1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/compat.h>
34 #include <linux/bit_spinlock.h>
35 #include <linux/xattr.h>
36 #include <linux/posix_acl.h>
37 #include <linux/falloc.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/mount.h>
41 #include <linux/btrfs.h>
42 #include <linux/blkdev.h>
43 #include <linux/posix_acl_xattr.h>
44 #include <linux/uio.h>
45 #include "ctree.h"
46 #include "disk-io.h"
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "print-tree.h"
50 #include "ordered-data.h"
51 #include "xattr.h"
52 #include "tree-log.h"
53 #include "volumes.h"
54 #include "compression.h"
55 #include "locking.h"
56 #include "free-space-cache.h"
57 #include "inode-map.h"
58 #include "backref.h"
59 #include "hash.h"
60 #include "props.h"
61 #include "qgroup.h"
62 #include "dedupe.h"
63
64 struct btrfs_iget_args {
65 struct btrfs_key *location;
66 struct btrfs_root *root;
67 };
68
69 struct btrfs_dio_data {
70 u64 outstanding_extents;
71 u64 reserve;
72 u64 unsubmitted_oe_range_start;
73 u64 unsubmitted_oe_range_end;
74 };
75
76 static const struct inode_operations btrfs_dir_inode_operations;
77 static const struct inode_operations btrfs_symlink_inode_operations;
78 static const struct inode_operations btrfs_dir_ro_inode_operations;
79 static const struct inode_operations btrfs_special_inode_operations;
80 static const struct inode_operations btrfs_file_inode_operations;
81 static const struct address_space_operations btrfs_aops;
82 static const struct address_space_operations btrfs_symlink_aops;
83 static const struct file_operations btrfs_dir_file_operations;
84 static const struct extent_io_ops btrfs_extent_io_ops;
85
86 static struct kmem_cache *btrfs_inode_cachep;
87 struct kmem_cache *btrfs_trans_handle_cachep;
88 struct kmem_cache *btrfs_transaction_cachep;
89 struct kmem_cache *btrfs_path_cachep;
90 struct kmem_cache *btrfs_free_space_cachep;
91
92 #define S_SHIFT 12
93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
94 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
95 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
96 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
97 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
98 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
99 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
100 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
101 };
102
103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
104 static int btrfs_truncate(struct inode *inode);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
106 static noinline int cow_file_range(struct inode *inode,
107 struct page *locked_page,
108 u64 start, u64 end, u64 delalloc_end,
109 int *page_started, unsigned long *nr_written,
110 int unlock, struct btrfs_dedupe_hash *hash);
111 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
112 u64 len, u64 orig_start,
113 u64 block_start, u64 block_len,
114 u64 orig_block_len, u64 ram_bytes,
115 int type);
116
117 static int btrfs_dirty_inode(struct inode *inode);
118
119 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
120 void btrfs_test_inode_set_ops(struct inode *inode)
121 {
122 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
123 }
124 #endif
125
126 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
127 struct inode *inode, struct inode *dir,
128 const struct qstr *qstr)
129 {
130 int err;
131
132 err = btrfs_init_acl(trans, inode, dir);
133 if (!err)
134 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
135 return err;
136 }
137
138 /*
139 * this does all the hard work for inserting an inline extent into
140 * the btree. The caller should have done a btrfs_drop_extents so that
141 * no overlapping inline items exist in the btree
142 */
143 static int insert_inline_extent(struct btrfs_trans_handle *trans,
144 struct btrfs_path *path, int extent_inserted,
145 struct btrfs_root *root, struct inode *inode,
146 u64 start, size_t size, size_t compressed_size,
147 int compress_type,
148 struct page **compressed_pages)
149 {
150 struct extent_buffer *leaf;
151 struct page *page = NULL;
152 char *kaddr;
153 unsigned long ptr;
154 struct btrfs_file_extent_item *ei;
155 int err = 0;
156 int ret;
157 size_t cur_size = size;
158 unsigned long offset;
159
160 if (compressed_size && compressed_pages)
161 cur_size = compressed_size;
162
163 inode_add_bytes(inode, size);
164
165 if (!extent_inserted) {
166 struct btrfs_key key;
167 size_t datasize;
168
169 key.objectid = btrfs_ino(inode);
170 key.offset = start;
171 key.type = BTRFS_EXTENT_DATA_KEY;
172
173 datasize = btrfs_file_extent_calc_inline_size(cur_size);
174 path->leave_spinning = 1;
175 ret = btrfs_insert_empty_item(trans, root, path, &key,
176 datasize);
177 if (ret) {
178 err = ret;
179 goto fail;
180 }
181 }
182 leaf = path->nodes[0];
183 ei = btrfs_item_ptr(leaf, path->slots[0],
184 struct btrfs_file_extent_item);
185 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
186 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
187 btrfs_set_file_extent_encryption(leaf, ei, 0);
188 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
189 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
190 ptr = btrfs_file_extent_inline_start(ei);
191
192 if (compress_type != BTRFS_COMPRESS_NONE) {
193 struct page *cpage;
194 int i = 0;
195 while (compressed_size > 0) {
196 cpage = compressed_pages[i];
197 cur_size = min_t(unsigned long, compressed_size,
198 PAGE_SIZE);
199
200 kaddr = kmap_atomic(cpage);
201 write_extent_buffer(leaf, kaddr, ptr, cur_size);
202 kunmap_atomic(kaddr);
203
204 i++;
205 ptr += cur_size;
206 compressed_size -= cur_size;
207 }
208 btrfs_set_file_extent_compression(leaf, ei,
209 compress_type);
210 } else {
211 page = find_get_page(inode->i_mapping,
212 start >> PAGE_SHIFT);
213 btrfs_set_file_extent_compression(leaf, ei, 0);
214 kaddr = kmap_atomic(page);
215 offset = start & (PAGE_SIZE - 1);
216 write_extent_buffer(leaf, kaddr + offset, ptr, size);
217 kunmap_atomic(kaddr);
218 put_page(page);
219 }
220 btrfs_mark_buffer_dirty(leaf);
221 btrfs_release_path(path);
222
223 /*
224 * we're an inline extent, so nobody can
225 * extend the file past i_size without locking
226 * a page we already have locked.
227 *
228 * We must do any isize and inode updates
229 * before we unlock the pages. Otherwise we
230 * could end up racing with unlink.
231 */
232 BTRFS_I(inode)->disk_i_size = inode->i_size;
233 ret = btrfs_update_inode(trans, root, inode);
234
235 return ret;
236 fail:
237 return err;
238 }
239
240
241 /*
242 * conditionally insert an inline extent into the file. This
243 * does the checks required to make sure the data is small enough
244 * to fit as an inline extent.
245 */
246 static noinline int cow_file_range_inline(struct btrfs_root *root,
247 struct inode *inode, u64 start,
248 u64 end, size_t compressed_size,
249 int compress_type,
250 struct page **compressed_pages)
251 {
252 struct btrfs_fs_info *fs_info = root->fs_info;
253 struct btrfs_trans_handle *trans;
254 u64 isize = i_size_read(inode);
255 u64 actual_end = min(end + 1, isize);
256 u64 inline_len = actual_end - start;
257 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
258 u64 data_len = inline_len;
259 int ret;
260 struct btrfs_path *path;
261 int extent_inserted = 0;
262 u32 extent_item_size;
263
264 if (compressed_size)
265 data_len = compressed_size;
266
267 if (start > 0 ||
268 actual_end > fs_info->sectorsize ||
269 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
270 (!compressed_size &&
271 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
272 end + 1 < isize ||
273 data_len > fs_info->max_inline) {
274 return 1;
275 }
276
277 path = btrfs_alloc_path();
278 if (!path)
279 return -ENOMEM;
280
281 trans = btrfs_join_transaction(root);
282 if (IS_ERR(trans)) {
283 btrfs_free_path(path);
284 return PTR_ERR(trans);
285 }
286 trans->block_rsv = &fs_info->delalloc_block_rsv;
287
288 if (compressed_size && compressed_pages)
289 extent_item_size = btrfs_file_extent_calc_inline_size(
290 compressed_size);
291 else
292 extent_item_size = btrfs_file_extent_calc_inline_size(
293 inline_len);
294
295 ret = __btrfs_drop_extents(trans, root, inode, path,
296 start, aligned_end, NULL,
297 1, 1, extent_item_size, &extent_inserted);
298 if (ret) {
299 btrfs_abort_transaction(trans, ret);
300 goto out;
301 }
302
303 if (isize > actual_end)
304 inline_len = min_t(u64, isize, actual_end);
305 ret = insert_inline_extent(trans, path, extent_inserted,
306 root, inode, start,
307 inline_len, compressed_size,
308 compress_type, compressed_pages);
309 if (ret && ret != -ENOSPC) {
310 btrfs_abort_transaction(trans, ret);
311 goto out;
312 } else if (ret == -ENOSPC) {
313 ret = 1;
314 goto out;
315 }
316
317 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
318 btrfs_delalloc_release_metadata(inode, end + 1 - start);
319 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
320 out:
321 /*
322 * Don't forget to free the reserved space, as for inlined extent
323 * it won't count as data extent, free them directly here.
324 * And at reserve time, it's always aligned to page size, so
325 * just free one page here.
326 */
327 btrfs_qgroup_free_data(inode, 0, PAGE_SIZE);
328 btrfs_free_path(path);
329 btrfs_end_transaction(trans);
330 return ret;
331 }
332
333 struct async_extent {
334 u64 start;
335 u64 ram_size;
336 u64 compressed_size;
337 struct page **pages;
338 unsigned long nr_pages;
339 int compress_type;
340 struct list_head list;
341 };
342
343 struct async_cow {
344 struct inode *inode;
345 struct btrfs_root *root;
346 struct page *locked_page;
347 u64 start;
348 u64 end;
349 struct list_head extents;
350 struct btrfs_work work;
351 };
352
353 static noinline int add_async_extent(struct async_cow *cow,
354 u64 start, u64 ram_size,
355 u64 compressed_size,
356 struct page **pages,
357 unsigned long nr_pages,
358 int compress_type)
359 {
360 struct async_extent *async_extent;
361
362 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
363 BUG_ON(!async_extent); /* -ENOMEM */
364 async_extent->start = start;
365 async_extent->ram_size = ram_size;
366 async_extent->compressed_size = compressed_size;
367 async_extent->pages = pages;
368 async_extent->nr_pages = nr_pages;
369 async_extent->compress_type = compress_type;
370 list_add_tail(&async_extent->list, &cow->extents);
371 return 0;
372 }
373
374 static inline int inode_need_compress(struct inode *inode)
375 {
376 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
377
378 /* force compress */
379 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
380 return 1;
381 /* bad compression ratios */
382 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
383 return 0;
384 if (btrfs_test_opt(fs_info, COMPRESS) ||
385 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
386 BTRFS_I(inode)->force_compress)
387 return 1;
388 return 0;
389 }
390
391 /*
392 * we create compressed extents in two phases. The first
393 * phase compresses a range of pages that have already been
394 * locked (both pages and state bits are locked).
395 *
396 * This is done inside an ordered work queue, and the compression
397 * is spread across many cpus. The actual IO submission is step
398 * two, and the ordered work queue takes care of making sure that
399 * happens in the same order things were put onto the queue by
400 * writepages and friends.
401 *
402 * If this code finds it can't get good compression, it puts an
403 * entry onto the work queue to write the uncompressed bytes. This
404 * makes sure that both compressed inodes and uncompressed inodes
405 * are written in the same order that the flusher thread sent them
406 * down.
407 */
408 static noinline void compress_file_range(struct inode *inode,
409 struct page *locked_page,
410 u64 start, u64 end,
411 struct async_cow *async_cow,
412 int *num_added)
413 {
414 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
415 struct btrfs_root *root = BTRFS_I(inode)->root;
416 u64 num_bytes;
417 u64 blocksize = fs_info->sectorsize;
418 u64 actual_end;
419 u64 isize = i_size_read(inode);
420 int ret = 0;
421 struct page **pages = NULL;
422 unsigned long nr_pages;
423 unsigned long nr_pages_ret = 0;
424 unsigned long total_compressed = 0;
425 unsigned long total_in = 0;
426 unsigned long max_compressed = SZ_128K;
427 unsigned long max_uncompressed = SZ_128K;
428 int i;
429 int will_compress;
430 int compress_type = fs_info->compress_type;
431 int redirty = 0;
432
433 /* if this is a small write inside eof, kick off a defrag */
434 if ((end - start + 1) < SZ_16K &&
435 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
436 btrfs_add_inode_defrag(NULL, inode);
437
438 actual_end = min_t(u64, isize, end + 1);
439 again:
440 will_compress = 0;
441 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
442 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_SIZE);
443
444 /*
445 * we don't want to send crud past the end of i_size through
446 * compression, that's just a waste of CPU time. So, if the
447 * end of the file is before the start of our current
448 * requested range of bytes, we bail out to the uncompressed
449 * cleanup code that can deal with all of this.
450 *
451 * It isn't really the fastest way to fix things, but this is a
452 * very uncommon corner.
453 */
454 if (actual_end <= start)
455 goto cleanup_and_bail_uncompressed;
456
457 total_compressed = actual_end - start;
458
459 /*
460 * skip compression for a small file range(<=blocksize) that
461 * isn't an inline extent, since it doesn't save disk space at all.
462 */
463 if (total_compressed <= blocksize &&
464 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
465 goto cleanup_and_bail_uncompressed;
466
467 /* we want to make sure that amount of ram required to uncompress
468 * an extent is reasonable, so we limit the total size in ram
469 * of a compressed extent to 128k. This is a crucial number
470 * because it also controls how easily we can spread reads across
471 * cpus for decompression.
472 *
473 * We also want to make sure the amount of IO required to do
474 * a random read is reasonably small, so we limit the size of
475 * a compressed extent to 128k.
476 */
477 total_compressed = min(total_compressed, max_uncompressed);
478 num_bytes = ALIGN(end - start + 1, blocksize);
479 num_bytes = max(blocksize, num_bytes);
480 total_in = 0;
481 ret = 0;
482
483 /*
484 * we do compression for mount -o compress and when the
485 * inode has not been flagged as nocompress. This flag can
486 * change at any time if we discover bad compression ratios.
487 */
488 if (inode_need_compress(inode)) {
489 WARN_ON(pages);
490 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
491 if (!pages) {
492 /* just bail out to the uncompressed code */
493 goto cont;
494 }
495
496 if (BTRFS_I(inode)->force_compress)
497 compress_type = BTRFS_I(inode)->force_compress;
498
499 /*
500 * we need to call clear_page_dirty_for_io on each
501 * page in the range. Otherwise applications with the file
502 * mmap'd can wander in and change the page contents while
503 * we are compressing them.
504 *
505 * If the compression fails for any reason, we set the pages
506 * dirty again later on.
507 */
508 extent_range_clear_dirty_for_io(inode, start, end);
509 redirty = 1;
510 ret = btrfs_compress_pages(compress_type,
511 inode->i_mapping, start,
512 total_compressed, pages,
513 nr_pages, &nr_pages_ret,
514 &total_in,
515 &total_compressed,
516 max_compressed);
517
518 if (!ret) {
519 unsigned long offset = total_compressed &
520 (PAGE_SIZE - 1);
521 struct page *page = pages[nr_pages_ret - 1];
522 char *kaddr;
523
524 /* zero the tail end of the last page, we might be
525 * sending it down to disk
526 */
527 if (offset) {
528 kaddr = kmap_atomic(page);
529 memset(kaddr + offset, 0,
530 PAGE_SIZE - offset);
531 kunmap_atomic(kaddr);
532 }
533 will_compress = 1;
534 }
535 }
536 cont:
537 if (start == 0) {
538 /* lets try to make an inline extent */
539 if (ret || total_in < (actual_end - start)) {
540 /* we didn't compress the entire range, try
541 * to make an uncompressed inline extent.
542 */
543 ret = cow_file_range_inline(root, inode, start, end,
544 0, 0, NULL);
545 } else {
546 /* try making a compressed inline extent */
547 ret = cow_file_range_inline(root, inode, start, end,
548 total_compressed,
549 compress_type, pages);
550 }
551 if (ret <= 0) {
552 unsigned long clear_flags = EXTENT_DELALLOC |
553 EXTENT_DEFRAG;
554 unsigned long page_error_op;
555
556 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
557 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
558
559 /*
560 * inline extent creation worked or returned error,
561 * we don't need to create any more async work items.
562 * Unlock and free up our temp pages.
563 */
564 extent_clear_unlock_delalloc(inode, start, end, end,
565 NULL, clear_flags,
566 PAGE_UNLOCK |
567 PAGE_CLEAR_DIRTY |
568 PAGE_SET_WRITEBACK |
569 page_error_op |
570 PAGE_END_WRITEBACK);
571 btrfs_free_reserved_data_space_noquota(inode, start,
572 end - start + 1);
573 goto free_pages_out;
574 }
575 }
576
577 if (will_compress) {
578 /*
579 * we aren't doing an inline extent round the compressed size
580 * up to a block size boundary so the allocator does sane
581 * things
582 */
583 total_compressed = ALIGN(total_compressed, blocksize);
584
585 /*
586 * one last check to make sure the compression is really a
587 * win, compare the page count read with the blocks on disk
588 */
589 total_in = ALIGN(total_in, PAGE_SIZE);
590 if (total_compressed >= total_in) {
591 will_compress = 0;
592 } else {
593 num_bytes = total_in;
594 *num_added += 1;
595
596 /*
597 * The async work queues will take care of doing actual
598 * allocation on disk for these compressed pages, and
599 * will submit them to the elevator.
600 */
601 add_async_extent(async_cow, start, num_bytes,
602 total_compressed, pages, nr_pages_ret,
603 compress_type);
604
605 if (start + num_bytes < end) {
606 start += num_bytes;
607 pages = NULL;
608 cond_resched();
609 goto again;
610 }
611 return;
612 }
613 }
614 if (pages) {
615 /*
616 * the compression code ran but failed to make things smaller,
617 * free any pages it allocated and our page pointer array
618 */
619 for (i = 0; i < nr_pages_ret; i++) {
620 WARN_ON(pages[i]->mapping);
621 put_page(pages[i]);
622 }
623 kfree(pages);
624 pages = NULL;
625 total_compressed = 0;
626 nr_pages_ret = 0;
627
628 /* flag the file so we don't compress in the future */
629 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
630 !(BTRFS_I(inode)->force_compress)) {
631 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
632 }
633 }
634 cleanup_and_bail_uncompressed:
635 /*
636 * No compression, but we still need to write the pages in the file
637 * we've been given so far. redirty the locked page if it corresponds
638 * to our extent and set things up for the async work queue to run
639 * cow_file_range to do the normal delalloc dance.
640 */
641 if (page_offset(locked_page) >= start &&
642 page_offset(locked_page) <= end)
643 __set_page_dirty_nobuffers(locked_page);
644 /* unlocked later on in the async handlers */
645
646 if (redirty)
647 extent_range_redirty_for_io(inode, start, end);
648 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
649 BTRFS_COMPRESS_NONE);
650 *num_added += 1;
651
652 return;
653
654 free_pages_out:
655 for (i = 0; i < nr_pages_ret; i++) {
656 WARN_ON(pages[i]->mapping);
657 put_page(pages[i]);
658 }
659 kfree(pages);
660 }
661
662 static void free_async_extent_pages(struct async_extent *async_extent)
663 {
664 int i;
665
666 if (!async_extent->pages)
667 return;
668
669 for (i = 0; i < async_extent->nr_pages; i++) {
670 WARN_ON(async_extent->pages[i]->mapping);
671 put_page(async_extent->pages[i]);
672 }
673 kfree(async_extent->pages);
674 async_extent->nr_pages = 0;
675 async_extent->pages = NULL;
676 }
677
678 /*
679 * phase two of compressed writeback. This is the ordered portion
680 * of the code, which only gets called in the order the work was
681 * queued. We walk all the async extents created by compress_file_range
682 * and send them down to the disk.
683 */
684 static noinline void submit_compressed_extents(struct inode *inode,
685 struct async_cow *async_cow)
686 {
687 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
688 struct async_extent *async_extent;
689 u64 alloc_hint = 0;
690 struct btrfs_key ins;
691 struct extent_map *em;
692 struct btrfs_root *root = BTRFS_I(inode)->root;
693 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
694 struct extent_io_tree *io_tree;
695 int ret = 0;
696
697 again:
698 while (!list_empty(&async_cow->extents)) {
699 async_extent = list_entry(async_cow->extents.next,
700 struct async_extent, list);
701 list_del(&async_extent->list);
702
703 io_tree = &BTRFS_I(inode)->io_tree;
704
705 retry:
706 /* did the compression code fall back to uncompressed IO? */
707 if (!async_extent->pages) {
708 int page_started = 0;
709 unsigned long nr_written = 0;
710
711 lock_extent(io_tree, async_extent->start,
712 async_extent->start +
713 async_extent->ram_size - 1);
714
715 /* allocate blocks */
716 ret = cow_file_range(inode, async_cow->locked_page,
717 async_extent->start,
718 async_extent->start +
719 async_extent->ram_size - 1,
720 async_extent->start +
721 async_extent->ram_size - 1,
722 &page_started, &nr_written, 0,
723 NULL);
724
725 /* JDM XXX */
726
727 /*
728 * if page_started, cow_file_range inserted an
729 * inline extent and took care of all the unlocking
730 * and IO for us. Otherwise, we need to submit
731 * all those pages down to the drive.
732 */
733 if (!page_started && !ret)
734 extent_write_locked_range(io_tree,
735 inode, async_extent->start,
736 async_extent->start +
737 async_extent->ram_size - 1,
738 btrfs_get_extent,
739 WB_SYNC_ALL);
740 else if (ret)
741 unlock_page(async_cow->locked_page);
742 kfree(async_extent);
743 cond_resched();
744 continue;
745 }
746
747 lock_extent(io_tree, async_extent->start,
748 async_extent->start + async_extent->ram_size - 1);
749
750 ret = btrfs_reserve_extent(root, async_extent->ram_size,
751 async_extent->compressed_size,
752 async_extent->compressed_size,
753 0, alloc_hint, &ins, 1, 1);
754 if (ret) {
755 free_async_extent_pages(async_extent);
756
757 if (ret == -ENOSPC) {
758 unlock_extent(io_tree, async_extent->start,
759 async_extent->start +
760 async_extent->ram_size - 1);
761
762 /*
763 * we need to redirty the pages if we decide to
764 * fallback to uncompressed IO, otherwise we
765 * will not submit these pages down to lower
766 * layers.
767 */
768 extent_range_redirty_for_io(inode,
769 async_extent->start,
770 async_extent->start +
771 async_extent->ram_size - 1);
772
773 goto retry;
774 }
775 goto out_free;
776 }
777 /*
778 * here we're doing allocation and writeback of the
779 * compressed pages
780 */
781 btrfs_drop_extent_cache(inode, async_extent->start,
782 async_extent->start +
783 async_extent->ram_size - 1, 0);
784
785 em = alloc_extent_map();
786 if (!em) {
787 ret = -ENOMEM;
788 goto out_free_reserve;
789 }
790 em->start = async_extent->start;
791 em->len = async_extent->ram_size;
792 em->orig_start = em->start;
793 em->mod_start = em->start;
794 em->mod_len = em->len;
795
796 em->block_start = ins.objectid;
797 em->block_len = ins.offset;
798 em->orig_block_len = ins.offset;
799 em->ram_bytes = async_extent->ram_size;
800 em->bdev = fs_info->fs_devices->latest_bdev;
801 em->compress_type = async_extent->compress_type;
802 set_bit(EXTENT_FLAG_PINNED, &em->flags);
803 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
804 em->generation = -1;
805
806 while (1) {
807 write_lock(&em_tree->lock);
808 ret = add_extent_mapping(em_tree, em, 1);
809 write_unlock(&em_tree->lock);
810 if (ret != -EEXIST) {
811 free_extent_map(em);
812 break;
813 }
814 btrfs_drop_extent_cache(inode, async_extent->start,
815 async_extent->start +
816 async_extent->ram_size - 1, 0);
817 }
818
819 if (ret)
820 goto out_free_reserve;
821
822 ret = btrfs_add_ordered_extent_compress(inode,
823 async_extent->start,
824 ins.objectid,
825 async_extent->ram_size,
826 ins.offset,
827 BTRFS_ORDERED_COMPRESSED,
828 async_extent->compress_type);
829 if (ret) {
830 btrfs_drop_extent_cache(inode, async_extent->start,
831 async_extent->start +
832 async_extent->ram_size - 1, 0);
833 goto out_free_reserve;
834 }
835 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
836
837 /*
838 * clear dirty, set writeback and unlock the pages.
839 */
840 extent_clear_unlock_delalloc(inode, async_extent->start,
841 async_extent->start +
842 async_extent->ram_size - 1,
843 async_extent->start +
844 async_extent->ram_size - 1,
845 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
846 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
847 PAGE_SET_WRITEBACK);
848 ret = btrfs_submit_compressed_write(inode,
849 async_extent->start,
850 async_extent->ram_size,
851 ins.objectid,
852 ins.offset, async_extent->pages,
853 async_extent->nr_pages);
854 if (ret) {
855 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
856 struct page *p = async_extent->pages[0];
857 const u64 start = async_extent->start;
858 const u64 end = start + async_extent->ram_size - 1;
859
860 p->mapping = inode->i_mapping;
861 tree->ops->writepage_end_io_hook(p, start, end,
862 NULL, 0);
863 p->mapping = NULL;
864 extent_clear_unlock_delalloc(inode, start, end, end,
865 NULL, 0,
866 PAGE_END_WRITEBACK |
867 PAGE_SET_ERROR);
868 free_async_extent_pages(async_extent);
869 }
870 alloc_hint = ins.objectid + ins.offset;
871 kfree(async_extent);
872 cond_resched();
873 }
874 return;
875 out_free_reserve:
876 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
877 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
878 out_free:
879 extent_clear_unlock_delalloc(inode, async_extent->start,
880 async_extent->start +
881 async_extent->ram_size - 1,
882 async_extent->start +
883 async_extent->ram_size - 1,
884 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
885 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
886 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
887 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
888 PAGE_SET_ERROR);
889 free_async_extent_pages(async_extent);
890 kfree(async_extent);
891 goto again;
892 }
893
894 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
895 u64 num_bytes)
896 {
897 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
898 struct extent_map *em;
899 u64 alloc_hint = 0;
900
901 read_lock(&em_tree->lock);
902 em = search_extent_mapping(em_tree, start, num_bytes);
903 if (em) {
904 /*
905 * if block start isn't an actual block number then find the
906 * first block in this inode and use that as a hint. If that
907 * block is also bogus then just don't worry about it.
908 */
909 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
910 free_extent_map(em);
911 em = search_extent_mapping(em_tree, 0, 0);
912 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
913 alloc_hint = em->block_start;
914 if (em)
915 free_extent_map(em);
916 } else {
917 alloc_hint = em->block_start;
918 free_extent_map(em);
919 }
920 }
921 read_unlock(&em_tree->lock);
922
923 return alloc_hint;
924 }
925
926 /*
927 * when extent_io.c finds a delayed allocation range in the file,
928 * the call backs end up in this code. The basic idea is to
929 * allocate extents on disk for the range, and create ordered data structs
930 * in ram to track those extents.
931 *
932 * locked_page is the page that writepage had locked already. We use
933 * it to make sure we don't do extra locks or unlocks.
934 *
935 * *page_started is set to one if we unlock locked_page and do everything
936 * required to start IO on it. It may be clean and already done with
937 * IO when we return.
938 */
939 static noinline int cow_file_range(struct inode *inode,
940 struct page *locked_page,
941 u64 start, u64 end, u64 delalloc_end,
942 int *page_started, unsigned long *nr_written,
943 int unlock, struct btrfs_dedupe_hash *hash)
944 {
945 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
946 struct btrfs_root *root = BTRFS_I(inode)->root;
947 u64 alloc_hint = 0;
948 u64 num_bytes;
949 unsigned long ram_size;
950 u64 disk_num_bytes;
951 u64 cur_alloc_size;
952 u64 blocksize = fs_info->sectorsize;
953 struct btrfs_key ins;
954 struct extent_map *em;
955 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
956 int ret = 0;
957
958 if (btrfs_is_free_space_inode(inode)) {
959 WARN_ON_ONCE(1);
960 ret = -EINVAL;
961 goto out_unlock;
962 }
963
964 num_bytes = ALIGN(end - start + 1, blocksize);
965 num_bytes = max(blocksize, num_bytes);
966 disk_num_bytes = num_bytes;
967
968 /* if this is a small write inside eof, kick off defrag */
969 if (num_bytes < SZ_64K &&
970 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
971 btrfs_add_inode_defrag(NULL, inode);
972
973 if (start == 0) {
974 /* lets try to make an inline extent */
975 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
976 NULL);
977 if (ret == 0) {
978 extent_clear_unlock_delalloc(inode, start, end,
979 delalloc_end, NULL,
980 EXTENT_LOCKED | EXTENT_DELALLOC |
981 EXTENT_DEFRAG, PAGE_UNLOCK |
982 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
983 PAGE_END_WRITEBACK);
984 btrfs_free_reserved_data_space_noquota(inode, start,
985 end - start + 1);
986 *nr_written = *nr_written +
987 (end - start + PAGE_SIZE) / PAGE_SIZE;
988 *page_started = 1;
989 goto out;
990 } else if (ret < 0) {
991 goto out_unlock;
992 }
993 }
994
995 BUG_ON(disk_num_bytes >
996 btrfs_super_total_bytes(fs_info->super_copy));
997
998 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
999 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1000
1001 while (disk_num_bytes > 0) {
1002 unsigned long op;
1003
1004 cur_alloc_size = disk_num_bytes;
1005 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1006 fs_info->sectorsize, 0, alloc_hint,
1007 &ins, 1, 1);
1008 if (ret < 0)
1009 goto out_unlock;
1010
1011 em = alloc_extent_map();
1012 if (!em) {
1013 ret = -ENOMEM;
1014 goto out_reserve;
1015 }
1016 em->start = start;
1017 em->orig_start = em->start;
1018 ram_size = ins.offset;
1019 em->len = ins.offset;
1020 em->mod_start = em->start;
1021 em->mod_len = em->len;
1022
1023 em->block_start = ins.objectid;
1024 em->block_len = ins.offset;
1025 em->orig_block_len = ins.offset;
1026 em->ram_bytes = ram_size;
1027 em->bdev = fs_info->fs_devices->latest_bdev;
1028 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1029 em->generation = -1;
1030
1031 while (1) {
1032 write_lock(&em_tree->lock);
1033 ret = add_extent_mapping(em_tree, em, 1);
1034 write_unlock(&em_tree->lock);
1035 if (ret != -EEXIST) {
1036 free_extent_map(em);
1037 break;
1038 }
1039 btrfs_drop_extent_cache(inode, start,
1040 start + ram_size - 1, 0);
1041 }
1042 if (ret)
1043 goto out_reserve;
1044
1045 cur_alloc_size = ins.offset;
1046 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1047 ram_size, cur_alloc_size, 0);
1048 if (ret)
1049 goto out_drop_extent_cache;
1050
1051 if (root->root_key.objectid ==
1052 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1053 ret = btrfs_reloc_clone_csums(inode, start,
1054 cur_alloc_size);
1055 if (ret)
1056 goto out_drop_extent_cache;
1057 }
1058
1059 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1060
1061 if (disk_num_bytes < cur_alloc_size)
1062 break;
1063
1064 /* we're not doing compressed IO, don't unlock the first
1065 * page (which the caller expects to stay locked), don't
1066 * clear any dirty bits and don't set any writeback bits
1067 *
1068 * Do set the Private2 bit so we know this page was properly
1069 * setup for writepage
1070 */
1071 op = unlock ? PAGE_UNLOCK : 0;
1072 op |= PAGE_SET_PRIVATE2;
1073
1074 extent_clear_unlock_delalloc(inode, start,
1075 start + ram_size - 1,
1076 delalloc_end, locked_page,
1077 EXTENT_LOCKED | EXTENT_DELALLOC,
1078 op);
1079 disk_num_bytes -= cur_alloc_size;
1080 num_bytes -= cur_alloc_size;
1081 alloc_hint = ins.objectid + ins.offset;
1082 start += cur_alloc_size;
1083 }
1084 out:
1085 return ret;
1086
1087 out_drop_extent_cache:
1088 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1089 out_reserve:
1090 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1091 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1092 out_unlock:
1093 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1094 locked_page,
1095 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1096 EXTENT_DELALLOC | EXTENT_DEFRAG,
1097 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1098 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1099 goto out;
1100 }
1101
1102 /*
1103 * work queue call back to started compression on a file and pages
1104 */
1105 static noinline void async_cow_start(struct btrfs_work *work)
1106 {
1107 struct async_cow *async_cow;
1108 int num_added = 0;
1109 async_cow = container_of(work, struct async_cow, work);
1110
1111 compress_file_range(async_cow->inode, async_cow->locked_page,
1112 async_cow->start, async_cow->end, async_cow,
1113 &num_added);
1114 if (num_added == 0) {
1115 btrfs_add_delayed_iput(async_cow->inode);
1116 async_cow->inode = NULL;
1117 }
1118 }
1119
1120 /*
1121 * work queue call back to submit previously compressed pages
1122 */
1123 static noinline void async_cow_submit(struct btrfs_work *work)
1124 {
1125 struct btrfs_fs_info *fs_info;
1126 struct async_cow *async_cow;
1127 struct btrfs_root *root;
1128 unsigned long nr_pages;
1129
1130 async_cow = container_of(work, struct async_cow, work);
1131
1132 root = async_cow->root;
1133 fs_info = root->fs_info;
1134 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1135 PAGE_SHIFT;
1136
1137 /*
1138 * atomic_sub_return implies a barrier for waitqueue_active
1139 */
1140 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1141 5 * SZ_1M &&
1142 waitqueue_active(&fs_info->async_submit_wait))
1143 wake_up(&fs_info->async_submit_wait);
1144
1145 if (async_cow->inode)
1146 submit_compressed_extents(async_cow->inode, async_cow);
1147 }
1148
1149 static noinline void async_cow_free(struct btrfs_work *work)
1150 {
1151 struct async_cow *async_cow;
1152 async_cow = container_of(work, struct async_cow, work);
1153 if (async_cow->inode)
1154 btrfs_add_delayed_iput(async_cow->inode);
1155 kfree(async_cow);
1156 }
1157
1158 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1159 u64 start, u64 end, int *page_started,
1160 unsigned long *nr_written)
1161 {
1162 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1163 struct async_cow *async_cow;
1164 struct btrfs_root *root = BTRFS_I(inode)->root;
1165 unsigned long nr_pages;
1166 u64 cur_end;
1167 int limit = 10 * SZ_1M;
1168
1169 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1170 1, 0, NULL, GFP_NOFS);
1171 while (start < end) {
1172 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1173 BUG_ON(!async_cow); /* -ENOMEM */
1174 async_cow->inode = igrab(inode);
1175 async_cow->root = root;
1176 async_cow->locked_page = locked_page;
1177 async_cow->start = start;
1178
1179 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1180 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1181 cur_end = end;
1182 else
1183 cur_end = min(end, start + SZ_512K - 1);
1184
1185 async_cow->end = cur_end;
1186 INIT_LIST_HEAD(&async_cow->extents);
1187
1188 btrfs_init_work(&async_cow->work,
1189 btrfs_delalloc_helper,
1190 async_cow_start, async_cow_submit,
1191 async_cow_free);
1192
1193 nr_pages = (cur_end - start + PAGE_SIZE) >>
1194 PAGE_SHIFT;
1195 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1196
1197 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1198
1199 if (atomic_read(&fs_info->async_delalloc_pages) > limit) {
1200 wait_event(fs_info->async_submit_wait,
1201 (atomic_read(&fs_info->async_delalloc_pages) <
1202 limit));
1203 }
1204
1205 while (atomic_read(&fs_info->async_submit_draining) &&
1206 atomic_read(&fs_info->async_delalloc_pages)) {
1207 wait_event(fs_info->async_submit_wait,
1208 (atomic_read(&fs_info->async_delalloc_pages) ==
1209 0));
1210 }
1211
1212 *nr_written += nr_pages;
1213 start = cur_end + 1;
1214 }
1215 *page_started = 1;
1216 return 0;
1217 }
1218
1219 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1220 u64 bytenr, u64 num_bytes)
1221 {
1222 int ret;
1223 struct btrfs_ordered_sum *sums;
1224 LIST_HEAD(list);
1225
1226 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1227 bytenr + num_bytes - 1, &list, 0);
1228 if (ret == 0 && list_empty(&list))
1229 return 0;
1230
1231 while (!list_empty(&list)) {
1232 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1233 list_del(&sums->list);
1234 kfree(sums);
1235 }
1236 return 1;
1237 }
1238
1239 /*
1240 * when nowcow writeback call back. This checks for snapshots or COW copies
1241 * of the extents that exist in the file, and COWs the file as required.
1242 *
1243 * If no cow copies or snapshots exist, we write directly to the existing
1244 * blocks on disk
1245 */
1246 static noinline int run_delalloc_nocow(struct inode *inode,
1247 struct page *locked_page,
1248 u64 start, u64 end, int *page_started, int force,
1249 unsigned long *nr_written)
1250 {
1251 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1252 struct btrfs_root *root = BTRFS_I(inode)->root;
1253 struct btrfs_trans_handle *trans;
1254 struct extent_buffer *leaf;
1255 struct btrfs_path *path;
1256 struct btrfs_file_extent_item *fi;
1257 struct btrfs_key found_key;
1258 u64 cow_start;
1259 u64 cur_offset;
1260 u64 extent_end;
1261 u64 extent_offset;
1262 u64 disk_bytenr;
1263 u64 num_bytes;
1264 u64 disk_num_bytes;
1265 u64 ram_bytes;
1266 int extent_type;
1267 int ret, err;
1268 int type;
1269 int nocow;
1270 int check_prev = 1;
1271 bool nolock;
1272 u64 ino = btrfs_ino(inode);
1273
1274 path = btrfs_alloc_path();
1275 if (!path) {
1276 extent_clear_unlock_delalloc(inode, start, end, end,
1277 locked_page,
1278 EXTENT_LOCKED | EXTENT_DELALLOC |
1279 EXTENT_DO_ACCOUNTING |
1280 EXTENT_DEFRAG, PAGE_UNLOCK |
1281 PAGE_CLEAR_DIRTY |
1282 PAGE_SET_WRITEBACK |
1283 PAGE_END_WRITEBACK);
1284 return -ENOMEM;
1285 }
1286
1287 nolock = btrfs_is_free_space_inode(inode);
1288
1289 if (nolock)
1290 trans = btrfs_join_transaction_nolock(root);
1291 else
1292 trans = btrfs_join_transaction(root);
1293
1294 if (IS_ERR(trans)) {
1295 extent_clear_unlock_delalloc(inode, start, end, end,
1296 locked_page,
1297 EXTENT_LOCKED | EXTENT_DELALLOC |
1298 EXTENT_DO_ACCOUNTING |
1299 EXTENT_DEFRAG, PAGE_UNLOCK |
1300 PAGE_CLEAR_DIRTY |
1301 PAGE_SET_WRITEBACK |
1302 PAGE_END_WRITEBACK);
1303 btrfs_free_path(path);
1304 return PTR_ERR(trans);
1305 }
1306
1307 trans->block_rsv = &fs_info->delalloc_block_rsv;
1308
1309 cow_start = (u64)-1;
1310 cur_offset = start;
1311 while (1) {
1312 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1313 cur_offset, 0);
1314 if (ret < 0)
1315 goto error;
1316 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1317 leaf = path->nodes[0];
1318 btrfs_item_key_to_cpu(leaf, &found_key,
1319 path->slots[0] - 1);
1320 if (found_key.objectid == ino &&
1321 found_key.type == BTRFS_EXTENT_DATA_KEY)
1322 path->slots[0]--;
1323 }
1324 check_prev = 0;
1325 next_slot:
1326 leaf = path->nodes[0];
1327 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1328 ret = btrfs_next_leaf(root, path);
1329 if (ret < 0)
1330 goto error;
1331 if (ret > 0)
1332 break;
1333 leaf = path->nodes[0];
1334 }
1335
1336 nocow = 0;
1337 disk_bytenr = 0;
1338 num_bytes = 0;
1339 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1340
1341 if (found_key.objectid > ino)
1342 break;
1343 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1344 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1345 path->slots[0]++;
1346 goto next_slot;
1347 }
1348 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1349 found_key.offset > end)
1350 break;
1351
1352 if (found_key.offset > cur_offset) {
1353 extent_end = found_key.offset;
1354 extent_type = 0;
1355 goto out_check;
1356 }
1357
1358 fi = btrfs_item_ptr(leaf, path->slots[0],
1359 struct btrfs_file_extent_item);
1360 extent_type = btrfs_file_extent_type(leaf, fi);
1361
1362 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1363 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1364 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1365 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1366 extent_offset = btrfs_file_extent_offset(leaf, fi);
1367 extent_end = found_key.offset +
1368 btrfs_file_extent_num_bytes(leaf, fi);
1369 disk_num_bytes =
1370 btrfs_file_extent_disk_num_bytes(leaf, fi);
1371 if (extent_end <= start) {
1372 path->slots[0]++;
1373 goto next_slot;
1374 }
1375 if (disk_bytenr == 0)
1376 goto out_check;
1377 if (btrfs_file_extent_compression(leaf, fi) ||
1378 btrfs_file_extent_encryption(leaf, fi) ||
1379 btrfs_file_extent_other_encoding(leaf, fi))
1380 goto out_check;
1381 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1382 goto out_check;
1383 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1384 goto out_check;
1385 if (btrfs_cross_ref_exist(trans, root, ino,
1386 found_key.offset -
1387 extent_offset, disk_bytenr))
1388 goto out_check;
1389 disk_bytenr += extent_offset;
1390 disk_bytenr += cur_offset - found_key.offset;
1391 num_bytes = min(end + 1, extent_end) - cur_offset;
1392 /*
1393 * if there are pending snapshots for this root,
1394 * we fall into common COW way.
1395 */
1396 if (!nolock) {
1397 err = btrfs_start_write_no_snapshoting(root);
1398 if (!err)
1399 goto out_check;
1400 }
1401 /*
1402 * force cow if csum exists in the range.
1403 * this ensure that csum for a given extent are
1404 * either valid or do not exist.
1405 */
1406 if (csum_exist_in_range(fs_info, disk_bytenr,
1407 num_bytes))
1408 goto out_check;
1409 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1410 goto out_check;
1411 nocow = 1;
1412 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1413 extent_end = found_key.offset +
1414 btrfs_file_extent_inline_len(leaf,
1415 path->slots[0], fi);
1416 extent_end = ALIGN(extent_end,
1417 fs_info->sectorsize);
1418 } else {
1419 BUG_ON(1);
1420 }
1421 out_check:
1422 if (extent_end <= start) {
1423 path->slots[0]++;
1424 if (!nolock && nocow)
1425 btrfs_end_write_no_snapshoting(root);
1426 if (nocow)
1427 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1428 goto next_slot;
1429 }
1430 if (!nocow) {
1431 if (cow_start == (u64)-1)
1432 cow_start = cur_offset;
1433 cur_offset = extent_end;
1434 if (cur_offset > end)
1435 break;
1436 path->slots[0]++;
1437 goto next_slot;
1438 }
1439
1440 btrfs_release_path(path);
1441 if (cow_start != (u64)-1) {
1442 ret = cow_file_range(inode, locked_page,
1443 cow_start, found_key.offset - 1,
1444 end, page_started, nr_written, 1,
1445 NULL);
1446 if (ret) {
1447 if (!nolock && nocow)
1448 btrfs_end_write_no_snapshoting(root);
1449 if (nocow)
1450 btrfs_dec_nocow_writers(fs_info,
1451 disk_bytenr);
1452 goto error;
1453 }
1454 cow_start = (u64)-1;
1455 }
1456
1457 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1458 struct extent_map *em;
1459 struct extent_map_tree *em_tree;
1460 em_tree = &BTRFS_I(inode)->extent_tree;
1461 em = alloc_extent_map();
1462 BUG_ON(!em); /* -ENOMEM */
1463 em->start = cur_offset;
1464 em->orig_start = found_key.offset - extent_offset;
1465 em->len = num_bytes;
1466 em->block_len = num_bytes;
1467 em->block_start = disk_bytenr;
1468 em->orig_block_len = disk_num_bytes;
1469 em->ram_bytes = ram_bytes;
1470 em->bdev = fs_info->fs_devices->latest_bdev;
1471 em->mod_start = em->start;
1472 em->mod_len = em->len;
1473 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1474 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1475 em->generation = -1;
1476 while (1) {
1477 write_lock(&em_tree->lock);
1478 ret = add_extent_mapping(em_tree, em, 1);
1479 write_unlock(&em_tree->lock);
1480 if (ret != -EEXIST) {
1481 free_extent_map(em);
1482 break;
1483 }
1484 btrfs_drop_extent_cache(inode, em->start,
1485 em->start + em->len - 1, 0);
1486 }
1487 type = BTRFS_ORDERED_PREALLOC;
1488 } else {
1489 type = BTRFS_ORDERED_NOCOW;
1490 }
1491
1492 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1493 num_bytes, num_bytes, type);
1494 if (nocow)
1495 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1496 BUG_ON(ret); /* -ENOMEM */
1497
1498 if (root->root_key.objectid ==
1499 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1500 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1501 num_bytes);
1502 if (ret) {
1503 if (!nolock && nocow)
1504 btrfs_end_write_no_snapshoting(root);
1505 goto error;
1506 }
1507 }
1508
1509 extent_clear_unlock_delalloc(inode, cur_offset,
1510 cur_offset + num_bytes - 1, end,
1511 locked_page, EXTENT_LOCKED |
1512 EXTENT_DELALLOC |
1513 EXTENT_CLEAR_DATA_RESV,
1514 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1515
1516 if (!nolock && nocow)
1517 btrfs_end_write_no_snapshoting(root);
1518 cur_offset = extent_end;
1519 if (cur_offset > end)
1520 break;
1521 }
1522 btrfs_release_path(path);
1523
1524 if (cur_offset <= end && cow_start == (u64)-1) {
1525 cow_start = cur_offset;
1526 cur_offset = end;
1527 }
1528
1529 if (cow_start != (u64)-1) {
1530 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1531 page_started, nr_written, 1, NULL);
1532 if (ret)
1533 goto error;
1534 }
1535
1536 error:
1537 err = btrfs_end_transaction(trans);
1538 if (!ret)
1539 ret = err;
1540
1541 if (ret && cur_offset < end)
1542 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1543 locked_page, EXTENT_LOCKED |
1544 EXTENT_DELALLOC | EXTENT_DEFRAG |
1545 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1546 PAGE_CLEAR_DIRTY |
1547 PAGE_SET_WRITEBACK |
1548 PAGE_END_WRITEBACK);
1549 btrfs_free_path(path);
1550 return ret;
1551 }
1552
1553 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1554 {
1555
1556 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1557 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1558 return 0;
1559
1560 /*
1561 * @defrag_bytes is a hint value, no spinlock held here,
1562 * if is not zero, it means the file is defragging.
1563 * Force cow if given extent needs to be defragged.
1564 */
1565 if (BTRFS_I(inode)->defrag_bytes &&
1566 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1567 EXTENT_DEFRAG, 0, NULL))
1568 return 1;
1569
1570 return 0;
1571 }
1572
1573 /*
1574 * extent_io.c call back to do delayed allocation processing
1575 */
1576 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1577 u64 start, u64 end, int *page_started,
1578 unsigned long *nr_written)
1579 {
1580 int ret;
1581 int force_cow = need_force_cow(inode, start, end);
1582
1583 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1584 ret = run_delalloc_nocow(inode, locked_page, start, end,
1585 page_started, 1, nr_written);
1586 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1587 ret = run_delalloc_nocow(inode, locked_page, start, end,
1588 page_started, 0, nr_written);
1589 } else if (!inode_need_compress(inode)) {
1590 ret = cow_file_range(inode, locked_page, start, end, end,
1591 page_started, nr_written, 1, NULL);
1592 } else {
1593 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1594 &BTRFS_I(inode)->runtime_flags);
1595 ret = cow_file_range_async(inode, locked_page, start, end,
1596 page_started, nr_written);
1597 }
1598 return ret;
1599 }
1600
1601 static void btrfs_split_extent_hook(struct inode *inode,
1602 struct extent_state *orig, u64 split)
1603 {
1604 u64 size;
1605
1606 /* not delalloc, ignore it */
1607 if (!(orig->state & EXTENT_DELALLOC))
1608 return;
1609
1610 size = orig->end - orig->start + 1;
1611 if (size > BTRFS_MAX_EXTENT_SIZE) {
1612 u64 num_extents;
1613 u64 new_size;
1614
1615 /*
1616 * See the explanation in btrfs_merge_extent_hook, the same
1617 * applies here, just in reverse.
1618 */
1619 new_size = orig->end - split + 1;
1620 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1621 BTRFS_MAX_EXTENT_SIZE);
1622 new_size = split - orig->start;
1623 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1624 BTRFS_MAX_EXTENT_SIZE);
1625 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1626 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1627 return;
1628 }
1629
1630 spin_lock(&BTRFS_I(inode)->lock);
1631 BTRFS_I(inode)->outstanding_extents++;
1632 spin_unlock(&BTRFS_I(inode)->lock);
1633 }
1634
1635 /*
1636 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1637 * extents so we can keep track of new extents that are just merged onto old
1638 * extents, such as when we are doing sequential writes, so we can properly
1639 * account for the metadata space we'll need.
1640 */
1641 static void btrfs_merge_extent_hook(struct inode *inode,
1642 struct extent_state *new,
1643 struct extent_state *other)
1644 {
1645 u64 new_size, old_size;
1646 u64 num_extents;
1647
1648 /* not delalloc, ignore it */
1649 if (!(other->state & EXTENT_DELALLOC))
1650 return;
1651
1652 if (new->start > other->start)
1653 new_size = new->end - other->start + 1;
1654 else
1655 new_size = other->end - new->start + 1;
1656
1657 /* we're not bigger than the max, unreserve the space and go */
1658 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1659 spin_lock(&BTRFS_I(inode)->lock);
1660 BTRFS_I(inode)->outstanding_extents--;
1661 spin_unlock(&BTRFS_I(inode)->lock);
1662 return;
1663 }
1664
1665 /*
1666 * We have to add up either side to figure out how many extents were
1667 * accounted for before we merged into one big extent. If the number of
1668 * extents we accounted for is <= the amount we need for the new range
1669 * then we can return, otherwise drop. Think of it like this
1670 *
1671 * [ 4k][MAX_SIZE]
1672 *
1673 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1674 * need 2 outstanding extents, on one side we have 1 and the other side
1675 * we have 1 so they are == and we can return. But in this case
1676 *
1677 * [MAX_SIZE+4k][MAX_SIZE+4k]
1678 *
1679 * Each range on their own accounts for 2 extents, but merged together
1680 * they are only 3 extents worth of accounting, so we need to drop in
1681 * this case.
1682 */
1683 old_size = other->end - other->start + 1;
1684 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1685 BTRFS_MAX_EXTENT_SIZE);
1686 old_size = new->end - new->start + 1;
1687 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1688 BTRFS_MAX_EXTENT_SIZE);
1689
1690 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1691 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1692 return;
1693
1694 spin_lock(&BTRFS_I(inode)->lock);
1695 BTRFS_I(inode)->outstanding_extents--;
1696 spin_unlock(&BTRFS_I(inode)->lock);
1697 }
1698
1699 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1700 struct inode *inode)
1701 {
1702 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1703
1704 spin_lock(&root->delalloc_lock);
1705 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1706 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1707 &root->delalloc_inodes);
1708 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1709 &BTRFS_I(inode)->runtime_flags);
1710 root->nr_delalloc_inodes++;
1711 if (root->nr_delalloc_inodes == 1) {
1712 spin_lock(&fs_info->delalloc_root_lock);
1713 BUG_ON(!list_empty(&root->delalloc_root));
1714 list_add_tail(&root->delalloc_root,
1715 &fs_info->delalloc_roots);
1716 spin_unlock(&fs_info->delalloc_root_lock);
1717 }
1718 }
1719 spin_unlock(&root->delalloc_lock);
1720 }
1721
1722 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1723 struct inode *inode)
1724 {
1725 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1726
1727 spin_lock(&root->delalloc_lock);
1728 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1729 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1730 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1731 &BTRFS_I(inode)->runtime_flags);
1732 root->nr_delalloc_inodes--;
1733 if (!root->nr_delalloc_inodes) {
1734 spin_lock(&fs_info->delalloc_root_lock);
1735 BUG_ON(list_empty(&root->delalloc_root));
1736 list_del_init(&root->delalloc_root);
1737 spin_unlock(&fs_info->delalloc_root_lock);
1738 }
1739 }
1740 spin_unlock(&root->delalloc_lock);
1741 }
1742
1743 /*
1744 * extent_io.c set_bit_hook, used to track delayed allocation
1745 * bytes in this file, and to maintain the list of inodes that
1746 * have pending delalloc work to be done.
1747 */
1748 static void btrfs_set_bit_hook(struct inode *inode,
1749 struct extent_state *state, unsigned *bits)
1750 {
1751
1752 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1753
1754 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1755 WARN_ON(1);
1756 /*
1757 * set_bit and clear bit hooks normally require _irqsave/restore
1758 * but in this case, we are only testing for the DELALLOC
1759 * bit, which is only set or cleared with irqs on
1760 */
1761 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1762 struct btrfs_root *root = BTRFS_I(inode)->root;
1763 u64 len = state->end + 1 - state->start;
1764 bool do_list = !btrfs_is_free_space_inode(inode);
1765
1766 if (*bits & EXTENT_FIRST_DELALLOC) {
1767 *bits &= ~EXTENT_FIRST_DELALLOC;
1768 } else {
1769 spin_lock(&BTRFS_I(inode)->lock);
1770 BTRFS_I(inode)->outstanding_extents++;
1771 spin_unlock(&BTRFS_I(inode)->lock);
1772 }
1773
1774 /* For sanity tests */
1775 if (btrfs_is_testing(fs_info))
1776 return;
1777
1778 __percpu_counter_add(&fs_info->delalloc_bytes, len,
1779 fs_info->delalloc_batch);
1780 spin_lock(&BTRFS_I(inode)->lock);
1781 BTRFS_I(inode)->delalloc_bytes += len;
1782 if (*bits & EXTENT_DEFRAG)
1783 BTRFS_I(inode)->defrag_bytes += len;
1784 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1785 &BTRFS_I(inode)->runtime_flags))
1786 btrfs_add_delalloc_inodes(root, inode);
1787 spin_unlock(&BTRFS_I(inode)->lock);
1788 }
1789 }
1790
1791 /*
1792 * extent_io.c clear_bit_hook, see set_bit_hook for why
1793 */
1794 static void btrfs_clear_bit_hook(struct inode *inode,
1795 struct extent_state *state,
1796 unsigned *bits)
1797 {
1798 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1799 u64 len = state->end + 1 - state->start;
1800 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1801 BTRFS_MAX_EXTENT_SIZE);
1802
1803 spin_lock(&BTRFS_I(inode)->lock);
1804 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1805 BTRFS_I(inode)->defrag_bytes -= len;
1806 spin_unlock(&BTRFS_I(inode)->lock);
1807
1808 /*
1809 * set_bit and clear bit hooks normally require _irqsave/restore
1810 * but in this case, we are only testing for the DELALLOC
1811 * bit, which is only set or cleared with irqs on
1812 */
1813 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1814 struct btrfs_root *root = BTRFS_I(inode)->root;
1815 bool do_list = !btrfs_is_free_space_inode(inode);
1816
1817 if (*bits & EXTENT_FIRST_DELALLOC) {
1818 *bits &= ~EXTENT_FIRST_DELALLOC;
1819 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1820 spin_lock(&BTRFS_I(inode)->lock);
1821 BTRFS_I(inode)->outstanding_extents -= num_extents;
1822 spin_unlock(&BTRFS_I(inode)->lock);
1823 }
1824
1825 /*
1826 * We don't reserve metadata space for space cache inodes so we
1827 * don't need to call dellalloc_release_metadata if there is an
1828 * error.
1829 */
1830 if (*bits & EXTENT_DO_ACCOUNTING &&
1831 root != fs_info->tree_root)
1832 btrfs_delalloc_release_metadata(inode, len);
1833
1834 /* For sanity tests. */
1835 if (btrfs_is_testing(fs_info))
1836 return;
1837
1838 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1839 && do_list && !(state->state & EXTENT_NORESERVE)
1840 && (*bits & (EXTENT_DO_ACCOUNTING |
1841 EXTENT_CLEAR_DATA_RESV)))
1842 btrfs_free_reserved_data_space_noquota(inode,
1843 state->start, len);
1844
1845 __percpu_counter_add(&fs_info->delalloc_bytes, -len,
1846 fs_info->delalloc_batch);
1847 spin_lock(&BTRFS_I(inode)->lock);
1848 BTRFS_I(inode)->delalloc_bytes -= len;
1849 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1850 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1851 &BTRFS_I(inode)->runtime_flags))
1852 btrfs_del_delalloc_inode(root, inode);
1853 spin_unlock(&BTRFS_I(inode)->lock);
1854 }
1855 }
1856
1857 /*
1858 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1859 * we don't create bios that span stripes or chunks
1860 *
1861 * return 1 if page cannot be merged to bio
1862 * return 0 if page can be merged to bio
1863 * return error otherwise
1864 */
1865 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1866 size_t size, struct bio *bio,
1867 unsigned long bio_flags)
1868 {
1869 struct inode *inode = page->mapping->host;
1870 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1871 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1872 u64 length = 0;
1873 u64 map_length;
1874 int ret;
1875
1876 if (bio_flags & EXTENT_BIO_COMPRESSED)
1877 return 0;
1878
1879 length = bio->bi_iter.bi_size;
1880 map_length = length;
1881 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1882 NULL, 0);
1883 if (ret < 0)
1884 return ret;
1885 if (map_length < length + size)
1886 return 1;
1887 return 0;
1888 }
1889
1890 /*
1891 * in order to insert checksums into the metadata in large chunks,
1892 * we wait until bio submission time. All the pages in the bio are
1893 * checksummed and sums are attached onto the ordered extent record.
1894 *
1895 * At IO completion time the cums attached on the ordered extent record
1896 * are inserted into the btree
1897 */
1898 static int __btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
1899 int mirror_num, unsigned long bio_flags,
1900 u64 bio_offset)
1901 {
1902 int ret = 0;
1903
1904 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1905 BUG_ON(ret); /* -ENOMEM */
1906 return 0;
1907 }
1908
1909 /*
1910 * in order to insert checksums into the metadata in large chunks,
1911 * we wait until bio submission time. All the pages in the bio are
1912 * checksummed and sums are attached onto the ordered extent record.
1913 *
1914 * At IO completion time the cums attached on the ordered extent record
1915 * are inserted into the btree
1916 */
1917 static int __btrfs_submit_bio_done(struct inode *inode, struct bio *bio,
1918 int mirror_num, unsigned long bio_flags,
1919 u64 bio_offset)
1920 {
1921 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1922 int ret;
1923
1924 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1925 if (ret) {
1926 bio->bi_error = ret;
1927 bio_endio(bio);
1928 }
1929 return ret;
1930 }
1931
1932 /*
1933 * extent_io.c submission hook. This does the right thing for csum calculation
1934 * on write, or reading the csums from the tree before a read
1935 */
1936 static int btrfs_submit_bio_hook(struct inode *inode, struct bio *bio,
1937 int mirror_num, unsigned long bio_flags,
1938 u64 bio_offset)
1939 {
1940 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1941 struct btrfs_root *root = BTRFS_I(inode)->root;
1942 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1943 int ret = 0;
1944 int skip_sum;
1945 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1946
1947 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1948
1949 if (btrfs_is_free_space_inode(inode))
1950 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1951
1952 if (bio_op(bio) != REQ_OP_WRITE) {
1953 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1954 if (ret)
1955 goto out;
1956
1957 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1958 ret = btrfs_submit_compressed_read(inode, bio,
1959 mirror_num,
1960 bio_flags);
1961 goto out;
1962 } else if (!skip_sum) {
1963 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
1964 if (ret)
1965 goto out;
1966 }
1967 goto mapit;
1968 } else if (async && !skip_sum) {
1969 /* csum items have already been cloned */
1970 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1971 goto mapit;
1972 /* we're doing a write, do the async checksumming */
1973 ret = btrfs_wq_submit_bio(fs_info, inode, bio, mirror_num,
1974 bio_flags, bio_offset,
1975 __btrfs_submit_bio_start,
1976 __btrfs_submit_bio_done);
1977 goto out;
1978 } else if (!skip_sum) {
1979 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1980 if (ret)
1981 goto out;
1982 }
1983
1984 mapit:
1985 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1986
1987 out:
1988 if (ret < 0) {
1989 bio->bi_error = ret;
1990 bio_endio(bio);
1991 }
1992 return ret;
1993 }
1994
1995 /*
1996 * given a list of ordered sums record them in the inode. This happens
1997 * at IO completion time based on sums calculated at bio submission time.
1998 */
1999 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2000 struct inode *inode, u64 file_offset,
2001 struct list_head *list)
2002 {
2003 struct btrfs_ordered_sum *sum;
2004
2005 list_for_each_entry(sum, list, list) {
2006 trans->adding_csums = 1;
2007 btrfs_csum_file_blocks(trans,
2008 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2009 trans->adding_csums = 0;
2010 }
2011 return 0;
2012 }
2013
2014 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2015 struct extent_state **cached_state, int dedupe)
2016 {
2017 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2018 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2019 cached_state);
2020 }
2021
2022 /* see btrfs_writepage_start_hook for details on why this is required */
2023 struct btrfs_writepage_fixup {
2024 struct page *page;
2025 struct btrfs_work work;
2026 };
2027
2028 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2029 {
2030 struct btrfs_writepage_fixup *fixup;
2031 struct btrfs_ordered_extent *ordered;
2032 struct extent_state *cached_state = NULL;
2033 struct page *page;
2034 struct inode *inode;
2035 u64 page_start;
2036 u64 page_end;
2037 int ret;
2038
2039 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2040 page = fixup->page;
2041 again:
2042 lock_page(page);
2043 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2044 ClearPageChecked(page);
2045 goto out_page;
2046 }
2047
2048 inode = page->mapping->host;
2049 page_start = page_offset(page);
2050 page_end = page_offset(page) + PAGE_SIZE - 1;
2051
2052 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2053 &cached_state);
2054
2055 /* already ordered? We're done */
2056 if (PagePrivate2(page))
2057 goto out;
2058
2059 ordered = btrfs_lookup_ordered_range(inode, page_start,
2060 PAGE_SIZE);
2061 if (ordered) {
2062 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2063 page_end, &cached_state, GFP_NOFS);
2064 unlock_page(page);
2065 btrfs_start_ordered_extent(inode, ordered, 1);
2066 btrfs_put_ordered_extent(ordered);
2067 goto again;
2068 }
2069
2070 ret = btrfs_delalloc_reserve_space(inode, page_start,
2071 PAGE_SIZE);
2072 if (ret) {
2073 mapping_set_error(page->mapping, ret);
2074 end_extent_writepage(page, ret, page_start, page_end);
2075 ClearPageChecked(page);
2076 goto out;
2077 }
2078
2079 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state,
2080 0);
2081 ClearPageChecked(page);
2082 set_page_dirty(page);
2083 out:
2084 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2085 &cached_state, GFP_NOFS);
2086 out_page:
2087 unlock_page(page);
2088 put_page(page);
2089 kfree(fixup);
2090 }
2091
2092 /*
2093 * There are a few paths in the higher layers of the kernel that directly
2094 * set the page dirty bit without asking the filesystem if it is a
2095 * good idea. This causes problems because we want to make sure COW
2096 * properly happens and the data=ordered rules are followed.
2097 *
2098 * In our case any range that doesn't have the ORDERED bit set
2099 * hasn't been properly setup for IO. We kick off an async process
2100 * to fix it up. The async helper will wait for ordered extents, set
2101 * the delalloc bit and make it safe to write the page.
2102 */
2103 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2104 {
2105 struct inode *inode = page->mapping->host;
2106 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2107 struct btrfs_writepage_fixup *fixup;
2108
2109 /* this page is properly in the ordered list */
2110 if (TestClearPagePrivate2(page))
2111 return 0;
2112
2113 if (PageChecked(page))
2114 return -EAGAIN;
2115
2116 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2117 if (!fixup)
2118 return -EAGAIN;
2119
2120 SetPageChecked(page);
2121 get_page(page);
2122 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2123 btrfs_writepage_fixup_worker, NULL, NULL);
2124 fixup->page = page;
2125 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2126 return -EBUSY;
2127 }
2128
2129 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2130 struct inode *inode, u64 file_pos,
2131 u64 disk_bytenr, u64 disk_num_bytes,
2132 u64 num_bytes, u64 ram_bytes,
2133 u8 compression, u8 encryption,
2134 u16 other_encoding, int extent_type)
2135 {
2136 struct btrfs_root *root = BTRFS_I(inode)->root;
2137 struct btrfs_file_extent_item *fi;
2138 struct btrfs_path *path;
2139 struct extent_buffer *leaf;
2140 struct btrfs_key ins;
2141 int extent_inserted = 0;
2142 int ret;
2143
2144 path = btrfs_alloc_path();
2145 if (!path)
2146 return -ENOMEM;
2147
2148 /*
2149 * we may be replacing one extent in the tree with another.
2150 * The new extent is pinned in the extent map, and we don't want
2151 * to drop it from the cache until it is completely in the btree.
2152 *
2153 * So, tell btrfs_drop_extents to leave this extent in the cache.
2154 * the caller is expected to unpin it and allow it to be merged
2155 * with the others.
2156 */
2157 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2158 file_pos + num_bytes, NULL, 0,
2159 1, sizeof(*fi), &extent_inserted);
2160 if (ret)
2161 goto out;
2162
2163 if (!extent_inserted) {
2164 ins.objectid = btrfs_ino(inode);
2165 ins.offset = file_pos;
2166 ins.type = BTRFS_EXTENT_DATA_KEY;
2167
2168 path->leave_spinning = 1;
2169 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2170 sizeof(*fi));
2171 if (ret)
2172 goto out;
2173 }
2174 leaf = path->nodes[0];
2175 fi = btrfs_item_ptr(leaf, path->slots[0],
2176 struct btrfs_file_extent_item);
2177 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2178 btrfs_set_file_extent_type(leaf, fi, extent_type);
2179 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2180 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2181 btrfs_set_file_extent_offset(leaf, fi, 0);
2182 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2183 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2184 btrfs_set_file_extent_compression(leaf, fi, compression);
2185 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2186 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2187
2188 btrfs_mark_buffer_dirty(leaf);
2189 btrfs_release_path(path);
2190
2191 inode_add_bytes(inode, num_bytes);
2192
2193 ins.objectid = disk_bytenr;
2194 ins.offset = disk_num_bytes;
2195 ins.type = BTRFS_EXTENT_ITEM_KEY;
2196 ret = btrfs_alloc_reserved_file_extent(trans, root->root_key.objectid,
2197 btrfs_ino(inode), file_pos,
2198 ram_bytes, &ins);
2199 /*
2200 * Release the reserved range from inode dirty range map, as it is
2201 * already moved into delayed_ref_head
2202 */
2203 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2204 out:
2205 btrfs_free_path(path);
2206
2207 return ret;
2208 }
2209
2210 /* snapshot-aware defrag */
2211 struct sa_defrag_extent_backref {
2212 struct rb_node node;
2213 struct old_sa_defrag_extent *old;
2214 u64 root_id;
2215 u64 inum;
2216 u64 file_pos;
2217 u64 extent_offset;
2218 u64 num_bytes;
2219 u64 generation;
2220 };
2221
2222 struct old_sa_defrag_extent {
2223 struct list_head list;
2224 struct new_sa_defrag_extent *new;
2225
2226 u64 extent_offset;
2227 u64 bytenr;
2228 u64 offset;
2229 u64 len;
2230 int count;
2231 };
2232
2233 struct new_sa_defrag_extent {
2234 struct rb_root root;
2235 struct list_head head;
2236 struct btrfs_path *path;
2237 struct inode *inode;
2238 u64 file_pos;
2239 u64 len;
2240 u64 bytenr;
2241 u64 disk_len;
2242 u8 compress_type;
2243 };
2244
2245 static int backref_comp(struct sa_defrag_extent_backref *b1,
2246 struct sa_defrag_extent_backref *b2)
2247 {
2248 if (b1->root_id < b2->root_id)
2249 return -1;
2250 else if (b1->root_id > b2->root_id)
2251 return 1;
2252
2253 if (b1->inum < b2->inum)
2254 return -1;
2255 else if (b1->inum > b2->inum)
2256 return 1;
2257
2258 if (b1->file_pos < b2->file_pos)
2259 return -1;
2260 else if (b1->file_pos > b2->file_pos)
2261 return 1;
2262
2263 /*
2264 * [------------------------------] ===> (a range of space)
2265 * |<--->| |<---->| =============> (fs/file tree A)
2266 * |<---------------------------->| ===> (fs/file tree B)
2267 *
2268 * A range of space can refer to two file extents in one tree while
2269 * refer to only one file extent in another tree.
2270 *
2271 * So we may process a disk offset more than one time(two extents in A)
2272 * and locate at the same extent(one extent in B), then insert two same
2273 * backrefs(both refer to the extent in B).
2274 */
2275 return 0;
2276 }
2277
2278 static void backref_insert(struct rb_root *root,
2279 struct sa_defrag_extent_backref *backref)
2280 {
2281 struct rb_node **p = &root->rb_node;
2282 struct rb_node *parent = NULL;
2283 struct sa_defrag_extent_backref *entry;
2284 int ret;
2285
2286 while (*p) {
2287 parent = *p;
2288 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2289
2290 ret = backref_comp(backref, entry);
2291 if (ret < 0)
2292 p = &(*p)->rb_left;
2293 else
2294 p = &(*p)->rb_right;
2295 }
2296
2297 rb_link_node(&backref->node, parent, p);
2298 rb_insert_color(&backref->node, root);
2299 }
2300
2301 /*
2302 * Note the backref might has changed, and in this case we just return 0.
2303 */
2304 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2305 void *ctx)
2306 {
2307 struct btrfs_file_extent_item *extent;
2308 struct old_sa_defrag_extent *old = ctx;
2309 struct new_sa_defrag_extent *new = old->new;
2310 struct btrfs_path *path = new->path;
2311 struct btrfs_key key;
2312 struct btrfs_root *root;
2313 struct sa_defrag_extent_backref *backref;
2314 struct extent_buffer *leaf;
2315 struct inode *inode = new->inode;
2316 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2317 int slot;
2318 int ret;
2319 u64 extent_offset;
2320 u64 num_bytes;
2321
2322 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2323 inum == btrfs_ino(inode))
2324 return 0;
2325
2326 key.objectid = root_id;
2327 key.type = BTRFS_ROOT_ITEM_KEY;
2328 key.offset = (u64)-1;
2329
2330 root = btrfs_read_fs_root_no_name(fs_info, &key);
2331 if (IS_ERR(root)) {
2332 if (PTR_ERR(root) == -ENOENT)
2333 return 0;
2334 WARN_ON(1);
2335 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2336 inum, offset, root_id);
2337 return PTR_ERR(root);
2338 }
2339
2340 key.objectid = inum;
2341 key.type = BTRFS_EXTENT_DATA_KEY;
2342 if (offset > (u64)-1 << 32)
2343 key.offset = 0;
2344 else
2345 key.offset = offset;
2346
2347 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2348 if (WARN_ON(ret < 0))
2349 return ret;
2350 ret = 0;
2351
2352 while (1) {
2353 cond_resched();
2354
2355 leaf = path->nodes[0];
2356 slot = path->slots[0];
2357
2358 if (slot >= btrfs_header_nritems(leaf)) {
2359 ret = btrfs_next_leaf(root, path);
2360 if (ret < 0) {
2361 goto out;
2362 } else if (ret > 0) {
2363 ret = 0;
2364 goto out;
2365 }
2366 continue;
2367 }
2368
2369 path->slots[0]++;
2370
2371 btrfs_item_key_to_cpu(leaf, &key, slot);
2372
2373 if (key.objectid > inum)
2374 goto out;
2375
2376 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2377 continue;
2378
2379 extent = btrfs_item_ptr(leaf, slot,
2380 struct btrfs_file_extent_item);
2381
2382 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2383 continue;
2384
2385 /*
2386 * 'offset' refers to the exact key.offset,
2387 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2388 * (key.offset - extent_offset).
2389 */
2390 if (key.offset != offset)
2391 continue;
2392
2393 extent_offset = btrfs_file_extent_offset(leaf, extent);
2394 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2395
2396 if (extent_offset >= old->extent_offset + old->offset +
2397 old->len || extent_offset + num_bytes <=
2398 old->extent_offset + old->offset)
2399 continue;
2400 break;
2401 }
2402
2403 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2404 if (!backref) {
2405 ret = -ENOENT;
2406 goto out;
2407 }
2408
2409 backref->root_id = root_id;
2410 backref->inum = inum;
2411 backref->file_pos = offset;
2412 backref->num_bytes = num_bytes;
2413 backref->extent_offset = extent_offset;
2414 backref->generation = btrfs_file_extent_generation(leaf, extent);
2415 backref->old = old;
2416 backref_insert(&new->root, backref);
2417 old->count++;
2418 out:
2419 btrfs_release_path(path);
2420 WARN_ON(ret);
2421 return ret;
2422 }
2423
2424 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2425 struct new_sa_defrag_extent *new)
2426 {
2427 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2428 struct old_sa_defrag_extent *old, *tmp;
2429 int ret;
2430
2431 new->path = path;
2432
2433 list_for_each_entry_safe(old, tmp, &new->head, list) {
2434 ret = iterate_inodes_from_logical(old->bytenr +
2435 old->extent_offset, fs_info,
2436 path, record_one_backref,
2437 old);
2438 if (ret < 0 && ret != -ENOENT)
2439 return false;
2440
2441 /* no backref to be processed for this extent */
2442 if (!old->count) {
2443 list_del(&old->list);
2444 kfree(old);
2445 }
2446 }
2447
2448 if (list_empty(&new->head))
2449 return false;
2450
2451 return true;
2452 }
2453
2454 static int relink_is_mergable(struct extent_buffer *leaf,
2455 struct btrfs_file_extent_item *fi,
2456 struct new_sa_defrag_extent *new)
2457 {
2458 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2459 return 0;
2460
2461 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2462 return 0;
2463
2464 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2465 return 0;
2466
2467 if (btrfs_file_extent_encryption(leaf, fi) ||
2468 btrfs_file_extent_other_encoding(leaf, fi))
2469 return 0;
2470
2471 return 1;
2472 }
2473
2474 /*
2475 * Note the backref might has changed, and in this case we just return 0.
2476 */
2477 static noinline int relink_extent_backref(struct btrfs_path *path,
2478 struct sa_defrag_extent_backref *prev,
2479 struct sa_defrag_extent_backref *backref)
2480 {
2481 struct btrfs_file_extent_item *extent;
2482 struct btrfs_file_extent_item *item;
2483 struct btrfs_ordered_extent *ordered;
2484 struct btrfs_trans_handle *trans;
2485 struct btrfs_root *root;
2486 struct btrfs_key key;
2487 struct extent_buffer *leaf;
2488 struct old_sa_defrag_extent *old = backref->old;
2489 struct new_sa_defrag_extent *new = old->new;
2490 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2491 struct inode *inode;
2492 struct extent_state *cached = NULL;
2493 int ret = 0;
2494 u64 start;
2495 u64 len;
2496 u64 lock_start;
2497 u64 lock_end;
2498 bool merge = false;
2499 int index;
2500
2501 if (prev && prev->root_id == backref->root_id &&
2502 prev->inum == backref->inum &&
2503 prev->file_pos + prev->num_bytes == backref->file_pos)
2504 merge = true;
2505
2506 /* step 1: get root */
2507 key.objectid = backref->root_id;
2508 key.type = BTRFS_ROOT_ITEM_KEY;
2509 key.offset = (u64)-1;
2510
2511 index = srcu_read_lock(&fs_info->subvol_srcu);
2512
2513 root = btrfs_read_fs_root_no_name(fs_info, &key);
2514 if (IS_ERR(root)) {
2515 srcu_read_unlock(&fs_info->subvol_srcu, index);
2516 if (PTR_ERR(root) == -ENOENT)
2517 return 0;
2518 return PTR_ERR(root);
2519 }
2520
2521 if (btrfs_root_readonly(root)) {
2522 srcu_read_unlock(&fs_info->subvol_srcu, index);
2523 return 0;
2524 }
2525
2526 /* step 2: get inode */
2527 key.objectid = backref->inum;
2528 key.type = BTRFS_INODE_ITEM_KEY;
2529 key.offset = 0;
2530
2531 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2532 if (IS_ERR(inode)) {
2533 srcu_read_unlock(&fs_info->subvol_srcu, index);
2534 return 0;
2535 }
2536
2537 srcu_read_unlock(&fs_info->subvol_srcu, index);
2538
2539 /* step 3: relink backref */
2540 lock_start = backref->file_pos;
2541 lock_end = backref->file_pos + backref->num_bytes - 1;
2542 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2543 &cached);
2544
2545 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2546 if (ordered) {
2547 btrfs_put_ordered_extent(ordered);
2548 goto out_unlock;
2549 }
2550
2551 trans = btrfs_join_transaction(root);
2552 if (IS_ERR(trans)) {
2553 ret = PTR_ERR(trans);
2554 goto out_unlock;
2555 }
2556
2557 key.objectid = backref->inum;
2558 key.type = BTRFS_EXTENT_DATA_KEY;
2559 key.offset = backref->file_pos;
2560
2561 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2562 if (ret < 0) {
2563 goto out_free_path;
2564 } else if (ret > 0) {
2565 ret = 0;
2566 goto out_free_path;
2567 }
2568
2569 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2570 struct btrfs_file_extent_item);
2571
2572 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2573 backref->generation)
2574 goto out_free_path;
2575
2576 btrfs_release_path(path);
2577
2578 start = backref->file_pos;
2579 if (backref->extent_offset < old->extent_offset + old->offset)
2580 start += old->extent_offset + old->offset -
2581 backref->extent_offset;
2582
2583 len = min(backref->extent_offset + backref->num_bytes,
2584 old->extent_offset + old->offset + old->len);
2585 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2586
2587 ret = btrfs_drop_extents(trans, root, inode, start,
2588 start + len, 1);
2589 if (ret)
2590 goto out_free_path;
2591 again:
2592 key.objectid = btrfs_ino(inode);
2593 key.type = BTRFS_EXTENT_DATA_KEY;
2594 key.offset = start;
2595
2596 path->leave_spinning = 1;
2597 if (merge) {
2598 struct btrfs_file_extent_item *fi;
2599 u64 extent_len;
2600 struct btrfs_key found_key;
2601
2602 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2603 if (ret < 0)
2604 goto out_free_path;
2605
2606 path->slots[0]--;
2607 leaf = path->nodes[0];
2608 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2609
2610 fi = btrfs_item_ptr(leaf, path->slots[0],
2611 struct btrfs_file_extent_item);
2612 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2613
2614 if (extent_len + found_key.offset == start &&
2615 relink_is_mergable(leaf, fi, new)) {
2616 btrfs_set_file_extent_num_bytes(leaf, fi,
2617 extent_len + len);
2618 btrfs_mark_buffer_dirty(leaf);
2619 inode_add_bytes(inode, len);
2620
2621 ret = 1;
2622 goto out_free_path;
2623 } else {
2624 merge = false;
2625 btrfs_release_path(path);
2626 goto again;
2627 }
2628 }
2629
2630 ret = btrfs_insert_empty_item(trans, root, path, &key,
2631 sizeof(*extent));
2632 if (ret) {
2633 btrfs_abort_transaction(trans, ret);
2634 goto out_free_path;
2635 }
2636
2637 leaf = path->nodes[0];
2638 item = btrfs_item_ptr(leaf, path->slots[0],
2639 struct btrfs_file_extent_item);
2640 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2641 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2642 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2643 btrfs_set_file_extent_num_bytes(leaf, item, len);
2644 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2645 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2646 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2647 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2648 btrfs_set_file_extent_encryption(leaf, item, 0);
2649 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2650
2651 btrfs_mark_buffer_dirty(leaf);
2652 inode_add_bytes(inode, len);
2653 btrfs_release_path(path);
2654
2655 ret = btrfs_inc_extent_ref(trans, fs_info, new->bytenr,
2656 new->disk_len, 0,
2657 backref->root_id, backref->inum,
2658 new->file_pos); /* start - extent_offset */
2659 if (ret) {
2660 btrfs_abort_transaction(trans, ret);
2661 goto out_free_path;
2662 }
2663
2664 ret = 1;
2665 out_free_path:
2666 btrfs_release_path(path);
2667 path->leave_spinning = 0;
2668 btrfs_end_transaction(trans);
2669 out_unlock:
2670 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2671 &cached, GFP_NOFS);
2672 iput(inode);
2673 return ret;
2674 }
2675
2676 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2677 {
2678 struct old_sa_defrag_extent *old, *tmp;
2679
2680 if (!new)
2681 return;
2682
2683 list_for_each_entry_safe(old, tmp, &new->head, list) {
2684 kfree(old);
2685 }
2686 kfree(new);
2687 }
2688
2689 static void relink_file_extents(struct new_sa_defrag_extent *new)
2690 {
2691 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2692 struct btrfs_path *path;
2693 struct sa_defrag_extent_backref *backref;
2694 struct sa_defrag_extent_backref *prev = NULL;
2695 struct inode *inode;
2696 struct btrfs_root *root;
2697 struct rb_node *node;
2698 int ret;
2699
2700 inode = new->inode;
2701 root = BTRFS_I(inode)->root;
2702
2703 path = btrfs_alloc_path();
2704 if (!path)
2705 return;
2706
2707 if (!record_extent_backrefs(path, new)) {
2708 btrfs_free_path(path);
2709 goto out;
2710 }
2711 btrfs_release_path(path);
2712
2713 while (1) {
2714 node = rb_first(&new->root);
2715 if (!node)
2716 break;
2717 rb_erase(node, &new->root);
2718
2719 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2720
2721 ret = relink_extent_backref(path, prev, backref);
2722 WARN_ON(ret < 0);
2723
2724 kfree(prev);
2725
2726 if (ret == 1)
2727 prev = backref;
2728 else
2729 prev = NULL;
2730 cond_resched();
2731 }
2732 kfree(prev);
2733
2734 btrfs_free_path(path);
2735 out:
2736 free_sa_defrag_extent(new);
2737
2738 atomic_dec(&fs_info->defrag_running);
2739 wake_up(&fs_info->transaction_wait);
2740 }
2741
2742 static struct new_sa_defrag_extent *
2743 record_old_file_extents(struct inode *inode,
2744 struct btrfs_ordered_extent *ordered)
2745 {
2746 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2747 struct btrfs_root *root = BTRFS_I(inode)->root;
2748 struct btrfs_path *path;
2749 struct btrfs_key key;
2750 struct old_sa_defrag_extent *old;
2751 struct new_sa_defrag_extent *new;
2752 int ret;
2753
2754 new = kmalloc(sizeof(*new), GFP_NOFS);
2755 if (!new)
2756 return NULL;
2757
2758 new->inode = inode;
2759 new->file_pos = ordered->file_offset;
2760 new->len = ordered->len;
2761 new->bytenr = ordered->start;
2762 new->disk_len = ordered->disk_len;
2763 new->compress_type = ordered->compress_type;
2764 new->root = RB_ROOT;
2765 INIT_LIST_HEAD(&new->head);
2766
2767 path = btrfs_alloc_path();
2768 if (!path)
2769 goto out_kfree;
2770
2771 key.objectid = btrfs_ino(inode);
2772 key.type = BTRFS_EXTENT_DATA_KEY;
2773 key.offset = new->file_pos;
2774
2775 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2776 if (ret < 0)
2777 goto out_free_path;
2778 if (ret > 0 && path->slots[0] > 0)
2779 path->slots[0]--;
2780
2781 /* find out all the old extents for the file range */
2782 while (1) {
2783 struct btrfs_file_extent_item *extent;
2784 struct extent_buffer *l;
2785 int slot;
2786 u64 num_bytes;
2787 u64 offset;
2788 u64 end;
2789 u64 disk_bytenr;
2790 u64 extent_offset;
2791
2792 l = path->nodes[0];
2793 slot = path->slots[0];
2794
2795 if (slot >= btrfs_header_nritems(l)) {
2796 ret = btrfs_next_leaf(root, path);
2797 if (ret < 0)
2798 goto out_free_path;
2799 else if (ret > 0)
2800 break;
2801 continue;
2802 }
2803
2804 btrfs_item_key_to_cpu(l, &key, slot);
2805
2806 if (key.objectid != btrfs_ino(inode))
2807 break;
2808 if (key.type != BTRFS_EXTENT_DATA_KEY)
2809 break;
2810 if (key.offset >= new->file_pos + new->len)
2811 break;
2812
2813 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2814
2815 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2816 if (key.offset + num_bytes < new->file_pos)
2817 goto next;
2818
2819 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2820 if (!disk_bytenr)
2821 goto next;
2822
2823 extent_offset = btrfs_file_extent_offset(l, extent);
2824
2825 old = kmalloc(sizeof(*old), GFP_NOFS);
2826 if (!old)
2827 goto out_free_path;
2828
2829 offset = max(new->file_pos, key.offset);
2830 end = min(new->file_pos + new->len, key.offset + num_bytes);
2831
2832 old->bytenr = disk_bytenr;
2833 old->extent_offset = extent_offset;
2834 old->offset = offset - key.offset;
2835 old->len = end - offset;
2836 old->new = new;
2837 old->count = 0;
2838 list_add_tail(&old->list, &new->head);
2839 next:
2840 path->slots[0]++;
2841 cond_resched();
2842 }
2843
2844 btrfs_free_path(path);
2845 atomic_inc(&fs_info->defrag_running);
2846
2847 return new;
2848
2849 out_free_path:
2850 btrfs_free_path(path);
2851 out_kfree:
2852 free_sa_defrag_extent(new);
2853 return NULL;
2854 }
2855
2856 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2857 u64 start, u64 len)
2858 {
2859 struct btrfs_block_group_cache *cache;
2860
2861 cache = btrfs_lookup_block_group(fs_info, start);
2862 ASSERT(cache);
2863
2864 spin_lock(&cache->lock);
2865 cache->delalloc_bytes -= len;
2866 spin_unlock(&cache->lock);
2867
2868 btrfs_put_block_group(cache);
2869 }
2870
2871 /* as ordered data IO finishes, this gets called so we can finish
2872 * an ordered extent if the range of bytes in the file it covers are
2873 * fully written.
2874 */
2875 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2876 {
2877 struct inode *inode = ordered_extent->inode;
2878 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2879 struct btrfs_root *root = BTRFS_I(inode)->root;
2880 struct btrfs_trans_handle *trans = NULL;
2881 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2882 struct extent_state *cached_state = NULL;
2883 struct new_sa_defrag_extent *new = NULL;
2884 int compress_type = 0;
2885 int ret = 0;
2886 u64 logical_len = ordered_extent->len;
2887 bool nolock;
2888 bool truncated = false;
2889
2890 nolock = btrfs_is_free_space_inode(inode);
2891
2892 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2893 ret = -EIO;
2894 goto out;
2895 }
2896
2897 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2898 ordered_extent->file_offset +
2899 ordered_extent->len - 1);
2900
2901 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2902 truncated = true;
2903 logical_len = ordered_extent->truncated_len;
2904 /* Truncated the entire extent, don't bother adding */
2905 if (!logical_len)
2906 goto out;
2907 }
2908
2909 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2910 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2911
2912 /*
2913 * For mwrite(mmap + memset to write) case, we still reserve
2914 * space for NOCOW range.
2915 * As NOCOW won't cause a new delayed ref, just free the space
2916 */
2917 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2918 ordered_extent->len);
2919 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2920 if (nolock)
2921 trans = btrfs_join_transaction_nolock(root);
2922 else
2923 trans = btrfs_join_transaction(root);
2924 if (IS_ERR(trans)) {
2925 ret = PTR_ERR(trans);
2926 trans = NULL;
2927 goto out;
2928 }
2929 trans->block_rsv = &fs_info->delalloc_block_rsv;
2930 ret = btrfs_update_inode_fallback(trans, root, inode);
2931 if (ret) /* -ENOMEM or corruption */
2932 btrfs_abort_transaction(trans, ret);
2933 goto out;
2934 }
2935
2936 lock_extent_bits(io_tree, ordered_extent->file_offset,
2937 ordered_extent->file_offset + ordered_extent->len - 1,
2938 &cached_state);
2939
2940 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2941 ordered_extent->file_offset + ordered_extent->len - 1,
2942 EXTENT_DEFRAG, 1, cached_state);
2943 if (ret) {
2944 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2945 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2946 /* the inode is shared */
2947 new = record_old_file_extents(inode, ordered_extent);
2948
2949 clear_extent_bit(io_tree, ordered_extent->file_offset,
2950 ordered_extent->file_offset + ordered_extent->len - 1,
2951 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2952 }
2953
2954 if (nolock)
2955 trans = btrfs_join_transaction_nolock(root);
2956 else
2957 trans = btrfs_join_transaction(root);
2958 if (IS_ERR(trans)) {
2959 ret = PTR_ERR(trans);
2960 trans = NULL;
2961 goto out_unlock;
2962 }
2963
2964 trans->block_rsv = &fs_info->delalloc_block_rsv;
2965
2966 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2967 compress_type = ordered_extent->compress_type;
2968 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2969 BUG_ON(compress_type);
2970 ret = btrfs_mark_extent_written(trans, inode,
2971 ordered_extent->file_offset,
2972 ordered_extent->file_offset +
2973 logical_len);
2974 } else {
2975 BUG_ON(root == fs_info->tree_root);
2976 ret = insert_reserved_file_extent(trans, inode,
2977 ordered_extent->file_offset,
2978 ordered_extent->start,
2979 ordered_extent->disk_len,
2980 logical_len, logical_len,
2981 compress_type, 0, 0,
2982 BTRFS_FILE_EXTENT_REG);
2983 if (!ret)
2984 btrfs_release_delalloc_bytes(fs_info,
2985 ordered_extent->start,
2986 ordered_extent->disk_len);
2987 }
2988 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2989 ordered_extent->file_offset, ordered_extent->len,
2990 trans->transid);
2991 if (ret < 0) {
2992 btrfs_abort_transaction(trans, ret);
2993 goto out_unlock;
2994 }
2995
2996 add_pending_csums(trans, inode, ordered_extent->file_offset,
2997 &ordered_extent->list);
2998
2999 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3000 ret = btrfs_update_inode_fallback(trans, root, inode);
3001 if (ret) { /* -ENOMEM or corruption */
3002 btrfs_abort_transaction(trans, ret);
3003 goto out_unlock;
3004 }
3005 ret = 0;
3006 out_unlock:
3007 unlock_extent_cached(io_tree, ordered_extent->file_offset,
3008 ordered_extent->file_offset +
3009 ordered_extent->len - 1, &cached_state, GFP_NOFS);
3010 out:
3011 if (root != fs_info->tree_root)
3012 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
3013 if (trans)
3014 btrfs_end_transaction(trans);
3015
3016 if (ret || truncated) {
3017 u64 start, end;
3018
3019 if (truncated)
3020 start = ordered_extent->file_offset + logical_len;
3021 else
3022 start = ordered_extent->file_offset;
3023 end = ordered_extent->file_offset + ordered_extent->len - 1;
3024 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
3025
3026 /* Drop the cache for the part of the extent we didn't write. */
3027 btrfs_drop_extent_cache(inode, start, end, 0);
3028
3029 /*
3030 * If the ordered extent had an IOERR or something else went
3031 * wrong we need to return the space for this ordered extent
3032 * back to the allocator. We only free the extent in the
3033 * truncated case if we didn't write out the extent at all.
3034 */
3035 if ((ret || !logical_len) &&
3036 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3037 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3038 btrfs_free_reserved_extent(fs_info,
3039 ordered_extent->start,
3040 ordered_extent->disk_len, 1);
3041 }
3042
3043
3044 /*
3045 * This needs to be done to make sure anybody waiting knows we are done
3046 * updating everything for this ordered extent.
3047 */
3048 btrfs_remove_ordered_extent(inode, ordered_extent);
3049
3050 /* for snapshot-aware defrag */
3051 if (new) {
3052 if (ret) {
3053 free_sa_defrag_extent(new);
3054 atomic_dec(&fs_info->defrag_running);
3055 } else {
3056 relink_file_extents(new);
3057 }
3058 }
3059
3060 /* once for us */
3061 btrfs_put_ordered_extent(ordered_extent);
3062 /* once for the tree */
3063 btrfs_put_ordered_extent(ordered_extent);
3064
3065 return ret;
3066 }
3067
3068 static void finish_ordered_fn(struct btrfs_work *work)
3069 {
3070 struct btrfs_ordered_extent *ordered_extent;
3071 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3072 btrfs_finish_ordered_io(ordered_extent);
3073 }
3074
3075 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3076 struct extent_state *state, int uptodate)
3077 {
3078 struct inode *inode = page->mapping->host;
3079 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3080 struct btrfs_ordered_extent *ordered_extent = NULL;
3081 struct btrfs_workqueue *wq;
3082 btrfs_work_func_t func;
3083
3084 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3085
3086 ClearPagePrivate2(page);
3087 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3088 end - start + 1, uptodate))
3089 return 0;
3090
3091 if (btrfs_is_free_space_inode(inode)) {
3092 wq = fs_info->endio_freespace_worker;
3093 func = btrfs_freespace_write_helper;
3094 } else {
3095 wq = fs_info->endio_write_workers;
3096 func = btrfs_endio_write_helper;
3097 }
3098
3099 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3100 NULL);
3101 btrfs_queue_work(wq, &ordered_extent->work);
3102
3103 return 0;
3104 }
3105
3106 static int __readpage_endio_check(struct inode *inode,
3107 struct btrfs_io_bio *io_bio,
3108 int icsum, struct page *page,
3109 int pgoff, u64 start, size_t len)
3110 {
3111 char *kaddr;
3112 u32 csum_expected;
3113 u32 csum = ~(u32)0;
3114
3115 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3116
3117 kaddr = kmap_atomic(page);
3118 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3119 btrfs_csum_final(csum, (u8 *)&csum);
3120 if (csum != csum_expected)
3121 goto zeroit;
3122
3123 kunmap_atomic(kaddr);
3124 return 0;
3125 zeroit:
3126 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3127 "csum failed ino %llu off %llu csum %u expected csum %u",
3128 btrfs_ino(inode), start, csum, csum_expected);
3129 memset(kaddr + pgoff, 1, len);
3130 flush_dcache_page(page);
3131 kunmap_atomic(kaddr);
3132 if (csum_expected == 0)
3133 return 0;
3134 return -EIO;
3135 }
3136
3137 /*
3138 * when reads are done, we need to check csums to verify the data is correct
3139 * if there's a match, we allow the bio to finish. If not, the code in
3140 * extent_io.c will try to find good copies for us.
3141 */
3142 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3143 u64 phy_offset, struct page *page,
3144 u64 start, u64 end, int mirror)
3145 {
3146 size_t offset = start - page_offset(page);
3147 struct inode *inode = page->mapping->host;
3148 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3149 struct btrfs_root *root = BTRFS_I(inode)->root;
3150
3151 if (PageChecked(page)) {
3152 ClearPageChecked(page);
3153 return 0;
3154 }
3155
3156 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3157 return 0;
3158
3159 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3160 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3161 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3162 return 0;
3163 }
3164
3165 phy_offset >>= inode->i_sb->s_blocksize_bits;
3166 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3167 start, (size_t)(end - start + 1));
3168 }
3169
3170 void btrfs_add_delayed_iput(struct inode *inode)
3171 {
3172 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3173 struct btrfs_inode *binode = BTRFS_I(inode);
3174
3175 if (atomic_add_unless(&inode->i_count, -1, 1))
3176 return;
3177
3178 spin_lock(&fs_info->delayed_iput_lock);
3179 if (binode->delayed_iput_count == 0) {
3180 ASSERT(list_empty(&binode->delayed_iput));
3181 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3182 } else {
3183 binode->delayed_iput_count++;
3184 }
3185 spin_unlock(&fs_info->delayed_iput_lock);
3186 }
3187
3188 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3189 {
3190
3191 spin_lock(&fs_info->delayed_iput_lock);
3192 while (!list_empty(&fs_info->delayed_iputs)) {
3193 struct btrfs_inode *inode;
3194
3195 inode = list_first_entry(&fs_info->delayed_iputs,
3196 struct btrfs_inode, delayed_iput);
3197 if (inode->delayed_iput_count) {
3198 inode->delayed_iput_count--;
3199 list_move_tail(&inode->delayed_iput,
3200 &fs_info->delayed_iputs);
3201 } else {
3202 list_del_init(&inode->delayed_iput);
3203 }
3204 spin_unlock(&fs_info->delayed_iput_lock);
3205 iput(&inode->vfs_inode);
3206 spin_lock(&fs_info->delayed_iput_lock);
3207 }
3208 spin_unlock(&fs_info->delayed_iput_lock);
3209 }
3210
3211 /*
3212 * This is called in transaction commit time. If there are no orphan
3213 * files in the subvolume, it removes orphan item and frees block_rsv
3214 * structure.
3215 */
3216 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3217 struct btrfs_root *root)
3218 {
3219 struct btrfs_fs_info *fs_info = root->fs_info;
3220 struct btrfs_block_rsv *block_rsv;
3221 int ret;
3222
3223 if (atomic_read(&root->orphan_inodes) ||
3224 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3225 return;
3226
3227 spin_lock(&root->orphan_lock);
3228 if (atomic_read(&root->orphan_inodes)) {
3229 spin_unlock(&root->orphan_lock);
3230 return;
3231 }
3232
3233 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3234 spin_unlock(&root->orphan_lock);
3235 return;
3236 }
3237
3238 block_rsv = root->orphan_block_rsv;
3239 root->orphan_block_rsv = NULL;
3240 spin_unlock(&root->orphan_lock);
3241
3242 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3243 btrfs_root_refs(&root->root_item) > 0) {
3244 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3245 root->root_key.objectid);
3246 if (ret)
3247 btrfs_abort_transaction(trans, ret);
3248 else
3249 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3250 &root->state);
3251 }
3252
3253 if (block_rsv) {
3254 WARN_ON(block_rsv->size > 0);
3255 btrfs_free_block_rsv(fs_info, block_rsv);
3256 }
3257 }
3258
3259 /*
3260 * This creates an orphan entry for the given inode in case something goes
3261 * wrong in the middle of an unlink/truncate.
3262 *
3263 * NOTE: caller of this function should reserve 5 units of metadata for
3264 * this function.
3265 */
3266 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3267 {
3268 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3269 struct btrfs_root *root = BTRFS_I(inode)->root;
3270 struct btrfs_block_rsv *block_rsv = NULL;
3271 int reserve = 0;
3272 int insert = 0;
3273 int ret;
3274
3275 if (!root->orphan_block_rsv) {
3276 block_rsv = btrfs_alloc_block_rsv(fs_info,
3277 BTRFS_BLOCK_RSV_TEMP);
3278 if (!block_rsv)
3279 return -ENOMEM;
3280 }
3281
3282 spin_lock(&root->orphan_lock);
3283 if (!root->orphan_block_rsv) {
3284 root->orphan_block_rsv = block_rsv;
3285 } else if (block_rsv) {
3286 btrfs_free_block_rsv(fs_info, block_rsv);
3287 block_rsv = NULL;
3288 }
3289
3290 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3291 &BTRFS_I(inode)->runtime_flags)) {
3292 #if 0
3293 /*
3294 * For proper ENOSPC handling, we should do orphan
3295 * cleanup when mounting. But this introduces backward
3296 * compatibility issue.
3297 */
3298 if (!xchg(&root->orphan_item_inserted, 1))
3299 insert = 2;
3300 else
3301 insert = 1;
3302 #endif
3303 insert = 1;
3304 atomic_inc(&root->orphan_inodes);
3305 }
3306
3307 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3308 &BTRFS_I(inode)->runtime_flags))
3309 reserve = 1;
3310 spin_unlock(&root->orphan_lock);
3311
3312 /* grab metadata reservation from transaction handle */
3313 if (reserve) {
3314 ret = btrfs_orphan_reserve_metadata(trans, inode);
3315 ASSERT(!ret);
3316 if (ret) {
3317 atomic_dec(&root->orphan_inodes);
3318 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3319 &BTRFS_I(inode)->runtime_flags);
3320 if (insert)
3321 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3322 &BTRFS_I(inode)->runtime_flags);
3323 return ret;
3324 }
3325 }
3326
3327 /* insert an orphan item to track this unlinked/truncated file */
3328 if (insert >= 1) {
3329 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3330 if (ret) {
3331 atomic_dec(&root->orphan_inodes);
3332 if (reserve) {
3333 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3334 &BTRFS_I(inode)->runtime_flags);
3335 btrfs_orphan_release_metadata(inode);
3336 }
3337 if (ret != -EEXIST) {
3338 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3339 &BTRFS_I(inode)->runtime_flags);
3340 btrfs_abort_transaction(trans, ret);
3341 return ret;
3342 }
3343 }
3344 ret = 0;
3345 }
3346
3347 /* insert an orphan item to track subvolume contains orphan files */
3348 if (insert >= 2) {
3349 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3350 root->root_key.objectid);
3351 if (ret && ret != -EEXIST) {
3352 btrfs_abort_transaction(trans, ret);
3353 return ret;
3354 }
3355 }
3356 return 0;
3357 }
3358
3359 /*
3360 * We have done the truncate/delete so we can go ahead and remove the orphan
3361 * item for this particular inode.
3362 */
3363 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3364 struct inode *inode)
3365 {
3366 struct btrfs_root *root = BTRFS_I(inode)->root;
3367 int delete_item = 0;
3368 int release_rsv = 0;
3369 int ret = 0;
3370
3371 spin_lock(&root->orphan_lock);
3372 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3373 &BTRFS_I(inode)->runtime_flags))
3374 delete_item = 1;
3375
3376 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3377 &BTRFS_I(inode)->runtime_flags))
3378 release_rsv = 1;
3379 spin_unlock(&root->orphan_lock);
3380
3381 if (delete_item) {
3382 atomic_dec(&root->orphan_inodes);
3383 if (trans)
3384 ret = btrfs_del_orphan_item(trans, root,
3385 btrfs_ino(inode));
3386 }
3387
3388 if (release_rsv)
3389 btrfs_orphan_release_metadata(inode);
3390
3391 return ret;
3392 }
3393
3394 /*
3395 * this cleans up any orphans that may be left on the list from the last use
3396 * of this root.
3397 */
3398 int btrfs_orphan_cleanup(struct btrfs_root *root)
3399 {
3400 struct btrfs_fs_info *fs_info = root->fs_info;
3401 struct btrfs_path *path;
3402 struct extent_buffer *leaf;
3403 struct btrfs_key key, found_key;
3404 struct btrfs_trans_handle *trans;
3405 struct inode *inode;
3406 u64 last_objectid = 0;
3407 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3408
3409 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3410 return 0;
3411
3412 path = btrfs_alloc_path();
3413 if (!path) {
3414 ret = -ENOMEM;
3415 goto out;
3416 }
3417 path->reada = READA_BACK;
3418
3419 key.objectid = BTRFS_ORPHAN_OBJECTID;
3420 key.type = BTRFS_ORPHAN_ITEM_KEY;
3421 key.offset = (u64)-1;
3422
3423 while (1) {
3424 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3425 if (ret < 0)
3426 goto out;
3427
3428 /*
3429 * if ret == 0 means we found what we were searching for, which
3430 * is weird, but possible, so only screw with path if we didn't
3431 * find the key and see if we have stuff that matches
3432 */
3433 if (ret > 0) {
3434 ret = 0;
3435 if (path->slots[0] == 0)
3436 break;
3437 path->slots[0]--;
3438 }
3439
3440 /* pull out the item */
3441 leaf = path->nodes[0];
3442 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3443
3444 /* make sure the item matches what we want */
3445 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3446 break;
3447 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3448 break;
3449
3450 /* release the path since we're done with it */
3451 btrfs_release_path(path);
3452
3453 /*
3454 * this is where we are basically btrfs_lookup, without the
3455 * crossing root thing. we store the inode number in the
3456 * offset of the orphan item.
3457 */
3458
3459 if (found_key.offset == last_objectid) {
3460 btrfs_err(fs_info,
3461 "Error removing orphan entry, stopping orphan cleanup");
3462 ret = -EINVAL;
3463 goto out;
3464 }
3465
3466 last_objectid = found_key.offset;
3467
3468 found_key.objectid = found_key.offset;
3469 found_key.type = BTRFS_INODE_ITEM_KEY;
3470 found_key.offset = 0;
3471 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3472 ret = PTR_ERR_OR_ZERO(inode);
3473 if (ret && ret != -ENOENT)
3474 goto out;
3475
3476 if (ret == -ENOENT && root == fs_info->tree_root) {
3477 struct btrfs_root *dead_root;
3478 struct btrfs_fs_info *fs_info = root->fs_info;
3479 int is_dead_root = 0;
3480
3481 /*
3482 * this is an orphan in the tree root. Currently these
3483 * could come from 2 sources:
3484 * a) a snapshot deletion in progress
3485 * b) a free space cache inode
3486 * We need to distinguish those two, as the snapshot
3487 * orphan must not get deleted.
3488 * find_dead_roots already ran before us, so if this
3489 * is a snapshot deletion, we should find the root
3490 * in the dead_roots list
3491 */
3492 spin_lock(&fs_info->trans_lock);
3493 list_for_each_entry(dead_root, &fs_info->dead_roots,
3494 root_list) {
3495 if (dead_root->root_key.objectid ==
3496 found_key.objectid) {
3497 is_dead_root = 1;
3498 break;
3499 }
3500 }
3501 spin_unlock(&fs_info->trans_lock);
3502 if (is_dead_root) {
3503 /* prevent this orphan from being found again */
3504 key.offset = found_key.objectid - 1;
3505 continue;
3506 }
3507 }
3508 /*
3509 * Inode is already gone but the orphan item is still there,
3510 * kill the orphan item.
3511 */
3512 if (ret == -ENOENT) {
3513 trans = btrfs_start_transaction(root, 1);
3514 if (IS_ERR(trans)) {
3515 ret = PTR_ERR(trans);
3516 goto out;
3517 }
3518 btrfs_debug(fs_info, "auto deleting %Lu",
3519 found_key.objectid);
3520 ret = btrfs_del_orphan_item(trans, root,
3521 found_key.objectid);
3522 btrfs_end_transaction(trans);
3523 if (ret)
3524 goto out;
3525 continue;
3526 }
3527
3528 /*
3529 * add this inode to the orphan list so btrfs_orphan_del does
3530 * the proper thing when we hit it
3531 */
3532 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3533 &BTRFS_I(inode)->runtime_flags);
3534 atomic_inc(&root->orphan_inodes);
3535
3536 /* if we have links, this was a truncate, lets do that */
3537 if (inode->i_nlink) {
3538 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3539 iput(inode);
3540 continue;
3541 }
3542 nr_truncate++;
3543
3544 /* 1 for the orphan item deletion. */
3545 trans = btrfs_start_transaction(root, 1);
3546 if (IS_ERR(trans)) {
3547 iput(inode);
3548 ret = PTR_ERR(trans);
3549 goto out;
3550 }
3551 ret = btrfs_orphan_add(trans, inode);
3552 btrfs_end_transaction(trans);
3553 if (ret) {
3554 iput(inode);
3555 goto out;
3556 }
3557
3558 ret = btrfs_truncate(inode);
3559 if (ret)
3560 btrfs_orphan_del(NULL, inode);
3561 } else {
3562 nr_unlink++;
3563 }
3564
3565 /* this will do delete_inode and everything for us */
3566 iput(inode);
3567 if (ret)
3568 goto out;
3569 }
3570 /* release the path since we're done with it */
3571 btrfs_release_path(path);
3572
3573 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3574
3575 if (root->orphan_block_rsv)
3576 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3577 (u64)-1);
3578
3579 if (root->orphan_block_rsv ||
3580 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3581 trans = btrfs_join_transaction(root);
3582 if (!IS_ERR(trans))
3583 btrfs_end_transaction(trans);
3584 }
3585
3586 if (nr_unlink)
3587 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3588 if (nr_truncate)
3589 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3590
3591 out:
3592 if (ret)
3593 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3594 btrfs_free_path(path);
3595 return ret;
3596 }
3597
3598 /*
3599 * very simple check to peek ahead in the leaf looking for xattrs. If we
3600 * don't find any xattrs, we know there can't be any acls.
3601 *
3602 * slot is the slot the inode is in, objectid is the objectid of the inode
3603 */
3604 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3605 int slot, u64 objectid,
3606 int *first_xattr_slot)
3607 {
3608 u32 nritems = btrfs_header_nritems(leaf);
3609 struct btrfs_key found_key;
3610 static u64 xattr_access = 0;
3611 static u64 xattr_default = 0;
3612 int scanned = 0;
3613
3614 if (!xattr_access) {
3615 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3616 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3617 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3618 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3619 }
3620
3621 slot++;
3622 *first_xattr_slot = -1;
3623 while (slot < nritems) {
3624 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3625
3626 /* we found a different objectid, there must not be acls */
3627 if (found_key.objectid != objectid)
3628 return 0;
3629
3630 /* we found an xattr, assume we've got an acl */
3631 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3632 if (*first_xattr_slot == -1)
3633 *first_xattr_slot = slot;
3634 if (found_key.offset == xattr_access ||
3635 found_key.offset == xattr_default)
3636 return 1;
3637 }
3638
3639 /*
3640 * we found a key greater than an xattr key, there can't
3641 * be any acls later on
3642 */
3643 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3644 return 0;
3645
3646 slot++;
3647 scanned++;
3648
3649 /*
3650 * it goes inode, inode backrefs, xattrs, extents,
3651 * so if there are a ton of hard links to an inode there can
3652 * be a lot of backrefs. Don't waste time searching too hard,
3653 * this is just an optimization
3654 */
3655 if (scanned >= 8)
3656 break;
3657 }
3658 /* we hit the end of the leaf before we found an xattr or
3659 * something larger than an xattr. We have to assume the inode
3660 * has acls
3661 */
3662 if (*first_xattr_slot == -1)
3663 *first_xattr_slot = slot;
3664 return 1;
3665 }
3666
3667 /*
3668 * read an inode from the btree into the in-memory inode
3669 */
3670 static int btrfs_read_locked_inode(struct inode *inode)
3671 {
3672 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3673 struct btrfs_path *path;
3674 struct extent_buffer *leaf;
3675 struct btrfs_inode_item *inode_item;
3676 struct btrfs_root *root = BTRFS_I(inode)->root;
3677 struct btrfs_key location;
3678 unsigned long ptr;
3679 int maybe_acls;
3680 u32 rdev;
3681 int ret;
3682 bool filled = false;
3683 int first_xattr_slot;
3684
3685 ret = btrfs_fill_inode(inode, &rdev);
3686 if (!ret)
3687 filled = true;
3688
3689 path = btrfs_alloc_path();
3690 if (!path) {
3691 ret = -ENOMEM;
3692 goto make_bad;
3693 }
3694
3695 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3696
3697 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3698 if (ret) {
3699 if (ret > 0)
3700 ret = -ENOENT;
3701 goto make_bad;
3702 }
3703
3704 leaf = path->nodes[0];
3705
3706 if (filled)
3707 goto cache_index;
3708
3709 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3710 struct btrfs_inode_item);
3711 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3712 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3713 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3714 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3715 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3716
3717 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3718 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3719
3720 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3721 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3722
3723 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3724 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3725
3726 BTRFS_I(inode)->i_otime.tv_sec =
3727 btrfs_timespec_sec(leaf, &inode_item->otime);
3728 BTRFS_I(inode)->i_otime.tv_nsec =
3729 btrfs_timespec_nsec(leaf, &inode_item->otime);
3730
3731 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3732 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3733 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3734
3735 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3736 inode->i_generation = BTRFS_I(inode)->generation;
3737 inode->i_rdev = 0;
3738 rdev = btrfs_inode_rdev(leaf, inode_item);
3739
3740 BTRFS_I(inode)->index_cnt = (u64)-1;
3741 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3742
3743 cache_index:
3744 /*
3745 * If we were modified in the current generation and evicted from memory
3746 * and then re-read we need to do a full sync since we don't have any
3747 * idea about which extents were modified before we were evicted from
3748 * cache.
3749 *
3750 * This is required for both inode re-read from disk and delayed inode
3751 * in delayed_nodes_tree.
3752 */
3753 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3754 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3755 &BTRFS_I(inode)->runtime_flags);
3756
3757 /*
3758 * We don't persist the id of the transaction where an unlink operation
3759 * against the inode was last made. So here we assume the inode might
3760 * have been evicted, and therefore the exact value of last_unlink_trans
3761 * lost, and set it to last_trans to avoid metadata inconsistencies
3762 * between the inode and its parent if the inode is fsync'ed and the log
3763 * replayed. For example, in the scenario:
3764 *
3765 * touch mydir/foo
3766 * ln mydir/foo mydir/bar
3767 * sync
3768 * unlink mydir/bar
3769 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3770 * xfs_io -c fsync mydir/foo
3771 * <power failure>
3772 * mount fs, triggers fsync log replay
3773 *
3774 * We must make sure that when we fsync our inode foo we also log its
3775 * parent inode, otherwise after log replay the parent still has the
3776 * dentry with the "bar" name but our inode foo has a link count of 1
3777 * and doesn't have an inode ref with the name "bar" anymore.
3778 *
3779 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3780 * but it guarantees correctness at the expense of occasional full
3781 * transaction commits on fsync if our inode is a directory, or if our
3782 * inode is not a directory, logging its parent unnecessarily.
3783 */
3784 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3785
3786 path->slots[0]++;
3787 if (inode->i_nlink != 1 ||
3788 path->slots[0] >= btrfs_header_nritems(leaf))
3789 goto cache_acl;
3790
3791 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3792 if (location.objectid != btrfs_ino(inode))
3793 goto cache_acl;
3794
3795 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3796 if (location.type == BTRFS_INODE_REF_KEY) {
3797 struct btrfs_inode_ref *ref;
3798
3799 ref = (struct btrfs_inode_ref *)ptr;
3800 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3801 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3802 struct btrfs_inode_extref *extref;
3803
3804 extref = (struct btrfs_inode_extref *)ptr;
3805 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3806 extref);
3807 }
3808 cache_acl:
3809 /*
3810 * try to precache a NULL acl entry for files that don't have
3811 * any xattrs or acls
3812 */
3813 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3814 btrfs_ino(inode), &first_xattr_slot);
3815 if (first_xattr_slot != -1) {
3816 path->slots[0] = first_xattr_slot;
3817 ret = btrfs_load_inode_props(inode, path);
3818 if (ret)
3819 btrfs_err(fs_info,
3820 "error loading props for ino %llu (root %llu): %d",
3821 btrfs_ino(inode),
3822 root->root_key.objectid, ret);
3823 }
3824 btrfs_free_path(path);
3825
3826 if (!maybe_acls)
3827 cache_no_acl(inode);
3828
3829 switch (inode->i_mode & S_IFMT) {
3830 case S_IFREG:
3831 inode->i_mapping->a_ops = &btrfs_aops;
3832 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3833 inode->i_fop = &btrfs_file_operations;
3834 inode->i_op = &btrfs_file_inode_operations;
3835 break;
3836 case S_IFDIR:
3837 inode->i_fop = &btrfs_dir_file_operations;
3838 inode->i_op = &btrfs_dir_inode_operations;
3839 break;
3840 case S_IFLNK:
3841 inode->i_op = &btrfs_symlink_inode_operations;
3842 inode_nohighmem(inode);
3843 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3844 break;
3845 default:
3846 inode->i_op = &btrfs_special_inode_operations;
3847 init_special_inode(inode, inode->i_mode, rdev);
3848 break;
3849 }
3850
3851 btrfs_update_iflags(inode);
3852 return 0;
3853
3854 make_bad:
3855 btrfs_free_path(path);
3856 make_bad_inode(inode);
3857 return ret;
3858 }
3859
3860 /*
3861 * given a leaf and an inode, copy the inode fields into the leaf
3862 */
3863 static void fill_inode_item(struct btrfs_trans_handle *trans,
3864 struct extent_buffer *leaf,
3865 struct btrfs_inode_item *item,
3866 struct inode *inode)
3867 {
3868 struct btrfs_map_token token;
3869
3870 btrfs_init_map_token(&token);
3871
3872 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3873 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3874 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3875 &token);
3876 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3877 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3878
3879 btrfs_set_token_timespec_sec(leaf, &item->atime,
3880 inode->i_atime.tv_sec, &token);
3881 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3882 inode->i_atime.tv_nsec, &token);
3883
3884 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3885 inode->i_mtime.tv_sec, &token);
3886 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3887 inode->i_mtime.tv_nsec, &token);
3888
3889 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3890 inode->i_ctime.tv_sec, &token);
3891 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3892 inode->i_ctime.tv_nsec, &token);
3893
3894 btrfs_set_token_timespec_sec(leaf, &item->otime,
3895 BTRFS_I(inode)->i_otime.tv_sec, &token);
3896 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3897 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3898
3899 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3900 &token);
3901 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3902 &token);
3903 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3904 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3905 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3906 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3907 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3908 }
3909
3910 /*
3911 * copy everything in the in-memory inode into the btree.
3912 */
3913 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3914 struct btrfs_root *root, struct inode *inode)
3915 {
3916 struct btrfs_inode_item *inode_item;
3917 struct btrfs_path *path;
3918 struct extent_buffer *leaf;
3919 int ret;
3920
3921 path = btrfs_alloc_path();
3922 if (!path)
3923 return -ENOMEM;
3924
3925 path->leave_spinning = 1;
3926 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3927 1);
3928 if (ret) {
3929 if (ret > 0)
3930 ret = -ENOENT;
3931 goto failed;
3932 }
3933
3934 leaf = path->nodes[0];
3935 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3936 struct btrfs_inode_item);
3937
3938 fill_inode_item(trans, leaf, inode_item, inode);
3939 btrfs_mark_buffer_dirty(leaf);
3940 btrfs_set_inode_last_trans(trans, inode);
3941 ret = 0;
3942 failed:
3943 btrfs_free_path(path);
3944 return ret;
3945 }
3946
3947 /*
3948 * copy everything in the in-memory inode into the btree.
3949 */
3950 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3951 struct btrfs_root *root, struct inode *inode)
3952 {
3953 struct btrfs_fs_info *fs_info = root->fs_info;
3954 int ret;
3955
3956 /*
3957 * If the inode is a free space inode, we can deadlock during commit
3958 * if we put it into the delayed code.
3959 *
3960 * The data relocation inode should also be directly updated
3961 * without delay
3962 */
3963 if (!btrfs_is_free_space_inode(inode)
3964 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3965 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3966 btrfs_update_root_times(trans, root);
3967
3968 ret = btrfs_delayed_update_inode(trans, root, inode);
3969 if (!ret)
3970 btrfs_set_inode_last_trans(trans, inode);
3971 return ret;
3972 }
3973
3974 return btrfs_update_inode_item(trans, root, inode);
3975 }
3976
3977 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3978 struct btrfs_root *root,
3979 struct inode *inode)
3980 {
3981 int ret;
3982
3983 ret = btrfs_update_inode(trans, root, inode);
3984 if (ret == -ENOSPC)
3985 return btrfs_update_inode_item(trans, root, inode);
3986 return ret;
3987 }
3988
3989 /*
3990 * unlink helper that gets used here in inode.c and in the tree logging
3991 * recovery code. It remove a link in a directory with a given name, and
3992 * also drops the back refs in the inode to the directory
3993 */
3994 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3995 struct btrfs_root *root,
3996 struct inode *dir, struct inode *inode,
3997 const char *name, int name_len)
3998 {
3999 struct btrfs_fs_info *fs_info = root->fs_info;
4000 struct btrfs_path *path;
4001 int ret = 0;
4002 struct extent_buffer *leaf;
4003 struct btrfs_dir_item *di;
4004 struct btrfs_key key;
4005 u64 index;
4006 u64 ino = btrfs_ino(inode);
4007 u64 dir_ino = btrfs_ino(dir);
4008
4009 path = btrfs_alloc_path();
4010 if (!path) {
4011 ret = -ENOMEM;
4012 goto out;
4013 }
4014
4015 path->leave_spinning = 1;
4016 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4017 name, name_len, -1);
4018 if (IS_ERR(di)) {
4019 ret = PTR_ERR(di);
4020 goto err;
4021 }
4022 if (!di) {
4023 ret = -ENOENT;
4024 goto err;
4025 }
4026 leaf = path->nodes[0];
4027 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4028 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4029 if (ret)
4030 goto err;
4031 btrfs_release_path(path);
4032
4033 /*
4034 * If we don't have dir index, we have to get it by looking up
4035 * the inode ref, since we get the inode ref, remove it directly,
4036 * it is unnecessary to do delayed deletion.
4037 *
4038 * But if we have dir index, needn't search inode ref to get it.
4039 * Since the inode ref is close to the inode item, it is better
4040 * that we delay to delete it, and just do this deletion when
4041 * we update the inode item.
4042 */
4043 if (BTRFS_I(inode)->dir_index) {
4044 ret = btrfs_delayed_delete_inode_ref(inode);
4045 if (!ret) {
4046 index = BTRFS_I(inode)->dir_index;
4047 goto skip_backref;
4048 }
4049 }
4050
4051 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4052 dir_ino, &index);
4053 if (ret) {
4054 btrfs_info(fs_info,
4055 "failed to delete reference to %.*s, inode %llu parent %llu",
4056 name_len, name, ino, dir_ino);
4057 btrfs_abort_transaction(trans, ret);
4058 goto err;
4059 }
4060 skip_backref:
4061 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4062 if (ret) {
4063 btrfs_abort_transaction(trans, ret);
4064 goto err;
4065 }
4066
4067 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
4068 inode, dir_ino);
4069 if (ret != 0 && ret != -ENOENT) {
4070 btrfs_abort_transaction(trans, ret);
4071 goto err;
4072 }
4073
4074 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4075 dir, index);
4076 if (ret == -ENOENT)
4077 ret = 0;
4078 else if (ret)
4079 btrfs_abort_transaction(trans, ret);
4080 err:
4081 btrfs_free_path(path);
4082 if (ret)
4083 goto out;
4084
4085 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4086 inode_inc_iversion(inode);
4087 inode_inc_iversion(dir);
4088 inode->i_ctime = dir->i_mtime =
4089 dir->i_ctime = current_time(inode);
4090 ret = btrfs_update_inode(trans, root, dir);
4091 out:
4092 return ret;
4093 }
4094
4095 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4096 struct btrfs_root *root,
4097 struct inode *dir, struct inode *inode,
4098 const char *name, int name_len)
4099 {
4100 int ret;
4101 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4102 if (!ret) {
4103 drop_nlink(inode);
4104 ret = btrfs_update_inode(trans, root, inode);
4105 }
4106 return ret;
4107 }
4108
4109 /*
4110 * helper to start transaction for unlink and rmdir.
4111 *
4112 * unlink and rmdir are special in btrfs, they do not always free space, so
4113 * if we cannot make our reservations the normal way try and see if there is
4114 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4115 * allow the unlink to occur.
4116 */
4117 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4118 {
4119 struct btrfs_root *root = BTRFS_I(dir)->root;
4120
4121 /*
4122 * 1 for the possible orphan item
4123 * 1 for the dir item
4124 * 1 for the dir index
4125 * 1 for the inode ref
4126 * 1 for the inode
4127 */
4128 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4129 }
4130
4131 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4132 {
4133 struct btrfs_root *root = BTRFS_I(dir)->root;
4134 struct btrfs_trans_handle *trans;
4135 struct inode *inode = d_inode(dentry);
4136 int ret;
4137
4138 trans = __unlink_start_trans(dir);
4139 if (IS_ERR(trans))
4140 return PTR_ERR(trans);
4141
4142 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4143
4144 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4145 dentry->d_name.name, dentry->d_name.len);
4146 if (ret)
4147 goto out;
4148
4149 if (inode->i_nlink == 0) {
4150 ret = btrfs_orphan_add(trans, inode);
4151 if (ret)
4152 goto out;
4153 }
4154
4155 out:
4156 btrfs_end_transaction(trans);
4157 btrfs_btree_balance_dirty(root->fs_info);
4158 return ret;
4159 }
4160
4161 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4162 struct btrfs_root *root,
4163 struct inode *dir, u64 objectid,
4164 const char *name, int name_len)
4165 {
4166 struct btrfs_fs_info *fs_info = root->fs_info;
4167 struct btrfs_path *path;
4168 struct extent_buffer *leaf;
4169 struct btrfs_dir_item *di;
4170 struct btrfs_key key;
4171 u64 index;
4172 int ret;
4173 u64 dir_ino = btrfs_ino(dir);
4174
4175 path = btrfs_alloc_path();
4176 if (!path)
4177 return -ENOMEM;
4178
4179 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4180 name, name_len, -1);
4181 if (IS_ERR_OR_NULL(di)) {
4182 if (!di)
4183 ret = -ENOENT;
4184 else
4185 ret = PTR_ERR(di);
4186 goto out;
4187 }
4188
4189 leaf = path->nodes[0];
4190 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4191 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4192 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4193 if (ret) {
4194 btrfs_abort_transaction(trans, ret);
4195 goto out;
4196 }
4197 btrfs_release_path(path);
4198
4199 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4200 root->root_key.objectid, dir_ino,
4201 &index, name, name_len);
4202 if (ret < 0) {
4203 if (ret != -ENOENT) {
4204 btrfs_abort_transaction(trans, ret);
4205 goto out;
4206 }
4207 di = btrfs_search_dir_index_item(root, path, dir_ino,
4208 name, name_len);
4209 if (IS_ERR_OR_NULL(di)) {
4210 if (!di)
4211 ret = -ENOENT;
4212 else
4213 ret = PTR_ERR(di);
4214 btrfs_abort_transaction(trans, ret);
4215 goto out;
4216 }
4217
4218 leaf = path->nodes[0];
4219 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4220 btrfs_release_path(path);
4221 index = key.offset;
4222 }
4223 btrfs_release_path(path);
4224
4225 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4226 if (ret) {
4227 btrfs_abort_transaction(trans, ret);
4228 goto out;
4229 }
4230
4231 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4232 inode_inc_iversion(dir);
4233 dir->i_mtime = dir->i_ctime = current_time(dir);
4234 ret = btrfs_update_inode_fallback(trans, root, dir);
4235 if (ret)
4236 btrfs_abort_transaction(trans, ret);
4237 out:
4238 btrfs_free_path(path);
4239 return ret;
4240 }
4241
4242 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4243 {
4244 struct inode *inode = d_inode(dentry);
4245 int err = 0;
4246 struct btrfs_root *root = BTRFS_I(dir)->root;
4247 struct btrfs_trans_handle *trans;
4248 u64 last_unlink_trans;
4249
4250 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4251 return -ENOTEMPTY;
4252 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4253 return -EPERM;
4254
4255 trans = __unlink_start_trans(dir);
4256 if (IS_ERR(trans))
4257 return PTR_ERR(trans);
4258
4259 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4260 err = btrfs_unlink_subvol(trans, root, dir,
4261 BTRFS_I(inode)->location.objectid,
4262 dentry->d_name.name,
4263 dentry->d_name.len);
4264 goto out;
4265 }
4266
4267 err = btrfs_orphan_add(trans, inode);
4268 if (err)
4269 goto out;
4270
4271 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4272
4273 /* now the directory is empty */
4274 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4275 dentry->d_name.name, dentry->d_name.len);
4276 if (!err) {
4277 btrfs_i_size_write(inode, 0);
4278 /*
4279 * Propagate the last_unlink_trans value of the deleted dir to
4280 * its parent directory. This is to prevent an unrecoverable
4281 * log tree in the case we do something like this:
4282 * 1) create dir foo
4283 * 2) create snapshot under dir foo
4284 * 3) delete the snapshot
4285 * 4) rmdir foo
4286 * 5) mkdir foo
4287 * 6) fsync foo or some file inside foo
4288 */
4289 if (last_unlink_trans >= trans->transid)
4290 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4291 }
4292 out:
4293 btrfs_end_transaction(trans);
4294 btrfs_btree_balance_dirty(root->fs_info);
4295
4296 return err;
4297 }
4298
4299 static int truncate_space_check(struct btrfs_trans_handle *trans,
4300 struct btrfs_root *root,
4301 u64 bytes_deleted)
4302 {
4303 struct btrfs_fs_info *fs_info = root->fs_info;
4304 int ret;
4305
4306 /*
4307 * This is only used to apply pressure to the enospc system, we don't
4308 * intend to use this reservation at all.
4309 */
4310 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4311 bytes_deleted *= fs_info->nodesize;
4312 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4313 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4314 if (!ret) {
4315 trace_btrfs_space_reservation(fs_info, "transaction",
4316 trans->transid,
4317 bytes_deleted, 1);
4318 trans->bytes_reserved += bytes_deleted;
4319 }
4320 return ret;
4321
4322 }
4323
4324 static int truncate_inline_extent(struct inode *inode,
4325 struct btrfs_path *path,
4326 struct btrfs_key *found_key,
4327 const u64 item_end,
4328 const u64 new_size)
4329 {
4330 struct extent_buffer *leaf = path->nodes[0];
4331 int slot = path->slots[0];
4332 struct btrfs_file_extent_item *fi;
4333 u32 size = (u32)(new_size - found_key->offset);
4334 struct btrfs_root *root = BTRFS_I(inode)->root;
4335
4336 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4337
4338 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4339 loff_t offset = new_size;
4340 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4341
4342 /*
4343 * Zero out the remaining of the last page of our inline extent,
4344 * instead of directly truncating our inline extent here - that
4345 * would be much more complex (decompressing all the data, then
4346 * compressing the truncated data, which might be bigger than
4347 * the size of the inline extent, resize the extent, etc).
4348 * We release the path because to get the page we might need to
4349 * read the extent item from disk (data not in the page cache).
4350 */
4351 btrfs_release_path(path);
4352 return btrfs_truncate_block(inode, offset, page_end - offset,
4353 0);
4354 }
4355
4356 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4357 size = btrfs_file_extent_calc_inline_size(size);
4358 btrfs_truncate_item(root->fs_info, path, size, 1);
4359
4360 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4361 inode_sub_bytes(inode, item_end + 1 - new_size);
4362
4363 return 0;
4364 }
4365
4366 /*
4367 * this can truncate away extent items, csum items and directory items.
4368 * It starts at a high offset and removes keys until it can't find
4369 * any higher than new_size
4370 *
4371 * csum items that cross the new i_size are truncated to the new size
4372 * as well.
4373 *
4374 * min_type is the minimum key type to truncate down to. If set to 0, this
4375 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4376 */
4377 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4378 struct btrfs_root *root,
4379 struct inode *inode,
4380 u64 new_size, u32 min_type)
4381 {
4382 struct btrfs_fs_info *fs_info = root->fs_info;
4383 struct btrfs_path *path;
4384 struct extent_buffer *leaf;
4385 struct btrfs_file_extent_item *fi;
4386 struct btrfs_key key;
4387 struct btrfs_key found_key;
4388 u64 extent_start = 0;
4389 u64 extent_num_bytes = 0;
4390 u64 extent_offset = 0;
4391 u64 item_end = 0;
4392 u64 last_size = new_size;
4393 u32 found_type = (u8)-1;
4394 int found_extent;
4395 int del_item;
4396 int pending_del_nr = 0;
4397 int pending_del_slot = 0;
4398 int extent_type = -1;
4399 int ret;
4400 int err = 0;
4401 u64 ino = btrfs_ino(inode);
4402 u64 bytes_deleted = 0;
4403 bool be_nice = 0;
4404 bool should_throttle = 0;
4405 bool should_end = 0;
4406
4407 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4408
4409 /*
4410 * for non-free space inodes and ref cows, we want to back off from
4411 * time to time
4412 */
4413 if (!btrfs_is_free_space_inode(inode) &&
4414 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4415 be_nice = 1;
4416
4417 path = btrfs_alloc_path();
4418 if (!path)
4419 return -ENOMEM;
4420 path->reada = READA_BACK;
4421
4422 /*
4423 * We want to drop from the next block forward in case this new size is
4424 * not block aligned since we will be keeping the last block of the
4425 * extent just the way it is.
4426 */
4427 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4428 root == fs_info->tree_root)
4429 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4430 fs_info->sectorsize),
4431 (u64)-1, 0);
4432
4433 /*
4434 * This function is also used to drop the items in the log tree before
4435 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4436 * it is used to drop the loged items. So we shouldn't kill the delayed
4437 * items.
4438 */
4439 if (min_type == 0 && root == BTRFS_I(inode)->root)
4440 btrfs_kill_delayed_inode_items(inode);
4441
4442 key.objectid = ino;
4443 key.offset = (u64)-1;
4444 key.type = (u8)-1;
4445
4446 search_again:
4447 /*
4448 * with a 16K leaf size and 128MB extents, you can actually queue
4449 * up a huge file in a single leaf. Most of the time that
4450 * bytes_deleted is > 0, it will be huge by the time we get here
4451 */
4452 if (be_nice && bytes_deleted > SZ_32M) {
4453 if (btrfs_should_end_transaction(trans)) {
4454 err = -EAGAIN;
4455 goto error;
4456 }
4457 }
4458
4459
4460 path->leave_spinning = 1;
4461 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4462 if (ret < 0) {
4463 err = ret;
4464 goto out;
4465 }
4466
4467 if (ret > 0) {
4468 /* there are no items in the tree for us to truncate, we're
4469 * done
4470 */
4471 if (path->slots[0] == 0)
4472 goto out;
4473 path->slots[0]--;
4474 }
4475
4476 while (1) {
4477 fi = NULL;
4478 leaf = path->nodes[0];
4479 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4480 found_type = found_key.type;
4481
4482 if (found_key.objectid != ino)
4483 break;
4484
4485 if (found_type < min_type)
4486 break;
4487
4488 item_end = found_key.offset;
4489 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4490 fi = btrfs_item_ptr(leaf, path->slots[0],
4491 struct btrfs_file_extent_item);
4492 extent_type = btrfs_file_extent_type(leaf, fi);
4493 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4494 item_end +=
4495 btrfs_file_extent_num_bytes(leaf, fi);
4496 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4497 item_end += btrfs_file_extent_inline_len(leaf,
4498 path->slots[0], fi);
4499 }
4500 item_end--;
4501 }
4502 if (found_type > min_type) {
4503 del_item = 1;
4504 } else {
4505 if (item_end < new_size)
4506 break;
4507 if (found_key.offset >= new_size)
4508 del_item = 1;
4509 else
4510 del_item = 0;
4511 }
4512 found_extent = 0;
4513 /* FIXME, shrink the extent if the ref count is only 1 */
4514 if (found_type != BTRFS_EXTENT_DATA_KEY)
4515 goto delete;
4516
4517 if (del_item)
4518 last_size = found_key.offset;
4519 else
4520 last_size = new_size;
4521
4522 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4523 u64 num_dec;
4524 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4525 if (!del_item) {
4526 u64 orig_num_bytes =
4527 btrfs_file_extent_num_bytes(leaf, fi);
4528 extent_num_bytes = ALIGN(new_size -
4529 found_key.offset,
4530 fs_info->sectorsize);
4531 btrfs_set_file_extent_num_bytes(leaf, fi,
4532 extent_num_bytes);
4533 num_dec = (orig_num_bytes -
4534 extent_num_bytes);
4535 if (test_bit(BTRFS_ROOT_REF_COWS,
4536 &root->state) &&
4537 extent_start != 0)
4538 inode_sub_bytes(inode, num_dec);
4539 btrfs_mark_buffer_dirty(leaf);
4540 } else {
4541 extent_num_bytes =
4542 btrfs_file_extent_disk_num_bytes(leaf,
4543 fi);
4544 extent_offset = found_key.offset -
4545 btrfs_file_extent_offset(leaf, fi);
4546
4547 /* FIXME blocksize != 4096 */
4548 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4549 if (extent_start != 0) {
4550 found_extent = 1;
4551 if (test_bit(BTRFS_ROOT_REF_COWS,
4552 &root->state))
4553 inode_sub_bytes(inode, num_dec);
4554 }
4555 }
4556 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4557 /*
4558 * we can't truncate inline items that have had
4559 * special encodings
4560 */
4561 if (!del_item &&
4562 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4563 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4564
4565 /*
4566 * Need to release path in order to truncate a
4567 * compressed extent. So delete any accumulated
4568 * extent items so far.
4569 */
4570 if (btrfs_file_extent_compression(leaf, fi) !=
4571 BTRFS_COMPRESS_NONE && pending_del_nr) {
4572 err = btrfs_del_items(trans, root, path,
4573 pending_del_slot,
4574 pending_del_nr);
4575 if (err) {
4576 btrfs_abort_transaction(trans,
4577 err);
4578 goto error;
4579 }
4580 pending_del_nr = 0;
4581 }
4582
4583 err = truncate_inline_extent(inode, path,
4584 &found_key,
4585 item_end,
4586 new_size);
4587 if (err) {
4588 btrfs_abort_transaction(trans, err);
4589 goto error;
4590 }
4591 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4592 &root->state)) {
4593 inode_sub_bytes(inode, item_end + 1 - new_size);
4594 }
4595 }
4596 delete:
4597 if (del_item) {
4598 if (!pending_del_nr) {
4599 /* no pending yet, add ourselves */
4600 pending_del_slot = path->slots[0];
4601 pending_del_nr = 1;
4602 } else if (pending_del_nr &&
4603 path->slots[0] + 1 == pending_del_slot) {
4604 /* hop on the pending chunk */
4605 pending_del_nr++;
4606 pending_del_slot = path->slots[0];
4607 } else {
4608 BUG();
4609 }
4610 } else {
4611 break;
4612 }
4613 should_throttle = 0;
4614
4615 if (found_extent &&
4616 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4617 root == fs_info->tree_root)) {
4618 btrfs_set_path_blocking(path);
4619 bytes_deleted += extent_num_bytes;
4620 ret = btrfs_free_extent(trans, fs_info, extent_start,
4621 extent_num_bytes, 0,
4622 btrfs_header_owner(leaf),
4623 ino, extent_offset);
4624 BUG_ON(ret);
4625 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4626 btrfs_async_run_delayed_refs(fs_info,
4627 trans->delayed_ref_updates * 2,
4628 trans->transid, 0);
4629 if (be_nice) {
4630 if (truncate_space_check(trans, root,
4631 extent_num_bytes)) {
4632 should_end = 1;
4633 }
4634 if (btrfs_should_throttle_delayed_refs(trans,
4635 fs_info))
4636 should_throttle = 1;
4637 }
4638 }
4639
4640 if (found_type == BTRFS_INODE_ITEM_KEY)
4641 break;
4642
4643 if (path->slots[0] == 0 ||
4644 path->slots[0] != pending_del_slot ||
4645 should_throttle || should_end) {
4646 if (pending_del_nr) {
4647 ret = btrfs_del_items(trans, root, path,
4648 pending_del_slot,
4649 pending_del_nr);
4650 if (ret) {
4651 btrfs_abort_transaction(trans, ret);
4652 goto error;
4653 }
4654 pending_del_nr = 0;
4655 }
4656 btrfs_release_path(path);
4657 if (should_throttle) {
4658 unsigned long updates = trans->delayed_ref_updates;
4659 if (updates) {
4660 trans->delayed_ref_updates = 0;
4661 ret = btrfs_run_delayed_refs(trans,
4662 fs_info,
4663 updates * 2);
4664 if (ret && !err)
4665 err = ret;
4666 }
4667 }
4668 /*
4669 * if we failed to refill our space rsv, bail out
4670 * and let the transaction restart
4671 */
4672 if (should_end) {
4673 err = -EAGAIN;
4674 goto error;
4675 }
4676 goto search_again;
4677 } else {
4678 path->slots[0]--;
4679 }
4680 }
4681 out:
4682 if (pending_del_nr) {
4683 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4684 pending_del_nr);
4685 if (ret)
4686 btrfs_abort_transaction(trans, ret);
4687 }
4688 error:
4689 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4690 ASSERT(last_size >= new_size);
4691 if (!err && last_size > new_size)
4692 last_size = new_size;
4693 btrfs_ordered_update_i_size(inode, last_size, NULL);
4694 }
4695
4696 btrfs_free_path(path);
4697
4698 if (be_nice && bytes_deleted > SZ_32M) {
4699 unsigned long updates = trans->delayed_ref_updates;
4700 if (updates) {
4701 trans->delayed_ref_updates = 0;
4702 ret = btrfs_run_delayed_refs(trans, fs_info,
4703 updates * 2);
4704 if (ret && !err)
4705 err = ret;
4706 }
4707 }
4708 return err;
4709 }
4710
4711 /*
4712 * btrfs_truncate_block - read, zero a chunk and write a block
4713 * @inode - inode that we're zeroing
4714 * @from - the offset to start zeroing
4715 * @len - the length to zero, 0 to zero the entire range respective to the
4716 * offset
4717 * @front - zero up to the offset instead of from the offset on
4718 *
4719 * This will find the block for the "from" offset and cow the block and zero the
4720 * part we want to zero. This is used with truncate and hole punching.
4721 */
4722 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4723 int front)
4724 {
4725 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4726 struct address_space *mapping = inode->i_mapping;
4727 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4728 struct btrfs_ordered_extent *ordered;
4729 struct extent_state *cached_state = NULL;
4730 char *kaddr;
4731 u32 blocksize = fs_info->sectorsize;
4732 pgoff_t index = from >> PAGE_SHIFT;
4733 unsigned offset = from & (blocksize - 1);
4734 struct page *page;
4735 gfp_t mask = btrfs_alloc_write_mask(mapping);
4736 int ret = 0;
4737 u64 block_start;
4738 u64 block_end;
4739
4740 if ((offset & (blocksize - 1)) == 0 &&
4741 (!len || ((len & (blocksize - 1)) == 0)))
4742 goto out;
4743
4744 ret = btrfs_delalloc_reserve_space(inode,
4745 round_down(from, blocksize), blocksize);
4746 if (ret)
4747 goto out;
4748
4749 again:
4750 page = find_or_create_page(mapping, index, mask);
4751 if (!page) {
4752 btrfs_delalloc_release_space(inode,
4753 round_down(from, blocksize),
4754 blocksize);
4755 ret = -ENOMEM;
4756 goto out;
4757 }
4758
4759 block_start = round_down(from, blocksize);
4760 block_end = block_start + blocksize - 1;
4761
4762 if (!PageUptodate(page)) {
4763 ret = btrfs_readpage(NULL, page);
4764 lock_page(page);
4765 if (page->mapping != mapping) {
4766 unlock_page(page);
4767 put_page(page);
4768 goto again;
4769 }
4770 if (!PageUptodate(page)) {
4771 ret = -EIO;
4772 goto out_unlock;
4773 }
4774 }
4775 wait_on_page_writeback(page);
4776
4777 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4778 set_page_extent_mapped(page);
4779
4780 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4781 if (ordered) {
4782 unlock_extent_cached(io_tree, block_start, block_end,
4783 &cached_state, GFP_NOFS);
4784 unlock_page(page);
4785 put_page(page);
4786 btrfs_start_ordered_extent(inode, ordered, 1);
4787 btrfs_put_ordered_extent(ordered);
4788 goto again;
4789 }
4790
4791 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4792 EXTENT_DIRTY | EXTENT_DELALLOC |
4793 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4794 0, 0, &cached_state, GFP_NOFS);
4795
4796 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4797 &cached_state, 0);
4798 if (ret) {
4799 unlock_extent_cached(io_tree, block_start, block_end,
4800 &cached_state, GFP_NOFS);
4801 goto out_unlock;
4802 }
4803
4804 if (offset != blocksize) {
4805 if (!len)
4806 len = blocksize - offset;
4807 kaddr = kmap(page);
4808 if (front)
4809 memset(kaddr + (block_start - page_offset(page)),
4810 0, offset);
4811 else
4812 memset(kaddr + (block_start - page_offset(page)) + offset,
4813 0, len);
4814 flush_dcache_page(page);
4815 kunmap(page);
4816 }
4817 ClearPageChecked(page);
4818 set_page_dirty(page);
4819 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4820 GFP_NOFS);
4821
4822 out_unlock:
4823 if (ret)
4824 btrfs_delalloc_release_space(inode, block_start,
4825 blocksize);
4826 unlock_page(page);
4827 put_page(page);
4828 out:
4829 return ret;
4830 }
4831
4832 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4833 u64 offset, u64 len)
4834 {
4835 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4836 struct btrfs_trans_handle *trans;
4837 int ret;
4838
4839 /*
4840 * Still need to make sure the inode looks like it's been updated so
4841 * that any holes get logged if we fsync.
4842 */
4843 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4844 BTRFS_I(inode)->last_trans = fs_info->generation;
4845 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4846 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4847 return 0;
4848 }
4849
4850 /*
4851 * 1 - for the one we're dropping
4852 * 1 - for the one we're adding
4853 * 1 - for updating the inode.
4854 */
4855 trans = btrfs_start_transaction(root, 3);
4856 if (IS_ERR(trans))
4857 return PTR_ERR(trans);
4858
4859 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4860 if (ret) {
4861 btrfs_abort_transaction(trans, ret);
4862 btrfs_end_transaction(trans);
4863 return ret;
4864 }
4865
4866 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4867 0, 0, len, 0, len, 0, 0, 0);
4868 if (ret)
4869 btrfs_abort_transaction(trans, ret);
4870 else
4871 btrfs_update_inode(trans, root, inode);
4872 btrfs_end_transaction(trans);
4873 return ret;
4874 }
4875
4876 /*
4877 * This function puts in dummy file extents for the area we're creating a hole
4878 * for. So if we are truncating this file to a larger size we need to insert
4879 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4880 * the range between oldsize and size
4881 */
4882 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4883 {
4884 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4885 struct btrfs_root *root = BTRFS_I(inode)->root;
4886 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4887 struct extent_map *em = NULL;
4888 struct extent_state *cached_state = NULL;
4889 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4890 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4891 u64 block_end = ALIGN(size, fs_info->sectorsize);
4892 u64 last_byte;
4893 u64 cur_offset;
4894 u64 hole_size;
4895 int err = 0;
4896
4897 /*
4898 * If our size started in the middle of a block we need to zero out the
4899 * rest of the block before we expand the i_size, otherwise we could
4900 * expose stale data.
4901 */
4902 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4903 if (err)
4904 return err;
4905
4906 if (size <= hole_start)
4907 return 0;
4908
4909 while (1) {
4910 struct btrfs_ordered_extent *ordered;
4911
4912 lock_extent_bits(io_tree, hole_start, block_end - 1,
4913 &cached_state);
4914 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4915 block_end - hole_start);
4916 if (!ordered)
4917 break;
4918 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4919 &cached_state, GFP_NOFS);
4920 btrfs_start_ordered_extent(inode, ordered, 1);
4921 btrfs_put_ordered_extent(ordered);
4922 }
4923
4924 cur_offset = hole_start;
4925 while (1) {
4926 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4927 block_end - cur_offset, 0);
4928 if (IS_ERR(em)) {
4929 err = PTR_ERR(em);
4930 em = NULL;
4931 break;
4932 }
4933 last_byte = min(extent_map_end(em), block_end);
4934 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4935 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4936 struct extent_map *hole_em;
4937 hole_size = last_byte - cur_offset;
4938
4939 err = maybe_insert_hole(root, inode, cur_offset,
4940 hole_size);
4941 if (err)
4942 break;
4943 btrfs_drop_extent_cache(inode, cur_offset,
4944 cur_offset + hole_size - 1, 0);
4945 hole_em = alloc_extent_map();
4946 if (!hole_em) {
4947 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4948 &BTRFS_I(inode)->runtime_flags);
4949 goto next;
4950 }
4951 hole_em->start = cur_offset;
4952 hole_em->len = hole_size;
4953 hole_em->orig_start = cur_offset;
4954
4955 hole_em->block_start = EXTENT_MAP_HOLE;
4956 hole_em->block_len = 0;
4957 hole_em->orig_block_len = 0;
4958 hole_em->ram_bytes = hole_size;
4959 hole_em->bdev = fs_info->fs_devices->latest_bdev;
4960 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4961 hole_em->generation = fs_info->generation;
4962
4963 while (1) {
4964 write_lock(&em_tree->lock);
4965 err = add_extent_mapping(em_tree, hole_em, 1);
4966 write_unlock(&em_tree->lock);
4967 if (err != -EEXIST)
4968 break;
4969 btrfs_drop_extent_cache(inode, cur_offset,
4970 cur_offset +
4971 hole_size - 1, 0);
4972 }
4973 free_extent_map(hole_em);
4974 }
4975 next:
4976 free_extent_map(em);
4977 em = NULL;
4978 cur_offset = last_byte;
4979 if (cur_offset >= block_end)
4980 break;
4981 }
4982 free_extent_map(em);
4983 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4984 GFP_NOFS);
4985 return err;
4986 }
4987
4988 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4989 {
4990 struct btrfs_root *root = BTRFS_I(inode)->root;
4991 struct btrfs_trans_handle *trans;
4992 loff_t oldsize = i_size_read(inode);
4993 loff_t newsize = attr->ia_size;
4994 int mask = attr->ia_valid;
4995 int ret;
4996
4997 /*
4998 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4999 * special case where we need to update the times despite not having
5000 * these flags set. For all other operations the VFS set these flags
5001 * explicitly if it wants a timestamp update.
5002 */
5003 if (newsize != oldsize) {
5004 inode_inc_iversion(inode);
5005 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5006 inode->i_ctime = inode->i_mtime =
5007 current_time(inode);
5008 }
5009
5010 if (newsize > oldsize) {
5011 /*
5012 * Don't do an expanding truncate while snapshoting is ongoing.
5013 * This is to ensure the snapshot captures a fully consistent
5014 * state of this file - if the snapshot captures this expanding
5015 * truncation, it must capture all writes that happened before
5016 * this truncation.
5017 */
5018 btrfs_wait_for_snapshot_creation(root);
5019 ret = btrfs_cont_expand(inode, oldsize, newsize);
5020 if (ret) {
5021 btrfs_end_write_no_snapshoting(root);
5022 return ret;
5023 }
5024
5025 trans = btrfs_start_transaction(root, 1);
5026 if (IS_ERR(trans)) {
5027 btrfs_end_write_no_snapshoting(root);
5028 return PTR_ERR(trans);
5029 }
5030
5031 i_size_write(inode, newsize);
5032 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5033 pagecache_isize_extended(inode, oldsize, newsize);
5034 ret = btrfs_update_inode(trans, root, inode);
5035 btrfs_end_write_no_snapshoting(root);
5036 btrfs_end_transaction(trans);
5037 } else {
5038
5039 /*
5040 * We're truncating a file that used to have good data down to
5041 * zero. Make sure it gets into the ordered flush list so that
5042 * any new writes get down to disk quickly.
5043 */
5044 if (newsize == 0)
5045 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5046 &BTRFS_I(inode)->runtime_flags);
5047
5048 /*
5049 * 1 for the orphan item we're going to add
5050 * 1 for the orphan item deletion.
5051 */
5052 trans = btrfs_start_transaction(root, 2);
5053 if (IS_ERR(trans))
5054 return PTR_ERR(trans);
5055
5056 /*
5057 * We need to do this in case we fail at _any_ point during the
5058 * actual truncate. Once we do the truncate_setsize we could
5059 * invalidate pages which forces any outstanding ordered io to
5060 * be instantly completed which will give us extents that need
5061 * to be truncated. If we fail to get an orphan inode down we
5062 * could have left over extents that were never meant to live,
5063 * so we need to guarantee from this point on that everything
5064 * will be consistent.
5065 */
5066 ret = btrfs_orphan_add(trans, inode);
5067 btrfs_end_transaction(trans);
5068 if (ret)
5069 return ret;
5070
5071 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5072 truncate_setsize(inode, newsize);
5073
5074 /* Disable nonlocked read DIO to avoid the end less truncate */
5075 btrfs_inode_block_unlocked_dio(inode);
5076 inode_dio_wait(inode);
5077 btrfs_inode_resume_unlocked_dio(inode);
5078
5079 ret = btrfs_truncate(inode);
5080 if (ret && inode->i_nlink) {
5081 int err;
5082
5083 /*
5084 * failed to truncate, disk_i_size is only adjusted down
5085 * as we remove extents, so it should represent the true
5086 * size of the inode, so reset the in memory size and
5087 * delete our orphan entry.
5088 */
5089 trans = btrfs_join_transaction(root);
5090 if (IS_ERR(trans)) {
5091 btrfs_orphan_del(NULL, inode);
5092 return ret;
5093 }
5094 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5095 err = btrfs_orphan_del(trans, inode);
5096 if (err)
5097 btrfs_abort_transaction(trans, err);
5098 btrfs_end_transaction(trans);
5099 }
5100 }
5101
5102 return ret;
5103 }
5104
5105 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5106 {
5107 struct inode *inode = d_inode(dentry);
5108 struct btrfs_root *root = BTRFS_I(inode)->root;
5109 int err;
5110
5111 if (btrfs_root_readonly(root))
5112 return -EROFS;
5113
5114 err = setattr_prepare(dentry, attr);
5115 if (err)
5116 return err;
5117
5118 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5119 err = btrfs_setsize(inode, attr);
5120 if (err)
5121 return err;
5122 }
5123
5124 if (attr->ia_valid) {
5125 setattr_copy(inode, attr);
5126 inode_inc_iversion(inode);
5127 err = btrfs_dirty_inode(inode);
5128
5129 if (!err && attr->ia_valid & ATTR_MODE)
5130 err = posix_acl_chmod(inode, inode->i_mode);
5131 }
5132
5133 return err;
5134 }
5135
5136 /*
5137 * While truncating the inode pages during eviction, we get the VFS calling
5138 * btrfs_invalidatepage() against each page of the inode. This is slow because
5139 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5140 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5141 * extent_state structures over and over, wasting lots of time.
5142 *
5143 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5144 * those expensive operations on a per page basis and do only the ordered io
5145 * finishing, while we release here the extent_map and extent_state structures,
5146 * without the excessive merging and splitting.
5147 */
5148 static void evict_inode_truncate_pages(struct inode *inode)
5149 {
5150 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5151 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5152 struct rb_node *node;
5153
5154 ASSERT(inode->i_state & I_FREEING);
5155 truncate_inode_pages_final(&inode->i_data);
5156
5157 write_lock(&map_tree->lock);
5158 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5159 struct extent_map *em;
5160
5161 node = rb_first(&map_tree->map);
5162 em = rb_entry(node, struct extent_map, rb_node);
5163 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5164 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5165 remove_extent_mapping(map_tree, em);
5166 free_extent_map(em);
5167 if (need_resched()) {
5168 write_unlock(&map_tree->lock);
5169 cond_resched();
5170 write_lock(&map_tree->lock);
5171 }
5172 }
5173 write_unlock(&map_tree->lock);
5174
5175 /*
5176 * Keep looping until we have no more ranges in the io tree.
5177 * We can have ongoing bios started by readpages (called from readahead)
5178 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5179 * still in progress (unlocked the pages in the bio but did not yet
5180 * unlocked the ranges in the io tree). Therefore this means some
5181 * ranges can still be locked and eviction started because before
5182 * submitting those bios, which are executed by a separate task (work
5183 * queue kthread), inode references (inode->i_count) were not taken
5184 * (which would be dropped in the end io callback of each bio).
5185 * Therefore here we effectively end up waiting for those bios and
5186 * anyone else holding locked ranges without having bumped the inode's
5187 * reference count - if we don't do it, when they access the inode's
5188 * io_tree to unlock a range it may be too late, leading to an
5189 * use-after-free issue.
5190 */
5191 spin_lock(&io_tree->lock);
5192 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5193 struct extent_state *state;
5194 struct extent_state *cached_state = NULL;
5195 u64 start;
5196 u64 end;
5197
5198 node = rb_first(&io_tree->state);
5199 state = rb_entry(node, struct extent_state, rb_node);
5200 start = state->start;
5201 end = state->end;
5202 spin_unlock(&io_tree->lock);
5203
5204 lock_extent_bits(io_tree, start, end, &cached_state);
5205
5206 /*
5207 * If still has DELALLOC flag, the extent didn't reach disk,
5208 * and its reserved space won't be freed by delayed_ref.
5209 * So we need to free its reserved space here.
5210 * (Refer to comment in btrfs_invalidatepage, case 2)
5211 *
5212 * Note, end is the bytenr of last byte, so we need + 1 here.
5213 */
5214 if (state->state & EXTENT_DELALLOC)
5215 btrfs_qgroup_free_data(inode, start, end - start + 1);
5216
5217 clear_extent_bit(io_tree, start, end,
5218 EXTENT_LOCKED | EXTENT_DIRTY |
5219 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5220 EXTENT_DEFRAG, 1, 1,
5221 &cached_state, GFP_NOFS);
5222
5223 cond_resched();
5224 spin_lock(&io_tree->lock);
5225 }
5226 spin_unlock(&io_tree->lock);
5227 }
5228
5229 void btrfs_evict_inode(struct inode *inode)
5230 {
5231 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5232 struct btrfs_trans_handle *trans;
5233 struct btrfs_root *root = BTRFS_I(inode)->root;
5234 struct btrfs_block_rsv *rsv, *global_rsv;
5235 int steal_from_global = 0;
5236 u64 min_size;
5237 int ret;
5238
5239 trace_btrfs_inode_evict(inode);
5240
5241 if (!root) {
5242 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5243 return;
5244 }
5245
5246 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5247
5248 evict_inode_truncate_pages(inode);
5249
5250 if (inode->i_nlink &&
5251 ((btrfs_root_refs(&root->root_item) != 0 &&
5252 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5253 btrfs_is_free_space_inode(inode)))
5254 goto no_delete;
5255
5256 if (is_bad_inode(inode)) {
5257 btrfs_orphan_del(NULL, inode);
5258 goto no_delete;
5259 }
5260 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5261 if (!special_file(inode->i_mode))
5262 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5263
5264 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5265
5266 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5267 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5268 &BTRFS_I(inode)->runtime_flags));
5269 goto no_delete;
5270 }
5271
5272 if (inode->i_nlink > 0) {
5273 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5274 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5275 goto no_delete;
5276 }
5277
5278 ret = btrfs_commit_inode_delayed_inode(inode);
5279 if (ret) {
5280 btrfs_orphan_del(NULL, inode);
5281 goto no_delete;
5282 }
5283
5284 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5285 if (!rsv) {
5286 btrfs_orphan_del(NULL, inode);
5287 goto no_delete;
5288 }
5289 rsv->size = min_size;
5290 rsv->failfast = 1;
5291 global_rsv = &fs_info->global_block_rsv;
5292
5293 btrfs_i_size_write(inode, 0);
5294
5295 /*
5296 * This is a bit simpler than btrfs_truncate since we've already
5297 * reserved our space for our orphan item in the unlink, so we just
5298 * need to reserve some slack space in case we add bytes and update
5299 * inode item when doing the truncate.
5300 */
5301 while (1) {
5302 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5303 BTRFS_RESERVE_FLUSH_LIMIT);
5304
5305 /*
5306 * Try and steal from the global reserve since we will
5307 * likely not use this space anyway, we want to try as
5308 * hard as possible to get this to work.
5309 */
5310 if (ret)
5311 steal_from_global++;
5312 else
5313 steal_from_global = 0;
5314 ret = 0;
5315
5316 /*
5317 * steal_from_global == 0: we reserved stuff, hooray!
5318 * steal_from_global == 1: we didn't reserve stuff, boo!
5319 * steal_from_global == 2: we've committed, still not a lot of
5320 * room but maybe we'll have room in the global reserve this
5321 * time.
5322 * steal_from_global == 3: abandon all hope!
5323 */
5324 if (steal_from_global > 2) {
5325 btrfs_warn(fs_info,
5326 "Could not get space for a delete, will truncate on mount %d",
5327 ret);
5328 btrfs_orphan_del(NULL, inode);
5329 btrfs_free_block_rsv(fs_info, rsv);
5330 goto no_delete;
5331 }
5332
5333 trans = btrfs_join_transaction(root);
5334 if (IS_ERR(trans)) {
5335 btrfs_orphan_del(NULL, inode);
5336 btrfs_free_block_rsv(fs_info, rsv);
5337 goto no_delete;
5338 }
5339
5340 /*
5341 * We can't just steal from the global reserve, we need to make
5342 * sure there is room to do it, if not we need to commit and try
5343 * again.
5344 */
5345 if (steal_from_global) {
5346 if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5347 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5348 min_size, 0);
5349 else
5350 ret = -ENOSPC;
5351 }
5352
5353 /*
5354 * Couldn't steal from the global reserve, we have too much
5355 * pending stuff built up, commit the transaction and try it
5356 * again.
5357 */
5358 if (ret) {
5359 ret = btrfs_commit_transaction(trans);
5360 if (ret) {
5361 btrfs_orphan_del(NULL, inode);
5362 btrfs_free_block_rsv(fs_info, rsv);
5363 goto no_delete;
5364 }
5365 continue;
5366 } else {
5367 steal_from_global = 0;
5368 }
5369
5370 trans->block_rsv = rsv;
5371
5372 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5373 if (ret != -ENOSPC && ret != -EAGAIN)
5374 break;
5375
5376 trans->block_rsv = &fs_info->trans_block_rsv;
5377 btrfs_end_transaction(trans);
5378 trans = NULL;
5379 btrfs_btree_balance_dirty(fs_info);
5380 }
5381
5382 btrfs_free_block_rsv(fs_info, rsv);
5383
5384 /*
5385 * Errors here aren't a big deal, it just means we leave orphan items
5386 * in the tree. They will be cleaned up on the next mount.
5387 */
5388 if (ret == 0) {
5389 trans->block_rsv = root->orphan_block_rsv;
5390 btrfs_orphan_del(trans, inode);
5391 } else {
5392 btrfs_orphan_del(NULL, inode);
5393 }
5394
5395 trans->block_rsv = &fs_info->trans_block_rsv;
5396 if (!(root == fs_info->tree_root ||
5397 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5398 btrfs_return_ino(root, btrfs_ino(inode));
5399
5400 btrfs_end_transaction(trans);
5401 btrfs_btree_balance_dirty(fs_info);
5402 no_delete:
5403 btrfs_remove_delayed_node(inode);
5404 clear_inode(inode);
5405 }
5406
5407 /*
5408 * this returns the key found in the dir entry in the location pointer.
5409 * If no dir entries were found, location->objectid is 0.
5410 */
5411 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5412 struct btrfs_key *location)
5413 {
5414 const char *name = dentry->d_name.name;
5415 int namelen = dentry->d_name.len;
5416 struct btrfs_dir_item *di;
5417 struct btrfs_path *path;
5418 struct btrfs_root *root = BTRFS_I(dir)->root;
5419 int ret = 0;
5420
5421 path = btrfs_alloc_path();
5422 if (!path)
5423 return -ENOMEM;
5424
5425 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5426 namelen, 0);
5427 if (IS_ERR(di))
5428 ret = PTR_ERR(di);
5429
5430 if (IS_ERR_OR_NULL(di))
5431 goto out_err;
5432
5433 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5434 out:
5435 btrfs_free_path(path);
5436 return ret;
5437 out_err:
5438 location->objectid = 0;
5439 goto out;
5440 }
5441
5442 /*
5443 * when we hit a tree root in a directory, the btrfs part of the inode
5444 * needs to be changed to reflect the root directory of the tree root. This
5445 * is kind of like crossing a mount point.
5446 */
5447 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5448 struct inode *dir,
5449 struct dentry *dentry,
5450 struct btrfs_key *location,
5451 struct btrfs_root **sub_root)
5452 {
5453 struct btrfs_path *path;
5454 struct btrfs_root *new_root;
5455 struct btrfs_root_ref *ref;
5456 struct extent_buffer *leaf;
5457 struct btrfs_key key;
5458 int ret;
5459 int err = 0;
5460
5461 path = btrfs_alloc_path();
5462 if (!path) {
5463 err = -ENOMEM;
5464 goto out;
5465 }
5466
5467 err = -ENOENT;
5468 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5469 key.type = BTRFS_ROOT_REF_KEY;
5470 key.offset = location->objectid;
5471
5472 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5473 if (ret) {
5474 if (ret < 0)
5475 err = ret;
5476 goto out;
5477 }
5478
5479 leaf = path->nodes[0];
5480 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5481 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5482 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5483 goto out;
5484
5485 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5486 (unsigned long)(ref + 1),
5487 dentry->d_name.len);
5488 if (ret)
5489 goto out;
5490
5491 btrfs_release_path(path);
5492
5493 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5494 if (IS_ERR(new_root)) {
5495 err = PTR_ERR(new_root);
5496 goto out;
5497 }
5498
5499 *sub_root = new_root;
5500 location->objectid = btrfs_root_dirid(&new_root->root_item);
5501 location->type = BTRFS_INODE_ITEM_KEY;
5502 location->offset = 0;
5503 err = 0;
5504 out:
5505 btrfs_free_path(path);
5506 return err;
5507 }
5508
5509 static void inode_tree_add(struct inode *inode)
5510 {
5511 struct btrfs_root *root = BTRFS_I(inode)->root;
5512 struct btrfs_inode *entry;
5513 struct rb_node **p;
5514 struct rb_node *parent;
5515 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5516 u64 ino = btrfs_ino(inode);
5517
5518 if (inode_unhashed(inode))
5519 return;
5520 parent = NULL;
5521 spin_lock(&root->inode_lock);
5522 p = &root->inode_tree.rb_node;
5523 while (*p) {
5524 parent = *p;
5525 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5526
5527 if (ino < btrfs_ino(&entry->vfs_inode))
5528 p = &parent->rb_left;
5529 else if (ino > btrfs_ino(&entry->vfs_inode))
5530 p = &parent->rb_right;
5531 else {
5532 WARN_ON(!(entry->vfs_inode.i_state &
5533 (I_WILL_FREE | I_FREEING)));
5534 rb_replace_node(parent, new, &root->inode_tree);
5535 RB_CLEAR_NODE(parent);
5536 spin_unlock(&root->inode_lock);
5537 return;
5538 }
5539 }
5540 rb_link_node(new, parent, p);
5541 rb_insert_color(new, &root->inode_tree);
5542 spin_unlock(&root->inode_lock);
5543 }
5544
5545 static void inode_tree_del(struct inode *inode)
5546 {
5547 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5548 struct btrfs_root *root = BTRFS_I(inode)->root;
5549 int empty = 0;
5550
5551 spin_lock(&root->inode_lock);
5552 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5553 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5554 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5555 empty = RB_EMPTY_ROOT(&root->inode_tree);
5556 }
5557 spin_unlock(&root->inode_lock);
5558
5559 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5560 synchronize_srcu(&fs_info->subvol_srcu);
5561 spin_lock(&root->inode_lock);
5562 empty = RB_EMPTY_ROOT(&root->inode_tree);
5563 spin_unlock(&root->inode_lock);
5564 if (empty)
5565 btrfs_add_dead_root(root);
5566 }
5567 }
5568
5569 void btrfs_invalidate_inodes(struct btrfs_root *root)
5570 {
5571 struct btrfs_fs_info *fs_info = root->fs_info;
5572 struct rb_node *node;
5573 struct rb_node *prev;
5574 struct btrfs_inode *entry;
5575 struct inode *inode;
5576 u64 objectid = 0;
5577
5578 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5579 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5580
5581 spin_lock(&root->inode_lock);
5582 again:
5583 node = root->inode_tree.rb_node;
5584 prev = NULL;
5585 while (node) {
5586 prev = node;
5587 entry = rb_entry(node, struct btrfs_inode, rb_node);
5588
5589 if (objectid < btrfs_ino(&entry->vfs_inode))
5590 node = node->rb_left;
5591 else if (objectid > btrfs_ino(&entry->vfs_inode))
5592 node = node->rb_right;
5593 else
5594 break;
5595 }
5596 if (!node) {
5597 while (prev) {
5598 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5599 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5600 node = prev;
5601 break;
5602 }
5603 prev = rb_next(prev);
5604 }
5605 }
5606 while (node) {
5607 entry = rb_entry(node, struct btrfs_inode, rb_node);
5608 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5609 inode = igrab(&entry->vfs_inode);
5610 if (inode) {
5611 spin_unlock(&root->inode_lock);
5612 if (atomic_read(&inode->i_count) > 1)
5613 d_prune_aliases(inode);
5614 /*
5615 * btrfs_drop_inode will have it removed from
5616 * the inode cache when its usage count
5617 * hits zero.
5618 */
5619 iput(inode);
5620 cond_resched();
5621 spin_lock(&root->inode_lock);
5622 goto again;
5623 }
5624
5625 if (cond_resched_lock(&root->inode_lock))
5626 goto again;
5627
5628 node = rb_next(node);
5629 }
5630 spin_unlock(&root->inode_lock);
5631 }
5632
5633 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5634 {
5635 struct btrfs_iget_args *args = p;
5636 inode->i_ino = args->location->objectid;
5637 memcpy(&BTRFS_I(inode)->location, args->location,
5638 sizeof(*args->location));
5639 BTRFS_I(inode)->root = args->root;
5640 return 0;
5641 }
5642
5643 static int btrfs_find_actor(struct inode *inode, void *opaque)
5644 {
5645 struct btrfs_iget_args *args = opaque;
5646 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5647 args->root == BTRFS_I(inode)->root;
5648 }
5649
5650 static struct inode *btrfs_iget_locked(struct super_block *s,
5651 struct btrfs_key *location,
5652 struct btrfs_root *root)
5653 {
5654 struct inode *inode;
5655 struct btrfs_iget_args args;
5656 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5657
5658 args.location = location;
5659 args.root = root;
5660
5661 inode = iget5_locked(s, hashval, btrfs_find_actor,
5662 btrfs_init_locked_inode,
5663 (void *)&args);
5664 return inode;
5665 }
5666
5667 /* Get an inode object given its location and corresponding root