projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: track transid for delayed ref flushing
[deliverable/linux.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/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
42 #include <linux/btrfs.h>
43 #include <linux/blkdev.h>
44 #include <linux/posix_acl_xattr.h>
45 #include <linux/uio.h>
46 #include "ctree.h"
47 #include "disk-io.h"
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
52 #include "xattr.h"
53 #include "tree-log.h"
54 #include "volumes.h"
55 #include "compression.h"
56 #include "locking.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
59 #include "backref.h"
60 #include "hash.h"
61 #include "props.h"
62 #include "qgroup.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, int *page_started,
109 unsigned long *nr_written, int unlock);
110 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
111 u64 len, u64 orig_start,
112 u64 block_start, u64 block_len,
113 u64 orig_block_len, u64 ram_bytes,
114 int type);
115
116 static int btrfs_dirty_inode(struct inode *inode);
117
118 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
119 void btrfs_test_inode_set_ops(struct inode *inode)
120 {
121 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
122 }
123 #endif
124
125 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
126 struct inode *inode, struct inode *dir,
127 const struct qstr *qstr)
128 {
129 int err;
130
131 err = btrfs_init_acl(trans, inode, dir);
132 if (!err)
133 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
134 return err;
135 }
136
137 /*
138 * this does all the hard work for inserting an inline extent into
139 * the btree. The caller should have done a btrfs_drop_extents so that
140 * no overlapping inline items exist in the btree
141 */
142 static int insert_inline_extent(struct btrfs_trans_handle *trans,
143 struct btrfs_path *path, int extent_inserted,
144 struct btrfs_root *root, struct inode *inode,
145 u64 start, size_t size, size_t compressed_size,
146 int compress_type,
147 struct page **compressed_pages)
148 {
149 struct extent_buffer *leaf;
150 struct page *page = NULL;
151 char *kaddr;
152 unsigned long ptr;
153 struct btrfs_file_extent_item *ei;
154 int err = 0;
155 int ret;
156 size_t cur_size = size;
157 unsigned long offset;
158
159 if (compressed_size && compressed_pages)
160 cur_size = compressed_size;
161
162 inode_add_bytes(inode, size);
163
164 if (!extent_inserted) {
165 struct btrfs_key key;
166 size_t datasize;
167
168 key.objectid = btrfs_ino(inode);
169 key.offset = start;
170 key.type = BTRFS_EXTENT_DATA_KEY;
171
172 datasize = btrfs_file_extent_calc_inline_size(cur_size);
173 path->leave_spinning = 1;
174 ret = btrfs_insert_empty_item(trans, root, path, &key,
175 datasize);
176 if (ret) {
177 err = ret;
178 goto fail;
179 }
180 }
181 leaf = path->nodes[0];
182 ei = btrfs_item_ptr(leaf, path->slots[0],
183 struct btrfs_file_extent_item);
184 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
185 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
186 btrfs_set_file_extent_encryption(leaf, ei, 0);
187 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
188 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
189 ptr = btrfs_file_extent_inline_start(ei);
190
191 if (compress_type != BTRFS_COMPRESS_NONE) {
192 struct page *cpage;
193 int i = 0;
194 while (compressed_size > 0) {
195 cpage = compressed_pages[i];
196 cur_size = min_t(unsigned long, compressed_size,
197 PAGE_SIZE);
198
199 kaddr = kmap_atomic(cpage);
200 write_extent_buffer(leaf, kaddr, ptr, cur_size);
201 kunmap_atomic(kaddr);
202
203 i++;
204 ptr += cur_size;
205 compressed_size -= cur_size;
206 }
207 btrfs_set_file_extent_compression(leaf, ei,
208 compress_type);
209 } else {
210 page = find_get_page(inode->i_mapping,
211 start >> PAGE_SHIFT);
212 btrfs_set_file_extent_compression(leaf, ei, 0);
213 kaddr = kmap_atomic(page);
214 offset = start & (PAGE_SIZE - 1);
215 write_extent_buffer(leaf, kaddr + offset, ptr, size);
216 kunmap_atomic(kaddr);
217 put_page(page);
218 }
219 btrfs_mark_buffer_dirty(leaf);
220 btrfs_release_path(path);
221
222 /*
223 * we're an inline extent, so nobody can
224 * extend the file past i_size without locking
225 * a page we already have locked.
226 *
227 * We must do any isize and inode updates
228 * before we unlock the pages. Otherwise we
229 * could end up racing with unlink.
230 */
231 BTRFS_I(inode)->disk_i_size = inode->i_size;
232 ret = btrfs_update_inode(trans, root, inode);
233
234 return ret;
235 fail:
236 return err;
237 }
238
239
240 /*
241 * conditionally insert an inline extent into the file. This
242 * does the checks required to make sure the data is small enough
243 * to fit as an inline extent.
244 */
245 static noinline int cow_file_range_inline(struct btrfs_root *root,
246 struct inode *inode, u64 start,
247 u64 end, size_t compressed_size,
248 int compress_type,
249 struct page **compressed_pages)
250 {
251 struct btrfs_trans_handle *trans;
252 u64 isize = i_size_read(inode);
253 u64 actual_end = min(end + 1, isize);
254 u64 inline_len = actual_end - start;
255 u64 aligned_end = ALIGN(end, root->sectorsize);
256 u64 data_len = inline_len;
257 int ret;
258 struct btrfs_path *path;
259 int extent_inserted = 0;
260 u32 extent_item_size;
261
262 if (compressed_size)
263 data_len = compressed_size;
264
265 if (start > 0 ||
266 actual_end > root->sectorsize ||
267 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
268 (!compressed_size &&
269 (actual_end & (root->sectorsize - 1)) == 0) ||
270 end + 1 < isize ||
271 data_len > root->fs_info->max_inline) {
272 return 1;
273 }
274
275 path = btrfs_alloc_path();
276 if (!path)
277 return -ENOMEM;
278
279 trans = btrfs_join_transaction(root);
280 if (IS_ERR(trans)) {
281 btrfs_free_path(path);
282 return PTR_ERR(trans);
283 }
284 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
285
286 if (compressed_size && compressed_pages)
287 extent_item_size = btrfs_file_extent_calc_inline_size(
288 compressed_size);
289 else
290 extent_item_size = btrfs_file_extent_calc_inline_size(
291 inline_len);
292
293 ret = __btrfs_drop_extents(trans, root, inode, path,
294 start, aligned_end, NULL,
295 1, 1, extent_item_size, &extent_inserted);
296 if (ret) {
297 btrfs_abort_transaction(trans, root, ret);
298 goto out;
299 }
300
301 if (isize > actual_end)
302 inline_len = min_t(u64, isize, actual_end);
303 ret = insert_inline_extent(trans, path, extent_inserted,
304 root, inode, start,
305 inline_len, compressed_size,
306 compress_type, compressed_pages);
307 if (ret && ret != -ENOSPC) {
308 btrfs_abort_transaction(trans, root, ret);
309 goto out;
310 } else if (ret == -ENOSPC) {
311 ret = 1;
312 goto out;
313 }
314
315 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
316 btrfs_delalloc_release_metadata(inode, end + 1 - start);
317 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
318 out:
319 /*
320 * Don't forget to free the reserved space, as for inlined extent
321 * it won't count as data extent, free them directly here.
322 * And at reserve time, it's always aligned to page size, so
323 * just free one page here.
324 */
325 btrfs_qgroup_free_data(inode, 0, PAGE_SIZE);
326 btrfs_free_path(path);
327 btrfs_end_transaction(trans, root);
328 return ret;
329 }
330
331 struct async_extent {
332 u64 start;
333 u64 ram_size;
334 u64 compressed_size;
335 struct page **pages;
336 unsigned long nr_pages;
337 int compress_type;
338 struct list_head list;
339 };
340
341 struct async_cow {
342 struct inode *inode;
343 struct btrfs_root *root;
344 struct page *locked_page;
345 u64 start;
346 u64 end;
347 struct list_head extents;
348 struct btrfs_work work;
349 };
350
351 static noinline int add_async_extent(struct async_cow *cow,
352 u64 start, u64 ram_size,
353 u64 compressed_size,
354 struct page **pages,
355 unsigned long nr_pages,
356 int compress_type)
357 {
358 struct async_extent *async_extent;
359
360 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
361 BUG_ON(!async_extent); /* -ENOMEM */
362 async_extent->start = start;
363 async_extent->ram_size = ram_size;
364 async_extent->compressed_size = compressed_size;
365 async_extent->pages = pages;
366 async_extent->nr_pages = nr_pages;
367 async_extent->compress_type = compress_type;
368 list_add_tail(&async_extent->list, &cow->extents);
369 return 0;
370 }
371
372 static inline int inode_need_compress(struct inode *inode)
373 {
374 struct btrfs_root *root = BTRFS_I(inode)->root;
375
376 /* force compress */
377 if (btrfs_test_opt(root, FORCE_COMPRESS))
378 return 1;
379 /* bad compression ratios */
380 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
381 return 0;
382 if (btrfs_test_opt(root, COMPRESS) ||
383 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
384 BTRFS_I(inode)->force_compress)
385 return 1;
386 return 0;
387 }
388
389 /*
390 * we create compressed extents in two phases. The first
391 * phase compresses a range of pages that have already been
392 * locked (both pages and state bits are locked).
393 *
394 * This is done inside an ordered work queue, and the compression
395 * is spread across many cpus. The actual IO submission is step
396 * two, and the ordered work queue takes care of making sure that
397 * happens in the same order things were put onto the queue by
398 * writepages and friends.
399 *
400 * If this code finds it can't get good compression, it puts an
401 * entry onto the work queue to write the uncompressed bytes. This
402 * makes sure that both compressed inodes and uncompressed inodes
403 * are written in the same order that the flusher thread sent them
404 * down.
405 */
406 static noinline void compress_file_range(struct inode *inode,
407 struct page *locked_page,
408 u64 start, u64 end,
409 struct async_cow *async_cow,
410 int *num_added)
411 {
412 struct btrfs_root *root = BTRFS_I(inode)->root;
413 u64 num_bytes;
414 u64 blocksize = root->sectorsize;
415 u64 actual_end;
416 u64 isize = i_size_read(inode);
417 int ret = 0;
418 struct page **pages = NULL;
419 unsigned long nr_pages;
420 unsigned long nr_pages_ret = 0;
421 unsigned long total_compressed = 0;
422 unsigned long total_in = 0;
423 unsigned long max_compressed = SZ_128K;
424 unsigned long max_uncompressed = SZ_128K;
425 int i;
426 int will_compress;
427 int compress_type = root->fs_info->compress_type;
428 int redirty = 0;
429
430 /* if this is a small write inside eof, kick off a defrag */
431 if ((end - start + 1) < SZ_16K &&
432 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
433 btrfs_add_inode_defrag(NULL, inode);
434
435 actual_end = min_t(u64, isize, end + 1);
436 again:
437 will_compress = 0;
438 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
439 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_SIZE);
440
441 /*
442 * we don't want to send crud past the end of i_size through
443 * compression, that's just a waste of CPU time. So, if the
444 * end of the file is before the start of our current
445 * requested range of bytes, we bail out to the uncompressed
446 * cleanup code that can deal with all of this.
447 *
448 * It isn't really the fastest way to fix things, but this is a
449 * very uncommon corner.
450 */
451 if (actual_end <= start)
452 goto cleanup_and_bail_uncompressed;
453
454 total_compressed = actual_end - start;
455
456 /*
457 * skip compression for a small file range(<=blocksize) that
458 * isn't an inline extent, since it doesn't save disk space at all.
459 */
460 if (total_compressed <= blocksize &&
461 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
462 goto cleanup_and_bail_uncompressed;
463
464 /* we want to make sure that amount of ram required to uncompress
465 * an extent is reasonable, so we limit the total size in ram
466 * of a compressed extent to 128k. This is a crucial number
467 * because it also controls how easily we can spread reads across
468 * cpus for decompression.
469 *
470 * We also want to make sure the amount of IO required to do
471 * a random read is reasonably small, so we limit the size of
472 * a compressed extent to 128k.
473 */
474 total_compressed = min(total_compressed, max_uncompressed);
475 num_bytes = ALIGN(end - start + 1, blocksize);
476 num_bytes = max(blocksize, num_bytes);
477 total_in = 0;
478 ret = 0;
479
480 /*
481 * we do compression for mount -o compress and when the
482 * inode has not been flagged as nocompress. This flag can
483 * change at any time if we discover bad compression ratios.
484 */
485 if (inode_need_compress(inode)) {
486 WARN_ON(pages);
487 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
488 if (!pages) {
489 /* just bail out to the uncompressed code */
490 goto cont;
491 }
492
493 if (BTRFS_I(inode)->force_compress)
494 compress_type = BTRFS_I(inode)->force_compress;
495
496 /*
497 * we need to call clear_page_dirty_for_io on each
498 * page in the range. Otherwise applications with the file
499 * mmap'd can wander in and change the page contents while
500 * we are compressing them.
501 *
502 * If the compression fails for any reason, we set the pages
503 * dirty again later on.
504 */
505 extent_range_clear_dirty_for_io(inode, start, end);
506 redirty = 1;
507 ret = btrfs_compress_pages(compress_type,
508 inode->i_mapping, start,
509 total_compressed, pages,
510 nr_pages, &nr_pages_ret,
511 &total_in,
512 &total_compressed,
513 max_compressed);
514
515 if (!ret) {
516 unsigned long offset = total_compressed &
517 (PAGE_SIZE - 1);
518 struct page *page = pages[nr_pages_ret - 1];
519 char *kaddr;
520
521 /* zero the tail end of the last page, we might be
522 * sending it down to disk
523 */
524 if (offset) {
525 kaddr = kmap_atomic(page);
526 memset(kaddr + offset, 0,
527 PAGE_SIZE - offset);
528 kunmap_atomic(kaddr);
529 }
530 will_compress = 1;
531 }
532 }
533 cont:
534 if (start == 0) {
535 /* lets try to make an inline extent */
536 if (ret || total_in < (actual_end - start)) {
537 /* we didn't compress the entire range, try
538 * to make an uncompressed inline extent.
539 */
540 ret = cow_file_range_inline(root, inode, start, end,
541 0, 0, NULL);
542 } else {
543 /* try making a compressed inline extent */
544 ret = cow_file_range_inline(root, inode, start, end,
545 total_compressed,
546 compress_type, pages);
547 }
548 if (ret <= 0) {
549 unsigned long clear_flags = EXTENT_DELALLOC |
550 EXTENT_DEFRAG;
551 unsigned long page_error_op;
552
553 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
554 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
555
556 /*
557 * inline extent creation worked or returned error,
558 * we don't need to create any more async work items.
559 * Unlock and free up our temp pages.
560 */
561 extent_clear_unlock_delalloc(inode, start, end, NULL,
562 clear_flags, PAGE_UNLOCK |
563 PAGE_CLEAR_DIRTY |
564 PAGE_SET_WRITEBACK |
565 page_error_op |
566 PAGE_END_WRITEBACK);
567 goto free_pages_out;
568 }
569 }
570
571 if (will_compress) {
572 /*
573 * we aren't doing an inline extent round the compressed size
574 * up to a block size boundary so the allocator does sane
575 * things
576 */
577 total_compressed = ALIGN(total_compressed, blocksize);
578
579 /*
580 * one last check to make sure the compression is really a
581 * win, compare the page count read with the blocks on disk
582 */
583 total_in = ALIGN(total_in, PAGE_SIZE);
584 if (total_compressed >= total_in) {
585 will_compress = 0;
586 } else {
587 num_bytes = total_in;
588 }
589 }
590 if (!will_compress && pages) {
591 /*
592 * the compression code ran but failed to make things smaller,
593 * free any pages it allocated and our page pointer array
594 */
595 for (i = 0; i < nr_pages_ret; i++) {
596 WARN_ON(pages[i]->mapping);
597 put_page(pages[i]);
598 }
599 kfree(pages);
600 pages = NULL;
601 total_compressed = 0;
602 nr_pages_ret = 0;
603
604 /* flag the file so we don't compress in the future */
605 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
606 !(BTRFS_I(inode)->force_compress)) {
607 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
608 }
609 }
610 if (will_compress) {
611 *num_added += 1;
612
613 /* the async work queues will take care of doing actual
614 * allocation on disk for these compressed pages,
615 * and will submit them to the elevator.
616 */
617 add_async_extent(async_cow, start, num_bytes,
618 total_compressed, pages, nr_pages_ret,
619 compress_type);
620
621 if (start + num_bytes < end) {
622 start += num_bytes;
623 pages = NULL;
624 cond_resched();
625 goto again;
626 }
627 } else {
628 cleanup_and_bail_uncompressed:
629 /*
630 * No compression, but we still need to write the pages in
631 * the file we've been given so far. redirty the locked
632 * page if it corresponds to our extent and set things up
633 * for the async work queue to run cow_file_range to do
634 * the normal delalloc dance
635 */
636 if (page_offset(locked_page) >= start &&
637 page_offset(locked_page) <= end) {
638 __set_page_dirty_nobuffers(locked_page);
639 /* unlocked later on in the async handlers */
640 }
641 if (redirty)
642 extent_range_redirty_for_io(inode, start, end);
643 add_async_extent(async_cow, start, end - start + 1,
644 0, NULL, 0, BTRFS_COMPRESS_NONE);
645 *num_added += 1;
646 }
647
648 return;
649
650 free_pages_out:
651 for (i = 0; i < nr_pages_ret; i++) {
652 WARN_ON(pages[i]->mapping);
653 put_page(pages[i]);
654 }
655 kfree(pages);
656 }
657
658 static void free_async_extent_pages(struct async_extent *async_extent)
659 {
660 int i;
661
662 if (!async_extent->pages)
663 return;
664
665 for (i = 0; i < async_extent->nr_pages; i++) {
666 WARN_ON(async_extent->pages[i]->mapping);
667 put_page(async_extent->pages[i]);
668 }
669 kfree(async_extent->pages);
670 async_extent->nr_pages = 0;
671 async_extent->pages = NULL;
672 }
673
674 /*
675 * phase two of compressed writeback. This is the ordered portion
676 * of the code, which only gets called in the order the work was
677 * queued. We walk all the async extents created by compress_file_range
678 * and send them down to the disk.
679 */
680 static noinline void submit_compressed_extents(struct inode *inode,
681 struct async_cow *async_cow)
682 {
683 struct async_extent *async_extent;
684 u64 alloc_hint = 0;
685 struct btrfs_key ins;
686 struct extent_map *em;
687 struct btrfs_root *root = BTRFS_I(inode)->root;
688 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
689 struct extent_io_tree *io_tree;
690 int ret = 0;
691
692 again:
693 while (!list_empty(&async_cow->extents)) {
694 async_extent = list_entry(async_cow->extents.next,
695 struct async_extent, list);
696 list_del(&async_extent->list);
697
698 io_tree = &BTRFS_I(inode)->io_tree;
699
700 retry:
701 /* did the compression code fall back to uncompressed IO? */
702 if (!async_extent->pages) {
703 int page_started = 0;
704 unsigned long nr_written = 0;
705
706 lock_extent(io_tree, async_extent->start,
707 async_extent->start +
708 async_extent->ram_size - 1);
709
710 /* allocate blocks */
711 ret = cow_file_range(inode, async_cow->locked_page,
712 async_extent->start,
713 async_extent->start +
714 async_extent->ram_size - 1,
715 &page_started, &nr_written, 0);
716
717 /* JDM XXX */
718
719 /*
720 * if page_started, cow_file_range inserted an
721 * inline extent and took care of all the unlocking
722 * and IO for us. Otherwise, we need to submit
723 * all those pages down to the drive.
724 */
725 if (!page_started && !ret)
726 extent_write_locked_range(io_tree,
727 inode, async_extent->start,
728 async_extent->start +
729 async_extent->ram_size - 1,
730 btrfs_get_extent,
731 WB_SYNC_ALL);
732 else if (ret)
733 unlock_page(async_cow->locked_page);
734 kfree(async_extent);
735 cond_resched();
736 continue;
737 }
738
739 lock_extent(io_tree, async_extent->start,
740 async_extent->start + async_extent->ram_size - 1);
741
742 ret = btrfs_reserve_extent(root,
743 async_extent->compressed_size,
744 async_extent->compressed_size,
745 0, alloc_hint, &ins, 1, 1);
746 if (ret) {
747 free_async_extent_pages(async_extent);
748
749 if (ret == -ENOSPC) {
750 unlock_extent(io_tree, async_extent->start,
751 async_extent->start +
752 async_extent->ram_size - 1);
753
754 /*
755 * we need to redirty the pages if we decide to
756 * fallback to uncompressed IO, otherwise we
757 * will not submit these pages down to lower
758 * layers.
759 */
760 extent_range_redirty_for_io(inode,
761 async_extent->start,
762 async_extent->start +
763 async_extent->ram_size - 1);
764
765 goto retry;
766 }
767 goto out_free;
768 }
769 /*
770 * here we're doing allocation and writeback of the
771 * compressed pages
772 */
773 btrfs_drop_extent_cache(inode, async_extent->start,
774 async_extent->start +
775 async_extent->ram_size - 1, 0);
776
777 em = alloc_extent_map();
778 if (!em) {
779 ret = -ENOMEM;
780 goto out_free_reserve;
781 }
782 em->start = async_extent->start;
783 em->len = async_extent->ram_size;
784 em->orig_start = em->start;
785 em->mod_start = em->start;
786 em->mod_len = em->len;
787
788 em->block_start = ins.objectid;
789 em->block_len = ins.offset;
790 em->orig_block_len = ins.offset;
791 em->ram_bytes = async_extent->ram_size;
792 em->bdev = root->fs_info->fs_devices->latest_bdev;
793 em->compress_type = async_extent->compress_type;
794 set_bit(EXTENT_FLAG_PINNED, &em->flags);
795 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
796 em->generation = -1;
797
798 while (1) {
799 write_lock(&em_tree->lock);
800 ret = add_extent_mapping(em_tree, em, 1);
801 write_unlock(&em_tree->lock);
802 if (ret != -EEXIST) {
803 free_extent_map(em);
804 break;
805 }
806 btrfs_drop_extent_cache(inode, async_extent->start,
807 async_extent->start +
808 async_extent->ram_size - 1, 0);
809 }
810
811 if (ret)
812 goto out_free_reserve;
813
814 ret = btrfs_add_ordered_extent_compress(inode,
815 async_extent->start,
816 ins.objectid,
817 async_extent->ram_size,
818 ins.offset,
819 BTRFS_ORDERED_COMPRESSED,
820 async_extent->compress_type);
821 if (ret) {
822 btrfs_drop_extent_cache(inode, async_extent->start,
823 async_extent->start +
824 async_extent->ram_size - 1, 0);
825 goto out_free_reserve;
826 }
827 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
828
829 /*
830 * clear dirty, set writeback and unlock the pages.
831 */
832 extent_clear_unlock_delalloc(inode, async_extent->start,
833 async_extent->start +
834 async_extent->ram_size - 1,
835 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
836 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
837 PAGE_SET_WRITEBACK);
838 ret = btrfs_submit_compressed_write(inode,
839 async_extent->start,
840 async_extent->ram_size,
841 ins.objectid,
842 ins.offset, async_extent->pages,
843 async_extent->nr_pages);
844 if (ret) {
845 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
846 struct page *p = async_extent->pages[0];
847 const u64 start = async_extent->start;
848 const u64 end = start + async_extent->ram_size - 1;
849
850 p->mapping = inode->i_mapping;
851 tree->ops->writepage_end_io_hook(p, start, end,
852 NULL, 0);
853 p->mapping = NULL;
854 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
855 PAGE_END_WRITEBACK |
856 PAGE_SET_ERROR);
857 free_async_extent_pages(async_extent);
858 }
859 alloc_hint = ins.objectid + ins.offset;
860 kfree(async_extent);
861 cond_resched();
862 }
863 return;
864 out_free_reserve:
865 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
866 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
867 out_free:
868 extent_clear_unlock_delalloc(inode, async_extent->start,
869 async_extent->start +
870 async_extent->ram_size - 1,
871 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
872 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
873 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
874 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
875 PAGE_SET_ERROR);
876 free_async_extent_pages(async_extent);
877 kfree(async_extent);
878 goto again;
879 }
880
881 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
882 u64 num_bytes)
883 {
884 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
885 struct extent_map *em;
886 u64 alloc_hint = 0;
887
888 read_lock(&em_tree->lock);
889 em = search_extent_mapping(em_tree, start, num_bytes);
890 if (em) {
891 /*
892 * if block start isn't an actual block number then find the
893 * first block in this inode and use that as a hint. If that
894 * block is also bogus then just don't worry about it.
895 */
896 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
897 free_extent_map(em);
898 em = search_extent_mapping(em_tree, 0, 0);
899 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
900 alloc_hint = em->block_start;
901 if (em)
902 free_extent_map(em);
903 } else {
904 alloc_hint = em->block_start;
905 free_extent_map(em);
906 }
907 }
908 read_unlock(&em_tree->lock);
909
910 return alloc_hint;
911 }
912
913 /*
914 * when extent_io.c finds a delayed allocation range in the file,
915 * the call backs end up in this code. The basic idea is to
916 * allocate extents on disk for the range, and create ordered data structs
917 * in ram to track those extents.
918 *
919 * locked_page is the page that writepage had locked already. We use
920 * it to make sure we don't do extra locks or unlocks.
921 *
922 * *page_started is set to one if we unlock locked_page and do everything
923 * required to start IO on it. It may be clean and already done with
924 * IO when we return.
925 */
926 static noinline int cow_file_range(struct inode *inode,
927 struct page *locked_page,
928 u64 start, u64 end, int *page_started,
929 unsigned long *nr_written,
930 int unlock)
931 {
932 struct btrfs_root *root = BTRFS_I(inode)->root;
933 u64 alloc_hint = 0;
934 u64 num_bytes;
935 unsigned long ram_size;
936 u64 disk_num_bytes;
937 u64 cur_alloc_size;
938 u64 blocksize = root->sectorsize;
939 struct btrfs_key ins;
940 struct extent_map *em;
941 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
942 int ret = 0;
943
944 if (btrfs_is_free_space_inode(inode)) {
945 WARN_ON_ONCE(1);
946 ret = -EINVAL;
947 goto out_unlock;
948 }
949
950 num_bytes = ALIGN(end - start + 1, blocksize);
951 num_bytes = max(blocksize, num_bytes);
952 disk_num_bytes = num_bytes;
953
954 /* if this is a small write inside eof, kick off defrag */
955 if (num_bytes < SZ_64K &&
956 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
957 btrfs_add_inode_defrag(NULL, inode);
958
959 if (start == 0) {
960 /* lets try to make an inline extent */
961 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
962 NULL);
963 if (ret == 0) {
964 extent_clear_unlock_delalloc(inode, start, end, NULL,
965 EXTENT_LOCKED | EXTENT_DELALLOC |
966 EXTENT_DEFRAG, PAGE_UNLOCK |
967 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
968 PAGE_END_WRITEBACK);
969
970 *nr_written = *nr_written +
971 (end - start + PAGE_SIZE) / PAGE_SIZE;
972 *page_started = 1;
973 goto out;
974 } else if (ret < 0) {
975 goto out_unlock;
976 }
977 }
978
979 BUG_ON(disk_num_bytes >
980 btrfs_super_total_bytes(root->fs_info->super_copy));
981
982 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
983 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
984
985 while (disk_num_bytes > 0) {
986 unsigned long op;
987
988 cur_alloc_size = disk_num_bytes;
989 ret = btrfs_reserve_extent(root, cur_alloc_size,
990 root->sectorsize, 0, alloc_hint,
991 &ins, 1, 1);
992 if (ret < 0)
993 goto out_unlock;
994
995 em = alloc_extent_map();
996 if (!em) {
997 ret = -ENOMEM;
998 goto out_reserve;
999 }
1000 em->start = start;
1001 em->orig_start = em->start;
1002 ram_size = ins.offset;
1003 em->len = ins.offset;
1004 em->mod_start = em->start;
1005 em->mod_len = em->len;
1006
1007 em->block_start = ins.objectid;
1008 em->block_len = ins.offset;
1009 em->orig_block_len = ins.offset;
1010 em->ram_bytes = ram_size;
1011 em->bdev = root->fs_info->fs_devices->latest_bdev;
1012 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1013 em->generation = -1;
1014
1015 while (1) {
1016 write_lock(&em_tree->lock);
1017 ret = add_extent_mapping(em_tree, em, 1);
1018 write_unlock(&em_tree->lock);
1019 if (ret != -EEXIST) {
1020 free_extent_map(em);
1021 break;
1022 }
1023 btrfs_drop_extent_cache(inode, start,
1024 start + ram_size - 1, 0);
1025 }
1026 if (ret)
1027 goto out_reserve;
1028
1029 cur_alloc_size = ins.offset;
1030 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1031 ram_size, cur_alloc_size, 0);
1032 if (ret)
1033 goto out_drop_extent_cache;
1034
1035 if (root->root_key.objectid ==
1036 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1037 ret = btrfs_reloc_clone_csums(inode, start,
1038 cur_alloc_size);
1039 if (ret)
1040 goto out_drop_extent_cache;
1041 }
1042
1043 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
1044
1045 if (disk_num_bytes < cur_alloc_size)
1046 break;
1047
1048 /* we're not doing compressed IO, don't unlock the first
1049 * page (which the caller expects to stay locked), don't
1050 * clear any dirty bits and don't set any writeback bits
1051 *
1052 * Do set the Private2 bit so we know this page was properly
1053 * setup for writepage
1054 */
1055 op = unlock ? PAGE_UNLOCK : 0;
1056 op |= PAGE_SET_PRIVATE2;
1057
1058 extent_clear_unlock_delalloc(inode, start,
1059 start + ram_size - 1, locked_page,
1060 EXTENT_LOCKED | EXTENT_DELALLOC,
1061 op);
1062 disk_num_bytes -= cur_alloc_size;
1063 num_bytes -= cur_alloc_size;
1064 alloc_hint = ins.objectid + ins.offset;
1065 start += cur_alloc_size;
1066 }
1067 out:
1068 return ret;
1069
1070 out_drop_extent_cache:
1071 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1072 out_reserve:
1073 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
1074 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1075 out_unlock:
1076 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1077 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1078 EXTENT_DELALLOC | EXTENT_DEFRAG,
1079 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1080 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1081 goto out;
1082 }
1083
1084 /*
1085 * work queue call back to started compression on a file and pages
1086 */
1087 static noinline void async_cow_start(struct btrfs_work *work)
1088 {
1089 struct async_cow *async_cow;
1090 int num_added = 0;
1091 async_cow = container_of(work, struct async_cow, work);
1092
1093 compress_file_range(async_cow->inode, async_cow->locked_page,
1094 async_cow->start, async_cow->end, async_cow,
1095 &num_added);
1096 if (num_added == 0) {
1097 btrfs_add_delayed_iput(async_cow->inode);
1098 async_cow->inode = NULL;
1099 }
1100 }
1101
1102 /*
1103 * work queue call back to submit previously compressed pages
1104 */
1105 static noinline void async_cow_submit(struct btrfs_work *work)
1106 {
1107 struct async_cow *async_cow;
1108 struct btrfs_root *root;
1109 unsigned long nr_pages;
1110
1111 async_cow = container_of(work, struct async_cow, work);
1112
1113 root = async_cow->root;
1114 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1115 PAGE_SHIFT;
1116
1117 /*
1118 * atomic_sub_return implies a barrier for waitqueue_active
1119 */
1120 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1121 5 * SZ_1M &&
1122 waitqueue_active(&root->fs_info->async_submit_wait))
1123 wake_up(&root->fs_info->async_submit_wait);
1124
1125 if (async_cow->inode)
1126 submit_compressed_extents(async_cow->inode, async_cow);
1127 }
1128
1129 static noinline void async_cow_free(struct btrfs_work *work)
1130 {
1131 struct async_cow *async_cow;
1132 async_cow = container_of(work, struct async_cow, work);
1133 if (async_cow->inode)
1134 btrfs_add_delayed_iput(async_cow->inode);
1135 kfree(async_cow);
1136 }
1137
1138 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1139 u64 start, u64 end, int *page_started,
1140 unsigned long *nr_written)
1141 {
1142 struct async_cow *async_cow;
1143 struct btrfs_root *root = BTRFS_I(inode)->root;
1144 unsigned long nr_pages;
1145 u64 cur_end;
1146 int limit = 10 * SZ_1M;
1147
1148 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1149 1, 0, NULL, GFP_NOFS);
1150 while (start < end) {
1151 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1152 BUG_ON(!async_cow); /* -ENOMEM */
1153 async_cow->inode = igrab(inode);
1154 async_cow->root = root;
1155 async_cow->locked_page = locked_page;
1156 async_cow->start = start;
1157
1158 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1159 !btrfs_test_opt(root, FORCE_COMPRESS))
1160 cur_end = end;
1161 else
1162 cur_end = min(end, start + SZ_512K - 1);
1163
1164 async_cow->end = cur_end;
1165 INIT_LIST_HEAD(&async_cow->extents);
1166
1167 btrfs_init_work(&async_cow->work,
1168 btrfs_delalloc_helper,
1169 async_cow_start, async_cow_submit,
1170 async_cow_free);
1171
1172 nr_pages = (cur_end - start + PAGE_SIZE) >>
1173 PAGE_SHIFT;
1174 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1175
1176 btrfs_queue_work(root->fs_info->delalloc_workers,
1177 &async_cow->work);
1178
1179 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1180 wait_event(root->fs_info->async_submit_wait,
1181 (atomic_read(&root->fs_info->async_delalloc_pages) <
1182 limit));
1183 }
1184
1185 while (atomic_read(&root->fs_info->async_submit_draining) &&
1186 atomic_read(&root->fs_info->async_delalloc_pages)) {
1187 wait_event(root->fs_info->async_submit_wait,
1188 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1189 0));
1190 }
1191
1192 *nr_written += nr_pages;
1193 start = cur_end + 1;
1194 }
1195 *page_started = 1;
1196 return 0;
1197 }
1198
1199 static noinline int csum_exist_in_range(struct btrfs_root *root,
1200 u64 bytenr, u64 num_bytes)
1201 {
1202 int ret;
1203 struct btrfs_ordered_sum *sums;
1204 LIST_HEAD(list);
1205
1206 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1207 bytenr + num_bytes - 1, &list, 0);
1208 if (ret == 0 && list_empty(&list))
1209 return 0;
1210
1211 while (!list_empty(&list)) {
1212 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1213 list_del(&sums->list);
1214 kfree(sums);
1215 }
1216 return 1;
1217 }
1218
1219 /*
1220 * when nowcow writeback call back. This checks for snapshots or COW copies
1221 * of the extents that exist in the file, and COWs the file as required.
1222 *
1223 * If no cow copies or snapshots exist, we write directly to the existing
1224 * blocks on disk
1225 */
1226 static noinline int run_delalloc_nocow(struct inode *inode,
1227 struct page *locked_page,
1228 u64 start, u64 end, int *page_started, int force,
1229 unsigned long *nr_written)
1230 {
1231 struct btrfs_root *root = BTRFS_I(inode)->root;
1232 struct btrfs_trans_handle *trans;
1233 struct extent_buffer *leaf;
1234 struct btrfs_path *path;
1235 struct btrfs_file_extent_item *fi;
1236 struct btrfs_key found_key;
1237 u64 cow_start;
1238 u64 cur_offset;
1239 u64 extent_end;
1240 u64 extent_offset;
1241 u64 disk_bytenr;
1242 u64 num_bytes;
1243 u64 disk_num_bytes;
1244 u64 ram_bytes;
1245 int extent_type;
1246 int ret, err;
1247 int type;
1248 int nocow;
1249 int check_prev = 1;
1250 bool nolock;
1251 u64 ino = btrfs_ino(inode);
1252
1253 path = btrfs_alloc_path();
1254 if (!path) {
1255 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1256 EXTENT_LOCKED | EXTENT_DELALLOC |
1257 EXTENT_DO_ACCOUNTING |
1258 EXTENT_DEFRAG, PAGE_UNLOCK |
1259 PAGE_CLEAR_DIRTY |
1260 PAGE_SET_WRITEBACK |
1261 PAGE_END_WRITEBACK);
1262 return -ENOMEM;
1263 }
1264
1265 nolock = btrfs_is_free_space_inode(inode);
1266
1267 if (nolock)
1268 trans = btrfs_join_transaction_nolock(root);
1269 else
1270 trans = btrfs_join_transaction(root);
1271
1272 if (IS_ERR(trans)) {
1273 extent_clear_unlock_delalloc(inode, start, end, 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 btrfs_free_path(path);
1281 return PTR_ERR(trans);
1282 }
1283
1284 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1285
1286 cow_start = (u64)-1;
1287 cur_offset = start;
1288 while (1) {
1289 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1290 cur_offset, 0);
1291 if (ret < 0)
1292 goto error;
1293 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1294 leaf = path->nodes[0];
1295 btrfs_item_key_to_cpu(leaf, &found_key,
1296 path->slots[0] - 1);
1297 if (found_key.objectid == ino &&
1298 found_key.type == BTRFS_EXTENT_DATA_KEY)
1299 path->slots[0]--;
1300 }
1301 check_prev = 0;
1302 next_slot:
1303 leaf = path->nodes[0];
1304 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1305 ret = btrfs_next_leaf(root, path);
1306 if (ret < 0)
1307 goto error;
1308 if (ret > 0)
1309 break;
1310 leaf = path->nodes[0];
1311 }
1312
1313 nocow = 0;
1314 disk_bytenr = 0;
1315 num_bytes = 0;
1316 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1317
1318 if (found_key.objectid > ino)
1319 break;
1320 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1321 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1322 path->slots[0]++;
1323 goto next_slot;
1324 }
1325 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1326 found_key.offset > end)
1327 break;
1328
1329 if (found_key.offset > cur_offset) {
1330 extent_end = found_key.offset;
1331 extent_type = 0;
1332 goto out_check;
1333 }
1334
1335 fi = btrfs_item_ptr(leaf, path->slots[0],
1336 struct btrfs_file_extent_item);
1337 extent_type = btrfs_file_extent_type(leaf, fi);
1338
1339 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1340 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1341 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1342 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1343 extent_offset = btrfs_file_extent_offset(leaf, fi);
1344 extent_end = found_key.offset +
1345 btrfs_file_extent_num_bytes(leaf, fi);
1346 disk_num_bytes =
1347 btrfs_file_extent_disk_num_bytes(leaf, fi);
1348 if (extent_end <= start) {
1349 path->slots[0]++;
1350 goto next_slot;
1351 }
1352 if (disk_bytenr == 0)
1353 goto out_check;
1354 if (btrfs_file_extent_compression(leaf, fi) ||
1355 btrfs_file_extent_encryption(leaf, fi) ||
1356 btrfs_file_extent_other_encoding(leaf, fi))
1357 goto out_check;
1358 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1359 goto out_check;
1360 if (btrfs_extent_readonly(root, disk_bytenr))
1361 goto out_check;
1362 if (btrfs_cross_ref_exist(trans, root, ino,
1363 found_key.offset -
1364 extent_offset, disk_bytenr))
1365 goto out_check;
1366 disk_bytenr += extent_offset;
1367 disk_bytenr += cur_offset - found_key.offset;
1368 num_bytes = min(end + 1, extent_end) - cur_offset;
1369 /*
1370 * if there are pending snapshots for this root,
1371 * we fall into common COW way.
1372 */
1373 if (!nolock) {
1374 err = btrfs_start_write_no_snapshoting(root);
1375 if (!err)
1376 goto out_check;
1377 }
1378 /*
1379 * force cow if csum exists in the range.
1380 * this ensure that csum for a given extent are
1381 * either valid or do not exist.
1382 */
1383 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1384 goto out_check;
1385 if (!btrfs_inc_nocow_writers(root->fs_info,
1386 disk_bytenr))
1387 goto out_check;
1388 nocow = 1;
1389 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1390 extent_end = found_key.offset +
1391 btrfs_file_extent_inline_len(leaf,
1392 path->slots[0], fi);
1393 extent_end = ALIGN(extent_end, root->sectorsize);
1394 } else {
1395 BUG_ON(1);
1396 }
1397 out_check:
1398 if (extent_end <= start) {
1399 path->slots[0]++;
1400 if (!nolock && nocow)
1401 btrfs_end_write_no_snapshoting(root);
1402 if (nocow)
1403 btrfs_dec_nocow_writers(root->fs_info,
1404 disk_bytenr);
1405 goto next_slot;
1406 }
1407 if (!nocow) {
1408 if (cow_start == (u64)-1)
1409 cow_start = cur_offset;
1410 cur_offset = extent_end;
1411 if (cur_offset > end)
1412 break;
1413 path->slots[0]++;
1414 goto next_slot;
1415 }
1416
1417 btrfs_release_path(path);
1418 if (cow_start != (u64)-1) {
1419 ret = cow_file_range(inode, locked_page,
1420 cow_start, found_key.offset - 1,
1421 page_started, nr_written, 1);
1422 if (ret) {
1423 if (!nolock && nocow)
1424 btrfs_end_write_no_snapshoting(root);
1425 if (nocow)
1426 btrfs_dec_nocow_writers(root->fs_info,
1427 disk_bytenr);
1428 goto error;
1429 }
1430 cow_start = (u64)-1;
1431 }
1432
1433 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1434 struct extent_map *em;
1435 struct extent_map_tree *em_tree;
1436 em_tree = &BTRFS_I(inode)->extent_tree;
1437 em = alloc_extent_map();
1438 BUG_ON(!em); /* -ENOMEM */
1439 em->start = cur_offset;
1440 em->orig_start = found_key.offset - extent_offset;
1441 em->len = num_bytes;
1442 em->block_len = num_bytes;
1443 em->block_start = disk_bytenr;
1444 em->orig_block_len = disk_num_bytes;
1445 em->ram_bytes = ram_bytes;
1446 em->bdev = root->fs_info->fs_devices->latest_bdev;
1447 em->mod_start = em->start;
1448 em->mod_len = em->len;
1449 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1450 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1451 em->generation = -1;
1452 while (1) {
1453 write_lock(&em_tree->lock);
1454 ret = add_extent_mapping(em_tree, em, 1);
1455 write_unlock(&em_tree->lock);
1456 if (ret != -EEXIST) {
1457 free_extent_map(em);
1458 break;
1459 }
1460 btrfs_drop_extent_cache(inode, em->start,
1461 em->start + em->len - 1, 0);
1462 }
1463 type = BTRFS_ORDERED_PREALLOC;
1464 } else {
1465 type = BTRFS_ORDERED_NOCOW;
1466 }
1467
1468 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1469 num_bytes, num_bytes, type);
1470 if (nocow)
1471 btrfs_dec_nocow_writers(root->fs_info, disk_bytenr);
1472 BUG_ON(ret); /* -ENOMEM */
1473
1474 if (root->root_key.objectid ==
1475 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1476 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1477 num_bytes);
1478 if (ret) {
1479 if (!nolock && nocow)
1480 btrfs_end_write_no_snapshoting(root);
1481 goto error;
1482 }
1483 }
1484
1485 extent_clear_unlock_delalloc(inode, cur_offset,
1486 cur_offset + num_bytes - 1,
1487 locked_page, EXTENT_LOCKED |
1488 EXTENT_DELALLOC, PAGE_UNLOCK |
1489 PAGE_SET_PRIVATE2);
1490 if (!nolock && nocow)
1491 btrfs_end_write_no_snapshoting(root);
1492 cur_offset = extent_end;
1493 if (cur_offset > end)
1494 break;
1495 }
1496 btrfs_release_path(path);
1497
1498 if (cur_offset <= end && cow_start == (u64)-1) {
1499 cow_start = cur_offset;
1500 cur_offset = end;
1501 }
1502
1503 if (cow_start != (u64)-1) {
1504 ret = cow_file_range(inode, locked_page, cow_start, end,
1505 page_started, nr_written, 1);
1506 if (ret)
1507 goto error;
1508 }
1509
1510 error:
1511 err = btrfs_end_transaction(trans, root);
1512 if (!ret)
1513 ret = err;
1514
1515 if (ret && cur_offset < end)
1516 extent_clear_unlock_delalloc(inode, cur_offset, end,
1517 locked_page, EXTENT_LOCKED |
1518 EXTENT_DELALLOC | EXTENT_DEFRAG |
1519 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1520 PAGE_CLEAR_DIRTY |
1521 PAGE_SET_WRITEBACK |
1522 PAGE_END_WRITEBACK);
1523 btrfs_free_path(path);
1524 return ret;
1525 }
1526
1527 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1528 {
1529
1530 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1531 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1532 return 0;
1533
1534 /*
1535 * @defrag_bytes is a hint value, no spinlock held here,
1536 * if is not zero, it means the file is defragging.
1537 * Force cow if given extent needs to be defragged.
1538 */
1539 if (BTRFS_I(inode)->defrag_bytes &&
1540 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1541 EXTENT_DEFRAG, 0, NULL))
1542 return 1;
1543
1544 return 0;
1545 }
1546
1547 /*
1548 * extent_io.c call back to do delayed allocation processing
1549 */
1550 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1551 u64 start, u64 end, int *page_started,
1552 unsigned long *nr_written)
1553 {
1554 int ret;
1555 int force_cow = need_force_cow(inode, start, end);
1556
1557 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1558 ret = run_delalloc_nocow(inode, locked_page, start, end,
1559 page_started, 1, nr_written);
1560 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1561 ret = run_delalloc_nocow(inode, locked_page, start, end,
1562 page_started, 0, nr_written);
1563 } else if (!inode_need_compress(inode)) {
1564 ret = cow_file_range(inode, locked_page, start, end,
1565 page_started, nr_written, 1);
1566 } else {
1567 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1568 &BTRFS_I(inode)->runtime_flags);
1569 ret = cow_file_range_async(inode, locked_page, start, end,
1570 page_started, nr_written);
1571 }
1572 return ret;
1573 }
1574
1575 static void btrfs_split_extent_hook(struct inode *inode,
1576 struct extent_state *orig, u64 split)
1577 {
1578 u64 size;
1579
1580 /* not delalloc, ignore it */
1581 if (!(orig->state & EXTENT_DELALLOC))
1582 return;
1583
1584 size = orig->end - orig->start + 1;
1585 if (size > BTRFS_MAX_EXTENT_SIZE) {
1586 u64 num_extents;
1587 u64 new_size;
1588
1589 /*
1590 * See the explanation in btrfs_merge_extent_hook, the same
1591 * applies here, just in reverse.
1592 */
1593 new_size = orig->end - split + 1;
1594 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1595 BTRFS_MAX_EXTENT_SIZE);
1596 new_size = split - orig->start;
1597 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1598 BTRFS_MAX_EXTENT_SIZE);
1599 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1600 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1601 return;
1602 }
1603
1604 spin_lock(&BTRFS_I(inode)->lock);
1605 BTRFS_I(inode)->outstanding_extents++;
1606 spin_unlock(&BTRFS_I(inode)->lock);
1607 }
1608
1609 /*
1610 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1611 * extents so we can keep track of new extents that are just merged onto old
1612 * extents, such as when we are doing sequential writes, so we can properly
1613 * account for the metadata space we'll need.
1614 */
1615 static void btrfs_merge_extent_hook(struct inode *inode,
1616 struct extent_state *new,
1617 struct extent_state *other)
1618 {
1619 u64 new_size, old_size;
1620 u64 num_extents;
1621
1622 /* not delalloc, ignore it */
1623 if (!(other->state & EXTENT_DELALLOC))
1624 return;
1625
1626 if (new->start > other->start)
1627 new_size = new->end - other->start + 1;
1628 else
1629 new_size = other->end - new->start + 1;
1630
1631 /* we're not bigger than the max, unreserve the space and go */
1632 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1633 spin_lock(&BTRFS_I(inode)->lock);
1634 BTRFS_I(inode)->outstanding_extents--;
1635 spin_unlock(&BTRFS_I(inode)->lock);
1636 return;
1637 }
1638
1639 /*
1640 * We have to add up either side to figure out how many extents were
1641 * accounted for before we merged into one big extent. If the number of
1642 * extents we accounted for is <= the amount we need for the new range
1643 * then we can return, otherwise drop. Think of it like this
1644 *
1645 * [ 4k][MAX_SIZE]
1646 *
1647 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1648 * need 2 outstanding extents, on one side we have 1 and the other side
1649 * we have 1 so they are == and we can return. But in this case
1650 *
1651 * [MAX_SIZE+4k][MAX_SIZE+4k]
1652 *
1653 * Each range on their own accounts for 2 extents, but merged together
1654 * they are only 3 extents worth of accounting, so we need to drop in
1655 * this case.
1656 */
1657 old_size = other->end - other->start + 1;
1658 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1659 BTRFS_MAX_EXTENT_SIZE);
1660 old_size = new->end - new->start + 1;
1661 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1662 BTRFS_MAX_EXTENT_SIZE);
1663
1664 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1665 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1666 return;
1667
1668 spin_lock(&BTRFS_I(inode)->lock);
1669 BTRFS_I(inode)->outstanding_extents--;
1670 spin_unlock(&BTRFS_I(inode)->lock);
1671 }
1672
1673 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1674 struct inode *inode)
1675 {
1676 spin_lock(&root->delalloc_lock);
1677 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1678 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1679 &root->delalloc_inodes);
1680 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1681 &BTRFS_I(inode)->runtime_flags);
1682 root->nr_delalloc_inodes++;
1683 if (root->nr_delalloc_inodes == 1) {
1684 spin_lock(&root->fs_info->delalloc_root_lock);
1685 BUG_ON(!list_empty(&root->delalloc_root));
1686 list_add_tail(&root->delalloc_root,
1687 &root->fs_info->delalloc_roots);
1688 spin_unlock(&root->fs_info->delalloc_root_lock);
1689 }
1690 }
1691 spin_unlock(&root->delalloc_lock);
1692 }
1693
1694 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1695 struct inode *inode)
1696 {
1697 spin_lock(&root->delalloc_lock);
1698 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1699 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1700 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1701 &BTRFS_I(inode)->runtime_flags);
1702 root->nr_delalloc_inodes--;
1703 if (!root->nr_delalloc_inodes) {
1704 spin_lock(&root->fs_info->delalloc_root_lock);
1705 BUG_ON(list_empty(&root->delalloc_root));
1706 list_del_init(&root->delalloc_root);
1707 spin_unlock(&root->fs_info->delalloc_root_lock);
1708 }
1709 }
1710 spin_unlock(&root->delalloc_lock);
1711 }
1712
1713 /*
1714 * extent_io.c set_bit_hook, used to track delayed allocation
1715 * bytes in this file, and to maintain the list of inodes that
1716 * have pending delalloc work to be done.
1717 */
1718 static void btrfs_set_bit_hook(struct inode *inode,
1719 struct extent_state *state, unsigned *bits)
1720 {
1721
1722 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1723 WARN_ON(1);
1724 /*
1725 * set_bit and clear bit hooks normally require _irqsave/restore
1726 * but in this case, we are only testing for the DELALLOC
1727 * bit, which is only set or cleared with irqs on
1728 */
1729 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1730 struct btrfs_root *root = BTRFS_I(inode)->root;
1731 u64 len = state->end + 1 - state->start;
1732 bool do_list = !btrfs_is_free_space_inode(inode);
1733
1734 if (*bits & EXTENT_FIRST_DELALLOC) {
1735 *bits &= ~EXTENT_FIRST_DELALLOC;
1736 } else {
1737 spin_lock(&BTRFS_I(inode)->lock);
1738 BTRFS_I(inode)->outstanding_extents++;
1739 spin_unlock(&BTRFS_I(inode)->lock);
1740 }
1741
1742 /* For sanity tests */
1743 if (btrfs_test_is_dummy_root(root))
1744 return;
1745
1746 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1747 root->fs_info->delalloc_batch);
1748 spin_lock(&BTRFS_I(inode)->lock);
1749 BTRFS_I(inode)->delalloc_bytes += len;
1750 if (*bits & EXTENT_DEFRAG)
1751 BTRFS_I(inode)->defrag_bytes += len;
1752 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1753 &BTRFS_I(inode)->runtime_flags))
1754 btrfs_add_delalloc_inodes(root, inode);
1755 spin_unlock(&BTRFS_I(inode)->lock);
1756 }
1757 }
1758
1759 /*
1760 * extent_io.c clear_bit_hook, see set_bit_hook for why
1761 */
1762 static void btrfs_clear_bit_hook(struct inode *inode,
1763 struct extent_state *state,
1764 unsigned *bits)
1765 {
1766 u64 len = state->end + 1 - state->start;
1767 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1768 BTRFS_MAX_EXTENT_SIZE);
1769
1770 spin_lock(&BTRFS_I(inode)->lock);
1771 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1772 BTRFS_I(inode)->defrag_bytes -= len;
1773 spin_unlock(&BTRFS_I(inode)->lock);
1774
1775 /*
1776 * set_bit and clear bit hooks normally require _irqsave/restore
1777 * but in this case, we are only testing for the DELALLOC
1778 * bit, which is only set or cleared with irqs on
1779 */
1780 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1781 struct btrfs_root *root = BTRFS_I(inode)->root;
1782 bool do_list = !btrfs_is_free_space_inode(inode);
1783
1784 if (*bits & EXTENT_FIRST_DELALLOC) {
1785 *bits &= ~EXTENT_FIRST_DELALLOC;
1786 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1787 spin_lock(&BTRFS_I(inode)->lock);
1788 BTRFS_I(inode)->outstanding_extents -= num_extents;
1789 spin_unlock(&BTRFS_I(inode)->lock);
1790 }
1791
1792 /*
1793 * We don't reserve metadata space for space cache inodes so we
1794 * don't need to call dellalloc_release_metadata if there is an
1795 * error.
1796 */
1797 if (*bits & EXTENT_DO_ACCOUNTING &&
1798 root != root->fs_info->tree_root)
1799 btrfs_delalloc_release_metadata(inode, len);
1800
1801 /* For sanity tests. */
1802 if (btrfs_test_is_dummy_root(root))
1803 return;
1804
1805 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1806 && do_list && !(state->state & EXTENT_NORESERVE))
1807 btrfs_free_reserved_data_space_noquota(inode,
1808 state->start, len);
1809
1810 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1811 root->fs_info->delalloc_batch);
1812 spin_lock(&BTRFS_I(inode)->lock);
1813 BTRFS_I(inode)->delalloc_bytes -= len;
1814 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1815 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1816 &BTRFS_I(inode)->runtime_flags))
1817 btrfs_del_delalloc_inode(root, inode);
1818 spin_unlock(&BTRFS_I(inode)->lock);
1819 }
1820 }
1821
1822 /*
1823 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1824 * we don't create bios that span stripes or chunks
1825 */
1826 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1827 size_t size, struct bio *bio,
1828 unsigned long bio_flags)
1829 {
1830 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1831 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1832 u64 length = 0;
1833 u64 map_length;
1834 int ret;
1835
1836 if (bio_flags & EXTENT_BIO_COMPRESSED)
1837 return 0;
1838
1839 length = bio->bi_iter.bi_size;
1840 map_length = length;
1841 ret = btrfs_map_block(root->fs_info, rw, logical,
1842 &map_length, NULL, 0);
1843 /* Will always return 0 with map_multi == NULL */
1844 BUG_ON(ret < 0);
1845 if (map_length < length + size)
1846 return 1;
1847 return 0;
1848 }
1849
1850 /*
1851 * in order to insert checksums into the metadata in large chunks,
1852 * we wait until bio submission time. All the pages in the bio are
1853 * checksummed and sums are attached onto the ordered extent record.
1854 *
1855 * At IO completion time the cums attached on the ordered extent record
1856 * are inserted into the btree
1857 */
1858 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1859 struct bio *bio, int mirror_num,
1860 unsigned long bio_flags,
1861 u64 bio_offset)
1862 {
1863 struct btrfs_root *root = BTRFS_I(inode)->root;
1864 int ret = 0;
1865
1866 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1867 BUG_ON(ret); /* -ENOMEM */
1868 return 0;
1869 }
1870
1871 /*
1872 * in order to insert checksums into the metadata in large chunks,
1873 * we wait until bio submission time. All the pages in the bio are
1874 * checksummed and sums are attached onto the ordered extent record.
1875 *
1876 * At IO completion time the cums attached on the ordered extent record
1877 * are inserted into the btree
1878 */
1879 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1880 int mirror_num, unsigned long bio_flags,
1881 u64 bio_offset)
1882 {
1883 struct btrfs_root *root = BTRFS_I(inode)->root;
1884 int ret;
1885
1886 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1887 if (ret) {
1888 bio->bi_error = ret;
1889 bio_endio(bio);
1890 }
1891 return ret;
1892 }
1893
1894 /*
1895 * extent_io.c submission hook. This does the right thing for csum calculation
1896 * on write, or reading the csums from the tree before a read
1897 */
1898 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1899 int mirror_num, unsigned long bio_flags,
1900 u64 bio_offset)
1901 {
1902 struct btrfs_root *root = BTRFS_I(inode)->root;
1903 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1904 int ret = 0;
1905 int skip_sum;
1906 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1907
1908 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1909
1910 if (btrfs_is_free_space_inode(inode))
1911 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1912
1913 if (!(rw & REQ_WRITE)) {
1914 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1915 if (ret)
1916 goto out;
1917
1918 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1919 ret = btrfs_submit_compressed_read(inode, bio,
1920 mirror_num,
1921 bio_flags);
1922 goto out;
1923 } else if (!skip_sum) {
1924 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1925 if (ret)
1926 goto out;
1927 }
1928 goto mapit;
1929 } else if (async && !skip_sum) {
1930 /* csum items have already been cloned */
1931 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1932 goto mapit;
1933 /* we're doing a write, do the async checksumming */
1934 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1935 inode, rw, bio, mirror_num,
1936 bio_flags, bio_offset,
1937 __btrfs_submit_bio_start,
1938 __btrfs_submit_bio_done);
1939 goto out;
1940 } else if (!skip_sum) {
1941 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1942 if (ret)
1943 goto out;
1944 }
1945
1946 mapit:
1947 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1948
1949 out:
1950 if (ret < 0) {
1951 bio->bi_error = ret;
1952 bio_endio(bio);
1953 }
1954 return ret;
1955 }
1956
1957 /*
1958 * given a list of ordered sums record them in the inode. This happens
1959 * at IO completion time based on sums calculated at bio submission time.
1960 */
1961 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1962 struct inode *inode, u64 file_offset,
1963 struct list_head *list)
1964 {
1965 struct btrfs_ordered_sum *sum;
1966
1967 list_for_each_entry(sum, list, list) {
1968 trans->adding_csums = 1;
1969 btrfs_csum_file_blocks(trans,
1970 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1971 trans->adding_csums = 0;
1972 }
1973 return 0;
1974 }
1975
1976 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1977 struct extent_state **cached_state)
1978 {
1979 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
1980 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1981 cached_state);
1982 }
1983
1984 /* see btrfs_writepage_start_hook for details on why this is required */
1985 struct btrfs_writepage_fixup {
1986 struct page *page;
1987 struct btrfs_work work;
1988 };
1989
1990 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1991 {
1992 struct btrfs_writepage_fixup *fixup;
1993 struct btrfs_ordered_extent *ordered;
1994 struct extent_state *cached_state = NULL;
1995 struct page *page;
1996 struct inode *inode;
1997 u64 page_start;
1998 u64 page_end;
1999 int ret;
2000
2001 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2002 page = fixup->page;
2003 again:
2004 lock_page(page);
2005 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2006 ClearPageChecked(page);
2007 goto out_page;
2008 }
2009
2010 inode = page->mapping->host;
2011 page_start = page_offset(page);
2012 page_end = page_offset(page) + PAGE_SIZE - 1;
2013
2014 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2015 &cached_state);
2016
2017 /* already ordered? We're done */
2018 if (PagePrivate2(page))
2019 goto out;
2020
2021 ordered = btrfs_lookup_ordered_range(inode, page_start,
2022 PAGE_SIZE);
2023 if (ordered) {
2024 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2025 page_end, &cached_state, GFP_NOFS);
2026 unlock_page(page);
2027 btrfs_start_ordered_extent(inode, ordered, 1);
2028 btrfs_put_ordered_extent(ordered);
2029 goto again;
2030 }
2031
2032 ret = btrfs_delalloc_reserve_space(inode, page_start,
2033 PAGE_SIZE);
2034 if (ret) {
2035 mapping_set_error(page->mapping, ret);
2036 end_extent_writepage(page, ret, page_start, page_end);
2037 ClearPageChecked(page);
2038 goto out;
2039 }
2040
2041 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2042 ClearPageChecked(page);
2043 set_page_dirty(page);
2044 out:
2045 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2046 &cached_state, GFP_NOFS);
2047 out_page:
2048 unlock_page(page);
2049 put_page(page);
2050 kfree(fixup);
2051 }
2052
2053 /*
2054 * There are a few paths in the higher layers of the kernel that directly
2055 * set the page dirty bit without asking the filesystem if it is a
2056 * good idea. This causes problems because we want to make sure COW
2057 * properly happens and the data=ordered rules are followed.
2058 *
2059 * In our case any range that doesn't have the ORDERED bit set
2060 * hasn't been properly setup for IO. We kick off an async process
2061 * to fix it up. The async helper will wait for ordered extents, set
2062 * the delalloc bit and make it safe to write the page.
2063 */
2064 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2065 {
2066 struct inode *inode = page->mapping->host;
2067 struct btrfs_writepage_fixup *fixup;
2068 struct btrfs_root *root = BTRFS_I(inode)->root;
2069
2070 /* this page is properly in the ordered list */
2071 if (TestClearPagePrivate2(page))
2072 return 0;
2073
2074 if (PageChecked(page))
2075 return -EAGAIN;
2076
2077 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2078 if (!fixup)
2079 return -EAGAIN;
2080
2081 SetPageChecked(page);
2082 get_page(page);
2083 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2084 btrfs_writepage_fixup_worker, NULL, NULL);
2085 fixup->page = page;
2086 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2087 return -EBUSY;
2088 }
2089
2090 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2091 struct inode *inode, u64 file_pos,
2092 u64 disk_bytenr, u64 disk_num_bytes,
2093 u64 num_bytes, u64 ram_bytes,
2094 u8 compression, u8 encryption,
2095 u16 other_encoding, int extent_type)
2096 {
2097 struct btrfs_root *root = BTRFS_I(inode)->root;
2098 struct btrfs_file_extent_item *fi;
2099 struct btrfs_path *path;
2100 struct extent_buffer *leaf;
2101 struct btrfs_key ins;
2102 int extent_inserted = 0;
2103 int ret;
2104
2105 path = btrfs_alloc_path();
2106 if (!path)
2107 return -ENOMEM;
2108
2109 /*
2110 * we may be replacing one extent in the tree with another.
2111 * The new extent is pinned in the extent map, and we don't want
2112 * to drop it from the cache until it is completely in the btree.
2113 *
2114 * So, tell btrfs_drop_extents to leave this extent in the cache.
2115 * the caller is expected to unpin it and allow it to be merged
2116 * with the others.
2117 */
2118 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2119 file_pos + num_bytes, NULL, 0,
2120 1, sizeof(*fi), &extent_inserted);
2121 if (ret)
2122 goto out;
2123
2124 if (!extent_inserted) {
2125 ins.objectid = btrfs_ino(inode);
2126 ins.offset = file_pos;
2127 ins.type = BTRFS_EXTENT_DATA_KEY;
2128
2129 path->leave_spinning = 1;
2130 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2131 sizeof(*fi));
2132 if (ret)
2133 goto out;
2134 }
2135 leaf = path->nodes[0];
2136 fi = btrfs_item_ptr(leaf, path->slots[0],
2137 struct btrfs_file_extent_item);
2138 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2139 btrfs_set_file_extent_type(leaf, fi, extent_type);
2140 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2141 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2142 btrfs_set_file_extent_offset(leaf, fi, 0);
2143 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2144 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2145 btrfs_set_file_extent_compression(leaf, fi, compression);
2146 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2147 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2148
2149 btrfs_mark_buffer_dirty(leaf);
2150 btrfs_release_path(path);
2151
2152 inode_add_bytes(inode, num_bytes);
2153
2154 ins.objectid = disk_bytenr;
2155 ins.offset = disk_num_bytes;
2156 ins.type = BTRFS_EXTENT_ITEM_KEY;
2157 ret = btrfs_alloc_reserved_file_extent(trans, root,
2158 root->root_key.objectid,
2159 btrfs_ino(inode), file_pos,
2160 ram_bytes, &ins);
2161 /*
2162 * Release the reserved range from inode dirty range map, as it is
2163 * already moved into delayed_ref_head
2164 */
2165 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2166 out:
2167 btrfs_free_path(path);
2168
2169 return ret;
2170 }
2171
2172 /* snapshot-aware defrag */
2173 struct sa_defrag_extent_backref {
2174 struct rb_node node;
2175 struct old_sa_defrag_extent *old;
2176 u64 root_id;
2177 u64 inum;
2178 u64 file_pos;
2179 u64 extent_offset;
2180 u64 num_bytes;
2181 u64 generation;
2182 };
2183
2184 struct old_sa_defrag_extent {
2185 struct list_head list;
2186 struct new_sa_defrag_extent *new;
2187
2188 u64 extent_offset;
2189 u64 bytenr;
2190 u64 offset;
2191 u64 len;
2192 int count;
2193 };
2194
2195 struct new_sa_defrag_extent {
2196 struct rb_root root;
2197 struct list_head head;
2198 struct btrfs_path *path;
2199 struct inode *inode;
2200 u64 file_pos;
2201 u64 len;
2202 u64 bytenr;
2203 u64 disk_len;
2204 u8 compress_type;
2205 };
2206
2207 static int backref_comp(struct sa_defrag_extent_backref *b1,
2208 struct sa_defrag_extent_backref *b2)
2209 {
2210 if (b1->root_id < b2->root_id)
2211 return -1;
2212 else if (b1->root_id > b2->root_id)
2213 return 1;
2214
2215 if (b1->inum < b2->inum)
2216 return -1;
2217 else if (b1->inum > b2->inum)
2218 return 1;
2219
2220 if (b1->file_pos < b2->file_pos)
2221 return -1;
2222 else if (b1->file_pos > b2->file_pos)
2223 return 1;
2224
2225 /*
2226 * [------------------------------] ===> (a range of space)
2227 * |<--->| |<---->| =============> (fs/file tree A)
2228 * |<---------------------------->| ===> (fs/file tree B)
2229 *
2230 * A range of space can refer to two file extents in one tree while
2231 * refer to only one file extent in another tree.
2232 *
2233 * So we may process a disk offset more than one time(two extents in A)
2234 * and locate at the same extent(one extent in B), then insert two same
2235 * backrefs(both refer to the extent in B).
2236 */
2237 return 0;
2238 }
2239
2240 static void backref_insert(struct rb_root *root,
2241 struct sa_defrag_extent_backref *backref)
2242 {
2243 struct rb_node **p = &root->rb_node;
2244 struct rb_node *parent = NULL;
2245 struct sa_defrag_extent_backref *entry;
2246 int ret;
2247
2248 while (*p) {
2249 parent = *p;
2250 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2251
2252 ret = backref_comp(backref, entry);
2253 if (ret < 0)
2254 p = &(*p)->rb_left;
2255 else
2256 p = &(*p)->rb_right;
2257 }
2258
2259 rb_link_node(&backref->node, parent, p);
2260 rb_insert_color(&backref->node, root);
2261 }
2262
2263 /*
2264 * Note the backref might has changed, and in this case we just return 0.
2265 */
2266 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2267 void *ctx)
2268 {
2269 struct btrfs_file_extent_item *extent;
2270 struct btrfs_fs_info *fs_info;
2271 struct old_sa_defrag_extent *old = ctx;
2272 struct new_sa_defrag_extent *new = old->new;
2273 struct btrfs_path *path = new->path;
2274 struct btrfs_key key;
2275 struct btrfs_root *root;
2276 struct sa_defrag_extent_backref *backref;
2277 struct extent_buffer *leaf;
2278 struct inode *inode = new->inode;
2279 int slot;
2280 int ret;
2281 u64 extent_offset;
2282 u64 num_bytes;
2283
2284 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2285 inum == btrfs_ino(inode))
2286 return 0;
2287
2288 key.objectid = root_id;
2289 key.type = BTRFS_ROOT_ITEM_KEY;
2290 key.offset = (u64)-1;
2291
2292 fs_info = BTRFS_I(inode)->root->fs_info;
2293 root = btrfs_read_fs_root_no_name(fs_info, &key);
2294 if (IS_ERR(root)) {
2295 if (PTR_ERR(root) == -ENOENT)
2296 return 0;
2297 WARN_ON(1);
2298 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2299 inum, offset, root_id);
2300 return PTR_ERR(root);
2301 }
2302
2303 key.objectid = inum;
2304 key.type = BTRFS_EXTENT_DATA_KEY;
2305 if (offset > (u64)-1 << 32)
2306 key.offset = 0;
2307 else
2308 key.offset = offset;
2309
2310 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2311 if (WARN_ON(ret < 0))
2312 return ret;
2313 ret = 0;
2314
2315 while (1) {
2316 cond_resched();
2317
2318 leaf = path->nodes[0];
2319 slot = path->slots[0];
2320
2321 if (slot >= btrfs_header_nritems(leaf)) {
2322 ret = btrfs_next_leaf(root, path);
2323 if (ret < 0) {
2324 goto out;
2325 } else if (ret > 0) {
2326 ret = 0;
2327 goto out;
2328 }
2329 continue;
2330 }
2331
2332 path->slots[0]++;
2333
2334 btrfs_item_key_to_cpu(leaf, &key, slot);
2335
2336 if (key.objectid > inum)
2337 goto out;
2338
2339 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2340 continue;
2341
2342 extent = btrfs_item_ptr(leaf, slot,
2343 struct btrfs_file_extent_item);
2344
2345 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2346 continue;
2347
2348 /*
2349 * 'offset' refers to the exact key.offset,
2350 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2351 * (key.offset - extent_offset).
2352 */
2353 if (key.offset != offset)
2354 continue;
2355
2356 extent_offset = btrfs_file_extent_offset(leaf, extent);
2357 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2358
2359 if (extent_offset >= old->extent_offset + old->offset +
2360 old->len || extent_offset + num_bytes <=
2361 old->extent_offset + old->offset)
2362 continue;
2363 break;
2364 }
2365
2366 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2367 if (!backref) {
2368 ret = -ENOENT;
2369 goto out;
2370 }
2371
2372 backref->root_id = root_id;
2373 backref->inum = inum;
2374 backref->file_pos = offset;
2375 backref->num_bytes = num_bytes;
2376 backref->extent_offset = extent_offset;
2377 backref->generation = btrfs_file_extent_generation(leaf, extent);
2378 backref->old = old;
2379 backref_insert(&new->root, backref);
2380 old->count++;
2381 out:
2382 btrfs_release_path(path);
2383 WARN_ON(ret);
2384 return ret;
2385 }
2386
2387 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2388 struct new_sa_defrag_extent *new)
2389 {
2390 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2391 struct old_sa_defrag_extent *old, *tmp;
2392 int ret;
2393
2394 new->path = path;
2395
2396 list_for_each_entry_safe(old, tmp, &new->head, list) {
2397 ret = iterate_inodes_from_logical(old->bytenr +
2398 old->extent_offset, fs_info,
2399 path, record_one_backref,
2400 old);
2401 if (ret < 0 && ret != -ENOENT)
2402 return false;
2403
2404 /* no backref to be processed for this extent */
2405 if (!old->count) {
2406 list_del(&old->list);
2407 kfree(old);
2408 }
2409 }
2410
2411 if (list_empty(&new->head))
2412 return false;
2413
2414 return true;
2415 }
2416
2417 static int relink_is_mergable(struct extent_buffer *leaf,
2418 struct btrfs_file_extent_item *fi,
2419 struct new_sa_defrag_extent *new)
2420 {
2421 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2422 return 0;
2423
2424 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2425 return 0;
2426
2427 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2428 return 0;
2429
2430 if (btrfs_file_extent_encryption(leaf, fi) ||
2431 btrfs_file_extent_other_encoding(leaf, fi))
2432 return 0;
2433
2434 return 1;
2435 }
2436
2437 /*
2438 * Note the backref might has changed, and in this case we just return 0.
2439 */
2440 static noinline int relink_extent_backref(struct btrfs_path *path,
2441 struct sa_defrag_extent_backref *prev,
2442 struct sa_defrag_extent_backref *backref)
2443 {
2444 struct btrfs_file_extent_item *extent;
2445 struct btrfs_file_extent_item *item;
2446 struct btrfs_ordered_extent *ordered;
2447 struct btrfs_trans_handle *trans;
2448 struct btrfs_fs_info *fs_info;
2449 struct btrfs_root *root;
2450 struct btrfs_key key;
2451 struct extent_buffer *leaf;
2452 struct old_sa_defrag_extent *old = backref->old;
2453 struct new_sa_defrag_extent *new = old->new;
2454 struct inode *src_inode = new->inode;
2455 struct inode *inode;
2456 struct extent_state *cached = NULL;
2457 int ret = 0;
2458 u64 start;
2459 u64 len;
2460 u64 lock_start;
2461 u64 lock_end;
2462 bool merge = false;
2463 int index;
2464
2465 if (prev && prev->root_id == backref->root_id &&
2466 prev->inum == backref->inum &&
2467 prev->file_pos + prev->num_bytes == backref->file_pos)
2468 merge = true;
2469
2470 /* step 1: get root */
2471 key.objectid = backref->root_id;
2472 key.type = BTRFS_ROOT_ITEM_KEY;
2473 key.offset = (u64)-1;
2474
2475 fs_info = BTRFS_I(src_inode)->root->fs_info;
2476 index = srcu_read_lock(&fs_info->subvol_srcu);
2477
2478 root = btrfs_read_fs_root_no_name(fs_info, &key);
2479 if (IS_ERR(root)) {
2480 srcu_read_unlock(&fs_info->subvol_srcu, index);
2481 if (PTR_ERR(root) == -ENOENT)
2482 return 0;
2483 return PTR_ERR(root);
2484 }
2485
2486 if (btrfs_root_readonly(root)) {
2487 srcu_read_unlock(&fs_info->subvol_srcu, index);
2488 return 0;
2489 }
2490
2491 /* step 2: get inode */
2492 key.objectid = backref->inum;
2493 key.type = BTRFS_INODE_ITEM_KEY;
2494 key.offset = 0;
2495
2496 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2497 if (IS_ERR(inode)) {
2498 srcu_read_unlock(&fs_info->subvol_srcu, index);
2499 return 0;
2500 }
2501
2502 srcu_read_unlock(&fs_info->subvol_srcu, index);
2503
2504 /* step 3: relink backref */
2505 lock_start = backref->file_pos;
2506 lock_end = backref->file_pos + backref->num_bytes - 1;
2507 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2508 &cached);
2509
2510 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2511 if (ordered) {
2512 btrfs_put_ordered_extent(ordered);
2513 goto out_unlock;
2514 }
2515
2516 trans = btrfs_join_transaction(root);
2517 if (IS_ERR(trans)) {
2518 ret = PTR_ERR(trans);
2519 goto out_unlock;
2520 }
2521
2522 key.objectid = backref->inum;
2523 key.type = BTRFS_EXTENT_DATA_KEY;
2524 key.offset = backref->file_pos;
2525
2526 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2527 if (ret < 0) {
2528 goto out_free_path;
2529 } else if (ret > 0) {
2530 ret = 0;
2531 goto out_free_path;
2532 }
2533
2534 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2535 struct btrfs_file_extent_item);
2536
2537 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2538 backref->generation)
2539 goto out_free_path;
2540
2541 btrfs_release_path(path);
2542
2543 start = backref->file_pos;
2544 if (backref->extent_offset < old->extent_offset + old->offset)
2545 start += old->extent_offset + old->offset -
2546 backref->extent_offset;
2547
2548 len = min(backref->extent_offset + backref->num_bytes,
2549 old->extent_offset + old->offset + old->len);
2550 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2551
2552 ret = btrfs_drop_extents(trans, root, inode, start,
2553 start + len, 1);
2554 if (ret)
2555 goto out_free_path;
2556 again:
2557 key.objectid = btrfs_ino(inode);
2558 key.type = BTRFS_EXTENT_DATA_KEY;
2559 key.offset = start;
2560
2561 path->leave_spinning = 1;
2562 if (merge) {
2563 struct btrfs_file_extent_item *fi;
2564 u64 extent_len;
2565 struct btrfs_key found_key;
2566
2567 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2568 if (ret < 0)
2569 goto out_free_path;
2570
2571 path->slots[0]--;
2572 leaf = path->nodes[0];
2573 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2574
2575 fi = btrfs_item_ptr(leaf, path->slots[0],
2576 struct btrfs_file_extent_item);
2577 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2578
2579 if (extent_len + found_key.offset == start &&
2580 relink_is_mergable(leaf, fi, new)) {
2581 btrfs_set_file_extent_num_bytes(leaf, fi,
2582 extent_len + len);
2583 btrfs_mark_buffer_dirty(leaf);
2584 inode_add_bytes(inode, len);
2585
2586 ret = 1;
2587 goto out_free_path;
2588 } else {
2589 merge = false;
2590 btrfs_release_path(path);
2591 goto again;
2592 }
2593 }
2594
2595 ret = btrfs_insert_empty_item(trans, root, path, &key,
2596 sizeof(*extent));
2597 if (ret) {
2598 btrfs_abort_transaction(trans, root, ret);
2599 goto out_free_path;
2600 }
2601
2602 leaf = path->nodes[0];
2603 item = btrfs_item_ptr(leaf, path->slots[0],
2604 struct btrfs_file_extent_item);
2605 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2606 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2607 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2608 btrfs_set_file_extent_num_bytes(leaf, item, len);
2609 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2610 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2611 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2612 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2613 btrfs_set_file_extent_encryption(leaf, item, 0);
2614 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2615
2616 btrfs_mark_buffer_dirty(leaf);
2617 inode_add_bytes(inode, len);
2618 btrfs_release_path(path);
2619
2620 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2621 new->disk_len, 0,
2622 backref->root_id, backref->inum,
2623 new->file_pos); /* start - extent_offset */
2624 if (ret) {
2625 btrfs_abort_transaction(trans, root, ret);
2626 goto out_free_path;
2627 }
2628
2629 ret = 1;
2630 out_free_path:
2631 btrfs_release_path(path);
2632 path->leave_spinning = 0;
2633 btrfs_end_transaction(trans, root);
2634 out_unlock:
2635 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2636 &cached, GFP_NOFS);
2637 iput(inode);
2638 return ret;
2639 }
2640
2641 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2642 {
2643 struct old_sa_defrag_extent *old, *tmp;
2644
2645 if (!new)
2646 return;
2647
2648 list_for_each_entry_safe(old, tmp, &new->head, list) {
2649 kfree(old);
2650 }
2651 kfree(new);
2652 }
2653
2654 static void relink_file_extents(struct new_sa_defrag_extent *new)
2655 {
2656 struct btrfs_path *path;
2657 struct sa_defrag_extent_backref *backref;
2658 struct sa_defrag_extent_backref *prev = NULL;
2659 struct inode *inode;
2660 struct btrfs_root *root;
2661 struct rb_node *node;
2662 int ret;
2663
2664 inode = new->inode;
2665 root = BTRFS_I(inode)->root;
2666
2667 path = btrfs_alloc_path();
2668 if (!path)
2669 return;
2670
2671 if (!record_extent_backrefs(path, new)) {
2672 btrfs_free_path(path);
2673 goto out;
2674 }
2675 btrfs_release_path(path);
2676
2677 while (1) {
2678 node = rb_first(&new->root);
2679 if (!node)
2680 break;
2681 rb_erase(node, &new->root);
2682
2683 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2684
2685 ret = relink_extent_backref(path, prev, backref);
2686 WARN_ON(ret < 0);
2687
2688 kfree(prev);
2689
2690 if (ret == 1)
2691 prev = backref;
2692 else
2693 prev = NULL;
2694 cond_resched();
2695 }
2696 kfree(prev);
2697
2698 btrfs_free_path(path);
2699 out:
2700 free_sa_defrag_extent(new);
2701
2702 atomic_dec(&root->fs_info->defrag_running);
2703 wake_up(&root->fs_info->transaction_wait);
2704 }
2705
2706 static struct new_sa_defrag_extent *
2707 record_old_file_extents(struct inode *inode,
2708 struct btrfs_ordered_extent *ordered)
2709 {
2710 struct btrfs_root *root = BTRFS_I(inode)->root;
2711 struct btrfs_path *path;
2712 struct btrfs_key key;
2713 struct old_sa_defrag_extent *old;
2714 struct new_sa_defrag_extent *new;
2715 int ret;
2716
2717 new = kmalloc(sizeof(*new), GFP_NOFS);
2718 if (!new)
2719 return NULL;
2720
2721 new->inode = inode;
2722 new->file_pos = ordered->file_offset;
2723 new->len = ordered->len;
2724 new->bytenr = ordered->start;
2725 new->disk_len = ordered->disk_len;
2726 new->compress_type = ordered->compress_type;
2727 new->root = RB_ROOT;
2728 INIT_LIST_HEAD(&new->head);
2729
2730 path = btrfs_alloc_path();
2731 if (!path)
2732 goto out_kfree;
2733
2734 key.objectid = btrfs_ino(inode);
2735 key.type = BTRFS_EXTENT_DATA_KEY;
2736 key.offset = new->file_pos;
2737
2738 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2739 if (ret < 0)
2740 goto out_free_path;
2741 if (ret > 0 && path->slots[0] > 0)
2742 path->slots[0]--;
2743
2744 /* find out all the old extents for the file range */
2745 while (1) {
2746 struct btrfs_file_extent_item *extent;
2747 struct extent_buffer *l;
2748 int slot;
2749 u64 num_bytes;
2750 u64 offset;
2751 u64 end;
2752 u64 disk_bytenr;
2753 u64 extent_offset;
2754
2755 l = path->nodes[0];
2756 slot = path->slots[0];
2757
2758 if (slot >= btrfs_header_nritems(l)) {
2759 ret = btrfs_next_leaf(root, path);
2760 if (ret < 0)
2761 goto out_free_path;
2762 else if (ret > 0)
2763 break;
2764 continue;
2765 }
2766
2767 btrfs_item_key_to_cpu(l, &key, slot);
2768
2769 if (key.objectid != btrfs_ino(inode))
2770 break;
2771 if (key.type != BTRFS_EXTENT_DATA_KEY)
2772 break;
2773 if (key.offset >= new->file_pos + new->len)
2774 break;
2775
2776 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2777
2778 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2779 if (key.offset + num_bytes < new->file_pos)
2780 goto next;
2781
2782 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2783 if (!disk_bytenr)
2784 goto next;
2785
2786 extent_offset = btrfs_file_extent_offset(l, extent);
2787
2788 old = kmalloc(sizeof(*old), GFP_NOFS);
2789 if (!old)
2790 goto out_free_path;
2791
2792 offset = max(new->file_pos, key.offset);
2793 end = min(new->file_pos + new->len, key.offset + num_bytes);
2794
2795 old->bytenr = disk_bytenr;
2796 old->extent_offset = extent_offset;
2797 old->offset = offset - key.offset;
2798 old->len = end - offset;
2799 old->new = new;
2800 old->count = 0;
2801 list_add_tail(&old->list, &new->head);
2802 next:
2803 path->slots[0]++;
2804 cond_resched();
2805 }
2806
2807 btrfs_free_path(path);
2808 atomic_inc(&root->fs_info->defrag_running);
2809
2810 return new;
2811
2812 out_free_path:
2813 btrfs_free_path(path);
2814 out_kfree:
2815 free_sa_defrag_extent(new);
2816 return NULL;
2817 }
2818
2819 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2820 u64 start, u64 len)
2821 {
2822 struct btrfs_block_group_cache *cache;
2823
2824 cache = btrfs_lookup_block_group(root->fs_info, start);
2825 ASSERT(cache);
2826
2827 spin_lock(&cache->lock);
2828 cache->delalloc_bytes -= len;
2829 spin_unlock(&cache->lock);
2830
2831 btrfs_put_block_group(cache);
2832 }
2833
2834 /* as ordered data IO finishes, this gets called so we can finish
2835 * an ordered extent if the range of bytes in the file it covers are
2836 * fully written.
2837 */
2838 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2839 {
2840 struct inode *inode = ordered_extent->inode;
2841 struct btrfs_root *root = BTRFS_I(inode)->root;
2842 struct btrfs_trans_handle *trans = NULL;
2843 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2844 struct extent_state *cached_state = NULL;
2845 struct new_sa_defrag_extent *new = NULL;
2846 int compress_type = 0;
2847 int ret = 0;
2848 u64 logical_len = ordered_extent->len;
2849 bool nolock;
2850 bool truncated = false;
2851
2852 nolock = btrfs_is_free_space_inode(inode);
2853
2854 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2855 ret = -EIO;
2856 goto out;
2857 }
2858
2859 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2860 ordered_extent->file_offset +
2861 ordered_extent->len - 1);
2862
2863 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2864 truncated = true;
2865 logical_len = ordered_extent->truncated_len;
2866 /* Truncated the entire extent, don't bother adding */
2867 if (!logical_len)
2868 goto out;
2869 }
2870
2871 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2872 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2873
2874 /*
2875 * For mwrite(mmap + memset to write) case, we still reserve
2876 * space for NOCOW range.
2877 * As NOCOW won't cause a new delayed ref, just free the space
2878 */
2879 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2880 ordered_extent->len);
2881 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2882 if (nolock)
2883 trans = btrfs_join_transaction_nolock(root);
2884 else
2885 trans = btrfs_join_transaction(root);
2886 if (IS_ERR(trans)) {
2887 ret = PTR_ERR(trans);
2888 trans = NULL;
2889 goto out;
2890 }
2891 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2892 ret = btrfs_update_inode_fallback(trans, root, inode);
2893 if (ret) /* -ENOMEM or corruption */
2894 btrfs_abort_transaction(trans, root, ret);
2895 goto out;
2896 }
2897
2898 lock_extent_bits(io_tree, ordered_extent->file_offset,
2899 ordered_extent->file_offset + ordered_extent->len - 1,
2900 &cached_state);
2901
2902 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2903 ordered_extent->file_offset + ordered_extent->len - 1,
2904 EXTENT_DEFRAG, 1, cached_state);
2905 if (ret) {
2906 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2907 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2908 /* the inode is shared */
2909 new = record_old_file_extents(inode, ordered_extent);
2910
2911 clear_extent_bit(io_tree, ordered_extent->file_offset,
2912 ordered_extent->file_offset + ordered_extent->len - 1,
2913 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2914 }
2915
2916 if (nolock)
2917 trans = btrfs_join_transaction_nolock(root);
2918 else
2919 trans = btrfs_join_transaction(root);
2920 if (IS_ERR(trans)) {
2921 ret = PTR_ERR(trans);
2922 trans = NULL;
2923 goto out_unlock;
2924 }
2925
2926 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2927
2928 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2929 compress_type = ordered_extent->compress_type;
2930 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2931 BUG_ON(compress_type);
2932 ret = btrfs_mark_extent_written(trans, inode,
2933 ordered_extent->file_offset,
2934 ordered_extent->file_offset +
2935 logical_len);
2936 } else {
2937 BUG_ON(root == root->fs_info->tree_root);
2938 ret = insert_reserved_file_extent(trans, inode,
2939 ordered_extent->file_offset,
2940 ordered_extent->start,
2941 ordered_extent->disk_len,
2942 logical_len, logical_len,
2943 compress_type, 0, 0,
2944 BTRFS_FILE_EXTENT_REG);
2945 if (!ret)
2946 btrfs_release_delalloc_bytes(root,
2947 ordered_extent->start,
2948 ordered_extent->disk_len);
2949 }
2950 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2951 ordered_extent->file_offset, ordered_extent->len,
2952 trans->transid);
2953 if (ret < 0) {
2954 btrfs_abort_transaction(trans, root, ret);
2955 goto out_unlock;
2956 }
2957
2958 add_pending_csums(trans, inode, ordered_extent->file_offset,
2959 &ordered_extent->list);
2960
2961 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2962 ret = btrfs_update_inode_fallback(trans, root, inode);
2963 if (ret) { /* -ENOMEM or corruption */
2964 btrfs_abort_transaction(trans, root, ret);
2965 goto out_unlock;
2966 }
2967 ret = 0;
2968 out_unlock:
2969 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2970 ordered_extent->file_offset +
2971 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2972 out:
2973 if (root != root->fs_info->tree_root)
2974 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2975 if (trans)
2976 btrfs_end_transaction(trans, root);
2977
2978 if (ret || truncated) {
2979 u64 start, end;
2980
2981 if (truncated)
2982 start = ordered_extent->file_offset + logical_len;
2983 else
2984 start = ordered_extent->file_offset;
2985 end = ordered_extent->file_offset + ordered_extent->len - 1;
2986 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2987
2988 /* Drop the cache for the part of the extent we didn't write. */
2989 btrfs_drop_extent_cache(inode, start, end, 0);
2990
2991 /*
2992 * If the ordered extent had an IOERR or something else went
2993 * wrong we need to return the space for this ordered extent
2994 * back to the allocator. We only free the extent in the
2995 * truncated case if we didn't write out the extent at all.
2996 */
2997 if ((ret || !logical_len) &&
2998 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2999 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3000 btrfs_free_reserved_extent(root, ordered_extent->start,
3001 ordered_extent->disk_len, 1);
3002 }
3003
3004
3005 /*
3006 * This needs to be done to make sure anybody waiting knows we are done
3007 * updating everything for this ordered extent.
3008 */
3009 btrfs_remove_ordered_extent(inode, ordered_extent);
3010
3011 /* for snapshot-aware defrag */
3012 if (new) {
3013 if (ret) {
3014 free_sa_defrag_extent(new);
3015 atomic_dec(&root->fs_info->defrag_running);
3016 } else {
3017 relink_file_extents(new);
3018 }
3019 }
3020
3021 /* once for us */
3022 btrfs_put_ordered_extent(ordered_extent);
3023 /* once for the tree */
3024 btrfs_put_ordered_extent(ordered_extent);
3025
3026 return ret;
3027 }
3028
3029 static void finish_ordered_fn(struct btrfs_work *work)
3030 {
3031 struct btrfs_ordered_extent *ordered_extent;
3032 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3033 btrfs_finish_ordered_io(ordered_extent);
3034 }
3035
3036 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3037 struct extent_state *state, int uptodate)
3038 {
3039 struct inode *inode = page->mapping->host;
3040 struct btrfs_root *root = BTRFS_I(inode)->root;
3041 struct btrfs_ordered_extent *ordered_extent = NULL;
3042 struct btrfs_workqueue *wq;
3043 btrfs_work_func_t func;
3044
3045 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3046
3047 ClearPagePrivate2(page);
3048 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3049 end - start + 1, uptodate))
3050 return 0;
3051
3052 if (btrfs_is_free_space_inode(inode)) {
3053 wq = root->fs_info->endio_freespace_worker;
3054 func = btrfs_freespace_write_helper;
3055 } else {
3056 wq = root->fs_info->endio_write_workers;
3057 func = btrfs_endio_write_helper;
3058 }
3059
3060 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3061 NULL);
3062 btrfs_queue_work(wq, &ordered_extent->work);
3063
3064 return 0;
3065 }
3066
3067 static int __readpage_endio_check(struct inode *inode,
3068 struct btrfs_io_bio *io_bio,
3069 int icsum, struct page *page,
3070 int pgoff, u64 start, size_t len)
3071 {
3072 char *kaddr;
3073 u32 csum_expected;
3074 u32 csum = ~(u32)0;
3075
3076 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3077
3078 kaddr = kmap_atomic(page);
3079 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3080 btrfs_csum_final(csum, (char *)&csum);
3081 if (csum != csum_expected)
3082 goto zeroit;
3083
3084 kunmap_atomic(kaddr);
3085 return 0;
3086 zeroit:
3087 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3088 "csum failed ino %llu off %llu csum %u expected csum %u",
3089 btrfs_ino(inode), start, csum, csum_expected);
3090 memset(kaddr + pgoff, 1, len);
3091 flush_dcache_page(page);
3092 kunmap_atomic(kaddr);
3093 if (csum_expected == 0)
3094 return 0;
3095 return -EIO;
3096 }
3097
3098 /*
3099 * when reads are done, we need to check csums to verify the data is correct
3100 * if there's a match, we allow the bio to finish. If not, the code in
3101 * extent_io.c will try to find good copies for us.
3102 */
3103 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3104 u64 phy_offset, struct page *page,
3105 u64 start, u64 end, int mirror)
3106 {
3107 size_t offset = start - page_offset(page);
3108 struct inode *inode = page->mapping->host;
3109 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3110 struct btrfs_root *root = BTRFS_I(inode)->root;
3111
3112 if (PageChecked(page)) {
3113 ClearPageChecked(page);
3114 return 0;
3115 }
3116
3117 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3118 return 0;
3119
3120 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3121 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3122 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3123 return 0;
3124 }
3125
3126 phy_offset >>= inode->i_sb->s_blocksize_bits;
3127 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3128 start, (size_t)(end - start + 1));
3129 }
3130
3131 void btrfs_add_delayed_iput(struct inode *inode)
3132 {
3133 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3134 struct btrfs_inode *binode = BTRFS_I(inode);
3135
3136 if (atomic_add_unless(&inode->i_count, -1, 1))
3137 return;
3138
3139 spin_lock(&fs_info->delayed_iput_lock);
3140 if (binode->delayed_iput_count == 0) {
3141 ASSERT(list_empty(&binode->delayed_iput));
3142 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3143 } else {
3144 binode->delayed_iput_count++;
3145 }
3146 spin_unlock(&fs_info->delayed_iput_lock);
3147 }
3148
3149 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3150 {
3151 struct btrfs_fs_info *fs_info = root->fs_info;
3152
3153 spin_lock(&fs_info->delayed_iput_lock);
3154 while (!list_empty(&fs_info->delayed_iputs)) {
3155 struct btrfs_inode *inode;
3156
3157 inode = list_first_entry(&fs_info->delayed_iputs,
3158 struct btrfs_inode, delayed_iput);
3159 if (inode->delayed_iput_count) {
3160 inode->delayed_iput_count--;
3161 list_move_tail(&inode->delayed_iput,
3162 &fs_info->delayed_iputs);
3163 } else {
3164 list_del_init(&inode->delayed_iput);
3165 }
3166 spin_unlock(&fs_info->delayed_iput_lock);
3167 iput(&inode->vfs_inode);
3168 spin_lock(&fs_info->delayed_iput_lock);
3169 }
3170 spin_unlock(&fs_info->delayed_iput_lock);
3171 }
3172
3173 /*
3174 * This is called in transaction commit time. If there are no orphan
3175 * files in the subvolume, it removes orphan item and frees block_rsv
3176 * structure.
3177 */
3178 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3179 struct btrfs_root *root)
3180 {
3181 struct btrfs_block_rsv *block_rsv;
3182 int ret;
3183
3184 if (atomic_read(&root->orphan_inodes) ||
3185 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3186 return;
3187
3188 spin_lock(&root->orphan_lock);
3189 if (atomic_read(&root->orphan_inodes)) {
3190 spin_unlock(&root->orphan_lock);
3191 return;
3192 }
3193
3194 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3195 spin_unlock(&root->orphan_lock);
3196 return;
3197 }
3198
3199 block_rsv = root->orphan_block_rsv;
3200 root->orphan_block_rsv = NULL;
3201 spin_unlock(&root->orphan_lock);
3202
3203 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3204 btrfs_root_refs(&root->root_item) > 0) {
3205 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3206 root->root_key.objectid);
3207 if (ret)
3208 btrfs_abort_transaction(trans, root, ret);
3209 else
3210 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3211 &root->state);
3212 }
3213
3214 if (block_rsv) {
3215 WARN_ON(block_rsv->size > 0);
3216 btrfs_free_block_rsv(root, block_rsv);
3217 }
3218 }
3219
3220 /*
3221 * This creates an orphan entry for the given inode in case something goes
3222 * wrong in the middle of an unlink/truncate.
3223 *
3224 * NOTE: caller of this function should reserve 5 units of metadata for
3225 * this function.
3226 */
3227 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3228 {
3229 struct btrfs_root *root = BTRFS_I(inode)->root;
3230 struct btrfs_block_rsv *block_rsv = NULL;
3231 int reserve = 0;
3232 int insert = 0;
3233 int ret;
3234
3235 if (!root->orphan_block_rsv) {
3236 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3237 if (!block_rsv)
3238 return -ENOMEM;
3239 }
3240
3241 spin_lock(&root->orphan_lock);
3242 if (!root->orphan_block_rsv) {
3243 root->orphan_block_rsv = block_rsv;
3244 } else if (block_rsv) {
3245 btrfs_free_block_rsv(root, block_rsv);
3246 block_rsv = NULL;
3247 }
3248
3249 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3250 &BTRFS_I(inode)->runtime_flags)) {
3251 #if 0
3252 /*
3253 * For proper ENOSPC handling, we should do orphan
3254 * cleanup when mounting. But this introduces backward
3255 * compatibility issue.
3256 */
3257 if (!xchg(&root->orphan_item_inserted, 1))
3258 insert = 2;
3259 else
3260 insert = 1;
3261 #endif
3262 insert = 1;
3263 atomic_inc(&root->orphan_inodes);
3264 }
3265
3266 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3267 &BTRFS_I(inode)->runtime_flags))
3268 reserve = 1;
3269 spin_unlock(&root->orphan_lock);
3270
3271 /* grab metadata reservation from transaction handle */
3272 if (reserve) {
3273 ret = btrfs_orphan_reserve_metadata(trans, inode);
3274 ASSERT(!ret);
3275 if (ret) {
3276 atomic_dec(&root->orphan_inodes);
3277 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3278 &BTRFS_I(inode)->runtime_flags);
3279 if (insert)
3280 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3281 &BTRFS_I(inode)->runtime_flags);
3282 return ret;
3283 }
3284 }
3285
3286 /* insert an orphan item to track this unlinked/truncated file */
3287 if (insert >= 1) {
3288 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3289 if (ret) {
3290 atomic_dec(&root->orphan_inodes);
3291 if (reserve) {
3292 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3293 &BTRFS_I(inode)->runtime_flags);
3294 btrfs_orphan_release_metadata(inode);
3295 }
3296 if (ret != -EEXIST) {
3297 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3298 &BTRFS_I(inode)->runtime_flags);
3299 btrfs_abort_transaction(trans, root, ret);
3300 return ret;
3301 }
3302 }
3303 ret = 0;
3304 }
3305
3306 /* insert an orphan item to track subvolume contains orphan files */
3307 if (insert >= 2) {
3308 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3309 root->root_key.objectid);
3310 if (ret && ret != -EEXIST) {
3311 btrfs_abort_transaction(trans, root, ret);
3312 return ret;
3313 }
3314 }
3315 return 0;
3316 }
3317
3318 /*
3319 * We have done the truncate/delete so we can go ahead and remove the orphan
3320 * item for this particular inode.
3321 */
3322 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3323 struct inode *inode)
3324 {
3325 struct btrfs_root *root = BTRFS_I(inode)->root;
3326 int delete_item = 0;
3327 int release_rsv = 0;
3328 int ret = 0;
3329
3330 spin_lock(&root->orphan_lock);
3331 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3332 &BTRFS_I(inode)->runtime_flags))
3333 delete_item = 1;
3334
3335 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3336 &BTRFS_I(inode)->runtime_flags))
3337 release_rsv = 1;
3338 spin_unlock(&root->orphan_lock);
3339
3340 if (delete_item) {
3341 atomic_dec(&root->orphan_inodes);
3342 if (trans)
3343 ret = btrfs_del_orphan_item(trans, root,
3344 btrfs_ino(inode));
3345 }
3346
3347 if (release_rsv)
3348 btrfs_orphan_release_metadata(inode);
3349
3350 return ret;
3351 }
3352
3353 /*
3354 * this cleans up any orphans that may be left on the list from the last use
3355 * of this root.
3356 */
3357 int btrfs_orphan_cleanup(struct btrfs_root *root)
3358 {
3359 struct btrfs_path *path;
3360 struct extent_buffer *leaf;
3361 struct btrfs_key key, found_key;
3362 struct btrfs_trans_handle *trans;
3363 struct inode *inode;
3364 u64 last_objectid = 0;
3365 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3366
3367 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3368 return 0;
3369
3370 path = btrfs_alloc_path();
3371 if (!path) {
3372 ret = -ENOMEM;
3373 goto out;
3374 }
3375 path->reada = READA_BACK;
3376
3377 key.objectid = BTRFS_ORPHAN_OBJECTID;
3378 key.type = BTRFS_ORPHAN_ITEM_KEY;
3379 key.offset = (u64)-1;
3380
3381 while (1) {
3382 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3383 if (ret < 0)
3384 goto out;
3385
3386 /*
3387 * if ret == 0 means we found what we were searching for, which
3388 * is weird, but possible, so only screw with path if we didn't
3389 * find the key and see if we have stuff that matches
3390 */
3391 if (ret > 0) {
3392 ret = 0;
3393 if (path->slots[0] == 0)
3394 break;
3395 path->slots[0]--;
3396 }
3397
3398 /* pull out the item */
3399 leaf = path->nodes[0];
3400 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3401
3402 /* make sure the item matches what we want */
3403 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3404 break;
3405 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3406 break;
3407
3408 /* release the path since we're done with it */
3409 btrfs_release_path(path);
3410
3411 /*
3412 * this is where we are basically btrfs_lookup, without the
3413 * crossing root thing. we store the inode number in the
3414 * offset of the orphan item.
3415 */
3416
3417 if (found_key.offset == last_objectid) {
3418 btrfs_err(root->fs_info,
3419 "Error removing orphan entry, stopping orphan cleanup");
3420 ret = -EINVAL;
3421 goto out;
3422 }
3423
3424 last_objectid = found_key.offset;
3425
3426 found_key.objectid = found_key.offset;
3427 found_key.type = BTRFS_INODE_ITEM_KEY;
3428 found_key.offset = 0;
3429 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3430 ret = PTR_ERR_OR_ZERO(inode);
3431 if (ret && ret != -ESTALE)
3432 goto out;
3433
3434 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3435 struct btrfs_root *dead_root;
3436 struct btrfs_fs_info *fs_info = root->fs_info;
3437 int is_dead_root = 0;
3438
3439 /*
3440 * this is an orphan in the tree root. Currently these
3441 * could come from 2 sources:
3442 * a) a snapshot deletion in progress
3443 * b) a free space cache inode
3444 * We need to distinguish those two, as the snapshot
3445 * orphan must not get deleted.
3446 * find_dead_roots already ran before us, so if this
3447 * is a snapshot deletion, we should find the root
3448 * in the dead_roots list
3449 */
3450 spin_lock(&fs_info->trans_lock);
3451 list_for_each_entry(dead_root, &fs_info->dead_roots,
3452 root_list) {
3453 if (dead_root->root_key.objectid ==
3454 found_key.objectid) {
3455 is_dead_root = 1;
3456 break;
3457 }
3458 }
3459 spin_unlock(&fs_info->trans_lock);
3460 if (is_dead_root) {
3461 /* prevent this orphan from being found again */
3462 key.offset = found_key.objectid - 1;
3463 continue;
3464 }
3465 }
3466 /*
3467 * Inode is already gone but the orphan item is still there,
3468 * kill the orphan item.
3469 */
3470 if (ret == -ESTALE) {
3471 trans = btrfs_start_transaction(root, 1);
3472 if (IS_ERR(trans)) {
3473 ret = PTR_ERR(trans);
3474 goto out;
3475 }
3476 btrfs_debug(root->fs_info, "auto deleting %Lu",
3477 found_key.objectid);
3478 ret = btrfs_del_orphan_item(trans, root,
3479 found_key.objectid);
3480 btrfs_end_transaction(trans, root);
3481 if (ret)
3482 goto out;
3483 continue;
3484 }
3485
3486 /*
3487 * add this inode to the orphan list so btrfs_orphan_del does
3488 * the proper thing when we hit it
3489 */
3490 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3491 &BTRFS_I(inode)->runtime_flags);
3492 atomic_inc(&root->orphan_inodes);
3493
3494 /* if we have links, this was a truncate, lets do that */
3495 if (inode->i_nlink) {
3496 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3497 iput(inode);
3498 continue;
3499 }
3500 nr_truncate++;
3501
3502 /* 1 for the orphan item deletion. */
3503 trans = btrfs_start_transaction(root, 1);
3504 if (IS_ERR(trans)) {
3505 iput(inode);
3506 ret = PTR_ERR(trans);
3507 goto out;
3508 }
3509 ret = btrfs_orphan_add(trans, inode);
3510 btrfs_end_transaction(trans, root);
3511 if (ret) {
3512 iput(inode);
3513 goto out;
3514 }
3515
3516 ret = btrfs_truncate(inode);
3517 if (ret)
3518 btrfs_orphan_del(NULL, inode);
3519 } else {
3520 nr_unlink++;
3521 }
3522
3523 /* this will do delete_inode and everything for us */
3524 iput(inode);
3525 if (ret)
3526 goto out;
3527 }
3528 /* release the path since we're done with it */
3529 btrfs_release_path(path);
3530
3531 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3532
3533 if (root->orphan_block_rsv)
3534 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3535 (u64)-1);
3536
3537 if (root->orphan_block_rsv ||
3538 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3539 trans = btrfs_join_transaction(root);
3540 if (!IS_ERR(trans))
3541 btrfs_end_transaction(trans, root);
3542 }
3543
3544 if (nr_unlink)
3545 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3546 if (nr_truncate)
3547 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3548
3549 out:
3550 if (ret)
3551 btrfs_err(root->fs_info,
3552 "could not do orphan cleanup %d", ret);
3553 btrfs_free_path(path);
3554 return ret;
3555 }
3556
3557 /*
3558 * very simple check to peek ahead in the leaf looking for xattrs. If we
3559 * don't find any xattrs, we know there can't be any acls.
3560 *
3561 * slot is the slot the inode is in, objectid is the objectid of the inode
3562 */
3563 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3564 int slot, u64 objectid,
3565 int *first_xattr_slot)
3566 {
3567 u32 nritems = btrfs_header_nritems(leaf);
3568 struct btrfs_key found_key;
3569 static u64 xattr_access = 0;
3570 static u64 xattr_default = 0;
3571 int scanned = 0;
3572
3573 if (!xattr_access) {
3574 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3575 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3576 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3577 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3578 }
3579
3580 slot++;
3581 *first_xattr_slot = -1;
3582 while (slot < nritems) {
3583 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3584
3585 /* we found a different objectid, there must not be acls */
3586 if (found_key.objectid != objectid)
3587 return 0;
3588
3589 /* we found an xattr, assume we've got an acl */
3590 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3591 if (*first_xattr_slot == -1)
3592 *first_xattr_slot = slot;
3593 if (found_key.offset == xattr_access ||
3594 found_key.offset == xattr_default)
3595 return 1;
3596 }
3597
3598 /*
3599 * we found a key greater than an xattr key, there can't
3600 * be any acls later on
3601 */
3602 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3603 return 0;
3604
3605 slot++;
3606 scanned++;
3607
3608 /*
3609 * it goes inode, inode backrefs, xattrs, extents,
3610 * so if there are a ton of hard links to an inode there can
3611 * be a lot of backrefs. Don't waste time searching too hard,
3612 * this is just an optimization
3613 */
3614 if (scanned >= 8)
3615 break;
3616 }
3617 /* we hit the end of the leaf before we found an xattr or
3618 * something larger than an xattr. We have to assume the inode
3619 * has acls
3620 */
3621 if (*first_xattr_slot == -1)
3622 *first_xattr_slot = slot;
3623 return 1;
3624 }
3625
3626 /*
3627 * read an inode from the btree into the in-memory inode
3628 */
3629 static void btrfs_read_locked_inode(struct inode *inode)
3630 {
3631 struct btrfs_path *path;
3632 struct extent_buffer *leaf;
3633 struct btrfs_inode_item *inode_item;
3634 struct btrfs_root *root = BTRFS_I(inode)->root;
3635 struct btrfs_key location;
3636 unsigned long ptr;
3637 int maybe_acls;
3638 u32 rdev;
3639 int ret;
3640 bool filled = false;
3641 int first_xattr_slot;
3642
3643 ret = btrfs_fill_inode(inode, &rdev);
3644 if (!ret)
3645 filled = true;
3646
3647 path = btrfs_alloc_path();
3648 if (!path)
3649 goto make_bad;
3650
3651 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3652
3653 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3654 if (ret)
3655 goto make_bad;
3656
3657 leaf = path->nodes[0];
3658
3659 if (filled)
3660 goto cache_index;
3661
3662 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3663 struct btrfs_inode_item);
3664 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3665 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3666 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3667 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3668 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3669
3670 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3671 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3672
3673 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3674 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3675
3676 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3677 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3678
3679 BTRFS_I(inode)->i_otime.tv_sec =
3680 btrfs_timespec_sec(leaf, &inode_item->otime);
3681 BTRFS_I(inode)->i_otime.tv_nsec =
3682 btrfs_timespec_nsec(leaf, &inode_item->otime);
3683
3684 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3685 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3686 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3687
3688 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3689 inode->i_generation = BTRFS_I(inode)->generation;
3690 inode->i_rdev = 0;
3691 rdev = btrfs_inode_rdev(leaf, inode_item);
3692
3693 BTRFS_I(inode)->index_cnt = (u64)-1;
3694 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3695
3696 cache_index:
3697 /*
3698 * If we were modified in the current generation and evicted from memory
3699 * and then re-read we need to do a full sync since we don't have any
3700 * idea about which extents were modified before we were evicted from
3701 * cache.
3702 *
3703 * This is required for both inode re-read from disk and delayed inode
3704 * in delayed_nodes_tree.
3705 */
3706 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3707 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3708 &BTRFS_I(inode)->runtime_flags);
3709
3710 /*
3711 * We don't persist the id of the transaction where an unlink operation
3712 * against the inode was last made. So here we assume the inode might
3713 * have been evicted, and therefore the exact value of last_unlink_trans
3714 * lost, and set it to last_trans to avoid metadata inconsistencies
3715 * between the inode and its parent if the inode is fsync'ed and the log
3716 * replayed. For example, in the scenario:
3717 *
3718 * touch mydir/foo
3719 * ln mydir/foo mydir/bar
3720 * sync
3721 * unlink mydir/bar
3722 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3723 * xfs_io -c fsync mydir/foo
3724 * <power failure>
3725 * mount fs, triggers fsync log replay
3726 *
3727 * We must make sure that when we fsync our inode foo we also log its
3728 * parent inode, otherwise after log replay the parent still has the
3729 * dentry with the "bar" name but our inode foo has a link count of 1
3730 * and doesn't have an inode ref with the name "bar" anymore.
3731 *
3732 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3733 * but it guarantees correctness at the expense of occasional full
3734 * transaction commits on fsync if our inode is a directory, or if our
3735 * inode is not a directory, logging its parent unnecessarily.
3736 */
3737 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3738
3739 path->slots[0]++;
3740 if (inode->i_nlink != 1 ||
3741 path->slots[0] >= btrfs_header_nritems(leaf))
3742 goto cache_acl;
3743
3744 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3745 if (location.objectid != btrfs_ino(inode))
3746 goto cache_acl;
3747
3748 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3749 if (location.type == BTRFS_INODE_REF_KEY) {
3750 struct btrfs_inode_ref *ref;
3751
3752 ref = (struct btrfs_inode_ref *)ptr;
3753 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3754 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3755 struct btrfs_inode_extref *extref;
3756
3757 extref = (struct btrfs_inode_extref *)ptr;
3758 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3759 extref);
3760 }
3761 cache_acl:
3762 /*
3763 * try to precache a NULL acl entry for files that don't have
3764 * any xattrs or acls
3765 */
3766 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3767 btrfs_ino(inode), &first_xattr_slot);
3768 if (first_xattr_slot != -1) {
3769 path->slots[0] = first_xattr_slot;
3770 ret = btrfs_load_inode_props(inode, path);
3771 if (ret)
3772 btrfs_err(root->fs_info,
3773 "error loading props for ino %llu (root %llu): %d",
3774 btrfs_ino(inode),
3775 root->root_key.objectid, ret);
3776 }
3777 btrfs_free_path(path);
3778
3779 if (!maybe_acls)
3780 cache_no_acl(inode);
3781
3782 switch (inode->i_mode & S_IFMT) {
3783 case S_IFREG:
3784 inode->i_mapping->a_ops = &btrfs_aops;
3785 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3786 inode->i_fop = &btrfs_file_operations;
3787 inode->i_op = &btrfs_file_inode_operations;
3788 break;
3789 case S_IFDIR:
3790 inode->i_fop = &btrfs_dir_file_operations;
3791 if (root == root->fs_info->tree_root)
3792 inode->i_op = &btrfs_dir_ro_inode_operations;
3793 else
3794 inode->i_op = &btrfs_dir_inode_operations;
3795 break;
3796 case S_IFLNK:
3797 inode->i_op = &btrfs_symlink_inode_operations;
3798 inode_nohighmem(inode);
3799 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3800 break;
3801 default:
3802 inode->i_op = &btrfs_special_inode_operations;
3803 init_special_inode(inode, inode->i_mode, rdev);
3804 break;
3805 }
3806
3807 btrfs_update_iflags(inode);
3808 return;
3809
3810 make_bad:
3811 btrfs_free_path(path);
3812 make_bad_inode(inode);
3813 }
3814
3815 /*
3816 * given a leaf and an inode, copy the inode fields into the leaf
3817 */
3818 static void fill_inode_item(struct btrfs_trans_handle *trans,
3819 struct extent_buffer *leaf,
3820 struct btrfs_inode_item *item,
3821 struct inode *inode)
3822 {
3823 struct btrfs_map_token token;
3824
3825 btrfs_init_map_token(&token);
3826
3827 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3828 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3829 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3830 &token);
3831 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3832 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3833
3834 btrfs_set_token_timespec_sec(leaf, &item->atime,
3835 inode->i_atime.tv_sec, &token);
3836 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3837 inode->i_atime.tv_nsec, &token);
3838
3839 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3840 inode->i_mtime.tv_sec, &token);
3841 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3842 inode->i_mtime.tv_nsec, &token);
3843
3844 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3845 inode->i_ctime.tv_sec, &token);
3846 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3847 inode->i_ctime.tv_nsec, &token);
3848
3849 btrfs_set_token_timespec_sec(leaf, &item->otime,
3850 BTRFS_I(inode)->i_otime.tv_sec, &token);
3851 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3852 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3853
3854 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3855 &token);
3856 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3857 &token);
3858 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3859 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3860 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3861 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3862 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3863 }
3864
3865 /*
3866 * copy everything in the in-memory inode into the btree.
3867 */
3868 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3869 struct btrfs_root *root, struct inode *inode)
3870 {
3871 struct btrfs_inode_item *inode_item;
3872 struct btrfs_path *path;
3873 struct extent_buffer *leaf;
3874 int ret;
3875
3876 path = btrfs_alloc_path();
3877 if (!path)
3878 return -ENOMEM;
3879
3880 path->leave_spinning = 1;
3881 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3882 1);
3883 if (ret) {
3884 if (ret > 0)
3885 ret = -ENOENT;
3886 goto failed;
3887 }
3888
3889 leaf = path->nodes[0];
3890 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3891 struct btrfs_inode_item);
3892
3893 fill_inode_item(trans, leaf, inode_item, inode);
3894 btrfs_mark_buffer_dirty(leaf);
3895 btrfs_set_inode_last_trans(trans, inode);
3896 ret = 0;
3897 failed:
3898 btrfs_free_path(path);
3899 return ret;
3900 }
3901
3902 /*
3903 * copy everything in the in-memory inode into the btree.
3904 */
3905 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3906 struct btrfs_root *root, struct inode *inode)
3907 {
3908 int ret;
3909
3910 /*
3911 * If the inode is a free space inode, we can deadlock during commit
3912 * if we put it into the delayed code.
3913 *
3914 * The data relocation inode should also be directly updated
3915 * without delay
3916 */
3917 if (!btrfs_is_free_space_inode(inode)
3918 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3919 && !root->fs_info->log_root_recovering) {
3920 btrfs_update_root_times(trans, root);
3921
3922 ret = btrfs_delayed_update_inode(trans, root, inode);
3923 if (!ret)
3924 btrfs_set_inode_last_trans(trans, inode);
3925 return ret;
3926 }
3927
3928 return btrfs_update_inode_item(trans, root, inode);
3929 }
3930
3931 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3932 struct btrfs_root *root,
3933 struct inode *inode)
3934 {
3935 int ret;
3936
3937 ret = btrfs_update_inode(trans, root, inode);
3938 if (ret == -ENOSPC)
3939 return btrfs_update_inode_item(trans, root, inode);
3940 return ret;
3941 }
3942
3943 /*
3944 * unlink helper that gets used here in inode.c and in the tree logging
3945 * recovery code. It remove a link in a directory with a given name, and
3946 * also drops the back refs in the inode to the directory
3947 */
3948 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3949 struct btrfs_root *root,
3950 struct inode *dir, struct inode *inode,
3951 const char *name, int name_len)
3952 {
3953 struct btrfs_path *path;
3954 int ret = 0;
3955 struct extent_buffer *leaf;
3956 struct btrfs_dir_item *di;
3957 struct btrfs_key key;
3958 u64 index;
3959 u64 ino = btrfs_ino(inode);
3960 u64 dir_ino = btrfs_ino(dir);
3961
3962 path = btrfs_alloc_path();
3963 if (!path) {
3964 ret = -ENOMEM;
3965 goto out;
3966 }
3967
3968 path->leave_spinning = 1;
3969 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3970 name, name_len, -1);
3971 if (IS_ERR(di)) {
3972 ret = PTR_ERR(di);
3973 goto err;
3974 }
3975 if (!di) {
3976 ret = -ENOENT;
3977 goto err;
3978 }
3979 leaf = path->nodes[0];
3980 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3981 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3982 if (ret)
3983 goto err;
3984 btrfs_release_path(path);
3985
3986 /*
3987 * If we don't have dir index, we have to get it by looking up
3988 * the inode ref, since we get the inode ref, remove it directly,
3989 * it is unnecessary to do delayed deletion.
3990 *
3991 * But if we have dir index, needn't search inode ref to get it.
3992 * Since the inode ref is close to the inode item, it is better
3993 * that we delay to delete it, and just do this deletion when
3994 * we update the inode item.
3995 */
3996 if (BTRFS_I(inode)->dir_index) {
3997 ret = btrfs_delayed_delete_inode_ref(inode);
3998 if (!ret) {
3999 index = BTRFS_I(inode)->dir_index;
4000 goto skip_backref;
4001 }
4002 }
4003
4004 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4005 dir_ino, &index);
4006 if (ret) {
4007 btrfs_info(root->fs_info,
4008 "failed to delete reference to %.*s, inode %llu parent %llu",
4009 name_len, name, ino, dir_ino);
4010 btrfs_abort_transaction(trans, root, ret);
4011 goto err;
4012 }
4013 skip_backref:
4014 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4015 if (ret) {
4016 btrfs_abort_transaction(trans, root, ret);
4017 goto err;
4018 }
4019
4020 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
4021 inode, dir_ino);
4022 if (ret != 0 && ret != -ENOENT) {
4023 btrfs_abort_transaction(trans, root, ret);
4024 goto err;
4025 }
4026
4027 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4028 dir, index);
4029 if (ret == -ENOENT)
4030 ret = 0;
4031 else if (ret)
4032 btrfs_abort_transaction(trans, root, ret);
4033 err:
4034 btrfs_free_path(path);
4035 if (ret)
4036 goto out;
4037
4038 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4039 inode_inc_iversion(inode);
4040 inode_inc_iversion(dir);
4041 inode->i_ctime = dir->i_mtime =
4042 dir->i_ctime = current_fs_time(inode->i_sb);
4043 ret = btrfs_update_inode(trans, root, dir);
4044 out:
4045 return ret;
4046 }
4047
4048 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4049 struct btrfs_root *root,
4050 struct inode *dir, struct inode *inode,
4051 const char *name, int name_len)
4052 {
4053 int ret;
4054 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4055 if (!ret) {
4056 drop_nlink(inode);
4057 ret = btrfs_update_inode(trans, root, inode);
4058 }
4059 return ret;
4060 }
4061
4062 /*
4063 * helper to start transaction for unlink and rmdir.
4064 *
4065 * unlink and rmdir are special in btrfs, they do not always free space, so
4066 * if we cannot make our reservations the normal way try and see if there is
4067 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4068 * allow the unlink to occur.
4069 */
4070 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4071 {
4072 struct btrfs_root *root = BTRFS_I(dir)->root;
4073
4074 /*
4075 * 1 for the possible orphan item
4076 * 1 for the dir item
4077 * 1 for the dir index
4078 * 1 for the inode ref
4079 * 1 for the inode
4080 */
4081 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4082 }
4083
4084 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4085 {
4086 struct btrfs_root *root = BTRFS_I(dir)->root;
4087 struct btrfs_trans_handle *trans;
4088 struct inode *inode = d_inode(dentry);
4089 int ret;
4090
4091 trans = __unlink_start_trans(dir);
4092 if (IS_ERR(trans))
4093 return PTR_ERR(trans);
4094
4095 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4096
4097 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4098 dentry->d_name.name, dentry->d_name.len);
4099 if (ret)
4100 goto out;
4101
4102 if (inode->i_nlink == 0) {
4103 ret = btrfs_orphan_add(trans, inode);
4104 if (ret)
4105 goto out;
4106 }
4107
4108 out:
4109 btrfs_end_transaction(trans, root);
4110 btrfs_btree_balance_dirty(root);
4111 return ret;
4112 }
4113
4114 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4115 struct btrfs_root *root,
4116 struct inode *dir, u64 objectid,
4117 const char *name, int name_len)
4118 {
4119 struct btrfs_path *path;
4120 struct extent_buffer *leaf;
4121 struct btrfs_dir_item *di;
4122 struct btrfs_key key;
4123 u64 index;
4124 int ret;
4125 u64 dir_ino = btrfs_ino(dir);
4126
4127 path = btrfs_alloc_path();
4128 if (!path)
4129 return -ENOMEM;
4130
4131 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4132 name, name_len, -1);
4133 if (IS_ERR_OR_NULL(di)) {
4134 if (!di)
4135 ret = -ENOENT;
4136 else
4137 ret = PTR_ERR(di);
4138 goto out;
4139 }
4140
4141 leaf = path->nodes[0];
4142 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4143 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4144 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4145 if (ret) {
4146 btrfs_abort_transaction(trans, root, ret);
4147 goto out;
4148 }
4149 btrfs_release_path(path);
4150
4151 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4152 objectid, root->root_key.objectid,
4153 dir_ino, &index, name, name_len);
4154 if (ret < 0) {
4155 if (ret != -ENOENT) {
4156 btrfs_abort_transaction(trans, root, ret);
4157 goto out;
4158 }
4159 di = btrfs_search_dir_index_item(root, path, dir_ino,
4160 name, name_len);
4161 if (IS_ERR_OR_NULL(di)) {
4162 if (!di)
4163 ret = -ENOENT;
4164 else
4165 ret = PTR_ERR(di);
4166 btrfs_abort_transaction(trans, root, ret);
4167 goto out;
4168 }
4169
4170 leaf = path->nodes[0];
4171 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4172 btrfs_release_path(path);
4173 index = key.offset;
4174 }
4175 btrfs_release_path(path);
4176
4177 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4178 if (ret) {
4179 btrfs_abort_transaction(trans, root, ret);
4180 goto out;
4181 }
4182
4183 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4184 inode_inc_iversion(dir);
4185 dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb);
4186 ret = btrfs_update_inode_fallback(trans, root, dir);
4187 if (ret)
4188 btrfs_abort_transaction(trans, root, ret);
4189 out:
4190 btrfs_free_path(path);
4191 return ret;
4192 }
4193
4194 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4195 {
4196 struct inode *inode = d_inode(dentry);
4197 int err = 0;
4198 struct btrfs_root *root = BTRFS_I(dir)->root;
4199 struct btrfs_trans_handle *trans;
4200
4201 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4202 return -ENOTEMPTY;
4203 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4204 return -EPERM;
4205
4206 trans = __unlink_start_trans(dir);
4207 if (IS_ERR(trans))
4208 return PTR_ERR(trans);
4209
4210 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4211 err = btrfs_unlink_subvol(trans, root, dir,
4212 BTRFS_I(inode)->location.objectid,
4213 dentry->d_name.name,
4214 dentry->d_name.len);
4215 goto out;
4216 }
4217
4218 err = btrfs_orphan_add(trans, inode);
4219 if (err)
4220 goto out;
4221
4222 /* now the directory is empty */
4223 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4224 dentry->d_name.name, dentry->d_name.len);
4225 if (!err)
4226 btrfs_i_size_write(inode, 0);
4227 out:
4228 btrfs_end_transaction(trans, root);
4229 btrfs_btree_balance_dirty(root);
4230
4231 return err;
4232 }
4233
4234 static int truncate_space_check(struct btrfs_trans_handle *trans,
4235 struct btrfs_root *root,
4236 u64 bytes_deleted)
4237 {
4238 int ret;
4239
4240 /*
4241 * This is only used to apply pressure to the enospc system, we don't
4242 * intend to use this reservation at all.
4243 */
4244 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4245 bytes_deleted *= root->nodesize;
4246 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4247 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4248 if (!ret) {
4249 trace_btrfs_space_reservation(root->fs_info, "transaction",
4250 trans->transid,
4251 bytes_deleted, 1);
4252 trans->bytes_reserved += bytes_deleted;
4253 }
4254 return ret;
4255
4256 }
4257
4258 static int truncate_inline_extent(struct inode *inode,
4259 struct btrfs_path *path,
4260 struct btrfs_key *found_key,
4261 const u64 item_end,
4262 const u64 new_size)
4263 {
4264 struct extent_buffer *leaf = path->nodes[0];
4265 int slot = path->slots[0];
4266 struct btrfs_file_extent_item *fi;
4267 u32 size = (u32)(new_size - found_key->offset);
4268 struct btrfs_root *root = BTRFS_I(inode)->root;
4269
4270 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4271
4272 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4273 loff_t offset = new_size;
4274 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4275
4276 /*
4277 * Zero out the remaining of the last page of our inline extent,
4278 * instead of directly truncating our inline extent here - that
4279 * would be much more complex (decompressing all the data, then
4280 * compressing the truncated data, which might be bigger than
4281 * the size of the inline extent, resize the extent, etc).
4282 * We release the path because to get the page we might need to
4283 * read the extent item from disk (data not in the page cache).
4284 */
4285 btrfs_release_path(path);
4286 return btrfs_truncate_block(inode, offset, page_end - offset,
4287 0);
4288 }
4289
4290 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4291 size = btrfs_file_extent_calc_inline_size(size);
4292 btrfs_truncate_item(root, path, size, 1);
4293
4294 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4295 inode_sub_bytes(inode, item_end + 1 - new_size);
4296
4297 return 0;
4298 }
4299
4300 /*
4301 * this can truncate away extent items, csum items and directory items.
4302 * It starts at a high offset and removes keys until it can't find
4303 * any higher than new_size
4304 *
4305 * csum items that cross the new i_size are truncated to the new size
4306 * as well.
4307 *
4308 * min_type is the minimum key type to truncate down to. If set to 0, this
4309 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4310 */
4311 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4312 struct btrfs_root *root,
4313 struct inode *inode,
4314 u64 new_size, u32 min_type)
4315 {
4316 struct btrfs_path *path;
4317 struct extent_buffer *leaf;
4318 struct btrfs_file_extent_item *fi;
4319 struct btrfs_key key;
4320 struct btrfs_key found_key;
4321 u64 extent_start = 0;
4322 u64 extent_num_bytes = 0;
4323 u64 extent_offset = 0;
4324 u64 item_end = 0;
4325 u64 last_size = new_size;
4326 u32 found_type = (u8)-1;
4327 int found_extent;
4328 int del_item;
4329 int pending_del_nr = 0;
4330 int pending_del_slot = 0;
4331 int extent_type = -1;
4332 int ret;
4333 int err = 0;
4334 u64 ino = btrfs_ino(inode);
4335 u64 bytes_deleted = 0;
4336 bool be_nice = 0;
4337 bool should_throttle = 0;
4338 bool should_end = 0;
4339
4340 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4341
4342 /*
4343 * for non-free space inodes and ref cows, we want to back off from
4344 * time to time
4345 */
4346 if (!btrfs_is_free_space_inode(inode) &&
4347 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4348 be_nice = 1;
4349
4350 path = btrfs_alloc_path();
4351 if (!path)
4352 return -ENOMEM;
4353 path->reada = READA_BACK;
4354
4355 /*
4356 * We want to drop from the next block forward in case this new size is
4357 * not block aligned since we will be keeping the last block of the
4358 * extent just the way it is.
4359 */
4360 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4361 root == root->fs_info->tree_root)
4362 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4363 root->sectorsize), (u64)-1, 0);
4364
4365 /*
4366 * This function is also used to drop the items in the log tree before
4367 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4368 * it is used to drop the loged items. So we shouldn't kill the delayed
4369 * items.
4370 */
4371 if (min_type == 0 && root == BTRFS_I(inode)->root)
4372 btrfs_kill_delayed_inode_items(inode);
4373
4374 key.objectid = ino;
4375 key.offset = (u64)-1;
4376 key.type = (u8)-1;
4377
4378 search_again:
4379 /*
4380 * with a 16K leaf size and 128MB extents, you can actually queue
4381 * up a huge file in a single leaf. Most of the time that
4382 * bytes_deleted is > 0, it will be huge by the time we get here
4383 */
4384 if (be_nice && bytes_deleted > SZ_32M) {
4385 if (btrfs_should_end_transaction(trans, root)) {
4386 err = -EAGAIN;
4387 goto error;
4388 }
4389 }
4390
4391
4392 path->leave_spinning = 1;
4393 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4394 if (ret < 0) {
4395 err = ret;
4396 goto out;
4397 }
4398
4399 if (ret > 0) {
4400 /* there are no items in the tree for us to truncate, we're
4401 * done
4402 */
4403 if (path->slots[0] == 0)
4404 goto out;
4405 path->slots[0]--;
4406 }
4407
4408 while (1) {
4409 fi = NULL;
4410 leaf = path->nodes[0];
4411 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4412 found_type = found_key.type;
4413
4414 if (found_key.objectid != ino)
4415 break;
4416
4417 if (found_type < min_type)
4418 break;
4419
4420 item_end = found_key.offset;
4421 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4422 fi = btrfs_item_ptr(leaf, path->slots[0],
4423 struct btrfs_file_extent_item);
4424 extent_type = btrfs_file_extent_type(leaf, fi);
4425 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4426 item_end +=
4427 btrfs_file_extent_num_bytes(leaf, fi);
4428 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4429 item_end += btrfs_file_extent_inline_len(leaf,
4430 path->slots[0], fi);
4431 }
4432 item_end--;
4433 }
4434 if (found_type > min_type) {
4435 del_item = 1;
4436 } else {
4437 if (item_end < new_size)
4438 break;
4439 if (found_key.offset >= new_size)
4440 del_item = 1;
4441 else
4442 del_item = 0;
4443 }
4444 found_extent = 0;
4445 /* FIXME, shrink the extent if the ref count is only 1 */
4446 if (found_type != BTRFS_EXTENT_DATA_KEY)
4447 goto delete;
4448
4449 if (del_item)
4450 last_size = found_key.offset;
4451 else
4452 last_size = new_size;
4453
4454 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4455 u64 num_dec;
4456 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4457 if (!del_item) {
4458 u64 orig_num_bytes =
4459 btrfs_file_extent_num_bytes(leaf, fi);
4460 extent_num_bytes = ALIGN(new_size -
4461 found_key.offset,
4462 root->sectorsize);
4463 btrfs_set_file_extent_num_bytes(leaf, fi,
4464 extent_num_bytes);
4465 num_dec = (orig_num_bytes -
4466 extent_num_bytes);
4467 if (test_bit(BTRFS_ROOT_REF_COWS,
4468 &root->state) &&
4469 extent_start != 0)
4470 inode_sub_bytes(inode, num_dec);
4471 btrfs_mark_buffer_dirty(leaf);
4472 } else {
4473 extent_num_bytes =
4474 btrfs_file_extent_disk_num_bytes(leaf,
4475 fi);
4476 extent_offset = found_key.offset -
4477 btrfs_file_extent_offset(leaf, fi);
4478
4479 /* FIXME blocksize != 4096 */
4480 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4481 if (extent_start != 0) {
4482 found_extent = 1;
4483 if (test_bit(BTRFS_ROOT_REF_COWS,
4484 &root->state))
4485 inode_sub_bytes(inode, num_dec);
4486 }
4487 }
4488 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4489 /*
4490 * we can't truncate inline items that have had
4491 * special encodings
4492 */
4493 if (!del_item &&
4494 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4495 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4496
4497 /*
4498 * Need to release path in order to truncate a
4499 * compressed extent. So delete any accumulated
4500 * extent items so far.
4501 */
4502 if (btrfs_file_extent_compression(leaf, fi) !=
4503 BTRFS_COMPRESS_NONE && pending_del_nr) {
4504 err = btrfs_del_items(trans, root, path,
4505 pending_del_slot,
4506 pending_del_nr);
4507 if (err) {
4508 btrfs_abort_transaction(trans,
4509 root,
4510 err);
4511 goto error;
4512 }
4513 pending_del_nr = 0;
4514 }
4515
4516 err = truncate_inline_extent(inode, path,
4517 &found_key,
4518 item_end,
4519 new_size);
4520 if (err) {
4521 btrfs_abort_transaction(trans,
4522 root, err);
4523 goto error;
4524 }
4525 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4526 &root->state)) {
4527 inode_sub_bytes(inode, item_end + 1 - new_size);
4528 }
4529 }
4530 delete:
4531 if (del_item) {
4532 if (!pending_del_nr) {
4533 /* no pending yet, add ourselves */
4534 pending_del_slot = path->slots[0];
4535 pending_del_nr = 1;
4536 } else if (pending_del_nr &&
4537 path->slots[0] + 1 == pending_del_slot) {
4538 /* hop on the pending chunk */
4539 pending_del_nr++;
4540 pending_del_slot = path->slots[0];
4541 } else {
4542 BUG();
4543 }
4544 } else {
4545 break;
4546 }
4547 should_throttle = 0;
4548
4549 if (found_extent &&
4550 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4551 root == root->fs_info->tree_root)) {
4552 btrfs_set_path_blocking(path);
4553 bytes_deleted += extent_num_bytes;
4554 ret = btrfs_free_extent(trans, root, extent_start,
4555 extent_num_bytes, 0,
4556 btrfs_header_owner(leaf),
4557 ino, extent_offset);
4558 BUG_ON(ret);
4559 if (btrfs_should_throttle_delayed_refs(trans, root))
4560 btrfs_async_run_delayed_refs(root,
4561 trans->transid,
4562 trans->delayed_ref_updates * 2, 0);
4563 if (be_nice) {
4564 if (truncate_space_check(trans, root,
4565 extent_num_bytes)) {
4566 should_end = 1;
4567 }
4568 if (btrfs_should_throttle_delayed_refs(trans,
4569 root)) {
4570 should_throttle = 1;
4571 }
4572 }
4573 }
4574
4575 if (found_type == BTRFS_INODE_ITEM_KEY)
4576 break;
4577
4578 if (path->slots[0] == 0 ||
4579 path->slots[0] != pending_del_slot ||
4580 should_throttle || should_end) {
4581 if (pending_del_nr) {
4582 ret = btrfs_del_items(trans, root, path,
4583 pending_del_slot,
4584 pending_del_nr);
4585 if (ret) {
4586 btrfs_abort_transaction(trans,
4587 root, ret);
4588 goto error;
4589 }
4590 pending_del_nr = 0;
4591 }
4592 btrfs_release_path(path);
4593 if (should_throttle) {
4594 unsigned long updates = trans->delayed_ref_updates;
4595 if (updates) {
4596 trans->delayed_ref_updates = 0;
4597 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4598 if (ret && !err)
4599 err = ret;
4600 }
4601 }
4602 /*
4603 * if we failed to refill our space rsv, bail out
4604 * and let the transaction restart
4605 */
4606 if (should_end) {
4607 err = -EAGAIN;
4608 goto error;
4609 }
4610 goto search_again;
4611 } else {
4612 path->slots[0]--;
4613 }
4614 }
4615 out:
4616 if (pending_del_nr) {
4617 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4618 pending_del_nr);
4619 if (ret)
4620 btrfs_abort_transaction(trans, root, ret);
4621 }
4622 error:
4623 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4624 btrfs_ordered_update_i_size(inode, last_size, NULL);
4625
4626 btrfs_free_path(path);
4627
4628 if (be_nice && bytes_deleted > SZ_32M) {
4629 unsigned long updates = trans->delayed_ref_updates;
4630 if (updates) {
4631 trans->delayed_ref_updates = 0;
4632 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4633 if (ret && !err)
4634 err = ret;
4635 }
4636 }
4637 return err;
4638 }
4639
4640 /*
4641 * btrfs_truncate_block - read, zero a chunk and write a block
4642 * @inode - inode that we're zeroing
4643 * @from - the offset to start zeroing
4644 * @len - the length to zero, 0 to zero the entire range respective to the
4645 * offset
4646 * @front - zero up to the offset instead of from the offset on
4647 *
4648 * This will find the block for the "from" offset and cow the block and zero the
4649 * part we want to zero. This is used with truncate and hole punching.
4650 */
4651 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4652 int front)
4653 {
4654 struct address_space *mapping = inode->i_mapping;
4655 struct btrfs_root *root = BTRFS_I(inode)->root;
4656 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4657 struct btrfs_ordered_extent *ordered;
4658 struct extent_state *cached_state = NULL;
4659 char *kaddr;
4660 u32 blocksize = root->sectorsize;
4661 pgoff_t index = from >> PAGE_SHIFT;
4662 unsigned offset = from & (blocksize - 1);
4663 struct page *page;
4664 gfp_t mask = btrfs_alloc_write_mask(mapping);
4665 int ret = 0;
4666 u64 block_start;
4667 u64 block_end;
4668
4669 if ((offset & (blocksize - 1)) == 0 &&
4670 (!len || ((len & (blocksize - 1)) == 0)))
4671 goto out;
4672
4673 ret = btrfs_delalloc_reserve_space(inode,
4674 round_down(from, blocksize), blocksize);
4675 if (ret)
4676 goto out;
4677
4678 again:
4679 page = find_or_create_page(mapping, index, mask);
4680 if (!page) {
4681 btrfs_delalloc_release_space(inode,
4682 round_down(from, blocksize),
4683 blocksize);
4684 ret = -ENOMEM;
4685 goto out;
4686 }
4687
4688 block_start = round_down(from, blocksize);
4689 block_end = block_start + blocksize - 1;
4690
4691 if (!PageUptodate(page)) {
4692 ret = btrfs_readpage(NULL, page);
4693 lock_page(page);
4694 if (page->mapping != mapping) {
4695 unlock_page(page);
4696 put_page(page);
4697 goto again;
4698 }
4699 if (!PageUptodate(page)) {
4700 ret = -EIO;
4701 goto out_unlock;
4702 }
4703 }
4704 wait_on_page_writeback(page);
4705
4706 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4707 set_page_extent_mapped(page);
4708
4709 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4710 if (ordered) {
4711 unlock_extent_cached(io_tree, block_start, block_end,
4712 &cached_state, GFP_NOFS);
4713 unlock_page(page);
4714 put_page(page);
4715 btrfs_start_ordered_extent(inode, ordered, 1);
4716 btrfs_put_ordered_extent(ordered);
4717 goto again;
4718 }
4719
4720 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4721 EXTENT_DIRTY | EXTENT_DELALLOC |
4722 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4723 0, 0, &cached_state, GFP_NOFS);
4724
4725 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4726 &cached_state);
4727 if (ret) {
4728 unlock_extent_cached(io_tree, block_start, block_end,
4729 &cached_state, GFP_NOFS);
4730 goto out_unlock;
4731 }
4732
4733 if (offset != blocksize) {
4734 if (!len)
4735 len = blocksize - offset;
4736 kaddr = kmap(page);
4737 if (front)
4738 memset(kaddr + (block_start - page_offset(page)),
4739 0, offset);
4740 else
4741 memset(kaddr + (block_start - page_offset(page)) + offset,
4742 0, len);
4743 flush_dcache_page(page);
4744 kunmap(page);
4745 }
4746 ClearPageChecked(page);
4747 set_page_dirty(page);
4748 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4749 GFP_NOFS);
4750
4751 out_unlock:
4752 if (ret)
4753 btrfs_delalloc_release_space(inode, block_start,
4754 blocksize);
4755 unlock_page(page);
4756 put_page(page);
4757 out:
4758 return ret;
4759 }
4760
4761 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4762 u64 offset, u64 len)
4763 {
4764 struct btrfs_trans_handle *trans;
4765 int ret;
4766
4767 /*
4768 * Still need to make sure the inode looks like it's been updated so
4769 * that any holes get logged if we fsync.
4770 */
4771 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4772 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4773 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4774 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4775 return 0;
4776 }
4777
4778 /*
4779 * 1 - for the one we're dropping
4780 * 1 - for the one we're adding
4781 * 1 - for updating the inode.
4782 */
4783 trans = btrfs_start_transaction(root, 3);
4784 if (IS_ERR(trans))
4785 return PTR_ERR(trans);
4786
4787 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4788 if (ret) {
4789 btrfs_abort_transaction(trans, root, ret);
4790 btrfs_end_transaction(trans, root);
4791 return ret;
4792 }
4793
4794 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4795 0, 0, len, 0, len, 0, 0, 0);
4796 if (ret)
4797 btrfs_abort_transaction(trans, root, ret);
4798 else
4799 btrfs_update_inode(trans, root, inode);
4800 btrfs_end_transaction(trans, root);
4801 return ret;
4802 }
4803
4804 /*
4805 * This function puts in dummy file extents for the area we're creating a hole
4806 * for. So if we are truncating this file to a larger size we need to insert
4807 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4808 * the range between oldsize and size
4809 */
4810 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4811 {
4812 struct btrfs_root *root = BTRFS_I(inode)->root;
4813 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4814 struct extent_map *em = NULL;
4815 struct extent_state *cached_state = NULL;
4816 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4817 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4818 u64 block_end = ALIGN(size, root->sectorsize);
4819 u64 last_byte;
4820 u64 cur_offset;
4821 u64 hole_size;
4822 int err = 0;
4823
4824 /*
4825 * If our size started in the middle of a block we need to zero out the
4826 * rest of the block before we expand the i_size, otherwise we could
4827 * expose stale data.
4828 */
4829 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4830 if (err)
4831 return err;
4832
4833 if (size <= hole_start)
4834 return 0;
4835
4836 while (1) {
4837 struct btrfs_ordered_extent *ordered;
4838
4839 lock_extent_bits(io_tree, hole_start, block_end - 1,
4840 &cached_state);
4841 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4842 block_end - hole_start);
4843 if (!ordered)
4844 break;
4845 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4846 &cached_state, GFP_NOFS);
4847 btrfs_start_ordered_extent(inode, ordered, 1);
4848 btrfs_put_ordered_extent(ordered);
4849 }
4850
4851 cur_offset = hole_start;
4852 while (1) {
4853 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4854 block_end - cur_offset, 0);
4855 if (IS_ERR(em)) {
4856 err = PTR_ERR(em);
4857 em = NULL;
4858 break;
4859 }
4860 last_byte = min(extent_map_end(em), block_end);
4861 last_byte = ALIGN(last_byte , root->sectorsize);
4862 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4863 struct extent_map *hole_em;
4864 hole_size = last_byte - cur_offset;
4865
4866 err = maybe_insert_hole(root, inode, cur_offset,
4867 hole_size);
4868 if (err)
4869 break;
4870 btrfs_drop_extent_cache(inode, cur_offset,
4871 cur_offset + hole_size - 1, 0);
4872 hole_em = alloc_extent_map();
4873 if (!hole_em) {
4874 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4875 &BTRFS_I(inode)->runtime_flags);
4876 goto next;
4877 }
4878 hole_em->start = cur_offset;
4879 hole_em->len = hole_size;
4880 hole_em->orig_start = cur_offset;
4881
4882 hole_em->block_start = EXTENT_MAP_HOLE;
4883 hole_em->block_len = 0;
4884 hole_em->orig_block_len = 0;
4885 hole_em->ram_bytes = hole_size;
4886 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4887 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4888 hole_em->generation = root->fs_info->generation;
4889
4890 while (1) {
4891 write_lock(&em_tree->lock);
4892 err = add_extent_mapping(em_tree, hole_em, 1);
4893 write_unlock(&em_tree->lock);
4894 if (err != -EEXIST)
4895 break;
4896 btrfs_drop_extent_cache(inode, cur_offset,
4897 cur_offset +
4898 hole_size - 1, 0);
4899 }
4900 free_extent_map(hole_em);
4901 }
4902 next:
4903 free_extent_map(em);
4904 em = NULL;
4905 cur_offset = last_byte;
4906 if (cur_offset >= block_end)
4907 break;
4908 }
4909 free_extent_map(em);
4910 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4911 GFP_NOFS);
4912 return err;
4913 }
4914
4915 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4916 {
4917 struct btrfs_root *root = BTRFS_I(inode)->root;
4918 struct btrfs_trans_handle *trans;
4919 loff_t oldsize = i_size_read(inode);
4920 loff_t newsize = attr->ia_size;
4921 int mask = attr->ia_valid;
4922 int ret;
4923
4924 /*
4925 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4926 * special case where we need to update the times despite not having
4927 * these flags set. For all other operations the VFS set these flags
4928 * explicitly if it wants a timestamp update.
4929 */
4930 if (newsize != oldsize) {
4931 inode_inc_iversion(inode);
4932 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4933 inode->i_ctime = inode->i_mtime =
4934 current_fs_time(inode->i_sb);
4935 }
4936
4937 if (newsize > oldsize) {
4938 /*
4939 * Don't do an expanding truncate while snapshoting is ongoing.
4940 * This is to ensure the snapshot captures a fully consistent
4941 * state of this file - if the snapshot captures this expanding
4942 * truncation, it must capture all writes that happened before
4943 * this truncation.
4944 */
4945 btrfs_wait_for_snapshot_creation(root);
4946 ret = btrfs_cont_expand(inode, oldsize, newsize);
4947 if (ret) {
4948 btrfs_end_write_no_snapshoting(root);
4949 return ret;
4950 }
4951
4952 trans = btrfs_start_transaction(root, 1);
4953 if (IS_ERR(trans)) {
4954 btrfs_end_write_no_snapshoting(root);
4955 return PTR_ERR(trans);
4956 }
4957
4958 i_size_write(inode, newsize);
4959 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4960 pagecache_isize_extended(inode, oldsize, newsize);
4961 ret = btrfs_update_inode(trans, root, inode);
4962 btrfs_end_write_no_snapshoting(root);
4963 btrfs_end_transaction(trans, root);
4964 } else {
4965
4966 /*
4967 * We're truncating a file that used to have good data down to
4968 * zero. Make sure it gets into the ordered flush list so that
4969 * any new writes get down to disk quickly.
4970 */
4971 if (newsize == 0)
4972 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4973 &BTRFS_I(inode)->runtime_flags);
4974
4975 /*
4976 * 1 for the orphan item we're going to add
4977 * 1 for the orphan item deletion.
4978 */
4979 trans = btrfs_start_transaction(root, 2);
4980 if (IS_ERR(trans))
4981 return PTR_ERR(trans);
4982
4983 /*
4984 * We need to do this in case we fail at _any_ point during the
4985 * actual truncate. Once we do the truncate_setsize we could
4986 * invalidate pages which forces any outstanding ordered io to
4987 * be instantly completed which will give us extents that need
4988 * to be truncated. If we fail to get an orphan inode down we
4989 * could have left over extents that were never meant to live,
4990 * so we need to guarantee from this point on that everything
4991 * will be consistent.
4992 */
4993 ret = btrfs_orphan_add(trans, inode);
4994 btrfs_end_transaction(trans, root);
4995 if (ret)
4996 return ret;
4997
4998 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4999 truncate_setsize(inode, newsize);
5000
5001 /* Disable nonlocked read DIO to avoid the end less truncate */
5002 btrfs_inode_block_unlocked_dio(inode);
5003 inode_dio_wait(inode);
5004 btrfs_inode_resume_unlocked_dio(inode);
5005
5006 ret = btrfs_truncate(inode);
5007 if (ret && inode->i_nlink) {
5008 int err;
5009
5010 /*
5011 * failed to truncate, disk_i_size is only adjusted down
5012 * as we remove extents, so it should represent the true
5013 * size of the inode, so reset the in memory size and
5014 * delete our orphan entry.
5015 */
5016 trans = btrfs_join_transaction(root);
5017 if (IS_ERR(trans)) {
5018 btrfs_orphan_del(NULL, inode);
5019 return ret;
5020 }
5021 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5022 err = btrfs_orphan_del(trans, inode);
5023 if (err)
5024 btrfs_abort_transaction(trans, root, err);
5025 btrfs_end_transaction(trans, root);
5026 }
5027 }
5028
5029 return ret;
5030 }
5031
5032 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5033 {
5034 struct inode *inode = d_inode(dentry);
5035 struct btrfs_root *root = BTRFS_I(inode)->root;
5036 int err;
5037
5038 if (btrfs_root_readonly(root))
5039 return -EROFS;
5040
5041 err = inode_change_ok(inode, attr);
5042 if (err)
5043 return err;
5044
5045 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5046 err = btrfs_setsize(inode, attr);
5047 if (err)
5048 return err;
5049 }
5050
5051 if (attr->ia_valid) {
5052 setattr_copy(inode, attr);
5053 inode_inc_iversion(inode);
5054 err = btrfs_dirty_inode(inode);
5055
5056 if (!err && attr->ia_valid & ATTR_MODE)
5057 err = posix_acl_chmod(inode, inode->i_mode);
5058 }
5059
5060 return err;
5061 }
5062
5063 /*
5064 * While truncating the inode pages during eviction, we get the VFS calling
5065 * btrfs_invalidatepage() against each page of the inode. This is slow because
5066 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5067 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5068 * extent_state structures over and over, wasting lots of time.
5069 *
5070 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5071 * those expensive operations on a per page basis and do only the ordered io
5072 * finishing, while we release here the extent_map and extent_state structures,
5073 * without the excessive merging and splitting.
5074 */
5075 static void evict_inode_truncate_pages(struct inode *inode)
5076 {
5077 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5078 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5079 struct rb_node *node;
5080
5081 ASSERT(inode->i_state & I_FREEING);
5082 truncate_inode_pages_final(&inode->i_data);
5083
5084 write_lock(&map_tree->lock);
5085 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5086 struct extent_map *em;
5087
5088 node = rb_first(&map_tree->map);
5089 em = rb_entry(node, struct extent_map, rb_node);
5090 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5091 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5092 remove_extent_mapping(map_tree, em);
5093 free_extent_map(em);
5094 if (need_resched()) {
5095 write_unlock(&map_tree->lock);
5096 cond_resched();
5097 write_lock(&map_tree->lock);
5098 }
5099 }
5100 write_unlock(&map_tree->lock);
5101
5102 /*
5103 * Keep looping until we have no more ranges in the io tree.
5104 * We can have ongoing bios started by readpages (called from readahead)
5105 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5106 * still in progress (unlocked the pages in the bio but did not yet
5107 * unlocked the ranges in the io tree). Therefore this means some
5108 * ranges can still be locked and eviction started because before
5109 * submitting those bios, which are executed by a separate task (work
5110 * queue kthread), inode references (inode->i_count) were not taken
5111 * (which would be dropped in the end io callback of each bio).
5112 * Therefore here we effectively end up waiting for those bios and
5113 * anyone else holding locked ranges without having bumped the inode's
5114 * reference count - if we don't do it, when they access the inode's
5115 * io_tree to unlock a range it may be too late, leading to an
5116 * use-after-free issue.
5117 */
5118 spin_lock(&io_tree->lock);
5119 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5120 struct extent_state *state;
5121 struct extent_state *cached_state = NULL;
5122 u64 start;
5123 u64 end;
5124
5125 node = rb_first(&io_tree->state);
5126 state = rb_entry(node, struct extent_state, rb_node);
5127 start = state->start;
5128 end = state->end;
5129 spin_unlock(&io_tree->lock);
5130
5131 lock_extent_bits(io_tree, start, end, &cached_state);
5132
5133 /*
5134 * If still has DELALLOC flag, the extent didn't reach disk,
5135 * and its reserved space won't be freed by delayed_ref.
5136 * So we need to free its reserved space here.
5137 * (Refer to comment in btrfs_invalidatepage, case 2)
5138 *
5139 * Note, end is the bytenr of last byte, so we need + 1 here.
5140 */
5141 if (state->state & EXTENT_DELALLOC)
5142 btrfs_qgroup_free_data(inode, start, end - start + 1);
5143
5144 clear_extent_bit(io_tree, start, end,
5145 EXTENT_LOCKED | EXTENT_DIRTY |
5146 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5147 EXTENT_DEFRAG, 1, 1,
5148 &cached_state, GFP_NOFS);
5149
5150 cond_resched();
5151 spin_lock(&io_tree->lock);
5152 }
5153 spin_unlock(&io_tree->lock);
5154 }
5155
5156 void btrfs_evict_inode(struct inode *inode)
5157 {
5158 struct btrfs_trans_handle *trans;
5159 struct btrfs_root *root = BTRFS_I(inode)->root;
5160 struct btrfs_block_rsv *rsv, *global_rsv;
5161 int steal_from_global = 0;
5162 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5163 int ret;
5164
5165 trace_btrfs_inode_evict(inode);
5166
5167 evict_inode_truncate_pages(inode);
5168
5169 if (inode->i_nlink &&
5170 ((btrfs_root_refs(&root->root_item) != 0 &&
5171 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5172 btrfs_is_free_space_inode(inode)))
5173 goto no_delete;
5174
5175 if (is_bad_inode(inode)) {
5176 btrfs_orphan_del(NULL, inode);
5177 goto no_delete;
5178 }
5179 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5180 if (!special_file(inode->i_mode))
5181 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5182
5183 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5184
5185 if (root->fs_info->log_root_recovering) {
5186 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5187 &BTRFS_I(inode)->runtime_flags));
5188 goto no_delete;
5189 }
5190
5191 if (inode->i_nlink > 0) {
5192 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5193 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5194 goto no_delete;
5195 }
5196
5197 ret = btrfs_commit_inode_delayed_inode(inode);
5198 if (ret) {
5199 btrfs_orphan_del(NULL, inode);
5200 goto no_delete;
5201 }
5202
5203 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5204 if (!rsv) {
5205 btrfs_orphan_del(NULL, inode);
5206 goto no_delete;
5207 }
5208 rsv->size = min_size;
5209 rsv->failfast = 1;
5210 global_rsv = &root->fs_info->global_block_rsv;
5211
5212 btrfs_i_size_write(inode, 0);
5213
5214 /*
5215 * This is a bit simpler than btrfs_truncate since we've already
5216 * reserved our space for our orphan item in the unlink, so we just
5217 * need to reserve some slack space in case we add bytes and update
5218 * inode item when doing the truncate.
5219 */
5220 while (1) {
5221 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5222 BTRFS_RESERVE_FLUSH_LIMIT);
5223
5224 /*
5225 * Try and steal from the global reserve since we will
5226 * likely not use this space anyway, we want to try as
5227 * hard as possible to get this to work.
5228 */
5229 if (ret)
5230 steal_from_global++;
5231 else
5232 steal_from_global = 0;
5233 ret = 0;
5234
5235 /*
5236 * steal_from_global == 0: we reserved stuff, hooray!
5237 * steal_from_global == 1: we didn't reserve stuff, boo!
5238 * steal_from_global == 2: we've committed, still not a lot of
5239 * room but maybe we'll have room in the global reserve this
5240 * time.
5241 * steal_from_global == 3: abandon all hope!
5242 */
5243 if (steal_from_global > 2) {
5244 btrfs_warn(root->fs_info,
5245 "Could not get space for a delete, will truncate on mount %d",
5246 ret);
5247 btrfs_orphan_del(NULL, inode);
5248 btrfs_free_block_rsv(root, rsv);
5249 goto no_delete;
5250 }
5251
5252 trans = btrfs_join_transaction(root);
5253 if (IS_ERR(trans)) {
5254 btrfs_orphan_del(NULL, inode);
5255 btrfs_free_block_rsv(root, rsv);
5256 goto no_delete;
5257 }
5258
5259 /*
5260 * We can't just steal from the global reserve, we need to make
5261 * sure there is room to do it, if not we need to commit and try
5262 * again.
5263 */
5264 if (steal_from_global) {
5265 if (!btrfs_check_space_for_delayed_refs(trans, root))
5266 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5267 min_size);
5268 else
5269 ret = -ENOSPC;
5270 }
5271
5272 /*
5273 * Couldn't steal from the global reserve, we have too much
5274 * pending stuff built up, commit the transaction and try it
5275 * again.
5276 */
5277 if (ret) {
5278 ret = btrfs_commit_transaction(trans, root);
5279 if (ret) {
5280 btrfs_orphan_del(NULL, inode);
5281 btrfs_free_block_rsv(root, rsv);
5282 goto no_delete;
5283 }
5284 continue;
5285 } else {
5286 steal_from_global = 0;
5287 }
5288
5289 trans->block_rsv = rsv;
5290
5291 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5292 if (ret != -ENOSPC && ret != -EAGAIN)
5293 break;
5294
5295 trans->block_rsv = &root->fs_info->trans_block_rsv;
5296 btrfs_end_transaction(trans, root);
5297 trans = NULL;
5298 btrfs_btree_balance_dirty(root);
5299 }
5300
5301 btrfs_free_block_rsv(root, rsv);
5302
5303 /*
5304 * Errors here aren't a big deal, it just means we leave orphan items
5305 * in the tree. They will be cleaned up on the next mount.
5306 */
5307 if (ret == 0) {
5308 trans->block_rsv = root->orphan_block_rsv;
5309 btrfs_orphan_del(trans, inode);
5310 } else {
5311 btrfs_orphan_del(NULL, inode);
5312 }
5313
5314 trans->block_rsv = &root->fs_info->trans_block_rsv;
5315 if (!(root == root->fs_info->tree_root ||
5316 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5317 btrfs_return_ino(root, btrfs_ino(inode));
5318
5319 btrfs_end_transaction(trans, root);
5320 btrfs_btree_balance_dirty(root);
5321 no_delete:
5322 btrfs_remove_delayed_node(inode);
5323 clear_inode(inode);
5324 }
5325
5326 /*
5327 * this returns the key found in the dir entry in the location pointer.
5328 * If no dir entries were found, location->objectid is 0.
5329 */
5330 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5331 struct btrfs_key *location)
5332 {
5333 const char *name = dentry->d_name.name;
5334 int namelen = dentry->d_name.len;
5335 struct btrfs_dir_item *di;
5336 struct btrfs_path *path;
5337 struct btrfs_root *root = BTRFS_I(dir)->root;
5338 int ret = 0;
5339
5340 path = btrfs_alloc_path();
5341 if (!path)
5342 return -ENOMEM;
5343
5344 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5345 namelen, 0);
5346 if (IS_ERR(di))
5347 ret = PTR_ERR(di);
5348
5349 if (IS_ERR_OR_NULL(di))
5350 goto out_err;
5351
5352 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5353 out:
5354 btrfs_free_path(path);
5355 return ret;
5356 out_err:
5357 location->objectid = 0;
5358 goto out;
5359 }
5360
5361 /*
5362 * when we hit a tree root in a directory, the btrfs part of the inode
5363 * needs to be changed to reflect the root directory of the tree root. This
5364 * is kind of like crossing a mount point.
5365 */
5366 static int fixup_tree_root_location(struct btrfs_root *root,
5367 struct inode *dir,
5368 struct dentry *dentry,
5369 struct btrfs_key *location,
5370 struct btrfs_root **sub_root)
5371 {
5372 struct btrfs_path *path;
5373 struct btrfs_root *new_root;
5374 struct btrfs_root_ref *ref;
5375 struct extent_buffer *leaf;
5376 struct btrfs_key key;
5377 int ret;
5378 int err = 0;
5379
5380 path = btrfs_alloc_path();
5381 if (!path) {
5382 err = -ENOMEM;
5383 goto out;
5384 }
5385
5386 err = -ENOENT;
5387 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5388 key.type = BTRFS_ROOT_REF_KEY;
5389 key.offset = location->objectid;
5390
5391 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5392 0, 0);
5393 if (ret) {
5394 if (ret < 0)
5395 err = ret;
5396 goto out;
5397 }
5398
5399 leaf = path->nodes[0];
5400 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5401 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5402 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5403 goto out;
5404
5405 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5406 (unsigned long)(ref + 1),
5407 dentry->d_name.len);
5408 if (ret)
5409 goto out;
5410
5411 btrfs_release_path(path);
5412
5413 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5414 if (IS_ERR(new_root)) {
5415 err = PTR_ERR(new_root);
5416 goto out;
5417 }
5418
5419 *sub_root = new_root;
5420 location->objectid = btrfs_root_dirid(&new_root->root_item);
5421 location->type = BTRFS_INODE_ITEM_KEY;
5422 location->offset = 0;
5423 err = 0;
5424 out:
5425 btrfs_free_path(path);
5426 return err;
5427 }
5428
5429 static void inode_tree_add(struct inode *inode)
5430 {
5431 struct btrfs_root *root = BTRFS_I(inode)->root;
5432 struct btrfs_inode *entry;
5433 struct rb_node **p;
5434 struct rb_node *parent;
5435 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5436 u64 ino = btrfs_ino(inode);
5437
5438 if (inode_unhashed(inode))
5439 return;
5440 parent = NULL;
5441 spin_lock(&root->inode_lock);
5442 p = &root->inode_tree.rb_node;
5443 while (*p) {
5444 parent = *p;
5445 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5446
5447 if (ino < btrfs_ino(&entry->vfs_inode))
5448 p = &parent->rb_left;
5449 else if (ino > btrfs_ino(&entry->vfs_inode))
5450 p = &parent->rb_right;
5451 else {
5452 WARN_ON(!(entry->vfs_inode.i_state &
5453 (I_WILL_FREE | I_FREEING)));
5454 rb_replace_node(parent, new, &root->inode_tree);
5455 RB_CLEAR_NODE(parent);
5456 spin_unlock(&root->inode_lock);
5457 return;
5458 }
5459 }
5460 rb_link_node(new, parent, p);
5461 rb_insert_color(new, &root->inode_tree);
5462 spin_unlock(&root->inode_lock);
5463 }
5464
5465 static void inode_tree_del(struct inode *inode)
5466 {
5467 struct btrfs_root *root = BTRFS_I(inode)->root;
5468 int empty = 0;
5469
5470 spin_lock(&root->inode_lock);
5471 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5472 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5473 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5474 empty = RB_EMPTY_ROOT(&root->inode_tree);
5475 }
5476 spin_unlock(&root->inode_lock);
5477
5478 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5479 synchronize_srcu(&root->fs_info->subvol_srcu);
5480 spin_lock(&root->inode_lock);
5481 empty = RB_EMPTY_ROOT(&root->inode_tree);
5482 spin_unlock(&root->inode_lock);
5483 if (empty)
5484 btrfs_add_dead_root(root);
5485 }
5486 }
5487
5488 void btrfs_invalidate_inodes(struct btrfs_root *root)
5489 {
5490 struct rb_node *node;
5491 struct rb_node *prev;
5492 struct btrfs_inode *entry;
5493 struct inode *inode;
5494 u64 objectid = 0;
5495
5496 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5497 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5498
5499 spin_lock(&root->inode_lock);
5500 again:
5501 node = root->inode_tree.rb_node;
5502 prev = NULL;
5503 while (node) {
5504 prev = node;
5505 entry = rb_entry(node, struct btrfs_inode, rb_node);
5506
5507 if (objectid < btrfs_ino(&entry->vfs_inode))
5508 node = node->rb_left;
5509 else if (objectid > btrfs_ino(&entry->vfs_inode))
5510 node = node->rb_right;
5511 else
5512 break;
5513 }
5514 if (!node) {
5515 while (prev) {
5516 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5517 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5518 node = prev;
5519 break;
5520 }
5521 prev = rb_next(prev);
5522 }
5523 }
5524 while (node) {
5525 entry = rb_entry(node, struct btrfs_inode, rb_node);
5526 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5527 inode = igrab(&entry->vfs_inode);
5528 if (inode) {
5529 spin_unlock(&root->inode_lock);
5530 if (atomic_read(&inode->i_count) > 1)
5531 d_prune_aliases(inode);
5532 /*
5533 * btrfs_drop_inode will have it removed from
5534 * the inode cache when its usage count
5535 * hits zero.
5536 */
5537 iput(inode);
5538 cond_resched();
5539 spin_lock(&root->inode_lock);
5540 goto again;
5541 }
5542
5543 if (cond_resched_lock(&root->inode_lock))
5544 goto again;
5545
5546 node = rb_next(node);
5547 }
5548 spin_unlock(&root->inode_lock);
5549 }
5550
5551 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5552 {
5553 struct btrfs_iget_args *args = p;
5554 inode->i_ino = args->location->objectid;
5555 memcpy(&BTRFS_I(inode)->location, args->location,
5556 sizeof(*args->location));
5557 BTRFS_I(inode)->root = args->root;
5558 return 0;
5559 }
5560
5561 static int btrfs_find_actor(struct inode *inode, void *opaque)
5562 {
5563 struct btrfs_iget_args *args = opaque;
5564 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5565 args->root == BTRFS_I(inode)->root;
5566 }
5567
5568 static struct inode *btrfs_iget_locked(struct super_block *s,
5569 struct btrfs_key *location,
5570 struct btrfs_root *root)
5571 {
5572 struct inode *inode;
5573 struct btrfs_iget_args args;
5574 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5575
5576 args.location = location;
5577 args.root = root;
5578
5579 inode = iget5_locked(s, hashval, btrfs_find_actor,
5580 btrfs_init_locked_inode,
5581 (void *)&args);
5582 return inode;
5583 }
5584
5585 /* Get an inode object given its location and corresponding root.
5586 * Returns in *is_new if the inode was read from disk
5587 */
5588 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5589 struct btrfs_root *root, int *new)
5590 {
5591 struct inode *inode;
5592
5593 inode = btrfs_iget_locked(s, location, root);
5594 if (!inode)
5595 return ERR_PTR(-ENOMEM);
5596
5597 if (inode->i_state & I_NEW) {
5598 btrfs_read_locked_inode(inode);
5599 if (!is_bad_inode(inode)) {
5600 inode_tree_add(inode);
5601 unlock_new_inode(inode);
5602 if (new)
5603 *new = 1;
5604 } else {
5605 unlock_new_inode(inode);
5606 iput(inode);
5607 inode = ERR_PTR(-ESTALE);
5608 }
5609 }
5610
5611 return inode;
5612 }
5613
5614 static struct inode *new_simple_dir(struct super_block *s,
5615 struct btrfs_key *key,
5616 struct btrfs_root *root)
5617 {
5618 struct inode *inode = new_inode(s);
5619
5620 if (!inode)
5621 return ERR_PTR(-ENOMEM);
5622
5623 BTRFS_I(inode)->root = root;
5624 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5625 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5626
5627 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5628 inode->i_op = &btrfs_dir_ro_inode_operations;
5629 inode->i_fop = &simple_dir_operations;
5630 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5631 inode->i_mtime = current_fs_time(inode->i_sb);
5632 inode->i_atime = inode->i_mtime;
5633 inode->i_ctime = inode->i_mtime;
5634 BTRFS_I(inode)->i_otime = inode->i_mtime;
5635
5636 return inode;
5637 }
5638
5639 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5640 {
5641 struct inode *inode;
5642 struct btrfs_root *root = BTRFS_I(dir)->root;
5643 struct btrfs_root *sub_root = root;
5644 struct btrfs_key location;
5645 int index;
5646 int ret = 0;
5647
5648 if (dentry->d_name.len > BTRFS_NAME_LEN)
5649 return ERR_PTR(-ENAMETOOLONG);
5650
5651 ret = btrfs_inode_by_name(dir, dentry, &location);
5652 if (ret < 0)
5653 return ERR_PTR(ret);
5654
5655 if (location.objectid == 0)
5656 return ERR_PTR(-ENOENT);
5657
5658 if (location.type == BTRFS_INODE_ITEM_KEY) {
5659 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5660 return inode;
5661 }
5662
5663 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5664
5665 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5666 ret = fixup_tree_root_location(root, dir, dentry,
5667 &location, &sub_root);
5668 if (ret < 0) {
5669 if (ret != -ENOENT)
5670 inode = ERR_PTR(ret);
5671 else
5672 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5673 } else {
5674 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5675 }
5676 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5677
5678 if (!IS_ERR(inode) && root != sub_root) {
5679 down_read(&root->fs_info->cleanup_work_sem);
5680 if (!(inode->i_sb->s_flags & MS_RDONLY))
5681 ret = btrfs_orphan_cleanup(sub_root);
5682 up_read(&root->fs_info->cleanup_work_sem);
5683 if (ret) {
5684 iput(inode);
5685 inode = ERR_PTR(ret);
5686 }
5687 }
5688
5689 return inode;
5690 }
5691
5692 static int btrfs_dentry_delete(const struct dentry *dentry)
5693 {
5694 struct btrfs_root *root;
5695 struct inode *inode = d_inode(dentry);
5696
5697 if (!inode && !IS_ROOT(dentry))
5698 inode = d_inode(dentry->d_parent);
5699
5700 if (inode) {
5701 root = BTRFS_I(inode)->root;
5702 if (btrfs_root_refs(&root->root_item) == 0)
5703 return 1;
5704
5705 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5706 return 1;
5707 }
5708 return 0;
5709 }
5710
5711 static void btrfs_dentry_release(struct dentry *dentry)
5712 {
5713 kfree(dentry->d_fsdata);
5714 }
5715
5716 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5717 unsigned int flags)
5718 {
5719 struct inode *inode;
5720
5721 inode = btrfs_lookup_dentry(dir, dentry);
5722 if (IS_ERR(inode)) {
5723 if (PTR_ERR(inode) == -ENOENT)
5724 inode = NULL;
5725 else
5726 return ERR_CAST(inode);
5727 }
5728
5729 return d_splice_alias(inode, dentry);
5730 }
5731
5732 unsigned char btrfs_filetype_table[] = {
5733 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5734 };
5735
5736 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5737 {
5738 struct inode *inode = file_inode(file);
5739 struct btrfs_root *root = BTRFS_I(inode)->root;
5740 struct btrfs_item *item;
5741 struct btrfs_dir_item *di;
5742 struct btrfs_key key;
5743 struct btrfs_key found_key;
5744 struct btrfs_path *path;
5745 struct list_head ins_list;
5746 struct list_head del_list;
5747 int ret;
5748 struct extent_buffer *leaf;
5749 int slot;
5750 unsigned char d_type;
5751 int over = 0;
5752 u32 di_cur;
5753 u32 di_total;
5754 u32 di_len;
5755 int key_type = BTRFS_DIR_INDEX_KEY;
5756 char tmp_name[32];
5757 char *name_ptr;
5758 int name_len;
5759 int is_curr = 0; /* ctx->pos points to the current index? */
5760 bool emitted;
5761
5762 /* FIXME, use a real flag for deciding about the key type */
5763 if (root->fs_info->tree_root == root)
5764 key_type = BTRFS_DIR_ITEM_KEY;
5765
5766 if (!dir_emit_dots(file, ctx))
5767 return 0;
5768
5769 path = btrfs_alloc_path();
5770 if (!path)
5771 return -ENOMEM;
5772
5773 path->reada = READA_FORWARD;
5774
5775 if (key_type == BTRFS_DIR_INDEX_KEY) {
5776 INIT_LIST_HEAD(&ins_list);
5777 INIT_LIST_HEAD(&del_list);
5778 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5779 }
5780
5781 key.type = key_type;
5782 key.offset = ctx->pos;
5783 key.objectid = btrfs_ino(inode);
5784
5785 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5786 if (ret < 0)
5787 goto err;
5788
5789 emitted = false;
5790 while (1) {
5791 leaf = path->nodes[0];
5792 slot = path->slots[0];
5793 if (slot >= btrfs_header_nritems(leaf)) {
5794 ret = btrfs_next_leaf(root, path);
5795 if (ret < 0)
5796 goto err;
5797 else if (ret > 0)
5798 break;
5799 continue;
5800 }
5801
5802 item = btrfs_item_nr(slot);
5803 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5804
5805 if (found_key.objectid != key.objectid)
5806 break;
5807 if (found_key.type != key_type)
5808 break;
5809 if (found_key.offset < ctx->pos)
5810 goto next;
5811 if (key_type == BTRFS_DIR_INDEX_KEY &&
5812 btrfs_should_delete_dir_index(&del_list,
5813 found_key.offset))
5814 goto next;
5815
5816 ctx->pos = found_key.offset;
5817 is_curr = 1;
5818
5819 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5820 di_cur = 0;
5821 di_total = btrfs_item_size(leaf, item);
5822
5823 while (di_cur < di_total) {
5824 struct btrfs_key location;
5825
5826 if (verify_dir_item(root, leaf, di))
5827 break;
5828
5829 name_len = btrfs_dir_name_len(leaf, di);
5830 if (name_len <= sizeof(tmp_name)) {
5831 name_ptr = tmp_name;
5832 } else {
5833 name_ptr = kmalloc(name_len, GFP_KERNEL);
5834 if (!name_ptr) {
5835 ret = -ENOMEM;
5836 goto err;
5837 }
5838 }
5839 read_extent_buffer(leaf, name_ptr,
5840 (unsigned long)(di + 1), name_len);
5841
5842 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5843 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5844
5845
5846 /* is this a reference to our own snapshot? If so
5847 * skip it.
5848 *
5849 * In contrast to old kernels, we insert the snapshot's
5850 * dir item and dir index after it has been created, so
5851 * we won't find a reference to our own snapshot. We
5852 * still keep the following code for backward
5853 * compatibility.
5854 */
5855 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5856 location.objectid == root->root_key.objectid) {
5857 over = 0;
5858 goto skip;
5859 }
5860 over = !dir_emit(ctx, name_ptr, name_len,
5861 location.objectid, d_type);
5862
5863 skip:
5864 if (name_ptr != tmp_name)
5865 kfree(name_ptr);
5866
5867 if (over)
5868 goto nopos;
5869 emitted = true;
5870 di_len = btrfs_dir_name_len(leaf, di) +
5871 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5872 di_cur += di_len;
5873 di = (struct btrfs_dir_item *)((char *)di + di_len);
5874 }
5875 next:
5876 path->slots[0]++;
5877 }
5878
5879 if (key_type == BTRFS_DIR_INDEX_KEY) {
5880 if (is_curr)
5881 ctx->pos++;
5882 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted);
5883 if (ret)
5884 goto nopos;
5885 }
5886
5887 /*
5888 * If we haven't emitted any dir entry, we must not touch ctx->pos as
5889 * it was was set to the termination value in previous call. We assume
5890 * that "." and ".." were emitted if we reach this point and set the
5891 * termination value as well for an empty directory.
5892 */
5893 if (ctx->pos > 2 && !emitted)
5894 goto nopos;
5895
5896 /* Reached end of directory/root. Bump pos past the last item. */
5897 ctx->pos++;
5898
5899 /*
5900 * Stop new entries from being returned after we return the last
5901 * entry.
5902 *
5903 * New directory entries are assigned a strictly increasing
5904 * offset. This means that new entries created during readdir
5905 * are *guaranteed* to be seen in the future by that readdir.
5906 * This has broken buggy programs which operate on names as
5907 * they're returned by readdir. Until we re-use freed offsets
5908 * we have this hack to stop new entries from being returned
5909 * under the assumption that they'll never reach this huge
5910 * offset.
5911 *
5912 * This is being careful not to overflow 32bit loff_t unless the
5913 * last entry requires it because doing so has broken 32bit apps
5914 * in the past.
5915 */
5916 if (key_type == BTRFS_DIR_INDEX_KEY) {
5917 if (ctx->pos >= INT_MAX)
5918 ctx->pos = LLONG_MAX;
5919 else
5920 ctx->pos = INT_MAX;
5921 }
5922 nopos:
5923 ret = 0;
5924 err:
5925 if (key_type == BTRFS_DIR_INDEX_KEY)
5926 btrfs_put_delayed_items(&ins_list, &del_list);
5927 btrfs_free_path(path);
5928 return ret;
5929 }
5930
5931 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5932 {
5933 struct btrfs_root *root = BTRFS_I(inode)->root;
5934 struct btrfs_trans_handle *trans;
5935 int ret = 0;
5936 bool nolock = false;
5937
5938 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5939 return 0;
5940
5941 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5942 nolock = true;
5943
5944 if (wbc->sync_mode == WB_SYNC_ALL) {
5945 if (nolock)
5946 trans = btrfs_join_transaction_nolock(root);
5947 else
5948 trans = btrfs_join_transaction(root);
5949 if (IS_ERR(trans))
5950 return PTR_ERR(trans);
5951 ret = btrfs_commit_transaction(trans, root);
5952 }
5953 return ret;
5954 }
5955
5956 /*
5957 * This is somewhat expensive, updating the tree every time the
5958 * inode changes. But, it is most likely to find the inode in cache.
5959 * FIXME, needs more benchmarking...there are no reasons other than performance
5960 * to keep or drop this code.
5961 */
5962 static int btrfs_dirty_inode(struct inode *inode)
5963 {
5964 struct btrfs_root *root = BTRFS_I(inode)->root;
5965 struct btrfs_trans_handle *trans;
5966 int ret;
5967
5968 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5969 return 0;
5970
5971 trans = btrfs_join_transaction(root);
5972 if (IS_ERR(trans))
5973 return PTR_ERR(trans);
5974
5975 ret = btrfs_update_inode(trans, root, inode);
5976 if (ret && ret == -ENOSPC) {
5977 /* whoops, lets try again with the full transaction */
5978 btrfs_end_transaction(trans, root);
5979 trans = btrfs_start_transaction(root, 1);
5980 if (IS_ERR(trans))
5981 return PTR_ERR(trans);
5982
5983 ret = btrfs_update_inode(trans, root, inode);
5984 }
5985 btrfs_end_transaction(trans, root);
5986 if (BTRFS_I(inode)->delayed_node)
5987 btrfs_balance_delayed_items(root);
5988
5989 return ret;
5990 }
5991
5992 /*
5993 * This is a copy of file_update_time. We need this so we can return error on
5994 * ENOSPC for updating the inode in the case of file write and mmap writes.
5995 */
5996 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5997 int flags)
5998 {
5999 struct btrfs_root *root = BTRFS_I(inode)->root;
6000
6001 if (btrfs_root_readonly(root))
6002 return -EROFS;
6003
6004 if (flags & S_VERSION)
6005 inode_inc_iversion(inode);
6006 if (flags & S_CTIME)
6007 inode->i_ctime = *now;
6008 if (flags & S_MTIME)
6009 inode->i_mtime = *now;
6010 if (flags & S_ATIME)
6011 inode->i_atime = *now;
6012 return btrfs_dirty_inode(inode);
6013 }
6014
6015 /*
6016 * find the highest existing sequence number in a directory
6017 * and then set the in-memory index_cnt variable to reflect
6018 * free sequence numbers
6019 */
6020 static int btrfs_set_inode_index_count(struct inode *inode)
6021 {
6022 struct btrfs_root *root = BTRFS_I(inode)->root;
6023 struct btrfs_key key, found_key;
6024 struct btrfs_path *path;
6025 struct extent_buffer *leaf;
6026 int ret;
6027
6028 key.objectid = btrfs_ino(inode);
6029 key.type = BTRFS_DIR_INDEX_KEY;
6030 key.offset = (u64)-1;
6031
6032 path = btrfs_alloc_path();
6033 if (!path)
6034 return -ENOMEM;
6035
6036 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6037 if (ret < 0)
6038 goto out;
6039 /* FIXME: we should be able to handle this */
6040 if (ret == 0)
6041 goto out;
6042 ret = 0;
6043
6044 /*
6045 * MAGIC NUMBER EXPLANATION:
6046 * since we search a directory based on f_pos we have to start at 2
6047 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6048 * else has to start at 2
6049 */
6050 if (path->slots[0] == 0) {
6051 BTRFS_I(inode)->index_cnt = 2;
6052 goto out;
6053 }
6054
6055 path->slots[0]--;
6056
6057 leaf = path->nodes[0];
6058 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6059
6060 if (found_key.objectid != btrfs_ino(inode) ||
6061 found_key.type != BTRFS_DIR_INDEX_KEY) {
6062 BTRFS_I(inode)->index_cnt = 2;
6063 goto out;
6064 }
6065
6066 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6067 out:
6068 btrfs_free_path(path);
6069 return ret;
6070 }
6071
6072 /*
6073 * helper to find a free sequence number in a given directory. This current
6074 * code is very simple, later versions will do smarter things in the btree
6075 */
6076 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6077 {
6078 int ret = 0;
6079
6080 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6081 ret = btrfs_inode_delayed_dir_index_count(dir);
6082 if (ret) {
6083 ret = btrfs_set_inode_index_count(dir);
6084 if (ret)
6085 return ret;
6086 }
6087 }
6088
6089 *index = BTRFS_I(dir)->index_cnt;
6090 BTRFS_I(dir)->index_cnt++;
6091
6092 return ret;
6093 }
6094
6095 static int btrfs_insert_inode_locked(struct inode *inode)
6096 {
6097 struct btrfs_iget_args args;
6098 args.location = &BTRFS_I(inode)->location;
6099 args.root = BTRFS_I(inode)->root;
6100
6101 return insert_inode_locked4(inode,
6102 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6103 btrfs_find_actor, &args);
6104 }
6105
6106 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6107 struct btrfs_root *root,
6108 struct inode *dir,
6109 const char *name, int name_len,
6110 u64 ref_objectid, u64 objectid,
6111 umode_t mode, u64 *index)
6112 {
6113 struct inode *inode;
6114 struct btrfs_inode_item *inode_item;
6115 struct btrfs_key *location;
6116 struct btrfs_path *path;
6117 struct btrfs_inode_ref *ref;
6118 struct btrfs_key key[2];
6119 u32 sizes[2];
6120 int nitems = name ? 2 : 1;
6121 unsigned long ptr;
6122 int ret;
6123
6124 path = btrfs_alloc_path();
6125 if (!path)
6126 return ERR_PTR(-ENOMEM);
6127
6128 inode = new_inode(root->fs_info->sb);
6129 if (!inode) {
6130 btrfs_free_path(path);
6131 return ERR_PTR(-ENOMEM);
6132 }
6133
6134 /*
6135 * O_TMPFILE, set link count to 0, so that after this point,
6136 * we fill in an inode item with the correct link count.
6137 */
6138 if (!name)
6139 set_nlink(inode, 0);
6140
6141 /*
6142 * we have to initialize this early, so we can reclaim the inode
6143 * number if we fail afterwards in this function.
6144 */
6145 inode->i_ino = objectid;
6146
6147 if (dir && name) {
6148 trace_btrfs_inode_request(dir);
6149
6150 ret = btrfs_set_inode_index(dir, index);
6151 if (ret) {
6152 btrfs_free_path(path);
6153 iput(inode);
6154 return ERR_PTR(ret);
6155 }
6156 } else if (dir) {
6157 *index = 0;
6158 }
6159 /*
6160 * index_cnt is ignored for everything but a dir,
6161 * btrfs_get_inode_index_count has an explanation for the magic
6162 * number
6163 */
6164 BTRFS_I(inode)->index_cnt = 2;
6165 BTRFS_I(inode)->dir_index = *index;
6166 BTRFS_I(inode)->root = root;
6167 BTRFS_I(inode)->generation = trans->transid;
6168 inode->i_generation = BTRFS_I(inode)->generation;
6169
6170 /*
6171 * We could have gotten an inode number from somebody who was fsynced
6172 * and then removed in this same transaction, so let's just set full
6173 * sync since it will be a full sync anyway and this will blow away the
6174 * old info in the log.
6175 */
6176 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6177
6178 key[0].objectid = objectid;
6179 key[0].type = BTRFS_INODE_ITEM_KEY;
6180 key[0].offset = 0;
6181
6182 sizes[0] = sizeof(struct btrfs_inode_item);
6183
6184 if (name) {
6185 /*
6186 * Start new inodes with an inode_ref. This is slightly more
6187 * efficient for small numbers of hard links since they will
6188 * be packed into one item. Extended refs will kick in if we
6189 * add more hard links than can fit in the ref item.
6190 */
6191 key[1].objectid = objectid;
6192 key[1].type = BTRFS_INODE_REF_KEY;
6193 key[1].offset = ref_objectid;
6194
6195 sizes[1] = name_len + sizeof(*ref);
6196 }
6197
6198 location = &BTRFS_I(inode)->location;
6199 location->objectid = objectid;
6200 location->offset = 0;
6201 location->type = BTRFS_INODE_ITEM_KEY;
6202
6203 ret = btrfs_insert_inode_locked(inode);
6204 if (ret < 0)
6205 goto fail;
6206
6207 path->leave_spinning = 1;
6208 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6209 if (ret != 0)
6210 goto fail_unlock;
6211
6212 inode_init_owner(inode, dir, mode);
6213 inode_set_bytes(inode, 0);
6214
6215 inode->i_mtime = current_fs_time(inode->i_sb);
6216 inode->i_atime = inode->i_mtime;
6217 inode->i_ctime = inode->i_mtime;
6218 BTRFS_I(inode)->i_otime = inode->i_mtime;
6219
6220 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6221 struct btrfs_inode_item);
6222 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6223 sizeof(*inode_item));
6224 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6225
6226 if (name) {
6227 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6228 struct btrfs_inode_ref);
6229 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6230 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6231 ptr = (unsigned long)(ref + 1);
6232 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6233 }
6234
6235 btrfs_mark_buffer_dirty(path->nodes[0]);
6236 btrfs_free_path(path);
6237
6238 btrfs_inherit_iflags(inode, dir);
6239
6240 if (S_ISREG(mode)) {
6241 if (btrfs_test_opt(root, NODATASUM))
6242 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6243 if (btrfs_test_opt(root, NODATACOW))
6244 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6245 BTRFS_INODE_NODATASUM;
6246 }
6247
6248 inode_tree_add(inode);
6249
6250 trace_btrfs_inode_new(inode);
6251 btrfs_set_inode_last_trans(trans, inode);
6252
6253 btrfs_update_root_times(trans, root);
6254
6255 ret = btrfs_inode_inherit_props(trans, inode, dir);
6256 if (ret)
6257 btrfs_err(root->fs_info,
6258 "error inheriting props for ino %llu (root %llu): %d",
6259 btrfs_ino(inode), root->root_key.objectid, ret);
6260
6261 return inode;
6262
6263 fail_unlock:
6264 unlock_new_inode(inode);
6265 fail:
6266 if (dir && name)
6267 BTRFS_I(dir)->index_cnt--;
6268 btrfs_free_path(path);
6269 iput(inode);
6270 return ERR_PTR(ret);
6271 }
6272
6273 static inline u8 btrfs_inode_type(struct inode *inode)
6274 {
6275 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6276 }
6277
6278 /*
6279 * utility function to add 'inode' into 'parent_inode' with
6280 * a give name and a given sequence number.
6281 * if 'add_backref' is true, also insert a backref from the
6282 * inode to the parent directory.
6283 */
6284 int btrfs_add_link(struct btrfs_trans_handle *trans,
6285 struct inode *parent_inode, struct inode *inode,
6286 const char *name, int name_len, int add_backref, u64 index)
6287 {
6288 int ret = 0;
6289 struct btrfs_key key;
6290 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6291 u64 ino = btrfs_ino(inode);
6292 u64 parent_ino = btrfs_ino(parent_inode);
6293
6294 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6295 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6296 } else {
6297 key.objectid = ino;
6298 key.type = BTRFS_INODE_ITEM_KEY;
6299 key.offset = 0;
6300 }
6301
6302 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6303 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6304 key.objectid, root->root_key.objectid,
6305 parent_ino, index, name, name_len);
6306 } else if (add_backref) {
6307 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6308 parent_ino, index);
6309 }
6310
6311 /* Nothing to clean up yet */
6312 if (ret)
6313 return ret;
6314
6315 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6316 parent_inode, &key,
6317 btrfs_inode_type(inode), index);
6318 if (ret == -EEXIST || ret == -EOVERFLOW)
6319 goto fail_dir_item;
6320 else if (ret) {
6321 btrfs_abort_transaction(trans, root, ret);
6322 return ret;
6323 }
6324
6325 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6326 name_len * 2);
6327 inode_inc_iversion(parent_inode);
6328 parent_inode->i_mtime = parent_inode->i_ctime =
6329 current_fs_time(parent_inode->i_sb);
6330 ret = btrfs_update_inode(trans, root, parent_inode);
6331 if (ret)
6332 btrfs_abort_transaction(trans, root, ret);
6333 return ret;
6334
6335 fail_dir_item:
6336 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6337 u64 local_index;
6338 int err;
6339 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6340 key.objectid, root->root_key.objectid,
6341 parent_ino, &local_index, name, name_len);
6342
6343 } else if (add_backref) {
6344 u64 local_index;
6345 int err;
6346
6347 err = btrfs_del_inode_ref(trans, root, name, name_len,
6348 ino, parent_ino, &local_index);
6349 }
6350 return ret;
6351 }
6352
6353 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6354 struct inode *dir, struct dentry *dentry,
6355 struct inode *inode, int backref, u64 index)
6356 {
6357 int err = btrfs_add_link(trans, dir, inode,
6358 dentry->d_name.name, dentry->d_name.len,
6359 backref, index);
6360 if (err > 0)
6361 err = -EEXIST;
6362 return err;
6363 }
6364
6365 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6366 umode_t mode, dev_t rdev)
6367 {
6368 struct btrfs_trans_handle *trans;
6369 struct btrfs_root *root = BTRFS_I(dir)->root;
6370 struct inode *inode = NULL;
6371 int err;
6372 int drop_inode = 0;
6373 u64 objectid;
6374 u64 index = 0;
6375
6376 /*
6377 * 2 for inode item and ref
6378 * 2 for dir items
6379 * 1 for xattr if selinux is on
6380 */
6381 trans = btrfs_start_transaction(root, 5);
6382 if (IS_ERR(trans))
6383 return PTR_ERR(trans);
6384
6385 err = btrfs_find_free_ino(root, &objectid);
6386 if (err)
6387 goto out_unlock;
6388
6389 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6390 dentry->d_name.len, btrfs_ino(dir), objectid,
6391 mode, &index);
6392 if (IS_ERR(inode)) {
6393 err = PTR_ERR(inode);
6394 goto out_unlock;
6395 }
6396
6397 /*
6398 * If the active LSM wants to access the inode during
6399 * d_instantiate it needs these. Smack checks to see
6400 * if the filesystem supports xattrs by looking at the
6401 * ops vector.
6402 */
6403 inode->i_op = &btrfs_special_inode_operations;
6404 init_special_inode(inode, inode->i_mode, rdev);
6405
6406 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6407 if (err)
6408 goto out_unlock_inode;
6409
6410 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6411 if (err) {
6412 goto out_unlock_inode;
6413 } else {
6414 btrfs_update_inode(trans, root, inode);
6415 unlock_new_inode(inode);
6416 d_instantiate(dentry, inode);
6417 }
6418
6419 out_unlock:
6420 btrfs_end_transaction(trans, root);
6421 btrfs_balance_delayed_items(root);
6422 btrfs_btree_balance_dirty(root);
6423 if (drop_inode) {
6424 inode_dec_link_count(inode);
6425 iput(inode);
6426 }
6427 return err;
6428
6429 out_unlock_inode:
6430 drop_inode = 1;
6431 unlock_new_inode(inode);
6432 goto out_unlock;
6433
6434 }
6435
6436 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6437 umode_t mode, bool excl)
6438 {
6439 struct btrfs_trans_handle *trans;
6440 struct btrfs_root *root = BTRFS_I(dir)->root;
6441 struct inode *inode = NULL;
6442 int drop_inode_on_err = 0;
6443 int err;
6444 u64 objectid;
6445 u64 index = 0;
6446
6447 /*
6448 * 2 for inode item and ref
6449 * 2 for dir items
6450 * 1 for xattr if selinux is on
6451 */
6452 trans = btrfs_start_transaction(root, 5);
6453 if (IS_ERR(trans))
6454 return PTR_ERR(trans);
6455
6456 err = btrfs_find_free_ino(root, &objectid);
6457 if (err)
6458 goto out_unlock;
6459
6460 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6461 dentry->d_name.len, btrfs_ino(dir), objectid,
6462 mode, &index);
6463 if (IS_ERR(inode)) {
6464 err = PTR_ERR(inode);
6465 goto out_unlock;
6466 }
6467 drop_inode_on_err = 1;
6468 /*
6469 * If the active LSM wants to access the inode during
6470 * d_instantiate it needs these. Smack checks to see
6471 * if the filesystem supports xattrs by looking at the
6472 * ops vector.
6473 */
6474 inode->i_fop = &btrfs_file_operations;
6475 inode->i_op = &btrfs_file_inode_operations;
6476 inode->i_mapping->a_ops = &btrfs_aops;
6477
6478 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6479 if (err)
6480 goto out_unlock_inode;
6481
6482 err = btrfs_update_inode(trans, root, inode);
6483 if (err)
6484 goto out_unlock_inode;
6485
6486 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6487 if (err)
6488 goto out_unlock_inode;
6489
6490 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6491 unlock_new_inode(inode);
6492 d_instantiate(dentry, inode);
6493
6494 out_unlock:
6495 btrfs_end_transaction(trans, root);
6496 if (err && drop_inode_on_err) {
6497 inode_dec_link_count(inode);
6498 iput(inode);
6499 }
6500 btrfs_balance_delayed_items(root);
6501 btrfs_btree_balance_dirty(root);
6502 return err;
6503
6504 out_unlock_inode:
6505 unlock_new_inode(inode);
6506 goto out_unlock;
6507
6508 }
6509
6510 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6511 struct dentry *dentry)
6512 {
6513 struct btrfs_trans_handle *trans = NULL;
6514 struct btrfs_root *root = BTRFS_I(dir)->root;
6515 struct inode *inode = d_inode(old_dentry);
6516 u64 index;
6517 int err;
6518 int drop_inode = 0;
6519
6520 /* do not allow sys_link's with other subvols of the same device */
6521 if (root->objectid != BTRFS_I(inode)->root->objectid)
6522 return -EXDEV;
6523
6524 if (inode->i_nlink >= BTRFS_LINK_MAX)
6525 return -EMLINK;
6526
6527 err = btrfs_set_inode_index(dir, &index);
6528 if (err)
6529 goto fail;
6530
6531 /*
6532 * 2 items for inode and inode ref
6533 * 2 items for dir items
6534 * 1 item for parent inode
6535 */
6536 trans = btrfs_start_transaction(root, 5);
6537 if (IS_ERR(trans)) {
6538 err = PTR_ERR(trans);
6539 trans = NULL;
6540 goto fail;
6541 }
6542
6543 /* There are several dir indexes for this inode, clear the cache. */
6544 BTRFS_I(inode)->dir_index = 0ULL;
6545 inc_nlink(inode);
6546 inode_inc_iversion(inode);
6547 inode->i_ctime = current_fs_time(inode->i_sb);
6548 ihold(inode);
6549 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6550
6551 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6552
6553 if (err) {
6554 drop_inode = 1;
6555 } else {
6556 struct dentry *parent = dentry->d_parent;
6557 err = btrfs_update_inode(trans, root, inode);
6558 if (err)
6559 goto fail;
6560 if (inode->i_nlink == 1) {
6561 /*
6562 * If new hard link count is 1, it's a file created
6563 * with open(2) O_TMPFILE flag.
6564 */
6565 err = btrfs_orphan_del(trans, inode);
6566 if (err)
6567 goto fail;
6568 }
6569 d_instantiate(dentry, inode);
6570 btrfs_log_new_name(trans, inode, NULL, parent);
6571 }
6572
6573 btrfs_balance_delayed_items(root);
6574 fail:
6575 if (trans)
6576 btrfs_end_transaction(trans, root);
6577 if (drop_inode) {
6578 inode_dec_link_count(inode);
6579 iput(inode);
6580 }
6581 btrfs_btree_balance_dirty(root);
6582 return err;
6583 }
6584
6585 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6586 {
6587 struct inode *inode = NULL;
6588 struct btrfs_trans_handle *trans;
6589 struct btrfs_root *root = BTRFS_I(dir)->root;
6590 int err = 0;
6591 int drop_on_err = 0;
6592 u64 objectid = 0;
6593 u64 index = 0;
6594
6595 /*
6596 * 2 items for inode and ref
6597 * 2 items for dir items
6598 * 1 for xattr if selinux is on
6599 */
6600 trans = btrfs_start_transaction(root, 5);
6601 if (IS_ERR(trans))
6602 return PTR_ERR(trans);
6603
6604 err = btrfs_find_free_ino(root, &objectid);
6605 if (err)
6606 goto out_fail;
6607
6608 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6609 dentry->d_name.len, btrfs_ino(dir), objectid,
6610 S_IFDIR | mode, &index);
6611 if (IS_ERR(inode)) {
6612 err = PTR_ERR(inode);
6613 goto out_fail;
6614 }
6615
6616 drop_on_err = 1;
6617 /* these must be set before we unlock the inode */
6618 inode->i_op = &btrfs_dir_inode_operations;
6619 inode->i_fop = &btrfs_dir_file_operations;
6620
6621 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6622 if (err)
6623 goto out_fail_inode;
6624
6625 btrfs_i_size_write(inode, 0);
6626 err = btrfs_update_inode(trans, root, inode);
6627 if (err)
6628 goto out_fail_inode;
6629
6630 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6631 dentry->d_name.len, 0, index);
6632 if (err)
6633 goto out_fail_inode;
6634
6635 d_instantiate(dentry, inode);
6636 /*
6637 * mkdir is special. We're unlocking after we call d_instantiate
6638 * to avoid a race with nfsd calling d_instantiate.
6639 */
6640 unlock_new_inode(inode);
6641 drop_on_err = 0;
6642
6643 out_fail:
6644 btrfs_end_transaction(trans, root);
6645 if (drop_on_err) {
6646 inode_dec_link_count(inode);
6647 iput(inode);
6648 }
6649 btrfs_balance_delayed_items(root);
6650 btrfs_btree_balance_dirty(root);
6651 return err;
6652
6653 out_fail_inode:
6654 unlock_new_inode(inode);
6655 goto out_fail;
6656 }
6657
6658 /* Find next extent map of a given extent map, caller needs to ensure locks */
6659 static struct extent_map *next_extent_map(struct extent_map *em)
6660 {
6661 struct rb_node *next;
6662
6663 next = rb_next(&em->rb_node);
6664 if (!next)
6665 return NULL;
6666 return container_of(next, struct extent_map, rb_node);
6667 }
6668
6669 static struct extent_map *prev_extent_map(struct extent_map *em)
6670 {
6671 struct rb_node *prev;
6672
6673 prev = rb_prev(&em->rb_node);
6674 if (!prev)
6675 return NULL;
6676 return container_of(prev, struct extent_map, rb_node);
6677 }
6678
6679 /* helper for btfs_get_extent. Given an existing extent in the tree,
6680 * the existing extent is the nearest extent to map_start,
6681 * and an extent that you want to insert, deal with overlap and insert
6682 * the best fitted new extent into the tree.
6683 */
6684 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6685 struct extent_map *existing,
6686 struct extent_map *em,
6687 u64 map_start)
6688 {
6689 struct extent_map *prev;
6690 struct extent_map *next;
6691 u64 start;
6692 u64 end;
6693 u64 start_diff;
6694
6695 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6696
6697 if (existing->start > map_start) {
6698 next = existing;
6699 prev = prev_extent_map(next);
6700 } else {
6701 prev = existing;
6702 next = next_extent_map(prev);
6703 }
6704
6705 start = prev ? extent_map_end(prev) : em->start;
6706 start = max_t(u64, start, em->start);
6707 end = next ? next->start : extent_map_end(em);
6708 end = min_t(u64, end, extent_map_end(em));
6709 start_diff = start - em->start;
6710 em->start = start;
6711 em->len = end - start;
6712 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6713 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6714 em->block_start += start_diff;
6715 em->block_len -= start_diff;
6716 }
6717 return add_extent_mapping(em_tree, em, 0);
6718 }
6719
6720 static noinline int uncompress_inline(struct btrfs_path *path,
6721 struct page *page,
6722 size_t pg_offset, u64 extent_offset,
6723 struct btrfs_file_extent_item *item)
6724 {
6725 int ret;
6726 struct extent_buffer *leaf = path->nodes[0];
6727 char *tmp;
6728 size_t max_size;
6729 unsigned long inline_size;
6730 unsigned long ptr;
6731 int compress_type;
6732
6733 WARN_ON(pg_offset != 0);
6734 compress_type = btrfs_file_extent_compression(leaf, item);
6735 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6736 inline_size = btrfs_file_extent_inline_item_len(leaf,
6737 btrfs_item_nr(path->slots[0]));
6738 tmp = kmalloc(inline_size, GFP_NOFS);
6739 if (!tmp)
6740 return -ENOMEM;
6741 ptr = btrfs_file_extent_inline_start(item);
6742
6743 read_extent_buffer(leaf, tmp, ptr, inline_size);
6744
6745 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6746 ret = btrfs_decompress(compress_type, tmp, page,
6747 extent_offset, inline_size, max_size);
6748 kfree(tmp);
6749 return ret;
6750 }
6751
6752 /*
6753 * a bit scary, this does extent mapping from logical file offset to the disk.
6754 * the ugly parts come from merging extents from the disk with the in-ram
6755 * representation. This gets more complex because of the data=ordered code,
6756 * where the in-ram extents might be locked pending data=ordered completion.
6757 *
6758 * This also copies inline extents directly into the page.
6759 */
6760
6761 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6762 size_t pg_offset, u64 start, u64 len,
6763 int create)
6764 {
6765 int ret;
6766 int err = 0;
6767 u64 extent_start = 0;
6768 u64 extent_end = 0;
6769 u64 objectid = btrfs_ino(inode);
6770 u32 found_type;
6771 struct btrfs_path *path = NULL;
6772 struct btrfs_root *root = BTRFS_I(inode)->root;
6773 struct btrfs_file_extent_item *item;
6774 struct extent_buffer *leaf;
6775 struct btrfs_key found_key;
6776 struct extent_map *em = NULL;
6777 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6778 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6779 struct btrfs_trans_handle *trans = NULL;
6780 const bool new_inline = !page || create;
6781
6782 again:
6783 read_lock(&em_tree->lock);
6784 em = lookup_extent_mapping(em_tree, start, len);
6785 if (em)
6786 em->bdev = root->fs_info->fs_devices->latest_bdev;
6787 read_unlock(&em_tree->lock);
6788
6789 if (em) {
6790 if (em->start > start || em->start + em->len <= start)
6791 free_extent_map(em);
6792 else if (em->block_start == EXTENT_MAP_INLINE && page)
6793 free_extent_map(em);
6794 else
6795 goto out;
6796 }
6797 em = alloc_extent_map();
6798 if (!em) {
6799 err = -ENOMEM;
6800 goto out;
6801 }
6802 em->bdev = root->fs_info->fs_devices->latest_bdev;
6803 em->start = EXTENT_MAP_HOLE;
6804 em->orig_start = EXTENT_MAP_HOLE;
6805 em->len = (u64)-1;
6806 em->block_len = (u64)-1;
6807
6808 if (!path) {
6809 path = btrfs_alloc_path();
6810 if (!path) {
6811 err = -ENOMEM;
6812 goto out;
6813 }
6814 /*
6815 * Chances are we'll be called again, so go ahead and do
6816 * readahead
6817 */
6818 path->reada = READA_FORWARD;
6819 }
6820
6821 ret = btrfs_lookup_file_extent(trans, root, path,
6822 objectid, start, trans != NULL);
6823 if (ret < 0) {
6824 err = ret;
6825 goto out;
6826 }
6827
6828 if (ret != 0) {
6829 if (path->slots[0] == 0)
6830 goto not_found;
6831 path->slots[0]--;
6832 }
6833
6834 leaf = path->nodes[0];
6835 item = btrfs_item_ptr(leaf, path->slots[0],
6836 struct btrfs_file_extent_item);
6837 /* are we inside the extent that was found? */
6838 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6839 found_type = found_key.type;
6840 if (found_key.objectid != objectid ||
6841 found_type != BTRFS_EXTENT_DATA_KEY) {
6842 /*
6843 * If we backup past the first extent we want to move forward
6844 * and see if there is an extent in front of us, otherwise we'll
6845 * say there is a hole for our whole search range which can
6846 * cause problems.
6847 */
6848 extent_end = start;
6849 goto next;
6850 }
6851
6852 found_type = btrfs_file_extent_type(leaf, item);
6853 extent_start = found_key.offset;
6854 if (found_type == BTRFS_FILE_EXTENT_REG ||
6855 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6856 extent_end = extent_start +
6857 btrfs_file_extent_num_bytes(leaf, item);
6858 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6859 size_t size;
6860 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6861 extent_end = ALIGN(extent_start + size, root->sectorsize);
6862 }
6863 next:
6864 if (start >= extent_end) {
6865 path->slots[0]++;
6866 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6867 ret = btrfs_next_leaf(root, path);
6868 if (ret < 0) {
6869 err = ret;
6870 goto out;
6871 }
6872 if (ret > 0)
6873 goto not_found;
6874 leaf = path->nodes[0];
6875 }
6876 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6877 if (found_key.objectid != objectid ||
6878 found_key.type != BTRFS_EXTENT_DATA_KEY)
6879 goto not_found;
6880 if (start + len <= found_key.offset)
6881 goto not_found;
6882 if (start > found_key.offset)
6883 goto next;
6884 em->start = start;
6885 em->orig_start = start;
6886 em->len = found_key.offset - start;
6887 goto not_found_em;
6888 }
6889
6890 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6891
6892 if (found_type == BTRFS_FILE_EXTENT_REG ||
6893 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6894 goto insert;
6895 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6896 unsigned long ptr;
6897 char *map;
6898 size_t size;
6899 size_t extent_offset;
6900 size_t copy_size;
6901
6902 if (new_inline)
6903 goto out;
6904
6905 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6906 extent_offset = page_offset(page) + pg_offset - extent_start;
6907 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6908 size - extent_offset);
6909 em->start = extent_start + extent_offset;
6910 em->len = ALIGN(copy_size, root->sectorsize);
6911 em->orig_block_len = em->len;
6912 em->orig_start = em->start;
6913 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6914 if (create == 0 && !PageUptodate(page)) {
6915 if (btrfs_file_extent_compression(leaf, item) !=
6916 BTRFS_COMPRESS_NONE) {
6917 ret = uncompress_inline(path, page, pg_offset,
6918 extent_offset, item);
6919 if (ret) {
6920 err = ret;
6921 goto out;
6922 }
6923 } else {
6924 map = kmap(page);
6925 read_extent_buffer(leaf, map + pg_offset, ptr,
6926 copy_size);
6927 if (pg_offset + copy_size < PAGE_SIZE) {
6928 memset(map + pg_offset + copy_size, 0,
6929 PAGE_SIZE - pg_offset -
6930 copy_size);
6931 }
6932 kunmap(page);
6933 }
6934 flush_dcache_page(page);
6935 } else if (create && PageUptodate(page)) {
6936 BUG();
6937 if (!trans) {
6938 kunmap(page);
6939 free_extent_map(em);
6940 em = NULL;
6941
6942 btrfs_release_path(path);
6943 trans = btrfs_join_transaction(root);
6944
6945 if (IS_ERR(trans))
6946 return ERR_CAST(trans);
6947 goto again;
6948 }
6949 map = kmap(page);
6950 write_extent_buffer(leaf, map + pg_offset, ptr,
6951 copy_size);
6952 kunmap(page);
6953 btrfs_mark_buffer_dirty(leaf);
6954 }
6955 set_extent_uptodate(io_tree, em->start,
6956 extent_map_end(em) - 1, NULL, GFP_NOFS);
6957 goto insert;
6958 }
6959 not_found:
6960 em->start = start;
6961 em->orig_start = start;
6962 em->len = len;
6963 not_found_em:
6964 em->block_start = EXTENT_MAP_HOLE;
6965 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6966 insert:
6967 btrfs_release_path(path);
6968 if (em->start > start || extent_map_end(em) <= start) {
6969 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6970 em->start, em->len, start, len);
6971 err = -EIO;
6972 goto out;
6973 }
6974
6975 err = 0;
6976 write_lock(&em_tree->lock);
6977 ret = add_extent_mapping(em_tree, em, 0);
6978 /* it is possible that someone inserted the extent into the tree
6979 * while we had the lock dropped. It is also possible that
6980 * an overlapping map exists in the tree
6981 */
6982 if (ret == -EEXIST) {
6983 struct extent_map *existing;
6984
6985 ret = 0;
6986
6987 existing = search_extent_mapping(em_tree, start, len);
6988 /*
6989 * existing will always be non-NULL, since there must be
6990 * extent causing the -EEXIST.
6991 */
6992 if (existing->start == em->start &&
6993 extent_map_end(existing) == extent_map_end(em) &&
6994 em->block_start == existing->block_start) {
6995 /*
6996 * these two extents are the same, it happens
6997 * with inlines especially
6998 */
6999 free_extent_map(em);
7000 em = existing;
7001 err = 0;
7002
7003 } else if (start >= extent_map_end(existing) ||
7004 start <= existing->start) {
7005 /*
7006 * The existing extent map is the one nearest to
7007 * the [start, start + len) range which overlaps
7008 */
7009 err = merge_extent_mapping(em_tree, existing,
7010 em, start);
7011 free_extent_map(existing);
7012 if (err) {
7013 free_extent_map(em);
7014 em = NULL;
7015 }
7016 } else {
7017 free_extent_map(em);
7018 em = existing;
7019 err = 0;
7020 }
7021 }
7022 write_unlock(&em_tree->lock);
7023 out:
7024
7025 trace_btrfs_get_extent(root, em);
7026
7027 btrfs_free_path(path);
7028 if (trans) {
7029 ret = btrfs_end_transaction(trans, root);
7030 if (!err)
7031 err = ret;
7032 }
7033 if (err) {
7034 free_extent_map(em);
7035 return ERR_PTR(err);
7036 }
7037 BUG_ON(!em); /* Error is always set */
7038 return em;
7039 }
7040
7041 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7042 size_t pg_offset, u64 start, u64 len,
7043 int create)
7044 {
7045 struct extent_map *em;
7046 struct extent_map *hole_em = NULL;
7047 u64 range_start = start;
7048 u64 end;
7049 u64 found;
7050 u64 found_end;
7051 int err = 0;
7052
7053 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7054 if (IS_ERR(em))
7055 return em;
7056 if (em) {
7057 /*
7058 * if our em maps to
7059 * - a hole or
7060 * - a pre-alloc extent,
7061 * there might actually be delalloc bytes behind it.
7062 */
7063 if (em->block_start != EXTENT_MAP_HOLE &&
7064 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7065 return em;
7066 else
7067 hole_em = em;
7068 }
7069
7070 /* check to see if we've wrapped (len == -1 or similar) */
7071 end = start + len;
7072 if (end < start)
7073 end = (u64)-1;
7074 else
7075 end -= 1;
7076
7077 em = NULL;
7078
7079 /* ok, we didn't find anything, lets look for delalloc */
7080 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7081 end, len, EXTENT_DELALLOC, 1);
7082 found_end = range_start + found;
7083 if (found_end < range_start)
7084 found_end = (u64)-1;
7085
7086 /*
7087 * we didn't find anything useful, return
7088 * the original results from get_extent()
7089 */
7090 if (range_start > end || found_end <= start) {
7091 em = hole_em;
7092 hole_em = NULL;
7093 goto out;
7094 }
7095
7096 /* adjust the range_start to make sure it doesn't
7097 * go backwards from the start they passed in
7098 */
7099 range_start = max(start, range_start);
7100 found = found_end - range_start;
7101
7102 if (found > 0) {
7103 u64 hole_start = start;
7104 u64 hole_len = len;
7105
7106 em = alloc_extent_map();
7107 if (!em) {
7108 err = -ENOMEM;
7109 goto out;
7110 }
7111 /*
7112 * when btrfs_get_extent can't find anything it
7113 * returns one huge hole
7114 *
7115 * make sure what it found really fits our range, and
7116 * adjust to make sure it is based on the start from
7117 * the caller
7118 */
7119 if (hole_em) {
7120 u64 calc_end = extent_map_end(hole_em);
7121
7122 if (calc_end <= start || (hole_em->start > end)) {
7123 free_extent_map(hole_em);
7124 hole_em = NULL;
7125 } else {
7126 hole_start = max(hole_em->start, start);
7127 hole_len = calc_end - hole_start;
7128 }
7129 }
7130 em->bdev = NULL;
7131 if (hole_em && range_start > hole_start) {
7132 /* our hole starts before our delalloc, so we
7133 * have to return just the parts of the hole
7134 * that go until the delalloc starts
7135 */
7136 em->len = min(hole_len,
7137 range_start - hole_start);
7138 em->start = hole_start;
7139 em->orig_start = hole_start;
7140 /*
7141 * don't adjust block start at all,
7142 * it is fixed at EXTENT_MAP_HOLE
7143 */
7144 em->block_start = hole_em->block_start;
7145 em->block_len = hole_len;
7146 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7147 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7148 } else {
7149 em->start = range_start;
7150 em->len = found;
7151 em->orig_start = range_start;
7152 em->block_start = EXTENT_MAP_DELALLOC;
7153 em->block_len = found;
7154 }
7155 } else if (hole_em) {
7156 return hole_em;
7157 }
7158 out:
7159
7160 free_extent_map(hole_em);
7161 if (err) {
7162 free_extent_map(em);
7163 return ERR_PTR(err);
7164 }
7165 return em;
7166 }
7167
7168 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7169 const u64 start,
7170 const u64 len,
7171 const u64 orig_start,
7172 const u64 block_start,
7173 const u64 block_len,
7174 const u64 orig_block_len,
7175 const u64 ram_bytes,
7176 const int type)
7177 {
7178 struct extent_map *em = NULL;
7179 int ret;
7180
7181 down_read(&BTRFS_I(inode)->dio_sem);
7182 if (type != BTRFS_ORDERED_NOCOW) {
7183 em = create_pinned_em(inode, start, len, orig_start,
7184 block_start, block_len, orig_block_len,
7185 ram_bytes, type);
7186 if (IS_ERR(em))
7187 goto out;
7188 }
7189 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7190 len, block_len, type);
7191 if (ret) {
7192 if (em) {
7193 free_extent_map(em);
7194 btrfs_drop_extent_cache(inode, start,
7195 start + len - 1, 0);
7196 }
7197 em = ERR_PTR(ret);
7198 }
7199 out:
7200 up_read(&BTRFS_I(inode)->dio_sem);
7201
7202 return em;
7203 }
7204
7205 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7206 u64 start, u64 len)
7207 {
7208 struct btrfs_root *root = BTRFS_I(inode)->root;
7209 struct extent_map *em;
7210 struct btrfs_key ins;
7211 u64 alloc_hint;
7212 int ret;
7213
7214 alloc_hint = get_extent_allocation_hint(inode, start, len);
7215 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7216 alloc_hint, &ins, 1, 1);
7217 if (ret)
7218 return ERR_PTR(ret);
7219
7220 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7221 ins.objectid, ins.offset, ins.offset,
7222 ins.offset, 0);
7223 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
7224 if (IS_ERR(em))
7225 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7226
7227 return em;
7228 }
7229
7230 /*
7231 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7232 * block must be cow'd
7233 */
7234 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7235 u64 *orig_start, u64 *orig_block_len,
7236 u64 *ram_bytes)
7237 {
7238 struct btrfs_trans_handle *trans;
7239 struct btrfs_path *path;
7240 int ret;
7241 struct extent_buffer *leaf;
7242 struct btrfs_root *root = BTRFS_I(inode)->root;
7243 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7244 struct btrfs_file_extent_item *fi;
7245 struct btrfs_key key;
7246 u64 disk_bytenr;
7247 u64 backref_offset;
7248 u64 extent_end;
7249 u64 num_bytes;
7250 int slot;
7251 int found_type;
7252 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7253
7254 path = btrfs_alloc_path();
7255 if (!path)
7256 return -ENOMEM;
7257
7258 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7259 offset, 0);
7260 if (ret < 0)
7261 goto out;
7262
7263 slot = path->slots[0];
7264 if (ret == 1) {
7265 if (slot == 0) {
7266 /* can't find the item, must cow */
7267 ret = 0;
7268 goto out;
7269 }
7270 slot--;
7271 }
7272 ret = 0;
7273 leaf = path->nodes[0];
7274 btrfs_item_key_to_cpu(leaf, &key, slot);
7275 if (key.objectid != btrfs_ino(inode) ||
7276 key.type != BTRFS_EXTENT_DATA_KEY) {
7277 /* not our file or wrong item type, must cow */
7278 goto out;
7279 }
7280
7281 if (key.offset > offset) {
7282 /* Wrong offset, must cow */
7283 goto out;
7284 }
7285
7286 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7287 found_type = btrfs_file_extent_type(leaf, fi);
7288 if (found_type != BTRFS_FILE_EXTENT_REG &&
7289 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7290 /* not a regular extent, must cow */
7291 goto out;
7292 }
7293
7294 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7295 goto out;
7296
7297 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7298 if (extent_end <= offset)
7299 goto out;
7300
7301 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7302 if (disk_bytenr == 0)
7303 goto out;
7304
7305 if (btrfs_file_extent_compression(leaf, fi) ||
7306 btrfs_file_extent_encryption(leaf, fi) ||
7307 btrfs_file_extent_other_encoding(leaf, fi))
7308 goto out;
7309
7310 backref_offset = btrfs_file_extent_offset(leaf, fi);
7311
7312 if (orig_start) {
7313 *orig_start = key.offset - backref_offset;
7314 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7315 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7316 }
7317
7318 if (btrfs_extent_readonly(root, disk_bytenr))
7319 goto out;
7320
7321 num_bytes = min(offset + *len, extent_end) - offset;
7322 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7323 u64 range_end;
7324
7325 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7326 ret = test_range_bit(io_tree, offset, range_end,
7327 EXTENT_DELALLOC, 0, NULL);
7328 if (ret) {
7329 ret = -EAGAIN;
7330 goto out;
7331 }
7332 }
7333
7334 btrfs_release_path(path);
7335
7336 /*
7337 * look for other files referencing this extent, if we
7338 * find any we must cow
7339 */
7340 trans = btrfs_join_transaction(root);
7341 if (IS_ERR(trans)) {
7342 ret = 0;
7343 goto out;
7344 }
7345
7346 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7347 key.offset - backref_offset, disk_bytenr);
7348 btrfs_end_transaction(trans, root);
7349 if (ret) {
7350 ret = 0;
7351 goto out;
7352 }
7353
7354 /*
7355 * adjust disk_bytenr and num_bytes to cover just the bytes
7356 * in this extent we are about to write. If there
7357 * are any csums in that range we have to cow in order
7358 * to keep the csums correct
7359 */
7360 disk_bytenr += backref_offset;
7361 disk_bytenr += offset - key.offset;
7362 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7363 goto out;
7364 /*
7365 * all of the above have passed, it is safe to overwrite this extent
7366 * without cow
7367 */
7368 *len = num_bytes;
7369 ret = 1;
7370 out:
7371 btrfs_free_path(path);
7372 return ret;
7373 }
7374
7375 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7376 {
7377 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7378 int found = false;
7379 void **pagep = NULL;
7380 struct page *page = NULL;
7381 int start_idx;
7382 int end_idx;
7383
7384 start_idx = start >> PAGE_SHIFT;
7385
7386 /*
7387 * end is the last byte in the last page. end == start is legal
7388 */
7389 end_idx = end >> PAGE_SHIFT;
7390
7391 rcu_read_lock();
7392
7393 /* Most of the code in this while loop is lifted from
7394 * find_get_page. It's been modified to begin searching from a
7395 * page and return just the first page found in that range. If the
7396 * found idx is less than or equal to the end idx then we know that
7397 * a page exists. If no pages are found or if those pages are
7398 * outside of the range then we're fine (yay!) */
7399 while (page == NULL &&
7400 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7401 page = radix_tree_deref_slot(pagep);
7402 if (unlikely(!page))
7403 break;
7404
7405 if (radix_tree_exception(page)) {
7406 if (radix_tree_deref_retry(page)) {
7407 page = NULL;
7408 continue;
7409 }
7410 /*
7411 * Otherwise, shmem/tmpfs must be storing a swap entry
7412 * here as an exceptional entry: so return it without
7413 * attempting to raise page count.
7414 */
7415 page = NULL;
7416 break; /* TODO: Is this relevant for this use case? */
7417 }
7418
7419 if (!page_cache_get_speculative(page)) {
7420 page = NULL;
7421 continue;
7422 }
7423
7424 /*
7425 * Has the page moved?
7426 * This is part of the lockless pagecache protocol. See
7427 * include/linux/pagemap.h for details.
7428 */
7429 if (unlikely(page != *pagep)) {
7430 put_page(page);
7431 page = NULL;
7432 }
7433 }
7434
7435 if (page) {
7436 if (page->index <= end_idx)
7437 found = true;
7438 put_page(page);
7439 }
7440
7441 rcu_read_unlock();
7442 return found;
7443 }
7444
7445 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7446 struct extent_state **cached_state, int writing)
7447 {
7448 struct btrfs_ordered_extent *ordered;
7449 int ret = 0;
7450
7451 while (1) {
7452 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7453 cached_state);
7454 /*
7455 * We're concerned with the entire range that we're going to be
7456 * doing DIO to, so we need to make sure there's no ordered
7457 * extents in this range.
7458 */
7459 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7460 lockend - lockstart + 1);
7461
7462 /*
7463 * We need to make sure there are no buffered pages in this
7464 * range either, we could have raced between the invalidate in
7465 * generic_file_direct_write and locking the extent. The
7466 * invalidate needs to happen so that reads after a write do not
7467 * get stale data.
7468 */
7469 if (!ordered &&
7470 (!writing ||
7471 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7472 break;
7473
7474 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7475 cached_state, GFP_NOFS);
7476
7477 if (ordered) {
7478 /*
7479 * If we are doing a DIO read and the ordered extent we
7480 * found is for a buffered write, we can not wait for it
7481 * to complete and retry, because if we do so we can
7482 * deadlock with concurrent buffered writes on page
7483 * locks. This happens only if our DIO read covers more
7484 * than one extent map, if at this point has already
7485 * created an ordered extent for a previous extent map
7486 * and locked its range in the inode's io tree, and a
7487 * concurrent write against that previous extent map's
7488 * range and this range started (we unlock the ranges
7489 * in the io tree only when the bios complete and
7490 * buffered writes always lock pages before attempting
7491 * to lock range in the io tree).
7492 */
7493 if (writing ||
7494 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7495 btrfs_start_ordered_extent(inode, ordered, 1);
7496 else
7497 ret = -ENOTBLK;
7498 btrfs_put_ordered_extent(ordered);
7499 } else {
7500 /*
7501 * We could trigger writeback for this range (and wait
7502 * for it to complete) and then invalidate the pages for
7503 * this range (through invalidate_inode_pages2_range()),
7504 * but that can lead us to a deadlock with a concurrent
7505 * call to readpages() (a buffered read or a defrag call
7506 * triggered a readahead) on a page lock due to an
7507 * ordered dio extent we created before but did not have
7508 * yet a corresponding bio submitted (whence it can not
7509 * complete), which makes readpages() wait for that
7510 * ordered extent to complete while holding a lock on
7511 * that page.
7512 */
7513 ret = -ENOTBLK;
7514 }
7515
7516 if (ret)
7517 break;
7518
7519 cond_resched();
7520 }
7521
7522 return ret;
7523 }
7524
7525 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7526 u64 len, u64 orig_start,
7527 u64 block_start, u64 block_len,
7528 u64 orig_block_len, u64 ram_bytes,
7529 int type)
7530 {
7531 struct extent_map_tree *em_tree;
7532 struct extent_map *em;
7533 struct btrfs_root *root = BTRFS_I(inode)->root;
7534 int ret;
7535
7536 em_tree = &BTRFS_I(inode)->extent_tree;
7537 em = alloc_extent_map();
7538 if (!em)
7539 return ERR_PTR(-ENOMEM);
7540
7541 em->start = start;
7542 em->orig_start = orig_start;
7543 em->mod_start = start;
7544 em->mod_len = len;
7545 em->len = len;
7546 em->block_len = block_len;
7547 em->block_start = block_start;
7548 em->bdev = root->fs_info->fs_devices->latest_bdev;
7549 em->orig_block_len = orig_block_len;
7550 em->ram_bytes = ram_bytes;
7551 em->generation = -1;
7552 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7553 if (type == BTRFS_ORDERED_PREALLOC)
7554 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7555
7556 do {
7557 btrfs_drop_extent_cache(inode, em->start,
7558 em->start + em->len - 1, 0);
7559 write_lock(&em_tree->lock);
7560 ret = add_extent_mapping(em_tree, em, 1);
7561 write_unlock(&em_tree->lock);
7562 } while (ret == -EEXIST);
7563
7564 if (ret) {
7565 free_extent_map(em);
7566 return ERR_PTR(ret);
7567 }
7568
7569 return em;
7570 }
7571
7572 static void adjust_dio_outstanding_extents(struct inode *inode,
7573 struct btrfs_dio_data *dio_data,
7574 const u64 len)
7575 {
7576 unsigned num_extents;
7577
7578 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7579 BTRFS_MAX_EXTENT_SIZE);
7580 /*
7581 * If we have an outstanding_extents count still set then we're
7582 * within our reservation, otherwise we need to adjust our inode
7583 * counter appropriately.
7584 */
7585 if (dio_data->outstanding_extents) {
7586 dio_data->outstanding_extents -= num_extents;
7587 } else {
7588 spin_lock(&BTRFS_I(inode)->lock);
7589 BTRFS_I(inode)->outstanding_extents += num_extents;
7590 spin_unlock(&BTRFS_I(inode)->lock);
7591 }
7592 }
7593
7594 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7595 struct buffer_head *bh_result, int create)
7596 {
7597 struct extent_map *em;
7598 struct btrfs_root *root = BTRFS_I(inode)->root;
7599 struct extent_state *cached_state = NULL;
7600 struct btrfs_dio_data *dio_data = NULL;
7601 u64 start = iblock << inode->i_blkbits;
7602 u64 lockstart, lockend;
7603 u64 len = bh_result->b_size;
7604 int unlock_bits = EXTENT_LOCKED;
7605 int ret = 0;
7606
7607 if (create)
7608 unlock_bits |= EXTENT_DIRTY;
7609 else
7610 len = min_t(u64, len, root->sectorsize);
7611
7612 lockstart = start;
7613 lockend = start + len - 1;
7614
7615 if (current->journal_info) {
7616 /*
7617 * Need to pull our outstanding extents and set journal_info to NULL so
7618 * that anything that needs to check if there's a transaction doesn't get
7619 * confused.
7620 */
7621 dio_data = current->journal_info;
7622 current->journal_info = NULL;
7623 }
7624
7625 /*
7626 * If this errors out it's because we couldn't invalidate pagecache for
7627 * this range and we need to fallback to buffered.
7628 */
7629 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7630 create)) {
7631 ret = -ENOTBLK;
7632 goto err;
7633 }
7634
7635 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7636 if (IS_ERR(em)) {
7637 ret = PTR_ERR(em);
7638 goto unlock_err;
7639 }
7640
7641 /*
7642 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7643 * io. INLINE is special, and we could probably kludge it in here, but
7644 * it's still buffered so for safety lets just fall back to the generic
7645 * buffered path.
7646 *
7647 * For COMPRESSED we _have_ to read the entire extent in so we can
7648 * decompress it, so there will be buffering required no matter what we
7649 * do, so go ahead and fallback to buffered.
7650 *
7651 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7652 * to buffered IO. Don't blame me, this is the price we pay for using
7653 * the generic code.
7654 */
7655 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7656 em->block_start == EXTENT_MAP_INLINE) {
7657 free_extent_map(em);
7658 ret = -ENOTBLK;
7659 goto unlock_err;
7660 }
7661
7662 /* Just a good old fashioned hole, return */
7663 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7664 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7665 free_extent_map(em);
7666 goto unlock_err;
7667 }
7668
7669 /*
7670 * We don't allocate a new extent in the following cases
7671 *
7672 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7673 * existing extent.
7674 * 2) The extent is marked as PREALLOC. We're good to go here and can
7675 * just use the extent.
7676 *
7677 */
7678 if (!create) {
7679 len = min(len, em->len - (start - em->start));
7680 lockstart = start + len;
7681 goto unlock;
7682 }
7683
7684 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7685 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7686 em->block_start != EXTENT_MAP_HOLE)) {
7687 int type;
7688 u64 block_start, orig_start, orig_block_len, ram_bytes;
7689
7690 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7691 type = BTRFS_ORDERED_PREALLOC;
7692 else
7693 type = BTRFS_ORDERED_NOCOW;
7694 len = min(len, em->len - (start - em->start));
7695 block_start = em->block_start + (start - em->start);
7696
7697 if (can_nocow_extent(inode, start, &len, &orig_start,
7698 &orig_block_len, &ram_bytes) == 1 &&
7699 btrfs_inc_nocow_writers(root->fs_info, block_start)) {
7700 struct extent_map *em2;
7701
7702 em2 = btrfs_create_dio_extent(inode, start, len,
7703 orig_start, block_start,
7704 len, orig_block_len,
7705 ram_bytes, type);
7706 btrfs_dec_nocow_writers(root->fs_info, block_start);
7707 if (type == BTRFS_ORDERED_PREALLOC) {
7708 free_extent_map(em);
7709 em = em2;
7710 }
7711 if (em2 && IS_ERR(em2)) {
7712 ret = PTR_ERR(em2);
7713 goto unlock_err;
7714 }
7715 goto unlock;
7716 }
7717 }
7718
7719 /*
7720 * this will cow the extent, reset the len in case we changed
7721 * it above
7722 */
7723 len = bh_result->b_size;
7724 free_extent_map(em);
7725 em = btrfs_new_extent_direct(inode, start, len);
7726 if (IS_ERR(em)) {
7727 ret = PTR_ERR(em);
7728 goto unlock_err;
7729 }
7730 len = min(len, em->len - (start - em->start));
7731 unlock:
7732 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7733 inode->i_blkbits;
7734 bh_result->b_size = len;
7735 bh_result->b_bdev = em->bdev;
7736 set_buffer_mapped(bh_result);
7737 if (create) {
7738 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7739 set_buffer_new(bh_result);
7740
7741 /*
7742 * Need to update the i_size under the extent lock so buffered
7743 * readers will get the updated i_size when we unlock.
7744 */
7745 if (start + len > i_size_read(inode))
7746 i_size_write(inode, start + len);
7747
7748 adjust_dio_outstanding_extents(inode, dio_data, len);
7749 btrfs_free_reserved_data_space(inode, start, len);
7750 WARN_ON(dio_data->reserve < len);
7751 dio_data->reserve -= len;
7752 dio_data->unsubmitted_oe_range_end = start + len;
7753 current->journal_info = dio_data;
7754 }
7755
7756 /*
7757 * In the case of write we need to clear and unlock the entire range,
7758 * in the case of read we need to unlock only the end area that we
7759 * aren't using if there is any left over space.
7760 */
7761 if (lockstart < lockend) {
7762 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7763 lockend, unlock_bits, 1, 0,
7764 &cached_state, GFP_NOFS);
7765 } else {
7766 free_extent_state(cached_state);
7767 }
7768
7769 free_extent_map(em);
7770
7771 return 0;
7772
7773 unlock_err:
7774 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7775 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7776 err:
7777 if (dio_data)
7778 current->journal_info = dio_data;
7779 /*
7780 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7781 * write less data then expected, so that we don't underflow our inode's
7782 * outstanding extents counter.
7783 */
7784 if (create && dio_data)
7785 adjust_dio_outstanding_extents(inode, dio_data, len);
7786
7787 return ret;
7788 }
7789
7790 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7791 int rw, int mirror_num)
7792 {
7793 struct btrfs_root *root = BTRFS_I(inode)->root;
7794 int ret;
7795
7796 BUG_ON(rw & REQ_WRITE);
7797
7798 bio_get(bio);
7799
7800 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7801 BTRFS_WQ_ENDIO_DIO_REPAIR);
7802 if (ret)
7803 goto err;
7804
7805 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7806 err:
7807 bio_put(bio);
7808 return ret;
7809 }
7810
7811 static int btrfs_check_dio_repairable(struct inode *inode,
7812 struct bio *failed_bio,
7813 struct io_failure_record *failrec,
7814 int failed_mirror)
7815 {
7816 int num_copies;
7817
7818 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7819 failrec->logical, failrec->len);
7820 if (num_copies == 1) {
7821 /*
7822 * we only have a single copy of the data, so don't bother with
7823 * all the retry and error correction code that follows. no
7824 * matter what the error is, it is very likely to persist.
7825 */
7826 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7827 num_copies, failrec->this_mirror, failed_mirror);
7828 return 0;
7829 }
7830
7831 failrec->failed_mirror = failed_mirror;
7832 failrec->this_mirror++;
7833 if (failrec->this_mirror == failed_mirror)
7834 failrec->this_mirror++;
7835
7836 if (failrec->this_mirror > num_copies) {
7837 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7838 num_copies, failrec->this_mirror, failed_mirror);
7839 return 0;
7840 }
7841
7842 return 1;
7843 }
7844
7845 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7846 struct page *page, unsigned int pgoff,
7847 u64 start, u64 end, int failed_mirror,
7848 bio_end_io_t *repair_endio, void *repair_arg)
7849 {
7850 struct io_failure_record *failrec;
7851 struct bio *bio;
7852 int isector;
7853 int read_mode;
7854 int ret;
7855
7856 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7857
7858 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7859 if (ret)
7860 return ret;
7861
7862 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7863 failed_mirror);
7864 if (!ret) {
7865 free_io_failure(inode, failrec);
7866 return -EIO;
7867 }
7868
7869 if ((failed_bio->bi_vcnt > 1)
7870 || (failed_bio->bi_io_vec->bv_len
7871 > BTRFS_I(inode)->root->sectorsize))
7872 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7873 else
7874 read_mode = READ_SYNC;
7875
7876 isector = start - btrfs_io_bio(failed_bio)->logical;
7877 isector >>= inode->i_sb->s_blocksize_bits;
7878 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7879 pgoff, isector, repair_endio, repair_arg);
7880 if (!bio) {
7881 free_io_failure(inode, failrec);
7882 return -EIO;
7883 }
7884
7885 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7886 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7887 read_mode, failrec->this_mirror, failrec->in_validation);
7888
7889 ret = submit_dio_repair_bio(inode, bio, read_mode,
7890 failrec->this_mirror);
7891 if (ret) {
7892 free_io_failure(inode, failrec);
7893 bio_put(bio);
7894 }
7895
7896 return ret;
7897 }
7898
7899 struct btrfs_retry_complete {
7900 struct completion done;
7901 struct inode *inode;
7902 u64 start;
7903 int uptodate;
7904 };
7905
7906 static void btrfs_retry_endio_nocsum(struct bio *bio)
7907 {
7908 struct btrfs_retry_complete *done = bio->bi_private;
7909 struct inode *inode;
7910 struct bio_vec *bvec;
7911 int i;
7912
7913 if (bio->bi_error)
7914 goto end;
7915
7916 ASSERT(bio->bi_vcnt == 1);
7917 inode = bio->bi_io_vec->bv_page->mapping->host;
7918 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7919
7920 done->uptodate = 1;
7921 bio_for_each_segment_all(bvec, bio, i)
7922 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7923 end:
7924 complete(&done->done);
7925 bio_put(bio);
7926 }
7927
7928 static int __btrfs_correct_data_nocsum(struct inode *inode,
7929 struct btrfs_io_bio *io_bio)
7930 {
7931 struct btrfs_fs_info *fs_info;
7932 struct bio_vec *bvec;
7933 struct btrfs_retry_complete done;
7934 u64 start;
7935 unsigned int pgoff;
7936 u32 sectorsize;
7937 int nr_sectors;
7938 int i;
7939 int ret;
7940
7941 fs_info = BTRFS_I(inode)->root->fs_info;
7942 sectorsize = BTRFS_I(inode)->root->sectorsize;
7943
7944 start = io_bio->logical;
7945 done.inode = inode;
7946
7947 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7948 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7949 pgoff = bvec->bv_offset;
7950
7951 next_block_or_try_again:
7952 done.uptodate = 0;
7953 done.start = start;
7954 init_completion(&done.done);
7955
7956 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7957 pgoff, start, start + sectorsize - 1,
7958 io_bio->mirror_num,
7959 btrfs_retry_endio_nocsum, &done);
7960 if (ret)
7961 return ret;
7962
7963 wait_for_completion(&done.done);
7964
7965 if (!done.uptodate) {
7966 /* We might have another mirror, so try again */
7967 goto next_block_or_try_again;
7968 }
7969
7970 start += sectorsize;
7971
7972 if (nr_sectors--) {
7973 pgoff += sectorsize;
7974 goto next_block_or_try_again;
7975 }
7976 }
7977
7978 return 0;
7979 }
7980
7981 static void btrfs_retry_endio(struct bio *bio)
7982 {
7983 struct btrfs_retry_complete *done = bio->bi_private;
7984 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7985 struct inode *inode;
7986 struct bio_vec *bvec;
7987 u64 start;
7988 int uptodate;
7989 int ret;
7990 int i;
7991
7992 if (bio->bi_error)
7993 goto end;
7994
7995 uptodate = 1;
7996
7997 start = done->start;
7998
7999 ASSERT(bio->bi_vcnt == 1);
8000 inode = bio->bi_io_vec->bv_page->mapping->host;
8001 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
8002
8003 bio_for_each_segment_all(bvec, bio, i) {
8004 ret = __readpage_endio_check(done->inode, io_bio, i,
8005 bvec->bv_page, bvec->bv_offset,
8006 done->start, bvec->bv_len);
8007 if (!ret)
8008 clean_io_failure(done->inode, done->start,
8009 bvec->bv_page, bvec->bv_offset);
8010 else
8011 uptodate = 0;
8012 }
8013
8014 done->uptodate = uptodate;
8015 end:
8016 complete(&done->done);
8017 bio_put(bio);
8018 }
8019
8020 static int __btrfs_subio_endio_read(struct inode *inode,
8021 struct btrfs_io_bio *io_bio, int err)
8022 {
8023 struct btrfs_fs_info *fs_info;
8024 struct bio_vec *bvec;
8025 struct btrfs_retry_complete done;
8026 u64 start;
8027 u64 offset = 0;
8028 u32 sectorsize;
8029 int nr_sectors;
8030 unsigned int pgoff;
8031 int csum_pos;
8032 int i;
8033 int ret;
8034
8035 fs_info = BTRFS_I(inode)->root->fs_info;
8036 sectorsize = BTRFS_I(inode)->root->sectorsize;
8037
8038 err = 0;
8039 start = io_bio->logical;
8040 done.inode = inode;
8041
8042 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
8043 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8044
8045 pgoff = bvec->bv_offset;
8046 next_block:
8047 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8048 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8049 bvec->bv_page, pgoff, start,
8050 sectorsize);
8051 if (likely(!ret))
8052 goto next;
8053 try_again:
8054 done.uptodate = 0;
8055 done.start = start;
8056 init_completion(&done.done);
8057
8058 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8059 pgoff, start, start + sectorsize - 1,
8060 io_bio->mirror_num,
8061 btrfs_retry_endio, &done);
8062 if (ret) {
8063 err = ret;
8064 goto next;
8065 }
8066
8067 wait_for_completion(&done.done);
8068
8069 if (!done.uptodate) {
8070 /* We might have another mirror, so try again */
8071 goto try_again;
8072 }
8073 next:
8074 offset += sectorsize;
8075 start += sectorsize;
8076
8077 ASSERT(nr_sectors);
8078
8079 if (--nr_sectors) {
8080 pgoff += sectorsize;
8081 goto next_block;
8082 }
8083 }
8084
8085 return err;
8086 }
8087
8088 static int btrfs_subio_endio_read(struct inode *inode,
8089 struct btrfs_io_bio *io_bio, int err)
8090 {
8091 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8092
8093 if (skip_csum) {
8094 if (unlikely(err))
8095 return __btrfs_correct_data_nocsum(inode, io_bio);
8096 else
8097 return 0;
8098 } else {
8099 return __btrfs_subio_endio_read(inode, io_bio, err);
8100 }
8101 }
8102
8103 static void btrfs_endio_direct_read(struct bio *bio)
8104 {
8105 struct btrfs_dio_private *dip = bio->bi_private;
8106 struct inode *inode = dip->inode;
8107 struct bio *dio_bio;
8108 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8109 int err = bio->bi_error;
8110
8111 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8112 err = btrfs_subio_endio_read(inode, io_bio, err);
8113
8114 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8115 dip->logical_offset + dip->bytes - 1);
8116 dio_bio = dip->dio_bio;
8117
8118 kfree(dip);
8119
8120 dio_bio->bi_error = bio->bi_error;
8121 dio_end_io(dio_bio, bio->bi_error);
8122
8123 if (io_bio->end_io)
8124 io_bio->end_io(io_bio, err);
8125 bio_put(bio);
8126 }
8127
8128 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8129 const u64 offset,
8130 const u64 bytes,
8131 const int uptodate)
8132 {
8133 struct btrfs_root *root = BTRFS_I(inode)->root;
8134 struct btrfs_ordered_extent *ordered = NULL;
8135 u64 ordered_offset = offset;
8136 u64 ordered_bytes = bytes;
8137 int ret;
8138
8139 again:
8140 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8141 &ordered_offset,
8142 ordered_bytes,
8143 uptodate);
8144 if (!ret)
8145 goto out_test;
8146
8147 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8148 finish_ordered_fn, NULL, NULL);
8149 btrfs_queue_work(root->fs_info->endio_write_workers,
8150 &ordered->work);
8151 out_test:
8152 /*
8153 * our bio might span multiple ordered extents. If we haven't
8154 * completed the accounting for the whole dio, go back and try again
8155 */
8156 if (ordered_offset < offset + bytes) {
8157 ordered_bytes = offset + bytes - ordered_offset;
8158 ordered = NULL;
8159 goto again;
8160 }
8161 }
8162
8163 static void btrfs_endio_direct_write(struct bio *bio)
8164 {
8165 struct btrfs_dio_private *dip = bio->bi_private;
8166 struct bio *dio_bio = dip->dio_bio;
8167
8168 btrfs_endio_direct_write_update_ordered(dip->inode,
8169 dip->logical_offset,
8170 dip->bytes,
8171 !bio->bi_error);
8172
8173 kfree(dip);
8174
8175 dio_bio->bi_error = bio->bi_error;
8176 dio_end_io(dio_bio, bio->bi_error);
8177 bio_put(bio);
8178 }
8179
8180 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8181 struct bio *bio, int mirror_num,
8182 unsigned long bio_flags, u64 offset)
8183 {
8184 int ret;
8185 struct btrfs_root *root = BTRFS_I(inode)->root;
8186 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8187 BUG_ON(ret); /* -ENOMEM */
8188 return 0;
8189 }
8190
8191 static void btrfs_end_dio_bio(struct bio *bio)
8192 {
8193 struct btrfs_dio_private *dip = bio->bi_private;
8194 int err = bio->bi_error;
8195
8196 if (err)
8197 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8198 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8199 btrfs_ino(dip->inode), bio->bi_rw,
8200 (unsigned long long)bio->bi_iter.bi_sector,
8201 bio->bi_iter.bi_size, err);
8202
8203 if (dip->subio_endio)
8204 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8205
8206 if (err) {
8207 dip->errors = 1;
8208
8209 /*
8210 * before atomic variable goto zero, we must make sure
8211 * dip->errors is perceived to be set.
8212 */
8213 smp_mb__before_atomic();
8214 }
8215
8216 /* if there are more bios still pending for this dio, just exit */
8217 if (!atomic_dec_and_test(&dip->pending_bios))
8218 goto out;
8219
8220 if (dip->errors) {
8221 bio_io_error(dip->orig_bio);
8222 } else {
8223 dip->dio_bio->bi_error = 0;
8224 bio_endio(dip->orig_bio);
8225 }
8226 out:
8227 bio_put(bio);
8228 }
8229
8230 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8231 u64 first_sector, gfp_t gfp_flags)
8232 {
8233 struct bio *bio;
8234 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8235 if (bio)
8236 bio_associate_current(bio);
8237 return bio;
8238 }
8239
8240 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8241 struct inode *inode,
8242 struct btrfs_dio_private *dip,
8243 struct bio *bio,
8244 u64 file_offset)
8245 {
8246 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8247 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8248 int ret;
8249
8250 /*
8251 * We load all the csum data we need when we submit
8252 * the first bio to reduce the csum tree search and
8253 * contention.
8254 */
8255 if (dip->logical_offset == file_offset) {
8256 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8257 file_offset);
8258 if (ret)
8259 return ret;
8260 }
8261
8262 if (bio == dip->orig_bio)
8263 return 0;
8264
8265 file_offset -= dip->logical_offset;
8266 file_offset >>= inode->i_sb->s_blocksize_bits;
8267 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8268
8269 return 0;
8270 }
8271
8272 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8273 int rw, u64 file_offset, int skip_sum,
8274 int async_submit)
8275 {
8276 struct btrfs_dio_private *dip = bio->bi_private;
8277 int write = rw & REQ_WRITE;
8278 struct btrfs_root *root = BTRFS_I(inode)->root;
8279 int ret;
8280
8281 if (async_submit)
8282 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8283
8284 bio_get(bio);
8285
8286 if (!write) {
8287 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8288 BTRFS_WQ_ENDIO_DATA);
8289 if (ret)
8290 goto err;
8291 }
8292
8293 if (skip_sum)
8294 goto map;
8295
8296 if (write && async_submit) {
8297 ret = btrfs_wq_submit_bio(root->fs_info,
8298 inode, rw, bio, 0, 0,
8299 file_offset,
8300 __btrfs_submit_bio_start_direct_io,
8301 __btrfs_submit_bio_done);
8302 goto err;
8303 } else if (write) {
8304 /*
8305 * If we aren't doing async submit, calculate the csum of the
8306 * bio now.
8307 */
8308 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8309 if (ret)
8310 goto err;
8311 } else {
8312 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8313 file_offset);
8314 if (ret)
8315 goto err;
8316 }
8317 map:
8318 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8319 err:
8320 bio_put(bio);
8321 return ret;
8322 }
8323
8324 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8325 int skip_sum)
8326 {
8327 struct inode *inode = dip->inode;
8328 struct btrfs_root *root = BTRFS_I(inode)->root;
8329 struct bio *bio;
8330 struct bio *orig_bio = dip->orig_bio;
8331 struct bio_vec *bvec = orig_bio->bi_io_vec;
8332 u64 start_sector = orig_bio->bi_iter.bi_sector;
8333 u64 file_offset = dip->logical_offset;
8334 u64 submit_len = 0;
8335 u64 map_length;
8336 u32 blocksize = root->sectorsize;
8337 int async_submit = 0;
8338 int nr_sectors;
8339 int ret;
8340 int i;
8341
8342 map_length = orig_bio->bi_iter.bi_size;
8343 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8344 &map_length, NULL, 0);
8345 if (ret)
8346 return -EIO;
8347
8348 if (map_length >= orig_bio->bi_iter.bi_size) {
8349 bio = orig_bio;
8350 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8351 goto submit;
8352 }
8353
8354 /* async crcs make it difficult to collect full stripe writes. */
8355 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8356 async_submit = 0;
8357 else
8358 async_submit = 1;
8359
8360 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8361 if (!bio)
8362 return -ENOMEM;
8363
8364 bio->bi_private = dip;
8365 bio->bi_end_io = btrfs_end_dio_bio;
8366 btrfs_io_bio(bio)->logical = file_offset;
8367 atomic_inc(&dip->pending_bios);
8368
8369 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8370 nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len);
8371 i = 0;
8372 next_block:
8373 if (unlikely(map_length < submit_len + blocksize ||
8374 bio_add_page(bio, bvec->bv_page, blocksize,
8375 bvec->bv_offset + (i * blocksize)) < blocksize)) {
8376 /*
8377 * inc the count before we submit the bio so
8378 * we know the end IO handler won't happen before
8379 * we inc the count. Otherwise, the dip might get freed
8380 * before we're done setting it up
8381 */
8382 atomic_inc(&dip->pending_bios);
8383 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8384 file_offset, skip_sum,
8385 async_submit);
8386 if (ret) {
8387 bio_put(bio);
8388 atomic_dec(&dip->pending_bios);
8389 goto out_err;
8390 }
8391
8392 start_sector += submit_len >> 9;
8393 file_offset += submit_len;
8394
8395 submit_len = 0;
8396
8397 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8398 start_sector, GFP_NOFS);
8399 if (!bio)
8400 goto out_err;
8401 bio->bi_private = dip;
8402 bio->bi_end_io = btrfs_end_dio_bio;
8403 btrfs_io_bio(bio)->logical = file_offset;
8404
8405 map_length = orig_bio->bi_iter.bi_size;
8406 ret = btrfs_map_block(root->fs_info, rw,
8407 start_sector << 9,
8408 &map_length, NULL, 0);
8409 if (ret) {
8410 bio_put(bio);
8411 goto out_err;
8412 }
8413
8414 goto next_block;
8415 } else {
8416 submit_len += blocksize;
8417 if (--nr_sectors) {
8418 i++;
8419 goto next_block;
8420 }
8421 bvec++;
8422 }
8423 }
8424
8425 submit:
8426 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8427 async_submit);
8428 if (!ret)
8429 return 0;
8430
8431 bio_put(bio);
8432 out_err:
8433 dip->errors = 1;
8434 /*
8435 * before atomic variable goto zero, we must
8436 * make sure dip->errors is perceived to be set.
8437 */
8438 smp_mb__before_atomic();
8439 if (atomic_dec_and_test(&dip->pending_bios))
8440 bio_io_error(dip->orig_bio);
8441
8442 /* bio_end_io() will handle error, so we needn't return it */
8443 return 0;
8444 }
8445
8446 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8447 struct inode *inode, loff_t file_offset)
8448 {
8449 struct btrfs_dio_private *dip = NULL;
8450 struct bio *io_bio = NULL;
8451 struct btrfs_io_bio *btrfs_bio;
8452 int skip_sum;
8453 int write = rw & REQ_WRITE;
8454 int ret = 0;
8455
8456 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8457
8458 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8459 if (!io_bio) {
8460 ret = -ENOMEM;
8461 goto free_ordered;
8462 }
8463
8464 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8465 if (!dip) {
8466 ret = -ENOMEM;
8467 goto free_ordered;
8468 }
8469
8470 dip->private = dio_bio->bi_private;
8471 dip->inode = inode;
8472 dip->logical_offset = file_offset;
8473 dip->bytes = dio_bio->bi_iter.bi_size;
8474 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8475 io_bio->bi_private = dip;
8476 dip->orig_bio = io_bio;
8477 dip->dio_bio = dio_bio;
8478 atomic_set(&dip->pending_bios, 0);
8479 btrfs_bio = btrfs_io_bio(io_bio);
8480 btrfs_bio->logical = file_offset;
8481
8482 if (write) {
8483 io_bio->bi_end_io = btrfs_endio_direct_write;
8484 } else {
8485 io_bio->bi_end_io = btrfs_endio_direct_read;
8486 dip->subio_endio = btrfs_subio_endio_read;
8487 }
8488
8489 /*
8490 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8491 * even if we fail to submit a bio, because in such case we do the
8492 * corresponding error handling below and it must not be done a second
8493 * time by btrfs_direct_IO().
8494 */
8495 if (write) {
8496 struct btrfs_dio_data *dio_data = current->journal_info;
8497
8498 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8499 dip->bytes;
8500 dio_data->unsubmitted_oe_range_start =
8501 dio_data->unsubmitted_oe_range_end;
8502 }
8503
8504 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8505 if (!ret)
8506 return;
8507
8508 if (btrfs_bio->end_io)
8509 btrfs_bio->end_io(btrfs_bio, ret);
8510
8511 free_ordered:
8512 /*
8513 * If we arrived here it means either we failed to submit the dip
8514 * or we either failed to clone the dio_bio or failed to allocate the
8515 * dip. If we cloned the dio_bio and allocated the dip, we can just
8516 * call bio_endio against our io_bio so that we get proper resource
8517 * cleanup if we fail to submit the dip, otherwise, we must do the
8518 * same as btrfs_endio_direct_[write|read] because we can't call these
8519 * callbacks - they require an allocated dip and a clone of dio_bio.
8520 */
8521 if (io_bio && dip) {
8522 io_bio->bi_error = -EIO;
8523 bio_endio(io_bio);
8524 /*
8525 * The end io callbacks free our dip, do the final put on io_bio
8526 * and all the cleanup and final put for dio_bio (through
8527 * dio_end_io()).
8528 */
8529 dip = NULL;
8530 io_bio = NULL;
8531 } else {
8532 if (write)
8533 btrfs_endio_direct_write_update_ordered(inode,
8534 file_offset,
8535 dio_bio->bi_iter.bi_size,
8536 0);
8537 else
8538 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8539 file_offset + dio_bio->bi_iter.bi_size - 1);
8540
8541 dio_bio->bi_error = -EIO;
8542 /*
8543 * Releases and cleans up our dio_bio, no need to bio_put()
8544 * nor bio_endio()/bio_io_error() against dio_bio.
8545 */
8546 dio_end_io(dio_bio, ret);
8547 }
8548 if (io_bio)
8549 bio_put(io_bio);
8550 kfree(dip);
8551 }
8552
8553 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8554 const struct iov_iter *iter, loff_t offset)
8555 {
8556 int seg;
8557 int i;
8558 unsigned blocksize_mask = root->sectorsize - 1;
8559 ssize_t retval = -EINVAL;
8560
8561 if (offset & blocksize_mask)
8562 goto out;
8563
8564 if (iov_iter_alignment(iter) & blocksize_mask)
8565 goto out;
8566
8567 /* If this is a write we don't need to check anymore */
8568 if (iov_iter_rw(iter) == WRITE)
8569 return 0;
8570 /*
8571 * Check to make sure we don't have duplicate iov_base's in this
8572 * iovec, if so return EINVAL, otherwise we'll get csum errors
8573 * when reading back.
8574 */
8575 for (seg = 0; seg < iter->nr_segs; seg++) {
8576 for (i = seg + 1; i < iter->nr_segs; i++) {
8577 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8578 goto out;
8579 }
8580 }
8581 retval = 0;
8582 out:
8583 return retval;
8584 }
8585
8586 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8587 loff_t offset)
8588 {
8589 struct file *file = iocb->ki_filp;
8590 struct inode *inode = file->f_mapping->host;
8591 struct btrfs_root *root = BTRFS_I(inode)->root;
8592 struct btrfs_dio_data dio_data = { 0 };
8593 size_t count = 0;
8594 int flags = 0;
8595 bool wakeup = true;
8596 bool relock = false;
8597 ssize_t ret;
8598
8599 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8600 return 0;
8601
8602 inode_dio_begin(inode);
8603 smp_mb__after_atomic();
8604
8605 /*
8606 * The generic stuff only does filemap_write_and_wait_range, which
8607 * isn't enough if we've written compressed pages to this area, so
8608 * we need to flush the dirty pages again to make absolutely sure
8609 * that any outstanding dirty pages are on disk.
8610 */
8611 count = iov_iter_count(iter);
8612 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8613 &BTRFS_I(inode)->runtime_flags))
8614 filemap_fdatawrite_range(inode->i_mapping, offset,
8615 offset + count - 1);
8616
8617 if (iov_iter_rw(iter) == WRITE) {
8618 /*
8619 * If the write DIO is beyond the EOF, we need update
8620 * the isize, but it is protected by i_mutex. So we can
8621 * not unlock the i_mutex at this case.
8622 */
8623 if (offset + count <= inode->i_size) {
8624 inode_unlock(inode);
8625 relock = true;
8626 }
8627 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8628 if (ret)
8629 goto out;
8630 dio_data.outstanding_extents = div64_u64(count +
8631 BTRFS_MAX_EXTENT_SIZE - 1,
8632 BTRFS_MAX_EXTENT_SIZE);
8633
8634 /*
8635 * We need to know how many extents we reserved so that we can
8636 * do the accounting properly if we go over the number we
8637 * originally calculated. Abuse current->journal_info for this.
8638 */
8639 dio_data.reserve = round_up(count, root->sectorsize);
8640 dio_data.unsubmitted_oe_range_start = (u64)offset;
8641 dio_data.unsubmitted_oe_range_end = (u64)offset;
8642 current->journal_info = &dio_data;
8643 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8644 &BTRFS_I(inode)->runtime_flags)) {
8645 inode_dio_end(inode);
8646 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8647 wakeup = false;
8648 }
8649
8650 ret = __blockdev_direct_IO(iocb, inode,
8651 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8652 iter, offset, btrfs_get_blocks_direct, NULL,
8653 btrfs_submit_direct, flags);
8654 if (iov_iter_rw(iter) == WRITE) {
8655 current->journal_info = NULL;
8656 if (ret < 0 && ret != -EIOCBQUEUED) {
8657 if (dio_data.reserve)
8658 btrfs_delalloc_release_space(inode, offset,
8659 dio_data.reserve);
8660 /*
8661 * On error we might have left some ordered extents
8662 * without submitting corresponding bios for them, so
8663 * cleanup them up to avoid other tasks getting them
8664 * and waiting for them to complete forever.
8665 */
8666 if (dio_data.unsubmitted_oe_range_start <
8667 dio_data.unsubmitted_oe_range_end)
8668 btrfs_endio_direct_write_update_ordered(inode,
8669 dio_data.unsubmitted_oe_range_start,
8670 dio_data.unsubmitted_oe_range_end -
8671 dio_data.unsubmitted_oe_range_start,
8672 0);
8673 } else if (ret >= 0 && (size_t)ret < count)
8674 btrfs_delalloc_release_space(inode, offset,
8675 count - (size_t)ret);
8676 }
8677 out:
8678 if (wakeup)
8679 inode_dio_end(inode);
8680 if (relock)
8681 inode_lock(inode);
8682
8683 return ret;
8684 }
8685
8686 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8687
8688 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8689 __u64 start, __u64 len)
8690 {
8691 int ret;
8692
8693 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8694 if (ret)
8695 return ret;
8696
8697 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8698 }
8699
8700 int btrfs_readpage(struct file *file, struct page *page)
8701 {
8702 struct extent_io_tree *tree;
8703 tree = &BTRFS_I(page->mapping->host)->io_tree;
8704 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8705 }
8706
8707 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8708 {
8709 struct extent_io_tree *tree;
8710 struct inode *inode = page->mapping->host;
8711 int ret;
8712
8713 if (current->flags & PF_MEMALLOC) {
8714 redirty_page_for_writepage(wbc, page);
8715 unlock_page(page);
8716 return 0;
8717 }
8718
8719 /*
8720 * If we are under memory pressure we will call this directly from the
8721 * VM, we need to make sure we have the inode referenced for the ordered
8722 * extent. If not just return like we didn't do anything.
8723 */
8724 if (!igrab(inode)) {
8725 redirty_page_for_writepage(wbc, page);
8726 return AOP_WRITEPAGE_ACTIVATE;
8727 }
8728 tree = &BTRFS_I(page->mapping->host)->io_tree;
8729 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8730 btrfs_add_delayed_iput(inode);
8731 return ret;
8732 }
8733
8734 static int btrfs_writepages(struct address_space *mapping,
8735 struct writeback_control *wbc)
8736 {
8737 struct extent_io_tree *tree;
8738
8739 tree = &BTRFS_I(mapping->host)->io_tree;
8740 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8741 }
8742
8743 static int
8744 btrfs_readpages(struct file *file, struct address_space *mapping,
8745 struct list_head *pages, unsigned nr_pages)
8746 {
8747 struct extent_io_tree *tree;
8748 tree = &BTRFS_I(mapping->host)->io_tree;
8749 return extent_readpages(tree, mapping, pages, nr_pages,
8750 btrfs_get_extent);
8751 }
8752 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8753 {
8754 struct extent_io_tree *tree;
8755 struct extent_map_tree *map;
8756 int ret;
8757
8758 tree = &BTRFS_I(page->mapping->host)->io_tree;
8759 map = &BTRFS_I(page->mapping->host)->extent_tree;
8760 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8761 if (ret == 1) {
8762 ClearPagePrivate(page);
8763 set_page_private(page, 0);
8764 put_page(page);
8765 }
8766 return ret;
8767 }
8768
8769 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8770 {
8771 if (PageWriteback(page) || PageDirty(page))
8772 return 0;
8773 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8774 }
8775
8776 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8777 unsigned int length)
8778 {
8779 struct inode *inode = page->mapping->host;
8780 struct extent_io_tree *tree;
8781 struct btrfs_ordered_extent *ordered;
8782 struct extent_state *cached_state = NULL;
8783 u64 page_start = page_offset(page);
8784 u64 page_end = page_start + PAGE_SIZE - 1;
8785 u64 start;
8786 u64 end;
8787 int inode_evicting = inode->i_state & I_FREEING;
8788
8789 /*
8790 * we have the page locked, so new writeback can't start,
8791 * and the dirty bit won't be cleared while we are here.
8792 *
8793 * Wait for IO on this page so that we can safely clear
8794 * the PagePrivate2 bit and do ordered accounting
8795 */
8796 wait_on_page_writeback(page);
8797
8798 tree = &BTRFS_I(inode)->io_tree;
8799 if (offset) {
8800 btrfs_releasepage(page, GFP_NOFS);
8801 return;
8802 }
8803
8804 if (!inode_evicting)
8805 lock_extent_bits(tree, page_start, page_end, &cached_state);
8806 again:
8807 start = page_start;
8808 ordered = btrfs_lookup_ordered_range(inode, start,
8809 page_end - start + 1);
8810 if (ordered) {
8811 end = min(page_end, ordered->file_offset + ordered->len - 1);
8812 /*
8813 * IO on this page will never be started, so we need
8814 * to account for any ordered extents now
8815 */
8816 if (!inode_evicting)
8817 clear_extent_bit(tree, start, end,
8818 EXTENT_DIRTY | EXTENT_DELALLOC |
8819 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8820 EXTENT_DEFRAG, 1, 0, &cached_state,
8821 GFP_NOFS);
8822 /*
8823 * whoever cleared the private bit is responsible
8824 * for the finish_ordered_io
8825 */
8826 if (TestClearPagePrivate2(page)) {
8827 struct btrfs_ordered_inode_tree *tree;
8828 u64 new_len;
8829
8830 tree = &BTRFS_I(inode)->ordered_tree;
8831
8832 spin_lock_irq(&tree->lock);
8833 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8834 new_len = start - ordered->file_offset;
8835 if (new_len < ordered->truncated_len)
8836 ordered->truncated_len = new_len;
8837 spin_unlock_irq(&tree->lock);
8838
8839 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8840 start,
8841 end - start + 1, 1))
8842 btrfs_finish_ordered_io(ordered);
8843 }
8844 btrfs_put_ordered_extent(ordered);
8845 if (!inode_evicting) {
8846 cached_state = NULL;
8847 lock_extent_bits(tree, start, end,
8848 &cached_state);
8849 }
8850
8851 start = end + 1;
8852 if (start < page_end)
8853 goto again;
8854 }
8855
8856 /*
8857 * Qgroup reserved space handler
8858 * Page here will be either
8859 * 1) Already written to disk
8860 * In this case, its reserved space is released from data rsv map
8861 * and will be freed by delayed_ref handler finally.
8862 * So even we call qgroup_free_data(), it won't decrease reserved
8863 * space.
8864 * 2) Not written to disk
8865 * This means the reserved space should be freed here.
8866 */
8867 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
8868 if (!inode_evicting) {
8869 clear_extent_bit(tree, page_start, page_end,
8870 EXTENT_LOCKED | EXTENT_DIRTY |
8871 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8872 EXTENT_DEFRAG, 1, 1,
8873 &cached_state, GFP_NOFS);
8874
8875 __btrfs_releasepage(page, GFP_NOFS);
8876 }
8877
8878 ClearPageChecked(page);
8879 if (PagePrivate(page)) {
8880 ClearPagePrivate(page);
8881 set_page_private(page, 0);
8882 put_page(page);
8883 }
8884 }
8885
8886 /*
8887 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8888 * called from a page fault handler when a page is first dirtied. Hence we must
8889 * be careful to check for EOF conditions here. We set the page up correctly
8890 * for a written page which means we get ENOSPC checking when writing into
8891 * holes and correct delalloc and unwritten extent mapping on filesystems that
8892 * support these features.
8893 *
8894 * We are not allowed to take the i_mutex here so we have to play games to
8895 * protect against truncate races as the page could now be beyond EOF. Because
8896 * vmtruncate() writes the inode size before removing pages, once we have the
8897 * page lock we can determine safely if the page is beyond EOF. If it is not
8898 * beyond EOF, then the page is guaranteed safe against truncation until we
8899 * unlock the page.
8900 */
8901 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8902 {
8903 struct page *page = vmf->page;
8904 struct inode *inode = file_inode(vma->vm_file);
8905 struct btrfs_root *root = BTRFS_I(inode)->root;
8906 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8907 struct btrfs_ordered_extent *ordered;
8908 struct extent_state *cached_state = NULL;
8909 char *kaddr;
8910 unsigned long zero_start;
8911 loff_t size;
8912 int ret;
8913 int reserved = 0;
8914 u64 reserved_space;
8915 u64 page_start;
8916 u64 page_end;
8917 u64 end;
8918
8919 reserved_space = PAGE_SIZE;
8920
8921 sb_start_pagefault(inode->i_sb);
8922 page_start = page_offset(page);
8923 page_end = page_start + PAGE_SIZE - 1;
8924 end = page_end;
8925
8926 /*
8927 * Reserving delalloc space after obtaining the page lock can lead to
8928 * deadlock. For example, if a dirty page is locked by this function
8929 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8930 * dirty page write out, then the btrfs_writepage() function could
8931 * end up waiting indefinitely to get a lock on the page currently
8932 * being processed by btrfs_page_mkwrite() function.
8933 */
8934 ret = btrfs_delalloc_reserve_space(inode, page_start,
8935 reserved_space);
8936 if (!ret) {
8937 ret = file_update_time(vma->vm_file);
8938 reserved = 1;
8939 }
8940 if (ret) {
8941 if (ret == -ENOMEM)
8942 ret = VM_FAULT_OOM;
8943 else /* -ENOSPC, -EIO, etc */
8944 ret = VM_FAULT_SIGBUS;
8945 if (reserved)
8946 goto out;
8947 goto out_noreserve;
8948 }
8949
8950 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8951 again:
8952 lock_page(page);
8953 size = i_size_read(inode);
8954
8955 if ((page->mapping != inode->i_mapping) ||
8956 (page_start >= size)) {
8957 /* page got truncated out from underneath us */
8958 goto out_unlock;
8959 }
8960 wait_on_page_writeback(page);
8961
8962 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8963 set_page_extent_mapped(page);
8964
8965 /*
8966 * we can't set the delalloc bits if there are pending ordered
8967 * extents. Drop our locks and wait for them to finish
8968 */
8969 ordered = btrfs_lookup_ordered_range(inode, page_start, page_end);
8970 if (ordered) {
8971 unlock_extent_cached(io_tree, page_start, page_end,
8972 &cached_state, GFP_NOFS);
8973 unlock_page(page);
8974 btrfs_start_ordered_extent(inode, ordered, 1);
8975 btrfs_put_ordered_extent(ordered);
8976 goto again;
8977 }
8978
8979 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8980 reserved_space = round_up(size - page_start, root->sectorsize);
8981 if (reserved_space < PAGE_SIZE) {
8982 end = page_start + reserved_space - 1;
8983 spin_lock(&BTRFS_I(inode)->lock);
8984 BTRFS_I(inode)->outstanding_extents++;
8985 spin_unlock(&BTRFS_I(inode)->lock);
8986 btrfs_delalloc_release_space(inode, page_start,
8987 PAGE_SIZE - reserved_space);
8988 }
8989 }
8990
8991 /*
8992 * XXX - page_mkwrite gets called every time the page is dirtied, even
8993 * if it was already dirty, so for space accounting reasons we need to
8994 * clear any delalloc bits for the range we are fixing to save. There
8995 * is probably a better way to do this, but for now keep consistent with
8996 * prepare_pages in the normal write path.
8997 */
8998 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8999 EXTENT_DIRTY | EXTENT_DELALLOC |
9000 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9001 0, 0, &cached_state, GFP_NOFS);
9002
9003 ret = btrfs_set_extent_delalloc(inode, page_start, end,
9004 &cached_state);
9005 if (ret) {
9006 unlock_extent_cached(io_tree, page_start, page_end,
9007 &cached_state, GFP_NOFS);
9008 ret = VM_FAULT_SIGBUS;
9009 goto out_unlock;
9010 }
9011 ret = 0;
9012
9013 /* page is wholly or partially inside EOF */
9014 if (page_start + PAGE_SIZE > size)
9015 zero_start = size & ~PAGE_MASK;
9016 else
9017 zero_start = PAGE_SIZE;
9018
9019 if (zero_start != PAGE_SIZE) {
9020 kaddr = kmap(page);
9021 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9022 flush_dcache_page(page);
9023 kunmap(page);
9024 }
9025 ClearPageChecked(page);
9026 set_page_dirty(page);
9027 SetPageUptodate(page);
9028
9029 BTRFS_I(inode)->last_trans = root->fs_info->generation;
9030 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9031 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9032
9033 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9034
9035 out_unlock:
9036 if (!ret) {
9037 sb_end_pagefault(inode->i_sb);
9038 return VM_FAULT_LOCKED;
9039 }
9040 unlock_page(page);
9041 out:
9042 btrfs_delalloc_release_space(inode, page_start, reserved_space);
9043 out_noreserve:
9044 sb_end_pagefault(inode->i_sb);
9045 return ret;
9046 }
9047
9048 static int btrfs_truncate(struct inode *inode)
9049 {
9050 struct btrfs_root *root = BTRFS_I(inode)->root;
9051 struct btrfs_block_rsv *rsv;
9052 int ret = 0;
9053 int err = 0;
9054 struct btrfs_trans_handle *trans;
9055 u64 mask = root->sectorsize - 1;
9056 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
9057
9058 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9059 (u64)-1);
9060 if (ret)
9061 return ret;
9062
9063 /*
9064 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9065 * 3 things going on here
9066 *
9067 * 1) We need to reserve space for our orphan item and the space to
9068 * delete our orphan item. Lord knows we don't want to have a dangling
9069 * orphan item because we didn't reserve space to remove it.
9070 *
9071 * 2) We need to reserve space to update our inode.
9072 *
9073 * 3) We need to have something to cache all the space that is going to
9074 * be free'd up by the truncate operation, but also have some slack
9075 * space reserved in case it uses space during the truncate (thank you
9076 * very much snapshotting).
9077 *
9078 * And we need these to all be separate. The fact is we can use a lot of
9079 * space doing the truncate, and we have no earthly idea how much space
9080 * we will use, so we need the truncate reservation to be separate so it
9081 * doesn't end up using space reserved for updating the inode or
9082 * removing the orphan item. We also need to be able to stop the
9083 * transaction and start a new one, which means we need to be able to
9084 * update the inode several times, and we have no idea of knowing how
9085 * many times that will be, so we can't just reserve 1 item for the
9086 * entirety of the operation, so that has to be done separately as well.
9087 * Then there is the orphan item, which does indeed need to be held on
9088 * to for the whole operation, and we need nobody to touch this reserved
9089 * space except the orphan code.
9090 *
9091 * So that leaves us with
9092 *
9093 * 1) root->orphan_block_rsv - for the orphan deletion.
9094 * 2) rsv - for the truncate reservation, which we will steal from the
9095 * transaction reservation.
9096 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9097 * updating the inode.
9098 */
9099 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
9100 if (!rsv)
9101 return -ENOMEM;
9102 rsv->size = min_size;
9103 rsv->failfast = 1;
9104
9105 /*
9106 * 1 for the truncate slack space
9107 * 1 for updating the inode.
9108 */
9109 trans = btrfs_start_transaction(root, 2);
9110 if (IS_ERR(trans)) {
9111 err = PTR_ERR(trans);
9112 goto out;
9113 }
9114
9115 /* Migrate the slack space for the truncate to our reserve */
9116 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
9117 min_size);
9118 BUG_ON(ret);
9119
9120 /*
9121 * So if we truncate and then write and fsync we normally would just
9122 * write the extents that changed, which is a problem if we need to
9123 * first truncate that entire inode. So set this flag so we write out
9124 * all of the extents in the inode to the sync log so we're completely
9125 * safe.
9126 */
9127 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9128 trans->block_rsv = rsv;
9129
9130 while (1) {
9131 ret = btrfs_truncate_inode_items(trans, root, inode,
9132 inode->i_size,
9133 BTRFS_EXTENT_DATA_KEY);
9134 if (ret != -ENOSPC && ret != -EAGAIN) {
9135 err = ret;
9136 break;
9137 }
9138
9139 trans->block_rsv = &root->fs_info->trans_block_rsv;
9140 ret = btrfs_update_inode(trans, root, inode);
9141 if (ret) {
9142 err = ret;
9143 break;
9144 }
9145
9146 btrfs_end_transaction(trans, root);
9147 btrfs_btree_balance_dirty(root);
9148
9149 trans = btrfs_start_transaction(root, 2);
9150 if (IS_ERR(trans)) {
9151 ret = err = PTR_ERR(trans);
9152 trans = NULL;
9153 break;
9154 }
9155
9156 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9157 rsv, min_size);
9158 BUG_ON(ret); /* shouldn't happen */
9159 trans->block_rsv = rsv;
9160 }
9161
9162 if (ret == 0 && inode->i_nlink > 0) {
9163 trans->block_rsv = root->orphan_block_rsv;
9164 ret = btrfs_orphan_del(trans, inode);
9165 if (ret)
9166 err = ret;
9167 }
9168
9169 if (trans) {
9170 trans->block_rsv = &root->fs_info->trans_block_rsv;
9171 ret = btrfs_update_inode(trans, root, inode);
9172 if (ret && !err)
9173 err = ret;
9174
9175 ret = btrfs_end_transaction(trans, root);
9176 btrfs_btree_balance_dirty(root);
9177 }
9178
9179 out:
9180 btrfs_free_block_rsv(root, rsv);
9181
9182 if (ret && !err)
9183 err = ret;
9184
9185 return err;
9186 }
9187
9188 /*
9189 * create a new subvolume directory/inode (helper for the ioctl).
9190 */
9191 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9192 struct btrfs_root *new_root,
9193 struct btrfs_root *parent_root,
9194 u64 new_dirid)
9195 {
9196 struct inode *inode;
9197 int err;
9198 u64 index = 0;
9199
9200 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9201 new_dirid, new_dirid,
9202 S_IFDIR | (~current_umask() & S_IRWXUGO),
9203 &index);
9204 if (IS_ERR(inode))
9205 return PTR_ERR(inode);
9206 inode->i_op = &btrfs_dir_inode_operations;
9207 inode->i_fop = &btrfs_dir_file_operations;
9208
9209 set_nlink(inode, 1);
9210 btrfs_i_size_write(inode, 0);
9211 unlock_new_inode(inode);
9212
9213 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9214 if (err)
9215 btrfs_err(new_root->fs_info,
9216 "error inheriting subvolume %llu properties: %d",
9217 new_root->root_key.objectid, err);
9218
9219 err = btrfs_update_inode(trans, new_root, inode);
9220
9221 iput(inode);
9222 return err;
9223 }
9224
9225 struct inode *btrfs_alloc_inode(struct super_block *sb)
9226 {
9227 struct btrfs_inode *ei;
9228 struct inode *inode;
9229
9230 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9231 if (!ei)
9232 return NULL;
9233
9234 ei->root = NULL;
9235 ei->generation = 0;
9236 ei->last_trans = 0;
9237 ei->last_sub_trans = 0;
9238 ei->logged_trans = 0;
9239 ei->delalloc_bytes = 0;
9240 ei->defrag_bytes = 0;
9241 ei->disk_i_size = 0;
9242 ei->flags = 0;
9243 ei->csum_bytes = 0;
9244 ei->index_cnt = (u64)-1;
9245 ei->dir_index = 0;
9246 ei->last_unlink_trans = 0;
9247 ei->last_log_commit = 0;
9248 ei->delayed_iput_count = 0;
9249
9250 spin_lock_init(&ei->lock);
9251 ei->outstanding_extents = 0;
9252 ei->reserved_extents = 0;
9253
9254 ei->runtime_flags = 0;
9255 ei->force_compress = BTRFS_COMPRESS_NONE;
9256
9257 ei->delayed_node = NULL;
9258
9259 ei->i_otime.tv_sec = 0;
9260 ei->i_otime.tv_nsec = 0;
9261
9262 inode = &ei->vfs_inode;
9263 extent_map_tree_init(&ei->extent_tree);
9264 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9265 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9266 ei->io_tree.track_uptodate = 1;
9267 ei->io_failure_tree.track_uptodate = 1;
9268 atomic_set(&ei->sync_writers, 0);
9269 mutex_init(&ei->log_mutex);
9270 mutex_init(&ei->delalloc_mutex);
9271 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9272 INIT_LIST_HEAD(&ei->delalloc_inodes);
9273 INIT_LIST_HEAD(&ei->delayed_iput);
9274 RB_CLEAR_NODE(&ei->rb_node);
9275 init_rwsem(&ei->dio_sem);
9276
9277 return inode;
9278 }
9279
9280 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9281 void btrfs_test_destroy_inode(struct inode *inode)
9282 {
9283 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9284 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9285 }
9286 #endif
9287
9288 static void btrfs_i_callback(struct rcu_head *head)
9289 {
9290 struct inode *inode = container_of(head, struct inode, i_rcu);
9291 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9292 }
9293
9294 void btrfs_destroy_inode(struct inode *inode)
9295 {
9296 struct btrfs_ordered_extent *ordered;
9297 struct btrfs_root *root = BTRFS_I(inode)->root;
9298
9299 WARN_ON(!hlist_empty(&inode->i_dentry));
9300 WARN_ON(inode->i_data.nrpages);
9301 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9302 WARN_ON(BTRFS_I(inode)->reserved_extents);
9303 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9304 WARN_ON(BTRFS_I(inode)->csum_bytes);
9305 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9306
9307 /*
9308 * This can happen where we create an inode, but somebody else also
9309 * created the same inode and we need to destroy the one we already
9310 * created.
9311 */
9312 if (!root)
9313 goto free;
9314
9315 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9316 &BTRFS_I(inode)->runtime_flags)) {
9317 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9318 btrfs_ino(inode));
9319 atomic_dec(&root->orphan_inodes);
9320 }
9321
9322 while (1) {
9323 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9324 if (!ordered)
9325 break;
9326 else {
9327 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9328 ordered->file_offset, ordered->len);
9329 btrfs_remove_ordered_extent(inode, ordered);
9330 btrfs_put_ordered_extent(ordered);
9331 btrfs_put_ordered_extent(ordered);
9332 }
9333 }
9334 btrfs_qgroup_check_reserved_leak(inode);
9335 inode_tree_del(inode);
9336 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9337 free:
9338 call_rcu(&inode->i_rcu, btrfs_i_callback);
9339 }
9340
9341 int btrfs_drop_inode(struct inode *inode)
9342 {
9343 struct btrfs_root *root = BTRFS_I(inode)->root;
9344
9345 if (root == NULL)
9346 return 1;
9347
9348 /* the snap/subvol tree is on deleting */
9349 if (btrfs_root_refs(&root->root_item) == 0)
9350 return 1;
9351 else
9352 return generic_drop_inode(inode);
9353 }
9354
9355 static void init_once(void *foo)
9356 {
9357 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9358
9359 inode_init_once(&ei->vfs_inode);
9360 }
9361
9362 void btrfs_destroy_cachep(void)
9363 {
9364 /*
9365 * Make sure all delayed rcu free inodes are flushed before we
9366 * destroy cache.
9367 */
9368 rcu_barrier();
9369 kmem_cache_destroy(btrfs_inode_cachep);
9370 kmem_cache_destroy(btrfs_trans_handle_cachep);
9371 kmem_cache_destroy(btrfs_transaction_cachep);
9372 kmem_cache_destroy(btrfs_path_cachep);
9373 kmem_cache_destroy(btrfs_free_space_cachep);
9374 }
9375
9376 int btrfs_init_cachep(void)
9377 {
9378 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9379 sizeof(struct btrfs_inode), 0,
9380 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9381 init_once);
9382 if (!btrfs_inode_cachep)
9383 goto fail;
9384
9385 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9386 sizeof(struct btrfs_trans_handle), 0,
9387 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9388 if (!btrfs_trans_handle_cachep)
9389 goto fail;
9390
9391 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9392 sizeof(struct btrfs_transaction), 0,
9393 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9394 if (!btrfs_transaction_cachep)
9395 goto fail;
9396
9397 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9398 sizeof(struct btrfs_path), 0,
9399 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9400 if (!btrfs_path_cachep)
9401 goto fail;
9402
9403 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9404 sizeof(struct btrfs_free_space), 0,
9405 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9406 if (!btrfs_free_space_cachep)
9407 goto fail;
9408
9409 return 0;
9410 fail:
9411 btrfs_destroy_cachep();
9412 return -ENOMEM;
9413 }
9414
9415 static int btrfs_getattr(struct vfsmount *mnt,
9416 struct dentry *dentry, struct kstat *stat)
9417 {
9418 u64 delalloc_bytes;
9419 struct inode *inode = d_inode(dentry);
9420 u32 blocksize = inode->i_sb->s_blocksize;
9421
9422 generic_fillattr(inode, stat);
9423 stat->dev = BTRFS_I(inode)->root->anon_dev;
9424
9425 spin_lock(&BTRFS_I(inode)->lock);
9426 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9427 spin_unlock(&BTRFS_I(inode)->lock);
9428 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9429 ALIGN(delalloc_bytes, blocksize)) >> 9;
9430 return 0;
9431 }
9432
9433 static int btrfs_rename_exchange(struct inode *old_dir,
9434 struct dentry *old_dentry,
9435 struct inode *new_dir,
9436 struct dentry *new_dentry)
9437 {
9438 struct btrfs_trans_handle *trans;
9439 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9440 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9441 struct inode *new_inode = new_dentry->d_inode;
9442 struct inode *old_inode = old_dentry->d_inode;
9443 struct timespec ctime = CURRENT_TIME;
9444 struct dentry *parent;
9445 u64 old_ino = btrfs_ino(old_inode);
9446 u64 new_ino = btrfs_ino(new_inode);
9447 u64 old_idx = 0;
9448 u64 new_idx = 0;
9449 u64 root_objectid;
9450 int ret;
9451 bool root_log_pinned = false;
9452 bool dest_log_pinned = false;
9453
9454 /* we only allow rename subvolume link between subvolumes */
9455 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9456 return -EXDEV;
9457
9458 /* close the race window with snapshot create/destroy ioctl */
9459 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9460 down_read(&root->fs_info->subvol_sem);
9461 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9462 down_read(&dest->fs_info->subvol_sem);
9463
9464 /*
9465 * We want to reserve the absolute worst case amount of items. So if
9466 * both inodes are subvols and we need to unlink them then that would
9467 * require 4 item modifications, but if they are both normal inodes it
9468 * would require 5 item modifications, so we'll assume their normal
9469 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9470 * should cover the worst case number of items we'll modify.
9471 */
9472 trans = btrfs_start_transaction(root, 12);
9473 if (IS_ERR(trans)) {
9474 ret = PTR_ERR(trans);
9475 goto out_notrans;
9476 }
9477
9478 /*
9479 * We need to find a free sequence number both in the source and
9480 * in the destination directory for the exchange.
9481 */
9482 ret = btrfs_set_inode_index(new_dir, &old_idx);
9483 if (ret)
9484 goto out_fail;
9485 ret = btrfs_set_inode_index(old_dir, &new_idx);
9486 if (ret)
9487 goto out_fail;
9488
9489 BTRFS_I(old_inode)->dir_index = 0ULL;
9490 BTRFS_I(new_inode)->dir_index = 0ULL;
9491
9492 /* Reference for the source. */
9493 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9494 /* force full log commit if subvolume involved. */
9495 btrfs_set_log_full_commit(root->fs_info, trans);
9496 } else {
9497 btrfs_pin_log_trans(root);
9498 root_log_pinned = true;
9499 ret = btrfs_insert_inode_ref(trans, dest,
9500 new_dentry->d_name.name,
9501 new_dentry->d_name.len,
9502 old_ino,
9503 btrfs_ino(new_dir), old_idx);
9504 if (ret)
9505 goto out_fail;
9506 }
9507
9508 /* And now for the dest. */
9509 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9510 /* force full log commit if subvolume involved. */
9511 btrfs_set_log_full_commit(dest->fs_info, trans);
9512 } else {
9513 btrfs_pin_log_trans(dest);
9514 dest_log_pinned = true;
9515 ret = btrfs_insert_inode_ref(trans, root,
9516 old_dentry->d_name.name,
9517 old_dentry->d_name.len,
9518 new_ino,
9519 btrfs_ino(old_dir), new_idx);
9520 if (ret)
9521 goto out_fail;
9522 }
9523
9524 /* Update inode version and ctime/mtime. */
9525 inode_inc_iversion(old_dir);
9526 inode_inc_iversion(new_dir);
9527 inode_inc_iversion(old_inode);
9528 inode_inc_iversion(new_inode);
9529 old_dir->i_ctime = old_dir->i_mtime = ctime;
9530 new_dir->i_ctime = new_dir->i_mtime = ctime;
9531 old_inode->i_ctime = ctime;
9532 new_inode->i_ctime = ctime;
9533
9534 if (old_dentry->d_parent != new_dentry->d_parent) {
9535 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9536 btrfs_record_unlink_dir(trans, new_dir, new_inode, 1);
9537 }
9538
9539 /* src is a subvolume */
9540 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9541 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9542 ret = btrfs_unlink_subvol(trans, root, old_dir,
9543 root_objectid,
9544 old_dentry->d_name.name,
9545 old_dentry->d_name.len);
9546 } else { /* src is an inode */
9547 ret = __btrfs_unlink_inode(trans, root, old_dir,
9548 old_dentry->d_inode,
9549 old_dentry->d_name.name,
9550 old_dentry->d_name.len);
9551 if (!ret)
9552 ret = btrfs_update_inode(trans, root, old_inode);
9553 }
9554 if (ret) {
9555 btrfs_abort_transaction(trans, root, ret);
9556 goto out_fail;
9557 }
9558
9559 /* dest is a subvolume */
9560 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9561 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9562 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9563 root_objectid,
9564 new_dentry->d_name.name,
9565 new_dentry->d_name.len);
9566 } else { /* dest is an inode */
9567 ret = __btrfs_unlink_inode(trans, dest, new_dir,
9568 new_dentry->d_inode,
9569 new_dentry->d_name.name,
9570 new_dentry->d_name.len);
9571 if (!ret)
9572 ret = btrfs_update_inode(trans, dest, new_inode);
9573 }
9574 if (ret) {
9575 btrfs_abort_transaction(trans, root, ret);
9576 goto out_fail;
9577 }
9578
9579 ret = btrfs_add_link(trans, new_dir, old_inode,
9580 new_dentry->d_name.name,
9581 new_dentry->d_name.len, 0, old_idx);
9582 if (ret) {
9583 btrfs_abort_transaction(trans, root, ret);
9584 goto out_fail;
9585 }
9586
9587 ret = btrfs_add_link(trans, old_dir, new_inode,
9588 old_dentry->d_name.name,
9589 old_dentry->d_name.len, 0, new_idx);
9590 if (ret) {
9591 btrfs_abort_transaction(trans, root, ret);
9592 goto out_fail;
9593 }
9594
9595 if (old_inode->i_nlink == 1)
9596 BTRFS_I(old_inode)->dir_index = old_idx;
9597 if (new_inode->i_nlink == 1)
9598 BTRFS_I(new_inode)->dir_index = new_idx;
9599
9600 if (root_log_pinned) {
9601 parent = new_dentry->d_parent;
9602 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9603 btrfs_end_log_trans(root);
9604 root_log_pinned = false;
9605 }
9606 if (dest_log_pinned) {
9607 parent = old_dentry->d_parent;
9608 btrfs_log_new_name(trans, new_inode, new_dir, parent);
9609 btrfs_end_log_trans(dest);
9610 dest_log_pinned = false;
9611 }
9612 out_fail:
9613 /*
9614 * If we have pinned a log and an error happened, we unpin tasks
9615 * trying to sync the log and force them to fallback to a transaction
9616 * commit if the log currently contains any of the inodes involved in
9617 * this rename operation (to ensure we do not persist a log with an
9618 * inconsistent state for any of these inodes or leading to any
9619 * inconsistencies when replayed). If the transaction was aborted, the
9620 * abortion reason is propagated to userspace when attempting to commit
9621 * the transaction. If the log does not contain any of these inodes, we
9622 * allow the tasks to sync it.
9623 */
9624 if (ret && (root_log_pinned || dest_log_pinned)) {
9625 if (btrfs_inode_in_log(old_dir, root->fs_info->generation) ||
9626 btrfs_inode_in_log(new_dir, root->fs_info->generation) ||
9627 btrfs_inode_in_log(old_inode, root->fs_info->generation) ||
9628 (new_inode &&
9629 btrfs_inode_in_log(new_inode, root->fs_info->generation)))
9630 btrfs_set_log_full_commit(root->fs_info, trans);
9631
9632 if (root_log_pinned) {
9633 btrfs_end_log_trans(root);
9634 root_log_pinned = false;
9635 }
9636 if (dest_log_pinned) {
9637 btrfs_end_log_trans(dest);
9638 dest_log_pinned = false;
9639 }
9640 }
9641 ret = btrfs_end_transaction(trans, root);
9642 out_notrans:
9643 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9644 up_read(&dest->fs_info->subvol_sem);
9645 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9646 up_read(&root->fs_info->subvol_sem);
9647
9648 return ret;
9649 }
9650
9651 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9652 struct btrfs_root *root,
9653 struct inode *dir,
9654 struct dentry *dentry)
9655 {
9656 int ret;
9657 struct inode *inode;
9658 u64 objectid;
9659 u64 index;
9660
9661 ret = btrfs_find_free_ino(root, &objectid);
9662 if (ret)
9663 return ret;
9664
9665 inode = btrfs_new_inode(trans, root, dir,
9666 dentry->d_name.name,
9667 dentry->d_name.len,
9668 btrfs_ino(dir),
9669 objectid,
9670 S_IFCHR | WHITEOUT_MODE,
9671 &index);
9672
9673 if (IS_ERR(inode)) {
9674 ret = PTR_ERR(inode);
9675 return ret;
9676 }
9677
9678 inode->i_op = &btrfs_special_inode_operations;
9679 init_special_inode(inode, inode->i_mode,
9680 WHITEOUT_DEV);
9681
9682 ret = btrfs_init_inode_security(trans, inode, dir,
9683 &dentry->d_name);
9684 if (ret)
9685 goto out;
9686
9687 ret = btrfs_add_nondir(trans, dir, dentry,
9688 inode, 0, index);
9689 if (ret)
9690 goto out;
9691
9692 ret = btrfs_update_inode(trans, root, inode);
9693 out:
9694 unlock_new_inode(inode);
9695 if (ret)
9696 inode_dec_link_count(inode);
9697 iput(inode);
9698
9699 return ret;
9700 }
9701
9702 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9703 struct inode *new_dir, struct dentry *new_dentry,
9704 unsigned int flags)
9705 {
9706 struct btrfs_trans_handle *trans;
9707 unsigned int trans_num_items;
9708 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9709 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9710 struct inode *new_inode = d_inode(new_dentry);
9711 struct inode *old_inode = d_inode(old_dentry);
9712 u64 index = 0;
9713 u64 root_objectid;
9714 int ret;
9715 u64 old_ino = btrfs_ino(old_inode);
9716 bool log_pinned = false;
9717
9718 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9719 return -EPERM;
9720
9721 /* we only allow rename subvolume link between subvolumes */
9722 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9723 return -EXDEV;
9724
9725 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9726 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9727 return -ENOTEMPTY;
9728
9729 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9730 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9731 return -ENOTEMPTY;
9732
9733
9734 /* check for collisions, even if the name isn't there */
9735 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9736 new_dentry->d_name.name,
9737 new_dentry->d_name.len);
9738
9739 if (ret) {
9740 if (ret == -EEXIST) {
9741 /* we shouldn't get
9742 * eexist without a new_inode */
9743 if (WARN_ON(!new_inode)) {
9744 return ret;
9745 }
9746 } else {
9747 /* maybe -EOVERFLOW */
9748 return ret;
9749 }
9750 }
9751 ret = 0;
9752
9753 /*
9754 * we're using rename to replace one file with another. Start IO on it
9755 * now so we don't add too much work to the end of the transaction
9756 */
9757 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9758 filemap_flush(old_inode->i_mapping);
9759
9760 /* close the racy window with snapshot create/destroy ioctl */
9761 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9762 down_read(&root->fs_info->subvol_sem);
9763 /*
9764 * We want to reserve the absolute worst case amount of items. So if
9765 * both inodes are subvols and we need to unlink them then that would
9766 * require 4 item modifications, but if they are both normal inodes it
9767 * would require 5 item modifications, so we'll assume they are normal
9768 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9769 * should cover the worst case number of items we'll modify.
9770 * If our rename has the whiteout flag, we need more 5 units for the
9771 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9772 * when selinux is enabled).
9773 */
9774 trans_num_items = 11;
9775 if (flags & RENAME_WHITEOUT)
9776 trans_num_items += 5;
9777 trans = btrfs_start_transaction(root, trans_num_items);
9778 if (IS_ERR(trans)) {
9779 ret = PTR_ERR(trans);
9780 goto out_notrans;
9781 }
9782
9783 if (dest != root)
9784 btrfs_record_root_in_trans(trans, dest);
9785
9786 ret = btrfs_set_inode_index(new_dir, &index);
9787 if (ret)
9788 goto out_fail;
9789
9790 BTRFS_I(old_inode)->dir_index = 0ULL;
9791 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9792 /* force full log commit if subvolume involved. */
9793 btrfs_set_log_full_commit(root->fs_info, trans);
9794 } else {
9795 btrfs_pin_log_trans(root);
9796 log_pinned = true;
9797 ret = btrfs_insert_inode_ref(trans, dest,
9798 new_dentry->d_name.name,
9799 new_dentry->d_name.len,
9800 old_ino,
9801 btrfs_ino(new_dir), index);
9802 if (ret)
9803 goto out_fail;
9804 }
9805
9806 inode_inc_iversion(old_dir);
9807 inode_inc_iversion(new_dir);
9808 inode_inc_iversion(old_inode);
9809 old_dir->i_ctime = old_dir->i_mtime =
9810 new_dir->i_ctime = new_dir->i_mtime =
9811 old_inode->i_ctime = current_fs_time(old_dir->i_sb);
9812
9813 if (old_dentry->d_parent != new_dentry->d_parent)
9814 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9815
9816 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9817 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9818 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9819 old_dentry->d_name.name,
9820 old_dentry->d_name.len);
9821 } else {
9822 ret = __btrfs_unlink_inode(trans, root, old_dir,
9823 d_inode(old_dentry),
9824 old_dentry->d_name.name,
9825 old_dentry->d_name.len);
9826 if (!ret)
9827 ret = btrfs_update_inode(trans, root, old_inode);
9828 }
9829 if (ret) {
9830 btrfs_abort_transaction(trans, root, ret);
9831 goto out_fail;
9832 }
9833
9834 if (new_inode) {
9835 inode_inc_iversion(new_inode);
9836 new_inode->i_ctime = current_fs_time(new_inode->i_sb);
9837 if (unlikely(btrfs_ino(new_inode) ==
9838 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9839 root_objectid = BTRFS_I(new_inode)->location.objectid;
9840 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9841 root_objectid,
9842 new_dentry->d_name.name,
9843 new_dentry->d_name.len);
9844 BUG_ON(new_inode->i_nlink == 0);
9845 } else {
9846 ret = btrfs_unlink_inode(trans, dest, new_dir,
9847 d_inode(new_dentry),
9848 new_dentry->d_name.name,
9849 new_dentry->d_name.len);
9850 }
9851 if (!ret && new_inode->i_nlink == 0)
9852 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9853 if (ret) {
9854 btrfs_abort_transaction(trans, root, ret);
9855 goto out_fail;
9856 }
9857 }
9858
9859 ret = btrfs_add_link(trans, new_dir, old_inode,
9860 new_dentry->d_name.name,
9861 new_dentry->d_name.len, 0, index);
9862 if (ret) {
9863 btrfs_abort_transaction(trans, root, ret);
9864 goto out_fail;
9865 }
9866
9867 if (old_inode->i_nlink == 1)
9868 BTRFS_I(old_inode)->dir_index = index;
9869
9870 if (log_pinned) {
9871 struct dentry *parent = new_dentry->d_parent;
9872
9873 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9874 btrfs_end_log_trans(root);
9875 log_pinned = false;
9876 }
9877
9878 if (flags & RENAME_WHITEOUT) {
9879 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9880 old_dentry);
9881
9882 if (ret) {
9883 btrfs_abort_transaction(trans, root, ret);
9884 goto out_fail;
9885 }
9886 }
9887 out_fail:
9888 /*
9889 * If we have pinned the log and an error happened, we unpin tasks
9890 * trying to sync the log and force them to fallback to a transaction
9891 * commit if the log currently contains any of the inodes involved in
9892 * this rename operation (to ensure we do not persist a log with an
9893 * inconsistent state for any of these inodes or leading to any
9894 * inconsistencies when replayed). If the transaction was aborted, the
9895 * abortion reason is propagated to userspace when attempting to commit
9896 * the transaction. If the log does not contain any of these inodes, we
9897 * allow the tasks to sync it.
9898 */
9899 if (ret && log_pinned) {
9900 if (btrfs_inode_in_log(old_dir, root->fs_info->generation) ||
9901 btrfs_inode_in_log(new_dir, root->fs_info->generation) ||
9902 btrfs_inode_in_log(old_inode, root->fs_info->generation) ||
9903 (new_inode &&
9904 btrfs_inode_in_log(new_inode, root->fs_info->generation)))
9905 btrfs_set_log_full_commit(root->fs_info, trans);
9906
9907 btrfs_end_log_trans(root);
9908 log_pinned = false;
9909 }
9910 btrfs_end_transaction(trans, root);
9911 out_notrans:
9912 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9913 up_read(&root->fs_info->subvol_sem);
9914
9915 return ret;
9916 }
9917
9918 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9919 struct inode *new_dir, struct dentry *new_dentry,
9920 unsigned int flags)
9921 {
9922 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9923 return -EINVAL;
9924
9925 if (flags & RENAME_EXCHANGE)
9926 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9927 new_dentry);
9928
9929 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9930 }
9931
9932 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9933 {
9934 struct btrfs_delalloc_work *delalloc_work;
9935 struct inode *inode;
9936
9937 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9938 work);
9939 inode = delalloc_work->inode;
9940 filemap_flush(inode->i_mapping);
9941 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9942 &BTRFS_I(inode)->runtime_flags))
9943 filemap_flush(inode->i_mapping);
9944
9945 if (delalloc_work->delay_iput)
9946 btrfs_add_delayed_iput(inode);
9947 else
9948 iput(inode);
9949 complete(&delalloc_work->completion);
9950 }
9951
9952 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9953 int delay_iput)
9954 {
9955 struct btrfs_delalloc_work *work;
9956
9957 work = kmalloc(sizeof(*work), GFP_NOFS);
9958 if (!work)
9959 return NULL;
9960
9961 init_completion(&work->completion);
9962 INIT_LIST_HEAD(&work->list);
9963 work->inode = inode;
9964 work->delay_iput = delay_iput;
9965 WARN_ON_ONCE(!inode);
9966 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9967 btrfs_run_delalloc_work, NULL, NULL);
9968
9969 return work;
9970 }
9971
9972 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9973 {
9974 wait_for_completion(&work->completion);
9975 kfree(work);
9976 }
9977
9978 /*
9979 * some fairly slow code that needs optimization. This walks the list
9980 * of all the inodes with pending delalloc and forces them to disk.
9981 */
9982 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9983 int nr)
9984 {
9985 struct btrfs_inode *binode;
9986 struct inode *inode;
9987 struct btrfs_delalloc_work *work, *next;
9988 struct list_head works;
9989 struct list_head splice;
9990 int ret = 0;
9991
9992 INIT_LIST_HEAD(&works);
9993 INIT_LIST_HEAD(&splice);
9994
9995 mutex_lock(&root->delalloc_mutex);
9996 spin_lock(&root->delalloc_lock);
9997 list_splice_init(&root->delalloc_inodes, &splice);
9998 while (!list_empty(&splice)) {
9999 binode = list_entry(splice.next, struct btrfs_inode,
10000 delalloc_inodes);
10001
10002 list_move_tail(&binode->delalloc_inodes,
10003 &root->delalloc_inodes);
10004 inode = igrab(&binode->vfs_inode);
10005 if (!inode) {
10006 cond_resched_lock(&root->delalloc_lock);
10007 continue;
10008 }
10009 spin_unlock(&root->delalloc_lock);
10010
10011 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10012 if (!work) {
10013 if (delay_iput)
10014 btrfs_add_delayed_iput(inode);
10015 else
10016 iput(inode);
10017 ret = -ENOMEM;
10018 goto out;
10019 }
10020 list_add_tail(&work->list, &works);
10021 btrfs_queue_work(root->fs_info->flush_workers,
10022 &work->work);
10023 ret++;
10024 if (nr != -1 && ret >= nr)
10025 goto out;
10026 cond_resched();
10027 spin_lock(&root->delalloc_lock);
10028 }
10029 spin_unlock(&root->delalloc_lock);
10030
10031 out:
10032 list_for_each_entry_safe(work, next, &works, list) {
10033 list_del_init(&work->list);
10034 btrfs_wait_and_free_delalloc_work(work);
10035 }
10036
10037 if (!list_empty_careful(&splice)) {
10038 spin_lock(&root->delalloc_lock);
10039 list_splice_tail(&splice, &root->delalloc_inodes);
10040 spin_unlock(&root->delalloc_lock);
10041 }
10042 mutex_unlock(&root->delalloc_mutex);
10043 return ret;
10044 }
10045
10046 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10047 {
10048 int ret;
10049
10050 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
10051 return -EROFS;
10052
10053 ret = __start_delalloc_inodes(root, delay_iput, -1);
10054 if (ret > 0)
10055 ret = 0;
10056 /*
10057 * the filemap_flush will queue IO into the worker threads, but
10058 * we have to make sure the IO is actually started and that
10059 * ordered extents get created before we return
10060 */
10061 atomic_inc(&root->fs_info->async_submit_draining);
10062 while (atomic_read(&root->fs_info->nr_async_submits) ||
10063 atomic_read(&root->fs_info->async_delalloc_pages)) {
10064 wait_event(root->fs_info->async_submit_wait,
10065 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
10066 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
10067 }
10068 atomic_dec(&root->fs_info->async_submit_draining);
10069 return ret;
10070 }
10071
10072 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10073 int nr)
10074 {
10075 struct btrfs_root *root;
10076 struct list_head splice;
10077 int ret;
10078
10079 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10080 return -EROFS;
10081
10082 INIT_LIST_HEAD(&splice);
10083
10084 mutex_lock(&fs_info->delalloc_root_mutex);
10085 spin_lock(&fs_info->delalloc_root_lock);
10086 list_splice_init(&fs_info->delalloc_roots, &splice);
10087 while (!list_empty(&splice) && nr) {
10088 root = list_first_entry(&splice, struct btrfs_root,
10089 delalloc_root);
10090 root = btrfs_grab_fs_root(root);
10091 BUG_ON(!root);
10092 list_move_tail(&root->delalloc_root,
10093 &fs_info->delalloc_roots);
10094 spin_unlock(&fs_info->delalloc_root_lock);
10095
10096 ret = __start_delalloc_inodes(root, delay_iput, nr);
10097 btrfs_put_fs_root(root);
10098 if (ret < 0)
10099 goto out;
10100
10101 if (nr != -1) {
10102 nr -= ret;
10103 WARN_ON(nr < 0);
10104 }
10105 spin_lock(&fs_info->delalloc_root_lock);
10106 }
10107 spin_unlock(&fs_info->delalloc_root_lock);
10108
10109 ret = 0;
10110 atomic_inc(&fs_info->async_submit_draining);
10111 while (atomic_read(&fs_info->nr_async_submits) ||
10112 atomic_read(&fs_info->async_delalloc_pages)) {
10113 wait_event(fs_info->async_submit_wait,
10114 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10115 atomic_read(&fs_info->async_delalloc_pages) == 0));
10116 }
10117 atomic_dec(&fs_info->async_submit_draining);
10118 out:
10119 if (!list_empty_careful(&splice)) {
10120 spin_lock(&fs_info->delalloc_root_lock);
10121 list_splice_tail(&splice, &fs_info->delalloc_roots);
10122 spin_unlock(&fs_info->delalloc_root_lock);
10123 }
10124 mutex_unlock(&fs_info->delalloc_root_mutex);
10125 return ret;
10126 }
10127
10128 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10129 const char *symname)
10130 {
10131 struct btrfs_trans_handle *trans;
10132 struct btrfs_root *root = BTRFS_I(dir)->root;
10133 struct btrfs_path *path;
10134 struct btrfs_key key;
10135 struct inode *inode = NULL;
10136 int err;
10137 int drop_inode = 0;
10138 u64 objectid;
10139 u64 index = 0;
10140 int name_len;
10141 int datasize;
10142 unsigned long ptr;
10143 struct btrfs_file_extent_item *ei;
10144 struct extent_buffer *leaf;
10145
10146 name_len = strlen(symname);
10147 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
10148 return -ENAMETOOLONG;
10149
10150 /*
10151 * 2 items for inode item and ref
10152 * 2 items for dir items
10153 * 1 item for updating parent inode item
10154 * 1 item for the inline extent item
10155 * 1 item for xattr if selinux is on
10156 */
10157 trans = btrfs_start_transaction(root, 7);
10158 if (IS_ERR(trans))
10159 return PTR_ERR(trans);
10160
10161 err = btrfs_find_free_ino(root, &objectid);
10162 if (err)
10163 goto out_unlock;
10164
10165 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10166 dentry->d_name.len, btrfs_ino(dir), objectid,
10167 S_IFLNK|S_IRWXUGO, &index);
10168 if (IS_ERR(inode)) {
10169 err = PTR_ERR(inode);
10170 goto out_unlock;
10171 }
10172
10173 /*
10174 * If the active LSM wants to access the inode during
10175 * d_instantiate it needs these. Smack checks to see
10176 * if the filesystem supports xattrs by looking at the
10177 * ops vector.
10178 */
10179 inode->i_fop = &btrfs_file_operations;
10180 inode->i_op = &btrfs_file_inode_operations;
10181 inode->i_mapping->a_ops = &btrfs_aops;
10182 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10183
10184 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10185 if (err)
10186 goto out_unlock_inode;
10187
10188 path = btrfs_alloc_path();
10189 if (!path) {
10190 err = -ENOMEM;
10191 goto out_unlock_inode;
10192 }
10193 key.objectid = btrfs_ino(inode);
10194 key.offset = 0;
10195 key.type = BTRFS_EXTENT_DATA_KEY;
10196 datasize = btrfs_file_extent_calc_inline_size(name_len);
10197 err = btrfs_insert_empty_item(trans, root, path, &key,
10198 datasize);
10199 if (err) {
10200 btrfs_free_path(path);
10201 goto out_unlock_inode;
10202 }
10203 leaf = path->nodes[0];
10204 ei = btrfs_item_ptr(leaf, path->slots[0],
10205 struct btrfs_file_extent_item);
10206 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10207 btrfs_set_file_extent_type(leaf, ei,
10208 BTRFS_FILE_EXTENT_INLINE);
10209 btrfs_set_file_extent_encryption(leaf, ei, 0);
10210 btrfs_set_file_extent_compression(leaf, ei, 0);
10211 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10212 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10213
10214 ptr = btrfs_file_extent_inline_start(ei);
10215 write_extent_buffer(leaf, symname, ptr, name_len);
10216 btrfs_mark_buffer_dirty(leaf);
10217 btrfs_free_path(path);
10218
10219 inode->i_op = &btrfs_symlink_inode_operations;
10220 inode_nohighmem(inode);
10221 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10222 inode_set_bytes(inode, name_len);
10223 btrfs_i_size_write(inode, name_len);
10224 err = btrfs_update_inode(trans, root, inode);
10225 /*
10226 * Last step, add directory indexes for our symlink inode. This is the
10227 * last step to avoid extra cleanup of these indexes if an error happens
10228 * elsewhere above.
10229 */
10230 if (!err)
10231 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
10232 if (err) {
10233 drop_inode = 1;
10234 goto out_unlock_inode;
10235 }
10236
10237 unlock_new_inode(inode);
10238 d_instantiate(dentry, inode);
10239
10240 out_unlock:
10241 btrfs_end_transaction(trans, root);
10242 if (drop_inode) {
10243 inode_dec_link_count(inode);
10244 iput(inode);
10245 }
10246 btrfs_btree_balance_dirty(root);
10247 return err;
10248
10249 out_unlock_inode:
10250 drop_inode = 1;
10251 unlock_new_inode(inode);
10252 goto out_unlock;
10253 }
10254
10255 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10256 u64 start, u64 num_bytes, u64 min_size,
10257 loff_t actual_len, u64 *alloc_hint,
10258 struct btrfs_trans_handle *trans)
10259 {
10260 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10261 struct extent_map *em;
10262 struct btrfs_root *root = BTRFS_I(inode)->root;
10263 struct btrfs_key ins;
10264 u64 cur_offset = start;
10265 u64 i_size;
10266 u64 cur_bytes;
10267 u64 last_alloc = (u64)-1;
10268 int ret = 0;
10269 bool own_trans = true;
10270
10271 if (trans)
10272 own_trans = false;
10273 while (num_bytes > 0) {
10274 if (own_trans) {
10275 trans = btrfs_start_transaction(root, 3);
10276 if (IS_ERR(trans)) {
10277 ret = PTR_ERR(trans);
10278 break;
10279 }
10280 }
10281
10282 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10283 cur_bytes = max(cur_bytes, min_size);
10284 /*
10285 * If we are severely fragmented we could end up with really
10286 * small allocations, so if the allocator is returning small
10287 * chunks lets make its job easier by only searching for those
10288 * sized chunks.
10289 */
10290 cur_bytes = min(cur_bytes, last_alloc);
10291 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
10292 *alloc_hint, &ins, 1, 0);
10293 if (ret) {
10294 if (own_trans)
10295 btrfs_end_transaction(trans, root);
10296 break;
10297 }
10298 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
10299
10300 last_alloc = ins.offset;
10301 ret = insert_reserved_file_extent(trans, inode,
10302 cur_offset, ins.objectid,
10303 ins.offset, ins.offset,
10304 ins.offset, 0, 0, 0,
10305 BTRFS_FILE_EXTENT_PREALLOC);
10306 if (ret) {
10307 btrfs_free_reserved_extent(root, ins.objectid,
10308 ins.offset, 0);
10309 btrfs_abort_transaction(trans, root, ret);
10310 if (own_trans)
10311 btrfs_end_transaction(trans, root);
10312 break;
10313 }
10314
10315 btrfs_drop_extent_cache(inode, cur_offset,
10316 cur_offset + ins.offset -1, 0);
10317
10318 em = alloc_extent_map();
10319 if (!em) {
10320 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10321 &BTRFS_I(inode)->runtime_flags);
10322 goto next;
10323 }
10324
10325 em->start = cur_offset;
10326 em->orig_start = cur_offset;
10327 em->len = ins.offset;
10328 em->block_start = ins.objectid;
10329 em->block_len = ins.offset;
10330 em->orig_block_len = ins.offset;
10331 em->ram_bytes = ins.offset;
10332 em->bdev = root->fs_info->fs_devices->latest_bdev;
10333 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10334 em->generation = trans->transid;
10335
10336 while (1) {
10337 write_lock(&em_tree->lock);
10338 ret = add_extent_mapping(em_tree, em, 1);
10339 write_unlock(&em_tree->lock);
10340 if (ret != -EEXIST)
10341 break;
10342 btrfs_drop_extent_cache(inode, cur_offset,
10343 cur_offset + ins.offset - 1,
10344 0);
10345 }
10346 free_extent_map(em);
10347 next:
10348 num_bytes -= ins.offset;
10349 cur_offset += ins.offset;
10350 *alloc_hint = ins.objectid + ins.offset;
10351
10352 inode_inc_iversion(inode);
10353 inode->i_ctime = current_fs_time(inode->i_sb);
10354 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10355 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10356 (actual_len > inode->i_size) &&
10357 (cur_offset > inode->i_size)) {
10358 if (cur_offset > actual_len)
10359 i_size = actual_len;
10360 else
10361 i_size = cur_offset;
10362 i_size_write(inode, i_size);
10363 btrfs_ordered_update_i_size(inode, i_size, NULL);
10364 }
10365
10366 ret = btrfs_update_inode(trans, root, inode);
10367
10368 if (ret) {
10369 btrfs_abort_transaction(trans, root, ret);
10370 if (own_trans)
10371 btrfs_end_transaction(trans, root);
10372 break;
10373 }
10374
10375 if (own_trans)
10376 btrfs_end_transaction(trans, root);
10377 }
10378 return ret;
10379 }
10380
10381 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10382 u64 start, u64 num_bytes, u64 min_size,
10383 loff_t actual_len, u64 *alloc_hint)
10384 {
10385 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10386 min_size, actual_len, alloc_hint,
10387 NULL);
10388 }
10389
10390 int btrfs_prealloc_file_range_trans(struct inode *inode,
10391 struct btrfs_trans_handle *trans, int mode,
10392 u64 start, u64 num_bytes, u64 min_size,
10393 loff_t actual_len, u64 *alloc_hint)
10394 {
10395 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10396 min_size, actual_len, alloc_hint, trans);
10397 }
10398
10399 static int btrfs_set_page_dirty(struct page *page)
10400 {
10401 return __set_page_dirty_nobuffers(page);
10402 }
10403
10404 static int btrfs_permission(struct inode *inode, int mask)
10405 {
10406 struct btrfs_root *root = BTRFS_I(inode)->root;
10407 umode_t mode = inode->i_mode;
10408
10409 if (mask & MAY_WRITE &&
10410 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10411 if (btrfs_root_readonly(root))
10412 return -EROFS;
10413 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10414 return -EACCES;
10415 }
10416 return generic_permission(inode, mask);
10417 }
10418
10419 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10420 {
10421 struct btrfs_trans_handle *trans;
10422 struct btrfs_root *root = BTRFS_I(dir)->root;
10423 struct inode *inode = NULL;
10424 u64 objectid;
10425 u64 index;
10426 int ret = 0;
10427
10428 /*
10429 * 5 units required for adding orphan entry
10430 */
10431 trans = btrfs_start_transaction(root, 5);
10432 if (IS_ERR(trans))
10433 return PTR_ERR(trans);
10434
10435 ret = btrfs_find_free_ino(root, &objectid);
10436 if (ret)
10437 goto out;
10438
10439 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10440 btrfs_ino(dir), objectid, mode, &index);
10441 if (IS_ERR(inode)) {
10442 ret = PTR_ERR(inode);
10443 inode = NULL;
10444 goto out;
10445 }
10446
10447 inode->i_fop = &btrfs_file_operations;
10448 inode->i_op = &btrfs_file_inode_operations;
10449
10450 inode->i_mapping->a_ops = &btrfs_aops;
10451 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10452
10453 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10454 if (ret)
10455 goto out_inode;
10456
10457 ret = btrfs_update_inode(trans, root, inode);
10458 if (ret)
10459 goto out_inode;
10460 ret = btrfs_orphan_add(trans, inode);
10461 if (ret)
10462 goto out_inode;
10463
10464 /*
10465 * We set number of links to 0 in btrfs_new_inode(), and here we set
10466 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10467 * through:
10468 *
10469 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10470 */
10471 set_nlink(inode, 1);
10472 unlock_new_inode(inode);
10473 d_tmpfile(dentry, inode);
10474 mark_inode_dirty(inode);
10475
10476 out:
10477 btrfs_end_transaction(trans, root);
10478 if (ret)
10479 iput(inode);
10480 btrfs_balance_delayed_items(root);
10481 btrfs_btree_balance_dirty(root);
10482 return ret;
10483
10484 out_inode:
10485 unlock_new_inode(inode);
10486 goto out;
10487
10488 }
10489
10490 /* Inspired by filemap_check_errors() */
10491 int btrfs_inode_check_errors(struct inode *inode)
10492 {
10493 int ret = 0;
10494
10495 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10496 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10497 ret = -ENOSPC;
10498 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10499 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10500 ret = -EIO;
10501
10502 return ret;
10503 }
10504
10505 static const struct inode_operations btrfs_dir_inode_operations = {
10506 .getattr = btrfs_getattr,
10507 .lookup = btrfs_lookup,
10508 .create = btrfs_create,
10509 .unlink = btrfs_unlink,
10510 .link = btrfs_link,
10511 .mkdir = btrfs_mkdir,
10512 .rmdir = btrfs_rmdir,
10513 .rename2 = btrfs_rename2,
10514 .symlink = btrfs_symlink,
10515 .setattr = btrfs_setattr,
10516 .mknod = btrfs_mknod,
10517 .setxattr = btrfs_setxattr,
10518 .getxattr = generic_getxattr,
10519 .listxattr = btrfs_listxattr,
10520 .removexattr = btrfs_removexattr,
10521 .permission = btrfs_permission,
10522 .get_acl = btrfs_get_acl,
10523 .set_acl = btrfs_set_acl,
10524 .update_time = btrfs_update_time,
10525 .tmpfile = btrfs_tmpfile,
10526 };
10527 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10528 .lookup = btrfs_lookup,
10529 .permission = btrfs_permission,
10530 .get_acl = btrfs_get_acl,
10531 .set_acl = btrfs_set_acl,
10532 .update_time = btrfs_update_time,
10533 };
10534
10535 static const struct file_operations btrfs_dir_file_operations = {
10536 .llseek = generic_file_llseek,
10537 .read = generic_read_dir,
10538 .iterate = btrfs_real_readdir,
10539 .unlocked_ioctl = btrfs_ioctl,
10540 #ifdef CONFIG_COMPAT
10541 .compat_ioctl = btrfs_compat_ioctl,
10542 #endif
10543 .release = btrfs_release_file,
10544 .fsync = btrfs_sync_file,
10545 };
10546
10547 static const struct extent_io_ops btrfs_extent_io_ops = {
10548 .fill_delalloc = run_delalloc_range,
10549 .submit_bio_hook = btrfs_submit_bio_hook,
10550 .merge_bio_hook = btrfs_merge_bio_hook,
10551 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10552 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10553 .writepage_start_hook = btrfs_writepage_start_hook,
10554 .set_bit_hook = btrfs_set_bit_hook,
10555 .clear_bit_hook = btrfs_clear_bit_hook,
10556 .merge_extent_hook = btrfs_merge_extent_hook,
10557 .split_extent_hook = btrfs_split_extent_hook,
10558 };
10559
10560 /*
10561 * btrfs doesn't support the bmap operation because swapfiles
10562 * use bmap to make a mapping of extents in the file. They assume
10563 * these extents won't change over the life of the file and they
10564 * use the bmap result to do IO directly to the drive.
10565 *
10566 * the btrfs bmap call would return logical addresses that aren't
10567 * suitable for IO and they also will change frequently as COW
10568 * operations happen. So, swapfile + btrfs == corruption.
10569 *
10570 * For now we're avoiding this by dropping bmap.
10571 */
10572 static const struct address_space_operations btrfs_aops = {
10573 .readpage = btrfs_readpage,
10574 .writepage = btrfs_writepage,
10575 .writepages = btrfs_writepages,
10576 .readpages = btrfs_readpages,
10577 .direct_IO = btrfs_direct_IO,
10578 .invalidatepage = btrfs_invalidatepage,
10579 .releasepage = btrfs_releasepage,
10580 .set_page_dirty = btrfs_set_page_dirty,
10581 .error_remove_page = generic_error_remove_page,
10582 };
10583
10584 static const struct address_space_operations btrfs_symlink_aops = {
10585 .readpage = btrfs_readpage,
10586 .writepage = btrfs_writepage,
10587 .invalidatepage = btrfs_invalidatepage,
10588 .releasepage = btrfs_releasepage,
10589 };
10590
10591 static const struct inode_operations btrfs_file_inode_operations = {
10592 .getattr = btrfs_getattr,
10593 .setattr = btrfs_setattr,
10594 .setxattr = btrfs_setxattr,
10595 .getxattr = generic_getxattr,
10596 .listxattr = btrfs_listxattr,
10597 .removexattr = btrfs_removexattr,
10598 .permission = btrfs_permission,
10599 .fiemap = btrfs_fiemap,
10600 .get_acl = btrfs_get_acl,
10601 .set_acl = btrfs_set_acl,
10602 .update_time = btrfs_update_time,
10603 };
10604 static const struct inode_operations btrfs_special_inode_operations = {
10605 .getattr = btrfs_getattr,
10606 .setattr = btrfs_setattr,
10607 .permission = btrfs_permission,
10608 .setxattr = btrfs_setxattr,
10609 .getxattr = generic_getxattr,
10610 .listxattr = btrfs_listxattr,
10611 .removexattr = btrfs_removexattr,
10612 .get_acl = btrfs_get_acl,
10613 .set_acl = btrfs_set_acl,
10614 .update_time = btrfs_update_time,
10615 };
10616 static const struct inode_operations btrfs_symlink_inode_operations = {
10617 .readlink = generic_readlink,
10618 .get_link = page_get_link,
10619 .getattr = btrfs_getattr,
10620 .setattr = btrfs_setattr,
10621 .permission = btrfs_permission,
10622 .setxattr = btrfs_setxattr,
10623 .getxattr = generic_getxattr,
10624 .listxattr = btrfs_listxattr,
10625 .removexattr = btrfs_removexattr,
10626 .update_time = btrfs_update_time,
10627 };
10628
10629 const struct dentry_operations btrfs_dentry_operations = {
10630 .d_delete = btrfs_dentry_delete,
10631 .d_release = btrfs_dentry_release,
10632 };
Linux kernel - Deliverables
This page took 0.405011 seconds and 5 git commands to generate.