]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/btrfs/inode.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'upstream-4.8-rc1' of git://git.infradead.org/linux-ubifs
[mirror_ubuntu-zesty-kernel.git] / fs / btrfs / inode.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/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(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, bio_op(bio), 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, struct bio *bio,
1859 int mirror_num, unsigned long bio_flags,
1860 u64 bio_offset)
1861 {
1862 struct btrfs_root *root = BTRFS_I(inode)->root;
1863 int ret = 0;
1864
1865 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1866 BUG_ON(ret); /* -ENOMEM */
1867 return 0;
1868 }
1869
1870 /*
1871 * in order to insert checksums into the metadata in large chunks,
1872 * we wait until bio submission time. All the pages in the bio are
1873 * checksummed and sums are attached onto the ordered extent record.
1874 *
1875 * At IO completion time the cums attached on the ordered extent record
1876 * are inserted into the btree
1877 */
1878 static int __btrfs_submit_bio_done(struct inode *inode, struct bio *bio,
1879 int mirror_num, unsigned long bio_flags,
1880 u64 bio_offset)
1881 {
1882 struct btrfs_root *root = BTRFS_I(inode)->root;
1883 int ret;
1884
1885 ret = btrfs_map_bio(root, bio, mirror_num, 1);
1886 if (ret) {
1887 bio->bi_error = ret;
1888 bio_endio(bio);
1889 }
1890 return ret;
1891 }
1892
1893 /*
1894 * extent_io.c submission hook. This does the right thing for csum calculation
1895 * on write, or reading the csums from the tree before a read
1896 */
1897 static int btrfs_submit_bio_hook(struct inode *inode, struct bio *bio,
1898 int mirror_num, unsigned long bio_flags,
1899 u64 bio_offset)
1900 {
1901 struct btrfs_root *root = BTRFS_I(inode)->root;
1902 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1903 int ret = 0;
1904 int skip_sum;
1905 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1906
1907 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1908
1909 if (btrfs_is_free_space_inode(inode))
1910 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1911
1912 if (bio_op(bio) != REQ_OP_WRITE) {
1913 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1914 if (ret)
1915 goto out;
1916
1917 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1918 ret = btrfs_submit_compressed_read(inode, bio,
1919 mirror_num,
1920 bio_flags);
1921 goto out;
1922 } else if (!skip_sum) {
1923 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1924 if (ret)
1925 goto out;
1926 }
1927 goto mapit;
1928 } else if (async && !skip_sum) {
1929 /* csum items have already been cloned */
1930 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1931 goto mapit;
1932 /* we're doing a write, do the async checksumming */
1933 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1934 inode, bio, mirror_num,
1935 bio_flags, bio_offset,
1936 __btrfs_submit_bio_start,
1937 __btrfs_submit_bio_done);
1938 goto out;
1939 } else if (!skip_sum) {
1940 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1941 if (ret)
1942 goto out;
1943 }
1944
1945 mapit:
1946 ret = btrfs_map_bio(root, bio, mirror_num, 0);
1947
1948 out:
1949 if (ret < 0) {
1950 bio->bi_error = ret;
1951 bio_endio(bio);
1952 }
1953 return ret;
1954 }
1955
1956 /*
1957 * given a list of ordered sums record them in the inode. This happens
1958 * at IO completion time based on sums calculated at bio submission time.
1959 */
1960 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1961 struct inode *inode, u64 file_offset,
1962 struct list_head *list)
1963 {
1964 struct btrfs_ordered_sum *sum;
1965
1966 list_for_each_entry(sum, list, list) {
1967 trans->adding_csums = 1;
1968 btrfs_csum_file_blocks(trans,
1969 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1970 trans->adding_csums = 0;
1971 }
1972 return 0;
1973 }
1974
1975 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1976 struct extent_state **cached_state)
1977 {
1978 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
1979 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1980 cached_state);
1981 }
1982
1983 /* see btrfs_writepage_start_hook for details on why this is required */
1984 struct btrfs_writepage_fixup {
1985 struct page *page;
1986 struct btrfs_work work;
1987 };
1988
1989 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1990 {
1991 struct btrfs_writepage_fixup *fixup;
1992 struct btrfs_ordered_extent *ordered;
1993 struct extent_state *cached_state = NULL;
1994 struct page *page;
1995 struct inode *inode;
1996 u64 page_start;
1997 u64 page_end;
1998 int ret;
1999
2000 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2001 page = fixup->page;
2002 again:
2003 lock_page(page);
2004 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2005 ClearPageChecked(page);
2006 goto out_page;
2007 }
2008
2009 inode = page->mapping->host;
2010 page_start = page_offset(page);
2011 page_end = page_offset(page) + PAGE_SIZE - 1;
2012
2013 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2014 &cached_state);
2015
2016 /* already ordered? We're done */
2017 if (PagePrivate2(page))
2018 goto out;
2019
2020 ordered = btrfs_lookup_ordered_range(inode, page_start,
2021 PAGE_SIZE);
2022 if (ordered) {
2023 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2024 page_end, &cached_state, GFP_NOFS);
2025 unlock_page(page);
2026 btrfs_start_ordered_extent(inode, ordered, 1);
2027 btrfs_put_ordered_extent(ordered);
2028 goto again;
2029 }
2030
2031 ret = btrfs_delalloc_reserve_space(inode, page_start,
2032 PAGE_SIZE);
2033 if (ret) {
2034 mapping_set_error(page->mapping, ret);
2035 end_extent_writepage(page, ret, page_start, page_end);
2036 ClearPageChecked(page);
2037 goto out;
2038 }
2039
2040 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2041 ClearPageChecked(page);
2042 set_page_dirty(page);
2043 out:
2044 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2045 &cached_state, GFP_NOFS);
2046 out_page:
2047 unlock_page(page);
2048 put_page(page);
2049 kfree(fixup);
2050 }
2051
2052 /*
2053 * There are a few paths in the higher layers of the kernel that directly
2054 * set the page dirty bit without asking the filesystem if it is a
2055 * good idea. This causes problems because we want to make sure COW
2056 * properly happens and the data=ordered rules are followed.
2057 *
2058 * In our case any range that doesn't have the ORDERED bit set
2059 * hasn't been properly setup for IO. We kick off an async process
2060 * to fix it up. The async helper will wait for ordered extents, set
2061 * the delalloc bit and make it safe to write the page.
2062 */
2063 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2064 {
2065 struct inode *inode = page->mapping->host;
2066 struct btrfs_writepage_fixup *fixup;
2067 struct btrfs_root *root = BTRFS_I(inode)->root;
2068
2069 /* this page is properly in the ordered list */
2070 if (TestClearPagePrivate2(page))
2071 return 0;
2072
2073 if (PageChecked(page))
2074 return -EAGAIN;
2075
2076 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2077 if (!fixup)
2078 return -EAGAIN;
2079
2080 SetPageChecked(page);
2081 get_page(page);
2082 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2083 btrfs_writepage_fixup_worker, NULL, NULL);
2084 fixup->page = page;
2085 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2086 return -EBUSY;
2087 }
2088
2089 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2090 struct inode *inode, u64 file_pos,
2091 u64 disk_bytenr, u64 disk_num_bytes,
2092 u64 num_bytes, u64 ram_bytes,
2093 u8 compression, u8 encryption,
2094 u16 other_encoding, int extent_type)
2095 {
2096 struct btrfs_root *root = BTRFS_I(inode)->root;
2097 struct btrfs_file_extent_item *fi;
2098 struct btrfs_path *path;
2099 struct extent_buffer *leaf;
2100 struct btrfs_key ins;
2101 int extent_inserted = 0;
2102 int ret;
2103
2104 path = btrfs_alloc_path();
2105 if (!path)
2106 return -ENOMEM;
2107
2108 /*
2109 * we may be replacing one extent in the tree with another.
2110 * The new extent is pinned in the extent map, and we don't want
2111 * to drop it from the cache until it is completely in the btree.
2112 *
2113 * So, tell btrfs_drop_extents to leave this extent in the cache.
2114 * the caller is expected to unpin it and allow it to be merged
2115 * with the others.
2116 */
2117 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2118 file_pos + num_bytes, NULL, 0,
2119 1, sizeof(*fi), &extent_inserted);
2120 if (ret)
2121 goto out;
2122
2123 if (!extent_inserted) {
2124 ins.objectid = btrfs_ino(inode);
2125 ins.offset = file_pos;
2126 ins.type = BTRFS_EXTENT_DATA_KEY;
2127
2128 path->leave_spinning = 1;
2129 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2130 sizeof(*fi));
2131 if (ret)
2132 goto out;
2133 }
2134 leaf = path->nodes[0];
2135 fi = btrfs_item_ptr(leaf, path->slots[0],
2136 struct btrfs_file_extent_item);
2137 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2138 btrfs_set_file_extent_type(leaf, fi, extent_type);
2139 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2140 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2141 btrfs_set_file_extent_offset(leaf, fi, 0);
2142 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2143 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2144 btrfs_set_file_extent_compression(leaf, fi, compression);
2145 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2146 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2147
2148 btrfs_mark_buffer_dirty(leaf);
2149 btrfs_release_path(path);
2150
2151 inode_add_bytes(inode, num_bytes);
2152
2153 ins.objectid = disk_bytenr;
2154 ins.offset = disk_num_bytes;
2155 ins.type = BTRFS_EXTENT_ITEM_KEY;
2156 ret = btrfs_alloc_reserved_file_extent(trans, root,
2157 root->root_key.objectid,
2158 btrfs_ino(inode), file_pos,
2159 ram_bytes, &ins);
2160 /*
2161 * Release the reserved range from inode dirty range map, as it is
2162 * already moved into delayed_ref_head
2163 */
2164 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2165 out:
2166 btrfs_free_path(path);
2167
2168 return ret;
2169 }
2170
2171 /* snapshot-aware defrag */
2172 struct sa_defrag_extent_backref {
2173 struct rb_node node;
2174 struct old_sa_defrag_extent *old;
2175 u64 root_id;
2176 u64 inum;
2177 u64 file_pos;
2178 u64 extent_offset;
2179 u64 num_bytes;
2180 u64 generation;
2181 };
2182
2183 struct old_sa_defrag_extent {
2184 struct list_head list;
2185 struct new_sa_defrag_extent *new;
2186
2187 u64 extent_offset;
2188 u64 bytenr;
2189 u64 offset;
2190 u64 len;
2191 int count;
2192 };
2193
2194 struct new_sa_defrag_extent {
2195 struct rb_root root;
2196 struct list_head head;
2197 struct btrfs_path *path;
2198 struct inode *inode;
2199 u64 file_pos;
2200 u64 len;
2201 u64 bytenr;
2202 u64 disk_len;
2203 u8 compress_type;
2204 };
2205
2206 static int backref_comp(struct sa_defrag_extent_backref *b1,
2207 struct sa_defrag_extent_backref *b2)
2208 {
2209 if (b1->root_id < b2->root_id)
2210 return -1;
2211 else if (b1->root_id > b2->root_id)
2212 return 1;
2213
2214 if (b1->inum < b2->inum)
2215 return -1;
2216 else if (b1->inum > b2->inum)
2217 return 1;
2218
2219 if (b1->file_pos < b2->file_pos)
2220 return -1;
2221 else if (b1->file_pos > b2->file_pos)
2222 return 1;
2223
2224 /*
2225 * [------------------------------] ===> (a range of space)
2226 * |<--->| |<---->| =============> (fs/file tree A)
2227 * |<---------------------------->| ===> (fs/file tree B)
2228 *
2229 * A range of space can refer to two file extents in one tree while
2230 * refer to only one file extent in another tree.
2231 *
2232 * So we may process a disk offset more than one time(two extents in A)
2233 * and locate at the same extent(one extent in B), then insert two same
2234 * backrefs(both refer to the extent in B).
2235 */
2236 return 0;
2237 }
2238
2239 static void backref_insert(struct rb_root *root,
2240 struct sa_defrag_extent_backref *backref)
2241 {
2242 struct rb_node **p = &root->rb_node;
2243 struct rb_node *parent = NULL;
2244 struct sa_defrag_extent_backref *entry;
2245 int ret;
2246
2247 while (*p) {
2248 parent = *p;
2249 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2250
2251 ret = backref_comp(backref, entry);
2252 if (ret < 0)
2253 p = &(*p)->rb_left;
2254 else
2255 p = &(*p)->rb_right;
2256 }
2257
2258 rb_link_node(&backref->node, parent, p);
2259 rb_insert_color(&backref->node, root);
2260 }
2261
2262 /*
2263 * Note the backref might has changed, and in this case we just return 0.
2264 */
2265 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2266 void *ctx)
2267 {
2268 struct btrfs_file_extent_item *extent;
2269 struct btrfs_fs_info *fs_info;
2270 struct old_sa_defrag_extent *old = ctx;
2271 struct new_sa_defrag_extent *new = old->new;
2272 struct btrfs_path *path = new->path;
2273 struct btrfs_key key;
2274 struct btrfs_root *root;
2275 struct sa_defrag_extent_backref *backref;
2276 struct extent_buffer *leaf;
2277 struct inode *inode = new->inode;
2278 int slot;
2279 int ret;
2280 u64 extent_offset;
2281 u64 num_bytes;
2282
2283 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2284 inum == btrfs_ino(inode))
2285 return 0;
2286
2287 key.objectid = root_id;
2288 key.type = BTRFS_ROOT_ITEM_KEY;
2289 key.offset = (u64)-1;
2290
2291 fs_info = BTRFS_I(inode)->root->fs_info;
2292 root = btrfs_read_fs_root_no_name(fs_info, &key);
2293 if (IS_ERR(root)) {
2294 if (PTR_ERR(root) == -ENOENT)
2295 return 0;
2296 WARN_ON(1);
2297 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2298 inum, offset, root_id);
2299 return PTR_ERR(root);
2300 }
2301
2302 key.objectid = inum;
2303 key.type = BTRFS_EXTENT_DATA_KEY;
2304 if (offset > (u64)-1 << 32)
2305 key.offset = 0;
2306 else
2307 key.offset = offset;
2308
2309 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2310 if (WARN_ON(ret < 0))
2311 return ret;
2312 ret = 0;
2313
2314 while (1) {
2315 cond_resched();
2316
2317 leaf = path->nodes[0];
2318 slot = path->slots[0];
2319
2320 if (slot >= btrfs_header_nritems(leaf)) {
2321 ret = btrfs_next_leaf(root, path);
2322 if (ret < 0) {
2323 goto out;
2324 } else if (ret > 0) {
2325 ret = 0;
2326 goto out;
2327 }
2328 continue;
2329 }
2330
2331 path->slots[0]++;
2332
2333 btrfs_item_key_to_cpu(leaf, &key, slot);
2334
2335 if (key.objectid > inum)
2336 goto out;
2337
2338 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2339 continue;
2340
2341 extent = btrfs_item_ptr(leaf, slot,
2342 struct btrfs_file_extent_item);
2343
2344 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2345 continue;
2346
2347 /*
2348 * 'offset' refers to the exact key.offset,
2349 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2350 * (key.offset - extent_offset).
2351 */
2352 if (key.offset != offset)
2353 continue;
2354
2355 extent_offset = btrfs_file_extent_offset(leaf, extent);
2356 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2357
2358 if (extent_offset >= old->extent_offset + old->offset +
2359 old->len || extent_offset + num_bytes <=
2360 old->extent_offset + old->offset)
2361 continue;
2362 break;
2363 }
2364
2365 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2366 if (!backref) {
2367 ret = -ENOENT;
2368 goto out;
2369 }
2370
2371 backref->root_id = root_id;
2372 backref->inum = inum;
2373 backref->file_pos = offset;
2374 backref->num_bytes = num_bytes;
2375 backref->extent_offset = extent_offset;
2376 backref->generation = btrfs_file_extent_generation(leaf, extent);
2377 backref->old = old;
2378 backref_insert(&new->root, backref);
2379 old->count++;
2380 out:
2381 btrfs_release_path(path);
2382 WARN_ON(ret);
2383 return ret;
2384 }
2385
2386 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2387 struct new_sa_defrag_extent *new)
2388 {
2389 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2390 struct old_sa_defrag_extent *old, *tmp;
2391 int ret;
2392
2393 new->path = path;
2394
2395 list_for_each_entry_safe(old, tmp, &new->head, list) {
2396 ret = iterate_inodes_from_logical(old->bytenr +
2397 old->extent_offset, fs_info,
2398 path, record_one_backref,
2399 old);
2400 if (ret < 0 && ret != -ENOENT)
2401 return false;
2402
2403 /* no backref to be processed for this extent */
2404 if (!old->count) {
2405 list_del(&old->list);
2406 kfree(old);
2407 }
2408 }
2409
2410 if (list_empty(&new->head))
2411 return false;
2412
2413 return true;
2414 }
2415
2416 static int relink_is_mergable(struct extent_buffer *leaf,
2417 struct btrfs_file_extent_item *fi,
2418 struct new_sa_defrag_extent *new)
2419 {
2420 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2421 return 0;
2422
2423 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2424 return 0;
2425
2426 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2427 return 0;
2428
2429 if (btrfs_file_extent_encryption(leaf, fi) ||
2430 btrfs_file_extent_other_encoding(leaf, fi))
2431 return 0;
2432
2433 return 1;
2434 }
2435
2436 /*
2437 * Note the backref might has changed, and in this case we just return 0.
2438 */
2439 static noinline int relink_extent_backref(struct btrfs_path *path,
2440 struct sa_defrag_extent_backref *prev,
2441 struct sa_defrag_extent_backref *backref)
2442 {
2443 struct btrfs_file_extent_item *extent;
2444 struct btrfs_file_extent_item *item;
2445 struct btrfs_ordered_extent *ordered;
2446 struct btrfs_trans_handle *trans;
2447 struct btrfs_fs_info *fs_info;
2448 struct btrfs_root *root;
2449 struct btrfs_key key;
2450 struct extent_buffer *leaf;
2451 struct old_sa_defrag_extent *old = backref->old;
2452 struct new_sa_defrag_extent *new = old->new;
2453 struct inode *src_inode = new->inode;
2454 struct inode *inode;
2455 struct extent_state *cached = NULL;
2456 int ret = 0;
2457 u64 start;
2458 u64 len;
2459 u64 lock_start;
2460 u64 lock_end;
2461 bool merge = false;
2462 int index;
2463
2464 if (prev && prev->root_id == backref->root_id &&
2465 prev->inum == backref->inum &&
2466 prev->file_pos + prev->num_bytes == backref->file_pos)
2467 merge = true;
2468
2469 /* step 1: get root */
2470 key.objectid = backref->root_id;
2471 key.type = BTRFS_ROOT_ITEM_KEY;
2472 key.offset = (u64)-1;
2473
2474 fs_info = BTRFS_I(src_inode)->root->fs_info;
2475 index = srcu_read_lock(&fs_info->subvol_srcu);
2476
2477 root = btrfs_read_fs_root_no_name(fs_info, &key);
2478 if (IS_ERR(root)) {
2479 srcu_read_unlock(&fs_info->subvol_srcu, index);
2480 if (PTR_ERR(root) == -ENOENT)
2481 return 0;
2482 return PTR_ERR(root);
2483 }
2484
2485 if (btrfs_root_readonly(root)) {
2486 srcu_read_unlock(&fs_info->subvol_srcu, index);
2487 return 0;
2488 }
2489
2490 /* step 2: get inode */
2491 key.objectid = backref->inum;
2492 key.type = BTRFS_INODE_ITEM_KEY;
2493 key.offset = 0;
2494
2495 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2496 if (IS_ERR(inode)) {
2497 srcu_read_unlock(&fs_info->subvol_srcu, index);
2498 return 0;
2499 }
2500
2501 srcu_read_unlock(&fs_info->subvol_srcu, index);
2502
2503 /* step 3: relink backref */
2504 lock_start = backref->file_pos;
2505 lock_end = backref->file_pos + backref->num_bytes - 1;
2506 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2507 &cached);
2508
2509 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2510 if (ordered) {
2511 btrfs_put_ordered_extent(ordered);
2512 goto out_unlock;
2513 }
2514
2515 trans = btrfs_join_transaction(root);
2516 if (IS_ERR(trans)) {
2517 ret = PTR_ERR(trans);
2518 goto out_unlock;
2519 }
2520
2521 key.objectid = backref->inum;
2522 key.type = BTRFS_EXTENT_DATA_KEY;
2523 key.offset = backref->file_pos;
2524
2525 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2526 if (ret < 0) {
2527 goto out_free_path;
2528 } else if (ret > 0) {
2529 ret = 0;
2530 goto out_free_path;
2531 }
2532
2533 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2534 struct btrfs_file_extent_item);
2535
2536 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2537 backref->generation)
2538 goto out_free_path;
2539
2540 btrfs_release_path(path);
2541
2542 start = backref->file_pos;
2543 if (backref->extent_offset < old->extent_offset + old->offset)
2544 start += old->extent_offset + old->offset -
2545 backref->extent_offset;
2546
2547 len = min(backref->extent_offset + backref->num_bytes,
2548 old->extent_offset + old->offset + old->len);
2549 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2550
2551 ret = btrfs_drop_extents(trans, root, inode, start,
2552 start + len, 1);
2553 if (ret)
2554 goto out_free_path;
2555 again:
2556 key.objectid = btrfs_ino(inode);
2557 key.type = BTRFS_EXTENT_DATA_KEY;
2558 key.offset = start;
2559
2560 path->leave_spinning = 1;
2561 if (merge) {
2562 struct btrfs_file_extent_item *fi;
2563 u64 extent_len;
2564 struct btrfs_key found_key;
2565
2566 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2567 if (ret < 0)
2568 goto out_free_path;
2569
2570 path->slots[0]--;
2571 leaf = path->nodes[0];
2572 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2573
2574 fi = btrfs_item_ptr(leaf, path->slots[0],
2575 struct btrfs_file_extent_item);
2576 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2577
2578 if (extent_len + found_key.offset == start &&
2579 relink_is_mergable(leaf, fi, new)) {
2580 btrfs_set_file_extent_num_bytes(leaf, fi,
2581 extent_len + len);
2582 btrfs_mark_buffer_dirty(leaf);
2583 inode_add_bytes(inode, len);
2584
2585 ret = 1;
2586 goto out_free_path;
2587 } else {
2588 merge = false;
2589 btrfs_release_path(path);
2590 goto again;
2591 }
2592 }
2593
2594 ret = btrfs_insert_empty_item(trans, root, path, &key,
2595 sizeof(*extent));
2596 if (ret) {
2597 btrfs_abort_transaction(trans, root, ret);
2598 goto out_free_path;
2599 }
2600
2601 leaf = path->nodes[0];
2602 item = btrfs_item_ptr(leaf, path->slots[0],
2603 struct btrfs_file_extent_item);
2604 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2605 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2606 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2607 btrfs_set_file_extent_num_bytes(leaf, item, len);
2608 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2609 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2610 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2611 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2612 btrfs_set_file_extent_encryption(leaf, item, 0);
2613 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2614
2615 btrfs_mark_buffer_dirty(leaf);
2616 inode_add_bytes(inode, len);
2617 btrfs_release_path(path);
2618
2619 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2620 new->disk_len, 0,
2621 backref->root_id, backref->inum,
2622 new->file_pos); /* start - extent_offset */
2623 if (ret) {
2624 btrfs_abort_transaction(trans, root, ret);
2625 goto out_free_path;
2626 }
2627
2628 ret = 1;
2629 out_free_path:
2630 btrfs_release_path(path);
2631 path->leave_spinning = 0;
2632 btrfs_end_transaction(trans, root);
2633 out_unlock:
2634 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2635 &cached, GFP_NOFS);
2636 iput(inode);
2637 return ret;
2638 }
2639
2640 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2641 {
2642 struct old_sa_defrag_extent *old, *tmp;
2643
2644 if (!new)
2645 return;
2646
2647 list_for_each_entry_safe(old, tmp, &new->head, list) {
2648 kfree(old);
2649 }
2650 kfree(new);
2651 }
2652
2653 static void relink_file_extents(struct new_sa_defrag_extent *new)
2654 {
2655 struct btrfs_path *path;
2656 struct sa_defrag_extent_backref *backref;
2657 struct sa_defrag_extent_backref *prev = NULL;
2658 struct inode *inode;
2659 struct btrfs_root *root;
2660 struct rb_node *node;
2661 int ret;
2662
2663 inode = new->inode;
2664 root = BTRFS_I(inode)->root;
2665
2666 path = btrfs_alloc_path();
2667 if (!path)
2668 return;
2669
2670 if (!record_extent_backrefs(path, new)) {
2671 btrfs_free_path(path);
2672 goto out;
2673 }
2674 btrfs_release_path(path);
2675
2676 while (1) {
2677 node = rb_first(&new->root);
2678 if (!node)
2679 break;
2680 rb_erase(node, &new->root);
2681
2682 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2683
2684 ret = relink_extent_backref(path, prev, backref);
2685 WARN_ON(ret < 0);
2686
2687 kfree(prev);
2688
2689 if (ret == 1)
2690 prev = backref;
2691 else
2692 prev = NULL;
2693 cond_resched();
2694 }
2695 kfree(prev);
2696
2697 btrfs_free_path(path);
2698 out:
2699 free_sa_defrag_extent(new);
2700
2701 atomic_dec(&root->fs_info->defrag_running);
2702 wake_up(&root->fs_info->transaction_wait);
2703 }
2704
2705 static struct new_sa_defrag_extent *
2706 record_old_file_extents(struct inode *inode,
2707 struct btrfs_ordered_extent *ordered)
2708 {
2709 struct btrfs_root *root = BTRFS_I(inode)->root;
2710 struct btrfs_path *path;
2711 struct btrfs_key key;
2712 struct old_sa_defrag_extent *old;
2713 struct new_sa_defrag_extent *new;
2714 int ret;
2715
2716 new = kmalloc(sizeof(*new), GFP_NOFS);
2717 if (!new)
2718 return NULL;
2719
2720 new->inode = inode;
2721 new->file_pos = ordered->file_offset;
2722 new->len = ordered->len;
2723 new->bytenr = ordered->start;
2724 new->disk_len = ordered->disk_len;
2725 new->compress_type = ordered->compress_type;
2726 new->root = RB_ROOT;
2727 INIT_LIST_HEAD(&new->head);
2728
2729 path = btrfs_alloc_path();
2730 if (!path)
2731 goto out_kfree;
2732
2733 key.objectid = btrfs_ino(inode);
2734 key.type = BTRFS_EXTENT_DATA_KEY;
2735 key.offset = new->file_pos;
2736
2737 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2738 if (ret < 0)
2739 goto out_free_path;
2740 if (ret > 0 && path->slots[0] > 0)
2741 path->slots[0]--;
2742
2743 /* find out all the old extents for the file range */
2744 while (1) {
2745 struct btrfs_file_extent_item *extent;
2746 struct extent_buffer *l;
2747 int slot;
2748 u64 num_bytes;
2749 u64 offset;
2750 u64 end;
2751 u64 disk_bytenr;
2752 u64 extent_offset;
2753
2754 l = path->nodes[0];
2755 slot = path->slots[0];
2756
2757 if (slot >= btrfs_header_nritems(l)) {
2758 ret = btrfs_next_leaf(root, path);
2759 if (ret < 0)
2760 goto out_free_path;
2761 else if (ret > 0)
2762 break;
2763 continue;
2764 }
2765
2766 btrfs_item_key_to_cpu(l, &key, slot);
2767
2768 if (key.objectid != btrfs_ino(inode))
2769 break;
2770 if (key.type != BTRFS_EXTENT_DATA_KEY)
2771 break;
2772 if (key.offset >= new->file_pos + new->len)
2773 break;
2774
2775 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2776
2777 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2778 if (key.offset + num_bytes < new->file_pos)
2779 goto next;
2780
2781 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2782 if (!disk_bytenr)
2783 goto next;
2784
2785 extent_offset = btrfs_file_extent_offset(l, extent);
2786
2787 old = kmalloc(sizeof(*old), GFP_NOFS);
2788 if (!old)
2789 goto out_free_path;
2790
2791 offset = max(new->file_pos, key.offset);
2792 end = min(new->file_pos + new->len, key.offset + num_bytes);
2793
2794 old->bytenr = disk_bytenr;
2795 old->extent_offset = extent_offset;
2796 old->offset = offset - key.offset;
2797 old->len = end - offset;
2798 old->new = new;
2799 old->count = 0;
2800 list_add_tail(&old->list, &new->head);
2801 next:
2802 path->slots[0]++;
2803 cond_resched();
2804 }
2805
2806 btrfs_free_path(path);
2807 atomic_inc(&root->fs_info->defrag_running);
2808
2809 return new;
2810
2811 out_free_path:
2812 btrfs_free_path(path);
2813 out_kfree:
2814 free_sa_defrag_extent(new);
2815 return NULL;
2816 }
2817
2818 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2819 u64 start, u64 len)
2820 {
2821 struct btrfs_block_group_cache *cache;
2822
2823 cache = btrfs_lookup_block_group(root->fs_info, start);
2824 ASSERT(cache);
2825
2826 spin_lock(&cache->lock);
2827 cache->delalloc_bytes -= len;
2828 spin_unlock(&cache->lock);
2829
2830 btrfs_put_block_group(cache);
2831 }
2832
2833 /* as ordered data IO finishes, this gets called so we can finish
2834 * an ordered extent if the range of bytes in the file it covers are
2835 * fully written.
2836 */
2837 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2838 {
2839 struct inode *inode = ordered_extent->inode;
2840 struct btrfs_root *root = BTRFS_I(inode)->root;
2841 struct btrfs_trans_handle *trans = NULL;
2842 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2843 struct extent_state *cached_state = NULL;
2844 struct new_sa_defrag_extent *new = NULL;
2845 int compress_type = 0;
2846 int ret = 0;
2847 u64 logical_len = ordered_extent->len;
2848 bool nolock;
2849 bool truncated = false;
2850
2851 nolock = btrfs_is_free_space_inode(inode);
2852
2853 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2854 ret = -EIO;
2855 goto out;
2856 }
2857
2858 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2859 ordered_extent->file_offset +
2860 ordered_extent->len - 1);
2861
2862 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2863 truncated = true;
2864 logical_len = ordered_extent->truncated_len;
2865 /* Truncated the entire extent, don't bother adding */
2866 if (!logical_len)
2867 goto out;
2868 }
2869
2870 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2871 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2872
2873 /*
2874 * For mwrite(mmap + memset to write) case, we still reserve
2875 * space for NOCOW range.
2876 * As NOCOW won't cause a new delayed ref, just free the space
2877 */
2878 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2879 ordered_extent->len);
2880 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2881 if (nolock)
2882 trans = btrfs_join_transaction_nolock(root);
2883 else
2884 trans = btrfs_join_transaction(root);
2885 if (IS_ERR(trans)) {
2886 ret = PTR_ERR(trans);
2887 trans = NULL;
2888 goto out;
2889 }
2890 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2891 ret = btrfs_update_inode_fallback(trans, root, inode);
2892 if (ret) /* -ENOMEM or corruption */
2893 btrfs_abort_transaction(trans, root, ret);
2894 goto out;
2895 }
2896
2897 lock_extent_bits(io_tree, ordered_extent->file_offset,
2898 ordered_extent->file_offset + ordered_extent->len - 1,
2899 &cached_state);
2900
2901 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2902 ordered_extent->file_offset + ordered_extent->len - 1,
2903 EXTENT_DEFRAG, 1, cached_state);
2904 if (ret) {
2905 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2906 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2907 /* the inode is shared */
2908 new = record_old_file_extents(inode, ordered_extent);
2909
2910 clear_extent_bit(io_tree, ordered_extent->file_offset,
2911 ordered_extent->file_offset + ordered_extent->len - 1,
2912 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2913 }
2914
2915 if (nolock)
2916 trans = btrfs_join_transaction_nolock(root);
2917 else
2918 trans = btrfs_join_transaction(root);
2919 if (IS_ERR(trans)) {
2920 ret = PTR_ERR(trans);
2921 trans = NULL;
2922 goto out_unlock;
2923 }
2924
2925 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2926
2927 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2928 compress_type = ordered_extent->compress_type;
2929 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2930 BUG_ON(compress_type);
2931 ret = btrfs_mark_extent_written(trans, inode,
2932 ordered_extent->file_offset,
2933 ordered_extent->file_offset +
2934 logical_len);
2935 } else {
2936 BUG_ON(root == root->fs_info->tree_root);
2937 ret = insert_reserved_file_extent(trans, inode,
2938 ordered_extent->file_offset,
2939 ordered_extent->start,
2940 ordered_extent->disk_len,
2941 logical_len, logical_len,
2942 compress_type, 0, 0,
2943 BTRFS_FILE_EXTENT_REG);
2944 if (!ret)
2945 btrfs_release_delalloc_bytes(root,
2946 ordered_extent->start,
2947 ordered_extent->disk_len);
2948 }
2949 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2950 ordered_extent->file_offset, ordered_extent->len,
2951 trans->transid);
2952 if (ret < 0) {
2953 btrfs_abort_transaction(trans, root, ret);
2954 goto out_unlock;
2955 }
2956
2957 add_pending_csums(trans, inode, ordered_extent->file_offset,
2958 &ordered_extent->list);
2959
2960 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2961 ret = btrfs_update_inode_fallback(trans, root, inode);
2962 if (ret) { /* -ENOMEM or corruption */
2963 btrfs_abort_transaction(trans, root, ret);
2964 goto out_unlock;
2965 }
2966 ret = 0;
2967 out_unlock:
2968 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2969 ordered_extent->file_offset +
2970 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2971 out:
2972 if (root != root->fs_info->tree_root)
2973 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2974 if (trans)
2975 btrfs_end_transaction(trans, root);
2976
2977 if (ret || truncated) {
2978 u64 start, end;
2979
2980 if (truncated)
2981 start = ordered_extent->file_offset + logical_len;
2982 else
2983 start = ordered_extent->file_offset;
2984 end = ordered_extent->file_offset + ordered_extent->len - 1;
2985 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2986
2987 /* Drop the cache for the part of the extent we didn't write. */
2988 btrfs_drop_extent_cache(inode, start, end, 0);
2989
2990 /*
2991 * If the ordered extent had an IOERR or something else went
2992 * wrong we need to return the space for this ordered extent
2993 * back to the allocator. We only free the extent in the
2994 * truncated case if we didn't write out the extent at all.
2995 */
2996 if ((ret || !logical_len) &&
2997 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2998 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2999 btrfs_free_reserved_extent(root, ordered_extent->start,
3000 ordered_extent->disk_len, 1);
3001 }
3002
3003
3004 /*
3005 * This needs to be done to make sure anybody waiting knows we are done
3006 * updating everything for this ordered extent.
3007 */
3008 btrfs_remove_ordered_extent(inode, ordered_extent);
3009
3010 /* for snapshot-aware defrag */
3011 if (new) {
3012 if (ret) {
3013 free_sa_defrag_extent(new);
3014 atomic_dec(&root->fs_info->defrag_running);
3015 } else {
3016 relink_file_extents(new);
3017 }
3018 }
3019
3020 /* once for us */
3021 btrfs_put_ordered_extent(ordered_extent);
3022 /* once for the tree */
3023 btrfs_put_ordered_extent(ordered_extent);
3024
3025 return ret;
3026 }
3027
3028 static void finish_ordered_fn(struct btrfs_work *work)
3029 {
3030 struct btrfs_ordered_extent *ordered_extent;
3031 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3032 btrfs_finish_ordered_io(ordered_extent);
3033 }
3034
3035 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3036 struct extent_state *state, int uptodate)
3037 {
3038 struct inode *inode = page->mapping->host;
3039 struct btrfs_root *root = BTRFS_I(inode)->root;
3040 struct btrfs_ordered_extent *ordered_extent = NULL;
3041 struct btrfs_workqueue *wq;
3042 btrfs_work_func_t func;
3043
3044 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3045
3046 ClearPagePrivate2(page);
3047 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3048 end - start + 1, uptodate))
3049 return 0;
3050
3051 if (btrfs_is_free_space_inode(inode)) {
3052 wq = root->fs_info->endio_freespace_worker;
3053 func = btrfs_freespace_write_helper;
3054 } else {
3055 wq = root->fs_info->endio_write_workers;
3056 func = btrfs_endio_write_helper;
3057 }
3058
3059 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3060 NULL);
3061 btrfs_queue_work(wq, &ordered_extent->work);
3062
3063 return 0;
3064 }
3065
3066 static int __readpage_endio_check(struct inode *inode,
3067 struct btrfs_io_bio *io_bio,
3068 int icsum, struct page *page,
3069 int pgoff, u64 start, size_t len)
3070 {
3071 char *kaddr;
3072 u32 csum_expected;
3073 u32 csum = ~(u32)0;
3074
3075 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3076
3077 kaddr = kmap_atomic(page);
3078 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3079 btrfs_csum_final(csum, (char *)&csum);
3080 if (csum != csum_expected)
3081 goto zeroit;
3082
3083 kunmap_atomic(kaddr);
3084 return 0;
3085 zeroit:
3086 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3087 "csum failed ino %llu off %llu csum %u expected csum %u",
3088 btrfs_ino(inode), start, csum, csum_expected);
3089 memset(kaddr + pgoff, 1, len);
3090 flush_dcache_page(page);
3091 kunmap_atomic(kaddr);
3092 if (csum_expected == 0)
3093 return 0;
3094 return -EIO;
3095 }
3096
3097 /*
3098 * when reads are done, we need to check csums to verify the data is correct
3099 * if there's a match, we allow the bio to finish. If not, the code in
3100 * extent_io.c will try to find good copies for us.
3101 */
3102 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3103 u64 phy_offset, struct page *page,
3104 u64 start, u64 end, int mirror)
3105 {
3106 size_t offset = start - page_offset(page);
3107 struct inode *inode = page->mapping->host;
3108 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3109 struct btrfs_root *root = BTRFS_I(inode)->root;
3110
3111 if (PageChecked(page)) {
3112 ClearPageChecked(page);
3113 return 0;
3114 }
3115
3116 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3117 return 0;
3118
3119 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3120 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3121 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3122 return 0;
3123 }
3124
3125 phy_offset >>= inode->i_sb->s_blocksize_bits;
3126 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3127 start, (size_t)(end - start + 1));
3128 }
3129
3130 void btrfs_add_delayed_iput(struct inode *inode)
3131 {
3132 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3133 struct btrfs_inode *binode = BTRFS_I(inode);
3134
3135 if (atomic_add_unless(&inode->i_count, -1, 1))
3136 return;
3137
3138 spin_lock(&fs_info->delayed_iput_lock);
3139 if (binode->delayed_iput_count == 0) {
3140 ASSERT(list_empty(&binode->delayed_iput));
3141 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3142 } else {
3143 binode->delayed_iput_count++;
3144 }
3145 spin_unlock(&fs_info->delayed_iput_lock);
3146 }
3147
3148 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3149 {
3150 struct btrfs_fs_info *fs_info = root->fs_info;
3151
3152 spin_lock(&fs_info->delayed_iput_lock);
3153 while (!list_empty(&fs_info->delayed_iputs)) {
3154 struct btrfs_inode *inode;
3155
3156 inode = list_first_entry(&fs_info->delayed_iputs,
3157 struct btrfs_inode, delayed_iput);
3158 if (inode->delayed_iput_count) {
3159 inode->delayed_iput_count--;
3160 list_move_tail(&inode->delayed_iput,
3161 &fs_info->delayed_iputs);
3162 } else {
3163 list_del_init(&inode->delayed_iput);
3164 }
3165 spin_unlock(&fs_info->delayed_iput_lock);
3166 iput(&inode->vfs_inode);
3167 spin_lock(&fs_info->delayed_iput_lock);
3168 }
3169 spin_unlock(&fs_info->delayed_iput_lock);
3170 }
3171
3172 /*
3173 * This is called in transaction commit time. If there are no orphan
3174 * files in the subvolume, it removes orphan item and frees block_rsv
3175 * structure.
3176 */
3177 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3178 struct btrfs_root *root)
3179 {
3180 struct btrfs_block_rsv *block_rsv;
3181 int ret;
3182
3183 if (atomic_read(&root->orphan_inodes) ||
3184 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3185 return;
3186
3187 spin_lock(&root->orphan_lock);
3188 if (atomic_read(&root->orphan_inodes)) {
3189 spin_unlock(&root->orphan_lock);
3190 return;
3191 }
3192
3193 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3194 spin_unlock(&root->orphan_lock);
3195 return;
3196 }
3197
3198 block_rsv = root->orphan_block_rsv;
3199 root->orphan_block_rsv = NULL;
3200 spin_unlock(&root->orphan_lock);
3201
3202 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3203 btrfs_root_refs(&root->root_item) > 0) {
3204 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3205 root->root_key.objectid);
3206 if (ret)
3207 btrfs_abort_transaction(trans, root, ret);
3208 else
3209 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3210 &root->state);
3211 }
3212
3213 if (block_rsv) {
3214 WARN_ON(block_rsv->size > 0);
3215 btrfs_free_block_rsv(root, block_rsv);
3216 }
3217 }
3218
3219 /*
3220 * This creates an orphan entry for the given inode in case something goes
3221 * wrong in the middle of an unlink/truncate.
3222 *
3223 * NOTE: caller of this function should reserve 5 units of metadata for
3224 * this function.
3225 */
3226 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3227 {
3228 struct btrfs_root *root = BTRFS_I(inode)->root;
3229 struct btrfs_block_rsv *block_rsv = NULL;
3230 int reserve = 0;
3231 int insert = 0;
3232 int ret;
3233
3234 if (!root->orphan_block_rsv) {
3235 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3236 if (!block_rsv)
3237 return -ENOMEM;
3238 }
3239
3240 spin_lock(&root->orphan_lock);
3241 if (!root->orphan_block_rsv) {
3242 root->orphan_block_rsv = block_rsv;
3243 } else if (block_rsv) {
3244 btrfs_free_block_rsv(root, block_rsv);
3245 block_rsv = NULL;
3246 }
3247
3248 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3249 &BTRFS_I(inode)->runtime_flags)) {
3250 #if 0
3251 /*
3252 * For proper ENOSPC handling, we should do orphan
3253 * cleanup when mounting. But this introduces backward
3254 * compatibility issue.
3255 */
3256 if (!xchg(&root->orphan_item_inserted, 1))
3257 insert = 2;
3258 else
3259 insert = 1;
3260 #endif
3261 insert = 1;
3262 atomic_inc(&root->orphan_inodes);
3263 }
3264
3265 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3266 &BTRFS_I(inode)->runtime_flags))
3267 reserve = 1;
3268 spin_unlock(&root->orphan_lock);
3269
3270 /* grab metadata reservation from transaction handle */
3271 if (reserve) {
3272 ret = btrfs_orphan_reserve_metadata(trans, inode);
3273 ASSERT(!ret);
3274 if (ret) {
3275 atomic_dec(&root->orphan_inodes);
3276 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3277 &BTRFS_I(inode)->runtime_flags);
3278 if (insert)
3279 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3280 &BTRFS_I(inode)->runtime_flags);
3281 return ret;
3282 }
3283 }
3284
3285 /* insert an orphan item to track this unlinked/truncated file */
3286 if (insert >= 1) {
3287 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3288 if (ret) {
3289 atomic_dec(&root->orphan_inodes);
3290 if (reserve) {
3291 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3292 &BTRFS_I(inode)->runtime_flags);
3293 btrfs_orphan_release_metadata(inode);
3294 }
3295 if (ret != -EEXIST) {
3296 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3297 &BTRFS_I(inode)->runtime_flags);
3298 btrfs_abort_transaction(trans, root, ret);
3299 return ret;
3300 }
3301 }
3302 ret = 0;
3303 }
3304
3305 /* insert an orphan item to track subvolume contains orphan files */
3306 if (insert >= 2) {
3307 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3308 root->root_key.objectid);
3309 if (ret && ret != -EEXIST) {
3310 btrfs_abort_transaction(trans, root, ret);
3311 return ret;
3312 }
3313 }
3314 return 0;
3315 }
3316
3317 /*
3318 * We have done the truncate/delete so we can go ahead and remove the orphan
3319 * item for this particular inode.
3320 */
3321 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3322 struct inode *inode)
3323 {
3324 struct btrfs_root *root = BTRFS_I(inode)->root;
3325 int delete_item = 0;
3326 int release_rsv = 0;
3327 int ret = 0;
3328
3329 spin_lock(&root->orphan_lock);
3330 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3331 &BTRFS_I(inode)->runtime_flags))
3332 delete_item = 1;
3333
3334 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3335 &BTRFS_I(inode)->runtime_flags))
3336 release_rsv = 1;
3337 spin_unlock(&root->orphan_lock);
3338
3339 if (delete_item) {
3340 atomic_dec(&root->orphan_inodes);
3341 if (trans)
3342 ret = btrfs_del_orphan_item(trans, root,
3343 btrfs_ino(inode));
3344 }
3345
3346 if (release_rsv)
3347 btrfs_orphan_release_metadata(inode);
3348
3349 return ret;
3350 }
3351
3352 /*
3353 * this cleans up any orphans that may be left on the list from the last use
3354 * of this root.
3355 */
3356 int btrfs_orphan_cleanup(struct btrfs_root *root)
3357 {
3358 struct btrfs_path *path;
3359 struct extent_buffer *leaf;
3360 struct btrfs_key key, found_key;
3361 struct btrfs_trans_handle *trans;
3362 struct inode *inode;
3363 u64 last_objectid = 0;
3364 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3365
3366 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3367 return 0;
3368
3369 path = btrfs_alloc_path();
3370 if (!path) {
3371 ret = -ENOMEM;
3372 goto out;
3373 }
3374 path->reada = READA_BACK;
3375
3376 key.objectid = BTRFS_ORPHAN_OBJECTID;
3377 key.type = BTRFS_ORPHAN_ITEM_KEY;
3378 key.offset = (u64)-1;
3379
3380 while (1) {
3381 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3382 if (ret < 0)
3383 goto out;
3384
3385 /*
3386 * if ret == 0 means we found what we were searching for, which
3387 * is weird, but possible, so only screw with path if we didn't
3388 * find the key and see if we have stuff that matches
3389 */
3390 if (ret > 0) {
3391 ret = 0;
3392 if (path->slots[0] == 0)
3393 break;
3394 path->slots[0]--;
3395 }
3396
3397 /* pull out the item */
3398 leaf = path->nodes[0];
3399 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3400
3401 /* make sure the item matches what we want */
3402 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3403 break;
3404 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3405 break;
3406
3407 /* release the path since we're done with it */
3408 btrfs_release_path(path);
3409
3410 /*
3411 * this is where we are basically btrfs_lookup, without the
3412 * crossing root thing. we store the inode number in the
3413 * offset of the orphan item.
3414 */
3415
3416 if (found_key.offset == last_objectid) {
3417 btrfs_err(root->fs_info,
3418 "Error removing orphan entry, stopping orphan cleanup");
3419 ret = -EINVAL;
3420 goto out;
3421 }
3422
3423 last_objectid = found_key.offset;
3424
3425 found_key.objectid = found_key.offset;
3426 found_key.type = BTRFS_INODE_ITEM_KEY;
3427 found_key.offset = 0;
3428 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3429 ret = PTR_ERR_OR_ZERO(inode);
3430 if (ret && ret != -ESTALE)
3431 goto out;
3432
3433 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3434 struct btrfs_root *dead_root;
3435 struct btrfs_fs_info *fs_info = root->fs_info;
3436 int is_dead_root = 0;
3437
3438 /*
3439 * this is an orphan in the tree root. Currently these
3440 * could come from 2 sources:
3441 * a) a snapshot deletion in progress
3442 * b) a free space cache inode
3443 * We need to distinguish those two, as the snapshot
3444 * orphan must not get deleted.
3445 * find_dead_roots already ran before us, so if this
3446 * is a snapshot deletion, we should find the root
3447 * in the dead_roots list
3448 */
3449 spin_lock(&fs_info->trans_lock);
3450 list_for_each_entry(dead_root, &fs_info->dead_roots,
3451 root_list) {
3452 if (dead_root->root_key.objectid ==
3453 found_key.objectid) {
3454 is_dead_root = 1;
3455 break;
3456 }
3457 }
3458 spin_unlock(&fs_info->trans_lock);
3459 if (is_dead_root) {
3460 /* prevent this orphan from being found again */
3461 key.offset = found_key.objectid - 1;
3462 continue;
3463 }
3464 }
3465 /*
3466 * Inode is already gone but the orphan item is still there,
3467 * kill the orphan item.
3468 */
3469 if (ret == -ESTALE) {
3470 trans = btrfs_start_transaction(root, 1);
3471 if (IS_ERR(trans)) {
3472 ret = PTR_ERR(trans);
3473 goto out;
3474 }
3475 btrfs_debug(root->fs_info, "auto deleting %Lu",
3476 found_key.objectid);
3477 ret = btrfs_del_orphan_item(trans, root,
3478 found_key.objectid);
3479 btrfs_end_transaction(trans, root);
3480 if (ret)
3481 goto out;
3482 continue;
3483 }
3484
3485 /*
3486 * add this inode to the orphan list so btrfs_orphan_del does
3487 * the proper thing when we hit it
3488 */
3489 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3490 &BTRFS_I(inode)->runtime_flags);
3491 atomic_inc(&root->orphan_inodes);
3492
3493 /* if we have links, this was a truncate, lets do that */
3494 if (inode->i_nlink) {
3495 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3496 iput(inode);
3497 continue;
3498 }
3499 nr_truncate++;
3500
3501 /* 1 for the orphan item deletion. */
3502 trans = btrfs_start_transaction(root, 1);
3503 if (IS_ERR(trans)) {
3504 iput(inode);
3505 ret = PTR_ERR(trans);
3506 goto out;
3507 }
3508 ret = btrfs_orphan_add(trans, inode);
3509 btrfs_end_transaction(trans, root);
3510 if (ret) {
3511 iput(inode);
3512 goto out;
3513 }
3514
3515 ret = btrfs_truncate(inode);
3516 if (ret)
3517 btrfs_orphan_del(NULL, inode);
3518 } else {
3519 nr_unlink++;
3520 }
3521
3522 /* this will do delete_inode and everything for us */
3523 iput(inode);
3524 if (ret)
3525 goto out;
3526 }
3527 /* release the path since we're done with it */
3528 btrfs_release_path(path);
3529
3530 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3531
3532 if (root->orphan_block_rsv)
3533 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3534 (u64)-1);
3535
3536 if (root->orphan_block_rsv ||
3537 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3538 trans = btrfs_join_transaction(root);
3539 if (!IS_ERR(trans))
3540 btrfs_end_transaction(trans, root);
3541 }
3542
3543 if (nr_unlink)
3544 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3545 if (nr_truncate)
3546 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3547
3548 out:
3549 if (ret)
3550 btrfs_err(root->fs_info,
3551 "could not do orphan cleanup %d", ret);
3552 btrfs_free_path(path);
3553 return ret;
3554 }
3555
3556 /*
3557 * very simple check to peek ahead in the leaf looking for xattrs. If we
3558 * don't find any xattrs, we know there can't be any acls.
3559 *
3560 * slot is the slot the inode is in, objectid is the objectid of the inode
3561 */
3562 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3563 int slot, u64 objectid,
3564 int *first_xattr_slot)
3565 {
3566 u32 nritems = btrfs_header_nritems(leaf);
3567 struct btrfs_key found_key;
3568 static u64 xattr_access = 0;
3569 static u64 xattr_default = 0;
3570 int scanned = 0;
3571
3572 if (!xattr_access) {
3573 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3574 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3575 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3576 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3577 }
3578
3579 slot++;
3580 *first_xattr_slot = -1;
3581 while (slot < nritems) {
3582 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3583
3584 /* we found a different objectid, there must not be acls */
3585 if (found_key.objectid != objectid)
3586 return 0;
3587
3588 /* we found an xattr, assume we've got an acl */
3589 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3590 if (*first_xattr_slot == -1)
3591 *first_xattr_slot = slot;
3592 if (found_key.offset == xattr_access ||
3593 found_key.offset == xattr_default)
3594 return 1;
3595 }
3596
3597 /*
3598 * we found a key greater than an xattr key, there can't
3599 * be any acls later on
3600 */
3601 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3602 return 0;
3603
3604 slot++;
3605 scanned++;
3606
3607 /*
3608 * it goes inode, inode backrefs, xattrs, extents,
3609 * so if there are a ton of hard links to an inode there can
3610 * be a lot of backrefs. Don't waste time searching too hard,
3611 * this is just an optimization
3612 */
3613 if (scanned >= 8)
3614 break;
3615 }
3616 /* we hit the end of the leaf before we found an xattr or
3617 * something larger than an xattr. We have to assume the inode
3618 * has acls
3619 */
3620 if (*first_xattr_slot == -1)
3621 *first_xattr_slot = slot;
3622 return 1;
3623 }
3624
3625 /*
3626 * read an inode from the btree into the in-memory inode
3627 */
3628 static void btrfs_read_locked_inode(struct inode *inode)
3629 {
3630 struct btrfs_path *path;
3631 struct extent_buffer *leaf;
3632 struct btrfs_inode_item *inode_item;
3633 struct btrfs_root *root = BTRFS_I(inode)->root;
3634 struct btrfs_key location;
3635 unsigned long ptr;
3636 int maybe_acls;
3637 u32 rdev;
3638 int ret;
3639 bool filled = false;
3640 int first_xattr_slot;
3641
3642 ret = btrfs_fill_inode(inode, &rdev);
3643 if (!ret)
3644 filled = true;
3645
3646 path = btrfs_alloc_path();
3647 if (!path)
3648 goto make_bad;
3649
3650 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3651
3652 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3653 if (ret)
3654 goto make_bad;
3655
3656 leaf = path->nodes[0];
3657
3658 if (filled)
3659 goto cache_index;
3660
3661 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3662 struct btrfs_inode_item);
3663 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3664 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3665 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3666 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3667 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3668
3669 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3670 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3671
3672 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3673 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3674
3675 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3676 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3677
3678 BTRFS_I(inode)->i_otime.tv_sec =
3679 btrfs_timespec_sec(leaf, &inode_item->otime);
3680 BTRFS_I(inode)->i_otime.tv_nsec =
3681 btrfs_timespec_nsec(leaf, &inode_item->otime);
3682
3683 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3684 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3685 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3686
3687 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3688 inode->i_generation = BTRFS_I(inode)->generation;
3689 inode->i_rdev = 0;
3690 rdev = btrfs_inode_rdev(leaf, inode_item);
3691
3692 BTRFS_I(inode)->index_cnt = (u64)-1;
3693 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3694
3695 cache_index:
3696 /*
3697 * If we were modified in the current generation and evicted from memory
3698 * and then re-read we need to do a full sync since we don't have any
3699 * idea about which extents were modified before we were evicted from
3700 * cache.
3701 *
3702 * This is required for both inode re-read from disk and delayed inode
3703 * in delayed_nodes_tree.
3704 */
3705 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3706 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3707 &BTRFS_I(inode)->runtime_flags);
3708
3709 /*
3710 * We don't persist the id of the transaction where an unlink operation
3711 * against the inode was last made. So here we assume the inode might
3712 * have been evicted, and therefore the exact value of last_unlink_trans
3713 * lost, and set it to last_trans to avoid metadata inconsistencies
3714 * between the inode and its parent if the inode is fsync'ed and the log
3715 * replayed. For example, in the scenario:
3716 *
3717 * touch mydir/foo
3718 * ln mydir/foo mydir/bar
3719 * sync
3720 * unlink mydir/bar
3721 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3722 * xfs_io -c fsync mydir/foo
3723 * <power failure>
3724 * mount fs, triggers fsync log replay
3725 *
3726 * We must make sure that when we fsync our inode foo we also log its
3727 * parent inode, otherwise after log replay the parent still has the
3728 * dentry with the "bar" name but our inode foo has a link count of 1
3729 * and doesn't have an inode ref with the name "bar" anymore.
3730 *
3731 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3732 * but it guarantees correctness at the expense of occasional full
3733 * transaction commits on fsync if our inode is a directory, or if our
3734 * inode is not a directory, logging its parent unnecessarily.
3735 */
3736 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3737
3738 path->slots[0]++;
3739 if (inode->i_nlink != 1 ||
3740 path->slots[0] >= btrfs_header_nritems(leaf))
3741 goto cache_acl;
3742
3743 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3744 if (location.objectid != btrfs_ino(inode))
3745 goto cache_acl;
3746
3747 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3748 if (location.type == BTRFS_INODE_REF_KEY) {
3749 struct btrfs_inode_ref *ref;
3750
3751 ref = (struct btrfs_inode_ref *)ptr;
3752 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3753 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3754 struct btrfs_inode_extref *extref;
3755
3756 extref = (struct btrfs_inode_extref *)ptr;
3757 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3758 extref);
3759 }
3760 cache_acl:
3761 /*
3762 * try to precache a NULL acl entry for files that don't have
3763 * any xattrs or acls
3764 */
3765 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3766 btrfs_ino(inode), &first_xattr_slot);
3767 if (first_xattr_slot != -1) {
3768 path->slots[0] = first_xattr_slot;
3769 ret = btrfs_load_inode_props(inode, path);
3770 if (ret)
3771 btrfs_err(root->fs_info,
3772 "error loading props for ino %llu (root %llu): %d",
3773 btrfs_ino(inode),
3774 root->root_key.objectid, ret);
3775 }
3776 btrfs_free_path(path);
3777
3778 if (!maybe_acls)
3779 cache_no_acl(inode);
3780
3781 switch (inode->i_mode & S_IFMT) {
3782 case S_IFREG:
3783 inode->i_mapping->a_ops = &btrfs_aops;
3784 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3785 inode->i_fop = &btrfs_file_operations;
3786 inode->i_op = &btrfs_file_inode_operations;
3787 break;
3788 case S_IFDIR:
3789 inode->i_fop = &btrfs_dir_file_operations;
3790 if (root == root->fs_info->tree_root)
3791 inode->i_op = &btrfs_dir_ro_inode_operations;
3792 else
3793 inode->i_op = &btrfs_dir_inode_operations;
3794 break;
3795 case S_IFLNK:
3796 inode->i_op = &btrfs_symlink_inode_operations;
3797 inode_nohighmem(inode);
3798 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3799 break;
3800 default:
3801 inode->i_op = &btrfs_special_inode_operations;
3802 init_special_inode(inode, inode->i_mode, rdev);
3803 break;
3804 }
3805
3806 btrfs_update_iflags(inode);
3807 return;
3808
3809 make_bad:
3810 btrfs_free_path(path);
3811 make_bad_inode(inode);
3812 }
3813
3814 /*
3815 * given a leaf and an inode, copy the inode fields into the leaf
3816 */
3817 static void fill_inode_item(struct btrfs_trans_handle *trans,
3818 struct extent_buffer *leaf,
3819 struct btrfs_inode_item *item,
3820 struct inode *inode)
3821 {
3822 struct btrfs_map_token token;
3823
3824 btrfs_init_map_token(&token);
3825
3826 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3827 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3828 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3829 &token);
3830 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3831 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3832
3833 btrfs_set_token_timespec_sec(leaf, &item->atime,
3834 inode->i_atime.tv_sec, &token);
3835 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3836 inode->i_atime.tv_nsec, &token);
3837
3838 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3839 inode->i_mtime.tv_sec, &token);
3840 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3841 inode->i_mtime.tv_nsec, &token);
3842
3843 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3844 inode->i_ctime.tv_sec, &token);
3845 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3846 inode->i_ctime.tv_nsec, &token);
3847
3848 btrfs_set_token_timespec_sec(leaf, &item->otime,
3849 BTRFS_I(inode)->i_otime.tv_sec, &token);
3850 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3851 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3852
3853 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3854 &token);
3855 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3856 &token);
3857 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3858 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3859 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3860 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3861 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3862 }
3863
3864 /*
3865 * copy everything in the in-memory inode into the btree.
3866 */
3867 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3868 struct btrfs_root *root, struct inode *inode)
3869 {
3870 struct btrfs_inode_item *inode_item;
3871 struct btrfs_path *path;
3872 struct extent_buffer *leaf;
3873 int ret;
3874
3875 path = btrfs_alloc_path();
3876 if (!path)
3877 return -ENOMEM;
3878
3879 path->leave_spinning = 1;
3880 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3881 1);
3882 if (ret) {
3883 if (ret > 0)
3884 ret = -ENOENT;
3885 goto failed;
3886 }
3887
3888 leaf = path->nodes[0];
3889 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3890 struct btrfs_inode_item);
3891
3892 fill_inode_item(trans, leaf, inode_item, inode);
3893 btrfs_mark_buffer_dirty(leaf);
3894 btrfs_set_inode_last_trans(trans, inode);
3895 ret = 0;
3896 failed:
3897 btrfs_free_path(path);
3898 return ret;
3899 }
3900
3901 /*
3902 * copy everything in the in-memory inode into the btree.
3903 */
3904 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3905 struct btrfs_root *root, struct inode *inode)
3906 {
3907 int ret;
3908
3909 /*
3910 * If the inode is a free space inode, we can deadlock during commit
3911 * if we put it into the delayed code.
3912 *
3913 * The data relocation inode should also be directly updated
3914 * without delay
3915 */
3916 if (!btrfs_is_free_space_inode(inode)
3917 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3918 && !root->fs_info->log_root_recovering) {
3919 btrfs_update_root_times(trans, root);
3920
3921 ret = btrfs_delayed_update_inode(trans, root, inode);
3922 if (!ret)
3923 btrfs_set_inode_last_trans(trans, inode);
3924 return ret;
3925 }
3926
3927 return btrfs_update_inode_item(trans, root, inode);
3928 }
3929
3930 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3931 struct btrfs_root *root,
3932 struct inode *inode)
3933 {
3934 int ret;
3935
3936 ret = btrfs_update_inode(trans, root, inode);
3937 if (ret == -ENOSPC)
3938 return btrfs_update_inode_item(trans, root, inode);
3939 return ret;
3940 }
3941
3942 /*
3943 * unlink helper that gets used here in inode.c and in the tree logging
3944 * recovery code. It remove a link in a directory with a given name, and
3945 * also drops the back refs in the inode to the directory
3946 */
3947 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3948 struct btrfs_root *root,
3949 struct inode *dir, struct inode *inode,
3950 const char *name, int name_len)
3951 {
3952 struct btrfs_path *path;
3953 int ret = 0;
3954 struct extent_buffer *leaf;
3955 struct btrfs_dir_item *di;
3956 struct btrfs_key key;
3957 u64 index;
3958 u64 ino = btrfs_ino(inode);
3959 u64 dir_ino = btrfs_ino(dir);
3960
3961 path = btrfs_alloc_path();
3962 if (!path) {
3963 ret = -ENOMEM;
3964 goto out;
3965 }
3966
3967 path->leave_spinning = 1;
3968 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3969 name, name_len, -1);
3970 if (IS_ERR(di)) {
3971 ret = PTR_ERR(di);
3972 goto err;
3973 }
3974 if (!di) {
3975 ret = -ENOENT;
3976 goto err;
3977 }
3978 leaf = path->nodes[0];
3979 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3980 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3981 if (ret)
3982 goto err;
3983 btrfs_release_path(path);
3984
3985 /*
3986 * If we don't have dir index, we have to get it by looking up
3987 * the inode ref, since we get the inode ref, remove it directly,
3988 * it is unnecessary to do delayed deletion.
3989 *
3990 * But if we have dir index, needn't search inode ref to get it.
3991 * Since the inode ref is close to the inode item, it is better
3992 * that we delay to delete it, and just do this deletion when
3993 * we update the inode item.
3994 */
3995 if (BTRFS_I(inode)->dir_index) {
3996 ret = btrfs_delayed_delete_inode_ref(inode);
3997 if (!ret) {
3998 index = BTRFS_I(inode)->dir_index;
3999 goto skip_backref;
4000 }
4001 }
4002
4003 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4004 dir_ino, &index);
4005 if (ret) {
4006 btrfs_info(root->fs_info,
4007 "failed to delete reference to %.*s, inode %llu parent %llu",
4008 name_len, name, ino, dir_ino);
4009 btrfs_abort_transaction(trans, root, ret);
4010 goto err;
4011 }
4012 skip_backref:
4013 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4014 if (ret) {
4015 btrfs_abort_transaction(trans, root, ret);
4016 goto err;
4017 }
4018
4019 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
4020 inode, dir_ino);
4021 if (ret != 0 && ret != -ENOENT) {
4022 btrfs_abort_transaction(trans, root, ret);
4023 goto err;
4024 }
4025
4026 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4027 dir, index);
4028 if (ret == -ENOENT)
4029 ret = 0;
4030 else if (ret)
4031 btrfs_abort_transaction(trans, root, ret);
4032 err:
4033 btrfs_free_path(path);
4034 if (ret)
4035 goto out;
4036
4037 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4038 inode_inc_iversion(inode);
4039 inode_inc_iversion(dir);
4040 inode->i_ctime = dir->i_mtime =
4041 dir->i_ctime = current_fs_time(inode->i_sb);
4042 ret = btrfs_update_inode(trans, root, dir);
4043 out:
4044 return ret;
4045 }
4046
4047 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4048 struct btrfs_root *root,
4049 struct inode *dir, struct inode *inode,
4050 const char *name, int name_len)
4051 {
4052 int ret;
4053 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4054 if (!ret) {
4055 drop_nlink(inode);
4056 ret = btrfs_update_inode(trans, root, inode);
4057 }
4058 return ret;
4059 }
4060
4061 /*
4062 * helper to start transaction for unlink and rmdir.
4063 *
4064 * unlink and rmdir are special in btrfs, they do not always free space, so
4065 * if we cannot make our reservations the normal way try and see if there is
4066 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4067 * allow the unlink to occur.
4068 */
4069 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4070 {
4071 struct btrfs_root *root = BTRFS_I(dir)->root;
4072
4073 /*
4074 * 1 for the possible orphan item
4075 * 1 for the dir item
4076 * 1 for the dir index
4077 * 1 for the inode ref
4078 * 1 for the inode
4079 */
4080 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4081 }
4082
4083 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4084 {
4085 struct btrfs_root *root = BTRFS_I(dir)->root;
4086 struct btrfs_trans_handle *trans;
4087 struct inode *inode = d_inode(dentry);
4088 int ret;
4089
4090 trans = __unlink_start_trans(dir);
4091 if (IS_ERR(trans))
4092 return PTR_ERR(trans);
4093
4094 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4095
4096 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4097 dentry->d_name.name, dentry->d_name.len);
4098 if (ret)
4099 goto out;
4100
4101 if (inode->i_nlink == 0) {
4102 ret = btrfs_orphan_add(trans, inode);
4103 if (ret)
4104 goto out;
4105 }
4106
4107 out:
4108 btrfs_end_transaction(trans, root);
4109 btrfs_btree_balance_dirty(root);
4110 return ret;
4111 }
4112
4113 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4114 struct btrfs_root *root,
4115 struct inode *dir, u64 objectid,
4116 const char *name, int name_len)
4117 {
4118 struct btrfs_path *path;
4119 struct extent_buffer *leaf;
4120 struct btrfs_dir_item *di;
4121 struct btrfs_key key;
4122 u64 index;
4123 int ret;
4124 u64 dir_ino = btrfs_ino(dir);
4125
4126 path = btrfs_alloc_path();
4127 if (!path)
4128 return -ENOMEM;
4129
4130 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4131 name, name_len, -1);
4132 if (IS_ERR_OR_NULL(di)) {
4133 if (!di)
4134 ret = -ENOENT;
4135 else
4136 ret = PTR_ERR(di);
4137 goto out;
4138 }
4139
4140 leaf = path->nodes[0];
4141 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4142 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4143 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4144 if (ret) {
4145 btrfs_abort_transaction(trans, root, ret);
4146 goto out;
4147 }
4148 btrfs_release_path(path);
4149
4150 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4151 objectid, root->root_key.objectid,
4152 dir_ino, &index, name, name_len);
4153 if (ret < 0) {
4154 if (ret != -ENOENT) {
4155 btrfs_abort_transaction(trans, root, ret);
4156 goto out;
4157 }
4158 di = btrfs_search_dir_index_item(root, path, dir_ino,
4159 name, name_len);
4160 if (IS_ERR_OR_NULL(di)) {
4161 if (!di)
4162 ret = -ENOENT;
4163 else
4164 ret = PTR_ERR(di);
4165 btrfs_abort_transaction(trans, root, ret);
4166 goto out;
4167 }
4168
4169 leaf = path->nodes[0];
4170 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4171 btrfs_release_path(path);
4172 index = key.offset;
4173 }
4174 btrfs_release_path(path);
4175
4176 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4177 if (ret) {
4178 btrfs_abort_transaction(trans, root, ret);
4179 goto out;
4180 }
4181
4182 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4183 inode_inc_iversion(dir);
4184 dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb);
4185 ret = btrfs_update_inode_fallback(trans, root, dir);
4186 if (ret)
4187 btrfs_abort_transaction(trans, root, ret);
4188 out:
4189 btrfs_free_path(path);
4190 return ret;
4191 }
4192
4193 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4194 {
4195 struct inode *inode = d_inode(dentry);
4196 int err = 0;
4197 struct btrfs_root *root = BTRFS_I(dir)->root;
4198 struct btrfs_trans_handle *trans;
4199
4200 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4201 return -ENOTEMPTY;
4202 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4203 return -EPERM;
4204
4205 trans = __unlink_start_trans(dir);
4206 if (IS_ERR(trans))
4207 return PTR_ERR(trans);
4208
4209 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4210 err = btrfs_unlink_subvol(trans, root, dir,
4211 BTRFS_I(inode)->location.objectid,
4212 dentry->d_name.name,
4213 dentry->d_name.len);
4214 goto out;
4215 }
4216
4217 err = btrfs_orphan_add(trans, inode);
4218 if (err)
4219 goto out;
4220
4221 /* now the directory is empty */
4222 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4223 dentry->d_name.name, dentry->d_name.len);
4224 if (!err)
4225 btrfs_i_size_write(inode, 0);
4226 out:
4227 btrfs_end_transaction(trans, root);
4228 btrfs_btree_balance_dirty(root);
4229
4230 return err;
4231 }
4232
4233 static int truncate_space_check(struct btrfs_trans_handle *trans,
4234 struct btrfs_root *root,
4235 u64 bytes_deleted)
4236 {
4237 int ret;
4238
4239 /*
4240 * This is only used to apply pressure to the enospc system, we don't
4241 * intend to use this reservation at all.
4242 */
4243 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4244 bytes_deleted *= root->nodesize;
4245 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4246 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4247 if (!ret) {
4248 trace_btrfs_space_reservation(root->fs_info, "transaction",
4249 trans->transid,
4250 bytes_deleted, 1);
4251 trans->bytes_reserved += bytes_deleted;
4252 }
4253 return ret;
4254
4255 }
4256
4257 static int truncate_inline_extent(struct inode *inode,
4258 struct btrfs_path *path,
4259 struct btrfs_key *found_key,
4260 const u64 item_end,
4261 const u64 new_size)
4262 {
4263 struct extent_buffer *leaf = path->nodes[0];
4264 int slot = path->slots[0];
4265 struct btrfs_file_extent_item *fi;
4266 u32 size = (u32)(new_size - found_key->offset);
4267 struct btrfs_root *root = BTRFS_I(inode)->root;
4268
4269 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4270
4271 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4272 loff_t offset = new_size;
4273 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4274
4275 /*
4276 * Zero out the remaining of the last page of our inline extent,
4277 * instead of directly truncating our inline extent here - that
4278 * would be much more complex (decompressing all the data, then
4279 * compressing the truncated data, which might be bigger than
4280 * the size of the inline extent, resize the extent, etc).
4281 * We release the path because to get the page we might need to
4282 * read the extent item from disk (data not in the page cache).
4283 */
4284 btrfs_release_path(path);
4285 return btrfs_truncate_block(inode, offset, page_end - offset,
4286 0);
4287 }
4288
4289 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4290 size = btrfs_file_extent_calc_inline_size(size);
4291 btrfs_truncate_item(root, path, size, 1);
4292
4293 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4294 inode_sub_bytes(inode, item_end + 1 - new_size);
4295
4296 return 0;
4297 }
4298
4299 /*
4300 * this can truncate away extent items, csum items and directory items.
4301 * It starts at a high offset and removes keys until it can't find
4302 * any higher than new_size
4303 *
4304 * csum items that cross the new i_size are truncated to the new size
4305 * as well.
4306 *
4307 * min_type is the minimum key type to truncate down to. If set to 0, this
4308 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4309 */
4310 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4311 struct btrfs_root *root,
4312 struct inode *inode,
4313 u64 new_size, u32 min_type)
4314 {
4315 struct btrfs_path *path;
4316 struct extent_buffer *leaf;
4317 struct btrfs_file_extent_item *fi;
4318 struct btrfs_key key;
4319 struct btrfs_key found_key;
4320 u64 extent_start = 0;
4321 u64 extent_num_bytes = 0;
4322 u64 extent_offset = 0;
4323 u64 item_end = 0;
4324 u64 last_size = new_size;
4325 u32 found_type = (u8)-1;
4326 int found_extent;
4327 int del_item;
4328 int pending_del_nr = 0;
4329 int pending_del_slot = 0;
4330 int extent_type = -1;
4331 int ret;
4332 int err = 0;
4333 u64 ino = btrfs_ino(inode);
4334 u64 bytes_deleted = 0;
4335 bool be_nice = 0;
4336 bool should_throttle = 0;
4337 bool should_end = 0;
4338
4339 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4340
4341 /*
4342 * for non-free space inodes and ref cows, we want to back off from
4343 * time to time
4344 */
4345 if (!btrfs_is_free_space_inode(inode) &&
4346 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4347 be_nice = 1;
4348
4349 path = btrfs_alloc_path();
4350 if (!path)
4351 return -ENOMEM;
4352 path->reada = READA_BACK;
4353
4354 /*
4355 * We want to drop from the next block forward in case this new size is
4356 * not block aligned since we will be keeping the last block of the
4357 * extent just the way it is.
4358 */
4359 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4360 root == root->fs_info->tree_root)
4361 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4362 root->sectorsize), (u64)-1, 0);
4363
4364 /*
4365 * This function is also used to drop the items in the log tree before
4366 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4367 * it is used to drop the loged items. So we shouldn't kill the delayed
4368 * items.
4369 */
4370 if (min_type == 0 && root == BTRFS_I(inode)->root)
4371 btrfs_kill_delayed_inode_items(inode);
4372
4373 key.objectid = ino;
4374 key.offset = (u64)-1;
4375 key.type = (u8)-1;
4376
4377 search_again:
4378 /*
4379 * with a 16K leaf size and 128MB extents, you can actually queue
4380 * up a huge file in a single leaf. Most of the time that
4381 * bytes_deleted is > 0, it will be huge by the time we get here
4382 */
4383 if (be_nice && bytes_deleted > SZ_32M) {
4384 if (btrfs_should_end_transaction(trans, root)) {
4385 err = -EAGAIN;
4386 goto error;
4387 }
4388 }
4389
4390
4391 path->leave_spinning = 1;
4392 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4393 if (ret < 0) {
4394 err = ret;
4395 goto out;
4396 }
4397
4398 if (ret > 0) {
4399 /* there are no items in the tree for us to truncate, we're
4400 * done
4401 */
4402 if (path->slots[0] == 0)
4403 goto out;
4404 path->slots[0]--;
4405 }
4406
4407 while (1) {
4408 fi = NULL;
4409 leaf = path->nodes[0];
4410 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4411 found_type = found_key.type;
4412
4413 if (found_key.objectid != ino)
4414 break;
4415
4416 if (found_type < min_type)
4417 break;
4418
4419 item_end = found_key.offset;
4420 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4421 fi = btrfs_item_ptr(leaf, path->slots[0],
4422 struct btrfs_file_extent_item);
4423 extent_type = btrfs_file_extent_type(leaf, fi);
4424 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4425 item_end +=
4426 btrfs_file_extent_num_bytes(leaf, fi);
4427 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4428 item_end += btrfs_file_extent_inline_len(leaf,
4429 path->slots[0], fi);
4430 }
4431 item_end--;
4432 }
4433 if (found_type > min_type) {
4434 del_item = 1;
4435 } else {
4436 if (item_end < new_size)
4437 break;
4438 if (found_key.offset >= new_size)
4439 del_item = 1;
4440 else
4441 del_item = 0;
4442 }
4443 found_extent = 0;
4444 /* FIXME, shrink the extent if the ref count is only 1 */
4445 if (found_type != BTRFS_EXTENT_DATA_KEY)
4446 goto delete;
4447
4448 if (del_item)
4449 last_size = found_key.offset;
4450 else
4451 last_size = new_size;
4452
4453 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4454 u64 num_dec;
4455 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4456 if (!del_item) {
4457 u64 orig_num_bytes =
4458 btrfs_file_extent_num_bytes(leaf, fi);
4459 extent_num_bytes = ALIGN(new_size -
4460 found_key.offset,
4461 root->sectorsize);
4462 btrfs_set_file_extent_num_bytes(leaf, fi,
4463 extent_num_bytes);
4464 num_dec = (orig_num_bytes -
4465 extent_num_bytes);
4466 if (test_bit(BTRFS_ROOT_REF_COWS,
4467 &root->state) &&
4468 extent_start != 0)
4469 inode_sub_bytes(inode, num_dec);
4470 btrfs_mark_buffer_dirty(leaf);
4471 } else {
4472 extent_num_bytes =
4473 btrfs_file_extent_disk_num_bytes(leaf,
4474 fi);
4475 extent_offset = found_key.offset -
4476 btrfs_file_extent_offset(leaf, fi);
4477
4478 /* FIXME blocksize != 4096 */
4479 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4480 if (extent_start != 0) {
4481 found_extent = 1;
4482 if (test_bit(BTRFS_ROOT_REF_COWS,
4483 &root->state))
4484 inode_sub_bytes(inode, num_dec);
4485 }
4486 }
4487 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4488 /*
4489 * we can't truncate inline items that have had
4490 * special encodings
4491 */
4492 if (!del_item &&
4493 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4494 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4495
4496 /*
4497 * Need to release path in order to truncate a
4498 * compressed extent. So delete any accumulated
4499 * extent items so far.
4500 */
4501 if (btrfs_file_extent_compression(leaf, fi) !=
4502 BTRFS_COMPRESS_NONE && pending_del_nr) {
4503 err = btrfs_del_items(trans, root, path,
4504 pending_del_slot,
4505 pending_del_nr);
4506 if (err) {
4507 btrfs_abort_transaction(trans,
4508 root,
4509 err);
4510 goto error;
4511 }
4512 pending_del_nr = 0;
4513 }
4514
4515 err = truncate_inline_extent(inode, path,
4516 &found_key,
4517 item_end,
4518 new_size);
4519 if (err) {
4520 btrfs_abort_transaction(trans,
4521 root, err);
4522 goto error;
4523 }
4524 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4525 &root->state)) {
4526 inode_sub_bytes(inode, item_end + 1 - new_size);
4527 }
4528 }
4529 delete:
4530 if (del_item) {
4531 if (!pending_del_nr) {
4532 /* no pending yet, add ourselves */
4533 pending_del_slot = path->slots[0];
4534 pending_del_nr = 1;
4535 } else if (pending_del_nr &&
4536 path->slots[0] + 1 == pending_del_slot) {
4537 /* hop on the pending chunk */
4538 pending_del_nr++;
4539 pending_del_slot = path->slots[0];
4540 } else {
4541 BUG();
4542 }
4543 } else {
4544 break;
4545 }
4546 should_throttle = 0;
4547
4548 if (found_extent &&
4549 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4550 root == root->fs_info->tree_root)) {
4551 btrfs_set_path_blocking(path);
4552 bytes_deleted += extent_num_bytes;
4553 ret = btrfs_free_extent(trans, root, extent_start,
4554 extent_num_bytes, 0,
4555 btrfs_header_owner(leaf),
4556 ino, extent_offset);
4557 BUG_ON(ret);
4558 if (btrfs_should_throttle_delayed_refs(trans, root))
4559 btrfs_async_run_delayed_refs(root,
4560 trans->transid,
4561 trans->delayed_ref_updates * 2, 0);
4562 if (be_nice) {
4563 if (truncate_space_check(trans, root,
4564 extent_num_bytes)) {
4565 should_end = 1;
4566 }
4567 if (btrfs_should_throttle_delayed_refs(trans,
4568 root)) {
4569 should_throttle = 1;
4570 }
4571 }
4572 }
4573
4574 if (found_type == BTRFS_INODE_ITEM_KEY)
4575 break;
4576
4577 if (path->slots[0] == 0 ||
4578 path->slots[0] != pending_del_slot ||
4579 should_throttle || should_end) {
4580 if (pending_del_nr) {
4581 ret = btrfs_del_items(trans, root, path,
4582 pending_del_slot,
4583 pending_del_nr);
4584 if (ret) {
4585 btrfs_abort_transaction(trans,
4586 root, ret);
4587 goto error;
4588 }
4589 pending_del_nr = 0;
4590 }
4591 btrfs_release_path(path);
4592 if (should_throttle) {
4593 unsigned long updates = trans->delayed_ref_updates;
4594 if (updates) {
4595 trans->delayed_ref_updates = 0;
4596 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4597 if (ret && !err)
4598 err = ret;
4599 }
4600 }
4601 /*
4602 * if we failed to refill our space rsv, bail out
4603 * and let the transaction restart
4604 */
4605 if (should_end) {
4606 err = -EAGAIN;
4607 goto error;
4608 }
4609 goto search_again;
4610 } else {
4611 path->slots[0]--;
4612 }
4613 }
4614 out:
4615 if (pending_del_nr) {
4616 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4617 pending_del_nr);
4618 if (ret)
4619 btrfs_abort_transaction(trans, root, ret);
4620 }
4621 error:
4622 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4623 btrfs_ordered_update_i_size(inode, last_size, NULL);
4624
4625 btrfs_free_path(path);
4626
4627 if (be_nice && bytes_deleted > SZ_32M) {
4628 unsigned long updates = trans->delayed_ref_updates;
4629 if (updates) {
4630 trans->delayed_ref_updates = 0;
4631 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4632 if (ret && !err)
4633 err = ret;
4634 }
4635 }
4636 return err;
4637 }
4638
4639 /*
4640 * btrfs_truncate_block - read, zero a chunk and write a block
4641 * @inode - inode that we're zeroing
4642 * @from - the offset to start zeroing
4643 * @len - the length to zero, 0 to zero the entire range respective to the
4644 * offset
4645 * @front - zero up to the offset instead of from the offset on
4646 *
4647 * This will find the block for the "from" offset and cow the block and zero the
4648 * part we want to zero. This is used with truncate and hole punching.
4649 */
4650 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4651 int front)
4652 {
4653 struct address_space *mapping = inode->i_mapping;
4654 struct btrfs_root *root = BTRFS_I(inode)->root;
4655 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4656 struct btrfs_ordered_extent *ordered;
4657 struct extent_state *cached_state = NULL;
4658 char *kaddr;
4659 u32 blocksize = root->sectorsize;
4660 pgoff_t index = from >> PAGE_SHIFT;
4661 unsigned offset = from & (blocksize - 1);
4662 struct page *page;
4663 gfp_t mask = btrfs_alloc_write_mask(mapping);
4664 int ret = 0;
4665 u64 block_start;
4666 u64 block_end;
4667
4668 if ((offset & (blocksize - 1)) == 0 &&
4669 (!len || ((len & (blocksize - 1)) == 0)))
4670 goto out;
4671
4672 ret = btrfs_delalloc_reserve_space(inode,
4673 round_down(from, blocksize), blocksize);
4674 if (ret)
4675 goto out;
4676
4677 again:
4678 page = find_or_create_page(mapping, index, mask);
4679 if (!page) {
4680 btrfs_delalloc_release_space(inode,
4681 round_down(from, blocksize),
4682 blocksize);
4683 ret = -ENOMEM;
4684 goto out;
4685 }
4686
4687 block_start = round_down(from, blocksize);
4688 block_end = block_start + blocksize - 1;
4689
4690 if (!PageUptodate(page)) {
4691 ret = btrfs_readpage(NULL, page);
4692 lock_page(page);
4693 if (page->mapping != mapping) {
4694 unlock_page(page);
4695 put_page(page);
4696 goto again;
4697 }
4698 if (!PageUptodate(page)) {
4699 ret = -EIO;
4700 goto out_unlock;
4701 }
4702 }
4703 wait_on_page_writeback(page);
4704
4705 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4706 set_page_extent_mapped(page);
4707
4708 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4709 if (ordered) {
4710 unlock_extent_cached(io_tree, block_start, block_end,
4711 &cached_state, GFP_NOFS);
4712 unlock_page(page);
4713 put_page(page);
4714 btrfs_start_ordered_extent(inode, ordered, 1);
4715 btrfs_put_ordered_extent(ordered);
4716 goto again;
4717 }
4718
4719 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4720 EXTENT_DIRTY | EXTENT_DELALLOC |
4721 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4722 0, 0, &cached_state, GFP_NOFS);
4723
4724 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4725 &cached_state);
4726 if (ret) {
4727 unlock_extent_cached(io_tree, block_start, block_end,
4728 &cached_state, GFP_NOFS);
4729 goto out_unlock;
4730 }
4731
4732 if (offset != blocksize) {
4733 if (!len)
4734 len = blocksize - offset;
4735 kaddr = kmap(page);
4736 if (front)
4737 memset(kaddr + (block_start - page_offset(page)),
4738 0, offset);
4739 else
4740 memset(kaddr + (block_start - page_offset(page)) + offset,
4741 0, len);
4742 flush_dcache_page(page);
4743 kunmap(page);
4744 }
4745 ClearPageChecked(page);
4746 set_page_dirty(page);
4747 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4748 GFP_NOFS);
4749
4750 out_unlock:
4751 if (ret)
4752 btrfs_delalloc_release_space(inode, block_start,
4753 blocksize);
4754 unlock_page(page);
4755 put_page(page);
4756 out:
4757 return ret;
4758 }
4759
4760 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4761 u64 offset, u64 len)
4762 {
4763 struct btrfs_trans_handle *trans;
4764 int ret;
4765
4766 /*
4767 * Still need to make sure the inode looks like it's been updated so
4768 * that any holes get logged if we fsync.
4769 */
4770 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4771 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4772 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4773 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4774 return 0;
4775 }
4776
4777 /*
4778 * 1 - for the one we're dropping
4779 * 1 - for the one we're adding
4780 * 1 - for updating the inode.
4781 */
4782 trans = btrfs_start_transaction(root, 3);
4783 if (IS_ERR(trans))
4784 return PTR_ERR(trans);
4785
4786 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4787 if (ret) {
4788 btrfs_abort_transaction(trans, root, ret);
4789 btrfs_end_transaction(trans, root);
4790 return ret;
4791 }
4792
4793 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4794 0, 0, len, 0, len, 0, 0, 0);
4795 if (ret)
4796 btrfs_abort_transaction(trans, root, ret);
4797 else
4798 btrfs_update_inode(trans, root, inode);
4799 btrfs_end_transaction(trans, root);
4800 return ret;
4801 }
4802
4803 /*
4804 * This function puts in dummy file extents for the area we're creating a hole
4805 * for. So if we are truncating this file to a larger size we need to insert
4806 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4807 * the range between oldsize and size
4808 */
4809 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4810 {
4811 struct btrfs_root *root = BTRFS_I(inode)->root;
4812 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4813 struct extent_map *em = NULL;
4814 struct extent_state *cached_state = NULL;
4815 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4816 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4817 u64 block_end = ALIGN(size, root->sectorsize);
4818 u64 last_byte;
4819 u64 cur_offset;
4820 u64 hole_size;
4821 int err = 0;
4822
4823 /*
4824 * If our size started in the middle of a block we need to zero out the
4825 * rest of the block before we expand the i_size, otherwise we could
4826 * expose stale data.
4827 */
4828 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4829 if (err)
4830 return err;
4831
4832 if (size <= hole_start)
4833 return 0;
4834
4835 while (1) {
4836 struct btrfs_ordered_extent *ordered;
4837
4838 lock_extent_bits(io_tree, hole_start, block_end - 1,
4839 &cached_state);
4840 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4841 block_end - hole_start);
4842 if (!ordered)
4843 break;
4844 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4845 &cached_state, GFP_NOFS);
4846 btrfs_start_ordered_extent(inode, ordered, 1);
4847 btrfs_put_ordered_extent(ordered);
4848 }
4849
4850 cur_offset = hole_start;
4851 while (1) {
4852 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4853 block_end - cur_offset, 0);
4854 if (IS_ERR(em)) {
4855 err = PTR_ERR(em);
4856 em = NULL;
4857 break;
4858 }
4859 last_byte = min(extent_map_end(em), block_end);
4860 last_byte = ALIGN(last_byte , root->sectorsize);
4861 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4862 struct extent_map *hole_em;
4863 hole_size = last_byte - cur_offset;
4864
4865 err = maybe_insert_hole(root, inode, cur_offset,
4866 hole_size);
4867 if (err)
4868 break;
4869 btrfs_drop_extent_cache(inode, cur_offset,
4870 cur_offset + hole_size - 1, 0);
4871 hole_em = alloc_extent_map();
4872 if (!hole_em) {
4873 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4874 &BTRFS_I(inode)->runtime_flags);
4875 goto next;
4876 }
4877 hole_em->start = cur_offset;
4878 hole_em->len = hole_size;
4879 hole_em->orig_start = cur_offset;
4880
4881 hole_em->block_start = EXTENT_MAP_HOLE;
4882 hole_em->block_len = 0;
4883 hole_em->orig_block_len = 0;
4884 hole_em->ram_bytes = hole_size;
4885 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4886 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4887 hole_em->generation = root->fs_info->generation;
4888
4889 while (1) {
4890 write_lock(&em_tree->lock);
4891 err = add_extent_mapping(em_tree, hole_em, 1);
4892 write_unlock(&em_tree->lock);
4893 if (err != -EEXIST)
4894 break;
4895 btrfs_drop_extent_cache(inode, cur_offset,
4896 cur_offset +
4897 hole_size - 1, 0);
4898 }
4899 free_extent_map(hole_em);
4900 }
4901 next:
4902 free_extent_map(em);
4903 em = NULL;
4904 cur_offset = last_byte;
4905 if (cur_offset >= block_end)
4906 break;
4907 }
4908 free_extent_map(em);
4909 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4910 GFP_NOFS);
4911 return err;
4912 }
4913
4914 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4915 {
4916 struct btrfs_root *root = BTRFS_I(inode)->root;
4917 struct btrfs_trans_handle *trans;
4918 loff_t oldsize = i_size_read(inode);
4919 loff_t newsize = attr->ia_size;
4920 int mask = attr->ia_valid;
4921 int ret;
4922
4923 /*
4924 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4925 * special case where we need to update the times despite not having
4926 * these flags set. For all other operations the VFS set these flags
4927 * explicitly if it wants a timestamp update.
4928 */
4929 if (newsize != oldsize) {
4930 inode_inc_iversion(inode);
4931 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4932 inode->i_ctime = inode->i_mtime =
4933 current_fs_time(inode->i_sb);
4934 }
4935
4936 if (newsize > oldsize) {
4937 /*
4938 * Don't do an expanding truncate while snapshoting is ongoing.
4939 * This is to ensure the snapshot captures a fully consistent
4940 * state of this file - if the snapshot captures this expanding
4941 * truncation, it must capture all writes that happened before
4942 * this truncation.
4943 */
4944 btrfs_wait_for_snapshot_creation(root);
4945 ret = btrfs_cont_expand(inode, oldsize, newsize);
4946 if (ret) {
4947 btrfs_end_write_no_snapshoting(root);
4948 return ret;
4949 }
4950
4951 trans = btrfs_start_transaction(root, 1);
4952 if (IS_ERR(trans)) {
4953 btrfs_end_write_no_snapshoting(root);
4954 return PTR_ERR(trans);
4955 }
4956
4957 i_size_write(inode, newsize);
4958 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4959 pagecache_isize_extended(inode, oldsize, newsize);
4960 ret = btrfs_update_inode(trans, root, inode);
4961 btrfs_end_write_no_snapshoting(root);
4962 btrfs_end_transaction(trans, root);
4963 } else {
4964
4965 /*
4966 * We're truncating a file that used to have good data down to
4967 * zero. Make sure it gets into the ordered flush list so that
4968 * any new writes get down to disk quickly.
4969 */
4970 if (newsize == 0)
4971 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4972 &BTRFS_I(inode)->runtime_flags);
4973
4974 /*
4975 * 1 for the orphan item we're going to add
4976 * 1 for the orphan item deletion.
4977 */
4978 trans = btrfs_start_transaction(root, 2);
4979 if (IS_ERR(trans))
4980 return PTR_ERR(trans);
4981
4982 /*
4983 * We need to do this in case we fail at _any_ point during the
4984 * actual truncate. Once we do the truncate_setsize we could
4985 * invalidate pages which forces any outstanding ordered io to
4986 * be instantly completed which will give us extents that need
4987 * to be truncated. If we fail to get an orphan inode down we
4988 * could have left over extents that were never meant to live,
4989 * so we need to guarantee from this point on that everything
4990 * will be consistent.
4991 */
4992 ret = btrfs_orphan_add(trans, inode);
4993 btrfs_end_transaction(trans, root);
4994 if (ret)
4995 return ret;
4996
4997 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4998 truncate_setsize(inode, newsize);
4999
5000 /* Disable nonlocked read DIO to avoid the end less truncate */
5001 btrfs_inode_block_unlocked_dio(inode);
5002 inode_dio_wait(inode);
5003 btrfs_inode_resume_unlocked_dio(inode);
5004
5005 ret = btrfs_truncate(inode);
5006 if (ret && inode->i_nlink) {
5007 int err;
5008
5009 /*
5010 * failed to truncate, disk_i_size is only adjusted down
5011 * as we remove extents, so it should represent the true
5012 * size of the inode, so reset the in memory size and
5013 * delete our orphan entry.
5014 */
5015 trans = btrfs_join_transaction(root);
5016 if (IS_ERR(trans)) {
5017 btrfs_orphan_del(NULL, inode);
5018 return ret;
5019 }
5020 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5021 err = btrfs_orphan_del(trans, inode);
5022 if (err)
5023 btrfs_abort_transaction(trans, root, err);
5024 btrfs_end_transaction(trans, root);
5025 }
5026 }
5027
5028 return ret;
5029 }
5030
5031 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5032 {
5033 struct inode *inode = d_inode(dentry);
5034 struct btrfs_root *root = BTRFS_I(inode)->root;
5035 int err;
5036
5037 if (btrfs_root_readonly(root))
5038 return -EROFS;
5039
5040 err = inode_change_ok(inode, attr);
5041 if (err)
5042 return err;
5043
5044 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5045 err = btrfs_setsize(inode, attr);
5046 if (err)
5047 return err;
5048 }
5049
5050 if (attr->ia_valid) {
5051 setattr_copy(inode, attr);
5052 inode_inc_iversion(inode);
5053 err = btrfs_dirty_inode(inode);
5054
5055 if (!err && attr->ia_valid & ATTR_MODE)
5056 err = posix_acl_chmod(inode, inode->i_mode);
5057 }
5058
5059 return err;
5060 }
5061
5062 /*
5063 * While truncating the inode pages during eviction, we get the VFS calling
5064 * btrfs_invalidatepage() against each page of the inode. This is slow because
5065 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5066 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5067 * extent_state structures over and over, wasting lots of time.
5068 *
5069 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5070 * those expensive operations on a per page basis and do only the ordered io
5071 * finishing, while we release here the extent_map and extent_state structures,
5072 * without the excessive merging and splitting.
5073 */
5074 static void evict_inode_truncate_pages(struct inode *inode)
5075 {
5076 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5077 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5078 struct rb_node *node;
5079
5080 ASSERT(inode->i_state & I_FREEING);
5081 truncate_inode_pages_final(&inode->i_data);
5082
5083 write_lock(&map_tree->lock);
5084 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5085 struct extent_map *em;
5086
5087 node = rb_first(&map_tree->map);
5088 em = rb_entry(node, struct extent_map, rb_node);
5089 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5090 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5091 remove_extent_mapping(map_tree, em);
5092 free_extent_map(em);
5093 if (need_resched()) {
5094 write_unlock(&map_tree->lock);
5095 cond_resched();
5096 write_lock(&map_tree->lock);
5097 }
5098 }
5099 write_unlock(&map_tree->lock);
5100
5101 /*
5102 * Keep looping until we have no more ranges in the io tree.
5103 * We can have ongoing bios started by readpages (called from readahead)
5104 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5105 * still in progress (unlocked the pages in the bio but did not yet
5106 * unlocked the ranges in the io tree). Therefore this means some
5107 * ranges can still be locked and eviction started because before
5108 * submitting those bios, which are executed by a separate task (work
5109 * queue kthread), inode references (inode->i_count) were not taken
5110 * (which would be dropped in the end io callback of each bio).
5111 * Therefore here we effectively end up waiting for those bios and
5112 * anyone else holding locked ranges without having bumped the inode's
5113 * reference count - if we don't do it, when they access the inode's
5114 * io_tree to unlock a range it may be too late, leading to an
5115 * use-after-free issue.
5116 */
5117 spin_lock(&io_tree->lock);
5118 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5119 struct extent_state *state;
5120 struct extent_state *cached_state = NULL;
5121 u64 start;
5122 u64 end;
5123
5124 node = rb_first(&io_tree->state);
5125 state = rb_entry(node, struct extent_state, rb_node);
5126 start = state->start;
5127 end = state->end;
5128 spin_unlock(&io_tree->lock);
5129
5130 lock_extent_bits(io_tree, start, end, &cached_state);
5131
5132 /*
5133 * If still has DELALLOC flag, the extent didn't reach disk,
5134 * and its reserved space won't be freed by delayed_ref.
5135 * So we need to free its reserved space here.
5136 * (Refer to comment in btrfs_invalidatepage, case 2)
5137 *
5138 * Note, end is the bytenr of last byte, so we need + 1 here.
5139 */
5140 if (state->state & EXTENT_DELALLOC)
5141 btrfs_qgroup_free_data(inode, start, end - start + 1);
5142
5143 clear_extent_bit(io_tree, start, end,
5144 EXTENT_LOCKED | EXTENT_DIRTY |
5145 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5146 EXTENT_DEFRAG, 1, 1,
5147 &cached_state, GFP_NOFS);
5148
5149 cond_resched();
5150 spin_lock(&io_tree->lock);
5151 }
5152 spin_unlock(&io_tree->lock);
5153 }
5154
5155 void btrfs_evict_inode(struct inode *inode)
5156 {
5157 struct btrfs_trans_handle *trans;
5158 struct btrfs_root *root = BTRFS_I(inode)->root;
5159 struct btrfs_block_rsv *rsv, *global_rsv;
5160 int steal_from_global = 0;
5161 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5162 int ret;
5163
5164 trace_btrfs_inode_evict(inode);
5165
5166 evict_inode_truncate_pages(inode);
5167
5168 if (inode->i_nlink &&
5169 ((btrfs_root_refs(&root->root_item) != 0 &&
5170 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5171 btrfs_is_free_space_inode(inode)))
5172 goto no_delete;
5173
5174 if (is_bad_inode(inode)) {
5175 btrfs_orphan_del(NULL, inode);
5176 goto no_delete;
5177 }
5178 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5179 if (!special_file(inode->i_mode))
5180 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5181
5182 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5183
5184 if (root->fs_info->log_root_recovering) {
5185 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5186 &BTRFS_I(inode)->runtime_flags));
5187 goto no_delete;
5188 }
5189
5190 if (inode->i_nlink > 0) {
5191 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5192 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5193 goto no_delete;
5194 }
5195
5196 ret = btrfs_commit_inode_delayed_inode(inode);
5197 if (ret) {
5198 btrfs_orphan_del(NULL, inode);
5199 goto no_delete;
5200 }
5201
5202 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5203 if (!rsv) {
5204 btrfs_orphan_del(NULL, inode);
5205 goto no_delete;
5206 }
5207 rsv->size = min_size;
5208 rsv->failfast = 1;
5209 global_rsv = &root->fs_info->global_block_rsv;
5210
5211 btrfs_i_size_write(inode, 0);
5212
5213 /*
5214 * This is a bit simpler than btrfs_truncate since we've already
5215 * reserved our space for our orphan item in the unlink, so we just
5216 * need to reserve some slack space in case we add bytes and update
5217 * inode item when doing the truncate.
5218 */
5219 while (1) {
5220 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5221 BTRFS_RESERVE_FLUSH_LIMIT);
5222
5223 /*
5224 * Try and steal from the global reserve since we will
5225 * likely not use this space anyway, we want to try as
5226 * hard as possible to get this to work.
5227 */
5228 if (ret)
5229 steal_from_global++;
5230 else
5231 steal_from_global = 0;
5232 ret = 0;
5233
5234 /*
5235 * steal_from_global == 0: we reserved stuff, hooray!
5236 * steal_from_global == 1: we didn't reserve stuff, boo!
5237 * steal_from_global == 2: we've committed, still not a lot of
5238 * room but maybe we'll have room in the global reserve this
5239 * time.
5240 * steal_from_global == 3: abandon all hope!
5241 */
5242 if (steal_from_global > 2) {
5243 btrfs_warn(root->fs_info,
5244 "Could not get space for a delete, will truncate on mount %d",
5245 ret);
5246 btrfs_orphan_del(NULL, inode);
5247 btrfs_free_block_rsv(root, rsv);
5248 goto no_delete;
5249 }
5250
5251 trans = btrfs_join_transaction(root);
5252 if (IS_ERR(trans)) {
5253 btrfs_orphan_del(NULL, inode);
5254 btrfs_free_block_rsv(root, rsv);
5255 goto no_delete;
5256 }
5257
5258 /*
5259 * We can't just steal from the global reserve, we need to make
5260 * sure there is room to do it, if not we need to commit and try
5261 * again.
5262 */
5263 if (steal_from_global) {
5264 if (!btrfs_check_space_for_delayed_refs(trans, root))
5265 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5266 min_size, 0);
5267 else
5268 ret = -ENOSPC;
5269 }
5270
5271 /*
5272 * Couldn't steal from the global reserve, we have too much
5273 * pending stuff built up, commit the transaction and try it
5274 * again.
5275 */
5276 if (ret) {
5277 ret = btrfs_commit_transaction(trans, root);
5278 if (ret) {
5279 btrfs_orphan_del(NULL, inode);
5280 btrfs_free_block_rsv(root, rsv);
5281 goto no_delete;
5282 }
5283 continue;
5284 } else {
5285 steal_from_global = 0;
5286 }
5287
5288 trans->block_rsv = rsv;
5289
5290 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5291 if (ret != -ENOSPC && ret != -EAGAIN)
5292 break;
5293
5294 trans->block_rsv = &root->fs_info->trans_block_rsv;
5295 btrfs_end_transaction(trans, root);
5296 trans = NULL;
5297 btrfs_btree_balance_dirty(root);
5298 }
5299
5300 btrfs_free_block_rsv(root, rsv);
5301
5302 /*
5303 * Errors here aren't a big deal, it just means we leave orphan items
5304 * in the tree. They will be cleaned up on the next mount.
5305 */
5306 if (ret == 0) {
5307 trans->block_rsv = root->orphan_block_rsv;
5308 btrfs_orphan_del(trans, inode);
5309 } else {
5310 btrfs_orphan_del(NULL, inode);
5311 }
5312
5313 trans->block_rsv = &root->fs_info->trans_block_rsv;
5314 if (!(root == root->fs_info->tree_root ||
5315 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5316 btrfs_return_ino(root, btrfs_ino(inode));
5317
5318 btrfs_end_transaction(trans, root);
5319 btrfs_btree_balance_dirty(root);
5320 no_delete:
5321 btrfs_remove_delayed_node(inode);
5322 clear_inode(inode);
5323 }
5324
5325 /*
5326 * this returns the key found in the dir entry in the location pointer.
5327 * If no dir entries were found, location->objectid is 0.
5328 */
5329 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5330 struct btrfs_key *location)
5331 {
5332 const char *name = dentry->d_name.name;
5333 int namelen = dentry->d_name.len;
5334 struct btrfs_dir_item *di;
5335 struct btrfs_path *path;
5336 struct btrfs_root *root = BTRFS_I(dir)->root;
5337 int ret = 0;
5338
5339 path = btrfs_alloc_path();
5340 if (!path)
5341 return -ENOMEM;
5342
5343 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5344 namelen, 0);
5345 if (IS_ERR(di))
5346 ret = PTR_ERR(di);
5347
5348 if (IS_ERR_OR_NULL(di))
5349 goto out_err;
5350
5351 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5352 out:
5353 btrfs_free_path(path);
5354 return ret;
5355 out_err:
5356 location->objectid = 0;
5357 goto out;
5358 }
5359
5360 /*
5361 * when we hit a tree root in a directory, the btrfs part of the inode
5362 * needs to be changed to reflect the root directory of the tree root. This
5363 * is kind of like crossing a mount point.
5364 */
5365 static int fixup_tree_root_location(struct btrfs_root *root,
5366 struct inode *dir,
5367 struct dentry *dentry,
5368 struct btrfs_key *location,
5369 struct btrfs_root **sub_root)
5370 {
5371 struct btrfs_path *path;
5372 struct btrfs_root *new_root;
5373 struct btrfs_root_ref *ref;
5374 struct extent_buffer *leaf;
5375 struct btrfs_key key;
5376 int ret;
5377 int err = 0;
5378
5379 path = btrfs_alloc_path();
5380 if (!path) {
5381 err = -ENOMEM;
5382 goto out;
5383 }
5384
5385 err = -ENOENT;
5386 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5387 key.type = BTRFS_ROOT_REF_KEY;
5388 key.offset = location->objectid;
5389
5390 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5391 0, 0);
5392 if (ret) {
5393 if (ret < 0)
5394 err = ret;
5395 goto out;
5396 }
5397
5398 leaf = path->nodes[0];
5399 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5400 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5401 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5402 goto out;
5403
5404 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5405 (unsigned long)(ref + 1),
5406 dentry->d_name.len);
5407 if (ret)
5408 goto out;
5409
5410 btrfs_release_path(path);
5411
5412 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5413 if (IS_ERR(new_root)) {
5414 err = PTR_ERR(new_root);
5415 goto out;
5416 }
5417
5418 *sub_root = new_root;
5419 location->objectid = btrfs_root_dirid(&new_root->root_item);
5420 location->type = BTRFS_INODE_ITEM_KEY;
5421 location->offset = 0;
5422 err = 0;
5423 out:
5424 btrfs_free_path(path);
5425 return err;
5426 }
5427
5428 static void inode_tree_add(struct inode *inode)
5429 {
5430 struct btrfs_root *root = BTRFS_I(inode)->root;
5431 struct btrfs_inode *entry;
5432 struct rb_node **p;
5433 struct rb_node *parent;
5434 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5435 u64 ino = btrfs_ino(inode);
5436
5437 if (inode_unhashed(inode))
5438 return;
5439 parent = NULL;
5440 spin_lock(&root->inode_lock);
5441 p = &root->inode_tree.rb_node;
5442 while (*p) {
5443 parent = *p;
5444 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5445
5446 if (ino < btrfs_ino(&entry->vfs_inode))
5447 p = &parent->rb_left;
5448 else if (ino > btrfs_ino(&entry->vfs_inode))
5449 p = &parent->rb_right;
5450 else {
5451 WARN_ON(!(entry->vfs_inode.i_state &
5452 (I_WILL_FREE | I_FREEING)));
5453 rb_replace_node(parent, new, &root->inode_tree);
5454 RB_CLEAR_NODE(parent);
5455 spin_unlock(&root->inode_lock);
5456 return;
5457 }
5458 }
5459 rb_link_node(new, parent, p);
5460 rb_insert_color(new, &root->inode_tree);
5461 spin_unlock(&root->inode_lock);
5462 }
5463
5464 static void inode_tree_del(struct inode *inode)
5465 {
5466 struct btrfs_root *root = BTRFS_I(inode)->root;
5467 int empty = 0;
5468
5469 spin_lock(&root->inode_lock);
5470 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5471 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5472 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5473 empty = RB_EMPTY_ROOT(&root->inode_tree);
5474 }
5475 spin_unlock(&root->inode_lock);
5476
5477 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5478 synchronize_srcu(&root->fs_info->subvol_srcu);
5479 spin_lock(&root->inode_lock);
5480 empty = RB_EMPTY_ROOT(&root->inode_tree);
5481 spin_unlock(&root->inode_lock);
5482 if (empty)
5483 btrfs_add_dead_root(root);
5484 }
5485 }
5486
5487 void btrfs_invalidate_inodes(struct btrfs_root *root)
5488 {
5489 struct rb_node *node;
5490 struct rb_node *prev;
5491 struct btrfs_inode *entry;
5492 struct inode *inode;
5493 u64 objectid = 0;
5494
5495 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5496 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5497
5498 spin_lock(&root->inode_lock);
5499 again:
5500 node = root->inode_tree.rb_node;
5501 prev = NULL;
5502 while (node) {
5503 prev = node;
5504 entry = rb_entry(node, struct btrfs_inode, rb_node);
5505
5506 if (objectid < btrfs_ino(&entry->vfs_inode))
5507 node = node->rb_left;
5508 else if (objectid > btrfs_ino(&entry->vfs_inode))
5509 node = node->rb_right;
5510 else
5511 break;
5512 }
5513 if (!node) {
5514 while (prev) {
5515 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5516 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5517 node = prev;
5518 break;
5519 }
5520 prev = rb_next(prev);
5521 }
5522 }
5523 while (node) {
5524 entry = rb_entry(node, struct btrfs_inode, rb_node);
5525 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5526 inode = igrab(&entry->vfs_inode);
5527 if (inode) {
5528 spin_unlock(&root->inode_lock);
5529 if (atomic_read(&inode->i_count) > 1)
5530 d_prune_aliases(inode);
5531 /*
5532 * btrfs_drop_inode will have it removed from
5533 * the inode cache when its usage count
5534 * hits zero.
5535 */
5536 iput(inode);
5537 cond_resched();
5538 spin_lock(&root->inode_lock);
5539 goto again;
5540 }
5541
5542 if (cond_resched_lock(&root->inode_lock))
5543 goto again;
5544
5545 node = rb_next(node);
5546 }
5547 spin_unlock(&root->inode_lock);
5548 }
5549
5550 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5551 {
5552 struct btrfs_iget_args *args = p;
5553 inode->i_ino = args->location->objectid;
5554 memcpy(&BTRFS_I(inode)->location, args->location,
5555 sizeof(*args->location));
5556 BTRFS_I(inode)->root = args->root;
5557 return 0;
5558 }
5559
5560 static int btrfs_find_actor(struct inode *inode, void *opaque)
5561 {
5562 struct btrfs_iget_args *args = opaque;
5563 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5564 args->root == BTRFS_I(inode)->root;
5565 }
5566
5567 static struct inode *btrfs_iget_locked(struct super_block *s,
5568 struct btrfs_key *location,
5569 struct btrfs_root *root)
5570 {
5571 struct inode *inode;
5572 struct btrfs_iget_args args;
5573 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5574
5575 args.location = location;
5576 args.root = root;
5577
5578 inode = iget5_locked(s, hashval, btrfs_find_actor,
5579 btrfs_init_locked_inode,
5580 (void *)&args);
5581 return inode;
5582 }
5583
5584 /* Get an inode object given its location and corresponding root.
5585 * Returns in *is_new if the inode was read from disk
5586 */
5587 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5588 struct btrfs_root *root, int *new)
5589 {
5590 struct inode *inode;
5591
5592 inode = btrfs_iget_locked(s, location, root);
5593 if (!inode)
5594 return ERR_PTR(-ENOMEM);
5595
5596 if (inode->i_state & I_NEW) {
5597 btrfs_read_locked_inode(inode);
5598 if (!is_bad_inode(inode)) {
5599 inode_tree_add(inode);
5600 unlock_new_inode(inode);
5601 if (new)
5602 *new = 1;
5603 } else {
5604 unlock_new_inode(inode);
5605 iput(inode);
5606 inode = ERR_PTR(-ESTALE);
5607 }
5608 }
5609
5610 return inode;
5611 }
5612
5613 static struct inode *new_simple_dir(struct super_block *s,
5614 struct btrfs_key *key,
5615 struct btrfs_root *root)
5616 {
5617 struct inode *inode = new_inode(s);
5618
5619 if (!inode)
5620 return ERR_PTR(-ENOMEM);
5621
5622 BTRFS_I(inode)->root = root;
5623 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5624 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5625
5626 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5627 inode->i_op = &btrfs_dir_ro_inode_operations;
5628 inode->i_fop = &simple_dir_operations;
5629 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5630 inode->i_mtime = current_fs_time(inode->i_sb);
5631 inode->i_atime = inode->i_mtime;
5632 inode->i_ctime = inode->i_mtime;
5633 BTRFS_I(inode)->i_otime = inode->i_mtime;
5634
5635 return inode;
5636 }
5637
5638 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5639 {
5640 struct inode *inode;
5641 struct btrfs_root *root = BTRFS_I(dir)->root;
5642 struct btrfs_root *sub_root = root;
5643 struct btrfs_key location;
5644 int index;
5645 int ret = 0;
5646
5647 if (dentry->d_name.len > BTRFS_NAME_LEN)
5648 return ERR_PTR(-ENAMETOOLONG);
5649
5650 ret = btrfs_inode_by_name(dir, dentry, &location);
5651 if (ret < 0)
5652 return ERR_PTR(ret);
5653
5654 if (location.objectid == 0)
5655 return ERR_PTR(-ENOENT);
5656
5657 if (location.type == BTRFS_INODE_ITEM_KEY) {
5658 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5659 return inode;
5660 }
5661
5662 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5663
5664 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5665 ret = fixup_tree_root_location(root, dir, dentry,
5666 &location, &sub_root);
5667 if (ret < 0) {
5668 if (ret != -ENOENT)
5669 inode = ERR_PTR(ret);
5670 else
5671 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5672 } else {
5673 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5674 }
5675 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5676
5677 if (!IS_ERR(inode) && root != sub_root) {
5678 down_read(&root->fs_info->cleanup_work_sem);
5679 if (!(inode->i_sb->s_flags & MS_RDONLY))
5680 ret = btrfs_orphan_cleanup(sub_root);
5681 up_read(&root->fs_info->cleanup_work_sem);
5682 if (ret) {
5683 iput(inode);
5684 inode = ERR_PTR(ret);
5685 }
5686 }
5687
5688 return inode;
5689 }
5690
5691 static int btrfs_dentry_delete(const struct dentry *dentry)
5692 {
5693 struct btrfs_root *root;
5694 struct inode *inode = d_inode(dentry);
5695
5696 if (!inode && !IS_ROOT(dentry))
5697 inode = d_inode(dentry->d_parent);
5698
5699 if (inode) {
5700 root = BTRFS_I(inode)->root;
5701 if (btrfs_root_refs(&root->root_item) == 0)
5702 return 1;
5703
5704 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5705 return 1;
5706 }
5707 return 0;
5708 }
5709
5710 static void btrfs_dentry_release(struct dentry *dentry)
5711 {
5712 kfree(dentry->d_fsdata);
5713 }
5714
5715 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5716 unsigned int flags)
5717 {
5718 struct inode *inode;
5719
5720 inode = btrfs_lookup_dentry(dir, dentry);
5721 if (IS_ERR(inode)) {
5722 if (PTR_ERR(inode) == -ENOENT)
5723 inode = NULL;
5724 else
5725 return ERR_CAST(inode);
5726 }
5727
5728 return d_splice_alias(inode, dentry);
5729 }
5730
5731 unsigned char btrfs_filetype_table[] = {
5732 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5733 };
5734
5735 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5736 {
5737 struct inode *inode = file_inode(file);
5738 struct btrfs_root *root = BTRFS_I(inode)->root;
5739 struct btrfs_item *item;
5740 struct btrfs_dir_item *di;
5741 struct btrfs_key key;
5742 struct btrfs_key found_key;
5743 struct btrfs_path *path;
5744 struct list_head ins_list;
5745 struct list_head del_list;
5746 int ret;
5747 struct extent_buffer *leaf;
5748 int slot;
5749 unsigned char d_type;
5750 int over = 0;
5751 u32 di_cur;
5752 u32 di_total;
5753 u32 di_len;
5754 int key_type = BTRFS_DIR_INDEX_KEY;
5755 char tmp_name[32];
5756 char *name_ptr;
5757 int name_len;
5758 int is_curr = 0; /* ctx->pos points to the current index? */
5759 bool emitted;
5760 bool put = false;
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 put = btrfs_readdir_get_delayed_items(inode, &ins_list,
5779 &del_list);
5780 }
5781
5782 key.type = key_type;
5783 key.offset = ctx->pos;
5784 key.objectid = btrfs_ino(inode);
5785
5786 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5787 if (ret < 0)
5788 goto err;
5789
5790 emitted = false;
5791 while (1) {
5792 leaf = path->nodes[0];
5793 slot = path->slots[0];
5794 if (slot >= btrfs_header_nritems(leaf)) {
5795 ret = btrfs_next_leaf(root, path);
5796 if (ret < 0)
5797 goto err;
5798 else if (ret > 0)
5799 break;
5800 continue;
5801 }
5802
5803 item = btrfs_item_nr(slot);
5804 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5805
5806 if (found_key.objectid != key.objectid)
5807 break;
5808 if (found_key.type != key_type)
5809 break;
5810 if (found_key.offset < ctx->pos)
5811 goto next;
5812 if (key_type == BTRFS_DIR_INDEX_KEY &&
5813 btrfs_should_delete_dir_index(&del_list,
5814 found_key.offset))
5815 goto next;
5816
5817 ctx->pos = found_key.offset;
5818 is_curr = 1;
5819
5820 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5821 di_cur = 0;
5822 di_total = btrfs_item_size(leaf, item);
5823
5824 while (di_cur < di_total) {
5825 struct btrfs_key location;
5826
5827 if (verify_dir_item(root, leaf, di))
5828 break;
5829
5830 name_len = btrfs_dir_name_len(leaf, di);
5831 if (name_len <= sizeof(tmp_name)) {
5832 name_ptr = tmp_name;
5833 } else {
5834 name_ptr = kmalloc(name_len, GFP_KERNEL);
5835 if (!name_ptr) {
5836 ret = -ENOMEM;
5837 goto err;
5838 }
5839 }
5840 read_extent_buffer(leaf, name_ptr,
5841 (unsigned long)(di + 1), name_len);
5842
5843 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5844 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5845
5846
5847 /* is this a reference to our own snapshot? If so
5848 * skip it.
5849 *
5850 * In contrast to old kernels, we insert the snapshot's
5851 * dir item and dir index after it has been created, so
5852 * we won't find a reference to our own snapshot. We
5853 * still keep the following code for backward
5854 * compatibility.
5855 */
5856 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5857 location.objectid == root->root_key.objectid) {
5858 over = 0;
5859 goto skip;
5860 }
5861 over = !dir_emit(ctx, name_ptr, name_len,
5862 location.objectid, d_type);
5863
5864 skip:
5865 if (name_ptr != tmp_name)
5866 kfree(name_ptr);
5867
5868 if (over)
5869 goto nopos;
5870 emitted = true;
5871 di_len = btrfs_dir_name_len(leaf, di) +
5872 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5873 di_cur += di_len;
5874 di = (struct btrfs_dir_item *)((char *)di + di_len);
5875 }
5876 next:
5877 path->slots[0]++;
5878 }
5879
5880 if (key_type == BTRFS_DIR_INDEX_KEY) {
5881 if (is_curr)
5882 ctx->pos++;
5883 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted);
5884 if (ret)
5885 goto nopos;
5886 }
5887
5888 /*
5889 * If we haven't emitted any dir entry, we must not touch ctx->pos as
5890 * it was was set to the termination value in previous call. We assume
5891 * that "." and ".." were emitted if we reach this point and set the
5892 * termination value as well for an empty directory.
5893 */
5894 if (ctx->pos > 2 && !emitted)
5895 goto nopos;
5896
5897 /* Reached end of directory/root. Bump pos past the last item. */
5898 ctx->pos++;
5899
5900 /*
5901 * Stop new entries from being returned after we return the last
5902 * entry.
5903 *
5904 * New directory entries are assigned a strictly increasing
5905 * offset. This means that new entries created during readdir
5906 * are *guaranteed* to be seen in the future by that readdir.
5907 * This has broken buggy programs which operate on names as
5908 * they're returned by readdir. Until we re-use freed offsets
5909 * we have this hack to stop new entries from being returned
5910 * under the assumption that they'll never reach this huge
5911 * offset.
5912 *
5913 * This is being careful not to overflow 32bit loff_t unless the
5914 * last entry requires it because doing so has broken 32bit apps
5915 * in the past.
5916 */
5917 if (key_type == BTRFS_DIR_INDEX_KEY) {
5918 if (ctx->pos >= INT_MAX)
5919 ctx->pos = LLONG_MAX;
5920 else
5921 ctx->pos = INT_MAX;
5922 }
5923 nopos:
5924 ret = 0;
5925 err:
5926 if (put)
5927 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5928 btrfs_free_path(path);
5929 return ret;
5930 }
5931
5932 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5933 {
5934 struct btrfs_root *root = BTRFS_I(inode)->root;
5935 struct btrfs_trans_handle *trans;
5936 int ret = 0;
5937 bool nolock = false;
5938
5939 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5940 return 0;
5941
5942 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5943 nolock = true;
5944
5945 if (wbc->sync_mode == WB_SYNC_ALL) {
5946 if (nolock)
5947 trans = btrfs_join_transaction_nolock(root);
5948 else
5949 trans = btrfs_join_transaction(root);
5950 if (IS_ERR(trans))
5951 return PTR_ERR(trans);
5952 ret = btrfs_commit_transaction(trans, root);
5953 }
5954 return ret;
5955 }
5956
5957 /*
5958 * This is somewhat expensive, updating the tree every time the
5959 * inode changes. But, it is most likely to find the inode in cache.
5960 * FIXME, needs more benchmarking...there are no reasons other than performance
5961 * to keep or drop this code.
5962 */
5963 static int btrfs_dirty_inode(struct inode *inode)
5964 {
5965 struct btrfs_root *root = BTRFS_I(inode)->root;
5966 struct btrfs_trans_handle *trans;
5967 int ret;
5968
5969 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5970 return 0;
5971
5972 trans = btrfs_join_transaction(root);
5973 if (IS_ERR(trans))
5974 return PTR_ERR(trans);
5975
5976 ret = btrfs_update_inode(trans, root, inode);
5977 if (ret && ret == -ENOSPC) {
5978 /* whoops, lets try again with the full transaction */
5979 btrfs_end_transaction(trans, root);
5980 trans = btrfs_start_transaction(root, 1);
5981 if (IS_ERR(trans))
5982 return PTR_ERR(trans);
5983
5984 ret = btrfs_update_inode(trans, root, inode);
5985 }
5986 btrfs_end_transaction(trans, root);
5987 if (BTRFS_I(inode)->delayed_node)
5988 btrfs_balance_delayed_items(root);
5989
5990 return ret;
5991 }
5992
5993 /*
5994 * This is a copy of file_update_time. We need this so we can return error on
5995 * ENOSPC for updating the inode in the case of file write and mmap writes.
5996 */
5997 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5998 int flags)
5999 {
6000 struct btrfs_root *root = BTRFS_I(inode)->root;
6001
6002 if (btrfs_root_readonly(root))
6003 return -EROFS;
6004
6005 if (flags & S_VERSION)
6006 inode_inc_iversion(inode);
6007 if (flags & S_CTIME)
6008 inode->i_ctime = *now;
6009 if (flags & S_MTIME)
6010 inode->i_mtime = *now;
6011 if (flags & S_ATIME)
6012 inode->i_atime = *now;
6013 return btrfs_dirty_inode(inode);
6014 }
6015
6016 /*
6017 * find the highest existing sequence number in a directory
6018 * and then set the in-memory index_cnt variable to reflect
6019 * free sequence numbers
6020 */
6021 static int btrfs_set_inode_index_count(struct inode *inode)
6022 {
6023 struct btrfs_root *root = BTRFS_I(inode)->root;
6024 struct btrfs_key key, found_key;
6025 struct btrfs_path *path;
6026 struct extent_buffer *leaf;
6027 int ret;
6028
6029 key.objectid = btrfs_ino(inode);
6030 key.type = BTRFS_DIR_INDEX_KEY;
6031 key.offset = (u64)-1;
6032
6033 path = btrfs_alloc_path();
6034 if (!path)
6035 return -ENOMEM;
6036
6037 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6038 if (ret < 0)
6039 goto out;
6040 /* FIXME: we should be able to handle this */
6041 if (ret == 0)
6042 goto out;
6043 ret = 0;
6044
6045 /*
6046 * MAGIC NUMBER EXPLANATION:
6047 * since we search a directory based on f_pos we have to start at 2
6048 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6049 * else has to start at 2
6050 */
6051 if (path->slots[0] == 0) {
6052 BTRFS_I(inode)->index_cnt = 2;
6053 goto out;
6054 }
6055
6056 path->slots[0]--;
6057
6058 leaf = path->nodes[0];
6059 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6060
6061 if (found_key.objectid != btrfs_ino(inode) ||
6062 found_key.type != BTRFS_DIR_INDEX_KEY) {
6063 BTRFS_I(inode)->index_cnt = 2;
6064 goto out;
6065 }
6066
6067 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6068 out:
6069 btrfs_free_path(path);
6070 return ret;
6071 }
6072
6073 /*
6074 * helper to find a free sequence number in a given directory. This current
6075 * code is very simple, later versions will do smarter things in the btree
6076 */
6077 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6078 {
6079 int ret = 0;
6080
6081 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6082 ret = btrfs_inode_delayed_dir_index_count(dir);
6083 if (ret) {
6084 ret = btrfs_set_inode_index_count(dir);
6085 if (ret)
6086 return ret;
6087 }
6088 }
6089
6090 *index = BTRFS_I(dir)->index_cnt;
6091 BTRFS_I(dir)->index_cnt++;
6092
6093 return ret;
6094 }
6095
6096 static int btrfs_insert_inode_locked(struct inode *inode)
6097 {
6098 struct btrfs_iget_args args;
6099 args.location = &BTRFS_I(inode)->location;
6100 args.root = BTRFS_I(inode)->root;
6101
6102 return insert_inode_locked4(inode,
6103 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6104 btrfs_find_actor, &args);
6105 }
6106
6107 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6108 struct btrfs_root *root,
6109 struct inode *dir,
6110 const char *name, int name_len,
6111 u64 ref_objectid, u64 objectid,
6112 umode_t mode, u64 *index)
6113 {
6114 struct inode *inode;
6115 struct btrfs_inode_item *inode_item;
6116 struct btrfs_key *location;
6117 struct btrfs_path *path;
6118 struct btrfs_inode_ref *ref;
6119 struct btrfs_key key[2];
6120 u32 sizes[2];
6121 int nitems = name ? 2 : 1;
6122 unsigned long ptr;
6123 int ret;
6124
6125 path = btrfs_alloc_path();
6126 if (!path)
6127 return ERR_PTR(-ENOMEM);
6128
6129 inode = new_inode(root->fs_info->sb);
6130 if (!inode) {
6131 btrfs_free_path(path);
6132 return ERR_PTR(-ENOMEM);
6133 }
6134
6135 /*
6136 * O_TMPFILE, set link count to 0, so that after this point,
6137 * we fill in an inode item with the correct link count.
6138 */
6139 if (!name)
6140 set_nlink(inode, 0);
6141
6142 /*
6143 * we have to initialize this early, so we can reclaim the inode
6144 * number if we fail afterwards in this function.
6145 */
6146 inode->i_ino = objectid;
6147
6148 if (dir && name) {
6149 trace_btrfs_inode_request(dir);
6150
6151 ret = btrfs_set_inode_index(dir, index);
6152 if (ret) {
6153 btrfs_free_path(path);
6154 iput(inode);
6155 return ERR_PTR(ret);
6156 }
6157 } else if (dir) {
6158 *index = 0;
6159 }
6160 /*
6161 * index_cnt is ignored for everything but a dir,
6162 * btrfs_get_inode_index_count has an explanation for the magic
6163 * number
6164 */
6165 BTRFS_I(inode)->index_cnt = 2;
6166 BTRFS_I(inode)->dir_index = *index;
6167 BTRFS_I(inode)->root = root;
6168 BTRFS_I(inode)->generation = trans->transid;
6169 inode->i_generation = BTRFS_I(inode)->generation;
6170
6171 /*
6172 * We could have gotten an inode number from somebody who was fsynced
6173 * and then removed in this same transaction, so let's just set full
6174 * sync since it will be a full sync anyway and this will blow away the
6175 * old info in the log.
6176 */
6177 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6178
6179 key[0].objectid = objectid;
6180 key[0].type = BTRFS_INODE_ITEM_KEY;
6181 key[0].offset = 0;
6182
6183 sizes[0] = sizeof(struct btrfs_inode_item);
6184
6185 if (name) {
6186 /*
6187 * Start new inodes with an inode_ref. This is slightly more
6188 * efficient for small numbers of hard links since they will
6189 * be packed into one item. Extended refs will kick in if we
6190 * add more hard links than can fit in the ref item.
6191 */
6192 key[1].objectid = objectid;
6193 key[1].type = BTRFS_INODE_REF_KEY;
6194 key[1].offset = ref_objectid;
6195
6196 sizes[1] = name_len + sizeof(*ref);
6197 }
6198
6199 location = &BTRFS_I(inode)->location;
6200 location->objectid = objectid;
6201 location->offset = 0;
6202 location->type = BTRFS_INODE_ITEM_KEY;
6203
6204 ret = btrfs_insert_inode_locked(inode);
6205 if (ret < 0)
6206 goto fail;
6207
6208 path->leave_spinning = 1;
6209 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6210 if (ret != 0)
6211 goto fail_unlock;
6212
6213 inode_init_owner(inode, dir, mode);
6214 inode_set_bytes(inode, 0);
6215
6216 inode->i_mtime = current_fs_time(inode->i_sb);
6217 inode->i_atime = inode->i_mtime;
6218 inode->i_ctime = inode->i_mtime;
6219 BTRFS_I(inode)->i_otime = inode->i_mtime;
6220
6221 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6222 struct btrfs_inode_item);
6223 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6224 sizeof(*inode_item));
6225 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6226
6227 if (name) {
6228 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6229 struct btrfs_inode_ref);
6230 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6231 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6232 ptr = (unsigned long)(ref + 1);
6233 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6234 }
6235
6236 btrfs_mark_buffer_dirty(path->nodes[0]);
6237 btrfs_free_path(path);
6238
6239 btrfs_inherit_iflags(inode, dir);
6240
6241 if (S_ISREG(mode)) {
6242 if (btrfs_test_opt(root, NODATASUM))
6243 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6244 if (btrfs_test_opt(root, NODATACOW))
6245 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6246 BTRFS_INODE_NODATASUM;
6247 }
6248
6249 inode_tree_add(inode);
6250
6251 trace_btrfs_inode_new(inode);
6252 btrfs_set_inode_last_trans(trans, inode);
6253
6254 btrfs_update_root_times(trans, root);
6255
6256 ret = btrfs_inode_inherit_props(trans, inode, dir);
6257 if (ret)
6258 btrfs_err(root->fs_info,
6259 "error inheriting props for ino %llu (root %llu): %d",
6260 btrfs_ino(inode), root->root_key.objectid, ret);
6261
6262 return inode;
6263
6264 fail_unlock:
6265 unlock_new_inode(inode);
6266 fail:
6267 if (dir && name)
6268 BTRFS_I(dir)->index_cnt--;
6269 btrfs_free_path(path);
6270 iput(inode);
6271 return ERR_PTR(ret);
6272 }
6273
6274 static inline u8 btrfs_inode_type(struct inode *inode)
6275 {
6276 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6277 }
6278
6279 /*
6280 * utility function to add 'inode' into 'parent_inode' with
6281 * a give name and a given sequence number.
6282 * if 'add_backref' is true, also insert a backref from the
6283 * inode to the parent directory.
6284 */
6285 int btrfs_add_link(struct btrfs_trans_handle *trans,
6286 struct inode *parent_inode, struct inode *inode,
6287 const char *name, int name_len, int add_backref, u64 index)
6288 {
6289 int ret = 0;
6290 struct btrfs_key key;
6291 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6292 u64 ino = btrfs_ino(inode);
6293 u64 parent_ino = btrfs_ino(parent_inode);
6294
6295 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6296 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6297 } else {
6298 key.objectid = ino;
6299 key.type = BTRFS_INODE_ITEM_KEY;
6300 key.offset = 0;
6301 }
6302
6303 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6304 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6305 key.objectid, root->root_key.objectid,
6306 parent_ino, index, name, name_len);
6307 } else if (add_backref) {
6308 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6309 parent_ino, index);
6310 }
6311
6312 /* Nothing to clean up yet */
6313 if (ret)
6314 return ret;
6315
6316 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6317 parent_inode, &key,
6318 btrfs_inode_type(inode), index);
6319 if (ret == -EEXIST || ret == -EOVERFLOW)
6320 goto fail_dir_item;
6321 else if (ret) {
6322 btrfs_abort_transaction(trans, root, ret);
6323 return ret;
6324 }
6325
6326 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6327 name_len * 2);
6328 inode_inc_iversion(parent_inode);
6329 parent_inode->i_mtime = parent_inode->i_ctime =
6330 current_fs_time(parent_inode->i_sb);
6331 ret = btrfs_update_inode(trans, root, parent_inode);
6332 if (ret)
6333 btrfs_abort_transaction(trans, root, ret);
6334 return ret;
6335
6336 fail_dir_item:
6337 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6338 u64 local_index;
6339 int err;
6340 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6341 key.objectid, root->root_key.objectid,
6342 parent_ino, &local_index, name, name_len);
6343
6344 } else if (add_backref) {
6345 u64 local_index;
6346 int err;
6347
6348 err = btrfs_del_inode_ref(trans, root, name, name_len,
6349 ino, parent_ino, &local_index);
6350 }
6351 return ret;
6352 }
6353
6354 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6355 struct inode *dir, struct dentry *dentry,
6356 struct inode *inode, int backref, u64 index)
6357 {
6358 int err = btrfs_add_link(trans, dir, inode,
6359 dentry->d_name.name, dentry->d_name.len,
6360 backref, index);
6361 if (err > 0)
6362 err = -EEXIST;
6363 return err;
6364 }
6365
6366 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6367 umode_t mode, dev_t rdev)
6368 {
6369 struct btrfs_trans_handle *trans;
6370 struct btrfs_root *root = BTRFS_I(dir)->root;
6371 struct inode *inode = NULL;
6372 int err;
6373 int drop_inode = 0;
6374 u64 objectid;
6375 u64 index = 0;
6376
6377 /*
6378 * 2 for inode item and ref
6379 * 2 for dir items
6380 * 1 for xattr if selinux is on
6381 */
6382 trans = btrfs_start_transaction(root, 5);
6383 if (IS_ERR(trans))
6384 return PTR_ERR(trans);
6385
6386 err = btrfs_find_free_ino(root, &objectid);
6387 if (err)
6388 goto out_unlock;
6389
6390 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6391 dentry->d_name.len, btrfs_ino(dir), objectid,
6392 mode, &index);
6393 if (IS_ERR(inode)) {
6394 err = PTR_ERR(inode);
6395 goto out_unlock;
6396 }
6397
6398 /*
6399 * If the active LSM wants to access the inode during
6400 * d_instantiate it needs these. Smack checks to see
6401 * if the filesystem supports xattrs by looking at the
6402 * ops vector.
6403 */
6404 inode->i_op = &btrfs_special_inode_operations;
6405 init_special_inode(inode, inode->i_mode, rdev);
6406
6407 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6408 if (err)
6409 goto out_unlock_inode;
6410
6411 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6412 if (err) {
6413 goto out_unlock_inode;
6414 } else {
6415 btrfs_update_inode(trans, root, inode);
6416 unlock_new_inode(inode);
6417 d_instantiate(dentry, inode);
6418 }
6419
6420 out_unlock:
6421 btrfs_end_transaction(trans, root);
6422 btrfs_balance_delayed_items(root);
6423 btrfs_btree_balance_dirty(root);
6424 if (drop_inode) {
6425 inode_dec_link_count(inode);
6426 iput(inode);
6427 }
6428 return err;
6429
6430 out_unlock_inode:
6431 drop_inode = 1;
6432 unlock_new_inode(inode);
6433 goto out_unlock;
6434
6435 }
6436
6437 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6438 umode_t mode, bool excl)
6439 {
6440 struct btrfs_trans_handle *trans;
6441 struct btrfs_root *root = BTRFS_I(dir)->root;
6442 struct inode *inode = NULL;
6443 int drop_inode_on_err = 0;
6444 int err;
6445 u64 objectid;
6446 u64 index = 0;
6447
6448 /*
6449 * 2 for inode item and ref
6450 * 2 for dir items
6451 * 1 for xattr if selinux is on
6452 */
6453 trans = btrfs_start_transaction(root, 5);
6454 if (IS_ERR(trans))
6455 return PTR_ERR(trans);
6456
6457 err = btrfs_find_free_ino(root, &objectid);
6458 if (err)
6459 goto out_unlock;
6460
6461 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6462 dentry->d_name.len, btrfs_ino(dir), objectid,
6463 mode, &index);
6464 if (IS_ERR(inode)) {
6465 err = PTR_ERR(inode);
6466 goto out_unlock;
6467 }
6468 drop_inode_on_err = 1;
6469 /*
6470 * If the active LSM wants to access the inode during
6471 * d_instantiate it needs these. Smack checks to see
6472 * if the filesystem supports xattrs by looking at the
6473 * ops vector.
6474 */
6475 inode->i_fop = &btrfs_file_operations;
6476 inode->i_op = &btrfs_file_inode_operations;
6477 inode->i_mapping->a_ops = &btrfs_aops;
6478
6479 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6480 if (err)
6481 goto out_unlock_inode;
6482
6483 err = btrfs_update_inode(trans, root, inode);
6484 if (err)
6485 goto out_unlock_inode;
6486
6487 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6488 if (err)
6489 goto out_unlock_inode;
6490
6491 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6492 unlock_new_inode(inode);
6493 d_instantiate(dentry, inode);
6494
6495 out_unlock:
6496 btrfs_end_transaction(trans, root);
6497 if (err && drop_inode_on_err) {
6498 inode_dec_link_count(inode);
6499 iput(inode);
6500 }
6501 btrfs_balance_delayed_items(root);
6502 btrfs_btree_balance_dirty(root);
6503 return err;
6504
6505 out_unlock_inode:
6506 unlock_new_inode(inode);
6507 goto out_unlock;
6508
6509 }
6510
6511 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6512 struct dentry *dentry)
6513 {
6514 struct btrfs_trans_handle *trans = NULL;
6515 struct btrfs_root *root = BTRFS_I(dir)->root;
6516 struct inode *inode = d_inode(old_dentry);
6517 u64 index;
6518 int err;
6519 int drop_inode = 0;
6520
6521 /* do not allow sys_link's with other subvols of the same device */
6522 if (root->objectid != BTRFS_I(inode)->root->objectid)
6523 return -EXDEV;
6524
6525 if (inode->i_nlink >= BTRFS_LINK_MAX)
6526 return -EMLINK;
6527
6528 err = btrfs_set_inode_index(dir, &index);
6529 if (err)
6530 goto fail;
6531
6532 /*
6533 * 2 items for inode and inode ref
6534 * 2 items for dir items
6535 * 1 item for parent inode
6536 */
6537 trans = btrfs_start_transaction(root, 5);
6538 if (IS_ERR(trans)) {
6539 err = PTR_ERR(trans);
6540 trans = NULL;
6541 goto fail;
6542 }
6543
6544 /* There are several dir indexes for this inode, clear the cache. */
6545 BTRFS_I(inode)->dir_index = 0ULL;
6546 inc_nlink(inode);
6547 inode_inc_iversion(inode);
6548 inode->i_ctime = current_fs_time(inode->i_sb);
6549 ihold(inode);
6550 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6551
6552 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6553
6554 if (err) {
6555 drop_inode = 1;
6556 } else {
6557 struct dentry *parent = dentry->d_parent;
6558 err = btrfs_update_inode(trans, root, inode);
6559 if (err)
6560 goto fail;
6561 if (inode->i_nlink == 1) {
6562 /*
6563 * If new hard link count is 1, it's a file created
6564 * with open(2) O_TMPFILE flag.
6565 */
6566 err = btrfs_orphan_del(trans, inode);
6567 if (err)
6568 goto fail;
6569 }
6570 d_instantiate(dentry, inode);
6571 btrfs_log_new_name(trans, inode, NULL, parent);
6572 }
6573
6574 btrfs_balance_delayed_items(root);
6575 fail:
6576 if (trans)
6577 btrfs_end_transaction(trans, root);
6578 if (drop_inode) {
6579 inode_dec_link_count(inode);
6580 iput(inode);
6581 }
6582 btrfs_btree_balance_dirty(root);
6583 return err;
6584 }
6585
6586 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6587 {
6588 struct inode *inode = NULL;
6589 struct btrfs_trans_handle *trans;
6590 struct btrfs_root *root = BTRFS_I(dir)->root;
6591 int err = 0;
6592 int drop_on_err = 0;
6593 u64 objectid = 0;
6594 u64 index = 0;
6595
6596 /*
6597 * 2 items for inode and ref
6598 * 2 items for dir items
6599 * 1 for xattr if selinux is on
6600 */
6601 trans = btrfs_start_transaction(root, 5);
6602 if (IS_ERR(trans))
6603 return PTR_ERR(trans);
6604
6605 err = btrfs_find_free_ino(root, &objectid);
6606 if (err)
6607 goto out_fail;
6608
6609 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6610 dentry->d_name.len, btrfs_ino(dir), objectid,
6611 S_IFDIR | mode, &index);
6612 if (IS_ERR(inode)) {
6613 err = PTR_ERR(inode);
6614 goto out_fail;
6615 }
6616
6617 drop_on_err = 1;
6618 /* these must be set before we unlock the inode */
6619 inode->i_op = &btrfs_dir_inode_operations;
6620 inode->i_fop = &btrfs_dir_file_operations;
6621
6622 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6623 if (err)
6624 goto out_fail_inode;
6625
6626 btrfs_i_size_write(inode, 0);
6627 err = btrfs_update_inode(trans, root, inode);
6628 if (err)
6629 goto out_fail_inode;
6630
6631 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6632 dentry->d_name.len, 0, index);
6633 if (err)
6634 goto out_fail_inode;
6635
6636 d_instantiate(dentry, inode);
6637 /*
6638 * mkdir is special. We're unlocking after we call d_instantiate
6639 * to avoid a race with nfsd calling d_instantiate.
6640 */
6641 unlock_new_inode(inode);
6642 drop_on_err = 0;
6643
6644 out_fail:
6645 btrfs_end_transaction(trans, root);
6646 if (drop_on_err) {
6647 inode_dec_link_count(inode);
6648 iput(inode);
6649 }
6650 btrfs_balance_delayed_items(root);
6651 btrfs_btree_balance_dirty(root);
6652 return err;
6653
6654 out_fail_inode:
6655 unlock_new_inode(inode);
6656 goto out_fail;
6657 }
6658
6659 /* Find next extent map of a given extent map, caller needs to ensure locks */
6660 static struct extent_map *next_extent_map(struct extent_map *em)
6661 {
6662 struct rb_node *next;
6663
6664 next = rb_next(&em->rb_node);
6665 if (!next)
6666 return NULL;
6667 return container_of(next, struct extent_map, rb_node);
6668 }
6669
6670 static struct extent_map *prev_extent_map(struct extent_map *em)
6671 {
6672 struct rb_node *prev;
6673
6674 prev = rb_prev(&em->rb_node);
6675 if (!prev)
6676 return NULL;
6677 return container_of(prev, struct extent_map, rb_node);
6678 }
6679
6680 /* helper for btfs_get_extent. Given an existing extent in the tree,
6681 * the existing extent is the nearest extent to map_start,
6682 * and an extent that you want to insert, deal with overlap and insert
6683 * the best fitted new extent into the tree.
6684 */
6685 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6686 struct extent_map *existing,
6687 struct extent_map *em,
6688 u64 map_start)
6689 {
6690 struct extent_map *prev;
6691 struct extent_map *next;
6692 u64 start;
6693 u64 end;
6694 u64 start_diff;
6695
6696 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6697
6698 if (existing->start > map_start) {
6699 next = existing;
6700 prev = prev_extent_map(next);
6701 } else {
6702 prev = existing;
6703 next = next_extent_map(prev);
6704 }
6705
6706 start = prev ? extent_map_end(prev) : em->start;
6707 start = max_t(u64, start, em->start);
6708 end = next ? next->start : extent_map_end(em);
6709 end = min_t(u64, end, extent_map_end(em));
6710 start_diff = start - em->start;
6711 em->start = start;
6712 em->len = end - start;
6713 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6714 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6715 em->block_start += start_diff;
6716 em->block_len -= start_diff;
6717 }
6718 return add_extent_mapping(em_tree, em, 0);
6719 }
6720
6721 static noinline int uncompress_inline(struct btrfs_path *path,
6722 struct page *page,
6723 size_t pg_offset, u64 extent_offset,
6724 struct btrfs_file_extent_item *item)
6725 {
6726 int ret;
6727 struct extent_buffer *leaf = path->nodes[0];
6728 char *tmp;
6729 size_t max_size;
6730 unsigned long inline_size;
6731 unsigned long ptr;
6732 int compress_type;
6733
6734 WARN_ON(pg_offset != 0);
6735 compress_type = btrfs_file_extent_compression(leaf, item);
6736 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6737 inline_size = btrfs_file_extent_inline_item_len(leaf,
6738 btrfs_item_nr(path->slots[0]));
6739 tmp = kmalloc(inline_size, GFP_NOFS);
6740 if (!tmp)
6741 return -ENOMEM;
6742 ptr = btrfs_file_extent_inline_start(item);
6743
6744 read_extent_buffer(leaf, tmp, ptr, inline_size);
6745
6746 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6747 ret = btrfs_decompress(compress_type, tmp, page,
6748 extent_offset, inline_size, max_size);
6749 kfree(tmp);
6750 return ret;
6751 }
6752
6753 /*
6754 * a bit scary, this does extent mapping from logical file offset to the disk.
6755 * the ugly parts come from merging extents from the disk with the in-ram
6756 * representation. This gets more complex because of the data=ordered code,
6757 * where the in-ram extents might be locked pending data=ordered completion.
6758 *
6759 * This also copies inline extents directly into the page.
6760 */
6761
6762 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6763 size_t pg_offset, u64 start, u64 len,
6764 int create)
6765 {
6766 int ret;
6767 int err = 0;
6768 u64 extent_start = 0;
6769 u64 extent_end = 0;
6770 u64 objectid = btrfs_ino(inode);
6771 u32 found_type;
6772 struct btrfs_path *path = NULL;
6773 struct btrfs_root *root = BTRFS_I(inode)->root;
6774 struct btrfs_file_extent_item *item;
6775 struct extent_buffer *leaf;
6776 struct btrfs_key found_key;
6777 struct extent_map *em = NULL;
6778 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6779 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6780 struct btrfs_trans_handle *trans = NULL;
6781 const bool new_inline = !page || create;
6782
6783 again:
6784 read_lock(&em_tree->lock);
6785 em = lookup_extent_mapping(em_tree, start, len);
6786 if (em)
6787 em->bdev = root->fs_info->fs_devices->latest_bdev;
6788 read_unlock(&em_tree->lock);
6789
6790 if (em) {
6791 if (em->start > start || em->start + em->len <= start)
6792 free_extent_map(em);
6793 else if (em->block_start == EXTENT_MAP_INLINE && page)
6794 free_extent_map(em);
6795 else
6796 goto out;
6797 }
6798 em = alloc_extent_map();
6799 if (!em) {
6800 err = -ENOMEM;
6801 goto out;
6802 }
6803 em->bdev = root->fs_info->fs_devices->latest_bdev;
6804 em->start = EXTENT_MAP_HOLE;
6805 em->orig_start = EXTENT_MAP_HOLE;
6806 em->len = (u64)-1;
6807 em->block_len = (u64)-1;
6808
6809 if (!path) {
6810 path = btrfs_alloc_path();
6811 if (!path) {
6812 err = -ENOMEM;
6813 goto out;
6814 }
6815 /*
6816 * Chances are we'll be called again, so go ahead and do
6817 * readahead
6818 */
6819 path->reada = READA_FORWARD;
6820 }
6821
6822 ret = btrfs_lookup_file_extent(trans, root, path,
6823 objectid, start, trans != NULL);
6824 if (ret < 0) {
6825 err = ret;
6826 goto out;
6827 }
6828
6829 if (ret != 0) {
6830 if (path->slots[0] == 0)
6831 goto not_found;
6832 path->slots[0]--;
6833 }
6834
6835 leaf = path->nodes[0];
6836 item = btrfs_item_ptr(leaf, path->slots[0],
6837 struct btrfs_file_extent_item);
6838 /* are we inside the extent that was found? */
6839 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6840 found_type = found_key.type;
6841 if (found_key.objectid != objectid ||
6842 found_type != BTRFS_EXTENT_DATA_KEY) {
6843 /*
6844 * If we backup past the first extent we want to move forward
6845 * and see if there is an extent in front of us, otherwise we'll
6846 * say there is a hole for our whole search range which can
6847 * cause problems.
6848 */
6849 extent_end = start;
6850 goto next;
6851 }
6852
6853 found_type = btrfs_file_extent_type(leaf, item);
6854 extent_start = found_key.offset;
6855 if (found_type == BTRFS_FILE_EXTENT_REG ||
6856 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6857 extent_end = extent_start +
6858 btrfs_file_extent_num_bytes(leaf, item);
6859 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6860 size_t size;
6861 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6862 extent_end = ALIGN(extent_start + size, root->sectorsize);
6863 }
6864 next:
6865 if (start >= extent_end) {
6866 path->slots[0]++;
6867 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6868 ret = btrfs_next_leaf(root, path);
6869 if (ret < 0) {
6870 err = ret;
6871 goto out;
6872 }
6873 if (ret > 0)
6874 goto not_found;
6875 leaf = path->nodes[0];
6876 }
6877 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6878 if (found_key.objectid != objectid ||
6879 found_key.type != BTRFS_EXTENT_DATA_KEY)
6880 goto not_found;
6881 if (start + len <= found_key.offset)
6882 goto not_found;
6883 if (start > found_key.offset)
6884 goto next;
6885 em->start = start;
6886 em->orig_start = start;
6887 em->len = found_key.offset - start;
6888 goto not_found_em;
6889 }
6890
6891 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6892
6893 if (found_type == BTRFS_FILE_EXTENT_REG ||
6894 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6895 goto insert;
6896 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6897 unsigned long ptr;
6898 char *map;
6899 size_t size;
6900 size_t extent_offset;
6901 size_t copy_size;
6902
6903 if (new_inline)
6904 goto out;
6905
6906 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6907 extent_offset = page_offset(page) + pg_offset - extent_start;
6908 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6909 size - extent_offset);
6910 em->start = extent_start + extent_offset;
6911 em->len = ALIGN(copy_size, root->sectorsize);
6912 em->orig_block_len = em->len;
6913 em->orig_start = em->start;
6914 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6915 if (create == 0 && !PageUptodate(page)) {
6916 if (btrfs_file_extent_compression(leaf, item) !=
6917 BTRFS_COMPRESS_NONE) {
6918 ret = uncompress_inline(path, page, pg_offset,
6919 extent_offset, item);
6920 if (ret) {
6921 err = ret;
6922 goto out;
6923 }
6924 } else {
6925 map = kmap(page);
6926 read_extent_buffer(leaf, map + pg_offset, ptr,
6927 copy_size);
6928 if (pg_offset + copy_size < PAGE_SIZE) {
6929 memset(map + pg_offset + copy_size, 0,
6930 PAGE_SIZE - pg_offset -
6931 copy_size);
6932 }
6933 kunmap(page);
6934 }
6935 flush_dcache_page(page);
6936 } else if (create && PageUptodate(page)) {
6937 BUG();
6938 if (!trans) {
6939 kunmap(page);
6940 free_extent_map(em);
6941 em = NULL;
6942
6943 btrfs_release_path(path);
6944 trans = btrfs_join_transaction(root);
6945
6946 if (IS_ERR(trans))
6947 return ERR_CAST(trans);
6948 goto again;
6949 }
6950 map = kmap(page);
6951 write_extent_buffer(leaf, map + pg_offset, ptr,
6952 copy_size);
6953 kunmap(page);
6954 btrfs_mark_buffer_dirty(leaf);
6955 }
6956 set_extent_uptodate(io_tree, em->start,
6957 extent_map_end(em) - 1, NULL, GFP_NOFS);
6958 goto insert;
6959 }
6960 not_found:
6961 em->start = start;
6962 em->orig_start = start;
6963 em->len = len;
6964 not_found_em:
6965 em->block_start = EXTENT_MAP_HOLE;
6966 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6967 insert:
6968 btrfs_release_path(path);
6969 if (em->start > start || extent_map_end(em) <= start) {
6970 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6971 em->start, em->len, start, len);
6972 err = -EIO;
6973 goto out;
6974 }
6975
6976 err = 0;
6977 write_lock(&em_tree->lock);
6978 ret = add_extent_mapping(em_tree, em, 0);
6979 /* it is possible that someone inserted the extent into the tree
6980 * while we had the lock dropped. It is also possible that
6981 * an overlapping map exists in the tree
6982 */
6983 if (ret == -EEXIST) {
6984 struct extent_map *existing;
6985
6986 ret = 0;
6987
6988 existing = search_extent_mapping(em_tree, start, len);
6989 /*
6990 * existing will always be non-NULL, since there must be
6991 * extent causing the -EEXIST.
6992 */
6993 if (existing->start == em->start &&
6994 extent_map_end(existing) == extent_map_end(em) &&
6995 em->block_start == existing->block_start) {
6996 /*
6997 * these two extents are the same, it happens
6998 * with inlines especially
6999 */
7000 free_extent_map(em);
7001 em = existing;
7002 err = 0;
7003
7004 } else if (start >= extent_map_end(existing) ||
7005 start <= existing->start) {
7006 /*
7007 * The existing extent map is the one nearest to
7008 * the [start, start + len) range which overlaps
7009 */
7010 err = merge_extent_mapping(em_tree, existing,
7011 em, start);
7012 free_extent_map(existing);
7013 if (err) {
7014 free_extent_map(em);
7015 em = NULL;
7016 }
7017 } else {
7018 free_extent_map(em);
7019 em = existing;
7020 err = 0;
7021 }
7022 }
7023 write_unlock(&em_tree->lock);
7024 out:
7025
7026 trace_btrfs_get_extent(root, em);
7027
7028 btrfs_free_path(path);
7029 if (trans) {
7030 ret = btrfs_end_transaction(trans, root);
7031 if (!err)
7032 err = ret;
7033 }
7034 if (err) {
7035 free_extent_map(em);
7036 return ERR_PTR(err);
7037 }
7038 BUG_ON(!em); /* Error is always set */
7039 return em;
7040 }
7041
7042 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7043 size_t pg_offset, u64 start, u64 len,
7044 int create)
7045 {
7046 struct extent_map *em;
7047 struct extent_map *hole_em = NULL;
7048 u64 range_start = start;
7049 u64 end;
7050 u64 found;
7051 u64 found_end;
7052 int err = 0;
7053
7054 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7055 if (IS_ERR(em))
7056 return em;
7057 if (em) {
7058 /*
7059 * if our em maps to
7060 * - a hole or
7061 * - a pre-alloc extent,
7062 * there might actually be delalloc bytes behind it.
7063 */
7064 if (em->block_start != EXTENT_MAP_HOLE &&
7065 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7066 return em;
7067 else
7068 hole_em = em;
7069 }
7070
7071 /* check to see if we've wrapped (len == -1 or similar) */
7072 end = start + len;
7073 if (end < start)
7074 end = (u64)-1;
7075 else
7076 end -= 1;
7077
7078 em = NULL;
7079
7080 /* ok, we didn't find anything, lets look for delalloc */
7081 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7082 end, len, EXTENT_DELALLOC, 1);
7083 found_end = range_start + found;
7084 if (found_end < range_start)
7085 found_end = (u64)-1;
7086
7087 /*
7088 * we didn't find anything useful, return
7089 * the original results from get_extent()
7090 */
7091 if (range_start > end || found_end <= start) {
7092 em = hole_em;
7093 hole_em = NULL;
7094 goto out;
7095 }
7096
7097 /* adjust the range_start to make sure it doesn't
7098 * go backwards from the start they passed in
7099 */
7100 range_start = max(start, range_start);
7101 found = found_end - range_start;
7102
7103 if (found > 0) {
7104 u64 hole_start = start;
7105 u64 hole_len = len;
7106
7107 em = alloc_extent_map();
7108 if (!em) {
7109 err = -ENOMEM;
7110 goto out;
7111 }
7112 /*
7113 * when btrfs_get_extent can't find anything it
7114 * returns one huge hole
7115 *
7116 * make sure what it found really fits our range, and
7117 * adjust to make sure it is based on the start from
7118 * the caller
7119 */
7120 if (hole_em) {
7121 u64 calc_end = extent_map_end(hole_em);
7122
7123 if (calc_end <= start || (hole_em->start > end)) {
7124 free_extent_map(hole_em);
7125 hole_em = NULL;
7126 } else {
7127 hole_start = max(hole_em->start, start);
7128 hole_len = calc_end - hole_start;
7129 }
7130 }
7131 em->bdev = NULL;
7132 if (hole_em && range_start > hole_start) {
7133 /* our hole starts before our delalloc, so we
7134 * have to return just the parts of the hole
7135 * that go until the delalloc starts
7136 */
7137 em->len = min(hole_len,
7138 range_start - hole_start);
7139 em->start = hole_start;
7140 em->orig_start = hole_start;
7141 /*
7142 * don't adjust block start at all,
7143 * it is fixed at EXTENT_MAP_HOLE
7144 */
7145 em->block_start = hole_em->block_start;
7146 em->block_len = hole_len;
7147 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7148 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7149 } else {
7150 em->start = range_start;
7151 em->len = found;
7152 em->orig_start = range_start;
7153 em->block_start = EXTENT_MAP_DELALLOC;
7154 em->block_len = found;
7155 }
7156 } else if (hole_em) {
7157 return hole_em;
7158 }
7159 out:
7160
7161 free_extent_map(hole_em);
7162 if (err) {
7163 free_extent_map(em);
7164 return ERR_PTR(err);
7165 }
7166 return em;
7167 }
7168
7169 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7170 const u64 start,
7171 const u64 len,
7172 const u64 orig_start,
7173 const u64 block_start,
7174 const u64 block_len,
7175 const u64 orig_block_len,
7176 const u64 ram_bytes,
7177 const int type)
7178 {
7179 struct extent_map *em = NULL;
7180 int ret;
7181
7182 down_read(&BTRFS_I(inode)->dio_sem);
7183 if (type != BTRFS_ORDERED_NOCOW) {
7184 em = create_pinned_em(inode, start, len, orig_start,
7185 block_start, block_len, orig_block_len,
7186 ram_bytes, type);
7187 if (IS_ERR(em))
7188 goto out;
7189 }
7190 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7191 len, block_len, type);
7192 if (ret) {
7193 if (em) {
7194 free_extent_map(em);
7195 btrfs_drop_extent_cache(inode, start,
7196 start + len - 1, 0);
7197 }
7198 em = ERR_PTR(ret);
7199 }
7200 out:
7201 up_read(&BTRFS_I(inode)->dio_sem);
7202
7203 return em;
7204 }
7205
7206 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7207 u64 start, u64 len)
7208 {
7209 struct btrfs_root *root = BTRFS_I(inode)->root;
7210 struct extent_map *em;
7211 struct btrfs_key ins;
7212 u64 alloc_hint;
7213 int ret;
7214
7215 alloc_hint = get_extent_allocation_hint(inode, start, len);
7216 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7217 alloc_hint, &ins, 1, 1);
7218 if (ret)
7219 return ERR_PTR(ret);
7220
7221 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7222 ins.objectid, ins.offset, ins.offset,
7223 ins.offset, 0);
7224 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
7225 if (IS_ERR(em))
7226 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7227
7228 return em;
7229 }
7230
7231 /*
7232 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7233 * block must be cow'd
7234 */
7235 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7236 u64 *orig_start, u64 *orig_block_len,
7237 u64 *ram_bytes)
7238 {
7239 struct btrfs_trans_handle *trans;
7240 struct btrfs_path *path;
7241 int ret;
7242 struct extent_buffer *leaf;
7243 struct btrfs_root *root = BTRFS_I(inode)->root;
7244 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7245 struct btrfs_file_extent_item *fi;
7246 struct btrfs_key key;
7247 u64 disk_bytenr;
7248 u64 backref_offset;
7249 u64 extent_end;
7250 u64 num_bytes;
7251 int slot;
7252 int found_type;
7253 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7254
7255 path = btrfs_alloc_path();
7256 if (!path)
7257 return -ENOMEM;
7258
7259 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7260 offset, 0);
7261 if (ret < 0)
7262 goto out;
7263
7264 slot = path->slots[0];
7265 if (ret == 1) {
7266 if (slot == 0) {
7267 /* can't find the item, must cow */
7268 ret = 0;
7269 goto out;
7270 }
7271 slot--;
7272 }
7273 ret = 0;
7274 leaf = path->nodes[0];
7275 btrfs_item_key_to_cpu(leaf, &key, slot);
7276 if (key.objectid != btrfs_ino(inode) ||
7277 key.type != BTRFS_EXTENT_DATA_KEY) {
7278 /* not our file or wrong item type, must cow */
7279 goto out;
7280 }
7281
7282 if (key.offset > offset) {
7283 /* Wrong offset, must cow */
7284 goto out;
7285 }
7286
7287 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7288 found_type = btrfs_file_extent_type(leaf, fi);
7289 if (found_type != BTRFS_FILE_EXTENT_REG &&
7290 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7291 /* not a regular extent, must cow */
7292 goto out;
7293 }
7294
7295 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7296 goto out;
7297
7298 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7299 if (extent_end <= offset)
7300 goto out;
7301
7302 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7303 if (disk_bytenr == 0)
7304 goto out;
7305
7306 if (btrfs_file_extent_compression(leaf, fi) ||
7307 btrfs_file_extent_encryption(leaf, fi) ||
7308 btrfs_file_extent_other_encoding(leaf, fi))
7309 goto out;
7310
7311 backref_offset = btrfs_file_extent_offset(leaf, fi);
7312
7313 if (orig_start) {
7314 *orig_start = key.offset - backref_offset;
7315 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7316 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7317 }
7318
7319 if (btrfs_extent_readonly(root, disk_bytenr))
7320 goto out;
7321
7322 num_bytes = min(offset + *len, extent_end) - offset;
7323 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7324 u64 range_end;
7325
7326 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7327 ret = test_range_bit(io_tree, offset, range_end,
7328 EXTENT_DELALLOC, 0, NULL);
7329 if (ret) {
7330 ret = -EAGAIN;
7331 goto out;
7332 }
7333 }
7334
7335 btrfs_release_path(path);
7336
7337 /*
7338 * look for other files referencing this extent, if we
7339 * find any we must cow
7340 */
7341 trans = btrfs_join_transaction(root);
7342 if (IS_ERR(trans)) {
7343 ret = 0;
7344 goto out;
7345 }
7346
7347 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7348 key.offset - backref_offset, disk_bytenr);
7349 btrfs_end_transaction(trans, root);
7350 if (ret) {
7351 ret = 0;
7352 goto out;
7353 }
7354
7355 /*
7356 * adjust disk_bytenr and num_bytes to cover just the bytes
7357 * in this extent we are about to write. If there
7358 * are any csums in that range we have to cow in order
7359 * to keep the csums correct
7360 */
7361 disk_bytenr += backref_offset;
7362 disk_bytenr += offset - key.offset;
7363 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7364 goto out;
7365 /*
7366 * all of the above have passed, it is safe to overwrite this extent
7367 * without cow
7368 */
7369 *len = num_bytes;
7370 ret = 1;
7371 out:
7372 btrfs_free_path(path);
7373 return ret;
7374 }
7375
7376 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7377 {
7378 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7379 int found = false;
7380 void **pagep = NULL;
7381 struct page *page = NULL;
7382 int start_idx;
7383 int end_idx;
7384
7385 start_idx = start >> PAGE_SHIFT;
7386
7387 /*
7388 * end is the last byte in the last page. end == start is legal
7389 */
7390 end_idx = end >> PAGE_SHIFT;
7391
7392 rcu_read_lock();
7393
7394 /* Most of the code in this while loop is lifted from
7395 * find_get_page. It's been modified to begin searching from a
7396 * page and return just the first page found in that range. If the
7397 * found idx is less than or equal to the end idx then we know that
7398 * a page exists. If no pages are found or if those pages are
7399 * outside of the range then we're fine (yay!) */
7400 while (page == NULL &&
7401 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7402 page = radix_tree_deref_slot(pagep);
7403 if (unlikely(!page))
7404 break;
7405
7406 if (radix_tree_exception(page)) {
7407 if (radix_tree_deref_retry(page)) {
7408 page = NULL;
7409 continue;
7410 }
7411 /*
7412 * Otherwise, shmem/tmpfs must be storing a swap entry
7413 * here as an exceptional entry: so return it without
7414 * attempting to raise page count.
7415 */
7416 page = NULL;
7417 break; /* TODO: Is this relevant for this use case? */
7418 }
7419
7420 if (!page_cache_get_speculative(page)) {
7421 page = NULL;
7422 continue;
7423 }
7424
7425 /*
7426 * Has the page moved?
7427 * This is part of the lockless pagecache protocol. See
7428 * include/linux/pagemap.h for details.
7429 */
7430 if (unlikely(page != *pagep)) {
7431 put_page(page);
7432 page = NULL;
7433 }
7434 }
7435
7436 if (page) {
7437 if (page->index <= end_idx)
7438 found = true;
7439 put_page(page);
7440 }
7441
7442 rcu_read_unlock();
7443 return found;
7444 }
7445
7446 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7447 struct extent_state **cached_state, int writing)
7448 {
7449 struct btrfs_ordered_extent *ordered;
7450 int ret = 0;
7451
7452 while (1) {
7453 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7454 cached_state);
7455 /*
7456 * We're concerned with the entire range that we're going to be
7457 * doing DIO to, so we need to make sure there's no ordered
7458 * extents in this range.
7459 */
7460 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7461 lockend - lockstart + 1);
7462
7463 /*
7464 * We need to make sure there are no buffered pages in this
7465 * range either, we could have raced between the invalidate in
7466 * generic_file_direct_write and locking the extent. The
7467 * invalidate needs to happen so that reads after a write do not
7468 * get stale data.
7469 */
7470 if (!ordered &&
7471 (!writing ||
7472 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7473 break;
7474
7475 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7476 cached_state, GFP_NOFS);
7477
7478 if (ordered) {
7479 /*
7480 * If we are doing a DIO read and the ordered extent we
7481 * found is for a buffered write, we can not wait for it
7482 * to complete and retry, because if we do so we can
7483 * deadlock with concurrent buffered writes on page
7484 * locks. This happens only if our DIO read covers more
7485 * than one extent map, if at this point has already
7486 * created an ordered extent for a previous extent map
7487 * and locked its range in the inode's io tree, and a
7488 * concurrent write against that previous extent map's
7489 * range and this range started (we unlock the ranges
7490 * in the io tree only when the bios complete and
7491 * buffered writes always lock pages before attempting
7492 * to lock range in the io tree).
7493 */
7494 if (writing ||
7495 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7496 btrfs_start_ordered_extent(inode, ordered, 1);
7497 else
7498 ret = -ENOTBLK;
7499 btrfs_put_ordered_extent(ordered);
7500 } else {
7501 /*
7502 * We could trigger writeback for this range (and wait
7503 * for it to complete) and then invalidate the pages for
7504 * this range (through invalidate_inode_pages2_range()),
7505 * but that can lead us to a deadlock with a concurrent
7506 * call to readpages() (a buffered read or a defrag call
7507 * triggered a readahead) on a page lock due to an
7508 * ordered dio extent we created before but did not have
7509 * yet a corresponding bio submitted (whence it can not
7510 * complete), which makes readpages() wait for that
7511 * ordered extent to complete while holding a lock on
7512 * that page.
7513 */
7514 ret = -ENOTBLK;
7515 }
7516
7517 if (ret)
7518 break;
7519
7520 cond_resched();
7521 }
7522
7523 return ret;
7524 }
7525
7526 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7527 u64 len, u64 orig_start,
7528 u64 block_start, u64 block_len,
7529 u64 orig_block_len, u64 ram_bytes,
7530 int type)
7531 {
7532 struct extent_map_tree *em_tree;
7533 struct extent_map *em;
7534 struct btrfs_root *root = BTRFS_I(inode)->root;
7535 int ret;
7536
7537 em_tree = &BTRFS_I(inode)->extent_tree;
7538 em = alloc_extent_map();
7539 if (!em)
7540 return ERR_PTR(-ENOMEM);
7541
7542 em->start = start;
7543 em->orig_start = orig_start;
7544 em->mod_start = start;
7545 em->mod_len = len;
7546 em->len = len;
7547 em->block_len = block_len;
7548 em->block_start = block_start;
7549 em->bdev = root->fs_info->fs_devices->latest_bdev;
7550 em->orig_block_len = orig_block_len;
7551 em->ram_bytes = ram_bytes;
7552 em->generation = -1;
7553 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7554 if (type == BTRFS_ORDERED_PREALLOC)
7555 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7556
7557 do {
7558 btrfs_drop_extent_cache(inode, em->start,
7559 em->start + em->len - 1, 0);
7560 write_lock(&em_tree->lock);
7561 ret = add_extent_mapping(em_tree, em, 1);
7562 write_unlock(&em_tree->lock);
7563 } while (ret == -EEXIST);
7564
7565 if (ret) {
7566 free_extent_map(em);
7567 return ERR_PTR(ret);
7568 }
7569
7570 return em;
7571 }
7572
7573 static void adjust_dio_outstanding_extents(struct inode *inode,
7574 struct btrfs_dio_data *dio_data,
7575 const u64 len)
7576 {
7577 unsigned num_extents;
7578
7579 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7580 BTRFS_MAX_EXTENT_SIZE);
7581 /*
7582 * If we have an outstanding_extents count still set then we're
7583 * within our reservation, otherwise we need to adjust our inode
7584 * counter appropriately.
7585 */
7586 if (dio_data->outstanding_extents) {
7587 dio_data->outstanding_extents -= num_extents;
7588 } else {
7589 spin_lock(&BTRFS_I(inode)->lock);
7590 BTRFS_I(inode)->outstanding_extents += num_extents;
7591 spin_unlock(&BTRFS_I(inode)->lock);
7592 }
7593 }
7594
7595 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7596 struct buffer_head *bh_result, int create)
7597 {
7598 struct extent_map *em;
7599 struct btrfs_root *root = BTRFS_I(inode)->root;
7600 struct extent_state *cached_state = NULL;
7601 struct btrfs_dio_data *dio_data = NULL;
7602 u64 start = iblock << inode->i_blkbits;
7603 u64 lockstart, lockend;
7604 u64 len = bh_result->b_size;
7605 int unlock_bits = EXTENT_LOCKED;
7606 int ret = 0;
7607
7608 if (create)
7609 unlock_bits |= EXTENT_DIRTY;
7610 else
7611 len = min_t(u64, len, root->sectorsize);
7612
7613 lockstart = start;
7614 lockend = start + len - 1;
7615
7616 if (current->journal_info) {
7617 /*
7618 * Need to pull our outstanding extents and set journal_info to NULL so
7619 * that anything that needs to check if there's a transaction doesn't get
7620 * confused.
7621 */
7622 dio_data = current->journal_info;
7623 current->journal_info = NULL;
7624 }
7625
7626 /*
7627 * If this errors out it's because we couldn't invalidate pagecache for
7628 * this range and we need to fallback to buffered.
7629 */
7630 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7631 create)) {
7632 ret = -ENOTBLK;
7633 goto err;
7634 }
7635
7636 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7637 if (IS_ERR(em)) {
7638 ret = PTR_ERR(em);
7639 goto unlock_err;
7640 }
7641
7642 /*
7643 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7644 * io. INLINE is special, and we could probably kludge it in here, but
7645 * it's still buffered so for safety lets just fall back to the generic
7646 * buffered path.
7647 *
7648 * For COMPRESSED we _have_ to read the entire extent in so we can
7649 * decompress it, so there will be buffering required no matter what we
7650 * do, so go ahead and fallback to buffered.
7651 *
7652 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7653 * to buffered IO. Don't blame me, this is the price we pay for using
7654 * the generic code.
7655 */
7656 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7657 em->block_start == EXTENT_MAP_INLINE) {
7658 free_extent_map(em);
7659 ret = -ENOTBLK;
7660 goto unlock_err;
7661 }
7662
7663 /* Just a good old fashioned hole, return */
7664 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7665 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7666 free_extent_map(em);
7667 goto unlock_err;
7668 }
7669
7670 /*
7671 * We don't allocate a new extent in the following cases
7672 *
7673 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7674 * existing extent.
7675 * 2) The extent is marked as PREALLOC. We're good to go here and can
7676 * just use the extent.
7677 *
7678 */
7679 if (!create) {
7680 len = min(len, em->len - (start - em->start));
7681 lockstart = start + len;
7682 goto unlock;
7683 }
7684
7685 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7686 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7687 em->block_start != EXTENT_MAP_HOLE)) {
7688 int type;
7689 u64 block_start, orig_start, orig_block_len, ram_bytes;
7690
7691 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7692 type = BTRFS_ORDERED_PREALLOC;
7693 else
7694 type = BTRFS_ORDERED_NOCOW;
7695 len = min(len, em->len - (start - em->start));
7696 block_start = em->block_start + (start - em->start);
7697
7698 if (can_nocow_extent(inode, start, &len, &orig_start,
7699 &orig_block_len, &ram_bytes) == 1 &&
7700 btrfs_inc_nocow_writers(root->fs_info, block_start)) {
7701 struct extent_map *em2;
7702
7703 em2 = btrfs_create_dio_extent(inode, start, len,
7704 orig_start, block_start,
7705 len, orig_block_len,
7706 ram_bytes, type);
7707 btrfs_dec_nocow_writers(root->fs_info, block_start);
7708 if (type == BTRFS_ORDERED_PREALLOC) {
7709 free_extent_map(em);
7710 em = em2;
7711 }
7712 if (em2 && IS_ERR(em2)) {
7713 ret = PTR_ERR(em2);
7714 goto unlock_err;
7715 }
7716 goto unlock;
7717 }
7718 }
7719
7720 /*
7721 * this will cow the extent, reset the len in case we changed
7722 * it above
7723 */
7724 len = bh_result->b_size;
7725 free_extent_map(em);
7726 em = btrfs_new_extent_direct(inode, start, len);
7727 if (IS_ERR(em)) {
7728 ret = PTR_ERR(em);
7729 goto unlock_err;
7730 }
7731 len = min(len, em->len - (start - em->start));
7732 unlock:
7733 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7734 inode->i_blkbits;
7735 bh_result->b_size = len;
7736 bh_result->b_bdev = em->bdev;
7737 set_buffer_mapped(bh_result);
7738 if (create) {
7739 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7740 set_buffer_new(bh_result);
7741
7742 /*
7743 * Need to update the i_size under the extent lock so buffered
7744 * readers will get the updated i_size when we unlock.
7745 */
7746 if (start + len > i_size_read(inode))
7747 i_size_write(inode, start + len);
7748
7749 adjust_dio_outstanding_extents(inode, dio_data, len);
7750 btrfs_free_reserved_data_space(inode, start, len);
7751 WARN_ON(dio_data->reserve < len);
7752 dio_data->reserve -= len;
7753 dio_data->unsubmitted_oe_range_end = start + len;
7754 current->journal_info = dio_data;
7755 }
7756
7757 /*
7758 * In the case of write we need to clear and unlock the entire range,
7759 * in the case of read we need to unlock only the end area that we
7760 * aren't using if there is any left over space.
7761 */
7762 if (lockstart < lockend) {
7763 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7764 lockend, unlock_bits, 1, 0,
7765 &cached_state, GFP_NOFS);
7766 } else {
7767 free_extent_state(cached_state);
7768 }
7769
7770 free_extent_map(em);
7771
7772 return 0;
7773
7774 unlock_err:
7775 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7776 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7777 err:
7778 if (dio_data)
7779 current->journal_info = dio_data;
7780 /*
7781 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7782 * write less data then expected, so that we don't underflow our inode's
7783 * outstanding extents counter.
7784 */
7785 if (create && dio_data)
7786 adjust_dio_outstanding_extents(inode, dio_data, len);
7787
7788 return ret;
7789 }
7790
7791 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7792 int mirror_num)
7793 {
7794 struct btrfs_root *root = BTRFS_I(inode)->root;
7795 int ret;
7796
7797 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7798
7799 bio_get(bio);
7800
7801 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7802 BTRFS_WQ_ENDIO_DIO_REPAIR);
7803 if (ret)
7804 goto err;
7805
7806 ret = btrfs_map_bio(root, bio, mirror_num, 0);
7807 err:
7808 bio_put(bio);
7809 return ret;
7810 }
7811
7812 static int btrfs_check_dio_repairable(struct inode *inode,
7813 struct bio *failed_bio,
7814 struct io_failure_record *failrec,
7815 int failed_mirror)
7816 {
7817 int num_copies;
7818
7819 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7820 failrec->logical, failrec->len);
7821 if (num_copies == 1) {
7822 /*
7823 * we only have a single copy of the data, so don't bother with
7824 * all the retry and error correction code that follows. no
7825 * matter what the error is, it is very likely to persist.
7826 */
7827 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7828 num_copies, failrec->this_mirror, failed_mirror);
7829 return 0;
7830 }
7831
7832 failrec->failed_mirror = failed_mirror;
7833 failrec->this_mirror++;
7834 if (failrec->this_mirror == failed_mirror)
7835 failrec->this_mirror++;
7836
7837 if (failrec->this_mirror > num_copies) {
7838 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7839 num_copies, failrec->this_mirror, failed_mirror);
7840 return 0;
7841 }
7842
7843 return 1;
7844 }
7845
7846 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7847 struct page *page, unsigned int pgoff,
7848 u64 start, u64 end, int failed_mirror,
7849 bio_end_io_t *repair_endio, void *repair_arg)
7850 {
7851 struct io_failure_record *failrec;
7852 struct bio *bio;
7853 int isector;
7854 int read_mode;
7855 int ret;
7856
7857 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7858
7859 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7860 if (ret)
7861 return ret;
7862
7863 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7864 failed_mirror);
7865 if (!ret) {
7866 free_io_failure(inode, failrec);
7867 return -EIO;
7868 }
7869
7870 if ((failed_bio->bi_vcnt > 1)
7871 || (failed_bio->bi_io_vec->bv_len
7872 > BTRFS_I(inode)->root->sectorsize))
7873 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7874 else
7875 read_mode = READ_SYNC;
7876
7877 isector = start - btrfs_io_bio(failed_bio)->logical;
7878 isector >>= inode->i_sb->s_blocksize_bits;
7879 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7880 pgoff, isector, repair_endio, repair_arg);
7881 if (!bio) {
7882 free_io_failure(inode, failrec);
7883 return -EIO;
7884 }
7885 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
7886
7887 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7888 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7889 read_mode, failrec->this_mirror, failrec->in_validation);
7890
7891 ret = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7892 if (ret) {
7893 free_io_failure(inode, failrec);
7894 bio_put(bio);
7895 }
7896
7897 return ret;
7898 }
7899
7900 struct btrfs_retry_complete {
7901 struct completion done;
7902 struct inode *inode;
7903 u64 start;
7904 int uptodate;
7905 };
7906
7907 static void btrfs_retry_endio_nocsum(struct bio *bio)
7908 {
7909 struct btrfs_retry_complete *done = bio->bi_private;
7910 struct inode *inode;
7911 struct bio_vec *bvec;
7912 int i;
7913
7914 if (bio->bi_error)
7915 goto end;
7916
7917 ASSERT(bio->bi_vcnt == 1);
7918 inode = bio->bi_io_vec->bv_page->mapping->host;
7919 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7920
7921 done->uptodate = 1;
7922 bio_for_each_segment_all(bvec, bio, i)
7923 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7924 end:
7925 complete(&done->done);
7926 bio_put(bio);
7927 }
7928
7929 static int __btrfs_correct_data_nocsum(struct inode *inode,
7930 struct btrfs_io_bio *io_bio)
7931 {
7932 struct btrfs_fs_info *fs_info;
7933 struct bio_vec *bvec;
7934 struct btrfs_retry_complete done;
7935 u64 start;
7936 unsigned int pgoff;
7937 u32 sectorsize;
7938 int nr_sectors;
7939 int i;
7940 int ret;
7941
7942 fs_info = BTRFS_I(inode)->root->fs_info;
7943 sectorsize = BTRFS_I(inode)->root->sectorsize;
7944
7945 start = io_bio->logical;
7946 done.inode = inode;
7947
7948 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7949 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7950 pgoff = bvec->bv_offset;
7951
7952 next_block_or_try_again:
7953 done.uptodate = 0;
7954 done.start = start;
7955 init_completion(&done.done);
7956
7957 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7958 pgoff, start, start + sectorsize - 1,
7959 io_bio->mirror_num,
7960 btrfs_retry_endio_nocsum, &done);
7961 if (ret)
7962 return ret;
7963
7964 wait_for_completion(&done.done);
7965
7966 if (!done.uptodate) {
7967 /* We might have another mirror, so try again */
7968 goto next_block_or_try_again;
7969 }
7970
7971 start += sectorsize;
7972
7973 if (nr_sectors--) {
7974 pgoff += sectorsize;
7975 goto next_block_or_try_again;
7976 }
7977 }
7978
7979 return 0;
7980 }
7981
7982 static void btrfs_retry_endio(struct bio *bio)
7983 {
7984 struct btrfs_retry_complete *done = bio->bi_private;
7985 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7986 struct inode *inode;
7987 struct bio_vec *bvec;
7988 u64 start;
7989 int uptodate;
7990 int ret;
7991 int i;
7992
7993 if (bio->bi_error)
7994 goto end;
7995
7996 uptodate = 1;
7997
7998 start = done->start;
7999
8000 ASSERT(bio->bi_vcnt == 1);
8001 inode = bio->bi_io_vec->bv_page->mapping->host;
8002 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
8003
8004 bio_for_each_segment_all(bvec, bio, i) {
8005 ret = __readpage_endio_check(done->inode, io_bio, i,
8006 bvec->bv_page, bvec->bv_offset,
8007 done->start, bvec->bv_len);
8008 if (!ret)
8009 clean_io_failure(done->inode, done->start,
8010 bvec->bv_page, bvec->bv_offset);
8011 else
8012 uptodate = 0;
8013 }
8014
8015 done->uptodate = uptodate;
8016 end:
8017 complete(&done->done);
8018 bio_put(bio);
8019 }
8020
8021 static int __btrfs_subio_endio_read(struct inode *inode,
8022 struct btrfs_io_bio *io_bio, int err)
8023 {
8024 struct btrfs_fs_info *fs_info;
8025 struct bio_vec *bvec;
8026 struct btrfs_retry_complete done;
8027 u64 start;
8028 u64 offset = 0;
8029 u32 sectorsize;
8030 int nr_sectors;
8031 unsigned int pgoff;
8032 int csum_pos;
8033 int i;
8034 int ret;
8035
8036 fs_info = BTRFS_I(inode)->root->fs_info;
8037 sectorsize = BTRFS_I(inode)->root->sectorsize;
8038
8039 err = 0;
8040 start = io_bio->logical;
8041 done.inode = inode;
8042
8043 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
8044 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8045
8046 pgoff = bvec->bv_offset;
8047 next_block:
8048 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8049 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8050 bvec->bv_page, pgoff, start,
8051 sectorsize);
8052 if (likely(!ret))
8053 goto next;
8054 try_again:
8055 done.uptodate = 0;
8056 done.start = start;
8057 init_completion(&done.done);
8058
8059 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8060 pgoff, start, start + sectorsize - 1,
8061 io_bio->mirror_num,
8062 btrfs_retry_endio, &done);
8063 if (ret) {
8064 err = ret;
8065 goto next;
8066 }
8067
8068 wait_for_completion(&done.done);
8069
8070 if (!done.uptodate) {
8071 /* We might have another mirror, so try again */
8072 goto try_again;
8073 }
8074 next:
8075 offset += sectorsize;
8076 start += sectorsize;
8077
8078 ASSERT(nr_sectors);
8079
8080 if (--nr_sectors) {
8081 pgoff += sectorsize;
8082 goto next_block;
8083 }
8084 }
8085
8086 return err;
8087 }
8088
8089 static int btrfs_subio_endio_read(struct inode *inode,
8090 struct btrfs_io_bio *io_bio, int err)
8091 {
8092 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8093
8094 if (skip_csum) {
8095 if (unlikely(err))
8096 return __btrfs_correct_data_nocsum(inode, io_bio);
8097 else
8098 return 0;
8099 } else {
8100 return __btrfs_subio_endio_read(inode, io_bio, err);
8101 }
8102 }
8103
8104 static void btrfs_endio_direct_read(struct bio *bio)
8105 {
8106 struct btrfs_dio_private *dip = bio->bi_private;
8107 struct inode *inode = dip->inode;
8108 struct bio *dio_bio;
8109 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8110 int err = bio->bi_error;
8111
8112 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8113 err = btrfs_subio_endio_read(inode, io_bio, err);
8114
8115 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8116 dip->logical_offset + dip->bytes - 1);
8117 dio_bio = dip->dio_bio;
8118
8119 kfree(dip);
8120
8121 dio_bio->bi_error = bio->bi_error;
8122 dio_end_io(dio_bio, bio->bi_error);
8123
8124 if (io_bio->end_io)
8125 io_bio->end_io(io_bio, err);
8126 bio_put(bio);
8127 }
8128
8129 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8130 const u64 offset,
8131 const u64 bytes,
8132 const int uptodate)
8133 {
8134 struct btrfs_root *root = BTRFS_I(inode)->root;
8135 struct btrfs_ordered_extent *ordered = NULL;
8136 u64 ordered_offset = offset;
8137 u64 ordered_bytes = bytes;
8138 int ret;
8139
8140 again:
8141 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8142 &ordered_offset,
8143 ordered_bytes,
8144 uptodate);
8145 if (!ret)
8146 goto out_test;
8147
8148 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8149 finish_ordered_fn, NULL, NULL);
8150 btrfs_queue_work(root->fs_info->endio_write_workers,
8151 &ordered->work);
8152 out_test:
8153 /*
8154 * our bio might span multiple ordered extents. If we haven't
8155 * completed the accounting for the whole dio, go back and try again
8156 */
8157 if (ordered_offset < offset + bytes) {
8158 ordered_bytes = offset + bytes - ordered_offset;
8159 ordered = NULL;
8160 goto again;
8161 }
8162 }
8163
8164 static void btrfs_endio_direct_write(struct bio *bio)
8165 {
8166 struct btrfs_dio_private *dip = bio->bi_private;
8167 struct bio *dio_bio = dip->dio_bio;
8168
8169 btrfs_endio_direct_write_update_ordered(dip->inode,
8170 dip->logical_offset,
8171 dip->bytes,
8172 !bio->bi_error);
8173
8174 kfree(dip);
8175
8176 dio_bio->bi_error = bio->bi_error;
8177 dio_end_io(dio_bio, bio->bi_error);
8178 bio_put(bio);
8179 }
8180
8181 static int __btrfs_submit_bio_start_direct_io(struct inode *inode,
8182 struct bio *bio, int mirror_num,
8183 unsigned long bio_flags, u64 offset)
8184 {
8185 int ret;
8186 struct btrfs_root *root = BTRFS_I(inode)->root;
8187 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8188 BUG_ON(ret); /* -ENOMEM */
8189 return 0;
8190 }
8191
8192 static void btrfs_end_dio_bio(struct bio *bio)
8193 {
8194 struct btrfs_dio_private *dip = bio->bi_private;
8195 int err = bio->bi_error;
8196
8197 if (err)
8198 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8199 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8200 btrfs_ino(dip->inode), bio_op(bio), bio->bi_rw,
8201 (unsigned long long)bio->bi_iter.bi_sector,
8202 bio->bi_iter.bi_size, err);
8203
8204 if (dip->subio_endio)
8205 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8206
8207 if (err) {
8208 dip->errors = 1;
8209
8210 /*
8211 * before atomic variable goto zero, we must make sure
8212 * dip->errors is perceived to be set.
8213 */
8214 smp_mb__before_atomic();
8215 }
8216
8217 /* if there are more bios still pending for this dio, just exit */
8218 if (!atomic_dec_and_test(&dip->pending_bios))
8219 goto out;
8220
8221 if (dip->errors) {
8222 bio_io_error(dip->orig_bio);
8223 } else {
8224 dip->dio_bio->bi_error = 0;
8225 bio_endio(dip->orig_bio);
8226 }
8227 out:
8228 bio_put(bio);
8229 }
8230
8231 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8232 u64 first_sector, gfp_t gfp_flags)
8233 {
8234 struct bio *bio;
8235 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8236 if (bio)
8237 bio_associate_current(bio);
8238 return bio;
8239 }
8240
8241 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8242 struct inode *inode,
8243 struct btrfs_dio_private *dip,
8244 struct bio *bio,
8245 u64 file_offset)
8246 {
8247 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8248 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8249 int ret;
8250
8251 /*
8252 * We load all the csum data we need when we submit
8253 * the first bio to reduce the csum tree search and
8254 * contention.
8255 */
8256 if (dip->logical_offset == file_offset) {
8257 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8258 file_offset);
8259 if (ret)
8260 return ret;
8261 }
8262
8263 if (bio == dip->orig_bio)
8264 return 0;
8265
8266 file_offset -= dip->logical_offset;
8267 file_offset >>= inode->i_sb->s_blocksize_bits;
8268 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8269
8270 return 0;
8271 }
8272
8273 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8274 u64 file_offset, int skip_sum,
8275 int async_submit)
8276 {
8277 struct btrfs_dio_private *dip = bio->bi_private;
8278 bool write = bio_op(bio) == REQ_OP_WRITE;
8279 struct btrfs_root *root = BTRFS_I(inode)->root;
8280 int ret;
8281
8282 if (async_submit)
8283 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8284
8285 bio_get(bio);
8286
8287 if (!write) {
8288 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8289 BTRFS_WQ_ENDIO_DATA);
8290 if (ret)
8291 goto err;
8292 }
8293
8294 if (skip_sum)
8295 goto map;
8296
8297 if (write && async_submit) {
8298 ret = btrfs_wq_submit_bio(root->fs_info,
8299 inode, bio, 0, 0, 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, bio, 0, async_submit);
8319 err:
8320 bio_put(bio);
8321 return ret;
8322 }
8323
8324 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip,
8325 int skip_sum)
8326 {
8327 struct inode *inode = dip->inode;
8328 struct btrfs_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, bio_op(orig_bio),
8344 start_sector << 9, &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_set_op_attrs(bio, bio_op(orig_bio), orig_bio->bi_rw);
8365 bio->bi_private = dip;
8366 bio->bi_end_io = btrfs_end_dio_bio;
8367 btrfs_io_bio(bio)->logical = file_offset;
8368 atomic_inc(&dip->pending_bios);
8369
8370 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8371 nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len);
8372 i = 0;
8373 next_block:
8374 if (unlikely(map_length < submit_len + blocksize ||
8375 bio_add_page(bio, bvec->bv_page, blocksize,
8376 bvec->bv_offset + (i * blocksize)) < blocksize)) {
8377 /*
8378 * inc the count before we submit the bio so
8379 * we know the end IO handler won't happen before
8380 * we inc the count. Otherwise, the dip might get freed
8381 * before we're done setting it up
8382 */
8383 atomic_inc(&dip->pending_bios);
8384 ret = __btrfs_submit_dio_bio(bio, inode,
8385 file_offset, skip_sum,
8386 async_submit);
8387 if (ret) {
8388 bio_put(bio);
8389 atomic_dec(&dip->pending_bios);
8390 goto out_err;
8391 }
8392
8393 start_sector += submit_len >> 9;
8394 file_offset += submit_len;
8395
8396 submit_len = 0;
8397
8398 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8399 start_sector, GFP_NOFS);
8400 if (!bio)
8401 goto out_err;
8402 bio_set_op_attrs(bio, bio_op(orig_bio), orig_bio->bi_rw);
8403 bio->bi_private = dip;
8404 bio->bi_end_io = btrfs_end_dio_bio;
8405 btrfs_io_bio(bio)->logical = file_offset;
8406
8407 map_length = orig_bio->bi_iter.bi_size;
8408 ret = btrfs_map_block(root->fs_info, bio_op(orig_bio),
8409 start_sector << 9,
8410 &map_length, NULL, 0);
8411 if (ret) {
8412 bio_put(bio);
8413 goto out_err;
8414 }
8415
8416 goto next_block;
8417 } else {
8418 submit_len += blocksize;
8419 if (--nr_sectors) {
8420 i++;
8421 goto next_block;
8422 }
8423 bvec++;
8424 }
8425 }
8426
8427 submit:
8428 ret = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8429 async_submit);
8430 if (!ret)
8431 return 0;
8432
8433 bio_put(bio);
8434 out_err:
8435 dip->errors = 1;
8436 /*
8437 * before atomic variable goto zero, we must
8438 * make sure dip->errors is perceived to be set.
8439 */
8440 smp_mb__before_atomic();
8441 if (atomic_dec_and_test(&dip->pending_bios))
8442 bio_io_error(dip->orig_bio);
8443
8444 /* bio_end_io() will handle error, so we needn't return it */
8445 return 0;
8446 }
8447
8448 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8449 loff_t file_offset)
8450 {
8451 struct btrfs_dio_private *dip = NULL;
8452 struct bio *io_bio = NULL;
8453 struct btrfs_io_bio *btrfs_bio;
8454 int skip_sum;
8455 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8456 int ret = 0;
8457
8458 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8459
8460 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8461 if (!io_bio) {
8462 ret = -ENOMEM;
8463 goto free_ordered;
8464 }
8465
8466 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8467 if (!dip) {
8468 ret = -ENOMEM;
8469 goto free_ordered;
8470 }
8471
8472 dip->private = dio_bio->bi_private;
8473 dip->inode = inode;
8474 dip->logical_offset = file_offset;
8475 dip->bytes = dio_bio->bi_iter.bi_size;
8476 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8477 io_bio->bi_private = dip;
8478 dip->orig_bio = io_bio;
8479 dip->dio_bio = dio_bio;
8480 atomic_set(&dip->pending_bios, 0);
8481 btrfs_bio = btrfs_io_bio(io_bio);
8482 btrfs_bio->logical = file_offset;
8483
8484 if (write) {
8485 io_bio->bi_end_io = btrfs_endio_direct_write;
8486 } else {
8487 io_bio->bi_end_io = btrfs_endio_direct_read;
8488 dip->subio_endio = btrfs_subio_endio_read;
8489 }
8490
8491 /*
8492 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8493 * even if we fail to submit a bio, because in such case we do the
8494 * corresponding error handling below and it must not be done a second
8495 * time by btrfs_direct_IO().
8496 */
8497 if (write) {
8498 struct btrfs_dio_data *dio_data = current->journal_info;
8499
8500 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8501 dip->bytes;
8502 dio_data->unsubmitted_oe_range_start =
8503 dio_data->unsubmitted_oe_range_end;
8504 }
8505
8506 ret = btrfs_submit_direct_hook(dip, skip_sum);
8507 if (!ret)
8508 return;
8509
8510 if (btrfs_bio->end_io)
8511 btrfs_bio->end_io(btrfs_bio, ret);
8512
8513 free_ordered:
8514 /*
8515 * If we arrived here it means either we failed to submit the dip
8516 * or we either failed to clone the dio_bio or failed to allocate the
8517 * dip. If we cloned the dio_bio and allocated the dip, we can just
8518 * call bio_endio against our io_bio so that we get proper resource
8519 * cleanup if we fail to submit the dip, otherwise, we must do the
8520 * same as btrfs_endio_direct_[write|read] because we can't call these
8521 * callbacks - they require an allocated dip and a clone of dio_bio.
8522 */
8523 if (io_bio && dip) {
8524 io_bio->bi_error = -EIO;
8525 bio_endio(io_bio);
8526 /*
8527 * The end io callbacks free our dip, do the final put on io_bio
8528 * and all the cleanup and final put for dio_bio (through
8529 * dio_end_io()).
8530 */
8531 dip = NULL;
8532 io_bio = NULL;
8533 } else {
8534 if (write)
8535 btrfs_endio_direct_write_update_ordered(inode,
8536 file_offset,
8537 dio_bio->bi_iter.bi_size,
8538 0);
8539 else
8540 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8541 file_offset + dio_bio->bi_iter.bi_size - 1);
8542
8543 dio_bio->bi_error = -EIO;
8544 /*
8545 * Releases and cleans up our dio_bio, no need to bio_put()
8546 * nor bio_endio()/bio_io_error() against dio_bio.
8547 */
8548 dio_end_io(dio_bio, ret);
8549 }
8550 if (io_bio)
8551 bio_put(io_bio);
8552 kfree(dip);
8553 }
8554
8555 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8556 const struct iov_iter *iter, loff_t offset)
8557 {
8558 int seg;
8559 int i;
8560 unsigned blocksize_mask = root->sectorsize - 1;
8561 ssize_t retval = -EINVAL;
8562
8563 if (offset & blocksize_mask)
8564 goto out;
8565
8566 if (iov_iter_alignment(iter) & blocksize_mask)
8567 goto out;
8568
8569 /* If this is a write we don't need to check anymore */
8570 if (iov_iter_rw(iter) == WRITE)
8571 return 0;
8572 /*
8573 * Check to make sure we don't have duplicate iov_base's in this
8574 * iovec, if so return EINVAL, otherwise we'll get csum errors
8575 * when reading back.
8576 */
8577 for (seg = 0; seg < iter->nr_segs; seg++) {
8578 for (i = seg + 1; i < iter->nr_segs; i++) {
8579 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8580 goto out;
8581 }
8582 }
8583 retval = 0;
8584 out:
8585 return retval;
8586 }
8587
8588 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8589 {
8590 struct file *file = iocb->ki_filp;
8591 struct inode *inode = file->f_mapping->host;
8592 struct btrfs_root *root = BTRFS_I(inode)->root;
8593 struct btrfs_dio_data dio_data = { 0 };
8594 loff_t offset = iocb->ki_pos;
8595 size_t count = 0;
8596 int flags = 0;
8597 bool wakeup = true;
8598 bool relock = false;
8599 ssize_t ret;
8600
8601 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8602 return 0;
8603
8604 inode_dio_begin(inode);
8605 smp_mb__after_atomic();
8606
8607 /*
8608 * The generic stuff only does filemap_write_and_wait_range, which
8609 * isn't enough if we've written compressed pages to this area, so
8610 * we need to flush the dirty pages again to make absolutely sure
8611 * that any outstanding dirty pages are on disk.
8612 */
8613 count = iov_iter_count(iter);
8614 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8615 &BTRFS_I(inode)->runtime_flags))
8616 filemap_fdatawrite_range(inode->i_mapping, offset,
8617 offset + count - 1);
8618
8619 if (iov_iter_rw(iter) == WRITE) {
8620 /*
8621 * If the write DIO is beyond the EOF, we need update
8622 * the isize, but it is protected by i_mutex. So we can
8623 * not unlock the i_mutex at this case.
8624 */
8625 if (offset + count <= inode->i_size) {
8626 inode_unlock(inode);
8627 relock = true;
8628 }
8629 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8630 if (ret)
8631 goto out;
8632 dio_data.outstanding_extents = div64_u64(count +
8633 BTRFS_MAX_EXTENT_SIZE - 1,
8634 BTRFS_MAX_EXTENT_SIZE);
8635
8636 /*
8637 * We need to know how many extents we reserved so that we can
8638 * do the accounting properly if we go over the number we
8639 * originally calculated. Abuse current->journal_info for this.
8640 */
8641 dio_data.reserve = round_up(count, root->sectorsize);
8642 dio_data.unsubmitted_oe_range_start = (u64)offset;
8643 dio_data.unsubmitted_oe_range_end = (u64)offset;
8644 current->journal_info = &dio_data;
8645 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8646 &BTRFS_I(inode)->runtime_flags)) {
8647 inode_dio_end(inode);
8648 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8649 wakeup = false;
8650 }
8651
8652 ret = __blockdev_direct_IO(iocb, inode,
8653 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8654 iter, btrfs_get_blocks_direct, NULL,
8655 btrfs_submit_direct, flags);
8656 if (iov_iter_rw(iter) == WRITE) {
8657 current->journal_info = NULL;
8658 if (ret < 0 && ret != -EIOCBQUEUED) {
8659 if (dio_data.reserve)
8660 btrfs_delalloc_release_space(inode, offset,
8661 dio_data.reserve);
8662 /*
8663 * On error we might have left some ordered extents
8664 * without submitting corresponding bios for them, so
8665 * cleanup them up to avoid other tasks getting them
8666 * and waiting for them to complete forever.
8667 */
8668 if (dio_data.unsubmitted_oe_range_start <
8669 dio_data.unsubmitted_oe_range_end)
8670 btrfs_endio_direct_write_update_ordered(inode,
8671 dio_data.unsubmitted_oe_range_start,
8672 dio_data.unsubmitted_oe_range_end -
8673 dio_data.unsubmitted_oe_range_start,
8674 0);
8675 } else if (ret >= 0 && (size_t)ret < count)
8676 btrfs_delalloc_release_space(inode, offset,
8677 count - (size_t)ret);
8678 }
8679 out:
8680 if (wakeup)
8681 inode_dio_end(inode);
8682 if (relock)
8683 inode_lock(inode);
8684
8685 return ret;
8686 }
8687
8688 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8689
8690 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8691 __u64 start, __u64 len)
8692 {
8693 int ret;
8694
8695 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8696 if (ret)
8697 return ret;
8698
8699 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8700 }
8701
8702 int btrfs_readpage(struct file *file, struct page *page)
8703 {
8704 struct extent_io_tree *tree;
8705 tree = &BTRFS_I(page->mapping->host)->io_tree;
8706 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8707 }
8708
8709 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8710 {
8711 struct extent_io_tree *tree;
8712 struct inode *inode = page->mapping->host;
8713 int ret;
8714
8715 if (current->flags & PF_MEMALLOC) {
8716 redirty_page_for_writepage(wbc, page);
8717 unlock_page(page);
8718 return 0;
8719 }
8720
8721 /*
8722 * If we are under memory pressure we will call this directly from the
8723 * VM, we need to make sure we have the inode referenced for the ordered
8724 * extent. If not just return like we didn't do anything.
8725 */
8726 if (!igrab(inode)) {
8727 redirty_page_for_writepage(wbc, page);
8728 return AOP_WRITEPAGE_ACTIVATE;
8729 }
8730 tree = &BTRFS_I(page->mapping->host)->io_tree;
8731 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8732 btrfs_add_delayed_iput(inode);
8733 return ret;
8734 }
8735
8736 static int btrfs_writepages(struct address_space *mapping,
8737 struct writeback_control *wbc)
8738 {
8739 struct extent_io_tree *tree;
8740
8741 tree = &BTRFS_I(mapping->host)->io_tree;
8742 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8743 }
8744
8745 static int
8746 btrfs_readpages(struct file *file, struct address_space *mapping,
8747 struct list_head *pages, unsigned nr_pages)
8748 {
8749 struct extent_io_tree *tree;
8750 tree = &BTRFS_I(mapping->host)->io_tree;
8751 return extent_readpages(tree, mapping, pages, nr_pages,
8752 btrfs_get_extent);
8753 }
8754 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8755 {
8756 struct extent_io_tree *tree;
8757 struct extent_map_tree *map;
8758 int ret;
8759
8760 tree = &BTRFS_I(page->mapping->host)->io_tree;
8761 map = &BTRFS_I(page->mapping->host)->extent_tree;
8762 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8763 if (ret == 1) {
8764 ClearPagePrivate(page);
8765 set_page_private(page, 0);
8766 put_page(page);
8767 }
8768 return ret;
8769 }
8770
8771 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8772 {
8773 if (PageWriteback(page) || PageDirty(page))
8774 return 0;
8775 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8776 }
8777
8778 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8779 unsigned int length)
8780 {
8781 struct inode *inode = page->mapping->host;
8782 struct extent_io_tree *tree;
8783 struct btrfs_ordered_extent *ordered;
8784 struct extent_state *cached_state = NULL;
8785 u64 page_start = page_offset(page);
8786 u64 page_end = page_start + PAGE_SIZE - 1;
8787 u64 start;
8788 u64 end;
8789 int inode_evicting = inode->i_state & I_FREEING;
8790
8791 /*
8792 * we have the page locked, so new writeback can't start,
8793 * and the dirty bit won't be cleared while we are here.
8794 *
8795 * Wait for IO on this page so that we can safely clear
8796 * the PagePrivate2 bit and do ordered accounting
8797 */
8798 wait_on_page_writeback(page);
8799
8800 tree = &BTRFS_I(inode)->io_tree;
8801 if (offset) {
8802 btrfs_releasepage(page, GFP_NOFS);
8803 return;
8804 }
8805
8806 if (!inode_evicting)
8807 lock_extent_bits(tree, page_start, page_end, &cached_state);
8808 again:
8809 start = page_start;
8810 ordered = btrfs_lookup_ordered_range(inode, start,
8811 page_end - start + 1);
8812 if (ordered) {
8813 end = min(page_end, ordered->file_offset + ordered->len - 1);
8814 /*
8815 * IO on this page will never be started, so we need
8816 * to account for any ordered extents now
8817 */
8818 if (!inode_evicting)
8819 clear_extent_bit(tree, start, end,
8820 EXTENT_DIRTY | EXTENT_DELALLOC |
8821 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8822 EXTENT_DEFRAG, 1, 0, &cached_state,
8823 GFP_NOFS);
8824 /*
8825 * whoever cleared the private bit is responsible
8826 * for the finish_ordered_io
8827 */
8828 if (TestClearPagePrivate2(page)) {
8829 struct btrfs_ordered_inode_tree *tree;
8830 u64 new_len;
8831
8832 tree = &BTRFS_I(inode)->ordered_tree;
8833
8834 spin_lock_irq(&tree->lock);
8835 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8836 new_len = start - ordered->file_offset;
8837 if (new_len < ordered->truncated_len)
8838 ordered->truncated_len = new_len;
8839 spin_unlock_irq(&tree->lock);
8840
8841 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8842 start,
8843 end - start + 1, 1))
8844 btrfs_finish_ordered_io(ordered);
8845 }
8846 btrfs_put_ordered_extent(ordered);
8847 if (!inode_evicting) {
8848 cached_state = NULL;
8849 lock_extent_bits(tree, start, end,
8850 &cached_state);
8851 }
8852
8853 start = end + 1;
8854 if (start < page_end)
8855 goto again;
8856 }
8857
8858 /*
8859 * Qgroup reserved space handler
8860 * Page here will be either
8861 * 1) Already written to disk
8862 * In this case, its reserved space is released from data rsv map
8863 * and will be freed by delayed_ref handler finally.
8864 * So even we call qgroup_free_data(), it won't decrease reserved
8865 * space.
8866 * 2) Not written to disk
8867 * This means the reserved space should be freed here.
8868 */
8869 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
8870 if (!inode_evicting) {
8871 clear_extent_bit(tree, page_start, page_end,
8872 EXTENT_LOCKED | EXTENT_DIRTY |
8873 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8874 EXTENT_DEFRAG, 1, 1,
8875 &cached_state, GFP_NOFS);
8876
8877 __btrfs_releasepage(page, GFP_NOFS);
8878 }
8879
8880 ClearPageChecked(page);
8881 if (PagePrivate(page)) {
8882 ClearPagePrivate(page);
8883 set_page_private(page, 0);
8884 put_page(page);
8885 }
8886 }
8887
8888 /*
8889 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8890 * called from a page fault handler when a page is first dirtied. Hence we must
8891 * be careful to check for EOF conditions here. We set the page up correctly
8892 * for a written page which means we get ENOSPC checking when writing into
8893 * holes and correct delalloc and unwritten extent mapping on filesystems that
8894 * support these features.
8895 *
8896 * We are not allowed to take the i_mutex here so we have to play games to
8897 * protect against truncate races as the page could now be beyond EOF. Because
8898 * vmtruncate() writes the inode size before removing pages, once we have the
8899 * page lock we can determine safely if the page is beyond EOF. If it is not
8900 * beyond EOF, then the page is guaranteed safe against truncation until we
8901 * unlock the page.
8902 */
8903 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8904 {
8905 struct page *page = vmf->page;
8906 struct inode *inode = file_inode(vma->vm_file);
8907 struct btrfs_root *root = BTRFS_I(inode)->root;
8908 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8909 struct btrfs_ordered_extent *ordered;
8910 struct extent_state *cached_state = NULL;
8911 char *kaddr;
8912 unsigned long zero_start;
8913 loff_t size;
8914 int ret;
8915 int reserved = 0;
8916 u64 reserved_space;
8917 u64 page_start;
8918 u64 page_end;
8919 u64 end;
8920
8921 reserved_space = PAGE_SIZE;
8922
8923 sb_start_pagefault(inode->i_sb);
8924 page_start = page_offset(page);
8925 page_end = page_start + PAGE_SIZE - 1;
8926 end = page_end;
8927
8928 /*
8929 * Reserving delalloc space after obtaining the page lock can lead to
8930 * deadlock. For example, if a dirty page is locked by this function
8931 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8932 * dirty page write out, then the btrfs_writepage() function could
8933 * end up waiting indefinitely to get a lock on the page currently
8934 * being processed by btrfs_page_mkwrite() function.
8935 */
8936 ret = btrfs_delalloc_reserve_space(inode, page_start,
8937 reserved_space);
8938 if (!ret) {
8939 ret = file_update_time(vma->vm_file);
8940 reserved = 1;
8941 }
8942 if (ret) {
8943 if (ret == -ENOMEM)
8944 ret = VM_FAULT_OOM;
8945 else /* -ENOSPC, -EIO, etc */
8946 ret = VM_FAULT_SIGBUS;
8947 if (reserved)
8948 goto out;
8949 goto out_noreserve;
8950 }
8951
8952 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8953 again:
8954 lock_page(page);
8955 size = i_size_read(inode);
8956
8957 if ((page->mapping != inode->i_mapping) ||
8958 (page_start >= size)) {
8959 /* page got truncated out from underneath us */
8960 goto out_unlock;
8961 }
8962 wait_on_page_writeback(page);
8963
8964 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8965 set_page_extent_mapped(page);
8966
8967 /*
8968 * we can't set the delalloc bits if there are pending ordered
8969 * extents. Drop our locks and wait for them to finish
8970 */
8971 ordered = btrfs_lookup_ordered_range(inode, page_start, page_end);
8972 if (ordered) {
8973 unlock_extent_cached(io_tree, page_start, page_end,
8974 &cached_state, GFP_NOFS);
8975 unlock_page(page);
8976 btrfs_start_ordered_extent(inode, ordered, 1);
8977 btrfs_put_ordered_extent(ordered);
8978 goto again;
8979 }
8980
8981 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8982 reserved_space = round_up(size - page_start, root->sectorsize);
8983 if (reserved_space < PAGE_SIZE) {
8984 end = page_start + reserved_space - 1;
8985 spin_lock(&BTRFS_I(inode)->lock);
8986 BTRFS_I(inode)->outstanding_extents++;
8987 spin_unlock(&BTRFS_I(inode)->lock);
8988 btrfs_delalloc_release_space(inode, page_start,
8989 PAGE_SIZE - reserved_space);
8990 }
8991 }
8992
8993 /*
8994 * XXX - page_mkwrite gets called every time the page is dirtied, even
8995 * if it was already dirty, so for space accounting reasons we need to
8996 * clear any delalloc bits for the range we are fixing to save. There
8997 * is probably a better way to do this, but for now keep consistent with
8998 * prepare_pages in the normal write path.
8999 */
9000 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9001 EXTENT_DIRTY | EXTENT_DELALLOC |
9002 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9003 0, 0, &cached_state, GFP_NOFS);
9004
9005 ret = btrfs_set_extent_delalloc(inode, page_start, end,
9006 &cached_state);
9007 if (ret) {
9008 unlock_extent_cached(io_tree, page_start, page_end,
9009 &cached_state, GFP_NOFS);
9010 ret = VM_FAULT_SIGBUS;
9011 goto out_unlock;
9012 }
9013 ret = 0;
9014
9015 /* page is wholly or partially inside EOF */
9016 if (page_start + PAGE_SIZE > size)
9017 zero_start = size & ~PAGE_MASK;
9018 else
9019 zero_start = PAGE_SIZE;
9020
9021 if (zero_start != PAGE_SIZE) {
9022 kaddr = kmap(page);
9023 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9024 flush_dcache_page(page);
9025 kunmap(page);
9026 }
9027 ClearPageChecked(page);
9028 set_page_dirty(page);
9029 SetPageUptodate(page);
9030
9031 BTRFS_I(inode)->last_trans = root->fs_info->generation;
9032 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9033 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9034
9035 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9036
9037 out_unlock:
9038 if (!ret) {
9039 sb_end_pagefault(inode->i_sb);
9040 return VM_FAULT_LOCKED;
9041 }
9042 unlock_page(page);
9043 out:
9044 btrfs_delalloc_release_space(inode, page_start, reserved_space);
9045 out_noreserve:
9046 sb_end_pagefault(inode->i_sb);
9047 return ret;
9048 }
9049
9050 static int btrfs_truncate(struct inode *inode)
9051 {
9052 struct btrfs_root *root = BTRFS_I(inode)->root;
9053 struct btrfs_block_rsv *rsv;
9054 int ret = 0;
9055 int err = 0;
9056 struct btrfs_trans_handle *trans;
9057 u64 mask = root->sectorsize - 1;
9058 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
9059
9060 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9061 (u64)-1);
9062 if (ret)
9063 return ret;
9064
9065 /*
9066 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9067 * 3 things going on here
9068 *
9069 * 1) We need to reserve space for our orphan item and the space to
9070 * delete our orphan item. Lord knows we don't want to have a dangling
9071 * orphan item because we didn't reserve space to remove it.
9072 *
9073 * 2) We need to reserve space to update our inode.
9074 *
9075 * 3) We need to have something to cache all the space that is going to
9076 * be free'd up by the truncate operation, but also have some slack
9077 * space reserved in case it uses space during the truncate (thank you
9078 * very much snapshotting).
9079 *
9080 * And we need these to all be separate. The fact is we can use a lot of
9081 * space doing the truncate, and we have no earthly idea how much space
9082 * we will use, so we need the truncate reservation to be separate so it
9083 * doesn't end up using space reserved for updating the inode or
9084 * removing the orphan item. We also need to be able to stop the
9085 * transaction and start a new one, which means we need to be able to
9086 * update the inode several times, and we have no idea of knowing how
9087 * many times that will be, so we can't just reserve 1 item for the
9088 * entirety of the operation, so that has to be done separately as well.
9089 * Then there is the orphan item, which does indeed need to be held on
9090 * to for the whole operation, and we need nobody to touch this reserved
9091 * space except the orphan code.
9092 *
9093 * So that leaves us with
9094 *
9095 * 1) root->orphan_block_rsv - for the orphan deletion.
9096 * 2) rsv - for the truncate reservation, which we will steal from the
9097 * transaction reservation.
9098 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9099 * updating the inode.
9100 */
9101 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
9102 if (!rsv)
9103 return -ENOMEM;
9104 rsv->size = min_size;
9105 rsv->failfast = 1;
9106
9107 /*
9108 * 1 for the truncate slack space
9109 * 1 for updating the inode.
9110 */
9111 trans = btrfs_start_transaction(root, 2);
9112 if (IS_ERR(trans)) {
9113 err = PTR_ERR(trans);
9114 goto out;
9115 }
9116
9117 /* Migrate the slack space for the truncate to our reserve */
9118 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
9119 min_size, 0);
9120 BUG_ON(ret);
9121
9122 /*
9123 * So if we truncate and then write and fsync we normally would just
9124 * write the extents that changed, which is a problem if we need to
9125 * first truncate that entire inode. So set this flag so we write out
9126 * all of the extents in the inode to the sync log so we're completely
9127 * safe.
9128 */
9129 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9130 trans->block_rsv = rsv;
9131
9132 while (1) {
9133 ret = btrfs_truncate_inode_items(trans, root, inode,
9134 inode->i_size,
9135 BTRFS_EXTENT_DATA_KEY);
9136 if (ret != -ENOSPC && ret != -EAGAIN) {
9137 err = ret;
9138 break;
9139 }
9140
9141 trans->block_rsv = &root->fs_info->trans_block_rsv;
9142 ret = btrfs_update_inode(trans, root, inode);
9143 if (ret) {
9144 err = ret;
9145 break;
9146 }
9147
9148 btrfs_end_transaction(trans, root);
9149 btrfs_btree_balance_dirty(root);
9150
9151 trans = btrfs_start_transaction(root, 2);
9152 if (IS_ERR(trans)) {
9153 ret = err = PTR_ERR(trans);
9154 trans = NULL;
9155 break;
9156 }
9157
9158 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9159 rsv, min_size, 0);
9160 BUG_ON(ret); /* shouldn't happen */
9161 trans->block_rsv = rsv;
9162 }
9163
9164 if (ret == 0 && inode->i_nlink > 0) {
9165 trans->block_rsv = root->orphan_block_rsv;
9166 ret = btrfs_orphan_del(trans, inode);
9167 if (ret)
9168 err = ret;
9169 }
9170
9171 if (trans) {
9172 trans->block_rsv = &root->fs_info->trans_block_rsv;
9173 ret = btrfs_update_inode(trans, root, inode);
9174 if (ret && !err)
9175 err = ret;
9176
9177 ret = btrfs_end_transaction(trans, root);
9178 btrfs_btree_balance_dirty(root);
9179 }
9180 out:
9181 btrfs_free_block_rsv(root, rsv);
9182
9183 if (ret && !err)
9184 err = ret;
9185
9186 return err;
9187 }
9188
9189 /*
9190 * create a new subvolume directory/inode (helper for the ioctl).
9191 */
9192 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9193 struct btrfs_root *new_root,
9194 struct btrfs_root *parent_root,
9195 u64 new_dirid)
9196 {
9197 struct inode *inode;
9198 int err;
9199 u64 index = 0;
9200
9201 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9202 new_dirid, new_dirid,
9203 S_IFDIR | (~current_umask() & S_IRWXUGO),
9204 &index);
9205 if (IS_ERR(inode))
9206 return PTR_ERR(inode);
9207 inode->i_op = &btrfs_dir_inode_operations;
9208 inode->i_fop = &btrfs_dir_file_operations;
9209
9210 set_nlink(inode, 1);
9211 btrfs_i_size_write(inode, 0);
9212 unlock_new_inode(inode);
9213
9214 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9215 if (err)
9216 btrfs_err(new_root->fs_info,
9217 "error inheriting subvolume %llu properties: %d",
9218 new_root->root_key.objectid, err);
9219
9220 err = btrfs_update_inode(trans, new_root, inode);
9221
9222 iput(inode);
9223 return err;
9224 }
9225
9226 struct inode *btrfs_alloc_inode(struct super_block *sb)
9227 {
9228 struct btrfs_inode *ei;
9229 struct inode *inode;
9230
9231 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9232 if (!ei)
9233 return NULL;
9234
9235 ei->root = NULL;
9236 ei->generation = 0;
9237 ei->last_trans = 0;
9238 ei->last_sub_trans = 0;
9239 ei->logged_trans = 0;
9240 ei->delalloc_bytes = 0;
9241 ei->defrag_bytes = 0;
9242 ei->disk_i_size = 0;
9243 ei->flags = 0;
9244 ei->csum_bytes = 0;
9245 ei->index_cnt = (u64)-1;
9246 ei->dir_index = 0;
9247 ei->last_unlink_trans = 0;
9248 ei->last_log_commit = 0;
9249 ei->delayed_iput_count = 0;
9250
9251 spin_lock_init(&ei->lock);
9252 ei->outstanding_extents = 0;
9253 ei->reserved_extents = 0;
9254
9255 ei->runtime_flags = 0;
9256 ei->force_compress = BTRFS_COMPRESS_NONE;
9257
9258 ei->delayed_node = NULL;
9259
9260 ei->i_otime.tv_sec = 0;
9261 ei->i_otime.tv_nsec = 0;
9262
9263 inode = &ei->vfs_inode;
9264 extent_map_tree_init(&ei->extent_tree);
9265 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9266 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9267 ei->io_tree.track_uptodate = 1;
9268 ei->io_failure_tree.track_uptodate = 1;
9269 atomic_set(&ei->sync_writers, 0);
9270 mutex_init(&ei->log_mutex);
9271 mutex_init(&ei->delalloc_mutex);
9272 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9273 INIT_LIST_HEAD(&ei->delalloc_inodes);
9274 INIT_LIST_HEAD(&ei->delayed_iput);
9275 RB_CLEAR_NODE(&ei->rb_node);
9276 init_rwsem(&ei->dio_sem);
9277
9278 return inode;
9279 }
9280
9281 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9282 void btrfs_test_destroy_inode(struct inode *inode)
9283 {
9284 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9285 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9286 }
9287 #endif
9288
9289 static void btrfs_i_callback(struct rcu_head *head)
9290 {
9291 struct inode *inode = container_of(head, struct inode, i_rcu);
9292 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9293 }
9294
9295 void btrfs_destroy_inode(struct inode *inode)
9296 {
9297 struct btrfs_ordered_extent *ordered;
9298 struct btrfs_root *root = BTRFS_I(inode)->root;
9299
9300 WARN_ON(!hlist_empty(&inode->i_dentry));
9301 WARN_ON(inode->i_data.nrpages);
9302 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9303 WARN_ON(BTRFS_I(inode)->reserved_extents);
9304 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9305 WARN_ON(BTRFS_I(inode)->csum_bytes);
9306 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9307
9308 /*
9309 * This can happen where we create an inode, but somebody else also
9310 * created the same inode and we need to destroy the one we already
9311 * created.
9312 */
9313 if (!root)
9314 goto free;
9315
9316 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9317 &BTRFS_I(inode)->runtime_flags)) {
9318 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9319 btrfs_ino(inode));
9320 atomic_dec(&root->orphan_inodes);
9321 }
9322
9323 while (1) {
9324 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9325 if (!ordered)
9326 break;
9327 else {
9328 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9329 ordered->file_offset, ordered->len);
9330 btrfs_remove_ordered_extent(inode, ordered);
9331 btrfs_put_ordered_extent(ordered);
9332 btrfs_put_ordered_extent(ordered);
9333 }
9334 }
9335 btrfs_qgroup_check_reserved_leak(inode);
9336 inode_tree_del(inode);
9337 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9338 free:
9339 call_rcu(&inode->i_rcu, btrfs_i_callback);
9340 }
9341
9342 int btrfs_drop_inode(struct inode *inode)
9343 {
9344 struct btrfs_root *root = BTRFS_I(inode)->root;
9345
9346 if (root == NULL)
9347 return 1;
9348
9349 /* the snap/subvol tree is on deleting */
9350 if (btrfs_root_refs(&root->root_item) == 0)
9351 return 1;
9352 else
9353 return generic_drop_inode(inode);
9354 }
9355
9356 static void init_once(void *foo)
9357 {
9358 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9359
9360 inode_init_once(&ei->vfs_inode);
9361 }
9362
9363 void btrfs_destroy_cachep(void)
9364 {
9365 /*
9366 * Make sure all delayed rcu free inodes are flushed before we
9367 * destroy cache.
9368 */
9369 rcu_barrier();
9370 kmem_cache_destroy(btrfs_inode_cachep);
9371 kmem_cache_destroy(btrfs_trans_handle_cachep);
9372 kmem_cache_destroy(btrfs_transaction_cachep);
9373 kmem_cache_destroy(btrfs_path_cachep);
9374 kmem_cache_destroy(btrfs_free_space_cachep);
9375 }
9376
9377 int btrfs_init_cachep(void)
9378 {
9379 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9380 sizeof(struct btrfs_inode), 0,
9381 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9382 init_once);
9383 if (!btrfs_inode_cachep)
9384 goto fail;
9385
9386 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9387 sizeof(struct btrfs_trans_handle), 0,
9388 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9389 if (!btrfs_trans_handle_cachep)
9390 goto fail;
9391
9392 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9393 sizeof(struct btrfs_transaction), 0,
9394 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9395 if (!btrfs_transaction_cachep)
9396 goto fail;
9397
9398 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9399 sizeof(struct btrfs_path), 0,
9400 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9401 if (!btrfs_path_cachep)
9402 goto fail;
9403
9404 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9405 sizeof(struct btrfs_free_space), 0,
9406 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9407 if (!btrfs_free_space_cachep)
9408 goto fail;
9409
9410 return 0;
9411 fail:
9412 btrfs_destroy_cachep();
9413 return -ENOMEM;
9414 }
9415
9416 static int btrfs_getattr(struct vfsmount *mnt,
9417 struct dentry *dentry, struct kstat *stat)
9418 {
9419 u64 delalloc_bytes;
9420 struct inode *inode = d_inode(dentry);
9421 u32 blocksize = inode->i_sb->s_blocksize;
9422
9423 generic_fillattr(inode, stat);
9424 stat->dev = BTRFS_I(inode)->root->anon_dev;
9425
9426 spin_lock(&BTRFS_I(inode)->lock);
9427 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9428 spin_unlock(&BTRFS_I(inode)->lock);
9429 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9430 ALIGN(delalloc_bytes, blocksize)) >> 9;
9431 return 0;
9432 }
9433
9434 static int btrfs_rename_exchange(struct inode *old_dir,
9435 struct dentry *old_dentry,
9436 struct inode *new_dir,
9437 struct dentry *new_dentry)
9438 {
9439 struct btrfs_trans_handle *trans;
9440 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9441 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9442 struct inode *new_inode = new_dentry->d_inode;
9443 struct inode *old_inode = old_dentry->d_inode;
9444 struct timespec ctime = CURRENT_TIME;
9445 struct dentry *parent;
9446 u64 old_ino = btrfs_ino(old_inode);
9447 u64 new_ino = btrfs_ino(new_inode);
9448 u64 old_idx = 0;
9449 u64 new_idx = 0;
9450 u64 root_objectid;
9451 int ret;
9452 bool root_log_pinned = false;
9453 bool dest_log_pinned = false;
9454
9455 /* we only allow rename subvolume link between subvolumes */
9456 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9457 return -EXDEV;
9458
9459 /* close the race window with snapshot create/destroy ioctl */
9460 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9461 down_read(&root->fs_info->subvol_sem);
9462 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9463 down_read(&dest->fs_info->subvol_sem);
9464
9465 /*
9466 * We want to reserve the absolute worst case amount of items. So if
9467 * both inodes are subvols and we need to unlink them then that would
9468 * require 4 item modifications, but if they are both normal inodes it
9469 * would require 5 item modifications, so we'll assume their normal
9470 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9471 * should cover the worst case number of items we'll modify.
9472 */
9473 trans = btrfs_start_transaction(root, 12);
9474 if (IS_ERR(trans)) {
9475 ret = PTR_ERR(trans);
9476 goto out_notrans;
9477 }
9478
9479 /*
9480 * We need to find a free sequence number both in the source and
9481 * in the destination directory for the exchange.
9482 */
9483 ret = btrfs_set_inode_index(new_dir, &old_idx);
9484 if (ret)
9485 goto out_fail;
9486 ret = btrfs_set_inode_index(old_dir, &new_idx);
9487 if (ret)
9488 goto out_fail;
9489
9490 BTRFS_I(old_inode)->dir_index = 0ULL;
9491 BTRFS_I(new_inode)->dir_index = 0ULL;
9492
9493 /* Reference for the source. */
9494 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9495 /* force full log commit if subvolume involved. */
9496 btrfs_set_log_full_commit(root->fs_info, trans);
9497 } else {
9498 btrfs_pin_log_trans(root);
9499 root_log_pinned = true;
9500 ret = btrfs_insert_inode_ref(trans, dest,
9501 new_dentry->d_name.name,
9502 new_dentry->d_name.len,
9503 old_ino,
9504 btrfs_ino(new_dir), old_idx);
9505 if (ret)
9506 goto out_fail;
9507 }
9508
9509 /* And now for the dest. */
9510 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9511 /* force full log commit if subvolume involved. */
9512 btrfs_set_log_full_commit(dest->fs_info, trans);
9513 } else {
9514 btrfs_pin_log_trans(dest);
9515 dest_log_pinned = true;
9516 ret = btrfs_insert_inode_ref(trans, root,
9517 old_dentry->d_name.name,
9518 old_dentry->d_name.len,
9519 new_ino,
9520 btrfs_ino(old_dir), new_idx);
9521 if (ret)
9522 goto out_fail;
9523 }
9524
9525 /* Update inode version and ctime/mtime. */
9526 inode_inc_iversion(old_dir);
9527 inode_inc_iversion(new_dir);
9528 inode_inc_iversion(old_inode);
9529 inode_inc_iversion(new_inode);
9530 old_dir->i_ctime = old_dir->i_mtime = ctime;
9531 new_dir->i_ctime = new_dir->i_mtime = ctime;
9532 old_inode->i_ctime = ctime;
9533 new_inode->i_ctime = ctime;
9534
9535 if (old_dentry->d_parent != new_dentry->d_parent) {
9536 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9537 btrfs_record_unlink_dir(trans, new_dir, new_inode, 1);
9538 }
9539
9540 /* src is a subvolume */
9541 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9542 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9543 ret = btrfs_unlink_subvol(trans, root, old_dir,
9544 root_objectid,
9545 old_dentry->d_name.name,
9546 old_dentry->d_name.len);
9547 } else { /* src is an inode */
9548 ret = __btrfs_unlink_inode(trans, root, old_dir,
9549 old_dentry->d_inode,
9550 old_dentry->d_name.name,
9551 old_dentry->d_name.len);
9552 if (!ret)
9553 ret = btrfs_update_inode(trans, root, old_inode);
9554 }
9555 if (ret) {
9556 btrfs_abort_transaction(trans, root, ret);
9557 goto out_fail;
9558 }
9559
9560 /* dest is a subvolume */
9561 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9562 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9563 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9564 root_objectid,
9565 new_dentry->d_name.name,
9566 new_dentry->d_name.len);
9567 } else { /* dest is an inode */
9568 ret = __btrfs_unlink_inode(trans, dest, new_dir,
9569 new_dentry->d_inode,
9570 new_dentry->d_name.name,
9571 new_dentry->d_name.len);
9572 if (!ret)
9573 ret = btrfs_update_inode(trans, dest, new_inode);
9574 }
9575 if (ret) {
9576 btrfs_abort_transaction(trans, root, ret);
9577 goto out_fail;
9578 }
9579
9580 ret = btrfs_add_link(trans, new_dir, old_inode,
9581 new_dentry->d_name.name,
9582 new_dentry->d_name.len, 0, old_idx);
9583 if (ret) {
9584 btrfs_abort_transaction(trans, root, ret);
9585 goto out_fail;
9586 }
9587
9588 ret = btrfs_add_link(trans, old_dir, new_inode,
9589 old_dentry->d_name.name,
9590 old_dentry->d_name.len, 0, new_idx);
9591 if (ret) {
9592 btrfs_abort_transaction(trans, root, ret);
9593 goto out_fail;
9594 }
9595
9596 if (old_inode->i_nlink == 1)
9597 BTRFS_I(old_inode)->dir_index = old_idx;
9598 if (new_inode->i_nlink == 1)
9599 BTRFS_I(new_inode)->dir_index = new_idx;
9600
9601 if (root_log_pinned) {
9602 parent = new_dentry->d_parent;
9603 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9604 btrfs_end_log_trans(root);
9605 root_log_pinned = false;
9606 }
9607 if (dest_log_pinned) {
9608 parent = old_dentry->d_parent;
9609 btrfs_log_new_name(trans, new_inode, new_dir, parent);
9610 btrfs_end_log_trans(dest);
9611 dest_log_pinned = false;
9612 }
9613 out_fail:
9614 /*
9615 * If we have pinned a log and an error happened, we unpin tasks
9616 * trying to sync the log and force them to fallback to a transaction
9617 * commit if the log currently contains any of the inodes involved in
9618 * this rename operation (to ensure we do not persist a log with an
9619 * inconsistent state for any of these inodes or leading to any
9620 * inconsistencies when replayed). If the transaction was aborted, the
9621 * abortion reason is propagated to userspace when attempting to commit
9622 * the transaction. If the log does not contain any of these inodes, we
9623 * allow the tasks to sync it.
9624 */
9625 if (ret && (root_log_pinned || dest_log_pinned)) {
9626 if (btrfs_inode_in_log(old_dir, root->fs_info->generation) ||
9627 btrfs_inode_in_log(new_dir, root->fs_info->generation) ||
9628 btrfs_inode_in_log(old_inode, root->fs_info->generation) ||
9629 (new_inode &&
9630 btrfs_inode_in_log(new_inode, root->fs_info->generation)))
9631 btrfs_set_log_full_commit(root->fs_info, trans);
9632
9633 if (root_log_pinned) {
9634 btrfs_end_log_trans(root);
9635 root_log_pinned = false;
9636 }
9637 if (dest_log_pinned) {
9638 btrfs_end_log_trans(dest);
9639 dest_log_pinned = false;
9640 }
9641 }
9642 ret = btrfs_end_transaction(trans, root);
9643 out_notrans:
9644 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9645 up_read(&dest->fs_info->subvol_sem);
9646 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9647 up_read(&root->fs_info->subvol_sem);
9648
9649 return ret;
9650 }
9651
9652 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9653 struct btrfs_root *root,
9654 struct inode *dir,
9655 struct dentry *dentry)
9656 {
9657 int ret;
9658 struct inode *inode;
9659 u64 objectid;
9660 u64 index;
9661
9662 ret = btrfs_find_free_ino(root, &objectid);
9663 if (ret)
9664 return ret;
9665
9666 inode = btrfs_new_inode(trans, root, dir,
9667 dentry->d_name.name,
9668 dentry->d_name.len,
9669 btrfs_ino(dir),
9670 objectid,
9671 S_IFCHR | WHITEOUT_MODE,
9672 &index);
9673
9674 if (IS_ERR(inode)) {
9675 ret = PTR_ERR(inode);
9676 return ret;
9677 }
9678
9679 inode->i_op = &btrfs_special_inode_operations;
9680 init_special_inode(inode, inode->i_mode,
9681 WHITEOUT_DEV);
9682
9683 ret = btrfs_init_inode_security(trans, inode, dir,
9684 &dentry->d_name);
9685 if (ret)
9686 goto out;
9687
9688 ret = btrfs_add_nondir(trans, dir, dentry,
9689 inode, 0, index);
9690 if (ret)
9691 goto out;
9692
9693 ret = btrfs_update_inode(trans, root, inode);
9694 out:
9695 unlock_new_inode(inode);
9696 if (ret)
9697 inode_dec_link_count(inode);
9698 iput(inode);
9699
9700 return ret;
9701 }
9702
9703 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9704 struct inode *new_dir, struct dentry *new_dentry,
9705 unsigned int flags)
9706 {
9707 struct btrfs_trans_handle *trans;
9708 unsigned int trans_num_items;
9709 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9710 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9711 struct inode *new_inode = d_inode(new_dentry);
9712 struct inode *old_inode = d_inode(old_dentry);
9713 u64 index = 0;
9714 u64 root_objectid;
9715 int ret;
9716 u64 old_ino = btrfs_ino(old_inode);
9717 bool log_pinned = false;
9718
9719 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9720 return -EPERM;
9721
9722 /* we only allow rename subvolume link between subvolumes */
9723 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9724 return -EXDEV;
9725
9726 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9727 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9728 return -ENOTEMPTY;
9729
9730 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9731 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9732 return -ENOTEMPTY;
9733
9734
9735 /* check for collisions, even if the name isn't there */
9736 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9737 new_dentry->d_name.name,
9738 new_dentry->d_name.len);
9739
9740 if (ret) {
9741 if (ret == -EEXIST) {
9742 /* we shouldn't get
9743 * eexist without a new_inode */
9744 if (WARN_ON(!new_inode)) {
9745 return ret;
9746 }
9747 } else {
9748 /* maybe -EOVERFLOW */
9749 return ret;
9750 }
9751 }
9752 ret = 0;
9753
9754 /*
9755 * we're using rename to replace one file with another. Start IO on it
9756 * now so we don't add too much work to the end of the transaction
9757 */
9758 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9759 filemap_flush(old_inode->i_mapping);
9760
9761 /* close the racy window with snapshot create/destroy ioctl */
9762 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9763 down_read(&root->fs_info->subvol_sem);
9764 /*
9765 * We want to reserve the absolute worst case amount of items. So if
9766 * both inodes are subvols and we need to unlink them then that would
9767 * require 4 item modifications, but if they are both normal inodes it
9768 * would require 5 item modifications, so we'll assume they are normal
9769 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9770 * should cover the worst case number of items we'll modify.
9771 * If our rename has the whiteout flag, we need more 5 units for the
9772 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9773 * when selinux is enabled).
9774 */
9775 trans_num_items = 11;
9776 if (flags & RENAME_WHITEOUT)
9777 trans_num_items += 5;
9778 trans = btrfs_start_transaction(root, trans_num_items);
9779 if (IS_ERR(trans)) {
9780 ret = PTR_ERR(trans);
9781 goto out_notrans;
9782 }
9783
9784 if (dest != root)
9785 btrfs_record_root_in_trans(trans, dest);
9786
9787 ret = btrfs_set_inode_index(new_dir, &index);
9788 if (ret)
9789 goto out_fail;
9790
9791 BTRFS_I(old_inode)->dir_index = 0ULL;
9792 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9793 /* force full log commit if subvolume involved. */
9794 btrfs_set_log_full_commit(root->fs_info, trans);
9795 } else {
9796 btrfs_pin_log_trans(root);
9797 log_pinned = true;
9798 ret = btrfs_insert_inode_ref(trans, dest,
9799 new_dentry->d_name.name,
9800 new_dentry->d_name.len,
9801 old_ino,
9802 btrfs_ino(new_dir), index);
9803 if (ret)
9804 goto out_fail;
9805 }
9806
9807 inode_inc_iversion(old_dir);
9808 inode_inc_iversion(new_dir);
9809 inode_inc_iversion(old_inode);
9810 old_dir->i_ctime = old_dir->i_mtime =
9811 new_dir->i_ctime = new_dir->i_mtime =
9812 old_inode->i_ctime = current_fs_time(old_dir->i_sb);
9813
9814 if (old_dentry->d_parent != new_dentry->d_parent)
9815 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9816
9817 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9818 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9819 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9820 old_dentry->d_name.name,
9821 old_dentry->d_name.len);
9822 } else {
9823 ret = __btrfs_unlink_inode(trans, root, old_dir,
9824 d_inode(old_dentry),
9825 old_dentry->d_name.name,
9826 old_dentry->d_name.len);
9827 if (!ret)
9828 ret = btrfs_update_inode(trans, root, old_inode);
9829 }
9830 if (ret) {
9831 btrfs_abort_transaction(trans, root, ret);
9832 goto out_fail;
9833 }
9834
9835 if (new_inode) {
9836 inode_inc_iversion(new_inode);
9837 new_inode->i_ctime = current_fs_time(new_inode->i_sb);
9838 if (unlikely(btrfs_ino(new_inode) ==
9839 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9840 root_objectid = BTRFS_I(new_inode)->location.objectid;
9841 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9842 root_objectid,
9843 new_dentry->d_name.name,
9844 new_dentry->d_name.len);
9845 BUG_ON(new_inode->i_nlink == 0);
9846 } else {
9847 ret = btrfs_unlink_inode(trans, dest, new_dir,
9848 d_inode(new_dentry),
9849 new_dentry->d_name.name,
9850 new_dentry->d_name.len);
9851 }
9852 if (!ret && new_inode->i_nlink == 0)
9853 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9854 if (ret) {
9855 btrfs_abort_transaction(trans, root, ret);
9856 goto out_fail;
9857 }
9858 }
9859
9860 ret = btrfs_add_link(trans, new_dir, old_inode,
9861 new_dentry->d_name.name,
9862 new_dentry->d_name.len, 0, index);
9863 if (ret) {
9864 btrfs_abort_transaction(trans, root, ret);
9865 goto out_fail;
9866 }
9867
9868 if (old_inode->i_nlink == 1)
9869 BTRFS_I(old_inode)->dir_index = index;
9870
9871 if (log_pinned) {
9872 struct dentry *parent = new_dentry->d_parent;
9873
9874 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9875 btrfs_end_log_trans(root);
9876 log_pinned = false;
9877 }
9878
9879 if (flags & RENAME_WHITEOUT) {
9880 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9881 old_dentry);
9882
9883 if (ret) {
9884 btrfs_abort_transaction(trans, root, ret);
9885 goto out_fail;
9886 }
9887 }
9888 out_fail:
9889 /*
9890 * If we have pinned the log and an error happened, we unpin tasks
9891 * trying to sync the log and force them to fallback to a transaction
9892 * commit if the log currently contains any of the inodes involved in
9893 * this rename operation (to ensure we do not persist a log with an
9894 * inconsistent state for any of these inodes or leading to any
9895 * inconsistencies when replayed). If the transaction was aborted, the
9896 * abortion reason is propagated to userspace when attempting to commit
9897 * the transaction. If the log does not contain any of these inodes, we
9898 * allow the tasks to sync it.
9899 */
9900 if (ret && log_pinned) {
9901 if (btrfs_inode_in_log(old_dir, root->fs_info->generation) ||
9902 btrfs_inode_in_log(new_dir, root->fs_info->generation) ||
9903 btrfs_inode_in_log(old_inode, root->fs_info->generation) ||
9904 (new_inode &&
9905 btrfs_inode_in_log(new_inode, root->fs_info->generation)))
9906 btrfs_set_log_full_commit(root->fs_info, trans);
9907
9908 btrfs_end_log_trans(root);
9909 log_pinned = false;
9910 }
9911 btrfs_end_transaction(trans, root);
9912 out_notrans:
9913 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9914 up_read(&root->fs_info->subvol_sem);
9915
9916 return ret;
9917 }
9918
9919 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9920 struct inode *new_dir, struct dentry *new_dentry,
9921 unsigned int flags)
9922 {
9923 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9924 return -EINVAL;
9925
9926 if (flags & RENAME_EXCHANGE)
9927 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9928 new_dentry);
9929
9930 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9931 }
9932
9933 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9934 {
9935 struct btrfs_delalloc_work *delalloc_work;
9936 struct inode *inode;
9937
9938 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9939 work);
9940 inode = delalloc_work->inode;
9941 filemap_flush(inode->i_mapping);
9942 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9943 &BTRFS_I(inode)->runtime_flags))
9944 filemap_flush(inode->i_mapping);
9945
9946 if (delalloc_work->delay_iput)
9947 btrfs_add_delayed_iput(inode);
9948 else
9949 iput(inode);
9950 complete(&delalloc_work->completion);
9951 }
9952
9953 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9954 int delay_iput)
9955 {
9956 struct btrfs_delalloc_work *work;
9957
9958 work = kmalloc(sizeof(*work), GFP_NOFS);
9959 if (!work)
9960 return NULL;
9961
9962 init_completion(&work->completion);
9963 INIT_LIST_HEAD(&work->list);
9964 work->inode = inode;
9965 work->delay_iput = delay_iput;
9966 WARN_ON_ONCE(!inode);
9967 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9968 btrfs_run_delalloc_work, NULL, NULL);
9969
9970 return work;
9971 }
9972
9973 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9974 {
9975 wait_for_completion(&work->completion);
9976 kfree(work);
9977 }
9978
9979 /*
9980 * some fairly slow code that needs optimization. This walks the list
9981 * of all the inodes with pending delalloc and forces them to disk.
9982 */
9983 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9984 int nr)
9985 {
9986 struct btrfs_inode *binode;
9987 struct inode *inode;
9988 struct btrfs_delalloc_work *work, *next;
9989 struct list_head works;
9990 struct list_head splice;
9991 int ret = 0;
9992
9993 INIT_LIST_HEAD(&works);
9994 INIT_LIST_HEAD(&splice);
9995
9996 mutex_lock(&root->delalloc_mutex);
9997 spin_lock(&root->delalloc_lock);
9998 list_splice_init(&root->delalloc_inodes, &splice);
9999 while (!list_empty(&splice)) {
10000 binode = list_entry(splice.next, struct btrfs_inode,
10001 delalloc_inodes);
10002
10003 list_move_tail(&binode->delalloc_inodes,
10004 &root->delalloc_inodes);
10005 inode = igrab(&binode->vfs_inode);
10006 if (!inode) {
10007 cond_resched_lock(&root->delalloc_lock);
10008 continue;
10009 }
10010 spin_unlock(&root->delalloc_lock);
10011
10012 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10013 if (!work) {
10014 if (delay_iput)
10015 btrfs_add_delayed_iput(inode);
10016 else
10017 iput(inode);
10018 ret = -ENOMEM;
10019 goto out;
10020 }
10021 list_add_tail(&work->list, &works);
10022 btrfs_queue_work(root->fs_info->flush_workers,
10023 &work->work);
10024 ret++;
10025 if (nr != -1 && ret >= nr)
10026 goto out;
10027 cond_resched();
10028 spin_lock(&root->delalloc_lock);
10029 }
10030 spin_unlock(&root->delalloc_lock);
10031
10032 out:
10033 list_for_each_entry_safe(work, next, &works, list) {
10034 list_del_init(&work->list);
10035 btrfs_wait_and_free_delalloc_work(work);
10036 }
10037
10038 if (!list_empty_careful(&splice)) {
10039 spin_lock(&root->delalloc_lock);
10040 list_splice_tail(&splice, &root->delalloc_inodes);
10041 spin_unlock(&root->delalloc_lock);
10042 }
10043 mutex_unlock(&root->delalloc_mutex);
10044 return ret;
10045 }
10046
10047 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10048 {
10049 int ret;
10050
10051 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
10052 return -EROFS;
10053
10054 ret = __start_delalloc_inodes(root, delay_iput, -1);
10055 if (ret > 0)
10056 ret = 0;
10057 /*
10058 * the filemap_flush will queue IO into the worker threads, but
10059 * we have to make sure the IO is actually started and that
10060 * ordered extents get created before we return
10061 */
10062 atomic_inc(&root->fs_info->async_submit_draining);
10063 while (atomic_read(&root->fs_info->nr_async_submits) ||
10064 atomic_read(&root->fs_info->async_delalloc_pages)) {
10065 wait_event(root->fs_info->async_submit_wait,
10066 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
10067 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
10068 }
10069 atomic_dec(&root->fs_info->async_submit_draining);
10070 return ret;
10071 }
10072
10073 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10074 int nr)
10075 {
10076 struct btrfs_root *root;
10077 struct list_head splice;
10078 int ret;
10079
10080 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10081 return -EROFS;
10082
10083 INIT_LIST_HEAD(&splice);
10084
10085 mutex_lock(&fs_info->delalloc_root_mutex);
10086 spin_lock(&fs_info->delalloc_root_lock);
10087 list_splice_init(&fs_info->delalloc_roots, &splice);
10088 while (!list_empty(&splice) && nr) {
10089 root = list_first_entry(&splice, struct btrfs_root,
10090 delalloc_root);
10091 root = btrfs_grab_fs_root(root);
10092 BUG_ON(!root);
10093 list_move_tail(&root->delalloc_root,
10094 &fs_info->delalloc_roots);
10095 spin_unlock(&fs_info->delalloc_root_lock);
10096
10097 ret = __start_delalloc_inodes(root, delay_iput, nr);
10098 btrfs_put_fs_root(root);
10099 if (ret < 0)
10100 goto out;
10101
10102 if (nr != -1) {
10103 nr -= ret;
10104 WARN_ON(nr < 0);
10105 }
10106 spin_lock(&fs_info->delalloc_root_lock);
10107 }
10108 spin_unlock(&fs_info->delalloc_root_lock);
10109
10110 ret = 0;
10111 atomic_inc(&fs_info->async_submit_draining);
10112 while (atomic_read(&fs_info->nr_async_submits) ||
10113 atomic_read(&fs_info->async_delalloc_pages)) {
10114 wait_event(fs_info->async_submit_wait,
10115 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10116 atomic_read(&fs_info->async_delalloc_pages) == 0));
10117 }
10118 atomic_dec(&fs_info->async_submit_draining);
10119 out:
10120 if (!list_empty_careful(&splice)) {
10121 spin_lock(&fs_info->delalloc_root_lock);
10122 list_splice_tail(&splice, &fs_info->delalloc_roots);
10123 spin_unlock(&fs_info->delalloc_root_lock);
10124 }
10125 mutex_unlock(&fs_info->delalloc_root_mutex);
10126 return ret;
10127 }
10128
10129 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10130 const char *symname)
10131 {
10132 struct btrfs_trans_handle *trans;
10133 struct btrfs_root *root = BTRFS_I(dir)->root;
10134 struct btrfs_path *path;
10135 struct btrfs_key key;
10136 struct inode *inode = NULL;
10137 int err;
10138 int drop_inode = 0;
10139 u64 objectid;
10140 u64 index = 0;
10141 int name_len;
10142 int datasize;
10143 unsigned long ptr;
10144 struct btrfs_file_extent_item *ei;
10145 struct extent_buffer *leaf;
10146
10147 name_len = strlen(symname);
10148 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
10149 return -ENAMETOOLONG;
10150
10151 /*
10152 * 2 items for inode item and ref
10153 * 2 items for dir items
10154 * 1 item for updating parent inode item
10155 * 1 item for the inline extent item
10156 * 1 item for xattr if selinux is on
10157 */
10158 trans = btrfs_start_transaction(root, 7);
10159 if (IS_ERR(trans))
10160 return PTR_ERR(trans);
10161
10162 err = btrfs_find_free_ino(root, &objectid);
10163 if (err)
10164 goto out_unlock;
10165
10166 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10167 dentry->d_name.len, btrfs_ino(dir), objectid,
10168 S_IFLNK|S_IRWXUGO, &index);
10169 if (IS_ERR(inode)) {
10170 err = PTR_ERR(inode);
10171 goto out_unlock;
10172 }
10173
10174 /*
10175 * If the active LSM wants to access the inode during
10176 * d_instantiate it needs these. Smack checks to see
10177 * if the filesystem supports xattrs by looking at the
10178 * ops vector.
10179 */
10180 inode->i_fop = &btrfs_file_operations;
10181 inode->i_op = &btrfs_file_inode_operations;
10182 inode->i_mapping->a_ops = &btrfs_aops;
10183 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10184
10185 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10186 if (err)
10187 goto out_unlock_inode;
10188
10189 path = btrfs_alloc_path();
10190 if (!path) {
10191 err = -ENOMEM;
10192 goto out_unlock_inode;
10193 }
10194 key.objectid = btrfs_ino(inode);
10195 key.offset = 0;
10196 key.type = BTRFS_EXTENT_DATA_KEY;
10197 datasize = btrfs_file_extent_calc_inline_size(name_len);
10198 err = btrfs_insert_empty_item(trans, root, path, &key,
10199 datasize);
10200 if (err) {
10201 btrfs_free_path(path);
10202 goto out_unlock_inode;
10203 }
10204 leaf = path->nodes[0];
10205 ei = btrfs_item_ptr(leaf, path->slots[0],
10206 struct btrfs_file_extent_item);
10207 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10208 btrfs_set_file_extent_type(leaf, ei,
10209 BTRFS_FILE_EXTENT_INLINE);
10210 btrfs_set_file_extent_encryption(leaf, ei, 0);
10211 btrfs_set_file_extent_compression(leaf, ei, 0);
10212 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10213 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10214
10215 ptr = btrfs_file_extent_inline_start(ei);
10216 write_extent_buffer(leaf, symname, ptr, name_len);
10217 btrfs_mark_buffer_dirty(leaf);
10218 btrfs_free_path(path);
10219
10220 inode->i_op = &btrfs_symlink_inode_operations;
10221 inode_nohighmem(inode);
10222 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10223 inode_set_bytes(inode, name_len);
10224 btrfs_i_size_write(inode, name_len);
10225 err = btrfs_update_inode(trans, root, inode);
10226 /*
10227 * Last step, add directory indexes for our symlink inode. This is the
10228 * last step to avoid extra cleanup of these indexes if an error happens
10229 * elsewhere above.
10230 */
10231 if (!err)
10232 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
10233 if (err) {
10234 drop_inode = 1;
10235 goto out_unlock_inode;
10236 }
10237
10238 unlock_new_inode(inode);
10239 d_instantiate(dentry, inode);
10240
10241 out_unlock:
10242 btrfs_end_transaction(trans, root);
10243 if (drop_inode) {
10244 inode_dec_link_count(inode);
10245 iput(inode);
10246 }
10247 btrfs_btree_balance_dirty(root);
10248 return err;
10249
10250 out_unlock_inode:
10251 drop_inode = 1;
10252 unlock_new_inode(inode);
10253 goto out_unlock;
10254 }
10255
10256 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10257 u64 start, u64 num_bytes, u64 min_size,
10258 loff_t actual_len, u64 *alloc_hint,
10259 struct btrfs_trans_handle *trans)
10260 {
10261 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10262 struct extent_map *em;
10263 struct btrfs_root *root = BTRFS_I(inode)->root;
10264 struct btrfs_key ins;
10265 u64 cur_offset = start;
10266 u64 i_size;
10267 u64 cur_bytes;
10268 u64 last_alloc = (u64)-1;
10269 int ret = 0;
10270 bool own_trans = true;
10271
10272 if (trans)
10273 own_trans = false;
10274 while (num_bytes > 0) {
10275 if (own_trans) {
10276 trans = btrfs_start_transaction(root, 3);
10277 if (IS_ERR(trans)) {
10278 ret = PTR_ERR(trans);
10279 break;
10280 }
10281 }
10282
10283 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10284 cur_bytes = max(cur_bytes, min_size);
10285 /*
10286 * If we are severely fragmented we could end up with really
10287 * small allocations, so if the allocator is returning small
10288 * chunks lets make its job easier by only searching for those
10289 * sized chunks.
10290 */
10291 cur_bytes = min(cur_bytes, last_alloc);
10292 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
10293 *alloc_hint, &ins, 1, 0);
10294 if (ret) {
10295 if (own_trans)
10296 btrfs_end_transaction(trans, root);
10297 break;
10298 }
10299 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
10300
10301 last_alloc = ins.offset;
10302 ret = insert_reserved_file_extent(trans, inode,
10303 cur_offset, ins.objectid,
10304 ins.offset, ins.offset,
10305 ins.offset, 0, 0, 0,
10306 BTRFS_FILE_EXTENT_PREALLOC);
10307 if (ret) {
10308 btrfs_free_reserved_extent(root, ins.objectid,
10309 ins.offset, 0);
10310 btrfs_abort_transaction(trans, root, ret);
10311 if (own_trans)
10312 btrfs_end_transaction(trans, root);
10313 break;
10314 }
10315
10316 btrfs_drop_extent_cache(inode, cur_offset,
10317 cur_offset + ins.offset -1, 0);
10318
10319 em = alloc_extent_map();
10320 if (!em) {
10321 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10322 &BTRFS_I(inode)->runtime_flags);
10323 goto next;
10324 }
10325
10326 em->start = cur_offset;
10327 em->orig_start = cur_offset;
10328 em->len = ins.offset;
10329 em->block_start = ins.objectid;
10330 em->block_len = ins.offset;
10331 em->orig_block_len = ins.offset;
10332 em->ram_bytes = ins.offset;
10333 em->bdev = root->fs_info->fs_devices->latest_bdev;
10334 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10335 em->generation = trans->transid;
10336
10337 while (1) {
10338 write_lock(&em_tree->lock);
10339 ret = add_extent_mapping(em_tree, em, 1);
10340 write_unlock(&em_tree->lock);
10341 if (ret != -EEXIST)
10342 break;
10343 btrfs_drop_extent_cache(inode, cur_offset,
10344 cur_offset + ins.offset - 1,
10345 0);
10346 }
10347 free_extent_map(em);
10348 next:
10349 num_bytes -= ins.offset;
10350 cur_offset += ins.offset;
10351 *alloc_hint = ins.objectid + ins.offset;
10352
10353 inode_inc_iversion(inode);
10354 inode->i_ctime = current_fs_time(inode->i_sb);
10355 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10356 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10357 (actual_len > inode->i_size) &&
10358 (cur_offset > inode->i_size)) {
10359 if (cur_offset > actual_len)
10360 i_size = actual_len;
10361 else
10362 i_size = cur_offset;
10363 i_size_write(inode, i_size);
10364 btrfs_ordered_update_i_size(inode, i_size, NULL);
10365 }
10366
10367 ret = btrfs_update_inode(trans, root, inode);
10368
10369 if (ret) {
10370 btrfs_abort_transaction(trans, root, ret);
10371 if (own_trans)
10372 btrfs_end_transaction(trans, root);
10373 break;
10374 }
10375
10376 if (own_trans)
10377 btrfs_end_transaction(trans, root);
10378 }
10379 return ret;
10380 }
10381
10382 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10383 u64 start, u64 num_bytes, u64 min_size,
10384 loff_t actual_len, u64 *alloc_hint)
10385 {
10386 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10387 min_size, actual_len, alloc_hint,
10388 NULL);
10389 }
10390
10391 int btrfs_prealloc_file_range_trans(struct inode *inode,
10392 struct btrfs_trans_handle *trans, int mode,
10393 u64 start, u64 num_bytes, u64 min_size,
10394 loff_t actual_len, u64 *alloc_hint)
10395 {
10396 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10397 min_size, actual_len, alloc_hint, trans);
10398 }
10399
10400 static int btrfs_set_page_dirty(struct page *page)
10401 {
10402 return __set_page_dirty_nobuffers(page);
10403 }
10404
10405 static int btrfs_permission(struct inode *inode, int mask)
10406 {
10407 struct btrfs_root *root = BTRFS_I(inode)->root;
10408 umode_t mode = inode->i_mode;
10409
10410 if (mask & MAY_WRITE &&
10411 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10412 if (btrfs_root_readonly(root))
10413 return -EROFS;
10414 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10415 return -EACCES;
10416 }
10417 return generic_permission(inode, mask);
10418 }
10419
10420 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10421 {
10422 struct btrfs_trans_handle *trans;
10423 struct btrfs_root *root = BTRFS_I(dir)->root;
10424 struct inode *inode = NULL;
10425 u64 objectid;
10426 u64 index;
10427 int ret = 0;
10428
10429 /*
10430 * 5 units required for adding orphan entry
10431 */
10432 trans = btrfs_start_transaction(root, 5);
10433 if (IS_ERR(trans))
10434 return PTR_ERR(trans);
10435
10436 ret = btrfs_find_free_ino(root, &objectid);
10437 if (ret)
10438 goto out;
10439
10440 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10441 btrfs_ino(dir), objectid, mode, &index);
10442 if (IS_ERR(inode)) {
10443 ret = PTR_ERR(inode);
10444 inode = NULL;
10445 goto out;
10446 }
10447
10448 inode->i_fop = &btrfs_file_operations;
10449 inode->i_op = &btrfs_file_inode_operations;
10450
10451 inode->i_mapping->a_ops = &btrfs_aops;
10452 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10453
10454 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10455 if (ret)
10456 goto out_inode;
10457
10458 ret = btrfs_update_inode(trans, root, inode);
10459 if (ret)
10460 goto out_inode;
10461 ret = btrfs_orphan_add(trans, inode);
10462 if (ret)
10463 goto out_inode;
10464
10465 /*
10466 * We set number of links to 0 in btrfs_new_inode(), and here we set
10467 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10468 * through:
10469 *
10470 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10471 */
10472 set_nlink(inode, 1);
10473 unlock_new_inode(inode);
10474 d_tmpfile(dentry, inode);
10475 mark_inode_dirty(inode);
10476
10477 out:
10478 btrfs_end_transaction(trans, root);
10479 if (ret)
10480 iput(inode);
10481 btrfs_balance_delayed_items(root);
10482 btrfs_btree_balance_dirty(root);
10483 return ret;
10484
10485 out_inode:
10486 unlock_new_inode(inode);
10487 goto out;
10488
10489 }
10490
10491 /* Inspired by filemap_check_errors() */
10492 int btrfs_inode_check_errors(struct inode *inode)
10493 {
10494 int ret = 0;
10495
10496 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10497 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10498 ret = -ENOSPC;
10499 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10500 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10501 ret = -EIO;
10502
10503 return ret;
10504 }
10505
10506 static const struct inode_operations btrfs_dir_inode_operations = {
10507 .getattr = btrfs_getattr,
10508 .lookup = btrfs_lookup,
10509 .create = btrfs_create,
10510 .unlink = btrfs_unlink,
10511 .link = btrfs_link,
10512 .mkdir = btrfs_mkdir,
10513 .rmdir = btrfs_rmdir,
10514 .rename2 = btrfs_rename2,
10515 .symlink = btrfs_symlink,
10516 .setattr = btrfs_setattr,
10517 .mknod = btrfs_mknod,
10518 .setxattr = generic_setxattr,
10519 .getxattr = generic_getxattr,
10520 .listxattr = btrfs_listxattr,
10521 .removexattr = generic_removexattr,
10522 .permission = btrfs_permission,
10523 .get_acl = btrfs_get_acl,
10524 .set_acl = btrfs_set_acl,
10525 .update_time = btrfs_update_time,
10526 .tmpfile = btrfs_tmpfile,
10527 };
10528 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10529 .lookup = btrfs_lookup,
10530 .permission = btrfs_permission,
10531 .get_acl = btrfs_get_acl,
10532 .set_acl = btrfs_set_acl,
10533 .update_time = btrfs_update_time,
10534 };
10535
10536 static const struct file_operations btrfs_dir_file_operations = {
10537 .llseek = generic_file_llseek,
10538 .read = generic_read_dir,
10539 .iterate_shared = btrfs_real_readdir,
10540 .unlocked_ioctl = btrfs_ioctl,
10541 #ifdef CONFIG_COMPAT
10542 .compat_ioctl = btrfs_compat_ioctl,
10543 #endif
10544 .release = btrfs_release_file,
10545 .fsync = btrfs_sync_file,
10546 };
10547
10548 static const struct extent_io_ops btrfs_extent_io_ops = {
10549 .fill_delalloc = run_delalloc_range,
10550 .submit_bio_hook = btrfs_submit_bio_hook,
10551 .merge_bio_hook = btrfs_merge_bio_hook,
10552 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10553 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10554 .writepage_start_hook = btrfs_writepage_start_hook,
10555 .set_bit_hook = btrfs_set_bit_hook,
10556 .clear_bit_hook = btrfs_clear_bit_hook,
10557 .merge_extent_hook = btrfs_merge_extent_hook,
10558 .split_extent_hook = btrfs_split_extent_hook,
10559 };
10560
10561 /*
10562 * btrfs doesn't support the bmap operation because swapfiles
10563 * use bmap to make a mapping of extents in the file. They assume
10564 * these extents won't change over the life of the file and they
10565 * use the bmap result to do IO directly to the drive.
10566 *
10567 * the btrfs bmap call would return logical addresses that aren't
10568 * suitable for IO and they also will change frequently as COW
10569 * operations happen. So, swapfile + btrfs == corruption.
10570 *
10571 * For now we're avoiding this by dropping bmap.
10572 */
10573 static const struct address_space_operations btrfs_aops = {
10574 .readpage = btrfs_readpage,
10575 .writepage = btrfs_writepage,
10576 .writepages = btrfs_writepages,
10577 .readpages = btrfs_readpages,
10578 .direct_IO = btrfs_direct_IO,
10579 .invalidatepage = btrfs_invalidatepage,
10580 .releasepage = btrfs_releasepage,
10581 .set_page_dirty = btrfs_set_page_dirty,
10582 .error_remove_page = generic_error_remove_page,
10583 };
10584
10585 static const struct address_space_operations btrfs_symlink_aops = {
10586 .readpage = btrfs_readpage,
10587 .writepage = btrfs_writepage,
10588 .invalidatepage = btrfs_invalidatepage,
10589 .releasepage = btrfs_releasepage,
10590 };
10591
10592 static const struct inode_operations btrfs_file_inode_operations = {
10593 .getattr = btrfs_getattr,
10594 .setattr = btrfs_setattr,
10595 .setxattr = generic_setxattr,
10596 .getxattr = generic_getxattr,
10597 .listxattr = btrfs_listxattr,
10598 .removexattr = generic_removexattr,
10599 .permission = btrfs_permission,
10600 .fiemap = btrfs_fiemap,
10601 .get_acl = btrfs_get_acl,
10602 .set_acl = btrfs_set_acl,
10603 .update_time = btrfs_update_time,
10604 };
10605 static const struct inode_operations btrfs_special_inode_operations = {
10606 .getattr = btrfs_getattr,
10607 .setattr = btrfs_setattr,
10608 .permission = btrfs_permission,
10609 .setxattr = generic_setxattr,
10610 .getxattr = generic_getxattr,
10611 .listxattr = btrfs_listxattr,
10612 .removexattr = generic_removexattr,
10613 .get_acl = btrfs_get_acl,
10614 .set_acl = btrfs_set_acl,
10615 .update_time = btrfs_update_time,
10616 };
10617 static const struct inode_operations btrfs_symlink_inode_operations = {
10618 .readlink = generic_readlink,
10619 .get_link = page_get_link,
10620 .getattr = btrfs_getattr,
10621 .setattr = btrfs_setattr,
10622 .permission = btrfs_permission,
10623 .setxattr = generic_setxattr,
10624 .getxattr = generic_getxattr,
10625 .listxattr = btrfs_listxattr,
10626 .removexattr = generic_removexattr,
10627 .update_time = btrfs_update_time,
10628 };
10629
10630 const struct dentry_operations btrfs_dentry_operations = {
10631 .d_delete = btrfs_dentry_delete,
10632 .d_release = btrfs_dentry_release,
10633 };
ubuntu-zesty-kernel mirror