in order to insert checksums into the metadata in large chunks,
1899 * we wait until bio submission time. All the pages in the bio are
1900 * checksummed and sums are attached onto the ordered extent record.
1901 *
1902 * At IO completion time the cums attached on the ordered extent record
1903 * are inserted into the btree
1904 */
1905 static int __btrfs_submit_bio_done(struct inode *inode, struct bio *bio,
1906 int mirror_num, unsigned long bio_flags,
1907 u64 bio_offset)
1908 {
1909 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1910 int ret;
1911
1912 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1913 if (ret) {
1914 bio->bi_error = ret;
1915 bio_endio(bio);
1916 }
1917 return ret;
1918 }
1919
1920 /*
1921 * extent_io.c submission hook. This does the right thing for csum calculation
1922 * on write, or reading the csums from the tree before a read
1923 */
1924 static int btrfs_submit_bio_hook(struct inode *inode, struct bio *bio,
1925 int mirror_num, unsigned long bio_flags,
1926 u64 bio_offset)
1927 {
1928 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1929 struct btrfs_root *root = BTRFS_I(inode)->root;
1930 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1931 int ret = 0;
1932 int skip_sum;
1933 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1934
1935 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1936
1937 if (btrfs_is_free_space_inode(inode))
1938 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1939
1940 if (bio_op(bio) != REQ_OP_WRITE) {
1941 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1942 if (ret)
1943 goto out;
1944
1945 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1946 ret = btrfs_submit_compressed_read(inode, bio,
1947 mirror_num,
1948 bio_flags);
1949 goto out;
1950 } else if (!skip_sum) {
1951 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
1952 if (ret)
1953 goto out;
1954 }
1955 goto mapit;
1956 } else if (async && !skip_sum) {
1957 /* csum items have already been cloned */
1958 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1959 goto mapit;
1960 /* we're doing a write, do the async checksumming */
1961 ret = btrfs_wq_submit_bio(fs_info, inode, bio, mirror_num,
1962 bio_flags, bio_offset,
1963 __btrfs_submit_bio_start,
1964 __btrfs_submit_bio_done);
1965 goto out;
1966 } else if (!skip_sum) {
1967 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1968 if (ret)
1969 goto out;
1970 }
1971
1972 mapit:
1973 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1974
1975 out:
1976 if (ret < 0) {
1977 bio->bi_error = ret;
1978 bio_endio(bio);
1979 }
1980 return ret;
1981 }
1982
1983 /*
1984 * given a list of ordered sums record them in the inode. This happens
1985 * at IO completion time based on sums calculated at bio submission time.
1986 */
1987 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1988 struct inode *inode, u64 file_offset,
1989 struct list_head *list)
1990 {
1991 struct btrfs_ordered_sum *sum;
1992
1993 list_for_each_entry(sum, list, list) {
1994 trans->adding_csums = 1;
1995 btrfs_csum_file_blocks(trans,
1996 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1997 trans->adding_csums = 0;
1998 }
1999 return 0;
2000 }
2001
2002 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2003 struct extent_state **cached_state, int dedupe)
2004 {
2005 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2006 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2007 cached_state);
2008 }
2009
2010 /* see btrfs_writepage_start_hook for details on why this is required */
2011 struct btrfs_writepage_fixup {
2012 struct page *page;
2013 struct btrfs_work work;
2014 };
2015
2016 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2017 {
2018 struct btrfs_writepage_fixup *fixup;
2019 struct btrfs_ordered_extent *ordered;
2020 struct extent_state *cached_state = NULL;
2021 struct page *page;
2022 struct inode *inode;
2023 u64 page_start;
2024 u64 page_end;
2025 int ret;
2026
2027 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2028 page = fixup->page;
2029 again:
2030 lock_page(page);
2031 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2032 ClearPageChecked(page);
2033 goto out_page;
2034 }
2035
2036 inode = page->mapping->host;
2037 page_start = page_offset(page);
2038 page_end = page_offset(page) + PAGE_SIZE - 1;
2039
2040 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2041 &cached_state);
2042
2043 /* already ordered? We're done */
2044 if (PagePrivate2(page))
2045 goto out;
2046
2047 ordered = btrfs_lookup_ordered_range(inode, page_start,
2048 PAGE_SIZE);
2049 if (ordered) {
2050 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2051 page_end, &cached_state, GFP_NOFS);
2052 unlock_page(page);
2053 btrfs_start_ordered_extent(inode, ordered, 1);
2054 btrfs_put_ordered_extent(ordered);
2055 goto again;
2056 }
2057
2058 ret = btrfs_delalloc_reserve_space(inode, page_start,
2059 PAGE_SIZE);
2060 if (ret) {
2061 mapping_set_error(page->mapping, ret);
2062 end_extent_writepage(page, ret, page_start, page_end);
2063 ClearPageChecked(page);
2064 goto out;
2065 }
2066
2067 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state,
2068 0);
2069 ClearPageChecked(page);
2070 set_page_dirty(page);
2071 out:
2072 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2073 &cached_state, GFP_NOFS);
2074 out_page:
2075 unlock_page(page);
2076 put_page(page);
2077 kfree(fixup);
2078 }
2079
2080 /*
2081 * There are a few paths in the higher layers of the kernel that directly
2082 * set the page dirty bit without asking the filesystem if it is a
2083 * good idea. This causes problems because we want to make sure COW
2084 * properly happens and the data=ordered rules are followed.
2085 *
2086 * In our case any range that doesn't have the ORDERED bit set
2087 * hasn't been properly setup for IO. We kick off an async process
2088 * to fix it up. The async helper will wait for ordered extents, set
2089 * the delalloc bit and make it safe to write the page.
2090 */
2091 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2092 {
2093 struct inode *inode = page->mapping->host;
2094 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2095 struct btrfs_writepage_fixup *fixup;
2096
2097 /* this page is properly in the ordered list */
2098 if (TestClearPagePrivate2(page))
2099 return 0;
2100
2101 if (PageChecked(page))
2102 return -EAGAIN;
2103
2104 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2105 if (!fixup)
2106 return -EAGAIN;
2107
2108 SetPageChecked(page);
2109 get_page(page);
2110 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2111 btrfs_writepage_fixup_worker, NULL, NULL);
2112 fixup->page = page;
2113 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2114 return -EBUSY;
2115 }
2116
2117 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2118 struct inode *inode, u64 file_pos,
2119 u64 disk_bytenr, u64 disk_num_bytes,
2120 u64 num_bytes, u64 ram_bytes,
2121 u8 compression, u8 encryption,
2122 u16 other_encoding, int extent_type)
2123 {
2124 struct btrfs_root *root = BTRFS_I(inode)->root;
2125 struct btrfs_file_extent_item *fi;
2126 struct btrfs_path *path;
2127 struct extent_buffer *leaf;
2128 struct btrfs_key ins;
2129 int extent_inserted = 0;
2130 int ret;
2131
2132 path = btrfs_alloc_path();
2133 if (!path)
2134 return -ENOMEM;
2135
2136 /*
2137 * we may be replacing one extent in the tree with another.
2138 * The new extent is pinned in the extent map, and we don't want
2139 * to drop it from the cache until it is completely in the btree.
2140 *
2141 * So, tell btrfs_drop_extents to leave this extent in the cache.
2142 * the caller is expected to unpin it and allow it to be merged
2143 * with the others.
2144 */
2145 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2146 file_pos + num_bytes, NULL, 0,
2147 1, sizeof(*fi), &extent_inserted);
2148 if (ret)
2149 goto out;
2150
2151 if (!extent_inserted) {
2152 ins.objectid = btrfs_ino(BTRFS_I(inode));
2153 ins.offset = file_pos;
2154 ins.type = BTRFS_EXTENT_DATA_KEY;
2155
2156 path->leave_spinning = 1;
2157 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2158 sizeof(*fi));
2159 if (ret)
2160 goto out;
2161 }
2162 leaf = path->nodes[0];
2163 fi = btrfs_item_ptr(leaf, path->slots[0],
2164 struct btrfs_file_extent_item);
2165 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2166 btrfs_set_file_extent_type(leaf, fi, extent_type);
2167 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2168 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2169 btrfs_set_file_extent_offset(leaf, fi, 0);
2170 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2171 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2172 btrfs_set_file_extent_compression(leaf, fi, compression);
2173 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2174 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2175
2176 btrfs_mark_buffer_dirty(leaf);
2177 btrfs_release_path(path);
2178
2179 inode_add_bytes(inode, num_bytes);
2180
2181 ins.objectid = disk_bytenr;
2182 ins.offset = disk_num_bytes;
2183 ins.type = BTRFS_EXTENT_ITEM_KEY;
2184 ret = btrfs_alloc_reserved_file_extent(trans, root->root_key.objectid,
2185 btrfs_ino(BTRFS_I(inode)), file_pos,
2186 ram_bytes, &ins);
2187 /*
2188 * Release the reserved range from inode dirty range map, as it is
2189 * already moved into delayed_ref_head
2190 */
2191 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2192 out:
2193 btrfs_free_path(path);
2194
2195 return ret;
2196 }
2197
2198 /* snapshot-aware defrag */
2199 struct sa_defrag_extent_backref {
2200 struct rb_node node;
2201 struct old_sa_defrag_extent *old;
2202 u64 root_id;
2203 u64 inum;
2204 u64 file_pos;
2205 u64 extent_offset;
2206 u64 num_bytes;
2207 u64 generation;
2208 };
2209
2210 struct old_sa_defrag_extent {
2211 struct list_head list;
2212 struct new_sa_defrag_extent *new;
2213
2214 u64 extent_offset;
2215 u64 bytenr;
2216 u64 offset;
2217 u64 len;
2218 int count;
2219 };
2220
2221 struct new_sa_defrag_extent {
2222 struct rb_root root;
2223 struct list_head head;
2224 struct btrfs_path *path;
2225 struct inode *inode;
2226 u64 file_pos;
2227 u64 len;
2228 u64 bytenr;
2229 u64 disk_len;
2230 u8 compress_type;
2231 };
2232
2233 static int backref_comp(struct sa_defrag_extent_backref *b1,
2234 struct sa_defrag_extent_backref *b2)
2235 {
2236 if (b1->root_id < b2->root_id)
2237 return -1;
2238 else if (b1->root_id > b2->root_id)
2239 return 1;
2240
2241 if (b1->inum < b2->inum)
2242 return -1;
2243 else if (b1->inum > b2->inum)
2244 return 1;
2245
2246 if (b1->file_pos < b2->file_pos)
2247 return -1;
2248 else if (b1->file_pos > b2->file_pos)
2249 return 1;
2250
2251 /*
2252 * [------------------------------] ===> (a range of space)
2253 * |<--->| |<---->| =============> (fs/file tree A)
2254 * |<---------------------------->| ===> (fs/file tree B)
2255 *
2256 * A range of space can refer to two file extents in one tree while
2257 * refer to only one file extent in another tree.
2258 *
2259 * So we may process a disk offset more than one time(two extents in A)
2260 * and locate at the same extent(one extent in B), then insert two same
2261 * backrefs(both refer to the extent in B).
2262 */
2263 return 0;
2264 }
2265
2266 static void backref_insert(struct rb_root *root,
2267 struct sa_defrag_extent_backref *backref)
2268 {
2269 struct rb_node **p = &root->rb_node;
2270 struct rb_node *parent = NULL;
2271 struct sa_defrag_extent_backref *entry;
2272 int ret;
2273
2274 while (*p) {
2275 parent = *p;
2276 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2277
2278 ret = backref_comp(backref, entry);
2279 if (ret < 0)
2280 p = &(*p)->rb_left;
2281 else
2282 p = &(*p)->rb_right;
2283 }
2284
2285 rb_link_node(&backref->node, parent, p);
2286 rb_insert_color(&backref->node, root);
2287 }
2288
2289 /*
2290 * Note the backref might has changed, and in this case we just return 0.
2291 */
2292 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2293 void *ctx)
2294 {
2295 struct btrfs_file_extent_item *extent;
2296 struct old_sa_defrag_extent *old = ctx;
2297 struct new_sa_defrag_extent *new = old->new;
2298 struct btrfs_path *path = new->path;
2299 struct btrfs_key key;
2300 struct btrfs_root *root;
2301 struct sa_defrag_extent_backref *backref;
2302 struct extent_buffer *leaf;
2303 struct inode *inode = new->inode;
2304 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2305 int slot;
2306 int ret;
2307 u64 extent_offset;
2308 u64 num_bytes;
2309
2310 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2311 inum == btrfs_ino(BTRFS_I(inode)))
2312 return 0;
2313
2314 key.objectid = root_id;
2315 key.type = BTRFS_ROOT_ITEM_KEY;
2316 key.offset = (u64)-1;
2317
2318 root = btrfs_read_fs_root_no_name(fs_info, &key);
2319 if (IS_ERR(root)) {
2320 if (PTR_ERR(root) == -ENOENT)
2321 return 0;
2322 WARN_ON(1);
2323 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2324 inum, offset, root_id);
2325 return PTR_ERR(root);
2326 }
2327
2328 key.objectid = inum;
2329 key.type = BTRFS_EXTENT_DATA_KEY;
2330 if (offset > (u64)-1 << 32)
2331 key.offset = 0;
2332 else
2333 key.offset = offset;
2334
2335 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2336 if (WARN_ON(ret < 0))
2337 return ret;
2338 ret = 0;
2339
2340 while (1) {
2341 cond_resched();
2342
2343 leaf = path->nodes[0];
2344 slot = path->slots[0];
2345
2346 if (slot >= btrfs_header_nritems(leaf)) {
2347 ret = btrfs_next_leaf(root, path);
2348 if (ret < 0) {
2349 goto out;
2350 } else if (ret > 0) {
2351 ret = 0;
2352 goto out;
2353 }
2354 continue;
2355 }
2356
2357 path->slots[0]++;
2358
2359 btrfs_item_key_to_cpu(leaf, &key, slot);
2360
2361 if (key.objectid > inum)
2362 goto out;
2363
2364 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2365 continue;
2366
2367 extent = btrfs_item_ptr(leaf, slot,
2368 struct btrfs_file_extent_item);
2369
2370 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2371 continue;
2372
2373 /*
2374 * 'offset' refers to the exact key.offset,
2375 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2376 * (key.offset - extent_offset).
2377 */
2378 if (key.offset != offset)
2379 continue;
2380
2381 extent_offset = btrfs_file_extent_offset(leaf, extent);
2382 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2383
2384 if (extent_offset >= old->extent_offset + old->offset +
2385 old->len || extent_offset + num_bytes <=
2386 old->extent_offset + old->offset)
2387 continue;
2388 break;
2389 }
2390
2391 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2392 if (!backref) {
2393 ret = -ENOENT;
2394 goto out;
2395 }
2396
2397 backref->root_id = root_id;
2398 backref->inum = inum;
2399 backref->file_pos = offset;
2400 backref->num_bytes = num_bytes;
2401 backref->extent_offset = extent_offset;
2402 backref->generation = btrfs_file_extent_generation(leaf, extent);
2403 backref->old = old;
2404 backref_insert(&new->root, backref);
2405 old->count++;
2406 out:
2407 btrfs_release_path(path);
2408 WARN_ON(ret);
2409 return ret;
2410 }
2411
2412 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2413 struct new_sa_defrag_extent *new)
2414 {
2415 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2416 struct old_sa_defrag_extent *old, *tmp;
2417 int ret;
2418
2419 new->path = path;
2420
2421 list_for_each_entry_safe(old, tmp, &new->head, list) {
2422 ret = iterate_inodes_from_logical(old->bytenr +
2423 old->extent_offset, fs_info,
2424 path, record_one_backref,
2425 old);
2426 if (ret < 0 && ret != -ENOENT)
2427 return false;
2428
2429 /* no backref to be processed for this extent */
2430 if (!old->count) {
2431 list_del(&old->list);
2432 kfree(old);
2433 }
2434 }
2435
2436 if (list_empty(&new->head))
2437 return false;
2438
2439 return true;
2440 }
2441
2442 static int relink_is_mergable(struct extent_buffer *leaf,
2443 struct btrfs_file_extent_item *fi,
2444 struct new_sa_defrag_extent *new)
2445 {
2446 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2447 return 0;
2448
2449 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2450 return 0;
2451
2452 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2453 return 0;
2454
2455 if (btrfs_file_extent_encryption(leaf, fi) ||
2456 btrfs_file_extent_other_encoding(leaf, fi))
2457 return 0;
2458
2459 return 1;
2460 }
2461
2462 /*
2463 * Note the backref might has changed, and in this case we just return 0.
2464 */
2465 static noinline int relink_extent_backref(struct btrfs_path *path,
2466 struct sa_defrag_extent_backref *prev,
2467 struct sa_defrag_extent_backref *backref)
2468 {
2469 struct btrfs_file_extent_item *extent;
2470 struct btrfs_file_extent_item *item;
2471 struct btrfs_ordered_extent *ordered;
2472 struct btrfs_trans_handle *trans;
2473 struct btrfs_root *root;
2474 struct btrfs_key key;
2475 struct extent_buffer *leaf;
2476 struct old_sa_defrag_extent *old = backref->old;
2477 struct new_sa_defrag_extent *new = old->new;
2478 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2479 struct inode *inode;
2480 struct extent_state *cached = NULL;
2481 int ret = 0;
2482 u64 start;
2483 u64 len;
2484 u64 lock_start;
2485 u64 lock_end;
2486 bool merge = false;
2487 int index;
2488
2489 if (prev && prev->root_id == backref->root_id &&
2490 prev->inum == backref->inum &&
2491 prev->file_pos + prev->num_bytes == backref->file_pos)
2492 merge = true;
2493
2494 /* step 1: get root */
2495 key.objectid = backref->root_id;
2496 key.type = BTRFS_ROOT_ITEM_KEY;
2497 key.offset = (u64)-1;
2498
2499 index = srcu_read_lock(&fs_info->subvol_srcu);
2500
2501 root = btrfs_read_fs_root_no_name(fs_info, &key);
2502 if (IS_ERR(root)) {
2503 srcu_read_unlock(&fs_info->subvol_srcu, index);
2504 if (PTR_ERR(root) == -ENOENT)
2505 return 0;
2506 return PTR_ERR(root);
2507 }
2508
2509 if (btrfs_root_readonly(root)) {
2510 srcu_read_unlock(&fs_info->subvol_srcu, index);
2511 return 0;
2512 }
2513
2514 /* step 2: get inode */
2515 key.objectid = backref->inum;
2516 key.type = BTRFS_INODE_ITEM_KEY;
2517 key.offset = 0;
2518
2519 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2520 if (IS_ERR(inode)) {
2521 srcu_read_unlock(&fs_info->subvol_srcu, index);
2522 return 0;
2523 }
2524
2525 srcu_read_unlock(&fs_info->subvol_srcu, index);
2526
2527 /* step 3: relink backref */
2528 lock_start = backref->file_pos;
2529 lock_end = backref->file_pos + backref->num_bytes - 1;
2530 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2531 &cached);
2532
2533 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2534 if (ordered) {
2535 btrfs_put_ordered_extent(ordered);
2536 goto out_unlock;
2537 }
2538
2539 trans = btrfs_join_transaction(root);
2540 if (IS_ERR(trans)) {
2541 ret = PTR_ERR(trans);
2542 goto out_unlock;
2543 }
2544
2545 key.objectid = backref->inum;
2546 key.type = BTRFS_EXTENT_DATA_KEY;
2547 key.offset = backref->file_pos;
2548
2549 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2550 if (ret < 0) {
2551 goto out_free_path;
2552 } else if (ret > 0) {
2553 ret = 0;
2554 goto out_free_path;
2555 }
2556
2557 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2558 struct btrfs_file_extent_item);
2559
2560 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2561 backref->generation)
2562 goto out_free_path;
2563
2564 btrfs_release_path(path);
2565
2566 start = backref->file_pos;
2567 if (backref->extent_offset < old->extent_offset + old->offset)
2568 start += old->extent_offset + old->offset -
2569 backref->extent_offset;
2570
2571 len = min(backref->extent_offset + backref->num_bytes,
2572 old->extent_offset + old->offset + old->len);
2573 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2574
2575 ret = btrfs_drop_extents(trans, root, inode, start,
2576 start + len, 1);
2577 if (ret)
2578 goto out_free_path;
2579 again:
2580 key.objectid = btrfs_ino(BTRFS_I(inode));
2581 key.type = BTRFS_EXTENT_DATA_KEY;
2582 key.offset = start;
2583
2584 path->leave_spinning = 1;
2585 if (merge) {
2586 struct btrfs_file_extent_item *fi;
2587 u64 extent_len;
2588 struct btrfs_key found_key;
2589
2590 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2591 if (ret < 0)
2592 goto out_free_path;
2593
2594 path->slots[0]--;
2595 leaf = path->nodes[0];
2596 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2597
2598 fi = btrfs_item_ptr(leaf, path->slots[0],
2599 struct btrfs_file_extent_item);
2600 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2601
2602 if (extent_len + found_key.offset == start &&
2603 relink_is_mergable(leaf, fi, new)) {
2604 btrfs_set_file_extent_num_bytes(leaf, fi,
2605 extent_len + len);
2606 btrfs_mark_buffer_dirty(leaf);
2607 inode_add_bytes(inode, len);
2608
2609 ret = 1;
2610 goto out_free_path;
2611 } else {
2612 merge = false;
2613 btrfs_release_path(path);
2614 goto again;
2615 }
2616 }
2617
2618 ret = btrfs_insert_empty_item(trans, root, path, &key,
2619 sizeof(*extent));
2620 if (ret) {
2621 btrfs_abort_transaction(trans, ret);
2622 goto out_free_path;
2623 }
2624
2625 leaf = path->nodes[0];
2626 item = btrfs_item_ptr(leaf, path->slots[0],
2627 struct btrfs_file_extent_item);
2628 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2629 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2630 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2631 btrfs_set_file_extent_num_bytes(leaf, item, len);
2632 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2633 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2634 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2635 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2636 btrfs_set_file_extent_encryption(leaf, item, 0);
2637 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2638
2639 btrfs_mark_buffer_dirty(leaf);
2640 inode_add_bytes(inode, len);
2641 btrfs_release_path(path);
2642
2643 ret = btrfs_inc_extent_ref(trans, fs_info, new->bytenr,
2644 new->disk_len, 0,
2645 backref->root_id, backref->inum,
2646 new->file_pos); /* start - extent_offset */
2647 if (ret) {
2648 btrfs_abort_transaction(trans, ret);
2649 goto out_free_path;
2650 }
2651
2652 ret = 1;
2653 out_free_path:
2654 btrfs_release_path(path);
2655 path->leave_spinning = 0;
2656 btrfs_end_transaction(trans);
2657 out_unlock:
2658 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2659 &cached, GFP_NOFS);
2660 iput(inode);
2661 return ret;
2662 }
2663
2664 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2665 {
2666 struct old_sa_defrag_extent *old, *tmp;
2667
2668 if (!new)
2669 return;
2670
2671 list_for_each_entry_safe(old, tmp, &new->head, list) {
2672 kfree(old);
2673 }
2674 kfree(new);
2675 }
2676
2677 static void relink_file_extents(struct new_sa_defrag_extent *new)
2678 {
2679 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2680 struct btrfs_path *path;
2681 struct sa_defrag_extent_backref *backref;
2682 struct sa_defrag_extent_backref *prev = NULL;
2683 struct inode *inode;
2684 struct btrfs_root *root;
2685 struct rb_node *node;
2686 int ret;
2687
2688 inode = new->inode;
2689 root = BTRFS_I(inode)->root;
2690
2691 path = btrfs_alloc_path();
2692 if (!path)
2693 return;
2694
2695 if (!record_extent_backrefs(path, new)) {
2696 btrfs_free_path(path);
2697 goto out;
2698 }
2699 btrfs_release_path(path);
2700
2701 while (1) {
2702 node = rb_first(&new->root);
2703 if (!node)
2704 break;
2705 rb_erase(node, &new->root);
2706
2707 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2708
2709 ret = relink_extent_backref(path, prev, backref);
2710 WARN_ON(ret < 0);
2711
2712 kfree(prev);
2713
2714 if (ret == 1)
2715 prev = backref;
2716 else
2717 prev = NULL;
2718 cond_resched();
2719 }
2720 kfree(prev);
2721
2722 btrfs_free_path(path);
2723 out:
2724 free_sa_defrag_extent(new);
2725
2726 atomic_dec(&fs_info->defrag_running);
2727 wake_up(&fs_info->transaction_wait);
2728 }
2729
2730 static struct new_sa_defrag_extent *
2731 record_old_file_extents(struct inode *inode,
2732 struct btrfs_ordered_extent *ordered)
2733 {
2734 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2735 struct btrfs_root *root = BTRFS_I(inode)->root;
2736 struct btrfs_path *path;
2737 struct btrfs_key key;
2738 struct old_sa_defrag_extent *old;
2739 struct new_sa_defrag_extent *new;
2740 int ret;
2741
2742 new = kmalloc(sizeof(*new), GFP_NOFS);
2743 if (!new)
2744 return NULL;
2745
2746 new->inode = inode;
2747 new->file_pos = ordered->file_offset;
2748 new->len = ordered->len;
2749 new->bytenr = ordered->start;
2750 new->disk_len = ordered->disk_len;
2751 new->compress_type = ordered->compress_type;
2752 new->root = RB_ROOT;
2753 INIT_LIST_HEAD(&new->head);
2754
2755 path = btrfs_alloc_path();
2756 if (!path)
2757 goto out_kfree;
2758
2759 key.objectid = btrfs_ino(BTRFS_I(inode));
2760 key.type = BTRFS_EXTENT_DATA_KEY;
2761 key.offset = new->file_pos;
2762
2763 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2764 if (ret < 0)
2765 goto out_free_path;
2766 if (ret > 0 && path->slots[0] > 0)
2767 path->slots[0]--;
2768
2769 /* find out all the old extents for the file range */
2770 while (1) {
2771 struct btrfs_file_extent_item *extent;
2772 struct extent_buffer *l;
2773 int slot;
2774 u64 num_bytes;
2775 u64 offset;
2776 u64 end;
2777 u64 disk_bytenr;
2778 u64 extent_offset;
2779
2780 l = path->nodes[0];
2781 slot = path->slots[0];
2782
2783 if (slot >= btrfs_header_nritems(l)) {
2784 ret = btrfs_next_leaf(root, path);
2785 if (ret < 0)
2786 goto out_free_path;
2787 else if (ret > 0)
2788 break;
2789 continue;
2790 }
2791
2792 btrfs_item_key_to_cpu(l, &key, slot);
2793
2794 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2795 break;
2796 if (key.type != BTRFS_EXTENT_DATA_KEY)
2797 break;
2798 if (key.offset >= new->file_pos + new->len)
2799 break;
2800
2801 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2802
2803 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2804 if (key.offset + num_bytes < new->file_pos)
2805 goto next;
2806
2807 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2808 if (!disk_bytenr)
2809 goto next;
2810
2811 extent_offset = btrfs_file_extent_offset(l, extent);
2812
2813 old = kmalloc(sizeof(*old), GFP_NOFS);
2814 if (!old)
2815 goto out_free_path;
2816
2817 offset = max(new->file_pos, key.offset);
2818 end = min(new->file_pos + new->len, key.offset + num_bytes);
2819
2820 old->bytenr = disk_bytenr;
2821 old->extent_offset = extent_offset;
2822 old->offset = offset - key.offset;
2823 old->len = end - offset;
2824 old->new = new;
2825 old->count = 0;
2826 list_add_tail(&old->list, &new->head);
2827 next:
2828 path->slots[0]++;
2829 cond_resched();
2830 }
2831
2832 btrfs_free_path(path);
2833 atomic_inc(&fs_info->defrag_running);
2834
2835 return new;
2836
2837 out_free_path:
2838 btrfs_free_path(path);
2839 out_kfree:
2840 free_sa_defrag_extent(new);
2841 return NULL;
2842 }
2843
2844 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2845 u64 start, u64 len)
2846 {
2847 struct btrfs_block_group_cache *cache;
2848
2849 cache = btrfs_lookup_block_group(fs_info, start);
2850 ASSERT(cache);
2851
2852 spin_lock(&cache->lock);
2853 cache->delalloc_bytes -= len;
2854 spin_unlock(&cache->lock);
2855
2856 btrfs_put_block_group(cache);
2857 }
2858
2859 /* as ordered data IO finishes, this gets called so we can finish
2860 * an ordered extent if the range of bytes in the file it covers are
2861 * fully written.
2862 */
2863 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2864 {
2865 struct inode *inode = ordered_extent->inode;
2866 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2867 struct btrfs_root *root = BTRFS_I(inode)->root;
2868 struct btrfs_trans_handle *trans = NULL;
2869 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2870 struct extent_state *cached_state = NULL;
2871 struct new_sa_defrag_extent *new = NULL;
2872 int compress_type = 0;
2873 int ret = 0;
2874 u64 logical_len = ordered_extent->len;
2875 bool nolock;
2876 bool truncated = false;
2877
2878 nolock = btrfs_is_free_space_inode(inode);
2879
2880 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2881 ret = -EIO;
2882 goto out;
2883 }
2884
2885 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2886 ordered_extent->file_offset +
2887 ordered_extent->len - 1);
2888
2889 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2890 truncated = true;
2891 logical_len = ordered_extent->truncated_len;
2892 /* Truncated the entire extent, don't bother adding */
2893 if (!logical_len)
2894 goto out;
2895 }
2896
2897 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2898 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2899
2900 /*
2901 * For mwrite(mmap + memset to write) case, we still reserve
2902 * space for NOCOW range.
2903 * As NOCOW won't cause a new delayed ref, just free the space
2904 */
2905 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2906 ordered_extent->len);
2907 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2908 if (nolock)
2909 trans = btrfs_join_transaction_nolock(root);
2910 else
2911 trans = btrfs_join_transaction(root);
2912 if (IS_ERR(trans)) {
2913 ret = PTR_ERR(trans);
2914 trans = NULL;
2915 goto out;
2916 }
2917 trans->block_rsv = &fs_info->delalloc_block_rsv;
2918 ret = btrfs_update_inode_fallback(trans, root, inode);
2919 if (ret) /* -ENOMEM or corruption */
2920 btrfs_abort_transaction(trans, ret);
2921 goto out;
2922 }
2923
2924 lock_extent_bits(io_tree, ordered_extent->file_offset,
2925 ordered_extent->file_offset + ordered_extent->len - 1,
2926 &cached_state);
2927
2928 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2929 ordered_extent->file_offset + ordered_extent->len - 1,
2930 EXTENT_DEFRAG, 1, cached_state);
2931 if (ret) {
2932 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2933 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2934 /* the inode is shared */
2935 new = record_old_file_extents(inode, ordered_extent);
2936
2937 clear_extent_bit(io_tree, ordered_extent->file_offset,
2938 ordered_extent->file_offset + ordered_extent->len - 1,
2939 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2940 }
2941
2942 if (nolock)
2943 trans = btrfs_join_transaction_nolock(root);
2944 else
2945 trans = btrfs_join_transaction(root);
2946 if (IS_ERR(trans)) {
2947 ret = PTR_ERR(trans);
2948 trans = NULL;
2949 goto out_unlock;
2950 }
2951
2952 trans->block_rsv = &fs_info->delalloc_block_rsv;
2953
2954 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2955 compress_type = ordered_extent->compress_type;
2956 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2957 BUG_ON(compress_type);
2958 ret = btrfs_mark_extent_written(trans, inode,
2959 ordered_extent->file_offset,
2960 ordered_extent->file_offset +
2961 logical_len);
2962 } else {
2963 BUG_ON(root == fs_info->tree_root);
2964 ret = insert_reserved_file_extent(trans, inode,
2965 ordered_extent->file_offset,
2966 ordered_extent->start,
2967 ordered_extent->disk_len,
2968 logical_len, logical_len,
2969 compress_type, 0, 0,
2970 BTRFS_FILE_EXTENT_REG);
2971 if (!ret)
2972 btrfs_release_delalloc_bytes(fs_info,
2973 ordered_extent->start,
2974 ordered_extent->disk_len);
2975 }
2976 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2977 ordered_extent->file_offset, ordered_extent->len,
2978 trans->transid);
2979 if (ret < 0) {
2980 btrfs_abort_transaction(trans, ret);
2981 goto out_unlock;
2982 }
2983
2984 add_pending_csums(trans, inode, ordered_extent->file_offset,
2985 &ordered_extent->list);
2986
2987 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2988 ret = btrfs_update_inode_fallback(trans, root, inode);
2989 if (ret) { /* -ENOMEM or corruption */
2990 btrfs_abort_transaction(trans, ret);
2991 goto out_unlock;
2992 }
2993 ret = 0;
2994 out_unlock:
2995 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2996 ordered_extent->file_offset +
2997 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2998 out:
2999 if (root != fs_info->tree_root)
3000 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
3001 if (trans)
3002 btrfs_end_transaction(trans);
3003
3004 if (ret || truncated) {
3005 u64 start, end;
3006
3007 if (truncated)
3008 start = ordered_extent->file_offset + logical_len;
3009 else
3010 start = ordered_extent->file_offset;
3011 end = ordered_extent->file_offset + ordered_extent->len - 1;
3012 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
3013
3014 /* Drop the cache for the part of the extent we didn't write. */
3015 btrfs_drop_extent_cache(inode, start, end, 0);
3016
3017 /*
3018 * If the ordered extent had an IOERR or something else went
3019 * wrong we need to return the space for this ordered extent
3020 * back to the allocator. We only free the extent in the
3021 * truncated case if we didn't write out the extent at all.
3022 */
3023 if ((ret || !logical_len) &&
3024 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3025 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3026 btrfs_free_reserved_extent(fs_info,
3027 ordered_extent->start,
3028 ordered_extent->disk_len, 1);
3029 }
3030
3031
3032 /*
3033 * This needs to be done to make sure anybody waiting knows we are done
3034 * updating everything for this ordered extent.
3035 */
3036 btrfs_remove_ordered_extent(inode, ordered_extent);
3037
3038 /* for snapshot-aware defrag */
3039 if (new) {
3040 if (ret) {
3041 free_sa_defrag_extent(new);
3042 atomic_dec(&fs_info->defrag_running);
3043 } else {
3044 relink_file_extents(new);
3045 }
3046 }
3047
3048 /* once for us */
3049 btrfs_put_ordered_extent(ordered_extent);
3050 /* once for the tree */
3051 btrfs_put_ordered_extent(ordered_extent);
3052
3053 return ret;
3054 }
3055
3056 static void finish_ordered_fn(struct btrfs_work *work)
3057 {
3058 struct btrfs_ordered_extent *ordered_extent;
3059 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3060 btrfs_finish_ordered_io(ordered_extent);
3061 }
3062
3063 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3064 struct extent_state *state, int uptodate)
3065 {
3066 struct inode *inode = page->mapping->host;
3067 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3068 struct btrfs_ordered_extent *ordered_extent = NULL;
3069 struct btrfs_workqueue *wq;
3070 btrfs_work_func_t func;
3071
3072 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3073
3074 ClearPagePrivate2(page);
3075 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3076 end - start + 1, uptodate))
3077 return 0;
3078
3079 if (btrfs_is_free_space_inode(inode)) {
3080 wq = fs_info->endio_freespace_worker;
3081 func = btrfs_freespace_write_helper;
3082 } else {
3083 wq = fs_info->endio_write_workers;
3084 func = btrfs_endio_write_helper;
3085 }
3086
3087 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3088 NULL);
3089 btrfs_queue_work(wq, &ordered_extent->work);
3090
3091 return 0;
3092 }
3093
3094 static int __readpage_endio_check(struct inode *inode,
3095 struct btrfs_io_bio *io_bio,
3096 int icsum, struct page *page,
3097 int pgoff, u64 start, size_t len)
3098 {
3099 char *kaddr;
3100 u32 csum_expected;
3101 u32 csum = ~(u32)0;
3102
3103 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3104
3105 kaddr = kmap_atomic(page);
3106 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3107 btrfs_csum_final(csum, (u8 *)&csum);
3108 if (csum != csum_expected)
3109 goto zeroit;
3110
3111 kunmap_atomic(kaddr);
3112 return 0;
3113 zeroit:
3114 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3115 "csum failed ino %llu off %llu csum %u expected csum %u",
3116 btrfs_ino(BTRFS_I(inode)), start, csum, csum_expected);
3117 memset(kaddr + pgoff, 1, len);
3118 flush_dcache_page(page);
3119 kunmap_atomic(kaddr);
3120 if (csum_expected == 0)
3121 return 0;
3122 return -EIO;
3123 }
3124
3125 /*
3126 * when reads are done, we need to check csums to verify the data is correct
3127 * if there's a match, we allow the bio to finish. If not, the code in
3128 * extent_io.c will try to find good copies for us.
3129 */
3130 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3131 u64 phy_offset, struct page *page,
3132 u64 start, u64 end, int mirror)
3133 {
3134 size_t offset = start - page_offset(page);
3135 struct inode *inode = page->mapping->host;
3136 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3137 struct btrfs_root *root = BTRFS_I(inode)->root;
3138
3139 if (PageChecked(page)) {
3140 ClearPageChecked(page);
3141 return 0;
3142 }
3143
3144 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3145 return 0;
3146
3147 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3148 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3149 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3150 return 0;
3151 }
3152
3153 phy_offset >>= inode->i_sb->s_blocksize_bits;
3154 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3155 start, (size_t)(end - start + 1));
3156 }
3157
3158 void btrfs_add_delayed_iput(struct inode *inode)
3159 {
3160 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3161 struct btrfs_inode *binode = BTRFS_I(inode);
3162
3163 if (atomic_add_unless(&inode->i_count, -1, 1))
3164 return;
3165
3166 spin_lock(&fs_info->delayed_iput_lock);
3167 if (binode->delayed_iput_count == 0) {
3168 ASSERT(list_empty(&binode->delayed_iput));
3169 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3170 } else {
3171 binode->delayed_iput_count++;
3172 }
3173 spin_unlock(&fs_info->delayed_iput_lock);
3174 }
3175
3176 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3177 {
3178
3179 spin_lock(&fs_info->delayed_iput_lock);
3180 while (!list_empty(&fs_info->delayed_iputs)) {
3181 struct btrfs_inode *inode;
3182
3183 inode = list_first_entry(&fs_info->delayed_iputs,
3184 struct btrfs_inode, delayed_iput);
3185 if (inode->delayed_iput_count) {
3186 inode->delayed_iput_count--;
3187 list_move_tail(&inode->delayed_iput,
3188 &fs_info->delayed_iputs);
3189 } else {
3190 list_del_init(&inode->delayed_iput);
3191 }
3192 spin_unlock(&fs_info->delayed_iput_lock);
3193 iput(&inode->vfs_inode);
3194 spin_lock(&fs_info->delayed_iput_lock);
3195 }
3196 spin_unlock(&fs_info->delayed_iput_lock);
3197 }
3198
3199 /*
3200 * This is called in transaction commit time. If there are no orphan
3201 * files in the subvolume, it removes orphan item and frees block_rsv
3202 * structure.
3203 */
3204 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3205 struct btrfs_root *root)
3206 {
3207 struct btrfs_fs_info *fs_info = root->fs_info;
3208 struct btrfs_block_rsv *block_rsv;
3209 int ret;
3210
3211 if (atomic_read(&root->orphan_inodes) ||
3212 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3213 return;
3214
3215 spin_lock(&root->orphan_lock);
3216 if (atomic_read(&root->orphan_inodes)) {
3217 spin_unlock(&root->orphan_lock);
3218 return;
3219 }
3220
3221 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3222 spin_unlock(&root->orphan_lock);
3223 return;
3224 }
3225
3226 block_rsv = root->orphan_block_rsv;
3227 root->orphan_block_rsv = NULL;
3228 spin_unlock(&root->orphan_lock);
3229
3230 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3231 btrfs_root_refs(&root->root_item) > 0) {
3232 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3233 root->root_key.objectid);
3234 if (ret)
3235 btrfs_abort_transaction(trans, ret);
3236 else
3237 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3238 &root->state);
3239 }
3240
3241 if (block_rsv) {
3242 WARN_ON(block_rsv->size > 0);
3243 btrfs_free_block_rsv(fs_info, block_rsv);
3244 }
3245 }
3246
3247 /*
3248 * This creates an orphan entry for the given inode in case something goes
3249 * wrong in the middle of an unlink/truncate.
3250 *
3251 * NOTE: caller of this function should reserve 5 units of metadata for
3252 * this function.
3253 */
3254 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3255 {
3256 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3257 struct btrfs_root *root = BTRFS_I(inode)->root;
3258 struct btrfs_block_rsv *block_rsv = NULL;
3259 int reserve = 0;
3260 int insert = 0;
3261 int ret;
3262
3263 if (!root->orphan_block_rsv) {
3264 block_rsv = btrfs_alloc_block_rsv(fs_info,
3265 BTRFS_BLOCK_RSV_TEMP);
3266 if (!block_rsv)
3267 return -ENOMEM;
3268 }
3269
3270 spin_lock(&root->orphan_lock);
3271 if (!root->orphan_block_rsv) {
3272 root->orphan_block_rsv = block_rsv;
3273 } else if (block_rsv) {
3274 btrfs_free_block_rsv(fs_info, block_rsv);
3275 block_rsv = NULL;
3276 }
3277
3278 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3279 &BTRFS_I(inode)->runtime_flags)) {
3280 #if 0
3281 /*
3282 * For proper ENOSPC handling, we should do orphan
3283 * cleanup when mounting. But this introduces backward
3284 * compatibility issue.
3285 */
3286 if (!xchg(&root->orphan_item_inserted, 1))
3287 insert = 2;
3288 else
3289 insert = 1;
3290 #endif
3291 insert = 1;
3292 atomic_inc(&root->orphan_inodes);
3293 }
3294
3295 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3296 &BTRFS_I(inode)->runtime_flags))
3297 reserve = 1;
3298 spin_unlock(&root->orphan_lock);
3299
3300 /* grab metadata reservation from transaction handle */
3301 if (reserve) {
3302 ret = btrfs_orphan_reserve_metadata(trans, inode);
3303 ASSERT(!ret);
3304 if (ret) {
3305 atomic_dec(&root->orphan_inodes);
3306 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3307 &BTRFS_I(inode)->runtime_flags);
3308 if (insert)
3309 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3310 &BTRFS_I(inode)->runtime_flags);
3311 return ret;
3312 }
3313 }
3314
3315 /* insert an orphan item to track this unlinked/truncated file */
3316 if (insert >= 1) {
3317 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(BTRFS_I(inode)));
3318 if (ret) {
3319 atomic_dec(&root->orphan_inodes);
3320 if (reserve) {
3321 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3322 &BTRFS_I(inode)->runtime_flags);
3323 btrfs_orphan_release_metadata(inode);
3324 }
3325 if (ret != -EEXIST) {
3326 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3327 &BTRFS_I(inode)->runtime_flags);
3328 btrfs_abort_transaction(trans, ret);
3329 return ret;
3330 }
3331 }
3332 ret = 0;
3333 }
3334
3335 /* insert an orphan item to track subvolume contains orphan files */
3336 if (insert >= 2) {
3337 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3338 root->root_key.objectid);
3339 if (ret && ret != -EEXIST) {
3340 btrfs_abort_transaction(trans, ret);
3341 return ret;
3342 }
3343 }
3344 return 0;
3345 }
3346
3347 /*
3348 * We have done the truncate/delete so we can go ahead and remove the orphan
3349 * item for this particular inode.
3350 */
3351 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3352 struct inode *inode)
3353 {
3354 struct btrfs_root *root = BTRFS_I(inode)->root;
3355 int delete_item = 0;
3356 int release_rsv = 0;
3357 int ret = 0;
3358
3359 spin_lock(&root->orphan_lock);
3360 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3361 &BTRFS_I(inode)->runtime_flags))
3362 delete_item = 1;
3363
3364 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3365 &BTRFS_I(inode)->runtime_flags))
3366 release_rsv = 1;
3367 spin_unlock(&root->orphan_lock);
3368
3369 if (delete_item) {
3370 atomic_dec(&root->orphan_inodes);
3371 if (trans)
3372 ret = btrfs_del_orphan_item(trans, root,
3373 btrfs_ino(BTRFS_I(inode)));
3374 }
3375
3376 if (release_rsv)
3377 btrfs_orphan_release_metadata(inode);
3378
3379 return ret;
3380 }
3381
3382 /*
3383 * this cleans up any orphans that may be left on the list from the last use
3384 * of this root.
3385 */
3386 int btrfs_orphan_cleanup(struct btrfs_root *root)
3387 {
3388 struct btrfs_fs_info *fs_info = root->fs_info;
3389 struct btrfs_path *path;
3390 struct extent_buffer *leaf;
3391 struct btrfs_key key, found_key;
3392 struct btrfs_trans_handle *trans;
3393 struct inode *inode;
3394 u64 last_objectid = 0;
3395 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3396
3397 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3398 return 0;
3399
3400 path = btrfs_alloc_path();
3401 if (!path) {
3402 ret = -ENOMEM;
3403 goto out;
3404 }
3405 path->reada = READA_BACK;
3406
3407 key.objectid = BTRFS_ORPHAN_OBJECTID;
3408 key.type = BTRFS_ORPHAN_ITEM_KEY;
3409 key.offset = (u64)-1;
3410
3411 while (1) {
3412 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3413 if (ret < 0)
3414 goto out;
3415
3416 /*
3417 * if ret == 0 means we found what we were searching for, which
3418 * is weird, but possible, so only screw with path if we didn't
3419 * find the key and see if we have stuff that matches
3420 */
3421 if (ret > 0) {
3422 ret = 0;
3423 if (path->slots[0] == 0)
3424 break;
3425 path->slots[0]--;
3426 }
3427
3428 /* pull out the item */
3429 leaf = path->nodes[0];
3430 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3431
3432 /* make sure the item matches what we want */
3433 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3434 break;
3435 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3436 break;
3437
3438 /* release the path since we're done with it */
3439 btrfs_release_path(path);
3440
3441 /*
3442 * this is where we are basically btrfs_lookup, without the
3443 * crossing root thing. we store the inode number in the
3444 * offset of the orphan item.
3445 */
3446
3447 if (found_key.offset == last_objectid) {
3448 btrfs_err(fs_info,
3449 "Error removing orphan entry, stopping orphan cleanup");
3450 ret = -EINVAL;
3451 goto out;
3452 }
3453
3454 last_objectid = found_key.offset;
3455
3456 found_key.objectid = found_key.offset;
3457 found_key.type = BTRFS_INODE_ITEM_KEY;
3458 found_key.offset = 0;
3459 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3460 ret = PTR_ERR_OR_ZERO(inode);
3461 if (ret && ret != -ENOENT)
3462 goto out;
3463
3464 if (ret == -ENOENT && root == fs_info->tree_root) {
3465 struct btrfs_root *dead_root;
3466 struct btrfs_fs_info *fs_info = root->fs_info;
3467 int is_dead_root = 0;
3468
3469 /*
3470 * this is an orphan in the tree root. Currently these
3471 * could come from 2 sources:
3472 * a) a snapshot deletion in progress
3473 * b) a free space cache inode
3474 * We need to distinguish those two, as the snapshot
3475 * orphan must not get deleted.
3476 * find_dead_roots already ran before us, so if this
3477 * is a snapshot deletion, we should find the root
3478 * in the dead_roots list
3479 */
3480 spin_lock(&fs_info->trans_lock);
3481 list_for_each_entry(dead_root, &fs_info->dead_roots,
3482 root_list) {
3483 if (dead_root->root_key.objectid ==
3484 found_key.objectid) {
3485 is_dead_root = 1;
3486 break;
3487 }
3488 }
3489 spin_unlock(&fs_info->trans_lock);
3490 if (is_dead_root) {
3491 /* prevent this orphan from being found again */
3492 key.offset = found_key.objectid - 1;
3493 continue;
3494 }
3495 }
3496 /*
3497 * Inode is already gone but the orphan item is still there,
3498 * kill the orphan item.
3499 */
3500 if (ret == -ENOENT) {
3501 trans = btrfs_start_transaction(root, 1);
3502 if (IS_ERR(trans)) {
3503 ret = PTR_ERR(trans);
3504 goto out;
3505 }
3506 btrfs_debug(fs_info, "auto deleting %Lu",
3507 found_key.objectid);
3508 ret = btrfs_del_orphan_item(trans, root,
3509 found_key.objectid);
3510 btrfs_end_transaction(trans);
3511 if (ret)
3512 goto out;
3513 continue;
3514 }
3515
3516 /*
3517 * add this inode to the orphan list so btrfs_orphan_del does
3518 * the proper thing when we hit it
3519 */
3520 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3521 &BTRFS_I(inode)->runtime_flags);
3522 atomic_inc(&root->orphan_inodes);
3523
3524 /* if we have links, this was a truncate, lets do that */
3525 if (inode->i_nlink) {
3526 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3527 iput(inode);
3528 continue;
3529 }
3530 nr_truncate++;
3531
3532 /* 1 for the orphan item deletion. */
3533 trans = btrfs_start_transaction(root, 1);
3534 if (IS_ERR(trans)) {
3535 iput(inode);
3536 ret = PTR_ERR(trans);
3537 goto out;
3538 }
3539 ret = btrfs_orphan_add(trans, inode);
3540 btrfs_end_transaction(trans);
3541 if (ret) {
3542 iput(inode);
3543 goto out;
3544 }
3545
3546 ret = btrfs_truncate(inode);
3547 if (ret)
3548 btrfs_orphan_del(NULL, inode);
3549 } else {
3550 nr_unlink++;
3551 }
3552
3553 /* this will do delete_inode and everything for us */
3554 iput(inode);
3555 if (ret)
3556 goto out;
3557 }
3558 /* release the path since we're done with it */
3559 btrfs_release_path(path);
3560
3561 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3562
3563 if (root->orphan_block_rsv)
3564 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3565 (u64)-1);
3566
3567 if (root->orphan_block_rsv ||
3568 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3569 trans = btrfs_join_transaction(root);
3570 if (!IS_ERR(trans))
3571 btrfs_end_transaction(trans);
3572 }
3573
3574 if (nr_unlink)
3575 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3576 if (nr_truncate)
3577 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3578
3579 out:
3580 if (ret)
3581 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3582 btrfs_free_path(path);
3583 return ret;
3584 }
3585
3586 /*
3587 * very simple check to peek ahead in the leaf looking for xattrs. If we
3588 * don't find any xattrs, we know there can't be any acls.
3589 *
3590 * slot is the slot the inode is in, objectid is the objectid of the inode
3591 */
3592 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3593 int slot, u64 objectid,
3594 int *first_xattr_slot)
3595 {
3596 u32 nritems = btrfs_header_nritems(leaf);
3597 struct btrfs_key found_key;
3598 static u64 xattr_access = 0;
3599 static u64 xattr_default = 0;
3600 int scanned = 0;
3601
3602 if (!xattr_access) {
3603 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3604 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3605 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3606 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3607 }
3608
3609 slot++;
3610 *first_xattr_slot = -1;
3611 while (slot < nritems) {
3612 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3613
3614 /* we found a different objectid, there must not be acls */
3615 if (found_key.objectid != objectid)
3616 return 0;
3617
3618 /* we found an xattr, assume we've got an acl */
3619 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3620 if (*first_xattr_slot == -1)
3621 *first_xattr_slot = slot;
3622 if (found_key.offset == xattr_access ||
3623 found_key.offset == xattr_default)
3624 return 1;
3625 }
3626
3627 /*
3628 * we found a key greater than an xattr key, there can't
3629 * be any acls later on
3630 */
3631 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3632 return 0;
3633
3634 slot++;
3635 scanned++;
3636
3637 /*
3638 * it goes inode, inode backrefs, xattrs, extents,
3639 * so if there are a ton of hard links to an inode there can
3640 * be a lot of backrefs. Don't waste time searching too hard,
3641 * this is just an optimization
3642 */
3643 if (scanned >= 8)
3644 break;
3645 }
3646 /* we hit the end of the leaf before we found an xattr or
3647 * something larger than an xattr. We have to assume the inode
3648 * has acls
3649 */
3650 if (*first_xattr_slot == -1)
3651 *first_xattr_slot = slot;
3652 return 1;
3653 }
3654
3655 /*
3656 * read an inode from the btree into the in-memory inode
3657 */
3658 static int btrfs_read_locked_inode(struct inode *inode)
3659 {
3660 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3661 struct btrfs_path *path;
3662 struct extent_buffer *leaf;
3663 struct btrfs_inode_item *inode_item;
3664 struct btrfs_root *root = BTRFS_I(inode)->root;
3665 struct btrfs_key location;
3666 unsigned long ptr;
3667 int maybe_acls;
3668 u32 rdev;
3669 int ret;
3670 bool filled = false;
3671 int first_xattr_slot;
3672
3673 ret = btrfs_fill_inode(inode, &rdev);
3674 if (!ret)
3675 filled = true;
3676
3677 path = btrfs_alloc_path();
3678 if (!path) {
3679 ret = -ENOMEM;
3680 goto make_bad;
3681 }
3682
3683 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3684
3685 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3686 if (ret) {
3687 if (ret > 0)
3688 ret = -ENOENT;
3689 goto make_bad;
3690 }
3691
3692 leaf = path->nodes[0];
3693
3694 if (filled)
3695 goto cache_index;
3696
3697 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3698 struct btrfs_inode_item);
3699 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3700 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3701 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3702 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3703 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3704
3705 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3706 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3707
3708 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3709 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3710
3711 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3712 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3713
3714 BTRFS_I(inode)->i_otime.tv_sec =
3715 btrfs_timespec_sec(leaf, &inode_item->otime);
3716 BTRFS_I(inode)->i_otime.tv_nsec =
3717 btrfs_timespec_nsec(leaf, &inode_item->otime);
3718
3719 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3720 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3721 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3722
3723 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3724 inode->i_generation = BTRFS_I(inode)->generation;
3725 inode->i_rdev = 0;
3726 rdev = btrfs_inode_rdev(leaf, inode_item);
3727
3728 BTRFS_I(inode)->index_cnt = (u64)-1;
3729 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3730
3731 cache_index:
3732 /*
3733 * If we were modified in the current generation and evicted from memory
3734 * and then re-read we need to do a full sync since we don't have any
3735 * idea about which extents were modified before we were evicted from
3736 * cache.
3737 *
3738 * This is required for both inode re-read from disk and delayed inode
3739 * in delayed_nodes_tree.
3740 */
3741 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3742 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3743 &BTRFS_I(inode)->runtime_flags);
3744
3745 /*
3746 * We don't persist the id of the transaction where an unlink operation
3747 * against the inode was last made. So here we assume the inode might
3748 * have been evicted, and therefore the exact value of last_unlink_trans
3749 * lost, and set it to last_trans to avoid metadata inconsistencies
3750 * between the inode and its parent if the inode is fsync'ed and the log
3751 * replayed. For example, in the scenario:
3752 *
3753 * touch mydir/foo
3754 * ln mydir/foo mydir/bar
3755 * sync
3756 * unlink mydir/bar
3757 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3758 * xfs_io -c fsync mydir/foo
3759 * <power failure>
3760 * mount fs, triggers fsync log replay
3761 *
3762 * We must make sure that when we fsync our inode foo we also log its
3763 * parent inode, otherwise after log replay the parent still has the
3764 * dentry with the "bar" name but our inode foo has a link count of 1
3765 * and doesn't have an inode ref with the name "bar" anymore.
3766 *
3767 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3768 * but it guarantees correctness at the expense of occasional full
3769 * transaction commits on fsync if our inode is a directory, or if our
3770 * inode is not a directory, logging its parent unnecessarily.
3771 */
3772 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3773
3774 path->slots[0]++;
3775 if (inode->i_nlink != 1 ||
3776 path->slots[0] >= btrfs_header_nritems(leaf))
3777 goto cache_acl;
3778
3779 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3780 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3781 goto cache_acl;
3782
3783 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3784 if (location.type == BTRFS_INODE_REF_KEY) {
3785 struct btrfs_inode_ref *ref;
3786
3787 ref = (struct btrfs_inode_ref *)ptr;
3788 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3789 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3790 struct btrfs_inode_extref *extref;
3791
3792 extref = (struct btrfs_inode_extref *)ptr;
3793 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3794 extref);
3795 }
3796 cache_acl:
3797 /*
3798 * try to precache a NULL acl entry for files that don't have
3799 * any xattrs or acls
3800 */
3801 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3802 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3803 if (first_xattr_slot != -1) {
3804 path->slots[0] = first_xattr_slot;
3805 ret = btrfs_load_inode_props(inode, path);
3806 if (ret)
3807 btrfs_err(fs_info,
3808 "error loading props for ino %llu (root %llu): %d",
3809 btrfs_ino(BTRFS_I(inode)),
3810 root->root_key.objectid, ret);
3811 }
3812 btrfs_free_path(path);
3813
3814 if (!maybe_acls)
3815 cache_no_acl(inode);
3816
3817 switch (inode->i_mode & S_IFMT) {
3818 case S_IFREG:
3819 inode->i_mapping->a_ops = &btrfs_aops;
3820 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3821 inode->i_fop = &btrfs_file_operations;
3822 inode->i_op = &btrfs_file_inode_operations;
3823 break;
3824 case S_IFDIR:
3825 inode->i_fop = &btrfs_dir_file_operations;
3826 inode->i_op = &btrfs_dir_inode_operations;
3827 break;
3828 case S_IFLNK:
3829 inode->i_op = &btrfs_symlink_inode_operations;
3830 inode_nohighmem(inode);
3831 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3832 break;
3833 default:
3834 inode->i_op = &btrfs_special_inode_operations;
3835 init_special_inode(inode, inode->i_mode, rdev);
3836 break;
3837 }
3838
3839 btrfs_update_iflags(inode);
3840 return 0;
3841
3842 make_bad:
3843 btrfs_free_path(path);
3844 make_bad_inode(inode);
3845 return ret;
3846 }
3847
3848 /*
3849 * given a leaf and an inode, copy the inode fields into the leaf
3850 */
3851 static void fill_inode_item(struct btrfs_trans_handle *trans,
3852 struct extent_buffer *leaf,
3853 struct btrfs_inode_item *item,
3854 struct inode *inode)
3855 {
3856 struct btrfs_map_token token;
3857
3858 btrfs_init_map_token(&token);
3859
3860 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3861 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3862 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3863 &token);
3864 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3865 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3866
3867 btrfs_set_token_timespec_sec(leaf, &item->atime,
3868 inode->i_atime.tv_sec, &token);
3869 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3870 inode->i_atime.tv_nsec, &token);
3871
3872 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3873 inode->i_mtime.tv_sec, &token);
3874 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3875 inode->i_mtime.tv_nsec, &token);
3876
3877 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3878 inode->i_ctime.tv_sec, &token);
3879 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3880 inode->i_ctime.tv_nsec, &token);
3881
3882 btrfs_set_token_timespec_sec(leaf, &item->otime,
3883 BTRFS_I(inode)->i_otime.tv_sec, &token);
3884 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3885 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3886
3887 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3888 &token);
3889 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3890 &token);
3891 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3892 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3893 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3894 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3895 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3896 }
3897
3898 /*
3899 * copy everything in the in-memory inode into the btree.
3900 */
3901 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3902 struct btrfs_root *root, struct inode *inode)
3903 {
3904 struct btrfs_inode_item *inode_item;
3905 struct btrfs_path *path;
3906 struct extent_buffer *leaf;
3907 int ret;
3908
3909 path = btrfs_alloc_path();
3910 if (!path)
3911 return -ENOMEM;
3912
3913 path->leave_spinning = 1;
3914 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3915 1);
3916 if (ret) {
3917 if (ret > 0)
3918 ret = -ENOENT;
3919 goto failed;
3920 }
3921
3922 leaf = path->nodes[0];
3923 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3924 struct btrfs_inode_item);
3925
3926 fill_inode_item(trans, leaf, inode_item, inode);
3927 btrfs_mark_buffer_dirty(leaf);
3928 btrfs_set_inode_last_trans(trans, inode);
3929 ret = 0;
3930 failed:
3931 btrfs_free_path(path);
3932 return ret;
3933 }
3934
3935 /*
3936 * copy everything in the in-memory inode into the btree.
3937 */
3938 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3939 struct btrfs_root *root, struct inode *inode)
3940 {
3941 struct btrfs_fs_info *fs_info = root->fs_info;
3942 int ret;
3943
3944 /*
3945 * If the inode is a free space inode, we can deadlock during commit
3946 * if we put it into the delayed code.
3947 *
3948 * The data relocation inode should also be directly updated
3949 * without delay
3950 */
3951 if (!btrfs_is_free_space_inode(inode)
3952 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3953 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3954 btrfs_update_root_times(trans, root);
3955
3956 ret = btrfs_delayed_update_inode(trans, root, inode);
3957 if (!ret)
3958 btrfs_set_inode_last_trans(trans, inode);
3959 return ret;
3960 }
3961
3962 return btrfs_update_inode_item(trans, root, inode);
3963 }
3964
3965 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3966 struct btrfs_root *root,
3967 struct inode *inode)
3968 {
3969 int ret;
3970
3971 ret = btrfs_update_inode(trans, root, inode);
3972 if (ret == -ENOSPC)
3973 return btrfs_update_inode_item(trans, root, inode);
3974 return ret;
3975 }
3976
3977 /*
3978 * unlink helper that gets used here in inode.c and in the tree logging
3979 * recovery code. It remove a link in a directory with a given name, and
3980 * also drops the back refs in the inode to the directory
3981 */
3982 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3983 struct btrfs_root *root,
3984 struct inode *dir, struct inode *inode,
3985 const char *name, int name_len)
3986 {
3987 struct btrfs_fs_info *fs_info = root->fs_info;
3988 struct btrfs_path *path;
3989 int ret = 0;
3990 struct extent_buffer *leaf;
3991 struct btrfs_dir_item *di;
3992 struct btrfs_key key;
3993 u64 index;
3994 u64 ino = btrfs_ino(BTRFS_I(inode));
3995 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
3996
3997 path = btrfs_alloc_path();
3998 if (!path) {
3999 ret = -ENOMEM;
4000 goto out;
4001 }
4002
4003 path->leave_spinning = 1;
4004 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4005 name, name_len, -1);
4006 if (IS_ERR(di)) {
4007 ret = PTR_ERR(di);
4008 goto err;
4009 }
4010 if (!di) {
4011 ret = -ENOENT;
4012 goto err;
4013 }
4014 leaf = path->nodes[0];
4015 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4016 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4017 if (ret)
4018 goto err;
4019 btrfs_release_path(path);
4020
4021 /*
4022 * If we don't have dir index, we have to get it by looking up
4023 * the inode ref, since we get the inode ref, remove it directly,
4024 * it is unnecessary to do delayed deletion.
4025 *
4026 * But if we have dir index, needn't search inode ref to get it.
4027 * Since the inode ref is close to the inode item, it is better
4028 * that we delay to delete it, and just do this deletion when
4029 * we update the inode item.
4030 */
4031 if (BTRFS_I(inode)->dir_index) {
4032 ret = btrfs_delayed_delete_inode_ref(BTRFS_I(inode));
4033 if (!ret) {
4034 index = BTRFS_I(inode)->dir_index;
4035 goto skip_backref;
4036 }
4037 }
4038
4039 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4040 dir_ino, &index);
4041 if (ret) {
4042 btrfs_info(fs_info,
4043 "failed to delete reference to %.*s, inode %llu parent %llu",
4044 name_len, name, ino, dir_ino);
4045 btrfs_abort_transaction(trans, ret);
4046 goto err;
4047 }
4048 skip_backref:
4049 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4050 if (ret) {
4051 btrfs_abort_transaction(trans, ret);
4052 goto err;
4053 }
4054
4055 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
4056 inode, dir_ino);
4057 if (ret != 0 && ret != -ENOENT) {
4058 btrfs_abort_transaction(trans, ret);
4059 goto err;
4060 }
4061
4062 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4063 dir, index);
4064 if (ret == -ENOENT)
4065 ret = 0;
4066 else if (ret)
4067 btrfs_abort_transaction(trans, ret);
4068 err:
4069 btrfs_free_path(path);
4070 if (ret)
4071 goto out;
4072
4073 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4074 inode_inc_iversion(inode);
4075 inode_inc_iversion(dir);
4076 inode->i_ctime = dir->i_mtime =
4077 dir->i_ctime = current_time(inode);
4078 ret = btrfs_update_inode(trans, root, dir);
4079 out:
4080 return ret;
4081 }
4082
4083 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4084 struct btrfs_root *root,
4085 struct inode *dir, struct inode *inode,
4086 const char *name, int name_len)
4087 {
4088 int ret;
4089 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4090 if (!ret) {
4091 drop_nlink(inode);
4092 ret = btrfs_update_inode(trans, root, inode);
4093 }
4094 return ret;
4095 }
4096
4097 /*
4098 * helper to start transaction for unlink and rmdir.
4099 *
4100 * unlink and rmdir are special in btrfs, they do not always free space, so
4101 * if we cannot make our reservations the normal way try and see if there is
4102 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4103 * allow the unlink to occur.
4104 */
4105 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4106 {
4107 struct btrfs_root *root = BTRFS_I(dir)->root;
4108
4109 /*
4110 * 1 for the possible orphan item
4111 * 1 for the dir item
4112 * 1 for the dir index
4113 * 1 for the inode ref
4114 * 1 for the inode
4115 */
4116 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4117 }
4118
4119 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4120 {
4121 struct btrfs_root *root = BTRFS_I(dir)->root;
4122 struct btrfs_trans_handle *trans;
4123 struct inode *inode = d_inode(dentry);
4124 int ret;
4125
4126 trans = __unlink_start_trans(dir);
4127 if (IS_ERR(trans))
4128 return PTR_ERR(trans);
4129
4130 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4131
4132 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4133 dentry->d_name.name, dentry->d_name.len);
4134 if (ret)
4135 goto out;
4136
4137 if (inode->i_nlink == 0) {
4138 ret = btrfs_orphan_add(trans, inode);
4139 if (ret)
4140 goto out;
4141 }
4142
4143 out:
4144 btrfs_end_transaction(trans);
4145 btrfs_btree_balance_dirty(root->fs_info);
4146 return ret;
4147 }
4148
4149 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4150 struct btrfs_root *root,
4151 struct inode *dir, u64 objectid,
4152 const char *name, int name_len)
4153 {
4154 struct btrfs_fs_info *fs_info = root->fs_info;
4155 struct btrfs_path *path;
4156 struct extent_buffer *leaf;
4157 struct btrfs_dir_item *di;
4158 struct btrfs_key key;
4159 u64 index;
4160 int ret;
4161 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4162
4163 path = btrfs_alloc_path();
4164 if (!path)
4165 return -ENOMEM;
4166
4167 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4168 name, name_len, -1);
4169 if (IS_ERR_OR_NULL(di)) {
4170 if (!di)
4171 ret = -ENOENT;
4172 else
4173 ret = PTR_ERR(di);
4174 goto out;
4175 }
4176
4177 leaf = path->nodes[0];
4178 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4179 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4180 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4181 if (ret) {
4182 btrfs_abort_transaction(trans, ret);
4183 goto out;
4184 }
4185 btrfs_release_path(path);
4186
4187 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4188 root->root_key.objectid, dir_ino,
4189 &index, name, name_len);
4190 if (ret < 0) {
4191 if (ret != -ENOENT) {
4192 btrfs_abort_transaction(trans, ret);
4193 goto out;
4194 }
4195 di = btrfs_search_dir_index_item(root, path, dir_ino,
4196 name, name_len);
4197 if (IS_ERR_OR_NULL(di)) {
4198 if (!di)
4199 ret = -ENOENT;
4200 else
4201 ret = PTR_ERR(di);
4202 btrfs_abort_transaction(trans, ret);
4203 goto out;
4204 }
4205
4206 leaf = path->nodes[0];
4207 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4208 btrfs_release_path(path);
4209 index = key.offset;
4210 }
4211 btrfs_release_path(path);
4212
4213 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4214 if (ret) {
4215 btrfs_abort_transaction(trans, ret);
4216 goto out;
4217 }
4218
4219 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4220 inode_inc_iversion(dir);
4221 dir->i_mtime = dir->i_ctime = current_time(dir);
4222 ret = btrfs_update_inode_fallback(trans, root, dir);
4223 if (ret)
4224 btrfs_abort_transaction(trans, ret);
4225 out:
4226 btrfs_free_path(path);
4227 return ret;
4228 }
4229
4230 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4231 {
4232 struct inode *inode = d_inode(dentry);
4233 int err = 0;
4234 struct btrfs_root *root = BTRFS_I(dir)->root;
4235 struct btrfs_trans_handle *trans;
4236 u64 last_unlink_trans;
4237
4238 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4239 return -ENOTEMPTY;
4240 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4241 return -EPERM;
4242
4243 trans = __unlink_start_trans(dir);
4244 if (IS_ERR(trans))
4245 return PTR_ERR(trans);
4246
4247 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4248 err = btrfs_unlink_subvol(trans, root, dir,
4249 BTRFS_I(inode)->location.objectid,
4250 dentry->d_name.name,
4251 dentry->d_name.len);
4252 goto out;
4253 }
4254
4255 err = btrfs_orphan_add(trans, inode);
4256 if (err)
4257 goto out;
4258
4259 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4260
4261 /* now the directory is empty */
4262 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4263 dentry->d_name.name, dentry->d_name.len);
4264 if (!err) {
4265 btrfs_i_size_write(inode, 0);
4266 /*
4267 * Propagate the last_unlink_trans value of the deleted dir to
4268 * its parent directory. This is to prevent an unrecoverable
4269 * log tree in the case we do something like this:
4270 * 1) create dir foo
4271 * 2) create snapshot under dir foo
4272 * 3) delete the snapshot
4273 * 4) rmdir foo
4274 * 5) mkdir foo
4275 * 6) fsync foo or some file inside foo
4276 */
4277 if (last_unlink_trans >= trans->transid)
4278 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4279 }
4280 out:
4281 btrfs_end_transaction(trans);
4282 btrfs_btree_balance_dirty(root->fs_info);
4283
4284 return err;
4285 }
4286
4287 static int truncate_space_check(struct btrfs_trans_handle *trans,
4288 struct btrfs_root *root,
4289 u64 bytes_deleted)
4290 {
4291 struct btrfs_fs_info *fs_info = root->fs_info;
4292 int ret;
4293
4294 /*
4295 * This is only used to apply pressure to the enospc system, we don't
4296 * intend to use this reservation at all.
4297 */
4298 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4299 bytes_deleted *= fs_info->nodesize;
4300 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4301 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4302 if (!ret) {
4303 trace_btrfs_space_reservation(fs_info, "transaction",
4304 trans->transid,
4305 bytes_deleted, 1);
4306 trans->bytes_reserved += bytes_deleted;
4307 }
4308 return ret;
4309
4310 }
4311
4312 static int truncate_inline_extent(struct inode *inode,
4313 struct btrfs_path *path,
4314 struct btrfs_key *found_key,
4315 const u64 item_end,
4316 const u64 new_size)
4317 {
4318 struct extent_buffer *leaf = path->nodes[0];
4319 int slot = path->slots[0];
4320 struct btrfs_file_extent_item *fi;
4321 u32 size = (u32)(new_size - found_key->offset);
4322 struct btrfs_root *root = BTRFS_I(inode)->root;
4323
4324 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4325
4326 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4327 loff_t offset = new_size;
4328 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4329
4330 /*
4331 * Zero out the remaining of the last page of our inline extent,
4332 * instead of directly truncating our inline extent here - that
4333 * would be much more complex (decompressing all the data, then
4334 * compressing the truncated data, which might be bigger than
4335 * the size of the inline extent, resize the extent, etc).
4336 * We release the path because to get the page we might need to
4337 * read the extent item from disk (data not in the page cache).
4338 */
4339 btrfs_release_path(path);
4340 return btrfs_truncate_block(inode, offset, page_end - offset,
4341 0);
4342 }
4343
4344 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4345 size = btrfs_file_extent_calc_inline_size(size);
4346 btrfs_truncate_item(root->fs_info, path, size, 1);
4347
4348 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4349 inode_sub_bytes(inode, item_end + 1 - new_size);
4350
4351 return 0;
4352 }
4353
4354 /*
4355 * this can truncate away extent items, csum items and directory items.
4356 * It starts at a high offset and removes keys until it can't find
4357 * any higher than new_size
4358 *
4359 * csum items that cross the new i_size are truncated to the new size
4360 * as well.
4361 *
4362 * min_type is the minimum key type to truncate down to. If set to 0, this
4363 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4364 */
4365 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4366 struct btrfs_root *root,
4367 struct inode *inode,
4368 u64 new_size, u32 min_type)
4369 {
4370 struct btrfs_fs_info *fs_info = root->fs_info;
4371 struct btrfs_path *path;
4372 struct extent_buffer *leaf;
4373 struct btrfs_file_extent_item *fi;
4374 struct btrfs_key key;
4375 struct btrfs_key found_key;
4376 u64 extent_start = 0;
4377 u64 extent_num_bytes = 0;
4378 u64 extent_offset = 0;
4379 u64 item_end = 0;
4380 u64 last_size = new_size;
4381 u32 found_type = (u8)-1;
4382 int found_extent;
4383 int del_item;
4384 int pending_del_nr = 0;
4385 int pending_del_slot = 0;
4386 int extent_type = -1;
4387 int ret;
4388 int err = 0;
4389 u64 ino = btrfs_ino(BTRFS_I(inode));
4390 u64 bytes_deleted = 0;
4391 bool be_nice = 0;
4392 bool should_throttle = 0;
4393 bool should_end = 0;
4394
4395 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4396
4397 /*
4398 * for non-free space inodes and ref cows, we want to back off from
4399 * time to time
4400 */
4401 if (!btrfs_is_free_space_inode(inode) &&
4402 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4403 be_nice = 1;
4404
4405 path = btrfs_alloc_path();
4406 if (!path)
4407 return -ENOMEM;
4408 path->reada = READA_BACK;
4409
4410 /*
4411 * We want to drop from the next block forward in case this new size is
4412 * not block aligned since we will be keeping the last block of the
4413 * extent just the way it is.
4414 */
4415 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4416 root == fs_info->tree_root)
4417 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4418 fs_info->sectorsize),
4419 (u64)-1, 0);
4420
4421 /*
4422 * This function is also used to drop the items in the log tree before
4423 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4424 * it is used to drop the loged items. So we shouldn't kill the delayed
4425 * items.
4426 */
4427 if (min_type == 0 && root == BTRFS_I(inode)->root)
4428 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4429
4430 key.objectid = ino;
4431 key.offset = (u64)-1;
4432 key.type = (u8)-1;
4433
4434 search_again:
4435 /*
4436 * with a 16K leaf size and 128MB extents, you can actually queue
4437 * up a huge file in a single leaf. Most of the time that
4438 * bytes_deleted is > 0, it will be huge by the time we get here
4439 */
4440 if (be_nice && bytes_deleted > SZ_32M) {
4441 if (btrfs_should_end_transaction(trans)) {
4442 err = -EAGAIN;
4443 goto error;
4444 }
4445 }
4446
4447
4448 path->leave_spinning = 1;
4449 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4450 if (ret < 0) {
4451 err = ret;
4452 goto out;
4453 }
4454
4455 if (ret > 0) {
4456 /* there are no items in the tree for us to truncate, we're
4457 * done
4458 */
4459 if (path->slots[0] == 0)
4460 goto out;
4461 path->slots[0]--;
4462 }
4463
4464 while (1) {
4465 fi = NULL;
4466 leaf = path->nodes[0];
4467 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4468 found_type = found_key.type;
4469
4470 if (found_key.objectid != ino)
4471 break;
4472
4473 if (found_type < min_type)
4474 break;
4475
4476 item_end = found_key.offset;
4477 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4478 fi = btrfs_item_ptr(leaf, path->slots[0],
4479 struct btrfs_file_extent_item);
4480 extent_type = btrfs_file_extent_type(leaf, fi);
4481 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4482 item_end +=
4483 btrfs_file_extent_num_bytes(leaf, fi);
4484 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4485 item_end += btrfs_file_extent_inline_len(leaf,
4486 path->slots[0], fi);
4487 }
4488 item_end--;
4489 }
4490 if (found_type > min_type) {
4491 del_item = 1;
4492 } else {
4493 if (item_end < new_size) {
4494 /*
4495 * With NO_HOLES mode, for the following mapping
4496 *
4497 * [0-4k][hole][8k-12k]
4498 *
4499 * if truncating isize down to 6k, it ends up
4500 * isize being 8k.
4501 */
4502 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
4503 last_size = new_size;
4504 break;
4505 }
4506 if (found_key.offset >= new_size)
4507 del_item = 1;
4508 else
4509 del_item = 0;
4510 }
4511 found_extent = 0;
4512 /* FIXME, shrink the extent if the ref count is only 1 */
4513 if (found_type != BTRFS_EXTENT_DATA_KEY)
4514 goto delete;
4515
4516 if (del_item)
4517 last_size = found_key.offset;
4518 else
4519 last_size = new_size;
4520
4521 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4522 u64 num_dec;
4523 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4524 if (!del_item) {
4525 u64 orig_num_bytes =
4526 btrfs_file_extent_num_bytes(leaf, fi);
4527 extent_num_bytes = ALIGN(new_size -
4528 found_key.offset,
4529 fs_info->sectorsize);
4530 btrfs_set_file_extent_num_bytes(leaf, fi,
4531 extent_num_bytes);
4532 num_dec = (orig_num_bytes -
4533 extent_num_bytes);
4534 if (test_bit(BTRFS_ROOT_REF_COWS,
4535 &root->state) &&
4536 extent_start != 0)
4537 inode_sub_bytes(inode, num_dec);
4538 btrfs_mark_buffer_dirty(leaf);
4539 } else {
4540 extent_num_bytes =
4541 btrfs_file_extent_disk_num_bytes(leaf,
4542 fi);
4543 extent_offset = found_key.offset -
4544 btrfs_file_extent_offset(leaf, fi);
4545
4546 /* FIXME blocksize != 4096 */
4547 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4548 if (extent_start != 0) {
4549 found_extent = 1;
4550 if (test_bit(BTRFS_ROOT_REF_COWS,
4551 &root->state))
4552 inode_sub_bytes(inode, num_dec);
4553 }
4554 }
4555 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4556 /*
4557 * we can't truncate inline items that have had
4558 * special encodings
4559 */
4560 if (!del_item &&
4561 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4562 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4563
4564 /*
4565 * Need to release path in order to truncate a
4566 * compressed extent. So delete any accumulated
4567 * extent items so far.
4568 */
4569 if (btrfs_file_extent_compression(leaf, fi) !=
4570 BTRFS_COMPRESS_NONE && pending_del_nr) {
4571 err = btrfs_del_items(trans, root, path,
4572 pending_del_slot,
4573 pending_del_nr);
4574 if (err) {
4575 btrfs_abort_transaction(trans,
4576 err);
4577 goto error;
4578 }
4579 pending_del_nr = 0;
4580 }
4581
4582 err = truncate_inline_extent(inode, path,
4583 &found_key,
4584 item_end,
4585 new_size);
4586 if (err) {
4587 btrfs_abort_transaction(trans, err);
4588 goto error;
4589 }
4590 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4591 &root->state)) {
4592 inode_sub_bytes(inode, item_end + 1 - new_size);
4593 }
4594 }
4595 delete:
4596 if (del_item) {
4597 if (!pending_del_nr) {
4598 /* no pending yet, add ourselves */
4599 pending_del_slot = path->slots[0];
4600 pending_del_nr = 1;
4601 } else if (pending_del_nr &&
4602 path->slots[0] + 1 == pending_del_slot) {
4603 /* hop on the pending chunk */
4604 pending_del_nr++;
4605 pending_del_slot = path->slots[0];
4606 } else {
4607 BUG();
4608 }
4609 } else {
4610 break;
4611 }
4612 should_throttle = 0;
4613
4614 if (found_extent &&
4615 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4616 root == fs_info->tree_root)) {
4617 btrfs_set_path_blocking(path);
4618 bytes_deleted += extent_num_bytes;
4619 ret = btrfs_free_extent(trans, fs_info, extent_start,
4620 extent_num_bytes, 0,
4621 btrfs_header_owner(leaf),
4622 ino, extent_offset);
4623 BUG_ON(ret);
4624 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4625 btrfs_async_run_delayed_refs(fs_info,
4626 trans->delayed_ref_updates * 2,
4627 trans->transid, 0);
4628 if (be_nice) {
4629 if (truncate_space_check(trans, root,
4630 extent_num_bytes)) {
4631 should_end = 1;
4632 }
4633 if (btrfs_should_throttle_delayed_refs(trans,
4634 fs_info))
4635 should_throttle = 1;
4636 }
4637 }
4638
4639 if (found_type == BTRFS_INODE_ITEM_KEY)
4640 break;
4641
4642 if (path->slots[0] == 0 ||
4643 path->slots[0] != pending_del_slot ||
4644 should_throttle || should_end) {
4645 if (pending_del_nr) {
4646 ret = btrfs_del_items(trans, root, path,
4647 pending_del_slot,
4648 pending_del_nr);
4649 if (ret) {
4650 btrfs_abort_transaction(trans, ret);
4651 goto error;
4652 }
4653 pending_del_nr = 0;
4654 }
4655 btrfs_release_path(path);
4656 if (should_throttle) {
4657 unsigned long updates = trans->delayed_ref_updates;
4658 if (updates) {
4659 trans->delayed_ref_updates = 0;
4660 ret = btrfs_run_delayed_refs(trans,
4661 fs_info,
4662 updates * 2);
4663 if (ret && !err)
4664 err = ret;
4665 }
4666 }
4667 /*
4668 * if we failed to refill our space rsv, bail out
4669 * and let the transaction restart
4670 */
4671 if (should_end) {
4672 err = -EAGAIN;
4673 goto error;
4674 }
4675 goto search_again;
4676 } else {
4677 path->slots[0]--;
4678 }
4679 }
4680 out:
4681 if (pending_del_nr) {
4682 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4683 pending_del_nr);
4684 if (ret)
4685 btrfs_abort_transaction(trans, ret);
4686 }
4687 error:
4688 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4689 btrfs_ordered_update_i_size(inode, last_size, NULL);
4690
4691 btrfs_free_path(path);
4692
4693 if (be_nice && bytes_deleted > SZ_32M) {
4694 unsigned long updates = trans->delayed_ref_updates;
4695 if (updates) {
4696 trans->delayed_ref_updates = 0;
4697 ret = btrfs_run_delayed_refs(trans, fs_info,
4698 updates * 2);
4699 if (ret && !err)
4700 err = ret;
4701 }
4702 }
4703 return err;
4704 }
4705
4706 /*
4707 * btrfs_truncate_block - read, zero a chunk and write a block
4708 * @inode - inode that we're zeroing
4709 * @from - the offset to start zeroing
4710 * @len - the length to zero, 0 to zero the entire range respective to the
4711 * offset
4712 * @front - zero up to the offset instead of from the offset on
4713 *
4714 * This will find the block for the "from" offset and cow the block and zero the
4715 * part we want to zero. This is used with truncate and hole punching.
4716 */
4717 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4718 int front)
4719 {
4720 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4721 struct address_space *mapping = inode->i_mapping;
4722 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4723 struct btrfs_ordered_extent *ordered;
4724 struct extent_state *cached_state = NULL;
4725 char *kaddr;
4726 u32 blocksize = fs_info->sectorsize;
4727 pgoff_t index = from >> PAGE_SHIFT;
4728 unsigned offset = from & (blocksize - 1);
4729 struct page *page;
4730 gfp_t mask = btrfs_alloc_write_mask(mapping);
4731 int ret = 0;
4732 u64 block_start;
4733 u64 block_end;
4734
4735 if ((offset & (blocksize - 1)) == 0 &&
4736 (!len || ((len & (blocksize - 1)) == 0)))
4737 goto out;
4738
4739 ret = btrfs_delalloc_reserve_space(inode,
4740 round_down(from, blocksize), blocksize);
4741 if (ret)
4742 goto out;
4743
4744 again:
4745 page = find_or_create_page(mapping, index, mask);
4746 if (!page) {
4747 btrfs_delalloc_release_space(inode,
4748 round_down(from, blocksize),
4749 blocksize);
4750 ret = -ENOMEM;
4751 goto out;
4752 }
4753
4754 block_start = round_down(from, blocksize);
4755 block_end = block_start + blocksize - 1;
4756
4757 if (!PageUptodate(page)) {
4758 ret = btrfs_readpage(NULL, page);
4759 lock_page(page);
4760 if (page->mapping != mapping) {
4761 unlock_page(page);
4762 put_page(page);
4763 goto again;
4764 }
4765 if (!PageUptodate(page)) {
4766 ret = -EIO;
4767 goto out_unlock;
4768 }
4769 }
4770 wait_on_page_writeback(page);
4771
4772 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4773 set_page_extent_mapped(page);
4774
4775 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4776 if (ordered) {
4777 unlock_extent_cached(io_tree, block_start, block_end,
4778 &cached_state, GFP_NOFS);
4779 unlock_page(page);
4780 put_page(page);
4781 btrfs_start_ordered_extent(inode, ordered, 1);
4782 btrfs_put_ordered_extent(ordered);
4783 goto again;
4784 }
4785
4786 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4787 EXTENT_DIRTY | EXTENT_DELALLOC |
4788 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4789 0, 0, &cached_state, GFP_NOFS);
4790
4791 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4792 &cached_state, 0);
4793 if (ret) {
4794 unlock_extent_cached(io_tree, block_start, block_end,
4795 &cached_state, GFP_NOFS);
4796 goto out_unlock;
4797 }
4798
4799 if (offset != blocksize) {
4800 if (!len)
4801 len = blocksize - offset;
4802 kaddr = kmap(page);
4803 if (front)
4804 memset(kaddr + (block_start - page_offset(page)),
4805 0, offset);
4806 else
4807 memset(kaddr + (block_start - page_offset(page)) + offset,
4808 0, len);
4809 flush_dcache_page(page);
4810 kunmap(page);
4811 }
4812 ClearPageChecked(page);
4813 set_page_dirty(page);
4814 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4815 GFP_NOFS);
4816
4817 out_unlock:
4818 if (ret)
4819 btrfs_delalloc_release_space(inode, block_start,
4820 blocksize);
4821 unlock_page(page);
4822 put_page(page);
4823 out:
4824 return ret;
4825 }
4826
4827 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4828 u64 offset, u64 len)
4829 {
4830 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4831 struct btrfs_trans_handle *trans;
4832 int ret;
4833
4834 /*
4835 * Still need to make sure the inode looks like it's been updated so
4836 * that any holes get logged if we fsync.
4837 */
4838 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4839 BTRFS_I(inode)->last_trans = fs_info->generation;
4840 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4841 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4842 return 0;
4843 }
4844
4845 /*
4846 * 1 - for the one we're dropping
4847 * 1 - for the one we're adding
4848 * 1 - for updating the inode.
4849 */
4850 trans = btrfs_start_transaction(root, 3);
4851 if (IS_ERR(trans))
4852 return PTR_ERR(trans);
4853
4854 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4855 if (ret) {
4856 btrfs_abort_transaction(trans, ret);
4857 btrfs_end_transaction(trans);
4858 return ret;
4859 }
4860
4861 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)), offset,
4862 0, 0, len, 0, len, 0, 0, 0);
4863 if (ret)
4864 btrfs_abort_transaction(trans, ret);
4865 else
4866 btrfs_update_inode(trans, root, inode);
4867 btrfs_end_transaction(trans);
4868 return ret;
4869 }
4870
4871 /*
4872 * This function puts in dummy file extents for the area we're creating a hole
4873 * for. So if we are truncating this file to a larger size we need to insert
4874 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4875 * the range between oldsize and size
4876 */
4877 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4878 {
4879 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4880 struct btrfs_root *root = BTRFS_I(inode)->root;
4881 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4882 struct extent_map *em = NULL;
4883 struct extent_state *cached_state = NULL;
4884 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4885 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4886 u64 block_end = ALIGN(size, fs_info->sectorsize);
4887 u64 last_byte;
4888 u64 cur_offset;
4889 u64 hole_size;
4890 int err = 0;
4891
4892 /*
4893 * If our size started in the middle of a block we need to zero out the
4894 * rest of the block before we expand the i_size, otherwise we could
4895 * expose stale data.
4896 */
4897 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4898 if (err)
4899 return err;
4900
4901 if (size <= hole_start)
4902 return 0;
4903
4904 while (1) {
4905 struct btrfs_ordered_extent *ordered;
4906
4907 lock_extent_bits(io_tree, hole_start, block_end - 1,
4908 &cached_state);
4909 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4910 block_end - hole_start);
4911 if (!ordered)
4912 break;
4913 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4914 &cached_state, GFP_NOFS);
4915 btrfs_start_ordered_extent(inode, ordered, 1);
4916 btrfs_put_ordered_extent(ordered);
4917 }
4918
4919 cur_offset = hole_start;
4920 while (1) {
4921 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4922 block_end - cur_offset, 0);
4923 if (IS_ERR(em)) {
4924 err = PTR_ERR(em);
4925 em = NULL;
4926 break;
4927 }
4928 last_byte = min(extent_map_end(em), block_end);
4929 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4930 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4931 struct extent_map *hole_em;
4932 hole_size = last_byte - cur_offset;
4933
4934 err = maybe_insert_hole(root, inode, cur_offset,
4935 hole_size);
4936 if (err)
4937 break;
4938 btrfs_drop_extent_cache(inode, cur_offset,
4939 cur_offset + hole_size - 1, 0);
4940 hole_em = alloc_extent_map();
4941 if (!hole_em) {
4942 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4943 &BTRFS_I(inode)->runtime_flags);
4944 goto next;
4945 }
4946 hole_em->start = cur_offset;
4947 hole_em->len = hole_size;
4948 hole_em->orig_start = cur_offset;
4949
4950 hole_em->block_start = EXTENT_MAP_HOLE;
4951 hole_em->block_len = 0;
4952 hole_em->orig_block_len = 0;
4953 hole_em->ram_bytes = hole_size;
4954 hole_em->bdev = fs_info->fs_devices->latest_bdev;
4955 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4956 hole_em->generation = fs_info->generation;
4957
4958 while (1) {
4959 write_lock(&em_tree->lock);
4960 err = add_extent_mapping(em_tree, hole_em, 1);
4961 write_unlock(&em_tree->lock);
4962 if (err != -EEXIST)
4963 break;
4964 btrfs_drop_extent_cache(inode, cur_offset,
4965 cur_offset +
4966 hole_size - 1, 0);
4967 }
4968 free_extent_map(hole_em);
4969 }
4970 next:
4971 free_extent_map(em);
4972 em = NULL;
4973 cur_offset = last_byte;
4974 if (cur_offset >= block_end)
4975 break;
4976 }
4977 free_extent_map(em);
4978 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4979 GFP_NOFS);
4980 return err;
4981 }
4982
4983 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4984 {
4985 struct btrfs_root *root = BTRFS_I(inode)->root;
4986 struct btrfs_trans_handle *trans;
4987 loff_t oldsize = i_size_read(inode);
4988 loff_t newsize = attr->ia_size;
4989 int mask = attr->ia_valid;
4990 int ret;
4991
4992 /*
4993 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4994 * special case where we need to update the times despite not having
4995 * these flags set. For all other operations the VFS set these flags
4996 * explicitly if it wants a timestamp update.
4997 */
4998 if (newsize != oldsize) {
4999 inode_inc_iversion(inode);
5000 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5001 inode->i_ctime = inode->i_mtime =
5002 current_time(inode);
5003 }
5004
5005 if (newsize > oldsize) {
5006 /*
5007 * Don't do an expanding truncate while snapshoting is ongoing.
5008 * This is to ensure the snapshot captures a fully consistent
5009 * state of this file - if the snapshot captures this expanding
5010 * truncation, it must capture all writes that happened before
5011 * this truncation.
5012 */
5013 btrfs_wait_for_snapshot_creation(root);
5014 ret = btrfs_cont_expand(inode, oldsize, newsize);
5015 if (ret) {
5016 btrfs_end_write_no_snapshoting(root);
5017 return ret;
5018 }
5019
5020 trans = btrfs_start_transaction(root, 1);
5021 if (IS_ERR(trans)) {
5022 btrfs_end_write_no_snapshoting(root);
5023 return PTR_ERR(trans);
5024 }
5025
5026 i_size_write(inode, newsize);
5027 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5028 pagecache_isize_extended(inode, oldsize, newsize);
5029 ret = btrfs_update_inode(trans, root, inode);
5030 btrfs_end_write_no_snapshoting(root);
5031 btrfs_end_transaction(trans);
5032 } else {
5033
5034 /*
5035 * We're truncating a file that used to have good data down to
5036 * zero. Make sure it gets into the ordered flush list so that
5037 * any new writes get down to disk quickly.
5038 */
5039 if (newsize == 0)
5040 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5041 &BTRFS_I(inode)->runtime_flags);
5042
5043 /*
5044 * 1 for the orphan item we're going to add
5045 * 1 for the orphan item deletion.
5046 */
5047 trans = btrfs_start_transaction(root, 2);
5048 if (IS_ERR(trans))
5049 return PTR_ERR(trans);
5050
5051 /*
5052 * We need to do this in case we fail at _any_ point during the
5053 * actual truncate. Once we do the truncate_setsize we could
5054 * invalidate pages which forces any outstanding ordered io to
5055 * be instantly completed which will give us extents that need
5056 * to be truncated. If we fail to get an orphan inode down we
5057 * could have left over extents that were never meant to live,
5058 * so we need to guarantee from this point on that everything
5059 * will be consistent.
5060 */
5061 ret = btrfs_orphan_add(trans, inode);
5062 btrfs_end_transaction(trans);
5063 if (ret)
5064 return ret;
5065
5066 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5067 truncate_setsize(inode, newsize);
5068
5069 /* Disable nonlocked read DIO to avoid the end less truncate */
5070 btrfs_inode_block_unlocked_dio(inode);
5071 inode_dio_wait(inode);
5072 btrfs_inode_resume_unlocked_dio(inode);
5073
5074 ret = btrfs_truncate(inode);
5075 if (ret && inode->i_nlink) {
5076 int err;
5077
5078 /*
5079 * failed to truncate, disk_i_size is only adjusted down
5080 * as we remove extents, so it should represent the true
5081 * size of the inode, so reset the in memory size and
5082 * delete our orphan entry.
5083 */
5084 trans = btrfs_join_transaction(root);
5085 if (IS_ERR(trans)) {
5086 btrfs_orphan_del(NULL, inode);
5087 return ret;
5088 }
5089 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5090 err = btrfs_orphan_del(trans, inode);
5091 if (err)
5092 btrfs_abort_transaction(trans, err);
5093 btrfs_end_transaction(trans);
5094 }
5095 }
5096
5097 return ret;
5098 }
5099
5100 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5101 {
5102 struct inode *inode = d_inode(dentry);
5103 struct btrfs_root *root = BTRFS_I(inode)->root;
5104 int err;
5105
5106 if (btrfs_root_readonly(root))
5107 return -EROFS;
5108
5109 err = setattr_prepare(dentry, attr);
5110 if (err)
5111 return err;
5112
5113 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5114 err = btrfs_setsize(inode, attr);
5115 if (err)
5116 return err;
5117 }
5118
5119 if (attr->ia_valid) {
5120 setattr_copy(inode, attr);
5121 inode_inc_iversion(inode);
5122 err = btrfs_dirty_inode(inode);
5123
5124 if (!err && attr->ia_valid & ATTR_MODE)
5125 err = posix_acl_chmod(inode, inode->i_mode);
5126 }
5127
5128 return err;
5129 }
5130
5131 /*
5132 * While truncating the inode pages during eviction, we get the VFS calling
5133 * btrfs_invalidatepage() against each page of the inode. This is slow because
5134 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5135 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5136 * extent_state structures over and over, wasting lots of time.
5137 *
5138 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5139 * those expensive operations on a per page basis and do only the ordered io
5140 * finishing, while we release here the extent_map and extent_state structures,
5141 * without the excessive merging and splitting.
5142 */
5143 static void evict_inode_truncate_pages(struct inode *inode)
5144 {
5145 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5146 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5147 struct rb_node *node;
5148
5149 ASSERT(inode->i_state & I_FREEING);
5150 truncate_inode_pages_final(&inode->i_data);
5151
5152 write_lock(&map_tree->lock);
5153 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5154 struct extent_map *em;
5155
5156 node = rb_first(&map_tree->map);
5157 em = rb_entry(node, struct extent_map, rb_node);
5158 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5159 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5160 remove_extent_mapping(map_tree, em);
5161 free_extent_map(em);
5162 if (need_resched()) {
5163 write_unlock(&map_tree->lock);
5164 cond_resched();
5165 write_lock(&map_tree->lock);
5166 }
5167 }
5168 write_unlock(&map_tree->lock);
5169
5170 /*
5171 * Keep looping until we have no more ranges in the io tree.
5172 * We can have ongoing bios started by readpages (called from readahead)
5173 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5174 * still in progress (unlocked the pages in the bio but did not yet
5175 * unlocked the ranges in the io tree). Therefore this means some
5176 * ranges can still be locked and eviction started because before
5177 * submitting those bios, which are executed by a separate task (work
5178 * queue kthread), inode references (inode->i_count) were not taken
5179 * (which would be dropped in the end io callback of each bio).
5180 * Therefore here we effectively end up waiting for those bios and
5181 * anyone else holding locked ranges without having bumped the inode's
5182 * reference count - if we don't do it, when they access the inode's
5183 * io_tree to unlock a range it may be too late, leading to an
5184 * use-after-free issue.
5185 */
5186 spin_lock(&io_tree->lock);
5187 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5188 struct extent_state *state;
5189 struct extent_state *cached_state = NULL;
5190 u64 start;
5191 u64 end;
5192
5193 node = rb_first(&io_tree->state);
5194 state = rb_entry(node, struct extent_state, rb_node);
5195 start = state->start;
5196 end = state->end;
5197 spin_unlock(&io_tree->lock);
5198
5199 lock_extent_bits(io_tree, start, end, &cached_state);
5200
5201 /*
5202 * If still has DELALLOC flag, the extent didn't reach disk,
5203 * and its reserved space won't be freed by delayed_ref.
5204 * So we need to free its reserved space here.
5205 * (Refer to comment in btrfs_invalidatepage, case 2)
5206 *
5207 * Note, end is the bytenr of last byte, so we need + 1 here.
5208 */
5209 if (state->state & EXTENT_DELALLOC)
5210 btrfs_qgroup_free_data(inode, start, end - start + 1);
5211
5212 clear_extent_bit(io_tree, start, end,
5213 EXTENT_LOCKED | EXTENT_DIRTY |
5214 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5215 EXTENT_DEFRAG, 1, 1,
5216 &cached_state, GFP_NOFS);
5217
5218 cond_resched();
5219 spin_lock(&io_tree->lock);
5220 }
5221 spin_unlock(&io_tree->lock);
5222 }
5223
5224 void btrfs_evict_inode(struct inode *inode)
5225 {
5226 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5227 struct btrfs_trans_handle *trans;
5228 struct btrfs_root *root = BTRFS_I(inode)->root;
5229 struct btrfs_block_rsv *rsv, *global_rsv;
5230 int steal_from_global = 0;
5231 u64 min_size;
5232 int ret;
5233
5234 trace_btrfs_inode_evict(inode);
5235
5236 if (!root) {
5237 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5238 return;
5239 }
5240
5241 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5242
5243 evict_inode_truncate_pages(inode);
5244
5245 if (inode->i_nlink &&
5246 ((btrfs_root_refs(&root->root_item) != 0 &&
5247 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5248 btrfs_is_free_space_inode(inode)))
5249 goto no_delete;
5250
5251 if (is_bad_inode(inode)) {
5252 btrfs_orphan_del(NULL, inode);
5253 goto no_delete;
5254 }
5255 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5256 if (!special_file(inode->i_mode))
5257 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5258
5259 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5260
5261 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5262 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5263 &BTRFS_I(inode)->runtime_flags));
5264 goto no_delete;
5265 }
5266
5267 if (inode->i_nlink > 0) {
5268 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5269 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5270 goto no_delete;
5271 }
5272
5273 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5274 if (ret) {
5275 btrfs_orphan_del(NULL, inode);
5276 goto no_delete;
5277 }
5278
5279 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5280 if (!rsv) {
5281 btrfs_orphan_del(NULL, inode);
5282 goto no_delete;
5283 }
5284 rsv->size = min_size;
5285 rsv->failfast = 1;
5286 global_rsv = &fs_info->global_block_rsv;
5287
5288 btrfs_i_size_write(inode, 0);
5289
5290 /*
5291 * This is a bit simpler than btrfs_truncate since we've already
5292 * reserved our space for our orphan item in the unlink, so we just
5293 * need to reserve some slack space in case we add bytes and update
5294 * inode item when doing the truncate.
5295 */
5296 while (1) {
5297 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5298 BTRFS_RESERVE_FLUSH_LIMIT);
5299
5300 /*
5301 * Try and steal from the global reserve since we will
5302 * likely not use this space anyway, we want to try as
5303 * hard as possible to get this to work.
5304 */
5305 if (ret)
5306 steal_from_global++;
5307 else
5308 steal_from_global = 0;
5309 ret = 0;
5310
5311 /*
5312 * steal_from_global == 0: we reserved stuff, hooray!
5313 * steal_from_global == 1: we didn't reserve stuff, boo!
5314 * steal_from_global == 2: we've committed, still not a lot of
5315 * room but maybe we'll have room in the global reserve this
5316 * time.
5317 * steal_from_global == 3: abandon all hope!
5318 */
5319 if (steal_from_global > 2) {
5320 btrfs_warn(fs_info,
5321 "Could not get space for a delete, will truncate on mount %d",
5322 ret);
5323 btrfs_orphan_del(NULL, inode);
5324 btrfs_free_block_rsv(fs_info, rsv);
5325 goto no_delete;
5326 }
5327
5328 trans = btrfs_join_transaction(root);
5329 if (IS_ERR(trans)) {
5330 btrfs_orphan_del(NULL, inode);
5331 btrfs_free_block_rsv(fs_info, rsv);
5332 goto no_delete;
5333 }
5334
5335 /*
5336 * We can't just steal from the global reserve, we need to make
5337 * sure there is room to do it, if not we need to commit and try
5338 * again.
5339 */
5340 if (steal_from_global) {
5341 if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5342 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5343 min_size, 0);
5344 else
5345 ret = -ENOSPC;
5346 }
5347
5348 /*
5349 * Couldn't steal from the global reserve, we have too much
5350 * pending stuff built up, commit the transaction and try it
5351 * again.
5352 */
5353 if (ret) {
5354 ret = btrfs_commit_transaction(trans);
5355 if (ret) {
5356 btrfs_orphan_del(NULL, inode);
5357 btrfs_free_block_rsv(fs_info, rsv);
5358 goto no_delete;
5359 }
5360 continue;
5361 } else {
5362 steal_from_global = 0;
5363 }
5364
5365 trans->block_rsv = rsv;
5366
5367 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5368 if (ret != -ENOSPC && ret != -EAGAIN)
5369 break;
5370
5371 trans->block_rsv = &fs_info->trans_block_rsv;
5372 btrfs_end_transaction(trans);
5373 trans = NULL;
5374 btrfs_btree_balance_dirty(fs_info);
5375 }
5376
5377 btrfs_free_block_rsv(fs_info, rsv);
5378
5379 /*
5380 * Errors here aren't a big deal, it just means we leave orphan items
5381 * in the tree. They will be cleaned up on the next mount.
5382 */
5383 if (ret == 0) {
5384 trans->block_rsv = root->orphan_block_rsv;
5385 btrfs_orphan_del(trans, inode);
5386 } else {
5387 btrfs_orphan_del(NULL, inode);
5388 }
5389
5390 trans->block_rsv = &fs_info->trans_block_rsv;
5391 if (!(root == fs_info->tree_root ||
5392 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5393 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5394
5395 btrfs_end_transaction(trans);
5396 btrfs_btree_balance_dirty(fs_info);
5397 no_delete:
5398 btrfs_remove_delayed_node(BTRFS_I(inode));
5399 clear_inode(inode);
5400 }
5401
5402 /*
5403 * this returns the key found in the dir entry in the location pointer.
5404 * If no dir entries were found, location->objectid is 0.
5405 */
5406 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5407 struct btrfs_key *location)
5408 {
5409 const char *name = dentry->d_name.name;
5410 int namelen = dentry->d_name.len;
5411 struct btrfs_dir_item *di;
5412 struct btrfs_path *path;
5413 struct btrfs_root *root = BTRFS_I(dir)->root;
5414 int ret = 0;
5415
5416 path = btrfs_alloc_path();
5417 if (!path)
5418 return -ENOMEM;
5419
5420 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)), name,
5421 namelen, 0);
5422 if (IS_ERR(di))
5423 ret = PTR_ERR(di);
5424
5425 if (IS_ERR_OR_NULL(di))
5426 goto out_err;
5427
5428 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5429 out:
5430 btrfs_free_path(path);
5431 return ret;
5432 out_err:
5433 location->objectid = 0;
5434 goto out;
5435 }
5436
5437 /*
5438 * when we hit a tree root in a directory, the btrfs part of the inode
5439 * needs to be changed to reflect the root directory of the tree root. This
5440 * is kind of like crossing a mount point.
5441 */
5442 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5443 struct inode *dir,
5444 struct dentry *dentry,
5445 struct btrfs_key *location,
5446 struct btrfs_root **sub_root)
5447 {
5448 struct btrfs_path *path;
5449 struct btrfs_root *new_root;
5450 struct btrfs_root_ref *ref;
5451 struct extent_buffer *leaf;
5452 struct btrfs_key key;
5453 int ret;
5454 int err = 0;
5455
5456 path = btrfs_alloc_path();
5457 if (!path) {
5458 err = -ENOMEM;
5459 goto out;
5460 }
5461
5462 err = -ENOENT;
5463 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5464 key.type = BTRFS_ROOT_REF_KEY;
5465 key.offset = location->objectid;
5466
5467 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5468 if (ret) {
5469 if (ret < 0)
5470 err = ret;
5471 goto out;
5472 }
5473
5474 leaf = path->nodes[0];
5475 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5476 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5477 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5478 goto out;
5479
5480 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5481 (unsigned long)(ref + 1),
5482 dentry->d_name.len);
5483 if (ret)
5484 goto out;
5485
5486 btrfs_release_path(path);
5487
5488 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5489 if (IS_ERR(new_root)) {
5490 err = PTR_ERR(new_root);
5491 goto out;
5492 }
5493
5494 *sub_root = new_root;
5495 location->objectid = btrfs_root_dirid(&new_root->root_item);
5496 location->type = BTRFS_INODE_ITEM_KEY;
5497 location->offset = 0;
5498 err = 0;
5499 out:
5500 btrfs_free_path(path);
5501 return err;
5502 }
5503
5504 static void inode_tree_add(struct inode *inode)
5505 {
5506 struct btrfs_root *root = BTRFS_I(inode)->root;
5507 struct btrfs_inode *entry;
5508 struct rb_node **p;
5509 struct rb_node *parent;
5510 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5511 u64 ino = btrfs_ino(BTRFS_I(inode));
5512
5513 if (inode_unhashed(inode))
5514 return;
5515 parent = NULL;
5516 spin_lock(&root->inode_lock);
5517 p = &root->inode_tree.rb_node;
5518 while (*p) {
5519 parent = *p;
5520 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5521
5522 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5523 p = &parent->rb_left;
5524 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5525 p = &parent->rb_right;
5526 else {
5527 WARN_ON(!(entry->vfs_inode.i_state &
5528 (I_WILL_FREE | I_FREEING)));
5529 rb_replace_node(parent, new, &root->inode_tree);
5530 RB_CLEAR_NODE(parent);
5531 spin_unlock(&root->inode_lock);
5532 return;
5533 }
5534 }
5535 rb_link_node(new, parent, p);
5536 rb_insert_color(new, &root->inode_tree);
5537 spin_unlock(&root->inode_lock);
5538 }
5539
5540 static void inode_tree_del(struct inode *inode)
5541 {
5542 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5543 struct btrfs_root *root = BTRFS_I(inode)->root;
5544 int empty = 0;
5545
5546 spin_lock(&root->inode_lock);
5547 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5548 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5549 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5550 empty = RB_EMPTY_ROOT(&root->inode_tree);
5551 }
5552 spin_unlock(&root->inode_lock);
5553
5554 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5555 synchronize_srcu(&fs_info->subvol_srcu);
5556 spin_lock(&root->inode_lock);
5557 empty = RB_EMPTY_ROOT(&root->inode_tree);
5558 spin_unlock(&root->inode_lock);
5559 if (empty)
5560 btrfs_add_dead_root(root);
5561 }
5562 }
5563
5564 void btrfs_invalidate_inodes(struct btrfs_root *root)
5565 {
5566 struct btrfs_fs_info *fs_info = root->fs_info;
5567 struct rb_node *node;
5568 struct rb_node *prev;
5569 struct btrfs_inode *entry;
5570 struct inode *inode;
5571 u64 objectid = 0;
5572
5573 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5574 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5575
5576 spin_lock(&root->inode_lock);
5577 again:
5578 node = root->inode_tree.rb_node;
5579 prev = NULL;
5580 while (node) {
5581 prev = node;
5582 entry = rb_entry(node, struct btrfs_inode, rb_node);
5583
5584 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5585 node = node->rb_left;
5586 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5587 node = node->rb_right;
5588 else
5589 break;
5590 }
5591 if (!node) {
5592 while (prev) {
5593 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5594 if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5595 node = prev;
5596 break;
5597 }
5598 prev = rb_next(prev);
5599 }
5600 }
5601 while (node) {
5602 entry = rb_entry(node, struct btrfs_inode, rb_node);
5603 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5604 inode = igrab(&entry->vfs_inode);
5605 if (inode) {
5606 spin_unlock(&root->inode_lock);
5607 if (atomic_read(&inode->i_count) > 1)
5608 d_prune_aliases(inode);
5609 /*
5610 * btrfs_drop_inode will have it removed from
5611 * the inode cache when its usage count
5612 * hits zero.
5613 */
5614 iput(inode);
5615 cond_resched();
5616 spin_lock(&root->inode_lock);
5617 goto again;
5618 }
5619
5620 if (cond_resched_lock(&root->inode_lock))
5621 goto again;
5622
5623 node = rb_next(node);
5624 }
5625 spin_unlock(&root->inode_lock);
5626 }
5627
5628 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5629 {
5630 struct btrfs_iget_args *args = p;
5631 inode->i_ino = args->location->objectid;
5632 memcpy(&BTRFS_I(inode)->location, args->location,
5633 sizeof(*args->location));
5634 BTRFS_I(inode)->root = args->root;
5635 return 0;
5636 }
5637
5638 static int btrfs_find_actor(struct inode *inode, void *opaque)
5639 {
5640 struct btrfs_iget_args *args = opaque;
5641 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5642 args->root == BTRFS_I(inode)->root;
5643 }
5644
5645 static struct inode *btrfs_iget_locked(struct super_block *s,
5646 struct btrfs_key *location,
5647 struct btrfs_root *root)
5648 {
5649 struct inode *inode;
5650 struct btrfs_iget_args args;
5651 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5652
5653 args.location = location;
5654 args.root = root;
5655
5656 inode = iget5_locked(s, hashval, btrfs_find_actor,
5657 btrfs_init_locked_inode,
5658 (void *)&args);
5659 return inode;
5660 }
5661
5662 /* Get an inode object given its location and corresponding root.
5663 * Returns in *is_new if the inode was read from disk
5664 */
5665 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5666 struct btrfs_root *root, int *new)
5667 {
5668 struct inode *inode;
5669
5670 inode = btrfs_iget_locked(s, location, root);
5671 if (!inode)
5672 return ERR_PTR(-ENOMEM);
5673
5674 if (inode->i_state & I_NEW) {
5675 int ret;
5676
5677 ret = btrfs_read_locked_inode(inode);
5678 if (!is_bad_inode(inode)) {
5679 inode_tree_add(inode);
5680 unlock_new_inode(inode);
5681 if (new)
5682 *new = 1;
5683 } else {
5684 unlock_new_inode(inode);
5685 iput(inode);
5686 ASSERT(ret < 0);
5687 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5688 }
5689 }
5690
5691 return inode;
5692 }
5693
5694 static struct inode *new_simple_dir(struct super_block *s,
5695 struct btrfs_key *key,
5696 struct btrfs_root *root)
5697 {
5698 struct inode *inode = new_inode(s);
5699
5700 if (!inode)
5701 return ERR_PTR(-ENOMEM);
5702
5703 BTRFS_I(inode)->root = root;
5704 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5705 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5706
5707 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5708 inode->i_op = &btrfs_dir_ro_inode_operations;
5709 inode->i_opflags &= ~IOP_XATTR;
5710 inode->i_fop = &simple_dir_operations;
5711 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5712 inode->i_mtime = current_time(inode);
5713 inode->i_atime = inode->i_mtime;
5714 inode->i_ctime = inode->i_mtime;
5715 BTRFS_I(inode)->i_otime = inode->i_mtime;
5716
5717 return inode;
5718 }
5719
5720 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5721 {
5722 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5723 struct inode *inode;
5724 struct btrfs_root *root = BTRFS_I(dir)->root;
5725 struct btrfs_root *sub_root = root;
5726 struct btrfs_key location;
5727 int index;
5728 int ret = 0;
5729
5730 if (dentry->d_name.len > BTRFS_NAME_LEN)
5731 return ERR_PTR(-ENAMETOOLONG);
5732
5733 ret = btrfs_inode_by_name(dir, dentry, &location);
5734 if (ret < 0)
5735 return ERR_PTR(ret);
5736
5737 if (location.objectid == 0)
5738 return ERR_PTR(-ENOENT);
5739
5740 if (location.type == BTRFS_INODE_ITEM_KEY) {
5741 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5742 return inode;
5743 }
5744
5745 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5746
5747 index = srcu_read_lock(&fs_info->subvol_srcu);
5748 ret = fixup_tree_root_location(fs_info, dir, dentry,
5749 &location, &sub_root);
5750 if (ret < 0) {
5751 if (ret != -ENOENT)
5752 inode = ERR_PTR(ret);
5753 else
5754 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5755 } else {
5756 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5757 }
5758 srcu_read_unlock(&fs_info->subvol_srcu, index);
5759
5760 if (!IS_ERR(inode) && root != sub_root) {
5761 down_read(&fs_info->cleanup_work_sem);
5762 if (!(inode->i_sb->s_flags & MS_RDONLY))
5763 ret = btrfs_orphan_cleanup(sub_root);
5764 up_read(&fs_info->cleanup_work_sem);
5765 if (ret) {
5766 iput(inode);
5767 inode = ERR_PTR(ret);
5768 }
5769 }
5770
5771 return inode;
5772 }
5773
5774 static int btrfs_dentry_delete(const struct dentry *dentry)
5775 {
5776 struct btrfs_root *root;
5777 struct inode *inode = d_inode(dentry);
5778
5779 if (!inode && !IS_ROOT(dentry))
5780 inode = d_inode(dentry->d_parent);
5781
5782 if (inode) {
5783 root = BTRFS_I(inode)->root;
5784 if (btrfs_root_refs(&root->root_item) == 0)
5785 return 1;
5786
5787 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5788 return 1;
5789 }
5790 return 0;
5791 }
5792
5793 static void btrfs_dentry_release(struct dentry *dentry)
5794 {
5795 kfree(dentry->d_fsdata);
5796 }
5797
5798 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5799 unsigned int flags)
5800 {
5801 struct inode *inode;
5802
5803 inode = btrfs_lookup_dentry(dir, dentry);
5804 if (IS_ERR(inode)) {
5805 if (PTR_ERR(inode) == -ENOENT)
5806 inode = NULL;
5807 else
5808 return ERR_CAST(inode);
5809 }
5810
5811 return d_splice_alias(inode, dentry);
5812 }
5813
5814 unsigned char btrfs_filetype_table[] = {
5815 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5816 };
5817
5818 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5819 {
5820 struct inode *inode = file_inode(file);
5821 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5822 struct btrfs_root *root = BTRFS_I(inode)->root;
5823 struct btrfs_item *item;
5824 struct btrfs_dir_item *di;
5825 struct btrfs_key key;
5826 struct btrfs_key found_key;
5827 struct btrfs_path *path;
5828 struct list_head ins_list;
5829 struct list_head del_list;
5830 int ret;
5831 struct extent_buffer *leaf;
5832 int slot;
5833 unsigned char d_type;
5834 int over = 0;
5835 char tmp_name[32];
5836 char *name_ptr;
5837 int name_len;
5838 bool put = false;
5839 struct btrfs_key location;
5840
5841 if (!dir_emit_dots(file, ctx))
5842 return 0;
5843
5844 path = btrfs_alloc_path();
5845 if (!path)
5846 return -ENOMEM;
5847
5848 path->reada = READA_FORWARD;
5849
5850 INIT_LIST_HEAD(&ins_list);
5851 INIT_LIST_HEAD(&del_list);
5852 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5853
5854 key.type = BTRFS_DIR_INDEX_KEY;
5855 key.offset = ctx->pos;
5856 key.objectid = btrfs_ino(BTRFS_I(inode));
5857
5858 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5859 if (ret < 0)
5860 goto err;
5861
5862 while (1) {
5863 leaf = path->nodes[0];
5864 slot = path->slots[0];
5865 if (slot >= btrfs_header_nritems(leaf)) {
5866 ret = btrfs_next_leaf(root, path);
5867 if (ret < 0)
5868 goto err;
5869 else if (ret > 0)
5870 break;
5871 continue;
5872 }
5873
5874 item = btrfs_item_nr(slot);
5875 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5876
5877 if (found_key.objectid != key.objectid)
5878 break;
5879 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5880 break;
5881 if (found_key.offset < ctx->pos)
5882 goto next;
5883 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5884 goto next;
5885
5886 ctx->pos = found_key.offset;
5887
5888 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5889 if (verify_dir_item(fs_info, leaf, di))
5890 goto next;
5891
5892 name_len = btrfs_dir_name_len(leaf, di);
5893 if (name_len <= sizeof(tmp_name)) {
5894 name_ptr = tmp_name;
5895 } else {
5896 name_ptr = kmalloc(name_len, GFP_KERNEL);
5897 if (!name_ptr) {
5898 ret = -ENOMEM;
5899 goto err;
5900 }
5901 }
5902 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5903 name_len);
5904
5905 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5906 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5907
5908 over = !dir_emit(ctx, name_ptr, name_len, location.objectid,
5909 d_type);
5910
5911 if (name_ptr != tmp_name)
5912 kfree(name_ptr);
5913
5914 if (over)
5915 goto nopos;
5916 ctx->pos++;
5917 next:
5918 path->slots[0]++;
5919 }
5920
5921 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5922 if (ret)
5923 goto nopos;
5924
5925 /*
5926 * Stop new entries from being returned after we return the last
5927 * entry.
5928 *
5929 * New directory entries are assigned a strictly increasing
5930 * offset. This means that new entries created during readdir
5931 * are *guaranteed* to be seen in the future by that readdir.
5932 * This has broken buggy programs which operate on names as
5933 * they're returned by readdir. Until we re-use freed offsets
5934 * we have this hack to stop new entries from being returned
5935 * under the assumption that they'll never reach this huge
5936 * offset.
5937 *
5938 * This is being careful not to overflow 32bit loff_t unless the
5939 * last entry requires it because doing so has broken 32bit apps
5940 * in the past.
5941 */
5942 if (ctx->pos >= INT_MAX)
5943 ctx->pos = LLONG_MAX;
5944 else
5945 ctx->pos = INT_MAX;
5946 nopos:
5947 ret = 0;
5948 err:
5949 if (put)
5950 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5951 btrfs_free_path(path);
5952 return ret;
5953 }
5954
5955 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5956 {
5957 struct btrfs_root *root = BTRFS_I(inode)->root;
5958 struct btrfs_trans_handle *trans;
5959 int ret = 0;
5960 bool nolock = false;
5961
5962 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5963 return 0;
5964
5965 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5966 nolock = true;
5967
5968 if (wbc->sync_mode == WB_SYNC_ALL) {
5969 if (nolock)
5970 trans = btrfs_join_transaction_nolock(root);
5971 else
5972 trans = btrfs_join_transaction(root);
5973 if (IS_ERR(trans))
5974 return PTR_ERR(trans);
5975 ret = btrfs_commit_transaction(trans);
5976 }
5977 return ret;
5978 }
5979
5980 /*
5981 * This is somewhat expensive, updating the tree every time the
5982 * inode changes. But, it is most likely to find the inode in cache.
5983 * FIXME, needs more benchmarking...there are no reasons other than performance
5984 * to keep or drop this code.
5985 */
5986 static int btrfs_dirty_inode(struct inode *inode)
5987 {
5988 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5989 struct btrfs_root *root = BTRFS_I(inode)->root;
5990 struct btrfs_trans_handle *trans;
5991 int ret;
5992
5993 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5994 return 0;
5995
5996 trans = btrfs_join_transaction(root);
5997 if (IS_ERR(trans))
5998 return PTR_ERR(trans);
5999
6000 ret = btrfs_update_inode(trans, root, inode);
6001 if (ret && ret == -ENOSPC) {
6002 /* whoops, lets try again with the full transaction */
6003 btrfs_end_transaction(trans);
6004 trans = btrfs_start_transaction(root, 1);
6005 if (IS_ERR(trans))
6006 return PTR_ERR(trans);
6007
6008 ret = btrfs_update_inode(trans, root, inode);
6009 }
6010 btrfs_end_transaction(trans);
6011 if (BTRFS_I(inode)->delayed_node)
6012 btrfs_balance_delayed_items(fs_info);
6013
6014 return ret;
6015 }
6016
6017 /*
6018 * This is a copy of file_update_time. We need this so we can return error on
6019 * ENOSPC for updating the inode in the case of file write and mmap writes.
6020 */
6021 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6022 int flags)
6023 {
6024 struct btrfs_root *root = BTRFS_I(inode)->root;
6025
6026 if (btrfs_root_readonly(root))
6027 return -EROFS;
6028
6029 if (flags & S_VERSION)
6030 inode_inc_iversion(inode);
6031 if (flags & S_CTIME)
6032 inode->i_ctime = *now;
6033 if (flags & S_MTIME)
6034 inode->i_mtime = *now;
6035 if (flags & S_ATIME)
6036 inode->i_atime = *now;
6037 return btrfs_dirty_inode(inode);
6038 }
6039
6040 /*
6041 * find the highest existing sequence number in a directory
6042 * and then set the in-memory index_cnt variable to reflect
6043 * free sequence numbers
6044 */
6045 static int btrfs_set_inode_index_count(struct inode *inode)
6046 {
6047 struct btrfs_root *root = BTRFS_I(inode)->root;
6048 struct btrfs_key key, found_key;
6049 struct btrfs_path *path;
6050 struct extent_buffer *leaf;
6051 int ret;
6052
6053 key.objectid = btrfs_ino(BTRFS_I(inode));
6054 key.type = BTRFS_DIR_INDEX_KEY;
6055 key.offset = (u64)-1;
6056
6057 path = btrfs_alloc_path();
6058 if (!path)
6059 return -ENOMEM;
6060
6061 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6062 if (ret < 0)
6063 goto out;
6064 /* FIXME: we should be able to handle this */
6065 if (ret == 0)
6066 goto out;
6067 ret = 0;
6068
6069 /*
6070 * MAGIC NUMBER EXPLANATION:
6071 * since we search a directory based on f_pos we have to start at 2
6072 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6073 * else has to start at 2
6074 */
6075 if (path->slots[0] == 0) {
6076 BTRFS_I(inode)->index_cnt = 2;
6077 goto out;
6078 }
6079
6080 path->slots[0]--;
6081
6082 leaf = path->nodes[0];
6083 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6084
6085 if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
6086 found_key.type != BTRFS_DIR_INDEX_KEY) {
6087 BTRFS_I(inode)->index_cnt = 2;
6088 goto out;
6089 }
6090
6091 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6092 out:
6093 btrfs_free_path(path);
6094 return ret;
6095 }
6096
6097 /*
6098 * helper to find a free sequence number in a given directory. This current
6099 * code is very simple, later versions will do smarter things in the btree
6100 */
6101 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6102 {
6103 int ret = 0;
6104
6105 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6106 ret = btrfs_inode_delayed_dir_index_count(dir);
6107 if (ret) {
6108 ret = btrfs_set_inode_index_count(dir);
6109 if (ret)
6110 return ret;
6111 }
6112 }
6113
6114 *index = BTRFS_I(dir)->index_cnt;
6115 BTRFS_I(dir)->index_cnt++;
6116
6117 return ret;
6118 }
6119
6120 static int btrfs_insert_inode_locked(struct inode *inode)
6121 {
6122 struct btrfs_iget_args args;
6123 args.location = &BTRFS_I(inode)->location;
6124 args.root = BTRFS_I(inode)->root;
6125
6126 return insert_inode_locked4(inode,
6127 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6128 btrfs_find_actor, &args);
6129 }
6130
6131 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6132 struct btrfs_root *root,
6133 struct inode *dir,
6134 const char *name, int name_len,
6135 u64 ref_objectid, u64 objectid,
6136 umode_t mode, u64 *index)
6137 {
6138 struct btrfs_fs_info *fs_info = root->fs_info;
6139 struct inode *inode;
6140 struct btrfs_inode_item *inode_item;
6141 struct btrfs_key *location;
6142 struct btrfs_path *path;
6143 struct btrfs_inode_ref *ref;
6144 struct btrfs_key key[2];
6145 u32 sizes[2];
6146 int nitems = name ? 2 : 1;
6147 unsigned long ptr;
6148 int ret;
6149
6150 path = btrfs_alloc_path();
6151 if (!path)
6152 return ERR_PTR(-ENOMEM);
6153
6154 inode = new_inode(fs_info->sb);
6155 if (!inode) {
6156 btrfs_free_path(path);
6157 return ERR_PTR(-ENOMEM);
6158 }
6159
6160 /*
6161 * O_TMPFILE, set link count to 0, so that after this point,
6162 * we fill in an inode item with the correct link count.
6163 */
6164 if (!name)
6165 set_nlink(inode, 0);
6166
6167 /*
6168 * we have to initialize this early, so we can reclaim the inode
6169 * number if we fail afterwards in this function.
6170 */
6171 inode->i_ino = objectid;
6172
6173 if (dir && name) {
6174 trace_btrfs_inode_request(dir);
6175
6176 ret = btrfs_set_inode_index(dir, index);
6177 if (ret) {
6178 btrfs_free_path(path);
6179 iput(inode);
6180 return ERR_PTR(ret);
6181 }
6182 } else if (dir) {
6183 *index = 0;
6184 }
6185 /*
6186 * index_cnt is ignored for everything but a dir,
6187 * btrfs_get_inode_index_count has an explanation for the magic
6188 * number
6189 */
6190 BTRFS_I(inode)->index_cnt = 2;
6191 BTRFS_I(inode)->dir_index = *index;
6192 BTRFS_I(inode)->root = root;
6193 BTRFS_I(inode)->generation = trans->transid;
6194 inode->i_generation = BTRFS_I(inode)->generation;
6195
6196 /*
6197 * We could have gotten an inode number from somebody who was fsynced
6198 * and then removed in this same transaction, so let's just set full
6199 * sync since it will be a full sync anyway and this will blow away the
6200 * old info in the log.
6201 */
6202 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6203
6204 key[0].objectid = objectid;
6205 key[0].type = BTRFS_INODE_ITEM_KEY;
6206 key[0].offset = 0;
6207
6208 sizes[0] = sizeof(struct btrfs_inode_item);
6209
6210 if (name) {
6211 /*
6212 * Start new inodes with an inode_ref. This is slightly more
6213 * efficient for small numbers of hard links since they will
6214 * be packed into one item. Extended refs will kick in if we
6215 * add more hard links than can fit in the ref item.
6216 */
6217 key[1].objectid = objectid;
6218 key[1].type = BTRFS_INODE_REF_KEY;
6219 key[1].offset = ref_objectid;
6220
6221 sizes[1] = name_len + sizeof(*ref);
6222 }
6223
6224 location = &BTRFS_I(inode)->location;
6225 location->objectid = objectid;
6226 location->offset = 0;
6227 location->type = BTRFS_INODE_ITEM_KEY;
6228
6229 ret = btrfs_insert_inode_locked(inode);
6230 if (ret < 0)
6231 goto fail;
6232
6233 path->leave_spinning = 1;
6234 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6235 if (ret != 0)
6236 goto fail_unlock;
6237
6238 inode_init_owner(inode, dir, mode);
6239 inode_set_bytes(inode, 0);
6240
6241 inode->i_mtime = current_time(inode);
6242 inode->i_atime = inode->i_mtime;
6243 inode->i_ctime = inode->i_mtime;
6244 BTRFS_I(inode)->i_otime = inode->i_mtime;
6245
6246 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6247 struct btrfs_inode_item);
6248 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6249 sizeof(*inode_item));
6250 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6251
6252 if (name) {
6253 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6254 struct btrfs_inode_ref);
6255 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6256 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6257 ptr = (unsigned long)(ref + 1);
6258 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6259 }
6260
6261 btrfs_mark_buffer_dirty(path->nodes[0]);
6262 btrfs_free_path(path);
6263
6264 btrfs_inherit_iflags(inode, dir);
6265
6266 if (S_ISREG(mode)) {
6267 if (btrfs_test_opt(fs_info, NODATASUM))
6268 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6269 if (btrfs_test_opt(fs_info, NODATACOW))
6270 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6271 BTRFS_INODE_NODATASUM;
6272 }
6273
6274 inode_tree_add(inode);
6275
6276 trace_btrfs_inode_new(inode);
6277 btrfs_set_inode_last_trans(trans, inode);
6278
6279 btrfs_update_root_times(trans, root);
6280
6281 ret = btrfs_inode_inherit_props(trans, inode, dir);
6282 if (ret)
6283 btrfs_err(fs_info,
6284 "error inheriting props for ino %llu (root %llu): %d",
6285 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6286
6287 return inode;
6288
6289 fail_unlock:
6290 unlock_new_inode(inode);
6291 fail:
6292 if (dir && name)
6293 BTRFS_I(dir)->index_cnt--;
6294 btrfs_free_path(path);
6295 iput(inode);
6296 return ERR_PTR(ret);
6297 }
6298
6299 static inline u8 btrfs_inode_type(struct inode *inode)
6300 {
6301 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6302 }
6303
6304 /*
6305 * utility function to add 'inode' into 'parent_inode' with
6306 * a give name and a given sequence number.
6307 * if 'add_backref' is true, also insert a backref from the
6308 * inode to the parent directory.
6309 */
6310 int btrfs_add_link(struct btrfs_trans_handle *trans,
6311 struct inode *parent_inode, struct inode *inode,
6312 const char *name, int name_len, int add_backref, u64 index)
6313 {
6314 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6315 int ret = 0;
6316 struct btrfs_key key;
6317 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6318 u64 ino = btrfs_ino(BTRFS_I(inode));
6319 u64 parent_ino = btrfs_ino(BTRFS_I(parent_inode));
6320
6321 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6322 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6323 } else {
6324 key.objectid = ino;
6325 key.type = BTRFS_INODE_ITEM_KEY;
6326 key.offset = 0;
6327 }
6328
6329 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6330 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6331 root->root_key.objectid, parent_ino,
6332 index, name, name_len);
6333 } else if (add_backref) {
6334 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6335 parent_ino, index);
6336 }
6337
6338 /* Nothing to clean up yet */
6339 if (ret)
6340 return ret;
6341
6342 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6343 parent_inode, &key,
6344 btrfs_inode_type(inode), index);
6345 if (ret == -EEXIST || ret == -EOVERFLOW)
6346 goto fail_dir_item;
6347 else if (ret) {
6348 btrfs_abort_transaction(trans, ret);
6349 return ret;
6350 }
6351
6352 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6353 name_len * 2);
6354 inode_inc_iversion(parent_inode);
6355 parent_inode->i_mtime = parent_inode->i_ctime =
6356 current_time(parent_inode);
6357 ret = btrfs_update_inode(trans, root, parent_inode);
6358 if (ret)
6359 btrfs_abort_transaction(trans, ret);
6360 return ret;
6361
6362 fail_dir_item:
6363 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6364 u64 local_index;
6365 int err;
6366 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6367 root->root_key.objectid, parent_ino,
6368 &local_index, name, name_len);
6369
6370 } else if (add_backref) {
6371 u64 local_index;
6372 int err;
6373
6374 err = btrfs_del_inode_ref(trans, root, name, name_len,
6375 ino, parent_ino, &local_index);
6376 }
6377 return ret;
6378 }
6379
6380 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6381 struct inode *dir, struct dentry *dentry,
6382 struct inode *inode, int backref, u64 index)
6383 {
6384 int err = btrfs_add_link(trans, dir, inode,
6385 dentry->d_name.name, dentry->d_name.len,
6386 backref, index);
6387 if (err > 0)
6388 err = -EEXIST;
6389 return err;
6390 }
6391
6392 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6393 umode_t mode, dev_t rdev)
6394 {
6395 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6396 struct btrfs_trans_handle *trans;
6397 struct btrfs_root *root = BTRFS_I(dir)->root;
6398 struct inode *inode = NULL;
6399 int err;
6400 int drop_inode = 0;
6401 u64 objectid;
6402 u64 index = 0;
6403
6404 /*
6405 * 2 for inode item and ref
6406 * 2 for dir items
6407 * 1 for xattr if selinux is on
6408 */
6409 trans = btrfs_start_transaction(root, 5);
6410 if (IS_ERR(trans))
6411 return PTR_ERR(trans);
6412
6413 err = btrfs_find_free_ino(root, &objectid);
6414 if (err)
6415 goto out_unlock;
6416
6417 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6418 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6419 mode, &index);
6420 if (IS_ERR(inode)) {
6421 err = PTR_ERR(inode);
6422 goto out_unlock;
6423 }
6424
6425 /*
6426 * If the active LSM wants to access the inode during
6427 * d_instantiate it needs these. Smack checks to see
6428 * if the filesystem supports xattrs by looking at the
6429 * ops vector.
6430 */
6431 inode->i_op = &btrfs_special_inode_operations;
6432 init_special_inode(inode, inode->i_mode, rdev);
6433
6434 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6435 if (err)
6436 goto out_unlock_inode;
6437
6438 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6439 if (err) {
6440 goto out_unlock_inode;
6441 } else {
6442 btrfs_update_inode(trans, root, inode);
6443 unlock_new_inode(inode);
6444 d_instantiate(dentry, inode);
6445 }
6446
6447 out_unlock:
6448 btrfs_end_transaction(trans);
6449 btrfs_balance_delayed_items(fs_info);
6450 btrfs_btree_balance_dirty(fs_info);
6451 if (drop_inode) {
6452 inode_dec_link_count(inode);
6453 iput(inode);
6454 }
6455 return err;
6456
6457 out_unlock_inode:
6458 drop_inode = 1;
6459 unlock_new_inode(inode);
6460 goto out_unlock;
6461
6462 }
6463
6464 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6465 umode_t mode, bool excl)
6466 {
6467 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6468 struct btrfs_trans_handle *trans;
6469 struct btrfs_root *root = BTRFS_I(dir)->root;
6470 struct inode *inode = NULL;
6471 int drop_inode_on_err = 0;
6472 int err;
6473 u64 objectid;
6474 u64 index = 0;
6475
6476 /*
6477 * 2 for inode item and ref
6478 * 2 for dir items
6479 * 1 for xattr if selinux is on
6480 */
6481 trans = btrfs_start_transaction(root, 5);
6482 if (IS_ERR(trans))
6483 return PTR_ERR(trans);
6484
6485 err = btrfs_find_free_ino(root, &objectid);
6486 if (err)
6487 goto out_unlock;
6488
6489 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6490 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6491 mode, &index);
6492 if (IS_ERR(inode)) {
6493 err = PTR_ERR(inode);
6494 goto out_unlock;
6495 }
6496 drop_inode_on_err = 1;
6497 /*
6498 * If the active LSM wants to access the inode during
6499 * d_instantiate it needs these. Smack checks to see
6500 * if the filesystem supports xattrs by looking at the
6501 * ops vector.
6502 */
6503 inode->i_fop = &btrfs_file_operations;
6504 inode->i_op = &btrfs_file_inode_operations;
6505 inode->i_mapping->a_ops = &btrfs_aops;
6506
6507 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6508 if (err)
6509 goto out_unlock_inode;
6510
6511 err = btrfs_update_inode(trans, root, inode);
6512 if (err)
6513 goto out_unlock_inode;
6514
6515 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6516 if (err)
6517 goto out_unlock_inode;
6518
6519 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6520 unlock_new_inode(inode);
6521 d_instantiate(dentry, inode);
6522
6523 out_unlock:
6524 btrfs_end_transaction(trans);
6525 if (err && drop_inode_on_err) {
6526 inode_dec_link_count(inode);
6527 iput(inode);
6528 }
6529 btrfs_balance_delayed_items(fs_info);
6530 btrfs_btree_balance_dirty(fs_info);
6531 return err;
6532
6533 out_unlock_inode:
6534 unlock_new_inode(inode);
6535 goto out_unlock;
6536
6537 }
6538
6539 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6540 struct dentry *dentry)
6541 {
6542 struct btrfs_trans_handle *trans = NULL;
6543 struct btrfs_root *root = BTRFS_I(dir)->root;
6544 struct inode *inode = d_inode(old_dentry);
6545 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6546 u64 index;
6547 int err;
6548 int drop_inode = 0;
6549
6550 /* do not allow sys_link's with other subvols of the same device */
6551 if (root->objectid != BTRFS_I(inode)->root->objectid)
6552 return -EXDEV;
6553
6554 if (inode->i_nlink >= BTRFS_LINK_MAX)
6555 return -EMLINK;
6556
6557 err = btrfs_set_inode_index(dir, &index);
6558 if (err)
6559 goto fail;
6560
6561 /*
6562 * 2 items for inode and inode ref
6563 * 2 items for dir items
6564 * 1 item for parent inode
6565 */
6566 trans = btrfs_start_transaction(root, 5);
6567 if (IS_ERR(trans)) {
6568 err = PTR_ERR(trans);
6569 trans = NULL;
6570 goto fail;
6571 }
6572
6573 /* There are several dir indexes for this inode, clear the cache. */
6574 BTRFS_I(inode)->dir_index = 0ULL;
6575 inc_nlink(inode);
6576 inode_inc_iversion(inode);
6577 inode->i_ctime = current_time(inode);
6578 ihold(inode);
6579 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6580
6581 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6582
6583 if (err) {
6584 drop_inode = 1;
6585 } else {
6586 struct dentry *parent = dentry->d_parent;
6587 err = btrfs_update_inode(trans, root, inode);
6588 if (err)
6589 goto fail;
6590 if (inode->i_nlink == 1) {
6591 /*
6592 * If new hard link count is 1, it's a file created
6593 * with open(2) O_TMPFILE flag.
6594 */
6595 err = btrfs_orphan_del(trans, inode);
6596 if (err)
6597 goto fail;
6598 }
6599 d_instantiate(dentry, inode);
6600 btrfs_log_new_name(trans, inode, NULL, parent);
6601 }
6602
6603 btrfs_balance_delayed_items(fs_info);
6604 fail:
6605 if (trans)
6606 btrfs_end_transaction(trans);
6607 if (drop_inode) {
6608 inode_dec_link_count(inode);
6609 iput(inode);
6610 }
6611 btrfs_btree_balance_dirty(fs_info);
6612 return err;
6613 }
6614
6615 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6616 {
6617 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6618 struct inode *inode = NULL;
6619 struct btrfs_trans_handle *trans;
6620 struct btrfs_root *root = BTRFS_I(dir)->root;
6621 int err = 0;
6622 int drop_on_err = 0;
6623 u64 objectid = 0;
6624 u64 index = 0;
6625
6626 /*
6627 * 2 items for inode and ref
6628 * 2 items for dir items
6629 * 1 for xattr if selinux is on
6630 */
6631 trans = btrfs_start_transaction(root, 5);
6632 if (IS_ERR(trans))
6633 return PTR_ERR(trans);
6634
6635 err = btrfs_find_free_ino(root, &objectid);
6636 if (err)
6637 goto out_fail;
6638
6639 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6640 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6641 S_IFDIR | mode, &index);
6642 if (IS_ERR(inode)) {
6643 err = PTR_ERR(inode);
6644 goto out_fail;
6645 }
6646
6647 drop_on_err = 1;
6648 /* these must be set before we unlock the inode */
6649 inode->i_op = &btrfs_dir_inode_operations;
6650 inode->i_fop = &btrfs_dir_file_operations;
6651
6652 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6653 if (err)
6654 goto out_fail_inode;
6655
6656 btrfs_i_size_write(inode, 0);
6657 err = btrfs_update_inode(trans, root, inode);
6658 if (err)
6659 goto out_fail_inode;
6660
6661 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6662 dentry->d_name.len, 0, index);
6663 if (err)
6664 goto out_fail_inode;
6665
6666 d_instantiate(dentry, inode);
6667 /*
6668 * mkdir is special. We're unlocking after we call d_instantiate
6669 * to avoid a race with nfsd calling d_instantiate.
6670 */
6671 unlock_new_inode(inode);
6672 drop_on_err = 0;
6673
6674 out_fail:
6675 btrfs_end_transaction(trans);
6676 if (drop_on_err) {
6677 inode_dec_link_count(inode);
6678 iput(inode);
6679 }
6680 btrfs_balance_delayed_items(fs_info);
6681 btrfs_btree_balance_dirty(fs_info);
6682 return err;
6683
6684 out_fail_inode:
6685 unlock_new_inode(inode);
6686 goto out_fail;
6687 }
6688
6689 /* Find next extent map of a given extent map, caller needs to ensure locks */
6690 static struct extent_map *next_extent_map(struct extent_map *em)
6691 {
6692 struct rb_node *next;
6693
6694 next = rb_next(&em->rb_node);
6695 if (!next)
6696 return NULL;
6697 return container_of(next, struct extent_map, rb_node);
6698 }
6699
6700 static struct extent_map *prev_extent_map(struct extent_map *em)
6701 {
6702 struct rb_node *prev;
6703
6704 prev = rb_prev(&em->rb_node);
6705 if (!prev)
6706 return NULL;
6707 return container_of(prev, struct extent_map, rb_node);
6708 }
6709
6710 /* helper for btfs_get_extent. Given an existing extent in the tree,
6711 * the existing extent is the nearest extent to map_start,
6712 * and an extent that you want to insert, deal with overlap and insert
6713 * the best fitted new extent into the tree.
6714 */
6715 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6716 struct extent_map *existing,
6717 struct extent_map *em,
6718 u64 map_start)
6719 {
6720 struct extent_map *prev;
6721 struct extent_map *next;
6722 u64 start;
6723 u64 end;
6724 u64 start_diff;
6725
6726 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6727
6728 if (existing->start > map_start) {
6729 next = existing;
6730 prev = prev_extent_map(next);
6731 } else {
6732 prev = existing;
6733 next = next_extent_map(prev);
6734 }
6735
6736 start = prev ? extent_map_end(prev) : em->start;
6737 start = max_t(u64, start, em->start);
6738 end = next ? next->start : extent_map_end(em);
6739 end = min_t(u64, end, extent_map_end(em));
6740 start_diff = start - em->start;
6741 em->start = start;
6742 em->len = end - start;
6743 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6744 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6745 em->block_start += start_diff;
6746 em->block_len -= start_diff;
6747 }
6748 return add_extent_mapping(em_tree, em, 0);
6749 }
6750
6751 static noinline int uncompress_inline(struct btrfs_path *path,
6752 struct page *page,
6753 size_t pg_offset, u64 extent_offset,
6754 struct btrfs_file_extent_item *item)
6755 {
6756 int ret;
6757 struct extent_buffer *leaf = path->nodes[0];
6758 char *tmp;
6759 size_t max_size;
6760 unsigned long inline_size;
6761 unsigned long ptr;
6762 int compress_type;
6763
6764 WARN_ON(pg_offset != 0);
6765 compress_type = btrfs_file_extent_compression(leaf, item);
6766 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6767 inline_size = btrfs_file_extent_inline_item_len(leaf,
6768 btrfs_item_nr(path->slots[0]));
6769 tmp = kmalloc(inline_size, GFP_NOFS);
6770 if (!tmp)
6771 return -ENOMEM;
6772 ptr = btrfs_file_extent_inline_start(item);
6773
6774 read_extent_buffer(leaf, tmp, ptr, inline_size);
6775
6776 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6777 ret = btrfs_decompress(compress_type, tmp, page,
6778 extent_offset, inline_size, max_size);
6779 kfree(tmp);
6780 return ret;
6781 }
6782
6783 /*
6784 * a bit scary, this does extent mapping from logical file offset to the disk.
6785 * the ugly parts come from merging extents from the disk with the in-ram
6786 * representation. This gets more complex because of the data=ordered code,
6787 * where the in-ram extents might be locked pending data=ordered completion.
6788 *
6789 * This also copies inline extents directly into the page.
6790 */
6791
6792 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6793 size_t pg_offset, u64 start, u64 len,
6794 int create)
6795 {
6796 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6797 int ret;
6798 int err = 0;
6799 u64 extent_start = 0;
6800 u64 extent_end = 0;
6801 u64 objectid = btrfs_ino(BTRFS_I(inode));
6802 u32 found_type;
6803 struct btrfs_path *path = NULL;
6804 struct btrfs_root *root = BTRFS_I(inode)->root;
6805 struct btrfs_file_extent_item *item;
6806 struct extent_buffer *leaf;
6807 struct btrfs_key found_key;
6808 struct extent_map *em = NULL;
6809 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6810 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6811 struct btrfs_trans_handle *trans = NULL;
6812 const bool new_inline = !page || create;
6813
6814 again:
6815 read_lock(&em_tree->lock);
6816 em = lookup_extent_mapping(em_tree, start, len);
6817 if (em)
6818 em->bdev = fs_info->fs_devices->latest_bdev;
6819 read_unlock(&em_tree->lock);
6820
6821 if (em) {
6822 if (em->start > start || em->start + em->len <= start)
6823 free_extent_map(em);
6824 else if (em->block_start == EXTENT_MAP_INLINE && page)
6825 free_extent_map(em);
6826 else
6827 goto out;
6828 }
6829 em = alloc_extent_map();
6830 if (!em) {
6831 err = -ENOMEM;
6832 goto out;
6833 }
6834 em->bdev = fs_info->fs_devices->latest_bdev;
6835 em->start = EXTENT_MAP_HOLE;
6836 em->orig_start = EXTENT_MAP_HOLE;
6837 em->len = (u64)-1;
6838 em->block_len = (u64)-1;
6839
6840 if (!path) {
6841 path = btrfs_alloc_path();
6842 if (!path) {
6843 err = -ENOMEM;
6844 goto out;
6845 }
6846 /*
6847 * Chances are we'll be called again, so go ahead and do
6848 * readahead
6849 */
6850 path->reada = READA_FORWARD;
6851 }
6852
6853 ret = btrfs_lookup_file_extent(trans, root, path,
6854 objectid, start, trans != NULL);
6855 if (ret < 0) {
6856 err = ret;
6857 goto out;
6858 }
6859
6860 if (ret != 0) {
6861 if (path->slots[0] == 0)
6862 goto not_found;
6863 path->slots[0]--;
6864 }
6865
6866 leaf = path->nodes[0];
6867 item = btrfs_item_ptr(leaf, path->slots[0],
6868 struct btrfs_file_extent_item);
6869 /* are we inside the extent that was found? */
6870 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6871 found_type = found_key.type;
6872 if (found_key.objectid != objectid ||
6873 found_type != BTRFS_EXTENT_DATA_KEY) {
6874 /*
6875 * If we backup past the first extent we want to move forward
6876 * and see if there is an extent in front of us, otherwise we'll
6877 * say there is a hole for our whole search range which can
6878 * cause problems.
6879 */
6880 extent_end = start;
6881 goto next;
6882 }
6883
6884 found_type = btrfs_file_extent_type(leaf, item);
6885 extent_start = found_key.offset;
6886 if (found_type == BTRFS_FILE_EXTENT_REG ||
6887 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6888 extent_end = extent_start +
6889 btrfs_file_extent_num_bytes(leaf, item);
6890 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6891 size_t size;
6892 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6893 extent_end = ALIGN(extent_start + size,
6894 fs_info->sectorsize);
6895 }
6896 next:
6897 if (start >= extent_end) {
6898 path->slots[0]++;
6899 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6900 ret = btrfs_next_leaf(root, path);
6901 if (ret < 0) {
6902 err = ret;
6903 goto out;
6904 }
6905 if (ret > 0)
6906 goto not_found;
6907 leaf = path->nodes[0];
6908 }
6909 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6910 if (found_key.objectid != objectid ||
6911 found_key.type != BTRFS_EXTENT_DATA_KEY)
6912 goto not_found;
6913 if (start + len <= found_key.offset)
6914 goto not_found;
6915 if (start > found_key.offset)
6916 goto next;
6917 em->start = start;
6918 em->orig_start = start;
6919 em->len = found_key.offset - start;
6920 goto not_found_em;
6921 }
6922
6923 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6924
6925 if (found_type == BTRFS_FILE_EXTENT_REG ||
6926 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6927 goto insert;
6928 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6929 unsigned long ptr;
6930 char *map;
6931 size_t size;
6932 size_t extent_offset;
6933 size_t copy_size;
6934
6935 if (new_inline)
6936 goto out;
6937
6938 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6939 extent_offset = page_offset(page) + pg_offset - extent_start;
6940 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6941 size - extent_offset);
6942 em->start = extent_start + extent_offset;
6943 em->len = ALIGN(copy_size, fs_info->sectorsize);
6944 em->orig_block_len = em->len;
6945 em->orig_start = em->start;
6946 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6947 if (create == 0 && !PageUptodate(page)) {
6948 if (btrfs_file_extent_compression(leaf, item) !=
6949 BTRFS_COMPRESS_NONE) {
6950 ret = uncompress_inline(path, page, pg_offset,
6951 extent_offset, item);
6952 if (ret) {
6953 err = ret;
6954 goto out;
6955 }
6956 } else {
6957 map = kmap(page);
6958 read_extent_buffer(leaf, map + pg_offset, ptr,
6959 copy_size);
6960 if (pg_offset + copy_size < PAGE_SIZE) {
6961 memset(map + pg_offset + copy_size, 0,
6962 PAGE_SIZE - pg_offset -
6963 copy_size);
6964 }
6965 kunmap(page);
6966 }
6967 flush_dcache_page(page);
6968 } else if (create && PageUptodate(page)) {
6969 BUG();
6970 if (!trans) {
6971 kunmap(page);
6972 free_extent_map(em);
6973 em = NULL;
6974
6975 btrfs_release_path(path);
6976 trans = btrfs_join_transaction(root);
6977
6978 if (IS_ERR(trans))
6979 return ERR_CAST(trans);
6980 goto again;
6981 }
6982 map = kmap(page);
6983 write_extent_buffer(leaf, map + pg_offset, ptr,
6984 copy_size);
6985 kunmap(page);
6986 btrfs_mark_buffer_dirty(leaf);
6987 }
6988 set_extent_uptodate(io_tree, em->start,
6989 extent_map_end(em) - 1, NULL, GFP_NOFS);
6990 goto insert;
6991 }
6992 not_found:
6993 em->start = start;
6994 em->orig_start = start;
6995 em->len = len;
6996 not_found_em:
6997 em->block_start = EXTENT_MAP_HOLE;
6998 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6999 insert:
7000 btrfs_release_path(path);
7001 if (em->start > start || extent_map_end(em) <= start) {
7002 btrfs_err(fs_info,
7003 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7004 em->start, em->len, start, len);
7005 err = -EIO;
7006 goto out;
7007 }
7008
7009 err = 0;
7010 write_lock(&em_tree->lock);
7011 ret = add_extent_mapping(em_tree, em, 0);
7012 /* it is possible that someone inserted the extent into the tree
7013 * while we had the lock dropped. It is also possible that
7014 * an overlapping map exists in the tree
7015 */
7016 if (ret == -EEXIST) {
7017 struct extent_map *existing;
7018
7019 ret = 0;
7020
7021 existing = search_extent_mapping(em_tree, start, len);
7022 /*
7023 * existing will always be non-NULL, since there must be
7024 * extent causing the -EEXIST.
7025 */
7026 if (existing->start == em->start &&
7027 extent_map_end(existing) >= extent_map_end(em) &&
7028 em->block_start == existing->block_start) {
7029 /*
7030 * The existing extent map already encompasses the
7031 * entire extent map we tried to add.
7032 */
7033 free_extent_map(em);
7034 em = existing;
7035 err = 0;
7036
7037 } else if (start >= extent_map_end(existing) ||
7038 start <= existing->start) {
7039 /*
7040 * The existing extent map is the one nearest to
7041 * the [start, start + len) range which overlaps
7042 */
7043 err = merge_extent_mapping(em_tree, existing,
7044 em, start);
7045 free_extent_map(existing);
7046 if (err) {
7047 free_extent_map(em);
7048 em = NULL;
7049 }
7050 } else {
7051 free_extent_map(em);
7052 em = existing;
7053 err = 0;
7054 }
7055 }
7056 write_unlock(&em_tree->lock);
7057 out:
7058
7059 trace_btrfs_get_extent(root, BTRFS_I(inode), em);
7060
7061 btrfs_free_path(path);
7062 if (trans) {
7063 ret = btrfs_end_transaction(trans);
7064 if (!err)
7065 err = ret;
7066 }
7067 if (err) {
7068 free_extent_map(em);
7069 return ERR_PTR(err);
7070 }
7071 BUG_ON(!em); /* Error is always set */
7072 return em;
7073 }
7074
7075 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7076 size_t pg_offset, u64 start, u64 len,
7077 int create)
7078 {
7079 struct extent_map *em;
7080 struct extent_map *hole_em = NULL;
7081 u64 range_start = start;
7082 u64 end;
7083 u64 found;
7084 u64 found_end;
7085 int err = 0;
7086
7087 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7088 if (IS_ERR(em))
7089 return em;
7090 if (em) {
7091 /*
7092 * if our em maps to
7093 * - a hole or
7094 * - a pre-alloc extent,
7095 * there might actually be delalloc bytes behind it.
7096 */
7097 if (em->block_start != EXTENT_MAP_HOLE &&
7098 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7099 return em;
7100 else
7101 hole_em = em;
7102 }
7103
7104 /* check to see if we've wrapped (len == -1 or similar) */
7105 end = start + len;
7106 if (end < start)
7107 end = (u64)-1;
7108 else
7109 end -= 1;
7110
7111 em = NULL;
7112
7113 /* ok, we didn't find anything, lets look for delalloc */
7114 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7115 end, len, EXTENT_DELALLOC, 1);
7116 found_end = range_start + found;
7117 if (found_end < range_start)
7118 found_end = (u64)-1;
7119
7120 /*
7121 * we didn't find anything useful, return
7122 * the original results from get_extent()
7123 */
7124 if (range_start > end || found_end <= start) {
7125 em = hole_em;
7126 hole_em = NULL;
7127 goto out;
7128 }
7129
7130 /* adjust the range_start to make sure it doesn't
7131 * go backwards from the start they passed in
7132 */
7133 range_start = max(start, range_start);
7134 found = found_end - range_start;
7135
7136 if (found > 0) {
7137 u64 hole_start = start;
7138 u64 hole_len = len;
7139
7140 em = alloc_extent_map();
7141 if (!em) {
7142 err = -ENOMEM;
7143 goto out;
7144 }
7145 /*
7146 * when btrfs_get_extent can't find anything it
7147 * returns one huge hole
7148 *
7149 * make sure what it found really fits our range, and
7150 * adjust to make sure it is based on the start from
7151 * the caller
7152 */
7153 if (hole_em) {
7154 u64 calc_end = extent_map_end(hole_em);
7155
7156 if (calc_end <= start || (hole_em->start > end)) {
7157 free_extent_map(hole_em);
7158 hole_em = NULL;
7159 } else {
7160 hole_start = max(hole_em->start, start);
7161 hole_len = calc_end - hole_start;
7162 }
7163 }
7164 em->bdev = NULL;
7165 if (hole_em && range_start > hole_start) {
7166 /* our hole starts before our delalloc, so we
7167 * have to return just the parts of the hole
7168 * that go until the delalloc starts
7169 */
7170 em->len = min(hole_len,
7171 range_start - hole_start);
7172 em->start = hole_start;
7173 em->orig_start = hole_start;
7174 /*
7175 * don't adjust block start at all,
7176 * it is fixed at EXTENT_MAP_HOLE
7177 */
7178 em->block_start = hole_em->block_start;
7179 em->block_len = hole_len;
7180 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7181 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7182 } else {
7183 em->start = range_start;
7184 em->len = found;
7185 em->orig_start = range_start;
7186 em->block_start = EXTENT_MAP_DELALLOC;
7187 em->block_len = found;
7188 }
7189 } else if (hole_em) {
7190 return hole_em;
7191 }
7192 out:
7193
7194 free_extent_map(hole_em);
7195 if (err) {
7196 free_extent_map(em);
7197 return ERR_PTR(err);
7198 }
7199 return em;
7200 }
7201
7202 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7203 const u64 start,
7204 const u64 len,
7205 const u64 orig_start,
7206 const u64 block_start,
7207 const u64 block_len,
7208 const u64 orig_block_len,
7209 const u64 ram_bytes,
7210 const int type)
7211 {
7212 struct extent_map *em = NULL;
7213 int ret;
7214
7215 if (type != BTRFS_ORDERED_NOCOW) {
7216 em = create_pinned_em(inode, start, len, orig_start,
7217 block_start, block_len, orig_block_len,
7218 ram_bytes, type);
7219 if (IS_ERR(em))
7220 goto out;
7221 }
7222 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7223 len, block_len, type);
7224 if (ret) {
7225 if (em) {
7226 free_extent_map(em);
7227 btrfs_drop_extent_cache(inode, start,
7228 start + len - 1, 0);
7229 }
7230 em = ERR_PTR(ret);
7231 }
7232 out:
7233
7234 return em;
7235 }
7236
7237 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7238 u64 start, u64 len)
7239 {
7240 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7241 struct btrfs_root *root = BTRFS_I(inode)->root;
7242 struct extent_map *em;
7243 struct btrfs_key ins;
7244 u64 alloc_hint;
7245 int ret;
7246
7247 alloc_hint = get_extent_allocation_hint(inode, start, len);
7248 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7249 0, alloc_hint, &ins, 1, 1);
7250 if (ret)
7251 return ERR_PTR(ret);
7252
7253 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7254 ins.objectid, ins.offset, ins.offset,
7255 ins.offset, 0);
7256 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7257 if (IS_ERR(em))
7258 btrfs_free_reserved_extent(fs_info, ins.objectid,
7259 ins.offset, 1);
7260
7261 return em;
7262 }
7263
7264 /*
7265 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7266 * block must be cow'd
7267 */
7268 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7269 u64 *orig_start, u64 *orig_block_len,
7270 u64 *ram_bytes)
7271 {
7272 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7273 struct btrfs_trans_handle *trans;
7274 struct btrfs_path *path;
7275 int ret;
7276 struct extent_buffer *leaf;
7277 struct btrfs_root *root = BTRFS_I(inode)->root;
7278 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7279 struct btrfs_file_extent_item *fi;
7280 struct btrfs_key key;
7281 u64 disk_bytenr;
7282 u64 backref_offset;
7283 u64 extent_end;
7284 u64 num_bytes;
7285 int slot;
7286 int found_type;
7287 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7288
7289 path = btrfs_alloc_path();
7290 if (!path)
7291 return -ENOMEM;
7292
7293 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(BTRFS_I(inode)),
7294 offset, 0);
7295 if (ret < 0)
7296 goto out;
7297
7298 slot = path->slots[0];
7299 if (ret == 1) {
7300 if (slot == 0) {
7301 /* can't find the item, must cow */
7302 ret = 0;
7303 goto out;
7304 }
7305 slot--;
7306 }
7307 ret = 0;
7308 leaf = path->nodes[0];
7309 btrfs_item_key_to_cpu(leaf, &key, slot);
7310 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7311 key.type != BTRFS_EXTENT_DATA_KEY) {
7312 /* not our file or wrong item type, must cow */
7313 goto out;
7314 }
7315
7316 if (key.offset > offset) {
7317 /* Wrong offset, must cow */
7318 goto out;
7319 }
7320
7321 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7322 found_type = btrfs_file_extent_type(leaf, fi);
7323 if (found_type != BTRFS_FILE_EXTENT_REG &&
7324 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7325 /* not a regular extent, must cow */
7326 goto out;
7327 }
7328
7329 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7330 goto out;
7331
7332 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7333 if (extent_end <= offset)
7334 goto out;
7335
7336 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7337 if (disk_bytenr == 0)
7338 goto out;
7339
7340 if (btrfs_file_extent_compression(leaf, fi) ||
7341 btrfs_file_extent_encryption(leaf, fi) ||
7342 btrfs_file_extent_other_encoding(leaf, fi))
7343 goto out;
7344
7345 backref_offset = btrfs_file_extent_offset(leaf, fi);
7346
7347 if (orig_start) {
7348 *orig_start = key.offset - backref_offset;
7349 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7350 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7351 }
7352
7353 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7354 goto out;
7355
7356 num_bytes = min(offset + *len, extent_end) - offset;
7357 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7358 u64 range_end;
7359
7360 range_end = round_up(offset + num_bytes,
7361 root->fs_info->sectorsize) - 1;
7362 ret = test_range_bit(io_tree, offset, range_end,
7363 EXTENT_DELALLOC, 0, NULL);
7364 if (ret) {
7365 ret = -EAGAIN;
7366 goto out;
7367 }
7368 }
7369
7370 btrfs_release_path(path);
7371
7372 /*
7373 * look for other files referencing this extent, if we
7374 * find any we must cow
7375 */
7376 trans = btrfs_join_transaction(root);
7377 if (IS_ERR(trans)) {
7378 ret = 0;
7379 goto out;
7380 }
7381
7382 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(BTRFS_I(inode)),
7383 key.offset - backref_offset, disk_bytenr);
7384 btrfs_end_transaction(trans);
7385 if (ret) {
7386 ret = 0;
7387 goto out;
7388 }
7389
7390 /*
7391 * adjust disk_bytenr and num_bytes to cover just the bytes
7392 * in this extent we are about to write. If there
7393 * are any csums in that range we have to cow in order
7394 * to keep the csums correct
7395 */
7396 disk_bytenr += backref_offset;
7397 disk_bytenr += offset - key.offset;
7398 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7399 goto out;
7400 /*
7401 * all of the above have passed, it is safe to overwrite this extent
7402 * without cow
7403 */
7404 *len = num_bytes;
7405 ret = 1;
7406 out:
7407 btrfs_free_path(path);
7408 return ret;
7409 }
7410
7411 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7412 {
7413 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7414 int found = false;
7415 void **pagep = NULL;
7416 struct page *page = NULL;
7417 int start_idx;
7418 int end_idx;
7419
7420 start_idx = start >> PAGE_SHIFT;
7421
7422 /*
7423 * end is the last byte in the last page. end == start is legal
7424 */
7425 end_idx = end >> PAGE_SHIFT;
7426
7427 rcu_read_lock();
7428
7429 /* Most of the code in this while loop is lifted from
7430 * find_get_page. It's been modified to begin searching from a
7431 * page and return just the first page found in that range. If the
7432 * found idx is less than or equal to the end idx then we know that
7433 * a page exists. If no pages are found or if those pages are
7434 * outside of the range then we're fine (yay!) */
7435 while (page == NULL &&
7436 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7437 page = radix_tree_deref_slot(pagep);
7438 if (unlikely(!page))
7439 break;
7440
7441 if (radix_tree_exception(page)) {
7442 if (radix_tree_deref_retry(page)) {
7443 page = NULL;
7444 continue;
7445 }
7446 /*
7447 * Otherwise, shmem/tmpfs must be storing a swap entry
7448 * here as an exceptional entry: so return it without
7449 * attempting to raise page count.
7450 */
7451 page = NULL;
7452 break; /* TODO: Is this relevant for this use case? */
7453 }
7454
7455 if (!page_cache_get_speculative(page)) {
7456 page = NULL;
7457 continue;
7458 }
7459
7460 /*
7461 * Has the page moved?
7462 * This is part of the lockless pagecache protocol. See
7463 * include/linux/pagemap.h for details.
7464 */
7465 if (unlikely(page != *pagep)) {
7466 put_page(page);
7467 page = NULL;
7468 }
7469 }
7470
7471 if (page) {
7472 if (page->index <= end_idx)
7473 found = true;
7474 put_page(page);
7475 }
7476
7477 rcu_read_unlock();
7478 return found;
7479 }
7480
7481 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7482 struct extent_state **cached_state, int writing)
7483 {
7484 struct btrfs_ordered_extent *ordered;
7485 int ret = 0;
7486
7487 while (1) {
7488 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7489 cached_state);
7490 /*
7491 * We're concerned with the entire range that we're going to be
7492 * doing DIO to, so we need to make sure there's no ordered
7493 * extents in this range.
7494 */
7495 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7496 lockend - lockstart + 1);
7497
7498 /*
7499 * We need to make sure there are no buffered pages in this
7500 * range either, we could have raced between the invalidate in
7501 * generic_file_direct_write and locking the extent. The
7502 * invalidate needs to happen so that reads after a write do not
7503 * get stale data.
7504 */
7505 if (!ordered &&
7506 (!writing ||
7507 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7508 break;
7509
7510 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7511 cached_state, GFP_NOFS);
7512
7513 if (ordered) {
7514 /*
7515 * If we are doing a DIO read and the ordered extent we
7516 * found is for a buffered write, we can not wait for it
7517 * to complete and retry, because if we do so we can
7518 * deadlock with concurrent buffered writes on page
7519 * locks. This happens only if our DIO read covers more
7520 * than one extent map, if at this point has already
7521 * created an ordered extent for a previous extent map
7522 * and locked its range in the inode's io tree, and a
7523 * concurrent write against that previous extent map's
7524 * range and this range started (we unlock the ranges
7525 * in the io tree only when the bios complete and
7526 * buffered writes always lock pages before attempting
7527 * to lock range in the io tree).
7528 */
7529 if (writing ||
7530 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7531 btrfs_start_ordered_extent(inode, ordered, 1);
7532 else
7533 ret = -ENOTBLK;
7534 btrfs_put_ordered_extent(ordered);
7535 } else {
7536 /*
7537 * We could trigger writeback for this range (and wait
7538 * for it to complete) and then invalidate the pages for
7539 * this range (through invalidate_inode_pages2_range()),
7540 * but that can lead us to a deadlock with a concurrent
7541 * call to readpages() (a buffered read or a defrag call
7542 * triggered a readahead) on a page lock due to an
7543 * ordered dio extent we created before but did not have
7544 * yet a corresponding bio submitted (whence it can not
7545 * complete), which makes readpages() wait for that
7546 * ordered extent to complete while holding a lock on
7547 * that page.
7548 */
7549 ret = -ENOTBLK;
7550 }
7551
7552 if (ret)
7553 break;
7554
7555 cond_resched();
7556 }
7557
7558 return ret;
7559 }
7560
7561 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7562 u64 len, u64 orig_start,
7563 u64 block_start, u64 block_len,
7564 u64 orig_block_len, u64 ram_bytes,
7565 int type)
7566 {
7567 struct extent_map_tree *em_tree;
7568 struct extent_map *em;
7569 struct btrfs_root *root = BTRFS_I(inode)->root;
7570 int ret;
7571
7572 em_tree = &BTRFS_I(inode)->extent_tree;
7573 em = alloc_extent_map();
7574 if (!em)
7575 return ERR_PTR(-ENOMEM);
7576
7577 em->start = start;
7578 em->orig_start = orig_start;
7579 em->mod_start = start;
7580 em->mod_len = len;
7581 em->len = len;
7582 em->block_len = block_len;
7583 em->block_start = block_start;
7584 em->bdev = root->fs_info->fs_devices->latest_bdev;
7585 em->orig_block_len = orig_block_len;
7586 em->ram_bytes = ram_bytes;
7587 em->generation = -1;
7588 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7589 if (type == BTRFS_ORDERED_PREALLOC)
7590 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7591
7592 do {
7593 btrfs_drop_extent_cache(inode, em->start,
7594 em->start + em->len - 1, 0);
7595 write_lock(&em_tree->lock);
7596 ret = add_extent_mapping(em_tree, em, 1);
7597 write_unlock(&em_tree->lock);
7598 } while (ret == -EEXIST);
7599
7600 if (ret) {
7601 free_extent_map(em);
7602 return ERR_PTR(ret);
7603 }
7604
7605 return em;
7606 }
7607
7608 static void adjust_dio_outstanding_extents(struct inode *inode,
7609 struct btrfs_dio_data *dio_data,
7610 const u64 len)
7611 {
7612 unsigned num_extents = count_max_extents(len);
7613
7614 /*
7615 * If we have an outstanding_extents count still set then we're
7616 * within our reservation, otherwise we need to adjust our inode
7617 * counter appropriately.
7618 */
7619 if (dio_data->outstanding_extents >= num_extents) {
7620 dio_data->outstanding_extents -= num_extents;
7621 } else {
7622 /*
7623 * If dio write length has been split due to no large enough
7624 * contiguous space, we need to compensate our inode counter
7625 * appropriately.
7626 */
7627 u64 num_needed = num_extents - dio_data->outstanding_extents;
7628
7629 spin_lock(&BTRFS_I(inode)->lock);
7630 BTRFS_I(inode)->outstanding_extents += num_needed;
7631 spin_unlock(&BTRFS_I(inode)->lock);
7632 }
7633 }
7634
7635 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7636 struct buffer_head *bh_result, int create)
7637 {
7638 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7639 struct extent_map *em;
7640 struct extent_state *cached_state = NULL;
7641 struct btrfs_dio_data *dio_data = NULL;
7642 u64 start = iblock << inode->i_blkbits;
7643 u64 lockstart, lockend;
7644 u64 len = bh_result->b_size;
7645 int unlock_bits = EXTENT_LOCKED;
7646 int ret = 0;
7647
7648 if (create)
7649 unlock_bits |= EXTENT_DIRTY;
7650 else
7651 len = min_t(u64, len, fs_info->sectorsize);
7652
7653 lockstart = start;
7654 lockend = start + len - 1;
7655
7656 if (current->journal_info) {
7657 /*
7658 * Need to pull our outstanding extents and set journal_info to NULL so
7659 * that anything that needs to check if there's a transaction doesn't get
7660 * confused.
7661 */
7662 dio_data = current->journal_info;
7663 current->journal_info = NULL;
7664 }
7665
7666 /*
7667 * If this errors out it's because we couldn't invalidate pagecache for
7668 * this range and we need to fallback to buffered.
7669 */
7670 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7671 create)) {
7672 ret = -ENOTBLK;
7673 goto err;
7674 }
7675
7676 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7677 if (IS_ERR(em)) {
7678 ret = PTR_ERR(em);
7679 goto unlock_err;
7680 }
7681
7682 /*
7683 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7684 * io. INLINE is special, and we could probably kludge it in here, but
7685 * it's still buffered so for safety lets just fall back to the generic
7686 * buffered path.
7687 *
7688 * For COMPRESSED we _have_ to read the entire extent in so we can
7689 * decompress it, so there will be buffering required no matter what we
7690 * do, so go ahead and fallback to buffered.
7691 *
7692 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7693 * to buffered IO. Don't blame me, this is the price we pay for using
7694 * the generic code.
7695 */
7696 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7697 em->block_start == EXTENT_MAP_INLINE) {
7698 free_extent_map(em);
7699 ret = -ENOTBLK;
7700 goto unlock_err;
7701 }
7702
7703 /* Just a good old fashioned hole, return */
7704 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7705 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7706 free_extent_map(em);
7707 goto unlock_err;
7708 }
7709
7710 /*
7711 * We don't allocate a new extent in the following cases
7712 *
7713 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7714 * existing extent.
7715 * 2) The extent is marked as PREALLOC. We're good to go here and can
7716 * just use the extent.
7717 *
7718 */
7719 if (!create) {
7720 len = min(len, em->len - (start - em->start));
7721 lockstart = start + len;
7722 goto unlock;
7723 }
7724
7725 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7726 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7727 em->block_start != EXTENT_MAP_HOLE)) {
7728 int type;
7729 u64 block_start, orig_start, orig_block_len, ram_bytes;
7730
7731 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7732 type = BTRFS_ORDERED_PREALLOC;
7733 else
7734 type = BTRFS_ORDERED_NOCOW;
7735 len = min(len, em->len - (start - em->start));
7736 block_start = em->block_start + (start - em->start);
7737
7738 if (can_nocow_extent(inode, start, &len, &orig_start,
7739 &orig_block_len, &ram_bytes) == 1 &&
7740 btrfs_inc_nocow_writers(fs_info, block_start)) {
7741 struct extent_map *em2;
7742
7743 em2 = btrfs_create_dio_extent(inode, start, len,
7744 orig_start, block_start,
7745 len, orig_block_len,
7746 ram_bytes, type);
7747 btrfs_dec_nocow_writers(fs_info, block_start);
7748 if (type == BTRFS_ORDERED_PREALLOC) {
7749 free_extent_map(em);
7750 em = em2;
7751 }
7752 if (em2 && IS_ERR(em2)) {
7753 ret = PTR_ERR(em2);
7754 goto unlock_err;
7755 }
7756 /*
7757 * For inode marked NODATACOW or extent marked PREALLOC,
7758 * use the existing or preallocated extent, so does not
7759 * need to adjust btrfs_space_info's bytes_may_use.
7760 */
7761 btrfs_free_reserved_data_space_noquota(inode,
7762 start, len);
7763 goto unlock;
7764 }
7765 }
7766
7767 /*
7768 * this will cow the extent, reset the len in case we changed
7769 * it above
7770 */
7771 len = bh_result->b_size;
7772 free_extent_map(em);
7773 em = btrfs_new_extent_direct(inode, start, len);
7774 if (IS_ERR(em)) {
7775 ret = PTR_ERR(em);
7776 goto unlock_err;
7777 }
7778 len = min(len, em->len - (start - em->start));
7779 unlock:
7780 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7781 inode->i_blkbits;
7782 bh_result->b_size = len;
7783 bh_result->b_bdev = em->bdev;
7784 set_buffer_mapped(bh_result);
7785 if (create) {
7786 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7787 set_buffer_new(bh_result);
7788
7789 /*
7790 * Need to update the i_size under the extent lock so buffered
7791 * readers will get the updated i_size when we unlock.
7792 */
7793 if (start + len > i_size_read(inode))
7794 i_size_write(inode, start + len);
7795
7796 adjust_dio_outstanding_extents(inode, dio_data, len);
7797 WARN_ON(dio_data->reserve < len);
7798 dio_data->reserve -= len;
7799 dio_data->unsubmitted_oe_range_end = start + len;
7800 current->journal_info = dio_data;
7801 }
7802
7803 /*
7804 * In the case of write we need to clear and unlock the entire range,
7805 * in the case of read we need to unlock only the end area that we
7806 * aren't using if there is any left over space.
7807 */
7808 if (lockstart < lockend) {
7809 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7810 lockend, unlock_bits, 1, 0,
7811 &cached_state, GFP_NOFS);
7812 } else {
7813 free_extent_state(cached_state);
7814 }
7815
7816 free_extent_map(em);
7817
7818 return 0;
7819
7820 unlock_err:
7821 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7822 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7823 err:
7824 if (dio_data)
7825 current->journal_info = dio_data;
7826 /*
7827 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7828 * write less data then expected, so that we don't underflow our inode's
7829 * outstanding extents counter.
7830 */
7831 if (create && dio_data)
7832 adjust_dio_outstanding_extents(inode, dio_data, len);
7833
7834 return ret;
7835 }
7836
7837 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7838 int mirror_num)
7839 {
7840 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7841 int ret;
7842
7843 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7844
7845 bio_get(bio);
7846
7847 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7848 if (ret)
7849 goto err;
7850
7851 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7852 err:
7853 bio_put(bio);
7854 return ret;
7855 }
7856
7857 static int btrfs_check_dio_repairable(struct inode *inode,
7858 struct bio *failed_bio,
7859 struct io_failure_record *failrec,
7860 int failed_mirror)
7861 {
7862 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7863 int num_copies;
7864
7865 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7866 if (num_copies == 1) {
7867 /*
7868 * we only have a single copy of the data, so don't bother with
7869 * all the retry and error correction code that follows. no
7870 * matter what the error is, it is very likely to persist.
7871 */
7872 btrfs_debug(fs_info,
7873 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7874 num_copies, failrec->this_mirror, failed_mirror);
7875 return 0;
7876 }
7877
7878 failrec->failed_mirror = failed_mirror;
7879 failrec->this_mirror++;
7880 if (failrec->this_mirror == failed_mirror)
7881 failrec->this_mirror++;
7882
7883 if (failrec->this_mirror > num_copies) {
7884 btrfs_debug(fs_info,
7885 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7886 num_copies, failrec->this_mirror, failed_mirror);
7887 return 0;
7888 }
7889
7890 return 1;
7891 }
7892
7893 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7894 struct page *page, unsigned int pgoff,
7895 u64 start, u64 end, int failed_mirror,
7896 bio_end_io_t *repair_endio, void *repair_arg)
7897 {
7898 struct io_failure_record *failrec;
7899 struct bio *bio;
7900 int isector;
7901 int read_mode = 0;
7902 int ret;
7903
7904 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7905
7906 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7907 if (ret)
7908 return ret;
7909
7910 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7911 failed_mirror);
7912 if (!ret) {
7913 free_io_failure(inode, failrec);
7914 return -EIO;
7915 }
7916
7917 if ((failed_bio->bi_vcnt > 1)
7918 || (failed_bio->bi_io_vec->bv_len
7919 > btrfs_inode_sectorsize(inode)))
7920 read_mode |= REQ_FAILFAST_DEV;
7921
7922 isector = start - btrfs_io_bio(failed_bio)->logical;
7923 isector >>= inode->i_sb->s_blocksize_bits;
7924 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7925 pgoff, isector, repair_endio, repair_arg);
7926 if (!bio) {
7927 free_io_failure(inode, failrec);
7928 return -EIO;
7929 }
7930 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
7931
7932 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7933 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7934 read_mode, failrec->this_mirror, failrec->in_validation);
7935
7936 ret = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7937 if (ret) {
7938 free_io_failure(inode, failrec);
7939 bio_put(bio);
7940 }
7941
7942 return ret;
7943 }
7944
7945 struct btrfs_retry_complete {
7946 struct completion done;
7947 struct inode *inode;
7948 u64 start;
7949 int uptodate;
7950 };
7951
7952 static void btrfs_retry_endio_nocsum(struct bio *bio)
7953 {
7954 struct btrfs_retry_complete *done = bio->bi_private;
7955 struct inode *inode;
7956 struct bio_vec *bvec;
7957 int i;
7958
7959 if (bio->bi_error)
7960 goto end;
7961
7962 ASSERT(bio->bi_vcnt == 1);
7963 inode = bio->bi_io_vec->bv_page->mapping->host;
7964 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
7965
7966 done->uptodate = 1;
7967 bio_for_each_segment_all(bvec, bio, i)
7968 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7969 end:
7970 complete(&done->done);
7971 bio_put(bio);
7972 }
7973
7974 static int __btrfs_correct_data_nocsum(struct inode *inode,
7975 struct btrfs_io_bio *io_bio)
7976 {
7977 struct btrfs_fs_info *fs_info;
7978 struct bio_vec *bvec;
7979 struct btrfs_retry_complete done;
7980 u64 start;
7981 unsigned int pgoff;
7982 u32 sectorsize;
7983 int nr_sectors;
7984 int i;
7985 int ret;
7986
7987 fs_info = BTRFS_I(inode)->root->fs_info;
7988 sectorsize = fs_info->sectorsize;
7989
7990 start = io_bio->logical;
7991 done.inode = inode;
7992
7993 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7994 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7995 pgoff = bvec->bv_offset;
7996
7997 next_block_or_try_again:
7998 done.uptodate = 0;
7999 done.start = start;
8000 init_completion(&done.done);
8001
8002 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8003 pgoff, start, start + sectorsize - 1,
8004 io_bio->mirror_num,
8005 btrfs_retry_endio_nocsum, &done);
8006 if (ret)
8007 return ret;
8008
8009 wait_for_completion(&done.done);
8010
8011 if (!done.uptodate) {
8012 /* We might have another mirror, so try again */
8013 goto next_block_or_try_again;
8014 }
8015
8016 start += sectorsize;
8017
8018 if (nr_sectors--) {
8019 pgoff += sectorsize;
8020 goto next_block_or_try_again;
8021 }
8022 }
8023
8024 return 0;
8025 }
8026
8027 static void btrfs_retry_endio(struct bio *bio)
8028 {
8029 struct btrfs_retry_complete *done = bio->bi_private;
8030 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8031 struct inode *inode;
8032 struct bio_vec *bvec;
8033 u64 start;
8034 int uptodate;
8035 int ret;
8036 int i;
8037
8038 if (bio->bi_error)
8039 goto end;
8040
8041 uptodate = 1;
8042
8043 start = done->start;
8044
8045 ASSERT(bio->bi_vcnt == 1);
8046 inode = bio->bi_io_vec->bv_page->mapping->host;
8047 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
8048
8049 bio_for_each_segment_all(bvec, bio, i) {
8050 ret = __readpage_endio_check(done->inode, io_bio, i,
8051 bvec->bv_page, bvec->bv_offset,
8052 done->start, bvec->bv_len);
8053 if (!ret)
8054 clean_io_failure(done->inode, done->start,
8055 bvec->bv_page, bvec->bv_offset);
8056 else
8057 uptodate = 0;
8058 }
8059
8060 done->uptodate = uptodate;
8061 end:
8062 complete(&done->done);
8063 bio_put(bio);
8064 }
8065
8066 static int __btrfs_subio_endio_read(struct inode *inode,
8067 struct btrfs_io_bio *io_bio, int err)
8068 {
8069 struct btrfs_fs_info *fs_info;
8070 struct bio_vec *bvec;
8071 struct btrfs_retry_complete done;
8072 u64 start;
8073 u64 offset = 0;
8074 u32 sectorsize;
8075 int nr_sectors;
8076 unsigned int pgoff;
8077 int csum_pos;
8078 int i;
8079 int ret;
8080
8081 fs_info = BTRFS_I(inode)->root->fs_info;
8082 sectorsize = fs_info->sectorsize;
8083
8084 err = 0;
8085 start = io_bio->logical;
8086 done.inode = inode;
8087
8088 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
8089 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8090
8091 pgoff = bvec->bv_offset;
8092 next_block:
8093 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8094 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8095 bvec->bv_page, pgoff, start,
8096 sectorsize);
8097 if (likely(!ret))
8098 goto next;
8099 try_again:
8100 done.uptodate = 0;
8101 done.start = start;
8102 init_completion(&done.done);
8103
8104 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8105 pgoff, start, start + sectorsize - 1,
8106 io_bio->mirror_num,
8107 btrfs_retry_endio, &done);
8108 if (ret) {
8109 err = ret;
8110 goto next;
8111 }
8112
8113 wait_for_completion(&done.done);
8114
8115 if (!done.uptodate) {
8116 /* We might have another mirror, so try again */
8117 goto try_again;
8118 }
8119 next:
8120 offset += sectorsize;
8121 start += sectorsize;
8122
8123 ASSERT(nr_sectors);
8124
8125 if (--nr_sectors) {
8126 pgoff += sectorsize;
8127 goto next_block;
8128 }
8129 }
8130
8131 return err;
8132 }
8133
8134 static int btrfs_subio_endio_read(struct inode *inode,
8135 struct btrfs_io_bio *io_bio, int err)
8136 {
8137 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8138
8139 if (skip_csum) {
8140 if (unlikely(err))
8141 return __btrfs_correct_data_nocsum(inode, io_bio);
8142 else
8143 return 0;
8144 } else {
8145 return __btrfs_subio_endio_read(inode, io_bio, err);
8146 }
8147 }
8148
8149 static void btrfs_endio_direct_read(struct bio *bio)
8150 {
8151 struct btrfs_dio_private *dip = bio->bi_private;
8152 struct inode *inode = dip->inode;
8153 struct bio *dio_bio;
8154 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8155 int err = bio->bi_error;
8156
8157 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8158 err = btrfs_subio_endio_read(inode, io_bio, err);
8159
8160 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8161 dip->logical_offset + dip->bytes - 1);
8162 dio_bio = dip->dio_bio;
8163
8164 kfree(dip);
8165
8166 dio_bio->bi_error = bio->bi_error;
8167 dio_end_io(dio_bio, bio->bi_error);
8168
8169 if (io_bio->end_io)
8170 io_bio->end_io(io_bio, err);
8171 bio_put(bio);
8172 }
8173
8174 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8175 const u64 offset,
8176 const u64 bytes,
8177 const int uptodate)
8178 {
8179 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8180 struct btrfs_ordered_extent *ordered = NULL;
8181 u64 ordered_offset = offset;
8182 u64 ordered_bytes = bytes;
8183 int ret;
8184
8185 again:
8186 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8187 &ordered_offset,
8188 ordered_bytes,
8189 uptodate);
8190 if (!ret)
8191 goto out_test;
8192
8193 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8194 finish_ordered_fn, NULL, NULL);
8195 btrfs_queue_work(fs_info->endio_write_workers, &ordered->work);
8196 out_test:
8197 /*
8198 * our bio might span multiple ordered extents. If we haven't
8199 * completed the accounting for the whole dio, go back and try again
8200 */
8201 if (ordered_offset < offset + bytes) {
8202 ordered_bytes = offset + bytes - ordered_offset;
8203 ordered = NULL;
8204 goto again;
8205 }
8206 }
8207
8208 static void btrfs_endio_direct_write(struct bio *bio)
8209 {
8210 struct btrfs_dio_private *dip = bio->bi_private;
8211 struct bio *dio_bio = dip->dio_bio;
8212
8213 btrfs_endio_direct_write_update_ordered(dip->inode,
8214 dip->logical_offset,
8215 dip->bytes,
8216 !bio->bi_error);
8217
8218 kfree(dip);
8219
8220 dio_bio->bi_error = bio->bi_error;
8221 dio_end_io(dio_bio, bio->bi_error);
8222 bio_put(bio);
8223 }
8224
8225 static int __btrfs_submit_bio_start_direct_io(struct inode *inode,
8226 struct bio *bio, int mirror_num,
8227 unsigned long bio_flags, u64 offset)
8228 {
8229 int ret;
8230 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8231 BUG_ON(ret); /* -ENOMEM */
8232 return 0;
8233 }
8234
8235 static void btrfs_end_dio_bio(struct bio *bio)
8236 {
8237 struct btrfs_dio_private *dip = bio->bi_private;
8238 int err = bio->bi_error;
8239
8240 if (err)
8241 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8242 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8243 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio), bio->bi_opf,
8244 (unsigned long long)bio->bi_iter.bi_sector,
8245 bio->bi_iter.bi_size, err);
8246
8247 if (dip->subio_endio)
8248 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8249
8250 if (err) {
8251 dip->errors = 1;
8252
8253 /*
8254 * before atomic variable goto zero, we must make sure
8255 * dip->errors is perceived to be set.
8256 */
8257 smp_mb__before_atomic();
8258 }
8259
8260 /* if there are more bios still pending for this dio, just exit */
8261 if (!atomic_dec_and_test(&dip->pending_bios))
8262 goto out;
8263
8264 if (dip->errors) {
8265 bio_io_error(dip->orig_bio);
8266 } else {
8267 dip->dio_bio->bi_error = 0;
8268 bio_endio(dip->orig_bio);
8269 }
8270 out:
8271 bio_put(bio);
8272 }
8273
8274 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8275 u64 first_sector, gfp_t gfp_flags)
8276 {
8277 struct bio *bio;
8278 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8279 if (bio)
8280 bio_associate_current(bio);
8281 return bio;
8282 }
8283
8284 static inline int btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8285 struct btrfs_dio_private *dip,
8286 struct bio *bio,
8287 u64 file_offset)
8288 {
8289 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8290 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8291 int ret;
8292
8293 /*
8294 * We load all the csum data we need when we submit
8295 * the first bio to reduce the csum tree search and
8296 * contention.
8297 */
8298 if (dip->logical_offset == file_offset) {
8299 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8300 file_offset);
8301 if (ret)
8302 return ret;
8303 }
8304
8305 if (bio == dip->orig_bio)
8306 return 0;
8307
8308 file_offset -= dip->logical_offset;
8309 file_offset >>= inode->i_sb->s_blocksize_bits;
8310 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8311
8312 return 0;
8313 }
8314
8315 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8316 u64 file_offset, int skip_sum,
8317 int async_submit)
8318 {
8319 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8320 struct btrfs_dio_private *dip = bio->bi_private;
8321 bool write = bio_op(bio) == REQ_OP_WRITE;
8322 int ret;
8323
8324 if (async_submit)
8325 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8326
8327 bio_get(bio);
8328
8329 if (!write) {
8330 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8331 if (ret)
8332 goto err;
8333 }
8334
8335 if (skip_sum)
8336 goto map;
8337
8338 if (write && async_submit) {
8339 ret = btrfs_wq_submit_bio(fs_info, inode, bio, 0, 0,
8340 file_offset,
8341 __btrfs_submit_bio_start_direct_io,
8342 __btrfs_submit_bio_done);
8343 goto err;
8344 } else if (write) {
8345 /*
8346 * If we aren't doing async submit, calculate the csum of the
8347 * bio now.
8348 */
8349 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8350 if (ret)
8351 goto err;
8352 } else {
8353 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8354 file_offset);
8355 if (ret)
8356 goto err;
8357 }
8358 map:
8359 ret = btrfs_map_bio(fs_info, bio, 0, async_submit);
8360 err:
8361 bio_put(bio);
8362 return ret;
8363 }
8364
8365 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip,
8366 int skip_sum)
8367 {
8368 struct inode *inode = dip->inode;
8369 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8370 struct btrfs_root *root = BTRFS_I(inode)->root;
8371 struct bio *bio;
8372 struct bio *orig_bio = dip->orig_bio;
8373 struct bio_vec *bvec;
8374 u64 start_sector = orig_bio->bi_iter.bi_sector;
8375 u64 file_offset = dip->logical_offset;
8376 u64 submit_len = 0;
8377 u64 map_length;
8378 u32 blocksize = fs_info->sectorsize;
8379 int async_submit = 0;
8380 int nr_sectors;
8381 int ret;
8382 int i, j;
8383
8384 map_length = orig_bio->bi_iter.bi_size;
8385 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8386 &map_length, NULL, 0);
8387 if (ret)
8388 return -EIO;
8389
8390 if (map_length >= orig_bio->bi_iter.bi_size) {
8391 bio = orig_bio;
8392 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8393 goto submit;
8394 }
8395
8396 /* async crcs make it difficult to collect full stripe writes. */
8397 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8398 async_submit = 0;
8399 else
8400 async_submit = 1;
8401
8402 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8403 if (!bio)
8404 return -ENOMEM;
8405
8406 bio->bi_opf = orig_bio->bi_opf;
8407 bio->bi_private = dip;
8408 bio->bi_end_io = btrfs_end_dio_bio;
8409 btrfs_io_bio(bio)->logical = file_offset;
8410 atomic_inc(&dip->pending_bios);
8411
8412 bio_for_each_segment_all(bvec, orig_bio, j) {
8413 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8414 i = 0;
8415 next_block:
8416 if (unlikely(map_length < submit_len + blocksize ||
8417 bio_add_page(bio, bvec->bv_page, blocksize,
8418 bvec->bv_offset + (i * blocksize)) < blocksize)) {
8419 /*
8420 * inc the count before we submit the bio so
8421 * we know the end IO handler won't happen before
8422 * we inc the count. Otherwise, the dip might get freed
8423 * before we're done setting it up
8424 */
8425 atomic_inc(&dip->pending_bios);
8426 ret = __btrfs_submit_dio_bio(bio, inode,
8427 file_offset, skip_sum,
8428 async_submit);
8429 if (ret) {
8430 bio_put(bio);
8431 atomic_dec(&dip->pending_bios);
8432 goto out_err;
8433 }
8434
8435 start_sector += submit_len >> 9;
8436 file_offset += submit_len;
8437
8438 submit_len = 0;
8439
8440 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8441 start_sector, GFP_NOFS);
8442 if (!bio)
8443 goto out_err;
8444 bio->bi_opf = orig_bio->bi_opf;
8445 bio->bi_private = dip;
8446 bio->bi_end_io = btrfs_end_dio_bio;
8447 btrfs_io_bio(bio)->logical = file_offset;
8448
8449 map_length = orig_bio->bi_iter.bi_size;
8450 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8451 start_sector << 9,
8452 &map_length, NULL, 0);
8453 if (ret) {
8454 bio_put(bio);
8455 goto out_err;
8456 }
8457
8458 goto next_block;
8459 } else {
8460 submit_len += blocksize;
8461 if (--nr_sectors) {
8462 i++;
8463 goto next_block;
8464 }
8465 }
8466 }
8467
8468 submit:
8469 ret = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8470 async_submit);
8471 if (!ret)
8472 return 0;
8473
8474 bio_put(bio);
8475 out_err:
8476 dip->errors = 1;
8477 /*
8478 * before atomic variable goto zero, we must
8479 * make sure dip->errors is perceived to be set.
8480 */
8481 smp_mb__before_atomic();
8482 if (atomic_dec_and_test(&dip->pending_bios))
8483 bio_io_error(dip->orig_bio);
8484
8485 /* bio_end_io() will handle error, so we needn't return it */
8486 return 0;
8487 }
8488
8489 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8490 loff_t file_offset)
8491 {
8492 struct btrfs_dio_private *dip = NULL;
8493 struct bio *io_bio = NULL;
8494 struct btrfs_io_bio *btrfs_bio;
8495 int skip_sum;
8496 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8497 int ret = 0;
8498
8499 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8500
8501 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8502 if (!io_bio) {
8503 ret = -ENOMEM;
8504 goto free_ordered;
8505 }
8506
8507 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8508 if (!dip) {
8509 ret = -ENOMEM;
8510 goto free_ordered;
8511 }
8512
8513 dip->private = dio_bio->bi_private;
8514 dip->inode = inode;
8515 dip->logical_offset = file_offset;
8516 dip->bytes = dio_bio->bi_iter.bi_size;
8517 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8518 io_bio->bi_private = dip;
8519 dip->orig_bio = io_bio;
8520 dip->dio_bio = dio_bio;
8521 atomic_set(&dip->pending_bios, 0);
8522 btrfs_bio = btrfs_io_bio(io_bio);
8523 btrfs_bio->logical = file_offset;
8524
8525 if (write) {
8526 io_bio->bi_end_io = btrfs_endio_direct_write;
8527 } else {
8528 io_bio->bi_end_io = btrfs_endio_direct_read;
8529 dip->subio_endio = btrfs_subio_endio_read;
8530 }
8531
8532 /*
8533 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8534 * even if we fail to submit a bio, because in such case we do the
8535 * corresponding error handling below and it must not be done a second
8536 * time by btrfs_direct_IO().
8537 */
8538 if (write) {
8539 struct btrfs_dio_data *dio_data = current->journal_info;
8540
8541 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8542 dip->bytes;
8543 dio_data->unsubmitted_oe_range_start =
8544 dio_data->unsubmitted_oe_range_end;
8545 }
8546
8547 ret = btrfs_submit_direct_hook(dip, skip_sum);
8548 if (!ret)
8549 return;
8550
8551 if (btrfs_bio->end_io)
8552 btrfs_bio->end_io(btrfs_bio, ret);
8553
8554 free_ordered:
8555 /*
8556 * If we arrived here it means either we failed to submit the dip
8557 * or we either failed to clone the dio_bio or failed to allocate the
8558 * dip. If we cloned the dio_bio and allocated the dip, we can just
8559 * call bio_endio against our io_bio so that we get proper resource
8560 * cleanup if we fail to submit the dip, otherwise, we must do the
8561 * same as btrfs_endio_direct_[write|read] because we can't call these
8562 * callbacks - they require an allocated dip and a clone of dio_bio.
8563 */
8564 if (io_bio && dip) {
8565 io_bio->bi_error = -EIO;
8566 bio_endio(io_bio);
8567 /*
8568 * The end io callbacks free our dip, do the final put on io_bio
8569 * and all the cleanup and final put for dio_bio (through
8570 * dio_end_io()).
8571 */
8572 dip = NULL;
8573 io_bio = NULL;
8574 } else {
8575 if (write)
8576 btrfs_endio_direct_write_update_ordered(inode,
8577 file_offset,
8578 dio_bio->bi_iter.bi_size,
8579 0);
8580 else
8581 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8582 file_offset + dio_bio->bi_iter.bi_size - 1);
8583
8584 dio_bio->bi_error = -EIO;
8585 /*
8586 * Releases and cleans up our dio_bio, no need to bio_put()
8587 * nor bio_endio()/bio_io_error() against dio_bio.
8588 */
8589 dio_end_io(dio_bio, ret);
8590 }
8591 if (io_bio)
8592 bio_put(io_bio);
8593 kfree(dip);
8594 }
8595
8596 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8597 struct kiocb *iocb,
8598 const struct iov_iter *iter, loff_t offset)
8599 {
8600 int seg;
8601 int i;
8602 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8603 ssize_t retval = -EINVAL;
8604
8605 if (offset & blocksize_mask)
8606 goto out;
8607
8608 if (iov_iter_alignment(iter) & blocksize_mask)
8609 goto out;
8610
8611 /* If this is a write we don't need to check anymore */
8612 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8613 return 0;
8614 /*
8615 * Check to make sure we don't have duplicate iov_base's in this
8616 * iovec, if so return EINVAL, otherwise we'll get csum errors
8617 * when reading back.
8618 */
8619 for (seg = 0; seg < iter->nr_segs; seg++) {
8620 for (i = seg + 1; i < iter->nr_segs; i++) {
8621 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8622 goto out;
8623 }
8624 }
8625 retval = 0;
8626 out:
8627 return retval;
8628 }
8629
8630 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8631 {
8632 struct file *file = iocb->ki_filp;
8633 struct inode *inode = file->f_mapping->host;
8634 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8635 struct btrfs_dio_data dio_data = { 0 };
8636 loff_t offset = iocb->ki_pos;
8637 size_t count = 0;
8638 int flags = 0;
8639 bool wakeup = true;
8640 bool relock = false;
8641 ssize_t ret;
8642
8643 if (check_direct_IO(fs_info, iocb, iter, offset))
8644 return 0;
8645
8646 inode_dio_begin(inode);
8647 smp_mb__after_atomic();
8648
8649 /*
8650 * The generic stuff only does filemap_write_and_wait_range, which
8651 * isn't enough if we've written compressed pages to this area, so
8652 * we need to flush the dirty pages again to make absolutely sure
8653 * that any outstanding dirty pages are on disk.
8654 */
8655 count = iov_iter_count(iter);
8656 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8657 &BTRFS_I(inode)->runtime_flags))
8658 filemap_fdatawrite_range(inode->i_mapping, offset,
8659 offset + count - 1);
8660
8661 if (iov_iter_rw(iter) == WRITE) {
8662 /*
8663 * If the write DIO is beyond the EOF, we need update
8664 * the isize, but it is protected by i_mutex. So we can
8665 * not unlock the i_mutex at this case.
8666 */
8667 if (offset + count <= inode->i_size) {
8668 inode_unlock(inode);
8669 relock = true;
8670 }
8671 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8672 if (ret)
8673 goto out;
8674 dio_data.outstanding_extents = count_max_extents(count);
8675
8676 /*
8677 * We need to know how many extents we reserved so that we can
8678 * do the accounting properly if we go over the number we
8679 * originally calculated. Abuse current->journal_info for this.
8680 */
8681 dio_data.reserve = round_up(count,
8682 fs_info->sectorsize);
8683 dio_data.unsubmitted_oe_range_start = (u64)offset;
8684 dio_data.unsubmitted_oe_range_end = (u64)offset;
8685 current->journal_info = &dio_data;
8686 down_read(&BTRFS_I(inode)->dio_sem);
8687 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8688 &BTRFS_I(inode)->runtime_flags)) {
8689 inode_dio_end(inode);
8690 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8691 wakeup = false;
8692 }
8693
8694 ret = __blockdev_direct_IO(iocb, inode,
8695 fs_info->fs_devices->latest_bdev,
8696 iter, btrfs_get_blocks_direct, NULL,
8697 btrfs_submit_direct, flags);
8698 if (iov_iter_rw(iter) == WRITE) {
8699 up_read(&BTRFS_I(inode)->dio_sem);
8700 current->journal_info = NULL;
8701 if (ret < 0 && ret != -EIOCBQUEUED) {
8702 if (dio_data.reserve)
8703 btrfs_delalloc_release_space(inode, offset,
8704 dio_data.reserve);
8705 /*
8706 * On error we might have left some ordered extents
8707 * without submitting corresponding bios for them, so
8708 * cleanup them up to avoid other tasks getting them
8709 * and waiting for them to complete forever.
8710 */
8711 if (dio_data.unsubmitted_oe_range_start <
8712 dio_data.unsubmitted_oe_range_end)
8713 btrfs_endio_direct_write_update_ordered(inode,
8714 dio_data.unsubmitted_oe_range_start,
8715 dio_data.unsubmitted_oe_range_end -
8716 dio_data.unsubmitted_oe_range_start,
8717 0);
8718 } else if (ret >= 0 && (size_t)ret < count)
8719 btrfs_delalloc_release_space(inode, offset,
8720 count - (size_t)ret);
8721 }
8722 out:
8723 if (wakeup)
8724 inode_dio_end(inode);
8725 if (relock)
8726 inode_lock(inode);
8727
8728 return ret;
8729 }
8730
8731 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8732
8733 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8734 __u64 start, __u64 len)
8735 {
8736 int ret;
8737
8738 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8739 if (ret)
8740 return ret;
8741
8742 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8743 }
8744
8745 int btrfs_readpage(struct file *file, struct page *page)
8746 {
8747 struct extent_io_tree *tree;
8748 tree = &BTRFS_I(page->mapping->host)->io_tree;
8749 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8750 }
8751
8752 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8753 {
8754 struct extent_io_tree *tree;
8755 struct inode *inode = page->mapping->host;
8756 int ret;
8757
8758 if (current->flags & PF_MEMALLOC) {
8759 redirty_page_for_writepage(wbc, page);
8760 unlock_page(page);
8761 return 0;
8762 }
8763
8764 /*
8765 * If we are under memory pressure we will call this directly from the
8766 * VM, we need to make sure we have the inode referenced for the ordered
8767 * extent. If not just return like we didn't do anything.
8768 */
8769 if (!igrab(inode)) {
8770 redirty_page_for_writepage(wbc, page);
8771 return AOP_WRITEPAGE_ACTIVATE;
8772 }
8773 tree = &BTRFS_I(page->mapping->host)->io_tree;
8774 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8775 btrfs_add_delayed_iput(inode);
8776 return ret;
8777 }
8778
8779 static int btrfs_writepages(struct address_space *mapping,
8780 struct writeback_control *wbc)
8781 {
8782 struct extent_io_tree *tree;
8783
8784 tree = &BTRFS_I(mapping->host)->io_tree;
8785 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8786 }
8787
8788 static int
8789 btrfs_readpages(struct file *file, struct address_space *mapping,
8790 struct list_head *pages, unsigned nr_pages)
8791 {
8792 struct extent_io_tree *tree;
8793 tree = &BTRFS_I(mapping->host)->io_tree;
8794 return extent_readpages(tree, mapping, pages, nr_pages,
8795 btrfs_get_extent);
8796 }
8797 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8798 {
8799 struct extent_io_tree *tree;
8800 struct extent_map_tree *map;
8801 int ret;
8802
8803 tree = &BTRFS_I(page->mapping->host)->io_tree;
8804 map = &BTRFS_I(page->mapping->host)->extent_tree;
8805 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8806 if (ret == 1) {
8807 ClearPagePrivate(page);
8808 set_page_private(page, 0);
8809 put_page(page);
8810 }
8811 return ret;
8812 }
8813
8814 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8815 {
8816 if (PageWriteback(page) || PageDirty(page))
8817 return 0;
8818 return __btrfs_releasepage(page, gfp_flags);
8819 }
8820
8821 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8822 unsigned int length)
8823 {
8824 struct inode *inode = page->mapping->host;
8825 struct extent_io_tree *tree;
8826 struct btrfs_ordered_extent *ordered;
8827 struct extent_state *cached_state = NULL;
8828 u64 page_start = page_offset(page);
8829 u64 page_end = page_start + PAGE_SIZE - 1;
8830 u64 start;
8831 u64 end;
8832 int inode_evicting = inode->i_state & I_FREEING;
8833
8834 /*
8835 * we have the page locked, so new writeback can't start,
8836 * and the dirty bit won't be cleared while we are here.
8837 *
8838 * Wait for IO on this page so that we can safely clear
8839 * the PagePrivate2 bit and do ordered accounting
8840 */
8841 wait_on_page_writeback(page);
8842
8843 tree = &BTRFS_I(inode)->io_tree;
8844 if (offset) {
8845 btrfs_releasepage(page, GFP_NOFS);
8846 return;
8847 }
8848
8849 if (!inode_evicting)
8850 lock_extent_bits(tree, page_start, page_end, &cached_state);
8851 again:
8852 start = page_start;
8853 ordered = btrfs_lookup_ordered_range(inode, start,
8854 page_end - start + 1);
8855 if (ordered) {
8856 end = min(page_end, ordered->file_offset + ordered->len - 1);
8857 /*
8858 * IO on this page will never be started, so we need
8859 * to account for any ordered extents now
8860 */
8861 if (!inode_evicting)
8862 clear_extent_bit(tree, start, end,
8863 EXTENT_DIRTY | EXTENT_DELALLOC |
8864 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8865 EXTENT_DEFRAG, 1, 0, &cached_state,
8866 GFP_NOFS);
8867 /*
8868 * whoever cleared the private bit is responsible
8869 * for the finish_ordered_io
8870 */
8871 if (TestClearPagePrivate2(page)) {
8872 struct btrfs_ordered_inode_tree *tree;
8873 u64 new_len;
8874
8875 tree = &BTRFS_I(inode)->ordered_tree;
8876
8877 spin_lock_irq(&tree->lock);
8878 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8879 new_len = start - ordered->file_offset;
8880 if (new_len < ordered->truncated_len)
8881 ordered->truncated_len = new_len;
8882 spin_unlock_irq(&tree->lock);
8883
8884 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8885 start,
8886 end - start + 1, 1))
8887 btrfs_finish_ordered_io(ordered);
8888 }
8889 btrfs_put_ordered_extent(ordered);
8890 if (!inode_evicting) {
8891 cached_state = NULL;
8892 lock_extent_bits(tree, start, end,
8893 &cached_state);
8894 }
8895
8896 start = end + 1;
8897 if (start < page_end)
8898 goto again;
8899 }
8900
8901 /*
8902 * Qgroup reserved space handler
8903 * Page here will be either
8904 * 1) Already written to disk
8905 * In this case, its reserved space is released from data rsv map
8906 * and will be freed by delayed_ref handler finally.
8907 * So even we call qgroup_free_data(), it won't decrease reserved
8908 * space.
8909 * 2) Not written to disk
8910 * This means the reserved space should be freed here. However,
8911 * if a truncate invalidates the page (by clearing PageDirty)
8912 * and the page is accounted for while allocating extent
8913 * in btrfs_check_data_free_space() we let delayed_ref to
8914 * free the entire extent.
8915 */
8916 if (PageDirty(page))
8917 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
8918 if (!inode_evicting) {
8919 clear_extent_bit(tree, page_start, page_end,
8920 EXTENT_LOCKED | EXTENT_DIRTY |
8921 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8922 EXTENT_DEFRAG, 1, 1,
8923 &cached_state, GFP_NOFS);
8924
8925 __btrfs_releasepage(page, GFP_NOFS);
8926 }
8927
8928 ClearPageChecked(page);
8929 if (PagePrivate(page)) {
8930 ClearPagePrivate(page);
8931 set_page_private(page, 0);
8932 put_page(page);
8933 }
8934 }
8935
8936 /*
8937 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8938 * called from a page fault handler when a page is first dirtied. Hence we must
8939 * be careful to check for EOF conditions here. We set the page up correctly
8940 * for a written page which means we get ENOSPC checking when writing into
8941 * holes and correct delalloc and unwritten extent mapping on filesystems that
8942 * support these features.
8943 *
8944 * We are not allowed to take the i_mutex here so we have to play games to
8945 * protect against truncate races as the page could now be beyond EOF. Because
8946 * vmtruncate() writes the inode size before removing pages, once we have the
8947 * page lock we can determine safely if the page is beyond EOF. If it is not
8948 * beyond EOF, then the page is guaranteed safe against truncation until we
8949 * unlock the page.
8950 */
8951 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8952 {
8953 struct page *page = vmf->page;
8954 struct inode *inode = file_inode(vma->vm_file);
8955 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8956 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8957 struct btrfs_ordered_extent *ordered;
8958 struct extent_state *cached_state = NULL;
8959 char *kaddr;
8960 unsigned long zero_start;
8961 loff_t size;
8962 int ret;
8963 int reserved = 0;
8964 u64 reserved_space;
8965 u64 page_start;
8966 u64 page_end;
8967 u64 end;
8968
8969 reserved_space = PAGE_SIZE;
8970
8971 sb_start_pagefault(inode->i_sb);
8972 page_start = page_offset(page);
8973 page_end = page_start + PAGE_SIZE - 1;
8974 end = page_end;
8975
8976 /*
8977 * Reserving delalloc space after obtaining the page lock can lead to
8978 * deadlock. For example, if a dirty page is locked by this function
8979 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8980 * dirty page write out, then the btrfs_writepage() function could
8981 * end up waiting indefinitely to get a lock on the page currently
8982 * being processed by btrfs_page_mkwrite() function.
8983 */
8984 ret = btrfs_delalloc_reserve_space(inode, page_start,
8985 reserved_space);
8986 if (!ret) {
8987 ret = file_update_time(vma->vm_file);
8988 reserved = 1;
8989 }
8990 if (ret) {
8991 if (ret == -ENOMEM)
8992 ret = VM_FAULT_OOM;
8993 else /* -ENOSPC, -EIO, etc */
8994 ret = VM_FAULT_SIGBUS;
8995 if (reserved)
8996 goto out;
8997 goto out_noreserve;
8998 }
8999
9000 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
9001 again:
9002 lock_page(page);
9003 size = i_size_read(inode);
9004
9005 if ((page->mapping != inode->i_mapping) ||
9006 (page_start >= size)) {
9007 /* page got truncated out from underneath us */
9008 goto out_unlock;
9009 }
9010 wait_on_page_writeback(page);
9011
9012 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
9013 set_page_extent_mapped(page);
9014
9015 /*
9016 * we can't set the delalloc bits if there are pending ordered
9017 * extents. Drop our locks and wait for them to finish
9018 */
9019 ordered = btrfs_lookup_ordered_range(inode, page_start, page_end);
9020 if (ordered) {
9021 unlock_extent_cached(io_tree, page_start, page_end,
9022 &cached_state, GFP_NOFS);
9023 unlock_page(page);
9024 btrfs_start_ordered_extent(inode, ordered, 1);
9025 btrfs_put_ordered_extent(ordered);
9026 goto again;
9027 }
9028
9029 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
9030 reserved_space = round_up(size - page_start,
9031 fs_info->sectorsize);
9032 if (reserved_space < PAGE_SIZE) {
9033 end = page_start + reserved_space - 1;
9034 spin_lock(&BTRFS_I(inode)->lock);
9035 BTRFS_I(inode)->outstanding_extents++;
9036 spin_unlock(&BTRFS_I(inode)->lock);
9037 btrfs_delalloc_release_space(inode, page_start,
9038 PAGE_SIZE - reserved_space);
9039 }
9040 }
9041
9042 /*
9043 * XXX - page_mkwrite gets called every time the page is dirtied, even
9044 * if it was already dirty, so for space accounting reasons we need to
9045 * clear any delalloc bits for the range we are fixing to save. There
9046 * is probably a better way to do this, but for now keep consistent with
9047 * prepare_pages in the normal write path.
9048 */
9049 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9050 EXTENT_DIRTY | EXTENT_DELALLOC |
9051 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9052 0, 0, &cached_state, GFP_NOFS);
9053
9054 ret = btrfs_set_extent_delalloc(inode, page_start, end,
9055 &cached_state, 0);
9056 if (ret) {
9057 unlock_extent_cached(io_tree, page_start, page_end,
9058 &cached_state, GFP_NOFS);
9059 ret = VM_FAULT_SIGBUS;
9060 goto out_unlock;
9061 }
9062 ret = 0;
9063
9064 /* page is wholly or partially inside EOF */
9065 if (page_start + PAGE_SIZE > size)
9066 zero_start = size & ~PAGE_MASK;
9067 else
9068 zero_start = PAGE_SIZE;
9069
9070 if (zero_start != PAGE_SIZE) {
9071 kaddr = kmap(page);
9072 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9073 flush_dcache_page(page);
9074 kunmap(page);
9075 }
9076 ClearPageChecked(page);
9077 set_page_dirty(page);
9078 SetPageUptodate(page);
9079
9080 BTRFS_I(inode)->last_trans = fs_info->generation;
9081 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9082 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9083
9084 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9085
9086 out_unlock:
9087 if (!ret) {
9088 sb_end_pagefault(inode->i_sb);
9089 return VM_FAULT_LOCKED;
9090 }
9091 unlock_page(page);
9092 out:
9093 btrfs_delalloc_release_space(inode, page_start, reserved_space);
9094 out_noreserve:
9095 sb_end_pagefault(inode->i_sb);
9096 return ret;
9097 }
9098
9099 static int btrfs_truncate(struct inode *inode)
9100 {
9101 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9102 struct btrfs_root *root = BTRFS_I(inode)->root;
9103 struct btrfs_block_rsv *rsv;
9104 int ret = 0;
9105 int err = 0;
9106 struct btrfs_trans_handle *trans;
9107 u64 mask = fs_info->sectorsize - 1;
9108 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9109
9110 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9111 (u64)-1);
9112 if (ret)
9113 return ret;
9114
9115 /*
9116 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9117 * 3 things going on here
9118 *
9119 * 1) We need to reserve space for our orphan item and the space to
9120 * delete our orphan item. Lord knows we don't want to have a dangling
9121 * orphan item because we didn't reserve space to remove it.
9122 *
9123 * 2) We need to reserve space to update our inode.
9124 *
9125 * 3) We need to have something to cache all the space that is going to
9126 * be free'd up by the truncate operation, but also have some slack
9127 * space reserved in case it uses space during the truncate (thank you
9128 * very much snapshotting).
9129 *
9130 * And we need these to all be separate. The fact is we can use a lot of
9131 * space doing the truncate, and we have no earthly idea how much space
9132 * we will use, so we need the truncate reservation to be separate so it
9133 * doesn't end up using space reserved for updating the inode or
9134 * removing the orphan item. We also need to be able to stop the
9135 * transaction and start a new one, which means we need to be able to
9136 * update the inode several times, and we have no idea of knowing how
9137 * many times that will be, so we can't just reserve 1 item for the
9138 * entirety of the operation, so that has to be done separately as well.
9139 * Then there is the orphan item, which does indeed need to be held on
9140 * to for the whole operation, and we need nobody to touch this reserved
9141 * space except the orphan code.
9142 *
9143 * So that leaves us with
9144 *
9145 * 1) root->orphan_block_rsv - for the orphan deletion.
9146 * 2) rsv - for the truncate reservation, which we will steal from the
9147 * transaction reservation.
9148 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9149 * updating the inode.
9150 */
9151 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9152 if (!rsv)
9153 return -ENOMEM;
9154 rsv->size = min_size;
9155 rsv->failfast = 1;
9156
9157 /*
9158 * 1 for the truncate slack space
9159 * 1 for updating the inode.
9160 */
9161 trans = btrfs_start_transaction(root, 2);
9162 if (IS_ERR(trans)) {
9163 err = PTR_ERR(trans);
9164 goto out;
9165 }
9166
9167 /* Migrate the slack space for the truncate to our reserve */
9168 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9169 min_size, 0);
9170 BUG_ON(ret);
9171
9172 /*
9173 * So if we truncate and then write and fsync we normally would just
9174 * write the extents that changed, which is a problem if we need to
9175 * first truncate that entire inode. So set this flag so we write out
9176 * all of the extents in the inode to the sync log so we're completely
9177 * safe.
9178 */
9179 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9180 trans->block_rsv = rsv;
9181
9182 while (1) {
9183 ret = btrfs_truncate_inode_items(trans, root, inode,
9184 inode->i_size,
9185 BTRFS_EXTENT_DATA_KEY);
9186 if (ret != -ENOSPC && ret != -EAGAIN) {
9187 err = ret;
9188 break;
9189 }
9190
9191 trans->block_rsv = &fs_info->trans_block_rsv;
9192 ret = btrfs_update_inode(trans, root, inode);
9193 if (ret) {
9194 err = ret;
9195 break;
9196 }
9197
9198 btrfs_end_transaction(trans);
9199 btrfs_btree_balance_dirty(fs_info);
9200
9201 trans = btrfs_start_transaction(root, 2);
9202 if (IS_ERR(trans)) {
9203 ret = err = PTR_ERR(trans);
9204 trans = NULL;
9205 break;
9206 }
9207
9208 btrfs_block_rsv_release(fs_info, rsv, -1);
9209 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9210 rsv, min_size, 0);
9211 BUG_ON(ret); /* shouldn't happen */
9212 trans->block_rsv = rsv;
9213 }
9214
9215 if (ret == 0 && inode->i_nlink > 0) {
9216 trans->block_rsv = root->orphan_block_rsv;
9217 ret = btrfs_orphan_del(trans, inode);
9218 if (ret)
9219 err = ret;
9220 }
9221
9222 if (trans) {
9223 trans->block_rsv = &fs_info->trans_block_rsv;
9224 ret = btrfs_update_inode(trans, root, inode);
9225 if (ret && !err)
9226 err = ret;
9227
9228 ret = btrfs_end_transaction(trans);
9229 btrfs_btree_balance_dirty(fs_info);
9230 }
9231 out:
9232 btrfs_free_block_rsv(fs_info, rsv);
9233
9234 if (ret && !err)
9235 err = ret;
9236
9237 return err;
9238 }
9239
9240 /*
9241 * create a new subvolume directory/inode (helper for the ioctl).
9242 */
9243 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9244 struct btrfs_root *new_root,
9245 struct btrfs_root *parent_root,
9246 u64 new_dirid)
9247 {
9248 struct inode *inode;
9249 int err;
9250 u64 index = 0;
9251
9252 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9253 new_dirid, new_dirid,
9254 S_IFDIR | (~current_umask() & S_IRWXUGO),
9255 &index);
9256 if (IS_ERR(inode))
9257 return PTR_ERR(inode);
9258 inode->i_op = &btrfs_dir_inode_operations;
9259 inode->i_fop = &btrfs_dir_file_operations;
9260
9261 set_nlink(inode, 1);
9262 btrfs_i_size_write(inode, 0);
9263 unlock_new_inode(inode);
9264
9265 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9266 if (err)
9267 btrfs_err(new_root->fs_info,
9268 "error inheriting subvolume %llu properties: %d",
9269 new_root->root_key.objectid, err);
9270
9271 err = btrfs_update_inode(trans, new_root, inode);
9272
9273 iput(inode);
9274 return err;
9275 }
9276
9277 struct inode *btrfs_alloc_inode(struct super_block *sb)
9278 {
9279 struct btrfs_inode *ei;
9280 struct inode *inode;
9281
9282 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9283 if (!ei)
9284 return NULL;
9285
9286 ei->root = NULL;
9287 ei->generation = 0;
9288 ei->last_trans = 0;
9289 ei->last_sub_trans = 0;
9290 ei->logged_trans = 0;
9291 ei->delalloc_bytes = 0;
9292 ei->defrag_bytes = 0;
9293 ei->disk_i_size = 0;
9294 ei->flags = 0;
9295 ei->csum_bytes = 0;
9296 ei->index_cnt = (u64)-1;
9297 ei->dir_index = 0;
9298 ei->last_unlink_trans = 0;
9299 ei->last_log_commit = 0;
9300 ei->delayed_iput_count = 0;
9301
9302 spin_lock_init(&ei->lock);
9303 ei->outstanding_extents = 0;
9304 ei->reserved_extents = 0;
9305
9306 ei->runtime_flags = 0;
9307 ei->force_compress = BTRFS_COMPRESS_NONE;
9308
9309 ei->delayed_node = NULL;
9310
9311 ei->i_otime.tv_sec = 0;
9312 ei->i_otime.tv_nsec = 0;
9313
9314 inode = &ei->vfs_inode;
9315 extent_map_tree_init(&ei->extent_tree);
9316 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9317 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9318 ei->io_tree.track_uptodate = 1;
9319 ei->io_failure_tree.track_uptodate = 1;
9320 atomic_set(&ei->sync_writers, 0);
9321 mutex_init(&ei->log_mutex);
9322 mutex_init(&ei->delalloc_mutex);
9323 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9324 INIT_LIST_HEAD(&ei->delalloc_inodes);
9325 INIT_LIST_HEAD(&ei->delayed_iput);
9326 RB_CLEAR_NODE(&ei->rb_node);
9327 init_rwsem(&ei->dio_sem);
9328
9329 return inode;
9330 }
9331
9332 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9333 void btrfs_test_destroy_inode(struct inode *inode)
9334 {
9335 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9336 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9337 }
9338 #endif
9339
9340 static void btrfs_i_callback(struct rcu_head *head)
9341 {
9342 struct inode *inode = container_of(head, struct inode, i_rcu);
9343 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9344 }
9345
9346 void btrfs_destroy_inode(struct inode *inode)
9347 {
9348 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9349 struct btrfs_ordered_extent *ordered;
9350 struct btrfs_root *root = BTRFS_I(inode)->root;
9351
9352 WARN_ON(!hlist_empty(&inode->i_dentry));
9353 WARN_ON(inode->i_data.nrpages);
9354 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9355 WARN_ON(BTRFS_I(inode)->reserved_extents);
9356 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9357 WARN_ON(BTRFS_I(inode)->csum_bytes);
9358 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9359
9360 /*
9361 * This can happen where we create an inode, but somebody else also
9362 * created the same inode and we need to destroy the one we already
9363 * created.
9364 */
9365 if (!root)
9366 goto free;
9367
9368 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9369 &BTRFS_I(inode)->runtime_flags)) {
9370 btrfs_info(fs_info, "inode %llu still on the orphan list",
9371 btrfs_ino(BTRFS_I(inode)));
9372 atomic_dec(&root->orphan_inodes);
9373 }
9374
9375 while (1) {
9376 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9377 if (!ordered)
9378 break;
9379 else {
9380 btrfs_err(fs_info,
9381 "found ordered extent %llu %llu on inode cleanup",
9382 ordered->file_offset, ordered->len);
9383 btrfs_remove_ordered_extent(inode, ordered);
9384 btrfs_put_ordered_extent(ordered);
9385 btrfs_put_ordered_extent(ordered);
9386 }
9387 }
9388 btrfs_qgroup_check_reserved_leak(inode);
9389 inode_tree_del(inode);
9390 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9391 free:
9392 call_rcu(&inode->i_rcu, btrfs_i_callback);
9393 }
9394
9395 int btrfs_drop_inode(struct inode *inode)
9396 {
9397 struct btrfs_root *root = BTRFS_I(inode)->root;
9398
9399 if (root == NULL)
9400 return 1;
9401
9402 /* the snap/subvol tree is on deleting */
9403 if (btrfs_root_refs(&root->root_item) == 0)
9404 return 1;
9405 else
9406 return generic_drop_inode(inode);
9407 }
9408
9409 static void init_once(void *foo)
9410 {
9411 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9412
9413 inode_init_once(&ei->vfs_inode);
9414 }
9415
9416 void btrfs_destroy_cachep(void)
9417 {
9418 /*
9419 * Make sure all delayed rcu free inodes are flushed before we
9420 * destroy cache.
9421 */
9422 rcu_barrier();
9423 kmem_cache_destroy(btrfs_inode_cachep);
9424 kmem_cache_destroy(btrfs_trans_handle_cachep);
9425 kmem_cache_destroy(btrfs_transaction_cachep);
9426 kmem_cache_destroy(btrfs_path_cachep);
9427 kmem_cache_destroy(btrfs_free_space_cachep);
9428 }
9429
9430 int btrfs_init_cachep(void)
9431 {
9432 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9433 sizeof(struct btrfs_inode), 0,
9434 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9435 init_once);
9436 if (!btrfs_inode_cachep)
9437 goto fail;
9438
9439 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9440 sizeof(struct btrfs_trans_handle), 0,
9441 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9442 if (!btrfs_trans_handle_cachep)
9443 goto fail;
9444
9445 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9446 sizeof(struct btrfs_transaction), 0,
9447 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9448 if (!btrfs_transaction_cachep)
9449 goto fail;
9450
9451 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9452 sizeof(struct btrfs_path), 0,
9453 SLAB_MEM_SPREAD, NULL);
9454 if (!btrfs_path_cachep)
9455 goto fail;
9456
9457 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9458 sizeof(struct btrfs_free_space), 0,
9459 SLAB_MEM_SPREAD, NULL);
9460 if (!btrfs_free_space_cachep)
9461 goto fail;
9462
9463 return 0;
9464 fail:
9465 btrfs_destroy_cachep();
9466 return -ENOMEM;
9467 }
9468
9469 static int btrfs_getattr(struct vfsmount *mnt,
9470 struct dentry *dentry, struct kstat *stat)
9471 {
9472 u64 delalloc_bytes;
9473 struct inode *inode = d_inode(dentry);
9474 u32 blocksize = inode->i_sb->s_blocksize;
9475
9476 generic_fillattr(inode, stat);
9477 stat->dev = BTRFS_I(inode)->root->anon_dev;
9478
9479 spin_lock(&BTRFS_I(inode)->lock);
9480 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9481 spin_unlock(&BTRFS_I(inode)->lock);
9482 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9483 ALIGN(delalloc_bytes, blocksize)) >> 9;
9484 return 0;
9485 }
9486
9487 static int btrfs_rename_exchange(struct inode *old_dir,
9488 struct dentry *old_dentry,
9489 struct inode *new_dir,
9490 struct dentry *new_dentry)
9491 {
9492 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9493 struct btrfs_trans_handle *trans;
9494 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9495 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9496 struct inode *new_inode = new_dentry->d_inode;
9497 struct inode *old_inode = old_dentry->d_inode;
9498 struct timespec ctime = current_time(old_inode);
9499 struct dentry *parent;
9500 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9501 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9502 u64 old_idx = 0;
9503 u64 new_idx = 0;
9504 u64 root_objectid;
9505 int ret;
9506 bool root_log_pinned = false;
9507 bool dest_log_pinned = false;
9508
9509 /* we only allow rename subvolume link between subvolumes */
9510 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9511 return -EXDEV;
9512
9513 /* close the race window with snapshot create/destroy ioctl */
9514 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9515 down_read(&fs_info->subvol_sem);
9516 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9517 down_read(&fs_info->subvol_sem);
9518
9519 /*
9520 * We want to reserve the absolute worst case amount of items. So if
9521 * both inodes are subvols and we need to unlink them then that would
9522 * require 4 item modifications, but if they are both normal inodes it
9523 * would require 5 item modifications, so we'll assume their normal
9524 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9525 * should cover the worst case number of items we'll modify.
9526 */
9527 trans = btrfs_start_transaction(root, 12);
9528 if (IS_ERR(trans)) {
9529 ret = PTR_ERR(trans);
9530 goto out_notrans;
9531 }
9532
9533 /*
9534 * We need to find a free sequence number both in the source and
9535 * in the destination directory for the exchange.
9536 */
9537 ret = btrfs_set_inode_index(new_dir, &old_idx);
9538 if (ret)
9539 goto out_fail;
9540 ret = btrfs_set_inode_index(old_dir, &new_idx);
9541 if (ret)
9542 goto out_fail;
9543
9544 BTRFS_I(old_inode)->dir_index = 0ULL;
9545 BTRFS_I(new_inode)->dir_index = 0ULL;
9546
9547 /* Reference for the source. */
9548 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9549 /* force full log commit if subvolume involved. */
9550 btrfs_set_log_full_commit(fs_info, trans);
9551 } else {
9552 btrfs_pin_log_trans(root);
9553 root_log_pinned = true;
9554 ret = btrfs_insert_inode_ref(trans, dest,
9555 new_dentry->d_name.name,
9556 new_dentry->d_name.len,
9557 old_ino,
9558 btrfs_ino(BTRFS_I(new_dir)), old_idx);
9559 if (ret)
9560 goto out_fail;
9561 }
9562
9563 /* And now for the dest. */
9564 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9565 /* force full log commit if subvolume involved. */
9566 btrfs_set_log_full_commit(fs_info, trans);
9567 } else {
9568 btrfs_pin_log_trans(dest);
9569 dest_log_pinned = true;
9570 ret = btrfs_insert_inode_ref(trans, root,
9571 old_dentry->d_name.name,
9572 old_dentry->d_name.len,
9573 new_ino,
9574 btrfs_ino(BTRFS_I(old_dir)), new_idx);
9575 if (ret)
9576 goto out_fail;
9577 }
9578
9579 /* Update inode version and ctime/mtime. */
9580 inode_inc_iversion(old_dir);
9581 inode_inc_iversion(new_dir);
9582 inode_inc_iversion(old_inode);
9583 inode_inc_iversion(new_inode);
9584 old_dir->i_ctime = old_dir->i_mtime = ctime;
9585 new_dir->i_ctime = new_dir->i_mtime = ctime;
9586 old_inode->i_ctime = ctime;
9587 new_inode->i_ctime = ctime;
9588
9589 if (old_dentry->d_parent != new_dentry->d_parent) {
9590 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9591 btrfs_record_unlink_dir(trans, new_dir, new_inode, 1);
9592 }
9593
9594 /* src is a subvolume */
9595 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9596 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9597 ret = btrfs_unlink_subvol(trans, root, old_dir,
9598 root_objectid,
9599 old_dentry->d_name.name,
9600 old_dentry->d_name.len);
9601 } else { /* src is an inode */
9602 ret = __btrfs_unlink_inode(trans, root, old_dir,
9603 old_dentry->d_inode,
9604 old_dentry->d_name.name,
9605 old_dentry->d_name.len);
9606 if (!ret)
9607 ret = btrfs_update_inode(trans, root, old_inode);
9608 }
9609 if (ret) {
9610 btrfs_abort_transaction(trans, ret);
9611 goto out_fail;
9612 }
9613
9614 /* dest is a subvolume */
9615 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9616 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9617 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9618 root_objectid,
9619 new_dentry->d_name.name,
9620 new_dentry->d_name.len);
9621 } else { /* dest is an inode */
9622 ret = __btrfs_unlink_inode(trans, dest, new_dir,
9623 new_dentry->d_inode,
9624 new_dentry->d_name.name,
9625 new_dentry->d_name.len);
9626 if (!ret)
9627 ret = btrfs_update_inode(trans, dest, new_inode);
9628 }
9629 if (ret) {
9630 btrfs_abort_transaction(trans, ret);
9631 goto out_fail;
9632 }
9633
9634 ret = btrfs_add_link(trans, new_dir, old_inode,
9635 new_dentry->d_name.name,
9636 new_dentry->d_name.len, 0, old_idx);
9637 if (ret) {
9638 btrfs_abort_transaction(trans, ret);
9639 goto out_fail;
9640 }
9641
9642 ret = btrfs_add_link(trans, old_dir, new_inode,
9643 old_dentry->d_name.name,
9644 old_dentry->d_name.len, 0, new_idx);
9645 if (ret) {
9646 btrfs_abort_transaction(trans, ret);
9647 goto out_fail;
9648 }
9649
9650 if (old_inode->i_nlink == 1)
9651 BTRFS_I(old_inode)->dir_index = old_idx;
9652 if (new_inode->i_nlink == 1)
9653 BTRFS_I(new_inode)->dir_index = new_idx;
9654
9655 if (root_log_pinned) {
9656 parent = new_dentry->d_parent;
9657 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9658 btrfs_end_log_trans(root);
9659 root_log_pinned = false;
9660 }
9661 if (dest_log_pinned) {
9662 parent = old_dentry->d_parent;
9663 btrfs_log_new_name(trans, new_inode, new_dir, parent);
9664 btrfs_end_log_trans(dest);
9665 dest_log_pinned = false;
9666 }
9667 out_fail:
9668 /*
9669 * If we have pinned a log and an error happened, we unpin tasks
9670 * trying to sync the log and force them to fallback to a transaction
9671 * commit if the log currently contains any of the inodes involved in
9672 * this rename operation (to ensure we do not persist a log with an
9673 * inconsistent state for any of these inodes or leading to any
9674 * inconsistencies when replayed). If the transaction was aborted, the
9675 * abortion reason is propagated to userspace when attempting to commit
9676 * the transaction. If the log does not contain any of these inodes, we
9677 * allow the tasks to sync it.
9678 */
9679 if (ret && (root_log_pinned || dest_log_pinned)) {
9680 if (btrfs_inode_in_log(old_dir, fs_info->generation) ||
9681 btrfs_inode_in_log(new_dir, fs_info->generation) ||
9682 btrfs_inode_in_log(old_inode, fs_info->generation) ||
9683 (new_inode &&
9684 btrfs_inode_in_log(new_inode, fs_info->generation)))
9685 btrfs_set_log_full_commit(fs_info, trans);
9686
9687 if (root_log_pinned) {
9688 btrfs_end_log_trans(root);
9689 root_log_pinned = false;
9690 }
9691 if (dest_log_pinned) {
9692 btrfs_end_log_trans(dest);
9693 dest_log_pinned = false;
9694 }
9695 }
9696 ret = btrfs_end_transaction(trans);
9697 out_notrans:
9698 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9699 up_read(&fs_info->subvol_sem);
9700 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9701 up_read(&fs_info->subvol_sem);
9702
9703 return ret;
9704 }
9705
9706 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9707 struct btrfs_root *root,
9708 struct inode *dir,
9709 struct dentry *dentry)
9710 {
9711 int ret;
9712 struct inode *inode;
9713 u64 objectid;
9714 u64 index;
9715
9716 ret = btrfs_find_free_ino(root, &objectid);
9717 if (ret)
9718 return ret;
9719
9720 inode = btrfs_new_inode(trans, root, dir,
9721 dentry->d_name.name,
9722 dentry->d_name.len,
9723 btrfs_ino(BTRFS_I(dir)),
9724 objectid,
9725 S_IFCHR | WHITEOUT_MODE,
9726 &index);
9727
9728 if (IS_ERR(inode)) {
9729 ret = PTR_ERR(inode);
9730 return ret;
9731 }
9732
9733 inode->i_op = &btrfs_special_inode_operations;
9734 init_special_inode(inode, inode->i_mode,
9735 WHITEOUT_DEV);
9736
9737 ret = btrfs_init_inode_security(trans, inode, dir,
9738 &dentry->d_name);
9739 if (ret)
9740 goto out;
9741
9742 ret = btrfs_add_nondir(trans, dir, dentry,
9743 inode, 0, index);
9744 if (ret)
9745 goto out;
9746
9747 ret = btrfs_update_inode(trans, root, inode);
9748 out:
9749 unlock_new_inode(inode);
9750 if (ret)
9751 inode_dec_link_count(inode);
9752 iput(inode);
9753
9754 return ret;
9755 }
9756
9757 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9758 struct inode *new_dir, struct dentry *new_dentry,
9759 unsigned int flags)
9760 {
9761 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9762 struct btrfs_trans_handle *trans;
9763 unsigned int trans_num_items;
9764 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9765 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9766 struct inode *new_inode = d_inode(new_dentry);
9767 struct inode *old_inode = d_inode(old_dentry);
9768 u64 index = 0;
9769 u64 root_objectid;
9770 int ret;
9771 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9772 bool log_pinned = false;
9773
9774 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9775 return -EPERM;
9776
9777 /* we only allow rename subvolume link between subvolumes */
9778 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9779 return -EXDEV;
9780
9781 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9782 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9783 return -ENOTEMPTY;
9784
9785 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9786 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9787 return -ENOTEMPTY;
9788
9789
9790 /* check for collisions, even if the name isn't there */
9791 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9792 new_dentry->d_name.name,
9793 new_dentry->d_name.len);
9794
9795 if (ret) {
9796 if (ret == -EEXIST) {
9797 /* we shouldn't get
9798 * eexist without a new_inode */
9799 if (WARN_ON(!new_inode)) {
9800 return ret;
9801 }
9802 } else {
9803 /* maybe -EOVERFLOW */
9804 return ret;
9805 }
9806 }
9807 ret = 0;
9808
9809 /*
9810 * we're using rename to replace one file with another. Start IO on it
9811 * now so we don't add too much work to the end of the transaction
9812 */
9813 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9814 filemap_flush(old_inode->i_mapping);
9815
9816 /* close the racy window with snapshot create/destroy ioctl */
9817 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9818 down_read(&fs_info->subvol_sem);
9819 /*
9820 * We want to reserve the absolute worst case amount of items. So if
9821 * both inodes are subvols and we need to unlink them then that would
9822 * require 4 item modifications, but if they are both normal inodes it
9823 * would require 5 item modifications, so we'll assume they are normal
9824 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9825 * should cover the worst case number of items we'll modify.
9826 * If our rename has the whiteout flag, we need more 5 units for the
9827 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9828 * when selinux is enabled).
9829 */
9830 trans_num_items = 11;
9831 if (flags & RENAME_WHITEOUT)
9832 trans_num_items += 5;
9833 trans = btrfs_start_transaction(root, trans_num_items);
9834 if (IS_ERR(trans)) {
9835 ret = PTR_ERR(trans);
9836 goto out_notrans;
9837 }
9838
9839 if (dest != root)
9840 btrfs_record_root_in_trans(trans, dest);
9841
9842 ret = btrfs_set_inode_index(new_dir, &index);
9843 if (ret)
9844 goto out_fail;
9845
9846 BTRFS_I(old_inode)->dir_index = 0ULL;
9847 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9848 /* force full log commit if subvolume involved. */
9849 btrfs_set_log_full_commit(fs_info, trans);
9850 } else {
9851 btrfs_pin_log_trans(root);
9852 log_pinned = true;
9853 ret = btrfs_insert_inode_ref(trans, dest,
9854 new_dentry->d_name.name,
9855 new_dentry->d_name.len,
9856 old_ino,
9857 btrfs_ino(BTRFS_I(new_dir)), index);
9858 if (ret)
9859 goto out_fail;
9860 }
9861
9862 inode_inc_iversion(old_dir);
9863 inode_inc_iversion(new_dir);
9864 inode_inc_iversion(old_inode);
9865 old_dir->i_ctime = old_dir->i_mtime =
9866 new_dir->i_ctime = new_dir->i_mtime =
9867 old_inode->i_ctime = current_time(old_dir);
9868
9869 if (old_dentry->d_parent != new_dentry->d_parent)
9870 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9871
9872 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9873 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9874 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9875 old_dentry->d_name.name,
9876 old_dentry->d_name.len);
9877 } else {
9878 ret = __btrfs_unlink_inode(trans, root, old_dir,
9879 d_inode(old_dentry),
9880 old_dentry->d_name.name,
9881 old_dentry->d_name.len);
9882 if (!ret)
9883 ret = btrfs_update_inode(trans, root, old_inode);
9884 }
9885 if (ret) {
9886 btrfs_abort_transaction(trans, ret);
9887 goto out_fail;
9888 }
9889
9890 if (new_inode) {
9891 inode_inc_iversion(new_inode);
9892 new_inode->i_ctime = current_time(new_inode);
9893 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9894 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9895 root_objectid = BTRFS_I(new_inode)->location.objectid;
9896 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9897 root_objectid,
9898 new_dentry->d_name.name,
9899 new_dentry->d_name.len);
9900 BUG_ON(new_inode->i_nlink == 0);
9901 } else {
9902 ret = btrfs_unlink_inode(trans, dest, new_dir,
9903 d_inode(new_dentry),
9904 new_dentry->d_name.name,
9905 new_dentry->d_name.len);
9906 }
9907 if (!ret && new_inode->i_nlink == 0)
9908 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9909 if (ret) {
9910 btrfs_abort_transaction(trans, ret);
9911 goto out_fail;
9912 }
9913 }
9914
9915 ret = btrfs_add_link(trans, new_dir, old_inode,
9916 new_dentry->d_name.name,
9917 new_dentry->d_name.len, 0, index);
9918 if (ret) {
9919 btrfs_abort_transaction(trans, ret);
9920 goto out_fail;
9921 }
9922
9923 if (old_inode->i_nlink == 1)
9924 BTRFS_I(old_inode)->dir_index = index;
9925
9926 if (log_pinned) {
9927 struct dentry *parent = new_dentry->d_parent;
9928
9929 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9930 btrfs_end_log_trans(root);
9931 log_pinned = false;
9932 }
9933
9934 if (flags & RENAME_WHITEOUT) {
9935 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9936 old_dentry);
9937
9938 if (ret) {
9939 btrfs_abort_transaction(trans, ret);
9940 goto out_fail;
9941 }
9942 }
9943 out_fail:
9944 /*
9945 * If we have pinned the log and an error happened, we unpin tasks
9946 * trying to sync the log and force them to fallback to a transaction
9947 * commit if the log currently contains any of the inodes involved in
9948 * this rename operation (to ensure we do not persist a log with an
9949 * inconsistent state for any of these inodes or leading to any
9950 * inconsistencies when replayed). If the transaction was aborted, the
9951 * abortion reason is propagated to userspace when attempting to commit
9952 * the transaction. If the log does not contain any of these inodes, we
9953 * allow the tasks to sync it.
9954 */
9955 if (ret && log_pinned) {
9956 if (btrfs_inode_in_log(old_dir, fs_info->generation) ||
9957 btrfs_inode_in_log(new_dir, fs_info->generation) ||
9958 btrfs_inode_in_log(old_inode, fs_info->generation) ||
9959 (new_inode &&
9960 btrfs_inode_in_log(new_inode, fs_info->generation)))
9961 btrfs_set_log_full_commit(fs_info, trans);
9962
9963 btrfs_end_log_trans(root);
9964 log_pinned = false;
9965 }
9966 btrfs_end_transaction(trans);
9967 out_notrans:
9968 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9969 up_read(&fs_info->subvol_sem);
9970
9971 return ret;
9972 }
9973
9974 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9975 struct inode *new_dir, struct dentry *new_dentry,
9976 unsigned int flags)
9977 {
9978 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9979 return -EINVAL;
9980
9981 if (flags & RENAME_EXCHANGE)
9982 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9983 new_dentry);
9984
9985 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9986 }
9987
9988 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9989 {
9990 struct btrfs_delalloc_work *delalloc_work;
9991 struct inode *inode;
9992
9993 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9994 work);
9995 inode = delalloc_work->inode;
9996 filemap_flush(inode->i_mapping);
9997 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9998 &BTRFS_I(inode)->runtime_flags))
9999 filemap_flush(inode->i_mapping);
10000
10001 if (delalloc_work->delay_iput)
10002 btrfs_add_delayed_iput(inode);
10003 else
10004 iput(inode);
10005 complete(&delalloc_work->completion);
10006 }
10007
10008 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
10009 int delay_iput)
10010 {
10011 struct btrfs_delalloc_work *work;
10012
10013 work = kmalloc(sizeof(*work), GFP_NOFS);
10014 if (!work)
10015 return NULL;
10016
10017 init_completion(&work->completion);
10018 INIT_LIST_HEAD(&work->list);
10019 work->inode = inode;
10020 work->delay_iput = delay_iput;
10021 WARN_ON_ONCE(!inode);
10022 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10023 btrfs_run_delalloc_work, NULL, NULL);
10024
10025 return work;
10026 }
10027
10028 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10029 {
10030 wait_for_completion(&work->completion);
10031 kfree(work);
10032 }
10033
10034 /*
10035 * some fairly slow code that needs optimization. This walks the list
10036 * of all the inodes with pending delalloc and forces them to disk.
10037 */
10038 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10039 int nr)
10040 {
10041 struct btrfs_inode *binode;
10042 struct inode *inode;
10043 struct btrfs_delalloc_work *work, *next;
10044 struct list_head works;
10045 struct list_head splice;
10046 int ret = 0;
10047
10048 INIT_LIST_HEAD(&works);
10049 INIT_LIST_HEAD(&splice);
10050
10051 mutex_lock(&root->delalloc_mutex);
10052 spin_lock(&root->delalloc_lock);
10053 list_splice_init(&root->delalloc_inodes, &splice);
10054 while (!list_empty(&splice)) {
10055 binode = list_entry(splice.next, struct btrfs_inode,
10056 delalloc_inodes);
10057
10058 list_move_tail(&binode->delalloc_inodes,
10059 &root->delalloc_inodes);
10060 inode = igrab(&binode->vfs_inode);
10061 if (!inode) {
10062 cond_resched_lock(&root->delalloc_lock);
10063 continue;
10064 }
10065 spin_unlock(&root->delalloc_lock);
10066
10067 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10068 if (!work) {
10069 if (delay_iput)
10070 btrfs_add_delayed_iput(inode);
10071 else
10072 iput(inode);
10073 ret = -ENOMEM;
10074 goto out;
10075 }
10076 list_add_tail(&work->list, &works);
10077 btrfs_queue_work(root->fs_info->flush_workers,
10078 &work->work);
10079 ret++;
10080 if (nr != -1 && ret >= nr)
10081 goto out;
10082 cond_resched();
10083 spin_lock(&root->delalloc_lock);
10084 }
10085 spin_unlock(&root->delalloc_lock);
10086
10087 out:
10088 list_for_each_entry_safe(work, next, &works, list) {
10089 list_del_init(&work->list);
10090 btrfs_wait_and_free_delalloc_work(work);
10091 }
10092
10093 if (!list_empty_careful(&splice)) {
10094 spin_lock(&root->delalloc_lock);
10095 list_splice_tail(&splice, &root->delalloc_inodes);
10096 spin_unlock(&root->delalloc_lock);
10097 }
10098 mutex_unlock(&root->delalloc_mutex);
10099 return ret;
10100 }
10101
10102 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10103 {
10104 struct btrfs_fs_info *fs_info = root->fs_info;
10105 int ret;
10106
10107 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10108 return -EROFS;
10109
10110 ret = __start_delalloc_inodes(root, delay_iput, -1);
10111 if (ret > 0)
10112 ret = 0;
10113 /*
10114 * the filemap_flush will queue IO into the worker threads, but
10115 * we have to make sure the IO is actually started and that
10116 * ordered extents get created before we return
10117 */
10118 atomic_inc(&fs_info->async_submit_draining);
10119 while (atomic_read(&fs_info->nr_async_submits) ||
10120 atomic_read(&fs_info->async_delalloc_pages)) {
10121 wait_event(fs_info->async_submit_wait,
10122 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10123 atomic_read(&fs_info->async_delalloc_pages) == 0));
10124 }
10125 atomic_dec(&fs_info->async_submit_draining);
10126 return ret;
10127 }
10128
10129 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10130 int nr)
10131 {
10132 struct btrfs_root *root;
10133 struct list_head splice;
10134 int ret;
10135
10136 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10137 return -EROFS;
10138
10139 INIT_LIST_HEAD(&splice);
10140
10141 mutex_lock(&fs_info->delalloc_root_mutex);
10142 spin_lock(&fs_info->delalloc_root_lock);
10143 list_splice_init(&fs_info->delalloc_roots, &splice);
10144 while (!list_empty(&splice) && nr) {
10145 root = list_first_entry(&splice, struct btrfs_root,
10146 delalloc_root);
10147 root = btrfs_grab_fs_root(root);
10148 BUG_ON(!root);
10149 list_move_tail(&root->delalloc_root,
10150 &fs_info->delalloc_roots);
10151 spin_unlock(&fs_info->delalloc_root_lock);
10152
10153 ret = __start_delalloc_inodes(root, delay_iput, nr);
10154 btrfs_put_fs_root(root);
10155 if (ret < 0)
10156 goto out;
10157
10158 if (nr != -1) {
10159 nr -= ret;
10160 WARN_ON(nr < 0);
10161 }
10162 spin_lock(&fs_info->delalloc_root_lock);
10163 }
10164 spin_unlock(&fs_info->delalloc_root_lock);
10165
10166 ret = 0;
10167 atomic_inc(&fs_info->async_submit_draining);
10168 while (atomic_read(&fs_info->nr_async_submits) ||
10169 atomic_read(&fs_info->async_delalloc_pages)) {
10170 wait_event(fs_info->async_submit_wait,
10171 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10172 atomic_read(&fs_info->async_delalloc_pages) == 0));
10173 }
10174 atomic_dec(&fs_info->async_submit_draining);
10175 out:
10176 if (!list_empty_careful(&splice)) {
10177 spin_lock(&fs_info->delalloc_root_lock);
10178 list_splice_tail(&splice, &fs_info->delalloc_roots);
10179 spin_unlock(&fs_info->delalloc_root_lock);
10180 }
10181 mutex_unlock(&fs_info->delalloc_root_mutex);
10182 return ret;
10183 }
10184
10185 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10186 const char *symname)
10187 {
10188 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10189 struct btrfs_trans_handle *trans;
10190 struct btrfs_root *root = BTRFS_I(dir)->root;
10191 struct btrfs_path *path;
10192 struct btrfs_key key;
10193 struct inode *inode = NULL;
10194 int err;
10195 int drop_inode = 0;
10196 u64 objectid;
10197 u64 index = 0;
10198 int name_len;
10199 int datasize;
10200 unsigned long ptr;
10201 struct btrfs_file_extent_item *ei;
10202 struct extent_buffer *leaf;
10203
10204 name_len = strlen(symname);
10205 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10206 return -ENAMETOOLONG;
10207
10208 /*
10209 * 2 items for inode item and ref
10210 * 2 items for dir items
10211 * 1 item for updating parent inode item
10212 * 1 item for the inline extent item
10213 * 1 item for xattr if selinux is on
10214 */
10215 trans = btrfs_start_transaction(root, 7);
10216 if (IS_ERR(trans))
10217 return PTR_ERR(trans);
10218
10219 err = btrfs_find_free_ino(root, &objectid);
10220 if (err)
10221 goto out_unlock;
10222
10223 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10224 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
10225 S_IFLNK|S_IRWXUGO, &index);
10226 if (IS_ERR(inode)) {
10227 err = PTR_ERR(inode);
10228 goto out_unlock;
10229 }
10230
10231 /*
10232 * If the active LSM wants to access the inode during
10233 * d_instantiate it needs these. Smack checks to see
10234 * if the filesystem supports xattrs by looking at the
10235 * ops vector.
10236 */
10237 inode->i_fop = &btrfs_file_operations;
10238 inode->i_op = &btrfs_file_inode_operations;
10239 inode->i_mapping->a_ops = &btrfs_aops;
10240 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10241
10242 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10243 if (err)
10244 goto out_unlock_inode;
10245
10246 path = btrfs_alloc_path();
10247 if (!path) {
10248 err = -ENOMEM;
10249 goto out_unlock_inode;
10250 }
10251 key.objectid = btrfs_ino(BTRFS_I(inode));
10252 key.offset = 0;
10253 key.type = BTRFS_EXTENT_DATA_KEY;
10254 datasize = btrfs_file_extent_calc_inline_size(name_len);
10255 err = btrfs_insert_empty_item(trans, root, path, &key,
10256 datasize);
10257 if (err) {
10258 btrfs_free_path(path);
10259 goto out_unlock_inode;
10260 }
10261 leaf = path->nodes[0];
10262 ei = btrfs_item_ptr(leaf, path->slots[0],
10263 struct btrfs_file_extent_item);
10264 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10265 btrfs_set_file_extent_type(leaf, ei,
10266 BTRFS_FILE_EXTENT_INLINE);
10267 btrfs_set_file_extent_encryption(leaf, ei, 0);
10268 btrfs_set_file_extent_compression(leaf, ei, 0);
10269 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10270 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10271
10272 ptr = btrfs_file_extent_inline_start(ei);
10273 write_extent_buffer(leaf, symname, ptr, name_len);
10274 btrfs_mark_buffer_dirty(leaf);
10275 btrfs_free_path(path);
10276
10277 inode->i_op = &btrfs_symlink_inode_operations;
10278 inode_nohighmem(inode);
10279 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10280 inode_set_bytes(inode, name_len);
10281 btrfs_i_size_write(inode, name_len);
10282 err = btrfs_update_inode(trans, root, inode);
10283 /*
10284 * Last step, add directory indexes for our symlink inode. This is the
10285 * last step to avoid extra cleanup of these indexes if an error happens
10286 * elsewhere above.
10287 */
10288 if (!err)
10289 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
10290 if (err) {
10291 drop_inode = 1;
10292 goto out_unlock_inode;
10293 }
10294
10295 unlock_new_inode(inode);
10296 d_instantiate(dentry, inode);
10297
10298 out_unlock:
10299 btrfs_end_transaction(trans);
10300 if (drop_inode) {
10301 inode_dec_link_count(inode);
10302 iput(inode);
10303 }
10304 btrfs_btree_balance_dirty(fs_info);
10305 return err;
10306
10307 out_unlock_inode:
10308 drop_inode = 1;
10309 unlock_new_inode(inode);
10310 goto out_unlock;
10311 }
10312
10313 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10314 u64 start, u64 num_bytes, u64 min_size,
10315 loff_t actual_len, u64 *alloc_hint,
10316 struct btrfs_trans_handle *trans)
10317 {
10318 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10319 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10320 struct extent_map *em;
10321 struct btrfs_root *root = BTRFS_I(inode)->root;
10322 struct btrfs_key ins;
10323 u64 cur_offset = start;
10324 u64 i_size;
10325 u64 cur_bytes;
10326 u64 last_alloc = (u64)-1;
10327 int ret = 0;
10328 bool own_trans = true;
10329 u64 end = start + num_bytes - 1;
10330
10331 if (trans)
10332 own_trans = false;
10333 while (num_bytes > 0) {
10334 if (own_trans) {
10335 trans = btrfs_start_transaction(root, 3);
10336 if (IS_ERR(trans)) {
10337 ret = PTR_ERR(trans);
10338 break;
10339 }
10340 }
10341
10342 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10343 cur_bytes = max(cur_bytes, min_size);
10344 /*
10345 * If we are severely fragmented we could end up with really
10346 * small allocations, so if the allocator is returning small
10347 * chunks lets make its job easier by only searching for those
10348 * sized chunks.
10349 */
10350 cur_bytes = min(cur_bytes, last_alloc);
10351 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10352 min_size, 0, *alloc_hint, &ins, 1, 0);
10353 if (ret) {
10354 if (own_trans)
10355 btrfs_end_transaction(trans);
10356 break;
10357 }
10358 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10359
10360 last_alloc = ins.offset;
10361 ret = insert_reserved_file_extent(trans, inode,
10362 cur_offset, ins.objectid,
10363 ins.offset, ins.offset,
10364 ins.offset, 0, 0, 0,
10365 BTRFS_FILE_EXTENT_PREALLOC);
10366 if (ret) {
10367 btrfs_free_reserved_extent(fs_info, ins.objectid,
10368 ins.offset, 0);
10369 btrfs_abort_transaction(trans, ret);
10370 if (own_trans)
10371 btrfs_end_transaction(trans);
10372 break;
10373 }
10374
10375 btrfs_drop_extent_cache(inode, cur_offset,
10376 cur_offset + ins.offset -1, 0);
10377
10378 em = alloc_extent_map();
10379 if (!em) {
10380 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10381 &BTRFS_I(inode)->runtime_flags);
10382 goto next;
10383 }
10384
10385 em->start = cur_offset;
10386 em->orig_start = cur_offset;
10387 em->len = ins.offset;
10388 em->block_start = ins.objectid;
10389 em->block_len = ins.offset;
10390 em->orig_block_len = ins.offset;
10391 em->ram_bytes = ins.offset;
10392 em->bdev = fs_info->fs_devices->latest_bdev;
10393 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10394 em->generation = trans->transid;
10395
10396 while (1) {
10397 write_lock(&em_tree->lock);
10398 ret = add_extent_mapping(em_tree, em, 1);
10399 write_unlock(&em_tree->lock);
10400 if (ret != -EEXIST)
10401 break;
10402 btrfs_drop_extent_cache(inode, cur_offset,
10403 cur_offset + ins.offset - 1,
10404 0);
10405 }
10406 free_extent_map(em);
10407 next:
10408 num_bytes -= ins.offset;
10409 cur_offset += ins.offset;
10410 *alloc_hint = ins.objectid + ins.offset;
10411
10412 inode_inc_iversion(inode);
10413 inode->i_ctime = current_time(inode);
10414 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10415 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10416 (actual_len > inode->i_size) &&
10417 (cur_offset > inode->i_size)) {
10418 if (cur_offset > actual_len)
10419 i_size = actual_len;
10420 else
10421 i_size = cur_offset;
10422 i_size_write(inode, i_size);
10423 btrfs_ordered_update_i_size(inode, i_size, NULL);
10424 }
10425
10426 ret = btrfs_update_inode(trans, root, inode);
10427
10428 if (ret) {
10429 btrfs_abort_transaction(trans, ret);
10430 if (own_trans)
10431 btrfs_end_transaction(trans);
10432 break;
10433 }
10434
10435 if (own_trans)
10436 btrfs_end_transaction(trans);
10437 }
10438 if (cur_offset < end)
10439 btrfs_free_reserved_data_space(inode, cur_offset,
10440 end - cur_offset + 1);
10441 return ret;
10442 }
10443
10444 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10445 u64 start, u64 num_bytes, u64 min_size,
10446 loff_t actual_len, u64 *alloc_hint)
10447 {
10448 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10449 min_size, actual_len, alloc_hint,
10450 NULL);
10451 }
10452
10453 int btrfs_prealloc_file_range_trans(struct inode *inode,
10454 struct btrfs_trans_handle *trans, int mode,
10455 u64 start, u64 num_bytes, u64 min_size,
10456 loff_t actual_len, u64 *alloc_hint)
10457 {
10458 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10459 min_size, actual_len, alloc_hint, trans);
10460 }
10461
10462 static int btrfs_set_page_dirty(struct page *page)
10463 {
10464 return __set_page_dirty_nobuffers(page);
10465 }
10466
10467 static int btrfs_permission(struct inode *inode, int mask)
10468 {
10469 struct btrfs_root *root = BTRFS_I(inode)->root;
10470 umode_t mode = inode->i_mode;
10471
10472 if (mask & MAY_WRITE &&
10473 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10474 if (btrfs_root_readonly(root))
10475 return -EROFS;
10476 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10477 return -EACCES;
10478 }
10479 return generic_permission(inode, mask);
10480 }
10481
10482 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10483 {
10484 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10485 struct btrfs_trans_handle *trans;
10486 struct btrfs_root *root = BTRFS_I(dir)->root;
10487 struct inode *inode = NULL;
10488 u64 objectid;
10489 u64 index;
10490 int ret = 0;
10491
10492 /*
10493 * 5 units required for adding orphan entry
10494 */
10495 trans = btrfs_start_transaction(root, 5);
10496 if (IS_ERR(trans))
10497 return PTR_ERR(trans);
10498
10499 ret = btrfs_find_free_ino(root, &objectid);
10500 if (ret)
10501 goto out;
10502
10503 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10504 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10505 if (IS_ERR(inode)) {
10506 ret = PTR_ERR(inode);
10507 inode = NULL;
10508 goto out;
10509 }
10510
10511 inode->i_fop = &btrfs_file_operations;
10512 inode->i_op = &btrfs_file_inode_operations;
10513
10514 inode->i_mapping->a_ops = &btrfs_aops;
10515 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10516
10517 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10518 if (ret)
10519 goto out_inode;
10520
10521 ret = btrfs_update_inode(trans, root, inode);
10522 if (ret)
10523 goto out_inode;
10524 ret = btrfs_orphan_add(trans, inode);
10525 if (ret)
10526 goto out_inode;
10527
10528 /*
10529 * We set number of links to 0 in btrfs_new_inode(), and here we set
10530 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10531 * through:
10532 *
10533 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10534 */
10535 set_nlink(inode, 1);
10536 unlock_new_inode(inode);
10537 d_tmpfile(dentry, inode);
10538 mark_inode_dirty(inode);
10539
10540 out:
10541 btrfs_end_transaction(trans);
10542 if (ret)
10543 iput(inode);
10544 btrfs_balance_delayed_items(fs_info);
10545 btrfs_btree_balance_dirty(fs_info);
10546 return ret;
10547
10548 out_inode:
10549 unlock_new_inode(inode);
10550 goto out;
10551
10552 }
10553
10554 static const struct inode_operations btrfs_dir_inode_operations = {
10555 .getattr = btrfs_getattr,
10556 .lookup = btrfs_lookup,
10557 .create = btrfs_create,
10558 .unlink = btrfs_unlink,
10559 .link = btrfs_link,
10560 .mkdir = btrfs_mkdir,
10561 .rmdir = btrfs_rmdir,
10562 .rename = btrfs_rename2,
10563 .symlink = btrfs_symlink,
10564 .setattr = btrfs_setattr,
10565 .mknod = btrfs_mknod,
10566 .listxattr = btrfs_listxattr,
10567 .permission = btrfs_permission,
10568 .get_acl = btrfs_get_acl,
10569 .set_acl = btrfs_set_acl,
10570 .update_time = btrfs_update_time,
10571 .tmpfile = btrfs_tmpfile,
10572 };
10573 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10574 .lookup = btrfs_lookup,
10575 .permission = btrfs_permission,
10576 .update_time = btrfs_update_time,
10577 };
10578
10579 static const struct file_operations btrfs_dir_file_operations = {
10580 .llseek = generic_file_llseek,
10581 .read = generic_read_dir,
10582 .iterate_shared = btrfs_real_readdir,
10583 .unlocked_ioctl = btrfs_ioctl,
10584 #ifdef CONFIG_COMPAT
10585 .compat_ioctl = btrfs_compat_ioctl,
10586 #endif
10587 .release = btrfs_release_file,
10588 .fsync = btrfs_sync_file,
10589 };
10590
10591 static const struct extent_io_ops btrfs_extent_io_ops = {
10592 .fill_delalloc = run_delalloc_range,
10593 .submit_bio_hook = btrfs_submit_bio_hook,
10594 .merge_bio_hook = btrfs_merge_bio_hook,
10595 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10596 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10597 .writepage_start_hook = btrfs_writepage_start_hook,
10598 .set_bit_hook = btrfs_set_bit_hook,
10599 .clear_bit_hook = btrfs_clear_bit_hook,
10600 .merge_extent_hook = btrfs_merge_extent_hook,
10601 .split_extent_hook = btrfs_split_extent_hook,
10602 };
10603
10604 /*
10605 * btrfs doesn't support the bmap operation because swapfiles
10606 * use bmap to make a mapping of extents in the file. They assume
10607 * these extents won't change over the life of the file and they
10608 * use the bmap result to do IO directly to the drive.
10609 *
10610 * the btrfs bmap call would return logical addresses that aren't
10611 * suitable for IO and they also will change frequently as COW
10612 * operations happen. So, swapfile + btrfs == corruption.
10613 *
10614 * For now we're avoiding this by dropping bmap.
10615 */
10616 static const struct address_space_operations btrfs_aops = {
10617 .readpage = btrfs_readpage,
10618 .writepage = btrfs_writepage,
10619 .writepages = btrfs_writepages,
10620 .readpages = btrfs_readpages,
10621 .direct_IO = btrfs_direct_IO,
10622 .invalidatepage = btrfs_invalidatepage,
10623 .releasepage = btrfs_releasepage,
10624 .set_page_dirty = btrfs_set_page_dirty,
10625 .error_remove_page = generic_error_remove_page,
10626 };
10627
10628 static const struct address_space_operations btrfs_symlink_aops = {
10629 .readpage = btrfs_readpage,
10630 .writepage = btrfs_writepage,
10631 .invalidatepage = btrfs_invalidatepage,
10632 .releasepage = btrfs_releasepage,
10633 };
10634
10635 static const struct inode_operations btrfs_file_inode_operations = {
10636 .getattr = btrfs_getattr,
10637 .setattr = btrfs_setattr,
10638 .listxattr = btrfs_listxattr,
10639 .permission = btrfs_permission,
10640 .fiemap = btrfs_fiemap,
10641 .get_acl = btrfs_get_acl,
10642 .set_acl = btrfs_set_acl,
10643 .update_time = btrfs_update_time,
10644 };
10645 static const struct inode_operations btrfs_special_inode_operations = {
10646 .getattr = btrfs_getattr,
10647 .setattr = btrfs_setattr,
10648 .permission = btrfs_permission,
10649 .listxattr = btrfs_listxattr,
10650 .get_acl = btrfs_get_acl,
10651 .set_acl = btrfs_set_acl,
10652 .update_time = btrfs_update_time,
10653 };
10654 static const struct inode_operations btrfs_symlink_inode_operations = {
10655 .get_link = page_get_link,
10656 .getattr = btrfs_getattr,
10657 .setattr = btrfs_setattr,
10658 .permission = btrfs_permission,
10659 .listxattr = btrfs_listxattr,
10660 .update_time = btrfs_update_time,
10661 };
10662
10663 const struct dentry_operations btrfs_dentry_operations = {
10664 .d_delete = btrfs_dentry_delete,
10665 .d_release = btrfs_dentry_release,
10666 };
ubuntu-bionic-kernel mirror