?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs: report errors when checksum is not found
[linux/fpc-iii.git] / fs / btrfs / inode.c
blob97970602c3d55908e47c6b031c1bdcbd527659d9
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/compat.h>
34 #include <linux/bit_spinlock.h>
35 #include <linux/xattr.h>
36 #include <linux/posix_acl.h>
37 #include <linux/falloc.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/mount.h>
41 #include <linux/btrfs.h>
42 #include <linux/blkdev.h>
43 #include <linux/posix_acl_xattr.h>
44 #include <linux/uio.h>
45 #include "ctree.h"
46 #include "disk-io.h"
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "print-tree.h"
50 #include "ordered-data.h"
51 #include "xattr.h"
52 #include "tree-log.h"
53 #include "volumes.h"
54 #include "compression.h"
55 #include "locking.h"
56 #include "free-space-cache.h"
57 #include "inode-map.h"
58 #include "backref.h"
59 #include "hash.h"
60 #include "props.h"
61 #include "qgroup.h"
62 #include "dedupe.h"
63
64 struct btrfs_iget_args {
65 struct btrfs_key *location;
66 struct btrfs_root *root;
67 };
68
69 struct btrfs_dio_data {
70 u64 outstanding_extents;
71 u64 reserve;
72 u64 unsubmitted_oe_range_start;
73 u64 unsubmitted_oe_range_end;
74 int overwrite;
75 };
76
77 static const struct inode_operations btrfs_dir_inode_operations;
78 static const struct inode_operations btrfs_symlink_inode_operations;
79 static const struct inode_operations btrfs_dir_ro_inode_operations;
80 static const struct inode_operations btrfs_special_inode_operations;
81 static const struct inode_operations btrfs_file_inode_operations;
82 static const struct address_space_operations btrfs_aops;
83 static const struct address_space_operations btrfs_symlink_aops;
84 static const struct file_operations btrfs_dir_file_operations;
85 static const struct extent_io_ops btrfs_extent_io_ops;
86
87 static struct kmem_cache *btrfs_inode_cachep;
88 struct kmem_cache *btrfs_trans_handle_cachep;
89 struct kmem_cache *btrfs_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_io_em(struct inode *inode, u64 start, u64 len,
112 u64 orig_start, u64 block_start,
113 u64 block_len, u64 orig_block_len,
114 u64 ram_bytes, int compress_type,
115 int type);
116
117 static void __endio_write_update_ordered(struct inode *inode,
118 const u64 offset, const u64 bytes,
119 const bool uptodate);
120
121 /*
122 * Cleanup all submitted ordered extents in specified range to handle errors
123 * from the fill_dellaloc() callback.
124 *
125 * NOTE: caller must ensure that when an error happens, it can not call
126 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
127 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
128 * to be released, which we want to happen only when finishing the ordered
129 * extent (btrfs_finish_ordered_io()). Also note that the caller of the
130 * fill_delalloc() callback already does proper cleanup for the first page of
131 * the range, that is, it invokes the callback writepage_end_io_hook() for the
132 * range of the first page.
133 */
134 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
135 const u64 offset,
136 const u64 bytes)
137 {
138 return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
139 bytes - PAGE_SIZE, false);
140 }
141
142 static int btrfs_dirty_inode(struct inode *inode);
143
144 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
145 void btrfs_test_inode_set_ops(struct inode *inode)
146 {
147 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
148 }
149 #endif
150
151 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
152 struct inode *inode, struct inode *dir,
153 const struct qstr *qstr)
154 {
155 int err;
156
157 err = btrfs_init_acl(trans, inode, dir);
158 if (!err)
159 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
160 return err;
161 }
162
163 /*
164 * this does all the hard work for inserting an inline extent into
165 * the btree. The caller should have done a btrfs_drop_extents so that
166 * no overlapping inline items exist in the btree
167 */
168 static int insert_inline_extent(struct btrfs_trans_handle *trans,
169 struct btrfs_path *path, int extent_inserted,
170 struct btrfs_root *root, struct inode *inode,
171 u64 start, size_t size, size_t compressed_size,
172 int compress_type,
173 struct page **compressed_pages)
174 {
175 struct extent_buffer *leaf;
176 struct page *page = NULL;
177 char *kaddr;
178 unsigned long ptr;
179 struct btrfs_file_extent_item *ei;
180 int ret;
181 size_t cur_size = size;
182 unsigned long offset;
183
184 if (compressed_size && compressed_pages)
185 cur_size = compressed_size;
186
187 inode_add_bytes(inode, size);
188
189 if (!extent_inserted) {
190 struct btrfs_key key;
191 size_t datasize;
192
193 key.objectid = btrfs_ino(BTRFS_I(inode));
194 key.offset = start;
195 key.type = BTRFS_EXTENT_DATA_KEY;
196
197 datasize = btrfs_file_extent_calc_inline_size(cur_size);
198 path->leave_spinning = 1;
199 ret = btrfs_insert_empty_item(trans, root, path, &key,
200 datasize);
201 if (ret)
202 goto fail;
203 }
204 leaf = path->nodes[0];
205 ei = btrfs_item_ptr(leaf, path->slots[0],
206 struct btrfs_file_extent_item);
207 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
208 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
209 btrfs_set_file_extent_encryption(leaf, ei, 0);
210 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
211 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
212 ptr = btrfs_file_extent_inline_start(ei);
213
214 if (compress_type != BTRFS_COMPRESS_NONE) {
215 struct page *cpage;
216 int i = 0;
217 while (compressed_size > 0) {
218 cpage = compressed_pages[i];
219 cur_size = min_t(unsigned long, compressed_size,
220 PAGE_SIZE);
221
222 kaddr = kmap_atomic(cpage);
223 write_extent_buffer(leaf, kaddr, ptr, cur_size);
224 kunmap_atomic(kaddr);
225
226 i++;
227 ptr += cur_size;
228 compressed_size -= cur_size;
229 }
230 btrfs_set_file_extent_compression(leaf, ei,
231 compress_type);
232 } else {
233 page = find_get_page(inode->i_mapping,
234 start >> PAGE_SHIFT);
235 btrfs_set_file_extent_compression(leaf, ei, 0);
236 kaddr = kmap_atomic(page);
237 offset = start & (PAGE_SIZE - 1);
238 write_extent_buffer(leaf, kaddr + offset, ptr, size);
239 kunmap_atomic(kaddr);
240 put_page(page);
241 }
242 btrfs_mark_buffer_dirty(leaf);
243 btrfs_release_path(path);
244
245 /*
246 * we're an inline extent, so nobody can
247 * extend the file past i_size without locking
248 * a page we already have locked.
249 *
250 * We must do any isize and inode updates
251 * before we unlock the pages. Otherwise we
252 * could end up racing with unlink.
253 */
254 BTRFS_I(inode)->disk_i_size = inode->i_size;
255 ret = btrfs_update_inode(trans, root, inode);
256
257 fail:
258 return ret;
259 }
260
261
262 /*
263 * conditionally insert an inline extent into the file. This
264 * does the checks required to make sure the data is small enough
265 * to fit as an inline extent.
266 */
267 static noinline int cow_file_range_inline(struct btrfs_root *root,
268 struct inode *inode, u64 start,
269 u64 end, size_t compressed_size,
270 int compress_type,
271 struct page **compressed_pages)
272 {
273 struct btrfs_fs_info *fs_info = root->fs_info;
274 struct btrfs_trans_handle *trans;
275 u64 isize = i_size_read(inode);
276 u64 actual_end = min(end + 1, isize);
277 u64 inline_len = actual_end - start;
278 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
279 u64 data_len = inline_len;
280 int ret;
281 struct btrfs_path *path;
282 int extent_inserted = 0;
283 u32 extent_item_size;
284
285 if (compressed_size)
286 data_len = compressed_size;
287
288 if (start > 0 ||
289 actual_end > fs_info->sectorsize ||
290 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
291 (!compressed_size &&
292 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
293 end + 1 < isize ||
294 data_len > fs_info->max_inline) {
295 return 1;
296 }
297
298 path = btrfs_alloc_path();
299 if (!path)
300 return -ENOMEM;
301
302 trans = btrfs_join_transaction(root);
303 if (IS_ERR(trans)) {
304 btrfs_free_path(path);
305 return PTR_ERR(trans);
306 }
307 trans->block_rsv = &fs_info->delalloc_block_rsv;
308
309 if (compressed_size && compressed_pages)
310 extent_item_size = btrfs_file_extent_calc_inline_size(
311 compressed_size);
312 else
313 extent_item_size = btrfs_file_extent_calc_inline_size(
314 inline_len);
315
316 ret = __btrfs_drop_extents(trans, root, inode, path,
317 start, aligned_end, NULL,
318 1, 1, extent_item_size, &extent_inserted);
319 if (ret) {
320 btrfs_abort_transaction(trans, ret);
321 goto out;
322 }
323
324 if (isize > actual_end)
325 inline_len = min_t(u64, isize, actual_end);
326 ret = insert_inline_extent(trans, path, extent_inserted,
327 root, inode, start,
328 inline_len, compressed_size,
329 compress_type, compressed_pages);
330 if (ret && ret != -ENOSPC) {
331 btrfs_abort_transaction(trans, ret);
332 goto out;
333 } else if (ret == -ENOSPC) {
334 ret = 1;
335 goto out;
336 }
337
338 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
339 btrfs_delalloc_release_metadata(BTRFS_I(inode), end + 1 - start);
340 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
341 out:
342 /*
343 * Don't forget to free the reserved space, as for inlined extent
344 * it won't count as data extent, free them directly here.
345 * And at reserve time, it's always aligned to page size, so
346 * just free one page here.
347 */
348 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
349 btrfs_free_path(path);
350 btrfs_end_transaction(trans);
351 return ret;
352 }
353
354 struct async_extent {
355 u64 start;
356 u64 ram_size;
357 u64 compressed_size;
358 struct page **pages;
359 unsigned long nr_pages;
360 int compress_type;
361 struct list_head list;
362 };
363
364 struct async_cow {
365 struct inode *inode;
366 struct btrfs_root *root;
367 struct page *locked_page;
368 u64 start;
369 u64 end;
370 struct list_head extents;
371 struct btrfs_work work;
372 };
373
374 static noinline int add_async_extent(struct async_cow *cow,
375 u64 start, u64 ram_size,
376 u64 compressed_size,
377 struct page **pages,
378 unsigned long nr_pages,
379 int compress_type)
380 {
381 struct async_extent *async_extent;
382
383 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
384 BUG_ON(!async_extent); /* -ENOMEM */
385 async_extent->start = start;
386 async_extent->ram_size = ram_size;
387 async_extent->compressed_size = compressed_size;
388 async_extent->pages = pages;
389 async_extent->nr_pages = nr_pages;
390 async_extent->compress_type = compress_type;
391 list_add_tail(&async_extent->list, &cow->extents);
392 return 0;
393 }
394
395 static inline int inode_need_compress(struct inode *inode)
396 {
397 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
398
399 /* force compress */
400 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
401 return 1;
402 /* bad compression ratios */
403 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
404 return 0;
405 if (btrfs_test_opt(fs_info, COMPRESS) ||
406 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
407 BTRFS_I(inode)->force_compress)
408 return 1;
409 return 0;
410 }
411
412 static inline void inode_should_defrag(struct btrfs_inode *inode,
413 u64 start, u64 end, u64 num_bytes, u64 small_write)
414 {
415 /* If this is a small write inside eof, kick off a defrag */
416 if (num_bytes < small_write &&
417 (start > 0 || end + 1 < inode->disk_i_size))
418 btrfs_add_inode_defrag(NULL, inode);
419 }
420
421 /*
422 * we create compressed extents in two phases. The first
423 * phase compresses a range of pages that have already been
424 * locked (both pages and state bits are locked).
425 *
426 * This is done inside an ordered work queue, and the compression
427 * is spread across many cpus. The actual IO submission is step
428 * two, and the ordered work queue takes care of making sure that
429 * happens in the same order things were put onto the queue by
430 * writepages and friends.
431 *
432 * If this code finds it can't get good compression, it puts an
433 * entry onto the work queue to write the uncompressed bytes. This
434 * makes sure that both compressed inodes and uncompressed inodes
435 * are written in the same order that the flusher thread sent them
436 * down.
437 */
438 static noinline void compress_file_range(struct inode *inode,
439 struct page *locked_page,
440 u64 start, u64 end,
441 struct async_cow *async_cow,
442 int *num_added)
443 {
444 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
445 struct btrfs_root *root = BTRFS_I(inode)->root;
446 u64 num_bytes;
447 u64 blocksize = fs_info->sectorsize;
448 u64 actual_end;
449 u64 isize = i_size_read(inode);
450 int ret = 0;
451 struct page **pages = NULL;
452 unsigned long nr_pages;
453 unsigned long total_compressed = 0;
454 unsigned long total_in = 0;
455 int i;
456 int will_compress;
457 int compress_type = fs_info->compress_type;
458 int redirty = 0;
459
460 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
461 SZ_16K);
462
463 actual_end = min_t(u64, isize, end + 1);
464 again:
465 will_compress = 0;
466 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
467 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
468 nr_pages = min_t(unsigned long, nr_pages,
469 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
470
471 /*
472 * we don't want to send crud past the end of i_size through
473 * compression, that's just a waste of CPU time. So, if the
474 * end of the file is before the start of our current
475 * requested range of bytes, we bail out to the uncompressed
476 * cleanup code that can deal with all of this.
477 *
478 * It isn't really the fastest way to fix things, but this is a
479 * very uncommon corner.
480 */
481 if (actual_end <= start)
482 goto cleanup_and_bail_uncompressed;
483
484 total_compressed = actual_end - start;
485
486 /*
487 * skip compression for a small file range(<=blocksize) that
488 * isn't an inline extent, since it doesn't save disk space at all.
489 */
490 if (total_compressed <= blocksize &&
491 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
492 goto cleanup_and_bail_uncompressed;
493
494 total_compressed = min_t(unsigned long, total_compressed,
495 BTRFS_MAX_UNCOMPRESSED);
496 num_bytes = ALIGN(end - start + 1, blocksize);
497 num_bytes = max(blocksize, num_bytes);
498 total_in = 0;
499 ret = 0;
500
501 /*
502 * we do compression for mount -o compress and when the
503 * inode has not been flagged as nocompress. This flag can
504 * change at any time if we discover bad compression ratios.
505 */
506 if (inode_need_compress(inode)) {
507 WARN_ON(pages);
508 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
509 if (!pages) {
510 /* just bail out to the uncompressed code */
511 goto cont;
512 }
513
514 if (BTRFS_I(inode)->force_compress)
515 compress_type = BTRFS_I(inode)->force_compress;
516
517 /*
518 * we need to call clear_page_dirty_for_io on each
519 * page in the range. Otherwise applications with the file
520 * mmap'd can wander in and change the page contents while
521 * we are compressing them.
522 *
523 * If the compression fails for any reason, we set the pages
524 * dirty again later on.
525 */
526 extent_range_clear_dirty_for_io(inode, start, end);
527 redirty = 1;
528 ret = btrfs_compress_pages(compress_type,
529 inode->i_mapping, start,
530 pages,
531 &nr_pages,
532 &total_in,
533 &total_compressed);
534
535 if (!ret) {
536 unsigned long offset = total_compressed &
537 (PAGE_SIZE - 1);
538 struct page *page = pages[nr_pages - 1];
539 char *kaddr;
540
541 /* zero the tail end of the last page, we might be
542 * sending it down to disk
543 */
544 if (offset) {
545 kaddr = kmap_atomic(page);
546 memset(kaddr + offset, 0,
547 PAGE_SIZE - offset);
548 kunmap_atomic(kaddr);
549 }
550 will_compress = 1;
551 }
552 }
553 cont:
554 if (start == 0) {
555 /* lets try to make an inline extent */
556 if (ret || total_in < (actual_end - start)) {
557 /* we didn't compress the entire range, try
558 * to make an uncompressed inline extent.
559 */
560 ret = cow_file_range_inline(root, inode, start, end,
561 0, BTRFS_COMPRESS_NONE, NULL);
562 } else {
563 /* try making a compressed inline extent */
564 ret = cow_file_range_inline(root, inode, start, end,
565 total_compressed,
566 compress_type, pages);
567 }
568 if (ret <= 0) {
569 unsigned long clear_flags = EXTENT_DELALLOC |
570 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG;
571 unsigned long page_error_op;
572
573 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
574 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
575
576 /*
577 * inline extent creation worked or returned error,
578 * we don't need to create any more async work items.
579 * Unlock and free up our temp pages.
580 */
581 extent_clear_unlock_delalloc(inode, start, end, end,
582 NULL, clear_flags,
583 PAGE_UNLOCK |
584 PAGE_CLEAR_DIRTY |
585 PAGE_SET_WRITEBACK |
586 page_error_op |
587 PAGE_END_WRITEBACK);
588 if (ret == 0)
589 btrfs_free_reserved_data_space_noquota(inode,
590 start,
591 end - start + 1);
592 goto free_pages_out;
593 }
594 }
595
596 if (will_compress) {
597 /*
598 * we aren't doing an inline extent round the compressed size
599 * up to a block size boundary so the allocator does sane
600 * things
601 */
602 total_compressed = ALIGN(total_compressed, blocksize);
603
604 /*
605 * one last check to make sure the compression is really a
606 * win, compare the page count read with the blocks on disk,
607 * compression must free at least one sector size
608 */
609 total_in = ALIGN(total_in, PAGE_SIZE);
610 if (total_compressed + blocksize <= total_in) {
611 num_bytes = total_in;
612 *num_added += 1;
613
614 /*
615 * The async work queues will take care of doing actual
616 * allocation on disk for these compressed pages, and
617 * will submit them to the elevator.
618 */
619 add_async_extent(async_cow, start, num_bytes,
620 total_compressed, pages, nr_pages,
621 compress_type);
622
623 if (start + num_bytes < end) {
624 start += num_bytes;
625 pages = NULL;
626 cond_resched();
627 goto again;
628 }
629 return;
630 }
631 }
632 if (pages) {
633 /*
634 * the compression code ran but failed to make things smaller,
635 * free any pages it allocated and our page pointer array
636 */
637 for (i = 0; i < nr_pages; i++) {
638 WARN_ON(pages[i]->mapping);
639 put_page(pages[i]);
640 }
641 kfree(pages);
642 pages = NULL;
643 total_compressed = 0;
644 nr_pages = 0;
645
646 /* flag the file so we don't compress in the future */
647 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
648 !(BTRFS_I(inode)->force_compress)) {
649 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
650 }
651 }
652 cleanup_and_bail_uncompressed:
653 /*
654 * No compression, but we still need to write the pages in the file
655 * we've been given so far. redirty the locked page if it corresponds
656 * to our extent and set things up for the async work queue to run
657 * cow_file_range to do the normal delalloc dance.
658 */
659 if (page_offset(locked_page) >= start &&
660 page_offset(locked_page) <= end)
661 __set_page_dirty_nobuffers(locked_page);
662 /* unlocked later on in the async handlers */
663
664 if (redirty)
665 extent_range_redirty_for_io(inode, start, end);
666 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
667 BTRFS_COMPRESS_NONE);
668 *num_added += 1;
669
670 return;
671
672 free_pages_out:
673 for (i = 0; i < nr_pages; i++) {
674 WARN_ON(pages[i]->mapping);
675 put_page(pages[i]);
676 }
677 kfree(pages);
678 }
679
680 static void free_async_extent_pages(struct async_extent *async_extent)
681 {
682 int i;
683
684 if (!async_extent->pages)
685 return;
686
687 for (i = 0; i < async_extent->nr_pages; i++) {
688 WARN_ON(async_extent->pages[i]->mapping);
689 put_page(async_extent->pages[i]);
690 }
691 kfree(async_extent->pages);
692 async_extent->nr_pages = 0;
693 async_extent->pages = NULL;
694 }
695
696 /*
697 * phase two of compressed writeback. This is the ordered portion
698 * of the code, which only gets called in the order the work was
699 * queued. We walk all the async extents created by compress_file_range
700 * and send them down to the disk.
701 */
702 static noinline void submit_compressed_extents(struct inode *inode,
703 struct async_cow *async_cow)
704 {
705 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
706 struct async_extent *async_extent;
707 u64 alloc_hint = 0;
708 struct btrfs_key ins;
709 struct extent_map *em;
710 struct btrfs_root *root = BTRFS_I(inode)->root;
711 struct extent_io_tree *io_tree;
712 int ret = 0;
713
714 again:
715 while (!list_empty(&async_cow->extents)) {
716 async_extent = list_entry(async_cow->extents.next,
717 struct async_extent, list);
718 list_del(&async_extent->list);
719
720 io_tree = &BTRFS_I(inode)->io_tree;
721
722 retry:
723 /* did the compression code fall back to uncompressed IO? */
724 if (!async_extent->pages) {
725 int page_started = 0;
726 unsigned long nr_written = 0;
727
728 lock_extent(io_tree, async_extent->start,
729 async_extent->start +
730 async_extent->ram_size - 1);
731
732 /* allocate blocks */
733 ret = cow_file_range(inode, async_cow->locked_page,
734 async_extent->start,
735 async_extent->start +
736 async_extent->ram_size - 1,
737 async_extent->start +
738 async_extent->ram_size - 1,
739 &page_started, &nr_written, 0,
740 NULL);
741
742 /* JDM XXX */
743
744 /*
745 * if page_started, cow_file_range inserted an
746 * inline extent and took care of all the unlocking
747 * and IO for us. Otherwise, we need to submit
748 * all those pages down to the drive.
749 */
750 if (!page_started && !ret)
751 extent_write_locked_range(io_tree,
752 inode, async_extent->start,
753 async_extent->start +
754 async_extent->ram_size - 1,
755 btrfs_get_extent,
756 WB_SYNC_ALL);
757 else if (ret)
758 unlock_page(async_cow->locked_page);
759 kfree(async_extent);
760 cond_resched();
761 continue;
762 }
763
764 lock_extent(io_tree, async_extent->start,
765 async_extent->start + async_extent->ram_size - 1);
766
767 ret = btrfs_reserve_extent(root, async_extent->ram_size,
768 async_extent->compressed_size,
769 async_extent->compressed_size,
770 0, alloc_hint, &ins, 1, 1);
771 if (ret) {
772 free_async_extent_pages(async_extent);
773
774 if (ret == -ENOSPC) {
775 unlock_extent(io_tree, async_extent->start,
776 async_extent->start +
777 async_extent->ram_size - 1);
778
779 /*
780 * we need to redirty the pages if we decide to
781 * fallback to uncompressed IO, otherwise we
782 * will not submit these pages down to lower
783 * layers.
784 */
785 extent_range_redirty_for_io(inode,
786 async_extent->start,
787 async_extent->start +
788 async_extent->ram_size - 1);
789
790 goto retry;
791 }
792 goto out_free;
793 }
794 /*
795 * here we're doing allocation and writeback of the
796 * compressed pages
797 */
798 em = create_io_em(inode, async_extent->start,
799 async_extent->ram_size, /* len */
800 async_extent->start, /* orig_start */
801 ins.objectid, /* block_start */
802 ins.offset, /* block_len */
803 ins.offset, /* orig_block_len */
804 async_extent->ram_size, /* ram_bytes */
805 async_extent->compress_type,
806 BTRFS_ORDERED_COMPRESSED);
807 if (IS_ERR(em))
808 /* ret value is not necessary due to void function */
809 goto out_free_reserve;
810 free_extent_map(em);
811
812 ret = btrfs_add_ordered_extent_compress(inode,
813 async_extent->start,
814 ins.objectid,
815 async_extent->ram_size,
816 ins.offset,
817 BTRFS_ORDERED_COMPRESSED,
818 async_extent->compress_type);
819 if (ret) {
820 btrfs_drop_extent_cache(BTRFS_I(inode),
821 async_extent->start,
822 async_extent->start +
823 async_extent->ram_size - 1, 0);
824 goto out_free_reserve;
825 }
826 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
827
828 /*
829 * clear dirty, set writeback and unlock the pages.
830 */
831 extent_clear_unlock_delalloc(inode, async_extent->start,
832 async_extent->start +
833 async_extent->ram_size - 1,
834 async_extent->start +
835 async_extent->ram_size - 1,
836 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
837 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
838 PAGE_SET_WRITEBACK);
839 if (btrfs_submit_compressed_write(inode,
840 async_extent->start,
841 async_extent->ram_size,
842 ins.objectid,
843 ins.offset, async_extent->pages,
844 async_extent->nr_pages)) {
845 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
846 struct page *p = async_extent->pages[0];
847 const u64 start = async_extent->start;
848 const u64 end = start + async_extent->ram_size - 1;
849
850 p->mapping = inode->i_mapping;
851 tree->ops->writepage_end_io_hook(p, start, end,
852 NULL, 0);
853 p->mapping = NULL;
854 extent_clear_unlock_delalloc(inode, start, end, end,
855 NULL, 0,
856 PAGE_END_WRITEBACK |
857 PAGE_SET_ERROR);
858 free_async_extent_pages(async_extent);
859 }
860 alloc_hint = ins.objectid + ins.offset;
861 kfree(async_extent);
862 cond_resched();
863 }
864 return;
865 out_free_reserve:
866 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
867 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
868 out_free:
869 extent_clear_unlock_delalloc(inode, async_extent->start,
870 async_extent->start +
871 async_extent->ram_size - 1,
872 async_extent->start +
873 async_extent->ram_size - 1,
874 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
875 EXTENT_DELALLOC_NEW |
876 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
877 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
878 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
879 PAGE_SET_ERROR);
880 free_async_extent_pages(async_extent);
881 kfree(async_extent);
882 goto again;
883 }
884
885 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
886 u64 num_bytes)
887 {
888 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
889 struct extent_map *em;
890 u64 alloc_hint = 0;
891
892 read_lock(&em_tree->lock);
893 em = search_extent_mapping(em_tree, start, num_bytes);
894 if (em) {
895 /*
896 * if block start isn't an actual block number then find the
897 * first block in this inode and use that as a hint. If that
898 * block is also bogus then just don't worry about it.
899 */
900 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
901 free_extent_map(em);
902 em = search_extent_mapping(em_tree, 0, 0);
903 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
904 alloc_hint = em->block_start;
905 if (em)
906 free_extent_map(em);
907 } else {
908 alloc_hint = em->block_start;
909 free_extent_map(em);
910 }
911 }
912 read_unlock(&em_tree->lock);
913
914 return alloc_hint;
915 }
916
917 /*
918 * when extent_io.c finds a delayed allocation range in the file,
919 * the call backs end up in this code. The basic idea is to
920 * allocate extents on disk for the range, and create ordered data structs
921 * in ram to track those extents.
922 *
923 * locked_page is the page that writepage had locked already. We use
924 * it to make sure we don't do extra locks or unlocks.
925 *
926 * *page_started is set to one if we unlock locked_page and do everything
927 * required to start IO on it. It may be clean and already done with
928 * IO when we return.
929 */
930 static noinline int cow_file_range(struct inode *inode,
931 struct page *locked_page,
932 u64 start, u64 end, u64 delalloc_end,
933 int *page_started, unsigned long *nr_written,
934 int unlock, struct btrfs_dedupe_hash *hash)
935 {
936 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
937 struct btrfs_root *root = BTRFS_I(inode)->root;
938 u64 alloc_hint = 0;
939 u64 num_bytes;
940 unsigned long ram_size;
941 u64 disk_num_bytes;
942 u64 cur_alloc_size = 0;
943 u64 blocksize = fs_info->sectorsize;
944 struct btrfs_key ins;
945 struct extent_map *em;
946 unsigned clear_bits;
947 unsigned long page_ops;
948 bool extent_reserved = false;
949 int ret = 0;
950
951 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
952 WARN_ON_ONCE(1);
953 ret = -EINVAL;
954 goto out_unlock;
955 }
956
957 num_bytes = ALIGN(end - start + 1, blocksize);
958 num_bytes = max(blocksize, num_bytes);
959 disk_num_bytes = num_bytes;
960
961 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
962
963 if (start == 0) {
964 /* lets try to make an inline extent */
965 ret = cow_file_range_inline(root, inode, start, end, 0,
966 BTRFS_COMPRESS_NONE, NULL);
967 if (ret == 0) {
968 extent_clear_unlock_delalloc(inode, start, end,
969 delalloc_end, NULL,
970 EXTENT_LOCKED | EXTENT_DELALLOC |
971 EXTENT_DELALLOC_NEW |
972 EXTENT_DEFRAG, PAGE_UNLOCK |
973 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
974 PAGE_END_WRITEBACK);
975 btrfs_free_reserved_data_space_noquota(inode, start,
976 end - start + 1);
977 *nr_written = *nr_written +
978 (end - start + PAGE_SIZE) / PAGE_SIZE;
979 *page_started = 1;
980 goto out;
981 } else if (ret < 0) {
982 goto out_unlock;
983 }
984 }
985
986 BUG_ON(disk_num_bytes >
987 btrfs_super_total_bytes(fs_info->super_copy));
988
989 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
990 btrfs_drop_extent_cache(BTRFS_I(inode), start,
991 start + num_bytes - 1, 0);
992
993 while (disk_num_bytes > 0) {
994 cur_alloc_size = disk_num_bytes;
995 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
996 fs_info->sectorsize, 0, alloc_hint,
997 &ins, 1, 1);
998 if (ret < 0)
999 goto out_unlock;
1000 cur_alloc_size = ins.offset;
1001 extent_reserved = true;
1002
1003 ram_size = ins.offset;
1004 em = create_io_em(inode, start, ins.offset, /* len */
1005 start, /* orig_start */
1006 ins.objectid, /* block_start */
1007 ins.offset, /* block_len */
1008 ins.offset, /* orig_block_len */
1009 ram_size, /* ram_bytes */
1010 BTRFS_COMPRESS_NONE, /* compress_type */
1011 BTRFS_ORDERED_REGULAR /* type */);
1012 if (IS_ERR(em))
1013 goto out_reserve;
1014 free_extent_map(em);
1015
1016 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1017 ram_size, cur_alloc_size, 0);
1018 if (ret)
1019 goto out_drop_extent_cache;
1020
1021 if (root->root_key.objectid ==
1022 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1023 ret = btrfs_reloc_clone_csums(inode, start,
1024 cur_alloc_size);
1025 /*
1026 * Only drop cache here, and process as normal.
1027 *
1028 * We must not allow extent_clear_unlock_delalloc()
1029 * at out_unlock label to free meta of this ordered
1030 * extent, as its meta should be freed by
1031 * btrfs_finish_ordered_io().
1032 *
1033 * So we must continue until @start is increased to
1034 * skip current ordered extent.
1035 */
1036 if (ret)
1037 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1038 start + ram_size - 1, 0);
1039 }
1040
1041 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1042
1043 /* we're not doing compressed IO, don't unlock the first
1044 * page (which the caller expects to stay locked), don't
1045 * clear any dirty bits and don't set any writeback bits
1046 *
1047 * Do set the Private2 bit so we know this page was properly
1048 * setup for writepage
1049 */
1050 page_ops = unlock ? PAGE_UNLOCK : 0;
1051 page_ops |= PAGE_SET_PRIVATE2;
1052
1053 extent_clear_unlock_delalloc(inode, start,
1054 start + ram_size - 1,
1055 delalloc_end, locked_page,
1056 EXTENT_LOCKED | EXTENT_DELALLOC,
1057 page_ops);
1058 if (disk_num_bytes < cur_alloc_size)
1059 disk_num_bytes = 0;
1060 else
1061 disk_num_bytes -= cur_alloc_size;
1062 num_bytes -= cur_alloc_size;
1063 alloc_hint = ins.objectid + ins.offset;
1064 start += cur_alloc_size;
1065 extent_reserved = false;
1066
1067 /*
1068 * btrfs_reloc_clone_csums() error, since start is increased
1069 * extent_clear_unlock_delalloc() at out_unlock label won't
1070 * free metadata of current ordered extent, we're OK to exit.
1071 */
1072 if (ret)
1073 goto out_unlock;
1074 }
1075 out:
1076 return ret;
1077
1078 out_drop_extent_cache:
1079 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1080 out_reserve:
1081 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1082 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1083 out_unlock:
1084 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1085 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1086 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1087 PAGE_END_WRITEBACK;
1088 /*
1089 * If we reserved an extent for our delalloc range (or a subrange) and
1090 * failed to create the respective ordered extent, then it means that
1091 * when we reserved the extent we decremented the extent's size from
1092 * the data space_info's bytes_may_use counter and incremented the
1093 * space_info's bytes_reserved counter by the same amount. We must make
1094 * sure extent_clear_unlock_delalloc() does not try to decrement again
1095 * the data space_info's bytes_may_use counter, therefore we do not pass
1096 * it the flag EXTENT_CLEAR_DATA_RESV.
1097 */
1098 if (extent_reserved) {
1099 extent_clear_unlock_delalloc(inode, start,
1100 start + cur_alloc_size,
1101 start + cur_alloc_size,
1102 locked_page,
1103 clear_bits,
1104 page_ops);
1105 start += cur_alloc_size;
1106 if (start >= end)
1107 goto out;
1108 }
1109 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1110 locked_page,
1111 clear_bits | EXTENT_CLEAR_DATA_RESV,
1112 page_ops);
1113 goto out;
1114 }
1115
1116 /*
1117 * work queue call back to started compression on a file and pages
1118 */
1119 static noinline void async_cow_start(struct btrfs_work *work)
1120 {
1121 struct async_cow *async_cow;
1122 int num_added = 0;
1123 async_cow = container_of(work, struct async_cow, work);
1124
1125 compress_file_range(async_cow->inode, async_cow->locked_page,
1126 async_cow->start, async_cow->end, async_cow,
1127 &num_added);
1128 if (num_added == 0) {
1129 btrfs_add_delayed_iput(async_cow->inode);
1130 async_cow->inode = NULL;
1131 }
1132 }
1133
1134 /*
1135 * work queue call back to submit previously compressed pages
1136 */
1137 static noinline void async_cow_submit(struct btrfs_work *work)
1138 {
1139 struct btrfs_fs_info *fs_info;
1140 struct async_cow *async_cow;
1141 struct btrfs_root *root;
1142 unsigned long nr_pages;
1143
1144 async_cow = container_of(work, struct async_cow, work);
1145
1146 root = async_cow->root;
1147 fs_info = root->fs_info;
1148 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1149 PAGE_SHIFT;
1150
1151 /*
1152 * atomic_sub_return implies a barrier for waitqueue_active
1153 */
1154 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1155 5 * SZ_1M &&
1156 waitqueue_active(&fs_info->async_submit_wait))
1157 wake_up(&fs_info->async_submit_wait);
1158
1159 if (async_cow->inode)
1160 submit_compressed_extents(async_cow->inode, async_cow);
1161 }
1162
1163 static noinline void async_cow_free(struct btrfs_work *work)
1164 {
1165 struct async_cow *async_cow;
1166 async_cow = container_of(work, struct async_cow, work);
1167 if (async_cow->inode)
1168 btrfs_add_delayed_iput(async_cow->inode);
1169 kfree(async_cow);
1170 }
1171
1172 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1173 u64 start, u64 end, int *page_started,
1174 unsigned long *nr_written)
1175 {
1176 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1177 struct async_cow *async_cow;
1178 struct btrfs_root *root = BTRFS_I(inode)->root;
1179 unsigned long nr_pages;
1180 u64 cur_end;
1181
1182 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1183 1, 0, NULL, GFP_NOFS);
1184 while (start < end) {
1185 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1186 BUG_ON(!async_cow); /* -ENOMEM */
1187 async_cow->inode = igrab(inode);
1188 async_cow->root = root;
1189 async_cow->locked_page = locked_page;
1190 async_cow->start = start;
1191
1192 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1193 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1194 cur_end = end;
1195 else
1196 cur_end = min(end, start + SZ_512K - 1);
1197
1198 async_cow->end = cur_end;
1199 INIT_LIST_HEAD(&async_cow->extents);
1200
1201 btrfs_init_work(&async_cow->work,
1202 btrfs_delalloc_helper,
1203 async_cow_start, async_cow_submit,
1204 async_cow_free);
1205
1206 nr_pages = (cur_end - start + PAGE_SIZE) >>
1207 PAGE_SHIFT;
1208 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1209
1210 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1211
1212 while (atomic_read(&fs_info->async_submit_draining) &&
1213 atomic_read(&fs_info->async_delalloc_pages)) {
1214 wait_event(fs_info->async_submit_wait,
1215 (atomic_read(&fs_info->async_delalloc_pages) ==
1216 0));
1217 }
1218
1219 *nr_written += nr_pages;
1220 start = cur_end + 1;
1221 }
1222 *page_started = 1;
1223 return 0;
1224 }
1225
1226 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1227 u64 bytenr, u64 num_bytes)
1228 {
1229 int ret;
1230 struct btrfs_ordered_sum *sums;
1231 LIST_HEAD(list);
1232
1233 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1234 bytenr + num_bytes - 1, &list, 0);
1235 if (ret == 0 && list_empty(&list))
1236 return 0;
1237
1238 while (!list_empty(&list)) {
1239 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1240 list_del(&sums->list);
1241 kfree(sums);
1242 }
1243 return 1;
1244 }
1245
1246 /*
1247 * when nowcow writeback call back. This checks for snapshots or COW copies
1248 * of the extents that exist in the file, and COWs the file as required.
1249 *
1250 * If no cow copies or snapshots exist, we write directly to the existing
1251 * blocks on disk
1252 */
1253 static noinline int run_delalloc_nocow(struct inode *inode,
1254 struct page *locked_page,
1255 u64 start, u64 end, int *page_started, int force,
1256 unsigned long *nr_written)
1257 {
1258 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1259 struct btrfs_root *root = BTRFS_I(inode)->root;
1260 struct extent_buffer *leaf;
1261 struct btrfs_path *path;
1262 struct btrfs_file_extent_item *fi;
1263 struct btrfs_key found_key;
1264 struct extent_map *em;
1265 u64 cow_start;
1266 u64 cur_offset;
1267 u64 extent_end;
1268 u64 extent_offset;
1269 u64 disk_bytenr;
1270 u64 num_bytes;
1271 u64 disk_num_bytes;
1272 u64 ram_bytes;
1273 int extent_type;
1274 int ret, err;
1275 int type;
1276 int nocow;
1277 int check_prev = 1;
1278 bool nolock;
1279 u64 ino = btrfs_ino(BTRFS_I(inode));
1280
1281 path = btrfs_alloc_path();
1282 if (!path) {
1283 extent_clear_unlock_delalloc(inode, start, end, end,
1284 locked_page,
1285 EXTENT_LOCKED | EXTENT_DELALLOC |
1286 EXTENT_DO_ACCOUNTING |
1287 EXTENT_DEFRAG, PAGE_UNLOCK |
1288 PAGE_CLEAR_DIRTY |
1289 PAGE_SET_WRITEBACK |
1290 PAGE_END_WRITEBACK);
1291 return -ENOMEM;
1292 }
1293
1294 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1295
1296 cow_start = (u64)-1;
1297 cur_offset = start;
1298 while (1) {
1299 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1300 cur_offset, 0);
1301 if (ret < 0)
1302 goto error;
1303 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1304 leaf = path->nodes[0];
1305 btrfs_item_key_to_cpu(leaf, &found_key,
1306 path->slots[0] - 1);
1307 if (found_key.objectid == ino &&
1308 found_key.type == BTRFS_EXTENT_DATA_KEY)
1309 path->slots[0]--;
1310 }
1311 check_prev = 0;
1312 next_slot:
1313 leaf = path->nodes[0];
1314 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1315 ret = btrfs_next_leaf(root, path);
1316 if (ret < 0)
1317 goto error;
1318 if (ret > 0)
1319 break;
1320 leaf = path->nodes[0];
1321 }
1322
1323 nocow = 0;
1324 disk_bytenr = 0;
1325 num_bytes = 0;
1326 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1327
1328 if (found_key.objectid > ino)
1329 break;
1330 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1331 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1332 path->slots[0]++;
1333 goto next_slot;
1334 }
1335 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1336 found_key.offset > end)
1337 break;
1338
1339 if (found_key.offset > cur_offset) {
1340 extent_end = found_key.offset;
1341 extent_type = 0;
1342 goto out_check;
1343 }
1344
1345 fi = btrfs_item_ptr(leaf, path->slots[0],
1346 struct btrfs_file_extent_item);
1347 extent_type = btrfs_file_extent_type(leaf, fi);
1348
1349 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1350 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1351 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1352 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1353 extent_offset = btrfs_file_extent_offset(leaf, fi);
1354 extent_end = found_key.offset +
1355 btrfs_file_extent_num_bytes(leaf, fi);
1356 disk_num_bytes =
1357 btrfs_file_extent_disk_num_bytes(leaf, fi);
1358 if (extent_end <= start) {
1359 path->slots[0]++;
1360 goto next_slot;
1361 }
1362 if (disk_bytenr == 0)
1363 goto out_check;
1364 if (btrfs_file_extent_compression(leaf, fi) ||
1365 btrfs_file_extent_encryption(leaf, fi) ||
1366 btrfs_file_extent_other_encoding(leaf, fi))
1367 goto out_check;
1368 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1369 goto out_check;
1370 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1371 goto out_check;
1372 if (btrfs_cross_ref_exist(root, ino,
1373 found_key.offset -
1374 extent_offset, disk_bytenr))
1375 goto out_check;
1376 disk_bytenr += extent_offset;
1377 disk_bytenr += cur_offset - found_key.offset;
1378 num_bytes = min(end + 1, extent_end) - cur_offset;
1379 /*
1380 * if there are pending snapshots for this root,
1381 * we fall into common COW way.
1382 */
1383 if (!nolock) {
1384 err = btrfs_start_write_no_snapshoting(root);
1385 if (!err)
1386 goto out_check;
1387 }
1388 /*
1389 * force cow if csum exists in the range.
1390 * this ensure that csum for a given extent are
1391 * either valid or do not exist.
1392 */
1393 if (csum_exist_in_range(fs_info, disk_bytenr,
1394 num_bytes)) {
1395 if (!nolock)
1396 btrfs_end_write_no_snapshoting(root);
1397 goto out_check;
1398 }
1399 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1400 if (!nolock)
1401 btrfs_end_write_no_snapshoting(root);
1402 goto out_check;
1403 }
1404 nocow = 1;
1405 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1406 extent_end = found_key.offset +
1407 btrfs_file_extent_inline_len(leaf,
1408 path->slots[0], fi);
1409 extent_end = ALIGN(extent_end,
1410 fs_info->sectorsize);
1411 } else {
1412 BUG_ON(1);
1413 }
1414 out_check:
1415 if (extent_end <= start) {
1416 path->slots[0]++;
1417 if (!nolock && nocow)
1418 btrfs_end_write_no_snapshoting(root);
1419 if (nocow)
1420 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1421 goto next_slot;
1422 }
1423 if (!nocow) {
1424 if (cow_start == (u64)-1)
1425 cow_start = cur_offset;
1426 cur_offset = extent_end;
1427 if (cur_offset > end)
1428 break;
1429 path->slots[0]++;
1430 goto next_slot;
1431 }
1432
1433 btrfs_release_path(path);
1434 if (cow_start != (u64)-1) {
1435 ret = cow_file_range(inode, locked_page,
1436 cow_start, found_key.offset - 1,
1437 end, page_started, nr_written, 1,
1438 NULL);
1439 if (ret) {
1440 if (!nolock && nocow)
1441 btrfs_end_write_no_snapshoting(root);
1442 if (nocow)
1443 btrfs_dec_nocow_writers(fs_info,
1444 disk_bytenr);
1445 goto error;
1446 }
1447 cow_start = (u64)-1;
1448 }
1449
1450 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1451 u64 orig_start = found_key.offset - extent_offset;
1452
1453 em = create_io_em(inode, cur_offset, num_bytes,
1454 orig_start,
1455 disk_bytenr, /* block_start */
1456 num_bytes, /* block_len */
1457 disk_num_bytes, /* orig_block_len */
1458 ram_bytes, BTRFS_COMPRESS_NONE,
1459 BTRFS_ORDERED_PREALLOC);
1460 if (IS_ERR(em)) {
1461 if (!nolock && nocow)
1462 btrfs_end_write_no_snapshoting(root);
1463 if (nocow)
1464 btrfs_dec_nocow_writers(fs_info,
1465 disk_bytenr);
1466 ret = PTR_ERR(em);
1467 goto error;
1468 }
1469 free_extent_map(em);
1470 }
1471
1472 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1473 type = BTRFS_ORDERED_PREALLOC;
1474 } else {
1475 type = BTRFS_ORDERED_NOCOW;
1476 }
1477
1478 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1479 num_bytes, num_bytes, type);
1480 if (nocow)
1481 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1482 BUG_ON(ret); /* -ENOMEM */
1483
1484 if (root->root_key.objectid ==
1485 BTRFS_DATA_RELOC_TREE_OBJECTID)
1486 /*
1487 * Error handled later, as we must prevent
1488 * extent_clear_unlock_delalloc() in error handler
1489 * from freeing metadata of created ordered extent.
1490 */
1491 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1492 num_bytes);
1493
1494 extent_clear_unlock_delalloc(inode, cur_offset,
1495 cur_offset + num_bytes - 1, end,
1496 locked_page, EXTENT_LOCKED |
1497 EXTENT_DELALLOC |
1498 EXTENT_CLEAR_DATA_RESV,
1499 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1500
1501 if (!nolock && nocow)
1502 btrfs_end_write_no_snapshoting(root);
1503 cur_offset = extent_end;
1504
1505 /*
1506 * btrfs_reloc_clone_csums() error, now we're OK to call error
1507 * handler, as metadata for created ordered extent will only
1508 * be freed by btrfs_finish_ordered_io().
1509 */
1510 if (ret)
1511 goto error;
1512 if (cur_offset > end)
1513 break;
1514 }
1515 btrfs_release_path(path);
1516
1517 if (cur_offset <= end && cow_start == (u64)-1) {
1518 cow_start = cur_offset;
1519 cur_offset = end;
1520 }
1521
1522 if (cow_start != (u64)-1) {
1523 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1524 page_started, nr_written, 1, NULL);
1525 if (ret)
1526 goto error;
1527 }
1528
1529 error:
1530 if (ret && cur_offset < end)
1531 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1532 locked_page, EXTENT_LOCKED |
1533 EXTENT_DELALLOC | EXTENT_DEFRAG |
1534 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1535 PAGE_CLEAR_DIRTY |
1536 PAGE_SET_WRITEBACK |
1537 PAGE_END_WRITEBACK);
1538 btrfs_free_path(path);
1539 return ret;
1540 }
1541
1542 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1543 {
1544
1545 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1546 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1547 return 0;
1548
1549 /*
1550 * @defrag_bytes is a hint value, no spinlock held here,
1551 * if is not zero, it means the file is defragging.
1552 * Force cow if given extent needs to be defragged.
1553 */
1554 if (BTRFS_I(inode)->defrag_bytes &&
1555 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1556 EXTENT_DEFRAG, 0, NULL))
1557 return 1;
1558
1559 return 0;
1560 }
1561
1562 /*
1563 * extent_io.c call back to do delayed allocation processing
1564 */
1565 static int run_delalloc_range(void *private_data, struct page *locked_page,
1566 u64 start, u64 end, int *page_started,
1567 unsigned long *nr_written)
1568 {
1569 struct inode *inode = private_data;
1570 int ret;
1571 int force_cow = need_force_cow(inode, start, end);
1572
1573 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1574 ret = run_delalloc_nocow(inode, locked_page, start, end,
1575 page_started, 1, nr_written);
1576 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1577 ret = run_delalloc_nocow(inode, locked_page, start, end,
1578 page_started, 0, nr_written);
1579 } else if (!inode_need_compress(inode)) {
1580 ret = cow_file_range(inode, locked_page, start, end, end,
1581 page_started, nr_written, 1, NULL);
1582 } else {
1583 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1584 &BTRFS_I(inode)->runtime_flags);
1585 ret = cow_file_range_async(inode, locked_page, start, end,
1586 page_started, nr_written);
1587 }
1588 if (ret)
1589 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1590 return ret;
1591 }
1592
1593 static void btrfs_split_extent_hook(void *private_data,
1594 struct extent_state *orig, u64 split)
1595 {
1596 struct inode *inode = private_data;
1597 u64 size;
1598
1599 /* not delalloc, ignore it */
1600 if (!(orig->state & EXTENT_DELALLOC))
1601 return;
1602
1603 size = orig->end - orig->start + 1;
1604 if (size > BTRFS_MAX_EXTENT_SIZE) {
1605 u32 num_extents;
1606 u64 new_size;
1607
1608 /*
1609 * See the explanation in btrfs_merge_extent_hook, the same
1610 * applies here, just in reverse.
1611 */
1612 new_size = orig->end - split + 1;
1613 num_extents = count_max_extents(new_size);
1614 new_size = split - orig->start;
1615 num_extents += count_max_extents(new_size);
1616 if (count_max_extents(size) >= num_extents)
1617 return;
1618 }
1619
1620 spin_lock(&BTRFS_I(inode)->lock);
1621 BTRFS_I(inode)->outstanding_extents++;
1622 spin_unlock(&BTRFS_I(inode)->lock);
1623 }
1624
1625 /*
1626 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1627 * extents so we can keep track of new extents that are just merged onto old
1628 * extents, such as when we are doing sequential writes, so we can properly
1629 * account for the metadata space we'll need.
1630 */
1631 static void btrfs_merge_extent_hook(void *private_data,
1632 struct extent_state *new,
1633 struct extent_state *other)
1634 {
1635 struct inode *inode = private_data;
1636 u64 new_size, old_size;
1637 u32 num_extents;
1638
1639 /* not delalloc, ignore it */
1640 if (!(other->state & EXTENT_DELALLOC))
1641 return;
1642
1643 if (new->start > other->start)
1644 new_size = new->end - other->start + 1;
1645 else
1646 new_size = other->end - new->start + 1;
1647
1648 /* we're not bigger than the max, unreserve the space and go */
1649 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1650 spin_lock(&BTRFS_I(inode)->lock);
1651 BTRFS_I(inode)->outstanding_extents--;
1652 spin_unlock(&BTRFS_I(inode)->lock);
1653 return;
1654 }
1655
1656 /*
1657 * We have to add up either side to figure out how many extents were
1658 * accounted for before we merged into one big extent. If the number of
1659 * extents we accounted for is <= the amount we need for the new range
1660 * then we can return, otherwise drop. Think of it like this
1661 *
1662 * [ 4k][MAX_SIZE]
1663 *
1664 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1665 * need 2 outstanding extents, on one side we have 1 and the other side
1666 * we have 1 so they are == and we can return. But in this case
1667 *
1668 * [MAX_SIZE+4k][MAX_SIZE+4k]
1669 *
1670 * Each range on their own accounts for 2 extents, but merged together
1671 * they are only 3 extents worth of accounting, so we need to drop in
1672 * this case.
1673 */
1674 old_size = other->end - other->start + 1;
1675 num_extents = count_max_extents(old_size);
1676 old_size = new->end - new->start + 1;
1677 num_extents += count_max_extents(old_size);
1678 if (count_max_extents(new_size) >= num_extents)
1679 return;
1680
1681 spin_lock(&BTRFS_I(inode)->lock);
1682 BTRFS_I(inode)->outstanding_extents--;
1683 spin_unlock(&BTRFS_I(inode)->lock);
1684 }
1685
1686 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1687 struct inode *inode)
1688 {
1689 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1690
1691 spin_lock(&root->delalloc_lock);
1692 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1693 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1694 &root->delalloc_inodes);
1695 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1696 &BTRFS_I(inode)->runtime_flags);
1697 root->nr_delalloc_inodes++;
1698 if (root->nr_delalloc_inodes == 1) {
1699 spin_lock(&fs_info->delalloc_root_lock);
1700 BUG_ON(!list_empty(&root->delalloc_root));
1701 list_add_tail(&root->delalloc_root,
1702 &fs_info->delalloc_roots);
1703 spin_unlock(&fs_info->delalloc_root_lock);
1704 }
1705 }
1706 spin_unlock(&root->delalloc_lock);
1707 }
1708
1709 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1710 struct btrfs_inode *inode)
1711 {
1712 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1713
1714 spin_lock(&root->delalloc_lock);
1715 if (!list_empty(&inode->delalloc_inodes)) {
1716 list_del_init(&inode->delalloc_inodes);
1717 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1718 &inode->runtime_flags);
1719 root->nr_delalloc_inodes--;
1720 if (!root->nr_delalloc_inodes) {
1721 spin_lock(&fs_info->delalloc_root_lock);
1722 BUG_ON(list_empty(&root->delalloc_root));
1723 list_del_init(&root->delalloc_root);
1724 spin_unlock(&fs_info->delalloc_root_lock);
1725 }
1726 }
1727 spin_unlock(&root->delalloc_lock);
1728 }
1729
1730 /*
1731 * extent_io.c set_bit_hook, used to track delayed allocation
1732 * bytes in this file, and to maintain the list of inodes that
1733 * have pending delalloc work to be done.
1734 */
1735 static void btrfs_set_bit_hook(void *private_data,
1736 struct extent_state *state, unsigned *bits)
1737 {
1738 struct inode *inode = private_data;
1739
1740 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1741
1742 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1743 WARN_ON(1);
1744 /*
1745 * set_bit and clear bit hooks normally require _irqsave/restore
1746 * but in this case, we are only testing for the DELALLOC
1747 * bit, which is only set or cleared with irqs on
1748 */
1749 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1750 struct btrfs_root *root = BTRFS_I(inode)->root;
1751 u64 len = state->end + 1 - state->start;
1752 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1753
1754 if (*bits & EXTENT_FIRST_DELALLOC) {
1755 *bits &= ~EXTENT_FIRST_DELALLOC;
1756 } else {
1757 spin_lock(&BTRFS_I(inode)->lock);
1758 BTRFS_I(inode)->outstanding_extents++;
1759 spin_unlock(&BTRFS_I(inode)->lock);
1760 }
1761
1762 /* For sanity tests */
1763 if (btrfs_is_testing(fs_info))
1764 return;
1765
1766 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1767 fs_info->delalloc_batch);
1768 spin_lock(&BTRFS_I(inode)->lock);
1769 BTRFS_I(inode)->delalloc_bytes += len;
1770 if (*bits & EXTENT_DEFRAG)
1771 BTRFS_I(inode)->defrag_bytes += len;
1772 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1773 &BTRFS_I(inode)->runtime_flags))
1774 btrfs_add_delalloc_inodes(root, inode);
1775 spin_unlock(&BTRFS_I(inode)->lock);
1776 }
1777
1778 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1779 (*bits & EXTENT_DELALLOC_NEW)) {
1780 spin_lock(&BTRFS_I(inode)->lock);
1781 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1782 state->start;
1783 spin_unlock(&BTRFS_I(inode)->lock);
1784 }
1785 }
1786
1787 /*
1788 * extent_io.c clear_bit_hook, see set_bit_hook for why
1789 */
1790 static void btrfs_clear_bit_hook(void *private_data,
1791 struct extent_state *state,
1792 unsigned *bits)
1793 {
1794 struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1795 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1796 u64 len = state->end + 1 - state->start;
1797 u32 num_extents = count_max_extents(len);
1798
1799 spin_lock(&inode->lock);
1800 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1801 inode->defrag_bytes -= len;
1802 spin_unlock(&inode->lock);
1803
1804 /*
1805 * set_bit and clear bit hooks normally require _irqsave/restore
1806 * but in this case, we are only testing for the DELALLOC
1807 * bit, which is only set or cleared with irqs on
1808 */
1809 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1810 struct btrfs_root *root = inode->root;
1811 bool do_list = !btrfs_is_free_space_inode(inode);
1812
1813 if (*bits & EXTENT_FIRST_DELALLOC) {
1814 *bits &= ~EXTENT_FIRST_DELALLOC;
1815 } else if (!(*bits & EXTENT_CLEAR_META_RESV)) {
1816 spin_lock(&inode->lock);
1817 inode->outstanding_extents -= num_extents;
1818 spin_unlock(&inode->lock);
1819 }
1820
1821 /*
1822 * We don't reserve metadata space for space cache inodes so we
1823 * don't need to call dellalloc_release_metadata if there is an
1824 * error.
1825 */
1826 if (*bits & EXTENT_CLEAR_META_RESV &&
1827 root != fs_info->tree_root)
1828 btrfs_delalloc_release_metadata(inode, len);
1829
1830 /* For sanity tests. */
1831 if (btrfs_is_testing(fs_info))
1832 return;
1833
1834 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1835 do_list && !(state->state & EXTENT_NORESERVE) &&
1836 (*bits & EXTENT_CLEAR_DATA_RESV))
1837 btrfs_free_reserved_data_space_noquota(
1838 &inode->vfs_inode,
1839 state->start, len);
1840
1841 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1842 fs_info->delalloc_batch);
1843 spin_lock(&inode->lock);
1844 inode->delalloc_bytes -= len;
1845 if (do_list && inode->delalloc_bytes == 0 &&
1846 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1847 &inode->runtime_flags))
1848 btrfs_del_delalloc_inode(root, inode);
1849 spin_unlock(&inode->lock);
1850 }
1851
1852 if ((state->state & EXTENT_DELALLOC_NEW) &&
1853 (*bits & EXTENT_DELALLOC_NEW)) {
1854 spin_lock(&inode->lock);
1855 ASSERT(inode->new_delalloc_bytes >= len);
1856 inode->new_delalloc_bytes -= len;
1857 spin_unlock(&inode->lock);
1858 }
1859 }
1860
1861 /*
1862 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1863 * we don't create bios that span stripes or chunks
1864 *
1865 * return 1 if page cannot be merged to bio
1866 * return 0 if page can be merged to bio
1867 * return error otherwise
1868 */
1869 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1870 size_t size, struct bio *bio,
1871 unsigned long bio_flags)
1872 {
1873 struct inode *inode = page->mapping->host;
1874 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1875 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1876 u64 length = 0;
1877 u64 map_length;
1878 int ret;
1879
1880 if (bio_flags & EXTENT_BIO_COMPRESSED)
1881 return 0;
1882
1883 length = bio->bi_iter.bi_size;
1884 map_length = length;
1885 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1886 NULL, 0);
1887 if (ret < 0)
1888 return ret;
1889 if (map_length < length + size)
1890 return 1;
1891 return 0;
1892 }
1893
1894 /*
1895 * in order to insert checksums into the metadata in large chunks,
1896 * we wait until bio submission time. All the pages in the bio are
1897 * checksummed and sums are attached onto the ordered extent record.
1898 *
1899 * At IO completion time the cums attached on the ordered extent record
1900 * are inserted into the btree
1901 */
1902 static blk_status_t __btrfs_submit_bio_start(void *private_data, struct bio *bio,
1903 int mirror_num, unsigned long bio_flags,
1904 u64 bio_offset)
1905 {
1906 struct inode *inode = private_data;
1907 blk_status_t ret = 0;
1908
1909 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1910 BUG_ON(ret); /* -ENOMEM */
1911 return 0;
1912 }
1913
1914 /*
1915 * in order to insert checksums into the metadata in large chunks,
1916 * we wait until bio submission time. All the pages in the bio are
1917 * checksummed and sums are attached onto the ordered extent record.
1918 *
1919 * At IO completion time the cums attached on the ordered extent record
1920 * are inserted into the btree
1921 */
1922 static blk_status_t __btrfs_submit_bio_done(void *private_data, struct bio *bio,
1923 int mirror_num, unsigned long bio_flags,
1924 u64 bio_offset)
1925 {
1926 struct inode *inode = private_data;
1927 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1928 blk_status_t ret;
1929
1930 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1931 if (ret) {
1932 bio->bi_status = ret;
1933 bio_endio(bio);
1934 }
1935 return ret;
1936 }
1937
1938 /*
1939 * extent_io.c submission hook. This does the right thing for csum calculation
1940 * on write, or reading the csums from the tree before a read
1941 */
1942 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1943 int mirror_num, unsigned long bio_flags,
1944 u64 bio_offset)
1945 {
1946 struct inode *inode = private_data;
1947 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1948 struct btrfs_root *root = BTRFS_I(inode)->root;
1949 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1950 blk_status_t ret = 0;
1951 int skip_sum;
1952 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1953
1954 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1955
1956 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1957 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1958
1959 if (bio_op(bio) != REQ_OP_WRITE) {
1960 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1961 if (ret)
1962 goto out;
1963
1964 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1965 ret = btrfs_submit_compressed_read(inode, bio,
1966 mirror_num,
1967 bio_flags);
1968 goto out;
1969 } else if (!skip_sum) {
1970 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
1971 if (ret)
1972 goto out;
1973 }
1974 goto mapit;
1975 } else if (async && !skip_sum) {
1976 /* csum items have already been cloned */
1977 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1978 goto mapit;
1979 /* we're doing a write, do the async checksumming */
1980 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
1981 bio_offset, inode,
1982 __btrfs_submit_bio_start,
1983 __btrfs_submit_bio_done);
1984 goto out;
1985 } else if (!skip_sum) {
1986 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1987 if (ret)
1988 goto out;
1989 }
1990
1991 mapit:
1992 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1993
1994 out:
1995 if (ret) {
1996 bio->bi_status = ret;
1997 bio_endio(bio);
1998 }
1999 return ret;
2000 }
2001
2002 /*
2003 * given a list of ordered sums record them in the inode. This happens
2004 * at IO completion time based on sums calculated at bio submission time.
2005 */
2006 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2007 struct inode *inode, struct list_head *list)
2008 {
2009 struct btrfs_ordered_sum *sum;
2010
2011 list_for_each_entry(sum, list, list) {
2012 trans->adding_csums = 1;
2013 btrfs_csum_file_blocks(trans,
2014 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2015 trans->adding_csums = 0;
2016 }
2017 return 0;
2018 }
2019
2020 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2021 struct extent_state **cached_state, int dedupe)
2022 {
2023 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2024 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2025 cached_state);
2026 }
2027
2028 /* see btrfs_writepage_start_hook for details on why this is required */
2029 struct btrfs_writepage_fixup {
2030 struct page *page;
2031 struct btrfs_work work;
2032 };
2033
2034 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2035 {
2036 struct btrfs_writepage_fixup *fixup;
2037 struct btrfs_ordered_extent *ordered;
2038 struct extent_state *cached_state = NULL;
2039 struct extent_changeset *data_reserved = NULL;
2040 struct page *page;
2041 struct inode *inode;
2042 u64 page_start;
2043 u64 page_end;
2044 int ret;
2045
2046 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2047 page = fixup->page;
2048 again:
2049 lock_page(page);
2050 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2051 ClearPageChecked(page);
2052 goto out_page;
2053 }
2054
2055 inode = page->mapping->host;
2056 page_start = page_offset(page);
2057 page_end = page_offset(page) + PAGE_SIZE - 1;
2058
2059 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2060 &cached_state);
2061
2062 /* already ordered? We're done */
2063 if (PagePrivate2(page))
2064 goto out;
2065
2066 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2067 PAGE_SIZE);
2068 if (ordered) {
2069 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2070 page_end, &cached_state, GFP_NOFS);
2071 unlock_page(page);
2072 btrfs_start_ordered_extent(inode, ordered, 1);
2073 btrfs_put_ordered_extent(ordered);
2074 goto again;
2075 }
2076
2077 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2078 PAGE_SIZE);
2079 if (ret) {
2080 mapping_set_error(page->mapping, ret);
2081 end_extent_writepage(page, ret, page_start, page_end);
2082 ClearPageChecked(page);
2083 goto out;
2084 }
2085
2086 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state,
2087 0);
2088 ClearPageChecked(page);
2089 set_page_dirty(page);
2090 out:
2091 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2092 &cached_state, GFP_NOFS);
2093 out_page:
2094 unlock_page(page);
2095 put_page(page);
2096 kfree(fixup);
2097 extent_changeset_free(data_reserved);
2098 }
2099
2100 /*
2101 * There are a few paths in the higher layers of the kernel that directly
2102 * set the page dirty bit without asking the filesystem if it is a
2103 * good idea. This causes problems because we want to make sure COW
2104 * properly happens and the data=ordered rules are followed.
2105 *
2106 * In our case any range that doesn't have the ORDERED bit set
2107 * hasn't been properly setup for IO. We kick off an async process
2108 * to fix it up. The async helper will wait for ordered extents, set
2109 * the delalloc bit and make it safe to write the page.
2110 */
2111 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2112 {
2113 struct inode *inode = page->mapping->host;
2114 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2115 struct btrfs_writepage_fixup *fixup;
2116
2117 /* this page is properly in the ordered list */
2118 if (TestClearPagePrivate2(page))
2119 return 0;
2120
2121 if (PageChecked(page))
2122 return -EAGAIN;
2123
2124 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2125 if (!fixup)
2126 return -EAGAIN;
2127
2128 SetPageChecked(page);
2129 get_page(page);
2130 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2131 btrfs_writepage_fixup_worker, NULL, NULL);
2132 fixup->page = page;
2133 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2134 return -EBUSY;
2135 }
2136
2137 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2138 struct inode *inode, u64 file_pos,
2139 u64 disk_bytenr, u64 disk_num_bytes,
2140 u64 num_bytes, u64 ram_bytes,
2141 u8 compression, u8 encryption,
2142 u16 other_encoding, int extent_type)
2143 {
2144 struct btrfs_root *root = BTRFS_I(inode)->root;
2145 struct btrfs_file_extent_item *fi;
2146 struct btrfs_path *path;
2147 struct extent_buffer *leaf;
2148 struct btrfs_key ins;
2149 u64 qg_released;
2150 int extent_inserted = 0;
2151 int ret;
2152
2153 path = btrfs_alloc_path();
2154 if (!path)
2155 return -ENOMEM;
2156
2157 /*
2158 * we may be replacing one extent in the tree with another.
2159 * The new extent is pinned in the extent map, and we don't want
2160 * to drop it from the cache until it is completely in the btree.
2161 *
2162 * So, tell btrfs_drop_extents to leave this extent in the cache.
2163 * the caller is expected to unpin it and allow it to be merged
2164 * with the others.
2165 */
2166 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2167 file_pos + num_bytes, NULL, 0,
2168 1, sizeof(*fi), &extent_inserted);
2169 if (ret)
2170 goto out;
2171
2172 if (!extent_inserted) {
2173 ins.objectid = btrfs_ino(BTRFS_I(inode));
2174 ins.offset = file_pos;
2175 ins.type = BTRFS_EXTENT_DATA_KEY;
2176
2177 path->leave_spinning = 1;
2178 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2179 sizeof(*fi));
2180 if (ret)
2181 goto out;
2182 }
2183 leaf = path->nodes[0];
2184 fi = btrfs_item_ptr(leaf, path->slots[0],
2185 struct btrfs_file_extent_item);
2186 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2187 btrfs_set_file_extent_type(leaf, fi, extent_type);
2188 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2189 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2190 btrfs_set_file_extent_offset(leaf, fi, 0);
2191 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2192 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2193 btrfs_set_file_extent_compression(leaf, fi, compression);
2194 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2195 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2196
2197 btrfs_mark_buffer_dirty(leaf);
2198 btrfs_release_path(path);
2199
2200 inode_add_bytes(inode, num_bytes);
2201
2202 ins.objectid = disk_bytenr;
2203 ins.offset = disk_num_bytes;
2204 ins.type = BTRFS_EXTENT_ITEM_KEY;
2205
2206 /*
2207 * Release the reserved range from inode dirty range map, as it is
2208 * already moved into delayed_ref_head
2209 */
2210 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2211 if (ret < 0)
2212 goto out;
2213 qg_released = ret;
2214 ret = btrfs_alloc_reserved_file_extent(trans, root->root_key.objectid,
2215 btrfs_ino(BTRFS_I(inode)), file_pos, qg_released, &ins);
2216 out:
2217 btrfs_free_path(path);
2218
2219 return ret;
2220 }
2221
2222 /* snapshot-aware defrag */
2223 struct sa_defrag_extent_backref {
2224 struct rb_node node;
2225 struct old_sa_defrag_extent *old;
2226 u64 root_id;
2227 u64 inum;
2228 u64 file_pos;
2229 u64 extent_offset;
2230 u64 num_bytes;
2231 u64 generation;
2232 };
2233
2234 struct old_sa_defrag_extent {
2235 struct list_head list;
2236 struct new_sa_defrag_extent *new;
2237
2238 u64 extent_offset;
2239 u64 bytenr;
2240 u64 offset;
2241 u64 len;
2242 int count;
2243 };
2244
2245 struct new_sa_defrag_extent {
2246 struct rb_root root;
2247 struct list_head head;
2248 struct btrfs_path *path;
2249 struct inode *inode;
2250 u64 file_pos;
2251 u64 len;
2252 u64 bytenr;
2253 u64 disk_len;
2254 u8 compress_type;
2255 };
2256
2257 static int backref_comp(struct sa_defrag_extent_backref *b1,
2258 struct sa_defrag_extent_backref *b2)
2259 {
2260 if (b1->root_id < b2->root_id)
2261 return -1;
2262 else if (b1->root_id > b2->root_id)
2263 return 1;
2264
2265 if (b1->inum < b2->inum)
2266 return -1;
2267 else if (b1->inum > b2->inum)
2268 return 1;
2269
2270 if (b1->file_pos < b2->file_pos)
2271 return -1;
2272 else if (b1->file_pos > b2->file_pos)
2273 return 1;
2274
2275 /*
2276 * [------------------------------] ===> (a range of space)
2277 * |<--->| |<---->| =============> (fs/file tree A)
2278 * |<---------------------------->| ===> (fs/file tree B)
2279 *
2280 * A range of space can refer to two file extents in one tree while
2281 * refer to only one file extent in another tree.
2282 *
2283 * So we may process a disk offset more than one time(two extents in A)
2284 * and locate at the same extent(one extent in B), then insert two same
2285 * backrefs(both refer to the extent in B).
2286 */
2287 return 0;
2288 }
2289
2290 static void backref_insert(struct rb_root *root,
2291 struct sa_defrag_extent_backref *backref)
2292 {
2293 struct rb_node **p = &root->rb_node;
2294 struct rb_node *parent = NULL;
2295 struct sa_defrag_extent_backref *entry;
2296 int ret;
2297
2298 while (*p) {
2299 parent = *p;
2300 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2301
2302 ret = backref_comp(backref, entry);
2303 if (ret < 0)
2304 p = &(*p)->rb_left;
2305 else
2306 p = &(*p)->rb_right;
2307 }
2308
2309 rb_link_node(&backref->node, parent, p);
2310 rb_insert_color(&backref->node, root);
2311 }
2312
2313 /*
2314 * Note the backref might has changed, and in this case we just return 0.
2315 */
2316 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2317 void *ctx)
2318 {
2319 struct btrfs_file_extent_item *extent;
2320 struct old_sa_defrag_extent *old = ctx;
2321 struct new_sa_defrag_extent *new = old->new;
2322 struct btrfs_path *path = new->path;
2323 struct btrfs_key key;
2324 struct btrfs_root *root;
2325 struct sa_defrag_extent_backref *backref;
2326 struct extent_buffer *leaf;
2327 struct inode *inode = new->inode;
2328 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2329 int slot;
2330 int ret;
2331 u64 extent_offset;
2332 u64 num_bytes;
2333
2334 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2335 inum == btrfs_ino(BTRFS_I(inode)))
2336 return 0;
2337
2338 key.objectid = root_id;
2339 key.type = BTRFS_ROOT_ITEM_KEY;
2340 key.offset = (u64)-1;
2341
2342 root = btrfs_read_fs_root_no_name(fs_info, &key);
2343 if (IS_ERR(root)) {
2344 if (PTR_ERR(root) == -ENOENT)
2345 return 0;
2346 WARN_ON(1);
2347 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2348 inum, offset, root_id);
2349 return PTR_ERR(root);
2350 }
2351
2352 key.objectid = inum;
2353 key.type = BTRFS_EXTENT_DATA_KEY;
2354 if (offset > (u64)-1 << 32)
2355 key.offset = 0;
2356 else
2357 key.offset = offset;
2358
2359 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2360 if (WARN_ON(ret < 0))
2361 return ret;
2362 ret = 0;
2363
2364 while (1) {
2365 cond_resched();
2366
2367 leaf = path->nodes[0];
2368 slot = path->slots[0];
2369
2370 if (slot >= btrfs_header_nritems(leaf)) {
2371 ret = btrfs_next_leaf(root, path);
2372 if (ret < 0) {
2373 goto out;
2374 } else if (ret > 0) {
2375 ret = 0;
2376 goto out;
2377 }
2378 continue;
2379 }
2380
2381 path->slots[0]++;
2382
2383 btrfs_item_key_to_cpu(leaf, &key, slot);
2384
2385 if (key.objectid > inum)
2386 goto out;
2387
2388 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2389 continue;
2390
2391 extent = btrfs_item_ptr(leaf, slot,
2392 struct btrfs_file_extent_item);
2393
2394 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2395 continue;
2396
2397 /*
2398 * 'offset' refers to the exact key.offset,
2399 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2400 * (key.offset - extent_offset).
2401 */
2402 if (key.offset != offset)
2403 continue;
2404
2405 extent_offset = btrfs_file_extent_offset(leaf, extent);
2406 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2407
2408 if (extent_offset >= old->extent_offset + old->offset +
2409 old->len || extent_offset + num_bytes <=
2410 old->extent_offset + old->offset)
2411 continue;
2412 break;
2413 }
2414
2415 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2416 if (!backref) {
2417 ret = -ENOENT;
2418 goto out;
2419 }
2420
2421 backref->root_id = root_id;
2422 backref->inum = inum;
2423 backref->file_pos = offset;
2424 backref->num_bytes = num_bytes;
2425 backref->extent_offset = extent_offset;
2426 backref->generation = btrfs_file_extent_generation(leaf, extent);
2427 backref->old = old;
2428 backref_insert(&new->root, backref);
2429 old->count++;
2430 out:
2431 btrfs_release_path(path);
2432 WARN_ON(ret);
2433 return ret;
2434 }
2435
2436 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2437 struct new_sa_defrag_extent *new)
2438 {
2439 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2440 struct old_sa_defrag_extent *old, *tmp;
2441 int ret;
2442
2443 new->path = path;
2444
2445 list_for_each_entry_safe(old, tmp, &new->head, list) {
2446 ret = iterate_inodes_from_logical(old->bytenr +
2447 old->extent_offset, fs_info,
2448 path, record_one_backref,
2449 old);
2450 if (ret < 0 && ret != -ENOENT)
2451 return false;
2452
2453 /* no backref to be processed for this extent */
2454 if (!old->count) {
2455 list_del(&old->list);
2456 kfree(old);
2457 }
2458 }
2459
2460 if (list_empty(&new->head))
2461 return false;
2462
2463 return true;
2464 }
2465
2466 static int relink_is_mergable(struct extent_buffer *leaf,
2467 struct btrfs_file_extent_item *fi,
2468 struct new_sa_defrag_extent *new)
2469 {
2470 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2471 return 0;
2472
2473 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2474 return 0;
2475
2476 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2477 return 0;
2478
2479 if (btrfs_file_extent_encryption(leaf, fi) ||
2480 btrfs_file_extent_other_encoding(leaf, fi))
2481 return 0;
2482
2483 return 1;
2484 }
2485
2486 /*
2487 * Note the backref might has changed, and in this case we just return 0.
2488 */
2489 static noinline int relink_extent_backref(struct btrfs_path *path,
2490 struct sa_defrag_extent_backref *prev,
2491 struct sa_defrag_extent_backref *backref)
2492 {
2493 struct btrfs_file_extent_item *extent;
2494 struct btrfs_file_extent_item *item;
2495 struct btrfs_ordered_extent *ordered;
2496 struct btrfs_trans_handle *trans;
2497 struct btrfs_root *root;
2498 struct btrfs_key key;
2499 struct extent_buffer *leaf;
2500 struct old_sa_defrag_extent *old = backref->old;
2501 struct new_sa_defrag_extent *new = old->new;
2502 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2503 struct inode *inode;
2504 struct extent_state *cached = NULL;
2505 int ret = 0;
2506 u64 start;
2507 u64 len;
2508 u64 lock_start;
2509 u64 lock_end;
2510 bool merge = false;
2511 int index;
2512
2513 if (prev && prev->root_id == backref->root_id &&
2514 prev->inum == backref->inum &&
2515 prev->file_pos + prev->num_bytes == backref->file_pos)
2516 merge = true;
2517
2518 /* step 1: get root */
2519 key.objectid = backref->root_id;
2520 key.type = BTRFS_ROOT_ITEM_KEY;
2521 key.offset = (u64)-1;
2522
2523 index = srcu_read_lock(&fs_info->subvol_srcu);
2524
2525 root = btrfs_read_fs_root_no_name(fs_info, &key);
2526 if (IS_ERR(root)) {
2527 srcu_read_unlock(&fs_info->subvol_srcu, index);
2528 if (PTR_ERR(root) == -ENOENT)
2529 return 0;
2530 return PTR_ERR(root);
2531 }
2532
2533 if (btrfs_root_readonly(root)) {
2534 srcu_read_unlock(&fs_info->subvol_srcu, index);
2535 return 0;
2536 }
2537
2538 /* step 2: get inode */
2539 key.objectid = backref->inum;
2540 key.type = BTRFS_INODE_ITEM_KEY;
2541 key.offset = 0;
2542
2543 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2544 if (IS_ERR(inode)) {
2545 srcu_read_unlock(&fs_info->subvol_srcu, index);
2546 return 0;
2547 }
2548
2549 srcu_read_unlock(&fs_info->subvol_srcu, index);
2550
2551 /* step 3: relink backref */
2552 lock_start = backref->file_pos;
2553 lock_end = backref->file_pos + backref->num_bytes - 1;
2554 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2555 &cached);
2556
2557 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2558 if (ordered) {
2559 btrfs_put_ordered_extent(ordered);
2560 goto out_unlock;
2561 }
2562
2563 trans = btrfs_join_transaction(root);
2564 if (IS_ERR(trans)) {
2565 ret = PTR_ERR(trans);
2566 goto out_unlock;
2567 }
2568
2569 key.objectid = backref->inum;
2570 key.type = BTRFS_EXTENT_DATA_KEY;
2571 key.offset = backref->file_pos;
2572
2573 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2574 if (ret < 0) {
2575 goto out_free_path;
2576 } else if (ret > 0) {
2577 ret = 0;
2578 goto out_free_path;
2579 }
2580
2581 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2582 struct btrfs_file_extent_item);
2583
2584 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2585 backref->generation)
2586 goto out_free_path;
2587
2588 btrfs_release_path(path);
2589
2590 start = backref->file_pos;
2591 if (backref->extent_offset < old->extent_offset + old->offset)
2592 start += old->extent_offset + old->offset -
2593 backref->extent_offset;
2594
2595 len = min(backref->extent_offset + backref->num_bytes,
2596 old->extent_offset + old->offset + old->len);
2597 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2598
2599 ret = btrfs_drop_extents(trans, root, inode, start,
2600 start + len, 1);
2601 if (ret)
2602 goto out_free_path;
2603 again:
2604 key.objectid = btrfs_ino(BTRFS_I(inode));
2605 key.type = BTRFS_EXTENT_DATA_KEY;
2606 key.offset = start;
2607
2608 path->leave_spinning = 1;
2609 if (merge) {
2610 struct btrfs_file_extent_item *fi;
2611 u64 extent_len;
2612 struct btrfs_key found_key;
2613
2614 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2615 if (ret < 0)
2616 goto out_free_path;
2617
2618 path->slots[0]--;
2619 leaf = path->nodes[0];
2620 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2621
2622 fi = btrfs_item_ptr(leaf, path->slots[0],
2623 struct btrfs_file_extent_item);
2624 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2625
2626 if (extent_len + found_key.offset == start &&
2627 relink_is_mergable(leaf, fi, new)) {
2628 btrfs_set_file_extent_num_bytes(leaf, fi,
2629 extent_len + len);
2630 btrfs_mark_buffer_dirty(leaf);
2631 inode_add_bytes(inode, len);
2632
2633 ret = 1;
2634 goto out_free_path;
2635 } else {
2636 merge = false;
2637 btrfs_release_path(path);
2638 goto again;
2639 }
2640 }
2641
2642 ret = btrfs_insert_empty_item(trans, root, path, &key,
2643 sizeof(*extent));
2644 if (ret) {
2645 btrfs_abort_transaction(trans, ret);
2646 goto out_free_path;
2647 }
2648
2649 leaf = path->nodes[0];
2650 item = btrfs_item_ptr(leaf, path->slots[0],
2651 struct btrfs_file_extent_item);
2652 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2653 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2654 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2655 btrfs_set_file_extent_num_bytes(leaf, item, len);
2656 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2657 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2658 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2659 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2660 btrfs_set_file_extent_encryption(leaf, item, 0);
2661 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2662
2663 btrfs_mark_buffer_dirty(leaf);
2664 inode_add_bytes(inode, len);
2665 btrfs_release_path(path);
2666
2667 ret = btrfs_inc_extent_ref(trans, fs_info, new->bytenr,
2668 new->disk_len, 0,
2669 backref->root_id, backref->inum,
2670 new->file_pos); /* start - extent_offset */
2671 if (ret) {
2672 btrfs_abort_transaction(trans, ret);
2673 goto out_free_path;
2674 }
2675
2676 ret = 1;
2677 out_free_path:
2678 btrfs_release_path(path);
2679 path->leave_spinning = 0;
2680 btrfs_end_transaction(trans);
2681 out_unlock:
2682 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2683 &cached, GFP_NOFS);
2684 iput(inode);
2685 return ret;
2686 }
2687
2688 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2689 {
2690 struct old_sa_defrag_extent *old, *tmp;
2691
2692 if (!new)
2693 return;
2694
2695 list_for_each_entry_safe(old, tmp, &new->head, list) {
2696 kfree(old);
2697 }
2698 kfree(new);
2699 }
2700
2701 static void relink_file_extents(struct new_sa_defrag_extent *new)
2702 {
2703 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2704 struct btrfs_path *path;
2705 struct sa_defrag_extent_backref *backref;
2706 struct sa_defrag_extent_backref *prev = NULL;
2707 struct inode *inode;
2708 struct btrfs_root *root;
2709 struct rb_node *node;
2710 int ret;
2711
2712 inode = new->inode;
2713 root = BTRFS_I(inode)->root;
2714
2715 path = btrfs_alloc_path();
2716 if (!path)
2717 return;
2718
2719 if (!record_extent_backrefs(path, new)) {
2720 btrfs_free_path(path);
2721 goto out;
2722 }
2723 btrfs_release_path(path);
2724
2725 while (1) {
2726 node = rb_first(&new->root);
2727 if (!node)
2728 break;
2729 rb_erase(node, &new->root);
2730
2731 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2732
2733 ret = relink_extent_backref(path, prev, backref);
2734 WARN_ON(ret < 0);
2735
2736 kfree(prev);
2737
2738 if (ret == 1)
2739 prev = backref;
2740 else
2741 prev = NULL;
2742 cond_resched();
2743 }
2744 kfree(prev);
2745
2746 btrfs_free_path(path);
2747 out:
2748 free_sa_defrag_extent(new);
2749
2750 atomic_dec(&fs_info->defrag_running);
2751 wake_up(&fs_info->transaction_wait);
2752 }
2753
2754 static struct new_sa_defrag_extent *
2755 record_old_file_extents(struct inode *inode,
2756 struct btrfs_ordered_extent *ordered)
2757 {
2758 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2759 struct btrfs_root *root = BTRFS_I(inode)->root;
2760 struct btrfs_path *path;
2761 struct btrfs_key key;
2762 struct old_sa_defrag_extent *old;
2763 struct new_sa_defrag_extent *new;
2764 int ret;
2765
2766 new = kmalloc(sizeof(*new), GFP_NOFS);
2767 if (!new)
2768 return NULL;
2769
2770 new->inode = inode;
2771 new->file_pos = ordered->file_offset;
2772 new->len = ordered->len;
2773 new->bytenr = ordered->start;
2774 new->disk_len = ordered->disk_len;
2775 new->compress_type = ordered->compress_type;
2776 new->root = RB_ROOT;
2777 INIT_LIST_HEAD(&new->head);
2778
2779 path = btrfs_alloc_path();
2780 if (!path)
2781 goto out_kfree;
2782
2783 key.objectid = btrfs_ino(BTRFS_I(inode));
2784 key.type = BTRFS_EXTENT_DATA_KEY;
2785 key.offset = new->file_pos;
2786
2787 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2788 if (ret < 0)
2789 goto out_free_path;
2790 if (ret > 0 && path->slots[0] > 0)
2791 path->slots[0]--;
2792
2793 /* find out all the old extents for the file range */
2794 while (1) {
2795 struct btrfs_file_extent_item *extent;
2796 struct extent_buffer *l;
2797 int slot;
2798 u64 num_bytes;
2799 u64 offset;
2800 u64 end;
2801 u64 disk_bytenr;
2802 u64 extent_offset;
2803
2804 l = path->nodes[0];
2805 slot = path->slots[0];
2806
2807 if (slot >= btrfs_header_nritems(l)) {
2808 ret = btrfs_next_leaf(root, path);
2809 if (ret < 0)
2810 goto out_free_path;
2811 else if (ret > 0)
2812 break;
2813 continue;
2814 }
2815
2816 btrfs_item_key_to_cpu(l, &key, slot);
2817
2818 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2819 break;
2820 if (key.type != BTRFS_EXTENT_DATA_KEY)
2821 break;
2822 if (key.offset >= new->file_pos + new->len)
2823 break;
2824
2825 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2826
2827 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2828 if (key.offset + num_bytes < new->file_pos)
2829 goto next;
2830
2831 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2832 if (!disk_bytenr)
2833 goto next;
2834
2835 extent_offset = btrfs_file_extent_offset(l, extent);
2836
2837 old = kmalloc(sizeof(*old), GFP_NOFS);
2838 if (!old)
2839 goto out_free_path;
2840
2841 offset = max(new->file_pos, key.offset);
2842 end = min(new->file_pos + new->len, key.offset + num_bytes);
2843
2844 old->bytenr = disk_bytenr;
2845 old->extent_offset = extent_offset;
2846 old->offset = offset - key.offset;
2847 old->len = end - offset;
2848 old->new = new;
2849 old->count = 0;
2850 list_add_tail(&old->list, &new->head);
2851 next:
2852 path->slots[0]++;
2853 cond_resched();
2854 }
2855
2856 btrfs_free_path(path);
2857 atomic_inc(&fs_info->defrag_running);
2858
2859 return new;
2860
2861 out_free_path:
2862 btrfs_free_path(path);
2863 out_kfree:
2864 free_sa_defrag_extent(new);
2865 return NULL;
2866 }
2867
2868 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2869 u64 start, u64 len)
2870 {
2871 struct btrfs_block_group_cache *cache;
2872
2873 cache = btrfs_lookup_block_group(fs_info, start);
2874 ASSERT(cache);
2875
2876 spin_lock(&cache->lock);
2877 cache->delalloc_bytes -= len;
2878 spin_unlock(&cache->lock);
2879
2880 btrfs_put_block_group(cache);
2881 }
2882
2883 /* as ordered data IO finishes, this gets called so we can finish
2884 * an ordered extent if the range of bytes in the file it covers are
2885 * fully written.
2886 */
2887 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2888 {
2889 struct inode *inode = ordered_extent->inode;
2890 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2891 struct btrfs_root *root = BTRFS_I(inode)->root;
2892 struct btrfs_trans_handle *trans = NULL;
2893 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2894 struct extent_state *cached_state = NULL;
2895 struct new_sa_defrag_extent *new = NULL;
2896 int compress_type = 0;
2897 int ret = 0;
2898 u64 logical_len = ordered_extent->len;
2899 bool nolock;
2900 bool truncated = false;
2901 bool range_locked = false;
2902 bool clear_new_delalloc_bytes = false;
2903
2904 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2905 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2906 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2907 clear_new_delalloc_bytes = true;
2908
2909 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2910
2911 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2912 ret = -EIO;
2913 goto out;
2914 }
2915
2916 btrfs_free_io_failure_record(BTRFS_I(inode),
2917 ordered_extent->file_offset,
2918 ordered_extent->file_offset +
2919 ordered_extent->len - 1);
2920
2921 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2922 truncated = true;
2923 logical_len = ordered_extent->truncated_len;
2924 /* Truncated the entire extent, don't bother adding */
2925 if (!logical_len)
2926 goto out;
2927 }
2928
2929 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2930 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2931
2932 /*
2933 * For mwrite(mmap + memset to write) case, we still reserve
2934 * space for NOCOW range.
2935 * As NOCOW won't cause a new delayed ref, just free the space
2936 */
2937 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2938 ordered_extent->len);
2939 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2940 if (nolock)
2941 trans = btrfs_join_transaction_nolock(root);
2942 else
2943 trans = btrfs_join_transaction(root);
2944 if (IS_ERR(trans)) {
2945 ret = PTR_ERR(trans);
2946 trans = NULL;
2947 goto out;
2948 }
2949 trans->block_rsv = &fs_info->delalloc_block_rsv;
2950 ret = btrfs_update_inode_fallback(trans, root, inode);
2951 if (ret) /* -ENOMEM or corruption */
2952 btrfs_abort_transaction(trans, ret);
2953 goto out;
2954 }
2955
2956 range_locked = true;
2957 lock_extent_bits(io_tree, ordered_extent->file_offset,
2958 ordered_extent->file_offset + ordered_extent->len - 1,
2959 &cached_state);
2960
2961 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2962 ordered_extent->file_offset + ordered_extent->len - 1,
2963 EXTENT_DEFRAG, 0, cached_state);
2964 if (ret) {
2965 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2966 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2967 /* the inode is shared */
2968 new = record_old_file_extents(inode, ordered_extent);
2969
2970 clear_extent_bit(io_tree, ordered_extent->file_offset,
2971 ordered_extent->file_offset + ordered_extent->len - 1,
2972 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2973 }
2974
2975 if (nolock)
2976 trans = btrfs_join_transaction_nolock(root);
2977 else
2978 trans = btrfs_join_transaction(root);
2979 if (IS_ERR(trans)) {
2980 ret = PTR_ERR(trans);
2981 trans = NULL;
2982 goto out;
2983 }
2984
2985 trans->block_rsv = &fs_info->delalloc_block_rsv;
2986
2987 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2988 compress_type = ordered_extent->compress_type;
2989 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2990 BUG_ON(compress_type);
2991 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
2992 ordered_extent->file_offset,
2993 ordered_extent->file_offset +
2994 logical_len);
2995 } else {
2996 BUG_ON(root == fs_info->tree_root);
2997 ret = insert_reserved_file_extent(trans, inode,
2998 ordered_extent->file_offset,
2999 ordered_extent->start,
3000 ordered_extent->disk_len,
3001 logical_len, logical_len,
3002 compress_type, 0, 0,
3003 BTRFS_FILE_EXTENT_REG);
3004 if (!ret)
3005 btrfs_release_delalloc_bytes(fs_info,
3006 ordered_extent->start,
3007 ordered_extent->disk_len);
3008 }
3009 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3010 ordered_extent->file_offset, ordered_extent->len,
3011 trans->transid);
3012 if (ret < 0) {
3013 btrfs_abort_transaction(trans, ret);
3014 goto out;
3015 }
3016
3017 add_pending_csums(trans, inode, &ordered_extent->list);
3018
3019 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3020 ret = btrfs_update_inode_fallback(trans, root, inode);
3021 if (ret) { /* -ENOMEM or corruption */
3022 btrfs_abort_transaction(trans, ret);
3023 goto out;
3024 }
3025 ret = 0;
3026 out:
3027 if (range_locked || clear_new_delalloc_bytes) {
3028 unsigned int clear_bits = 0;
3029
3030 if (range_locked)
3031 clear_bits |= EXTENT_LOCKED;
3032 if (clear_new_delalloc_bytes)
3033 clear_bits |= EXTENT_DELALLOC_NEW;
3034 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3035 ordered_extent->file_offset,
3036 ordered_extent->file_offset +
3037 ordered_extent->len - 1,
3038 clear_bits,
3039 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3040 0, &cached_state, GFP_NOFS);
3041 }
3042
3043 if (root != fs_info->tree_root)
3044 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3045 ordered_extent->len);
3046 if (trans)
3047 btrfs_end_transaction(trans);
3048
3049 if (ret || truncated) {
3050 u64 start, end;
3051
3052 if (truncated)
3053 start = ordered_extent->file_offset + logical_len;
3054 else
3055 start = ordered_extent->file_offset;
3056 end = ordered_extent->file_offset + ordered_extent->len - 1;
3057 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
3058
3059 /* Drop the cache for the part of the extent we didn't write. */
3060 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3061
3062 /*
3063 * If the ordered extent had an IOERR or something else went
3064 * wrong we need to return the space for this ordered extent
3065 * back to the allocator. We only free the extent in the
3066 * truncated case if we didn't write out the extent at all.
3067 */
3068 if ((ret || !logical_len) &&
3069 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3070 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3071 btrfs_free_reserved_extent(fs_info,
3072 ordered_extent->start,
3073 ordered_extent->disk_len, 1);
3074 }
3075
3076
3077 /*
3078 * This needs to be done to make sure anybody waiting knows we are done
3079 * updating everything for this ordered extent.
3080 */
3081 btrfs_remove_ordered_extent(inode, ordered_extent);
3082
3083 /* for snapshot-aware defrag */
3084 if (new) {
3085 if (ret) {
3086 free_sa_defrag_extent(new);
3087 atomic_dec(&fs_info->defrag_running);
3088 } else {
3089 relink_file_extents(new);
3090 }
3091 }
3092
3093 /* once for us */
3094 btrfs_put_ordered_extent(ordered_extent);
3095 /* once for the tree */
3096 btrfs_put_ordered_extent(ordered_extent);
3097
3098 return ret;
3099 }
3100
3101 static void finish_ordered_fn(struct btrfs_work *work)
3102 {
3103 struct btrfs_ordered_extent *ordered_extent;
3104 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3105 btrfs_finish_ordered_io(ordered_extent);
3106 }
3107
3108 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3109 struct extent_state *state, int uptodate)
3110 {
3111 struct inode *inode = page->mapping->host;
3112 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3113 struct btrfs_ordered_extent *ordered_extent = NULL;
3114 struct btrfs_workqueue *wq;
3115 btrfs_work_func_t func;
3116
3117 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3118
3119 ClearPagePrivate2(page);
3120 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3121 end - start + 1, uptodate))
3122 return;
3123
3124 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3125 wq = fs_info->endio_freespace_worker;
3126 func = btrfs_freespace_write_helper;
3127 } else {
3128 wq = fs_info->endio_write_workers;
3129 func = btrfs_endio_write_helper;
3130 }
3131
3132 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3133 NULL);
3134 btrfs_queue_work(wq, &ordered_extent->work);
3135 }
3136
3137 static int __readpage_endio_check(struct inode *inode,
3138 struct btrfs_io_bio *io_bio,
3139 int icsum, struct page *page,
3140 int pgoff, u64 start, size_t len)
3141 {
3142 char *kaddr;
3143 u32 csum_expected;
3144 u32 csum = ~(u32)0;
3145
3146 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3147
3148 kaddr = kmap_atomic(page);
3149 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3150 btrfs_csum_final(csum, (u8 *)&csum);
3151 if (csum != csum_expected)
3152 goto zeroit;
3153
3154 kunmap_atomic(kaddr);
3155 return 0;
3156 zeroit:
3157 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3158 io_bio->mirror_num);
3159 memset(kaddr + pgoff, 1, len);
3160 flush_dcache_page(page);
3161 kunmap_atomic(kaddr);
3162 return -EIO;
3163 }
3164
3165 /*
3166 * when reads are done, we need to check csums to verify the data is correct
3167 * if there's a match, we allow the bio to finish. If not, the code in
3168 * extent_io.c will try to find good copies for us.
3169 */
3170 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3171 u64 phy_offset, struct page *page,
3172 u64 start, u64 end, int mirror)
3173 {
3174 size_t offset = start - page_offset(page);
3175 struct inode *inode = page->mapping->host;
3176 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3177 struct btrfs_root *root = BTRFS_I(inode)->root;
3178
3179 if (PageChecked(page)) {
3180 ClearPageChecked(page);
3181 return 0;
3182 }
3183
3184 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3185 return 0;
3186
3187 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3188 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3189 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3190 return 0;
3191 }
3192
3193 phy_offset >>= inode->i_sb->s_blocksize_bits;
3194 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3195 start, (size_t)(end - start + 1));
3196 }
3197
3198 void btrfs_add_delayed_iput(struct inode *inode)
3199 {
3200 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3201 struct btrfs_inode *binode = BTRFS_I(inode);
3202
3203 if (atomic_add_unless(&inode->i_count, -1, 1))
3204 return;
3205
3206 spin_lock(&fs_info->delayed_iput_lock);
3207 if (binode->delayed_iput_count == 0) {
3208 ASSERT(list_empty(&binode->delayed_iput));
3209 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3210 } else {
3211 binode->delayed_iput_count++;
3212 }
3213 spin_unlock(&fs_info->delayed_iput_lock);
3214 }
3215
3216 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3217 {
3218
3219 spin_lock(&fs_info->delayed_iput_lock);
3220 while (!list_empty(&fs_info->delayed_iputs)) {
3221 struct btrfs_inode *inode;
3222
3223 inode = list_first_entry(&fs_info->delayed_iputs,
3224 struct btrfs_inode, delayed_iput);
3225 if (inode->delayed_iput_count) {
3226 inode->delayed_iput_count--;
3227 list_move_tail(&inode->delayed_iput,
3228 &fs_info->delayed_iputs);
3229 } else {
3230 list_del_init(&inode->delayed_iput);
3231 }
3232 spin_unlock(&fs_info->delayed_iput_lock);
3233 iput(&inode->vfs_inode);
3234 spin_lock(&fs_info->delayed_iput_lock);
3235 }
3236 spin_unlock(&fs_info->delayed_iput_lock);
3237 }
3238
3239 /*
3240 * This is called in transaction commit time. If there are no orphan
3241 * files in the subvolume, it removes orphan item and frees block_rsv
3242 * structure.
3243 */
3244 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3245 struct btrfs_root *root)
3246 {
3247 struct btrfs_fs_info *fs_info = root->fs_info;
3248 struct btrfs_block_rsv *block_rsv;
3249 int ret;
3250
3251 if (atomic_read(&root->orphan_inodes) ||
3252 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3253 return;
3254
3255 spin_lock(&root->orphan_lock);
3256 if (atomic_read(&root->orphan_inodes)) {
3257 spin_unlock(&root->orphan_lock);
3258 return;
3259 }
3260
3261 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3262 spin_unlock(&root->orphan_lock);
3263 return;
3264 }
3265
3266 block_rsv = root->orphan_block_rsv;
3267 root->orphan_block_rsv = NULL;
3268 spin_unlock(&root->orphan_lock);
3269
3270 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3271 btrfs_root_refs(&root->root_item) > 0) {
3272 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3273 root->root_key.objectid);
3274 if (ret)
3275 btrfs_abort_transaction(trans, ret);
3276 else
3277 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3278 &root->state);
3279 }
3280
3281 if (block_rsv) {
3282 WARN_ON(block_rsv->size > 0);
3283 btrfs_free_block_rsv(fs_info, block_rsv);
3284 }
3285 }
3286
3287 /*
3288 * This creates an orphan entry for the given inode in case something goes
3289 * wrong in the middle of an unlink/truncate.
3290 *
3291 * NOTE: caller of this function should reserve 5 units of metadata for
3292 * this function.
3293 */
3294 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3295 struct btrfs_inode *inode)
3296 {
3297 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3298 struct btrfs_root *root = inode->root;
3299 struct btrfs_block_rsv *block_rsv = NULL;
3300 int reserve = 0;
3301 int insert = 0;
3302 int ret;
3303
3304 if (!root->orphan_block_rsv) {
3305 block_rsv = btrfs_alloc_block_rsv(fs_info,
3306 BTRFS_BLOCK_RSV_TEMP);
3307 if (!block_rsv)
3308 return -ENOMEM;
3309 }
3310
3311 spin_lock(&root->orphan_lock);
3312 if (!root->orphan_block_rsv) {
3313 root->orphan_block_rsv = block_rsv;
3314 } else if (block_rsv) {
3315 btrfs_free_block_rsv(fs_info, block_rsv);
3316 block_rsv = NULL;
3317 }
3318
3319 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3320 &inode->runtime_flags)) {
3321 #if 0
3322 /*
3323 * For proper ENOSPC handling, we should do orphan
3324 * cleanup when mounting. But this introduces backward
3325 * compatibility issue.
3326 */
3327 if (!xchg(&root->orphan_item_inserted, 1))
3328 insert = 2;
3329 else
3330 insert = 1;
3331 #endif
3332 insert = 1;
3333 atomic_inc(&root->orphan_inodes);
3334 }
3335
3336 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3337 &inode->runtime_flags))
3338 reserve = 1;
3339 spin_unlock(&root->orphan_lock);
3340
3341 /* grab metadata reservation from transaction handle */
3342 if (reserve) {
3343 ret = btrfs_orphan_reserve_metadata(trans, inode);
3344 ASSERT(!ret);
3345 if (ret) {
3346 atomic_dec(&root->orphan_inodes);
3347 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3348 &inode->runtime_flags);
3349 if (insert)
3350 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3351 &inode->runtime_flags);
3352 return ret;
3353 }
3354 }
3355
3356 /* insert an orphan item to track this unlinked/truncated file */
3357 if (insert >= 1) {
3358 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3359 if (ret) {
3360 atomic_dec(&root->orphan_inodes);
3361 if (reserve) {
3362 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3363 &inode->runtime_flags);
3364 btrfs_orphan_release_metadata(inode);
3365 }
3366 if (ret != -EEXIST) {
3367 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3368 &inode->runtime_flags);
3369 btrfs_abort_transaction(trans, ret);
3370 return ret;
3371 }
3372 }
3373 ret = 0;
3374 }
3375
3376 /* insert an orphan item to track subvolume contains orphan files */
3377 if (insert >= 2) {
3378 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3379 root->root_key.objectid);
3380 if (ret && ret != -EEXIST) {
3381 btrfs_abort_transaction(trans, ret);
3382 return ret;
3383 }
3384 }
3385 return 0;
3386 }
3387
3388 /*
3389 * We have done the truncate/delete so we can go ahead and remove the orphan
3390 * item for this particular inode.
3391 */
3392 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3393 struct btrfs_inode *inode)
3394 {
3395 struct btrfs_root *root = inode->root;
3396 int delete_item = 0;
3397 int release_rsv = 0;
3398 int ret = 0;
3399
3400 spin_lock(&root->orphan_lock);
3401 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3402 &inode->runtime_flags))
3403 delete_item = 1;
3404
3405 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3406 &inode->runtime_flags))
3407 release_rsv = 1;
3408 spin_unlock(&root->orphan_lock);
3409
3410 if (delete_item) {
3411 atomic_dec(&root->orphan_inodes);
3412 if (trans)
3413 ret = btrfs_del_orphan_item(trans, root,
3414 btrfs_ino(inode));
3415 }
3416
3417 if (release_rsv)
3418 btrfs_orphan_release_metadata(inode);
3419
3420 return ret;
3421 }
3422
3423 /*
3424 * this cleans up any orphans that may be left on the list from the last use
3425 * of this root.
3426 */
3427 int btrfs_orphan_cleanup(struct btrfs_root *root)
3428 {
3429 struct btrfs_fs_info *fs_info = root->fs_info;
3430 struct btrfs_path *path;
3431 struct extent_buffer *leaf;
3432 struct btrfs_key key, found_key;
3433 struct btrfs_trans_handle *trans;
3434 struct inode *inode;
3435 u64 last_objectid = 0;
3436 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3437
3438 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3439 return 0;
3440
3441 path = btrfs_alloc_path();
3442 if (!path) {
3443 ret = -ENOMEM;
3444 goto out;
3445 }
3446 path->reada = READA_BACK;
3447
3448 key.objectid = BTRFS_ORPHAN_OBJECTID;
3449 key.type = BTRFS_ORPHAN_ITEM_KEY;
3450 key.offset = (u64)-1;
3451
3452 while (1) {
3453 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3454 if (ret < 0)
3455 goto out;
3456
3457 /*
3458 * if ret == 0 means we found what we were searching for, which
3459 * is weird, but possible, so only screw with path if we didn't
3460 * find the key and see if we have stuff that matches
3461 */
3462 if (ret > 0) {
3463 ret = 0;
3464 if (path->slots[0] == 0)
3465 break;
3466 path->slots[0]--;
3467 }
3468
3469 /* pull out the item */
3470 leaf = path->nodes[0];
3471 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3472
3473 /* make sure the item matches what we want */
3474 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3475 break;
3476 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3477 break;
3478
3479 /* release the path since we're done with it */
3480 btrfs_release_path(path);
3481
3482 /*
3483 * this is where we are basically btrfs_lookup, without the
3484 * crossing root thing. we store the inode number in the
3485 * offset of the orphan item.
3486 */
3487
3488 if (found_key.offset == last_objectid) {
3489 btrfs_err(fs_info,
3490 "Error removing orphan entry, stopping orphan cleanup");
3491 ret = -EINVAL;
3492 goto out;
3493 }
3494
3495 last_objectid = found_key.offset;
3496
3497 found_key.objectid = found_key.offset;
3498 found_key.type = BTRFS_INODE_ITEM_KEY;
3499 found_key.offset = 0;
3500 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3501 ret = PTR_ERR_OR_ZERO(inode);
3502 if (ret && ret != -ENOENT)
3503 goto out;
3504
3505 if (ret == -ENOENT && root == fs_info->tree_root) {
3506 struct btrfs_root *dead_root;
3507 struct btrfs_fs_info *fs_info = root->fs_info;
3508 int is_dead_root = 0;
3509
3510 /*
3511 * this is an orphan in the tree root. Currently these
3512 * could come from 2 sources:
3513 * a) a snapshot deletion in progress
3514 * b) a free space cache inode
3515 * We need to distinguish those two, as the snapshot
3516 * orphan must not get deleted.
3517 * find_dead_roots already ran before us, so if this
3518 * is a snapshot deletion, we should find the root
3519 * in the dead_roots list
3520 */
3521 spin_lock(&fs_info->trans_lock);
3522 list_for_each_entry(dead_root, &fs_info->dead_roots,
3523 root_list) {
3524 if (dead_root->root_key.objectid ==
3525 found_key.objectid) {
3526 is_dead_root = 1;
3527 break;
3528 }
3529 }
3530 spin_unlock(&fs_info->trans_lock);
3531 if (is_dead_root) {
3532 /* prevent this orphan from being found again */
3533 key.offset = found_key.objectid - 1;
3534 continue;
3535 }
3536 }
3537 /*
3538 * Inode is already gone but the orphan item is still there,
3539 * kill the orphan item.
3540 */
3541 if (ret == -ENOENT) {
3542 trans = btrfs_start_transaction(root, 1);
3543 if (IS_ERR(trans)) {
3544 ret = PTR_ERR(trans);
3545 goto out;
3546 }
3547 btrfs_debug(fs_info, "auto deleting %Lu",
3548 found_key.objectid);
3549 ret = btrfs_del_orphan_item(trans, root,
3550 found_key.objectid);
3551 btrfs_end_transaction(trans);
3552 if (ret)
3553 goto out;
3554 continue;
3555 }
3556
3557 /*
3558 * add this inode to the orphan list so btrfs_orphan_del does
3559 * the proper thing when we hit it
3560 */
3561 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3562 &BTRFS_I(inode)->runtime_flags);
3563 atomic_inc(&root->orphan_inodes);
3564
3565 /* if we have links, this was a truncate, lets do that */
3566 if (inode->i_nlink) {
3567 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3568 iput(inode);
3569 continue;
3570 }
3571 nr_truncate++;
3572
3573 /* 1 for the orphan item deletion. */
3574 trans = btrfs_start_transaction(root, 1);
3575 if (IS_ERR(trans)) {
3576 iput(inode);
3577 ret = PTR_ERR(trans);
3578 goto out;
3579 }
3580 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3581 btrfs_end_transaction(trans);
3582 if (ret) {
3583 iput(inode);
3584 goto out;
3585 }
3586
3587 ret = btrfs_truncate(inode);
3588 if (ret)
3589 btrfs_orphan_del(NULL, BTRFS_I(inode));
3590 } else {
3591 nr_unlink++;
3592 }
3593
3594 /* this will do delete_inode and everything for us */
3595 iput(inode);
3596 if (ret)
3597 goto out;
3598 }
3599 /* release the path since we're done with it */
3600 btrfs_release_path(path);
3601
3602 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3603
3604 if (root->orphan_block_rsv)
3605 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3606 (u64)-1);
3607
3608 if (root->orphan_block_rsv ||
3609 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3610 trans = btrfs_join_transaction(root);
3611 if (!IS_ERR(trans))
3612 btrfs_end_transaction(trans);
3613 }
3614
3615 if (nr_unlink)
3616 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3617 if (nr_truncate)
3618 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3619
3620 out:
3621 if (ret)
3622 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3623 btrfs_free_path(path);
3624 return ret;
3625 }
3626
3627 /*
3628 * very simple check to peek ahead in the leaf looking for xattrs. If we
3629 * don't find any xattrs, we know there can't be any acls.
3630 *
3631 * slot is the slot the inode is in, objectid is the objectid of the inode
3632 */
3633 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3634 int slot, u64 objectid,
3635 int *first_xattr_slot)
3636 {
3637 u32 nritems = btrfs_header_nritems(leaf);
3638 struct btrfs_key found_key;
3639 static u64 xattr_access = 0;
3640 static u64 xattr_default = 0;
3641 int scanned = 0;
3642
3643 if (!xattr_access) {
3644 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3645 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3646 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3647 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3648 }
3649
3650 slot++;
3651 *first_xattr_slot = -1;
3652 while (slot < nritems) {
3653 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3654
3655 /* we found a different objectid, there must not be acls */
3656 if (found_key.objectid != objectid)
3657 return 0;
3658
3659 /* we found an xattr, assume we've got an acl */
3660 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3661 if (*first_xattr_slot == -1)
3662 *first_xattr_slot = slot;
3663 if (found_key.offset == xattr_access ||
3664 found_key.offset == xattr_default)
3665 return 1;
3666 }
3667
3668 /*
3669 * we found a key greater than an xattr key, there can't
3670 * be any acls later on
3671 */
3672 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3673 return 0;
3674
3675 slot++;
3676 scanned++;
3677
3678 /*
3679 * it goes inode, inode backrefs, xattrs, extents,
3680 * so if there are a ton of hard links to an inode there can
3681 * be a lot of backrefs. Don't waste time searching too hard,
3682 * this is just an optimization
3683 */
3684 if (scanned >= 8)
3685 break;
3686 }
3687 /* we hit the end of the leaf before we found an xattr or
3688 * something larger than an xattr. We have to assume the inode
3689 * has acls
3690 */
3691 if (*first_xattr_slot == -1)
3692 *first_xattr_slot = slot;
3693 return 1;
3694 }
3695
3696 /*
3697 * read an inode from the btree into the in-memory inode
3698 */
3699 static int btrfs_read_locked_inode(struct inode *inode)
3700 {
3701 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3702 struct btrfs_path *path;
3703 struct extent_buffer *leaf;
3704 struct btrfs_inode_item *inode_item;
3705 struct btrfs_root *root = BTRFS_I(inode)->root;
3706 struct btrfs_key location;
3707 unsigned long ptr;
3708 int maybe_acls;
3709 u32 rdev;
3710 int ret;
3711 bool filled = false;
3712 int first_xattr_slot;
3713
3714 ret = btrfs_fill_inode(inode, &rdev);
3715 if (!ret)
3716 filled = true;
3717
3718 path = btrfs_alloc_path();
3719 if (!path) {
3720 ret = -ENOMEM;
3721 goto make_bad;
3722 }
3723
3724 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3725
3726 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3727 if (ret) {
3728 if (ret > 0)
3729 ret = -ENOENT;
3730 goto make_bad;
3731 }
3732
3733 leaf = path->nodes[0];
3734
3735 if (filled)
3736 goto cache_index;
3737
3738 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3739 struct btrfs_inode_item);
3740 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3741 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3742 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3743 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3744 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3745
3746 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3747 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3748
3749 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3750 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3751
3752 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3753 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3754
3755 BTRFS_I(inode)->i_otime.tv_sec =
3756 btrfs_timespec_sec(leaf, &inode_item->otime);
3757 BTRFS_I(inode)->i_otime.tv_nsec =
3758 btrfs_timespec_nsec(leaf, &inode_item->otime);
3759
3760 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3761 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3762 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3763
3764 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3765 inode->i_generation = BTRFS_I(inode)->generation;
3766 inode->i_rdev = 0;
3767 rdev = btrfs_inode_rdev(leaf, inode_item);
3768
3769 BTRFS_I(inode)->index_cnt = (u64)-1;
3770 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3771
3772 cache_index:
3773 /*
3774 * If we were modified in the current generation and evicted from memory
3775 * and then re-read we need to do a full sync since we don't have any
3776 * idea about which extents were modified before we were evicted from
3777 * cache.
3778 *
3779 * This is required for both inode re-read from disk and delayed inode
3780 * in delayed_nodes_tree.
3781 */
3782 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3783 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3784 &BTRFS_I(inode)->runtime_flags);
3785
3786 /*
3787 * We don't persist the id of the transaction where an unlink operation
3788 * against the inode was last made. So here we assume the inode might
3789 * have been evicted, and therefore the exact value of last_unlink_trans
3790 * lost, and set it to last_trans to avoid metadata inconsistencies
3791 * between the inode and its parent if the inode is fsync'ed and the log
3792 * replayed. For example, in the scenario:
3793 *
3794 * touch mydir/foo
3795 * ln mydir/foo mydir/bar
3796 * sync
3797 * unlink mydir/bar
3798 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3799 * xfs_io -c fsync mydir/foo
3800 * <power failure>
3801 * mount fs, triggers fsync log replay
3802 *
3803 * We must make sure that when we fsync our inode foo we also log its
3804 * parent inode, otherwise after log replay the parent still has the
3805 * dentry with the "bar" name but our inode foo has a link count of 1
3806 * and doesn't have an inode ref with the name "bar" anymore.
3807 *
3808 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3809 * but it guarantees correctness at the expense of occasional full
3810 * transaction commits on fsync if our inode is a directory, or if our
3811 * inode is not a directory, logging its parent unnecessarily.
3812 */
3813 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3814
3815 path->slots[0]++;
3816 if (inode->i_nlink != 1 ||
3817 path->slots[0] >= btrfs_header_nritems(leaf))
3818 goto cache_acl;
3819
3820 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3821 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3822 goto cache_acl;
3823
3824 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3825 if (location.type == BTRFS_INODE_REF_KEY) {
3826 struct btrfs_inode_ref *ref;
3827
3828 ref = (struct btrfs_inode_ref *)ptr;
3829 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3830 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3831 struct btrfs_inode_extref *extref;
3832
3833 extref = (struct btrfs_inode_extref *)ptr;
3834 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3835 extref);
3836 }
3837 cache_acl:
3838 /*
3839 * try to precache a NULL acl entry for files that don't have
3840 * any xattrs or acls
3841 */
3842 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3843 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3844 if (first_xattr_slot != -1) {
3845 path->slots[0] = first_xattr_slot;
3846 ret = btrfs_load_inode_props(inode, path);
3847 if (ret)
3848 btrfs_err(fs_info,
3849 "error loading props for ino %llu (root %llu): %d",
3850 btrfs_ino(BTRFS_I(inode)),
3851 root->root_key.objectid, ret);
3852 }
3853 btrfs_free_path(path);
3854
3855 if (!maybe_acls)
3856 cache_no_acl(inode);
3857
3858 switch (inode->i_mode & S_IFMT) {
3859 case S_IFREG:
3860 inode->i_mapping->a_ops = &btrfs_aops;
3861 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3862 inode->i_fop = &btrfs_file_operations;
3863 inode->i_op = &btrfs_file_inode_operations;
3864 break;
3865 case S_IFDIR:
3866 inode->i_fop = &btrfs_dir_file_operations;
3867 inode->i_op = &btrfs_dir_inode_operations;
3868 break;
3869 case S_IFLNK:
3870 inode->i_op = &btrfs_symlink_inode_operations;
3871 inode_nohighmem(inode);
3872 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3873 break;
3874 default:
3875 inode->i_op = &btrfs_special_inode_operations;
3876 init_special_inode(inode, inode->i_mode, rdev);
3877 break;
3878 }
3879
3880 btrfs_update_iflags(inode);
3881 return 0;
3882
3883 make_bad:
3884 btrfs_free_path(path);
3885 make_bad_inode(inode);
3886 return ret;
3887 }
3888
3889 /*
3890 * given a leaf and an inode, copy the inode fields into the leaf
3891 */
3892 static void fill_inode_item(struct btrfs_trans_handle *trans,
3893 struct extent_buffer *leaf,
3894 struct btrfs_inode_item *item,
3895 struct inode *inode)
3896 {
3897 struct btrfs_map_token token;
3898
3899 btrfs_init_map_token(&token);
3900
3901 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3902 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3903 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3904 &token);
3905 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3906 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3907
3908 btrfs_set_token_timespec_sec(leaf, &item->atime,
3909 inode->i_atime.tv_sec, &token);
3910 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3911 inode->i_atime.tv_nsec, &token);
3912
3913 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3914 inode->i_mtime.tv_sec, &token);
3915 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3916 inode->i_mtime.tv_nsec, &token);
3917
3918 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3919 inode->i_ctime.tv_sec, &token);
3920 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3921 inode->i_ctime.tv_nsec, &token);
3922
3923 btrfs_set_token_timespec_sec(leaf, &item->otime,
3924 BTRFS_I(inode)->i_otime.tv_sec, &token);
3925 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3926 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3927
3928 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3929 &token);
3930 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3931 &token);
3932 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3933 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3934 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3935 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3936 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3937 }
3938
3939 /*
3940 * copy everything in the in-memory inode into the btree.
3941 */
3942 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3943 struct btrfs_root *root, struct inode *inode)
3944 {
3945 struct btrfs_inode_item *inode_item;
3946 struct btrfs_path *path;
3947 struct extent_buffer *leaf;
3948 int ret;
3949
3950 path = btrfs_alloc_path();
3951 if (!path)
3952 return -ENOMEM;
3953
3954 path->leave_spinning = 1;
3955 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3956 1);
3957 if (ret) {
3958 if (ret > 0)
3959 ret = -ENOENT;
3960 goto failed;
3961 }
3962
3963 leaf = path->nodes[0];
3964 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3965 struct btrfs_inode_item);
3966
3967 fill_inode_item(trans, leaf, inode_item, inode);
3968 btrfs_mark_buffer_dirty(leaf);
3969 btrfs_set_inode_last_trans(trans, inode);
3970 ret = 0;
3971 failed:
3972 btrfs_free_path(path);
3973 return ret;
3974 }
3975
3976 /*
3977 * copy everything in the in-memory inode into the btree.
3978 */
3979 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3980 struct btrfs_root *root, struct inode *inode)
3981 {
3982 struct btrfs_fs_info *fs_info = root->fs_info;
3983 int ret;
3984
3985 /*
3986 * If the inode is a free space inode, we can deadlock during commit
3987 * if we put it into the delayed code.
3988 *
3989 * The data relocation inode should also be directly updated
3990 * without delay
3991 */
3992 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3993 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3994 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3995 btrfs_update_root_times(trans, root);
3996
3997 ret = btrfs_delayed_update_inode(trans, root, inode);
3998 if (!ret)
3999 btrfs_set_inode_last_trans(trans, inode);
4000 return ret;
4001 }
4002
4003 return btrfs_update_inode_item(trans, root, inode);
4004 }
4005
4006 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4007 struct btrfs_root *root,
4008 struct inode *inode)
4009 {
4010 int ret;
4011
4012 ret = btrfs_update_inode(trans, root, inode);
4013 if (ret == -ENOSPC)
4014 return btrfs_update_inode_item(trans, root, inode);
4015 return ret;
4016 }
4017
4018 /*
4019 * unlink helper that gets used here in inode.c and in the tree logging
4020 * recovery code. It remove a link in a directory with a given name, and
4021 * also drops the back refs in the inode to the directory
4022 */
4023 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4024 struct btrfs_root *root,
4025 struct btrfs_inode *dir,
4026 struct btrfs_inode *inode,
4027 const char *name, int name_len)
4028 {
4029 struct btrfs_fs_info *fs_info = root->fs_info;
4030 struct btrfs_path *path;
4031 int ret = 0;
4032 struct extent_buffer *leaf;
4033 struct btrfs_dir_item *di;
4034 struct btrfs_key key;
4035 u64 index;
4036 u64 ino = btrfs_ino(inode);
4037 u64 dir_ino = btrfs_ino(dir);
4038
4039 path = btrfs_alloc_path();
4040 if (!path) {
4041 ret = -ENOMEM;
4042 goto out;
4043 }
4044
4045 path->leave_spinning = 1;
4046 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4047 name, name_len, -1);
4048 if (IS_ERR(di)) {
4049 ret = PTR_ERR(di);
4050 goto err;
4051 }
4052 if (!di) {
4053 ret = -ENOENT;
4054 goto err;
4055 }
4056 leaf = path->nodes[0];
4057 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4058 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4059 if (ret)
4060 goto err;
4061 btrfs_release_path(path);
4062
4063 /*
4064 * If we don't have dir index, we have to get it by looking up
4065 * the inode ref, since we get the inode ref, remove it directly,
4066 * it is unnecessary to do delayed deletion.
4067 *
4068 * But if we have dir index, needn't search inode ref to get it.
4069 * Since the inode ref is close to the inode item, it is better
4070 * that we delay to delete it, and just do this deletion when
4071 * we update the inode item.
4072 */
4073 if (inode->dir_index) {
4074 ret = btrfs_delayed_delete_inode_ref(inode);
4075 if (!ret) {
4076 index = inode->dir_index;
4077 goto skip_backref;
4078 }
4079 }
4080
4081 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4082 dir_ino, &index);
4083 if (ret) {
4084 btrfs_info(fs_info,
4085 "failed to delete reference to %.*s, inode %llu parent %llu",
4086 name_len, name, ino, dir_ino);
4087 btrfs_abort_transaction(trans, ret);
4088 goto err;
4089 }
4090 skip_backref:
4091 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4092 if (ret) {
4093 btrfs_abort_transaction(trans, ret);
4094 goto err;
4095 }
4096
4097 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4098 dir_ino);
4099 if (ret != 0 && ret != -ENOENT) {
4100 btrfs_abort_transaction(trans, ret);
4101 goto err;
4102 }
4103
4104 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4105 index);
4106 if (ret == -ENOENT)
4107 ret = 0;
4108 else if (ret)
4109 btrfs_abort_transaction(trans, ret);
4110 err:
4111 btrfs_free_path(path);
4112 if (ret)
4113 goto out;
4114
4115 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4116 inode_inc_iversion(&inode->vfs_inode);
4117 inode_inc_iversion(&dir->vfs_inode);
4118 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4119 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4120 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4121 out:
4122 return ret;
4123 }
4124
4125 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4126 struct btrfs_root *root,
4127 struct btrfs_inode *dir, struct btrfs_inode *inode,
4128 const char *name, int name_len)
4129 {
4130 int ret;
4131 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4132 if (!ret) {
4133 drop_nlink(&inode->vfs_inode);
4134 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4135 }
4136 return ret;
4137 }
4138
4139 /*
4140 * helper to start transaction for unlink and rmdir.
4141 *
4142 * unlink and rmdir are special in btrfs, they do not always free space, so
4143 * if we cannot make our reservations the normal way try and see if there is
4144 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4145 * allow the unlink to occur.
4146 */
4147 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4148 {
4149 struct btrfs_root *root = BTRFS_I(dir)->root;
4150
4151 /*
4152 * 1 for the possible orphan item
4153 * 1 for the dir item
4154 * 1 for the dir index
4155 * 1 for the inode ref
4156 * 1 for the inode
4157 */
4158 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4159 }
4160
4161 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4162 {
4163 struct btrfs_root *root = BTRFS_I(dir)->root;
4164 struct btrfs_trans_handle *trans;
4165 struct inode *inode = d_inode(dentry);
4166 int ret;
4167
4168 trans = __unlink_start_trans(dir);
4169 if (IS_ERR(trans))
4170 return PTR_ERR(trans);
4171
4172 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4173 0);
4174
4175 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4176 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4177 dentry->d_name.len);
4178 if (ret)
4179 goto out;
4180
4181 if (inode->i_nlink == 0) {
4182 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4183 if (ret)
4184 goto out;
4185 }
4186
4187 out:
4188 btrfs_end_transaction(trans);
4189 btrfs_btree_balance_dirty(root->fs_info);
4190 return ret;
4191 }
4192
4193 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4194 struct btrfs_root *root,
4195 struct inode *dir, u64 objectid,
4196 const char *name, int name_len)
4197 {
4198 struct btrfs_fs_info *fs_info = root->fs_info;
4199 struct btrfs_path *path;
4200 struct extent_buffer *leaf;
4201 struct btrfs_dir_item *di;
4202 struct btrfs_key key;
4203 u64 index;
4204 int ret;
4205 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4206
4207 path = btrfs_alloc_path();
4208 if (!path)
4209 return -ENOMEM;
4210
4211 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4212 name, name_len, -1);
4213 if (IS_ERR_OR_NULL(di)) {
4214 if (!di)
4215 ret = -ENOENT;
4216 else
4217 ret = PTR_ERR(di);
4218 goto out;
4219 }
4220
4221 leaf = path->nodes[0];
4222 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4223 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4224 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4225 if (ret) {
4226 btrfs_abort_transaction(trans, ret);
4227 goto out;
4228 }
4229 btrfs_release_path(path);
4230
4231 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4232 root->root_key.objectid, dir_ino,
4233 &index, name, name_len);
4234 if (ret < 0) {
4235 if (ret != -ENOENT) {
4236 btrfs_abort_transaction(trans, ret);
4237 goto out;
4238 }
4239 di = btrfs_search_dir_index_item(root, path, dir_ino,
4240 name, name_len);
4241 if (IS_ERR_OR_NULL(di)) {
4242 if (!di)
4243 ret = -ENOENT;
4244 else
4245 ret = PTR_ERR(di);
4246 btrfs_abort_transaction(trans, ret);
4247 goto out;
4248 }
4249
4250 leaf = path->nodes[0];
4251 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4252 btrfs_release_path(path);
4253 index = key.offset;
4254 }
4255 btrfs_release_path(path);
4256
4257 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4258 if (ret) {
4259 btrfs_abort_transaction(trans, ret);
4260 goto out;
4261 }
4262
4263 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4264 inode_inc_iversion(dir);
4265 dir->i_mtime = dir->i_ctime = current_time(dir);
4266 ret = btrfs_update_inode_fallback(trans, root, dir);
4267 if (ret)
4268 btrfs_abort_transaction(trans, ret);
4269 out:
4270 btrfs_free_path(path);
4271 return ret;
4272 }
4273
4274 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4275 {
4276 struct inode *inode = d_inode(dentry);
4277 int err = 0;
4278 struct btrfs_root *root = BTRFS_I(dir)->root;
4279 struct btrfs_trans_handle *trans;
4280 u64 last_unlink_trans;
4281
4282 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4283 return -ENOTEMPTY;
4284 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4285 return -EPERM;
4286
4287 trans = __unlink_start_trans(dir);
4288 if (IS_ERR(trans))
4289 return PTR_ERR(trans);
4290
4291 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4292 err = btrfs_unlink_subvol(trans, root, dir,
4293 BTRFS_I(inode)->location.objectid,
4294 dentry->d_name.name,
4295 dentry->d_name.len);
4296 goto out;
4297 }
4298
4299 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4300 if (err)
4301 goto out;
4302
4303 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4304
4305 /* now the directory is empty */
4306 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4307 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4308 dentry->d_name.len);
4309 if (!err) {
4310 btrfs_i_size_write(BTRFS_I(inode), 0);
4311 /*
4312 * Propagate the last_unlink_trans value of the deleted dir to
4313 * its parent directory. This is to prevent an unrecoverable
4314 * log tree in the case we do something like this:
4315 * 1) create dir foo
4316 * 2) create snapshot under dir foo
4317 * 3) delete the snapshot
4318 * 4) rmdir foo
4319 * 5) mkdir foo
4320 * 6) fsync foo or some file inside foo
4321 */
4322 if (last_unlink_trans >= trans->transid)
4323 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4324 }
4325 out:
4326 btrfs_end_transaction(trans);
4327 btrfs_btree_balance_dirty(root->fs_info);
4328
4329 return err;
4330 }
4331
4332 static int truncate_space_check(struct btrfs_trans_handle *trans,
4333 struct btrfs_root *root,
4334 u64 bytes_deleted)
4335 {
4336 struct btrfs_fs_info *fs_info = root->fs_info;
4337 int ret;
4338
4339 /*
4340 * This is only used to apply pressure to the enospc system, we don't
4341 * intend to use this reservation at all.
4342 */
4343 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4344 bytes_deleted *= fs_info->nodesize;
4345 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4346 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4347 if (!ret) {
4348 trace_btrfs_space_reservation(fs_info, "transaction",
4349 trans->transid,
4350 bytes_deleted, 1);
4351 trans->bytes_reserved += bytes_deleted;
4352 }
4353 return ret;
4354
4355 }
4356
4357 static int truncate_inline_extent(struct inode *inode,
4358 struct btrfs_path *path,
4359 struct btrfs_key *found_key,
4360 const u64 item_end,
4361 const u64 new_size)
4362 {
4363 struct extent_buffer *leaf = path->nodes[0];
4364 int slot = path->slots[0];
4365 struct btrfs_file_extent_item *fi;
4366 u32 size = (u32)(new_size - found_key->offset);
4367 struct btrfs_root *root = BTRFS_I(inode)->root;
4368
4369 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4370
4371 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4372 loff_t offset = new_size;
4373 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4374
4375 /*
4376 * Zero out the remaining of the last page of our inline extent,
4377 * instead of directly truncating our inline extent here - that
4378 * would be much more complex (decompressing all the data, then
4379 * compressing the truncated data, which might be bigger than
4380 * the size of the inline extent, resize the extent, etc).
4381 * We release the path because to get the page we might need to
4382 * read the extent item from disk (data not in the page cache).
4383 */
4384 btrfs_release_path(path);
4385 return btrfs_truncate_block(inode, offset, page_end - offset,
4386 0);
4387 }
4388
4389 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4390 size = btrfs_file_extent_calc_inline_size(size);
4391 btrfs_truncate_item(root->fs_info, path, size, 1);
4392
4393 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4394 inode_sub_bytes(inode, item_end + 1 - new_size);
4395
4396 return 0;
4397 }
4398
4399 /*
4400 * this can truncate away extent items, csum items and directory items.
4401 * It starts at a high offset and removes keys until it can't find
4402 * any higher than new_size
4403 *
4404 * csum items that cross the new i_size are truncated to the new size
4405 * as well.
4406 *
4407 * min_type is the minimum key type to truncate down to. If set to 0, this
4408 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4409 */
4410 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4411 struct btrfs_root *root,
4412 struct inode *inode,
4413 u64 new_size, u32 min_type)
4414 {
4415 struct btrfs_fs_info *fs_info = root->fs_info;
4416 struct btrfs_path *path;
4417 struct extent_buffer *leaf;
4418 struct btrfs_file_extent_item *fi;
4419 struct btrfs_key key;
4420 struct btrfs_key found_key;
4421 u64 extent_start = 0;
4422 u64 extent_num_bytes = 0;
4423 u64 extent_offset = 0;
4424 u64 item_end = 0;
4425 u64 last_size = new_size;
4426 u32 found_type = (u8)-1;
4427 int found_extent;
4428 int del_item;
4429 int pending_del_nr = 0;
4430 int pending_del_slot = 0;
4431 int extent_type = -1;
4432 int ret;
4433 int err = 0;
4434 u64 ino = btrfs_ino(BTRFS_I(inode));
4435 u64 bytes_deleted = 0;
4436 bool be_nice = 0;
4437 bool should_throttle = 0;
4438 bool should_end = 0;
4439
4440 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4441
4442 /*
4443 * for non-free space inodes and ref cows, we want to back off from
4444 * time to time
4445 */
4446 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4447 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4448 be_nice = 1;
4449
4450 path = btrfs_alloc_path();
4451 if (!path)
4452 return -ENOMEM;
4453 path->reada = READA_BACK;
4454
4455 /*
4456 * We want to drop from the next block forward in case this new size is
4457 * not block aligned since we will be keeping the last block of the
4458 * extent just the way it is.
4459 */
4460 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4461 root == fs_info->tree_root)
4462 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4463 fs_info->sectorsize),
4464 (u64)-1, 0);
4465
4466 /*
4467 * This function is also used to drop the items in the log tree before
4468 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4469 * it is used to drop the loged items. So we shouldn't kill the delayed
4470 * items.
4471 */
4472 if (min_type == 0 && root == BTRFS_I(inode)->root)
4473 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4474
4475 key.objectid = ino;
4476 key.offset = (u64)-1;
4477 key.type = (u8)-1;
4478
4479 search_again:
4480 /*
4481 * with a 16K leaf size and 128MB extents, you can actually queue
4482 * up a huge file in a single leaf. Most of the time that
4483 * bytes_deleted is > 0, it will be huge by the time we get here
4484 */
4485 if (be_nice && bytes_deleted > SZ_32M) {
4486 if (btrfs_should_end_transaction(trans)) {
4487 err = -EAGAIN;
4488 goto error;
4489 }
4490 }
4491
4492
4493 path->leave_spinning = 1;
4494 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4495 if (ret < 0) {
4496 err = ret;
4497 goto out;
4498 }
4499
4500 if (ret > 0) {
4501 /* there are no items in the tree for us to truncate, we're
4502 * done
4503 */
4504 if (path->slots[0] == 0)
4505 goto out;
4506 path->slots[0]--;
4507 }
4508
4509 while (1) {
4510 fi = NULL;
4511 leaf = path->nodes[0];
4512 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4513 found_type = found_key.type;
4514
4515 if (found_key.objectid != ino)
4516 break;
4517
4518 if (found_type < min_type)
4519 break;
4520
4521 item_end = found_key.offset;
4522 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4523 fi = btrfs_item_ptr(leaf, path->slots[0],
4524 struct btrfs_file_extent_item);
4525 extent_type = btrfs_file_extent_type(leaf, fi);
4526 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4527 item_end +=
4528 btrfs_file_extent_num_bytes(leaf, fi);
4529
4530 trace_btrfs_truncate_show_fi_regular(
4531 BTRFS_I(inode), leaf, fi,
4532 found_key.offset);
4533 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4534 item_end += btrfs_file_extent_inline_len(leaf,
4535 path->slots[0], fi);
4536
4537 trace_btrfs_truncate_show_fi_inline(
4538 BTRFS_I(inode), leaf, fi, path->slots[0],
4539 found_key.offset);
4540 }
4541 item_end--;
4542 }
4543 if (found_type > min_type) {
4544 del_item = 1;
4545 } else {
4546 if (item_end < new_size)
4547 break;
4548 if (found_key.offset >= new_size)
4549 del_item = 1;
4550 else
4551 del_item = 0;
4552 }
4553 found_extent = 0;
4554 /* FIXME, shrink the extent if the ref count is only 1 */
4555 if (found_type != BTRFS_EXTENT_DATA_KEY)
4556 goto delete;
4557
4558 if (del_item)
4559 last_size = found_key.offset;
4560 else
4561 last_size = new_size;
4562
4563 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4564 u64 num_dec;
4565 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4566 if (!del_item) {
4567 u64 orig_num_bytes =
4568 btrfs_file_extent_num_bytes(leaf, fi);
4569 extent_num_bytes = ALIGN(new_size -
4570 found_key.offset,
4571 fs_info->sectorsize);
4572 btrfs_set_file_extent_num_bytes(leaf, fi,
4573 extent_num_bytes);
4574 num_dec = (orig_num_bytes -
4575 extent_num_bytes);
4576 if (test_bit(BTRFS_ROOT_REF_COWS,
4577 &root->state) &&
4578 extent_start != 0)
4579 inode_sub_bytes(inode, num_dec);
4580 btrfs_mark_buffer_dirty(leaf);
4581 } else {
4582 extent_num_bytes =
4583 btrfs_file_extent_disk_num_bytes(leaf,
4584 fi);
4585 extent_offset = found_key.offset -
4586 btrfs_file_extent_offset(leaf, fi);
4587
4588 /* FIXME blocksize != 4096 */
4589 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4590 if (extent_start != 0) {
4591 found_extent = 1;
4592 if (test_bit(BTRFS_ROOT_REF_COWS,
4593 &root->state))
4594 inode_sub_bytes(inode, num_dec);
4595 }
4596 }
4597 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4598 /*
4599 * we can't truncate inline items that have had
4600 * special encodings
4601 */
4602 if (!del_item &&
4603 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4604 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4605
4606 /*
4607 * Need to release path in order to truncate a
4608 * compressed extent. So delete any accumulated
4609 * extent items so far.
4610 */
4611 if (btrfs_file_extent_compression(leaf, fi) !=
4612 BTRFS_COMPRESS_NONE && pending_del_nr) {
4613 err = btrfs_del_items(trans, root, path,
4614 pending_del_slot,
4615 pending_del_nr);
4616 if (err) {
4617 btrfs_abort_transaction(trans,
4618 err);
4619 goto error;
4620 }
4621 pending_del_nr = 0;
4622 }
4623
4624 err = truncate_inline_extent(inode, path,
4625 &found_key,
4626 item_end,
4627 new_size);
4628 if (err) {
4629 btrfs_abort_transaction(trans, err);
4630 goto error;
4631 }
4632 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4633 &root->state)) {
4634 inode_sub_bytes(inode, item_end + 1 - new_size);
4635 }
4636 }
4637 delete:
4638 if (del_item) {
4639 if (!pending_del_nr) {
4640 /* no pending yet, add ourselves */
4641 pending_del_slot = path->slots[0];
4642 pending_del_nr = 1;
4643 } else if (pending_del_nr &&
4644 path->slots[0] + 1 == pending_del_slot) {
4645 /* hop on the pending chunk */
4646 pending_del_nr++;
4647 pending_del_slot = path->slots[0];
4648 } else {
4649 BUG();
4650 }
4651 } else {
4652 break;
4653 }
4654 should_throttle = 0;
4655
4656 if (found_extent &&
4657 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4658 root == fs_info->tree_root)) {
4659 btrfs_set_path_blocking(path);
4660 bytes_deleted += extent_num_bytes;
4661 ret = btrfs_free_extent(trans, fs_info, extent_start,
4662 extent_num_bytes, 0,
4663 btrfs_header_owner(leaf),
4664 ino, extent_offset);
4665 BUG_ON(ret);
4666 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4667 btrfs_async_run_delayed_refs(fs_info,
4668 trans->delayed_ref_updates * 2,
4669 trans->transid, 0);
4670 if (be_nice) {
4671 if (truncate_space_check(trans, root,
4672 extent_num_bytes)) {
4673 should_end = 1;
4674 }
4675 if (btrfs_should_throttle_delayed_refs(trans,
4676 fs_info))
4677 should_throttle = 1;
4678 }
4679 }
4680
4681 if (found_type == BTRFS_INODE_ITEM_KEY)
4682 break;
4683
4684 if (path->slots[0] == 0 ||
4685 path->slots[0] != pending_del_slot ||
4686 should_throttle || should_end) {
4687 if (pending_del_nr) {
4688 ret = btrfs_del_items(trans, root, path,
4689 pending_del_slot,
4690 pending_del_nr);
4691 if (ret) {
4692 btrfs_abort_transaction(trans, ret);
4693 goto error;
4694 }
4695 pending_del_nr = 0;
4696 }
4697 btrfs_release_path(path);
4698 if (should_throttle) {
4699 unsigned long updates = trans->delayed_ref_updates;
4700 if (updates) {
4701 trans->delayed_ref_updates = 0;
4702 ret = btrfs_run_delayed_refs(trans,
4703 fs_info,
4704 updates * 2);
4705 if (ret && !err)
4706 err = ret;
4707 }
4708 }
4709 /*
4710 * if we failed to refill our space rsv, bail out
4711 * and let the transaction restart
4712 */
4713 if (should_end) {
4714 err = -EAGAIN;
4715 goto error;
4716 }
4717 goto search_again;
4718 } else {
4719 path->slots[0]--;
4720 }
4721 }
4722 out:
4723 if (pending_del_nr) {
4724 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4725 pending_del_nr);
4726 if (ret)
4727 btrfs_abort_transaction(trans, ret);
4728 }
4729 error:
4730 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4731 ASSERT(last_size >= new_size);
4732 if (!err && last_size > new_size)
4733 last_size = new_size;
4734 btrfs_ordered_update_i_size(inode, last_size, NULL);
4735 }
4736
4737 btrfs_free_path(path);
4738
4739 if (be_nice && bytes_deleted > SZ_32M) {
4740 unsigned long updates = trans->delayed_ref_updates;
4741 if (updates) {
4742 trans->delayed_ref_updates = 0;
4743 ret = btrfs_run_delayed_refs(trans, fs_info,
4744 updates * 2);
4745 if (ret && !err)
4746 err = ret;
4747 }
4748 }
4749 return err;
4750 }
4751
4752 /*
4753 * btrfs_truncate_block - read, zero a chunk and write a block
4754 * @inode - inode that we're zeroing
4755 * @from - the offset to start zeroing
4756 * @len - the length to zero, 0 to zero the entire range respective to the
4757 * offset
4758 * @front - zero up to the offset instead of from the offset on
4759 *
4760 * This will find the block for the "from" offset and cow the block and zero the
4761 * part we want to zero. This is used with truncate and hole punching.
4762 */
4763 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4764 int front)
4765 {
4766 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4767 struct address_space *mapping = inode->i_mapping;
4768 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4769 struct btrfs_ordered_extent *ordered;
4770 struct extent_state *cached_state = NULL;
4771 struct extent_changeset *data_reserved = NULL;
4772 char *kaddr;
4773 u32 blocksize = fs_info->sectorsize;
4774 pgoff_t index = from >> PAGE_SHIFT;
4775 unsigned offset = from & (blocksize - 1);
4776 struct page *page;
4777 gfp_t mask = btrfs_alloc_write_mask(mapping);
4778 int ret = 0;
4779 u64 block_start;
4780 u64 block_end;
4781
4782 if ((offset & (blocksize - 1)) == 0 &&
4783 (!len || ((len & (blocksize - 1)) == 0)))
4784 goto out;
4785
4786 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4787 round_down(from, blocksize), blocksize);
4788 if (ret)
4789 goto out;
4790
4791 again:
4792 page = find_or_create_page(mapping, index, mask);
4793 if (!page) {
4794 btrfs_delalloc_release_space(inode, data_reserved,
4795 round_down(from, blocksize),
4796 blocksize);
4797 ret = -ENOMEM;
4798 goto out;
4799 }
4800
4801 block_start = round_down(from, blocksize);
4802 block_end = block_start + blocksize - 1;
4803
4804 if (!PageUptodate(page)) {
4805 ret = btrfs_readpage(NULL, page);
4806 lock_page(page);
4807 if (page->mapping != mapping) {
4808 unlock_page(page);
4809 put_page(page);
4810 goto again;
4811 }
4812 if (!PageUptodate(page)) {
4813 ret = -EIO;
4814 goto out_unlock;
4815 }
4816 }
4817 wait_on_page_writeback(page);
4818
4819 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4820 set_page_extent_mapped(page);
4821
4822 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4823 if (ordered) {
4824 unlock_extent_cached(io_tree, block_start, block_end,
4825 &cached_state, GFP_NOFS);
4826 unlock_page(page);
4827 put_page(page);
4828 btrfs_start_ordered_extent(inode, ordered, 1);
4829 btrfs_put_ordered_extent(ordered);
4830 goto again;
4831 }
4832
4833 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4834 EXTENT_DIRTY | EXTENT_DELALLOC |
4835 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4836 0, 0, &cached_state, GFP_NOFS);
4837
4838 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4839 &cached_state, 0);
4840 if (ret) {
4841 unlock_extent_cached(io_tree, block_start, block_end,
4842 &cached_state, GFP_NOFS);
4843 goto out_unlock;
4844 }
4845
4846 if (offset != blocksize) {
4847 if (!len)
4848 len = blocksize - offset;
4849 kaddr = kmap(page);
4850 if (front)
4851 memset(kaddr + (block_start - page_offset(page)),
4852 0, offset);
4853 else
4854 memset(kaddr + (block_start - page_offset(page)) + offset,
4855 0, len);
4856 flush_dcache_page(page);
4857 kunmap(page);
4858 }
4859 ClearPageChecked(page);
4860 set_page_dirty(page);
4861 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4862 GFP_NOFS);
4863
4864 out_unlock:
4865 if (ret)
4866 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4867 blocksize);
4868 unlock_page(page);
4869 put_page(page);
4870 out:
4871 extent_changeset_free(data_reserved);
4872 return ret;
4873 }
4874
4875 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4876 u64 offset, u64 len)
4877 {
4878 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4879 struct btrfs_trans_handle *trans;
4880 int ret;
4881
4882 /*
4883 * Still need to make sure the inode looks like it's been updated so
4884 * that any holes get logged if we fsync.
4885 */
4886 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4887 BTRFS_I(inode)->last_trans = fs_info->generation;
4888 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4889 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4890 return 0;
4891 }
4892
4893 /*
4894 * 1 - for the one we're dropping
4895 * 1 - for the one we're adding
4896 * 1 - for updating the inode.
4897 */
4898 trans = btrfs_start_transaction(root, 3);
4899 if (IS_ERR(trans))
4900 return PTR_ERR(trans);
4901
4902 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4903 if (ret) {
4904 btrfs_abort_transaction(trans, ret);
4905 btrfs_end_transaction(trans);
4906 return ret;
4907 }
4908
4909 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4910 offset, 0, 0, len, 0, len, 0, 0, 0);
4911 if (ret)
4912 btrfs_abort_transaction(trans, ret);
4913 else
4914 btrfs_update_inode(trans, root, inode);
4915 btrfs_end_transaction(trans);
4916 return ret;
4917 }
4918
4919 /*
4920 * This function puts in dummy file extents for the area we're creating a hole
4921 * for. So if we are truncating this file to a larger size we need to insert
4922 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4923 * the range between oldsize and size
4924 */
4925 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4926 {
4927 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4928 struct btrfs_root *root = BTRFS_I(inode)->root;
4929 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4930 struct extent_map *em = NULL;
4931 struct extent_state *cached_state = NULL;
4932 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4933 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4934 u64 block_end = ALIGN(size, fs_info->sectorsize);
4935 u64 last_byte;
4936 u64 cur_offset;
4937 u64 hole_size;
4938 int err = 0;
4939
4940 /*
4941 * If our size started in the middle of a block we need to zero out the
4942 * rest of the block before we expand the i_size, otherwise we could
4943 * expose stale data.
4944 */
4945 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4946 if (err)
4947 return err;
4948
4949 if (size <= hole_start)
4950 return 0;
4951
4952 while (1) {
4953 struct btrfs_ordered_extent *ordered;
4954
4955 lock_extent_bits(io_tree, hole_start, block_end - 1,
4956 &cached_state);
4957 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4958 block_end - hole_start);
4959 if (!ordered)
4960 break;
4961 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4962 &cached_state, GFP_NOFS);
4963 btrfs_start_ordered_extent(inode, ordered, 1);
4964 btrfs_put_ordered_extent(ordered);
4965 }
4966
4967 cur_offset = hole_start;
4968 while (1) {
4969 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4970 block_end - cur_offset, 0);
4971 if (IS_ERR(em)) {
4972 err = PTR_ERR(em);
4973 em = NULL;
4974 break;
4975 }
4976 last_byte = min(extent_map_end(em), block_end);
4977 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4978 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4979 struct extent_map *hole_em;
4980 hole_size = last_byte - cur_offset;
4981
4982 err = maybe_insert_hole(root, inode, cur_offset,
4983 hole_size);
4984 if (err)
4985 break;
4986 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4987 cur_offset + hole_size - 1, 0);
4988 hole_em = alloc_extent_map();
4989 if (!hole_em) {
4990 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4991 &BTRFS_I(inode)->runtime_flags);
4992 goto next;
4993 }
4994 hole_em->start = cur_offset;
4995 hole_em->len = hole_size;
4996 hole_em->orig_start = cur_offset;
4997
4998 hole_em->block_start = EXTENT_MAP_HOLE;
4999 hole_em->block_len = 0;
5000 hole_em->orig_block_len = 0;
5001 hole_em->ram_bytes = hole_size;
5002 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5003 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5004 hole_em->generation = fs_info->generation;
5005
5006 while (1) {
5007 write_lock(&em_tree->lock);
5008 err = add_extent_mapping(em_tree, hole_em, 1);
5009 write_unlock(&em_tree->lock);
5010 if (err != -EEXIST)
5011 break;
5012 btrfs_drop_extent_cache(BTRFS_I(inode),
5013 cur_offset,
5014 cur_offset +
5015 hole_size - 1, 0);
5016 }
5017 free_extent_map(hole_em);
5018 }
5019 next:
5020 free_extent_map(em);
5021 em = NULL;
5022 cur_offset = last_byte;
5023 if (cur_offset >= block_end)
5024 break;
5025 }
5026 free_extent_map(em);
5027 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
5028 GFP_NOFS);
5029 return err;
5030 }
5031
5032 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5033 {
5034 struct btrfs_root *root = BTRFS_I(inode)->root;
5035 struct btrfs_trans_handle *trans;
5036 loff_t oldsize = i_size_read(inode);
5037 loff_t newsize = attr->ia_size;
5038 int mask = attr->ia_valid;
5039 int ret;
5040
5041 /*
5042 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5043 * special case where we need to update the times despite not having
5044 * these flags set. For all other operations the VFS set these flags
5045 * explicitly if it wants a timestamp update.
5046 */
5047 if (newsize != oldsize) {
5048 inode_inc_iversion(inode);
5049 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5050 inode->i_ctime = inode->i_mtime =
5051 current_time(inode);
5052 }
5053
5054 if (newsize > oldsize) {
5055 /*
5056 * Don't do an expanding truncate while snapshoting is ongoing.
5057 * This is to ensure the snapshot captures a fully consistent
5058 * state of this file - if the snapshot captures this expanding
5059 * truncation, it must capture all writes that happened before
5060 * this truncation.
5061 */
5062 btrfs_wait_for_snapshot_creation(root);
5063 ret = btrfs_cont_expand(inode, oldsize, newsize);
5064 if (ret) {
5065 btrfs_end_write_no_snapshoting(root);
5066 return ret;
5067 }
5068
5069 trans = btrfs_start_transaction(root, 1);
5070 if (IS_ERR(trans)) {
5071 btrfs_end_write_no_snapshoting(root);
5072 return PTR_ERR(trans);
5073 }
5074
5075 i_size_write(inode, newsize);
5076 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5077 pagecache_isize_extended(inode, oldsize, newsize);
5078 ret = btrfs_update_inode(trans, root, inode);
5079 btrfs_end_write_no_snapshoting(root);
5080 btrfs_end_transaction(trans);
5081 } else {
5082
5083 /*
5084 * We're truncating a file that used to have good data down to
5085 * zero. Make sure it gets into the ordered flush list so that
5086 * any new writes get down to disk quickly.
5087 */
5088 if (newsize == 0)
5089 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5090 &BTRFS_I(inode)->runtime_flags);
5091
5092 /*
5093 * 1 for the orphan item we're going to add
5094 * 1 for the orphan item deletion.
5095 */
5096 trans = btrfs_start_transaction(root, 2);
5097 if (IS_ERR(trans))
5098 return PTR_ERR(trans);
5099
5100 /*
5101 * We need to do this in case we fail at _any_ point during the
5102 * actual truncate. Once we do the truncate_setsize we could
5103 * invalidate pages which forces any outstanding ordered io to
5104 * be instantly completed which will give us extents that need
5105 * to be truncated. If we fail to get an orphan inode down we
5106 * could have left over extents that were never meant to live,
5107 * so we need to guarantee from this point on that everything
5108 * will be consistent.
5109 */
5110 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
5111 btrfs_end_transaction(trans);
5112 if (ret)
5113 return ret;
5114
5115 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5116 truncate_setsize(inode, newsize);
5117
5118 /* Disable nonlocked read DIO to avoid the end less truncate */
5119 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5120 inode_dio_wait(inode);
5121 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5122
5123 ret = btrfs_truncate(inode);
5124 if (ret && inode->i_nlink) {
5125 int err;
5126
5127 /* To get a stable disk_i_size */
5128 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5129 if (err) {
5130 btrfs_orphan_del(NULL, BTRFS_I(inode));
5131 return err;
5132 }
5133
5134 /*
5135 * failed to truncate, disk_i_size is only adjusted down
5136 * as we remove extents, so it should represent the true
5137 * size of the inode, so reset the in memory size and
5138 * delete our orphan entry.
5139 */
5140 trans = btrfs_join_transaction(root);
5141 if (IS_ERR(trans)) {
5142 btrfs_orphan_del(NULL, BTRFS_I(inode));
5143 return ret;
5144 }
5145 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5146 err = btrfs_orphan_del(trans, BTRFS_I(inode));
5147 if (err)
5148 btrfs_abort_transaction(trans, err);
5149 btrfs_end_transaction(trans);
5150 }
5151 }
5152
5153 return ret;
5154 }
5155
5156 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5157 {
5158 struct inode *inode = d_inode(dentry);
5159 struct btrfs_root *root = BTRFS_I(inode)->root;
5160 int err;
5161
5162 if (btrfs_root_readonly(root))
5163 return -EROFS;
5164
5165 err = setattr_prepare(dentry, attr);
5166 if (err)
5167 return err;
5168
5169 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5170 err = btrfs_setsize(inode, attr);
5171 if (err)
5172 return err;
5173 }
5174
5175 if (attr->ia_valid) {
5176 setattr_copy(inode, attr);
5177 inode_inc_iversion(inode);
5178 err = btrfs_dirty_inode(inode);
5179
5180 if (!err && attr->ia_valid & ATTR_MODE)
5181 err = posix_acl_chmod(inode, inode->i_mode);
5182 }
5183
5184 return err;
5185 }
5186
5187 /*
5188 * While truncating the inode pages during eviction, we get the VFS calling
5189 * btrfs_invalidatepage() against each page of the inode. This is slow because
5190 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5191 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5192 * extent_state structures over and over, wasting lots of time.
5193 *
5194 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5195 * those expensive operations on a per page basis and do only the ordered io
5196 * finishing, while we release here the extent_map and extent_state structures,
5197 * without the excessive merging and splitting.
5198 */
5199 static void evict_inode_truncate_pages(struct inode *inode)
5200 {
5201 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5202 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5203 struct rb_node *node;
5204
5205 ASSERT(inode->i_state & I_FREEING);
5206 truncate_inode_pages_final(&inode->i_data);
5207
5208 write_lock(&map_tree->lock);
5209 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5210 struct extent_map *em;
5211
5212 node = rb_first(&map_tree->map);
5213 em = rb_entry(node, struct extent_map, rb_node);
5214 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5215 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5216 remove_extent_mapping(map_tree, em);
5217 free_extent_map(em);
5218 if (need_resched()) {
5219 write_unlock(&map_tree->lock);
5220 cond_resched();
5221 write_lock(&map_tree->lock);
5222 }
5223 }
5224 write_unlock(&map_tree->lock);
5225
5226 /*
5227 * Keep looping until we have no more ranges in the io tree.
5228 * We can have ongoing bios started by readpages (called from readahead)
5229 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5230 * still in progress (unlocked the pages in the bio but did not yet
5231 * unlocked the ranges in the io tree). Therefore this means some
5232 * ranges can still be locked and eviction started because before
5233 * submitting those bios, which are executed by a separate task (work
5234 * queue kthread), inode references (inode->i_count) were not taken
5235 * (which would be dropped in the end io callback of each bio).
5236 * Therefore here we effectively end up waiting for those bios and
5237 * anyone else holding locked ranges without having bumped the inode's
5238 * reference count - if we don't do it, when they access the inode's
5239 * io_tree to unlock a range it may be too late, leading to an
5240 * use-after-free issue.
5241 */
5242 spin_lock(&io_tree->lock);
5243 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5244 struct extent_state *state;
5245 struct extent_state *cached_state = NULL;
5246 u64 start;
5247 u64 end;
5248
5249 node = rb_first(&io_tree->state);
5250 state = rb_entry(node, struct extent_state, rb_node);
5251 start = state->start;
5252 end = state->end;
5253 spin_unlock(&io_tree->lock);
5254
5255 lock_extent_bits(io_tree, start, end, &cached_state);
5256
5257 /*
5258 * If still has DELALLOC flag, the extent didn't reach disk,
5259 * and its reserved space won't be freed by delayed_ref.
5260 * So we need to free its reserved space here.
5261 * (Refer to comment in btrfs_invalidatepage, case 2)
5262 *
5263 * Note, end is the bytenr of last byte, so we need + 1 here.
5264 */
5265 if (state->state & EXTENT_DELALLOC)
5266 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5267
5268 clear_extent_bit(io_tree, start, end,
5269 EXTENT_LOCKED | EXTENT_DIRTY |
5270 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5271 EXTENT_DEFRAG, 1, 1,
5272 &cached_state, GFP_NOFS);
5273
5274 cond_resched();
5275 spin_lock(&io_tree->lock);
5276 }
5277 spin_unlock(&io_tree->lock);
5278 }
5279
5280 void btrfs_evict_inode(struct inode *inode)
5281 {
5282 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5283 struct btrfs_trans_handle *trans;
5284 struct btrfs_root *root = BTRFS_I(inode)->root;
5285 struct btrfs_block_rsv *rsv, *global_rsv;
5286 int steal_from_global = 0;
5287 u64 min_size;
5288 int ret;
5289
5290 trace_btrfs_inode_evict(inode);
5291
5292 if (!root) {
5293 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5294 return;
5295 }
5296
5297 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5298
5299 evict_inode_truncate_pages(inode);
5300
5301 if (inode->i_nlink &&
5302 ((btrfs_root_refs(&root->root_item) != 0 &&
5303 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5304 btrfs_is_free_space_inode(BTRFS_I(inode))))
5305 goto no_delete;
5306
5307 if (is_bad_inode(inode)) {
5308 btrfs_orphan_del(NULL, BTRFS_I(inode));
5309 goto no_delete;
5310 }
5311 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5312 if (!special_file(inode->i_mode))
5313 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5314
5315 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5316
5317 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5318 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5319 &BTRFS_I(inode)->runtime_flags));
5320 goto no_delete;
5321 }
5322
5323 if (inode->i_nlink > 0) {
5324 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5325 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5326 goto no_delete;
5327 }
5328
5329 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5330 if (ret) {
5331 btrfs_orphan_del(NULL, BTRFS_I(inode));
5332 goto no_delete;
5333 }
5334
5335 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5336 if (!rsv) {
5337 btrfs_orphan_del(NULL, BTRFS_I(inode));
5338 goto no_delete;
5339 }
5340 rsv->size = min_size;
5341 rsv->failfast = 1;
5342 global_rsv = &fs_info->global_block_rsv;
5343
5344 btrfs_i_size_write(BTRFS_I(inode), 0);
5345
5346 /*
5347 * This is a bit simpler than btrfs_truncate since we've already
5348 * reserved our space for our orphan item in the unlink, so we just
5349 * need to reserve some slack space in case we add bytes and update
5350 * inode item when doing the truncate.
5351 */
5352 while (1) {
5353 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5354 BTRFS_RESERVE_FLUSH_LIMIT);
5355
5356 /*
5357 * Try and steal from the global reserve since we will
5358 * likely not use this space anyway, we want to try as
5359 * hard as possible to get this to work.
5360 */
5361 if (ret)
5362 steal_from_global++;
5363 else
5364 steal_from_global = 0;
5365 ret = 0;
5366
5367 /*
5368 * steal_from_global == 0: we reserved stuff, hooray!
5369 * steal_from_global == 1: we didn't reserve stuff, boo!
5370 * steal_from_global == 2: we've committed, still not a lot of
5371 * room but maybe we'll have room in the global reserve this
5372 * time.
5373 * steal_from_global == 3: abandon all hope!
5374 */
5375 if (steal_from_global > 2) {
5376 btrfs_warn(fs_info,
5377 "Could not get space for a delete, will truncate on mount %d",
5378 ret);
5379 btrfs_orphan_del(NULL, BTRFS_I(inode));
5380 btrfs_free_block_rsv(fs_info, rsv);
5381 goto no_delete;
5382 }
5383
5384 trans = btrfs_join_transaction(root);
5385 if (IS_ERR(trans)) {
5386 btrfs_orphan_del(NULL, BTRFS_I(inode));
5387 btrfs_free_block_rsv(fs_info, rsv);
5388 goto no_delete;
5389 }
5390
5391 /*
5392 * We can't just steal from the global reserve, we need to make
5393 * sure there is room to do it, if not we need to commit and try
5394 * again.
5395 */
5396 if (steal_from_global) {
5397 if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5398 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5399 min_size, 0);
5400 else
5401 ret = -ENOSPC;
5402 }
5403
5404 /*
5405 * Couldn't steal from the global reserve, we have too much
5406 * pending stuff built up, commit the transaction and try it
5407 * again.
5408 */
5409 if (ret) {
5410 ret = btrfs_commit_transaction(trans);
5411 if (ret) {
5412 btrfs_orphan_del(NULL, BTRFS_I(inode));
5413 btrfs_free_block_rsv(fs_info, rsv);
5414 goto no_delete;
5415 }
5416 continue;
5417 } else {
5418 steal_from_global = 0;
5419 }
5420
5421 trans->block_rsv = rsv;
5422
5423 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5424 if (ret != -ENOSPC && ret != -EAGAIN)
5425 break;
5426
5427 trans->block_rsv = &fs_info->trans_block_rsv;
5428 btrfs_end_transaction(trans);
5429 trans = NULL;
5430 btrfs_btree_balance_dirty(fs_info);
5431 }
5432
5433 btrfs_free_block_rsv(fs_info, rsv);
5434
5435 /*
5436 * Errors here aren't a big deal, it just means we leave orphan items
5437 * in the tree. They will be cleaned up on the next mount.
5438 */
5439 if (ret == 0) {
5440 trans->block_rsv = root->orphan_block_rsv;
5441 btrfs_orphan_del(trans, BTRFS_I(inode));
5442 } else {
5443 btrfs_orphan_del(NULL, BTRFS_I(inode));
5444 }
5445
5446 trans->block_rsv = &fs_info->trans_block_rsv;
5447 if (!(root == fs_info->tree_root ||
5448 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5449 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5450
5451 btrfs_end_transaction(trans);
5452 btrfs_btree_balance_dirty(fs_info);
5453 no_delete:
5454 btrfs_remove_delayed_node(BTRFS_I(inode));
5455 clear_inode(inode);
5456 }
5457
5458 /*
5459 * this returns the key found in the dir entry in the location pointer.
5460 * If no dir entries were found, location->objectid is 0.
5461 */
5462 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5463 struct btrfs_key *location)
5464 {
5465 const char *name = dentry->d_name.name;
5466 int namelen = dentry->d_name.len;
5467 struct btrfs_dir_item *di;
5468 struct btrfs_path *path;
5469 struct btrfs_root *root = BTRFS_I(dir)->root;
5470 int ret = 0;
5471
5472 path = btrfs_alloc_path();
5473 if (!path)
5474 return -ENOMEM;
5475
5476 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5477 name, namelen, 0);
5478 if (IS_ERR(di))
5479 ret = PTR_ERR(di);
5480
5481 if (IS_ERR_OR_NULL(di))
5482 goto out_err;
5483
5484 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5485 out:
5486 btrfs_free_path(path);
5487 return ret;
5488 out_err:
5489 location->objectid = 0;
5490 goto out;
5491 }
5492
5493 /*
5494 * when we hit a tree root in a directory, the btrfs part of the inode
5495 * needs to be changed to reflect the root directory of the tree root. This
5496 * is kind of like crossing a mount point.
5497 */
5498 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5499 struct inode *dir,
5500 struct dentry *dentry,
5501 struct btrfs_key *location,
5502 struct btrfs_root **sub_root)
5503 {
5504 struct btrfs_path *path;
5505 struct btrfs_root *new_root;
5506 struct btrfs_root_ref *ref;
5507 struct extent_buffer *leaf;
5508 struct btrfs_key key;
5509 int ret;
5510 int err = 0;
5511
5512 path = btrfs_alloc_path();
5513 if (!path) {
5514 err = -ENOMEM;
5515 goto out;
5516 }
5517
5518 err = -ENOENT;
5519 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5520 key.type = BTRFS_ROOT_REF_KEY;
5521 key.offset = location->objectid;
5522
5523 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5524 if (ret) {
5525 if (ret < 0)
5526 err = ret;
5527 goto out;
5528 }
5529
5530 leaf = path->nodes[0];
5531 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5532 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5533 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5534 goto out;
5535
5536 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5537 (unsigned long)(ref + 1),
5538 dentry->d_name.len);
5539 if (ret)
5540 goto out;
5541
5542 btrfs_release_path(path);
5543
5544 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5545 if (IS_ERR(new_root)) {
5546 err = PTR_ERR(new_root);
5547 goto out;
5548 }
5549
5550 *sub_root = new_root;
5551 location->objectid = btrfs_root_dirid(&new_root->root_item);
5552 location->type = BTRFS_INODE_ITEM_KEY;
5553 location->offset = 0;
5554 err = 0;
5555 out:
5556 btrfs_free_path(path);
5557 return err;
5558 }
5559
5560 static void inode_tree_add(struct inode *inode)
5561 {
5562 struct btrfs_root *root = BTRFS_I(inode)->root;
5563 struct btrfs_inode *entry;
5564 struct rb_node **p;
5565 struct rb_node *parent;
5566 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5567 u64 ino = btrfs_ino(BTRFS_I(inode));
5568
5569 if (inode_unhashed(inode))
5570 return;
5571 parent = NULL;
5572 spin_lock(&root->inode_lock);
5573 p = &root->inode_tree.rb_node;
5574 while (*p) {
5575 parent = *p;
5576 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5577
5578 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5579 p = &parent->rb_left;
5580 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5581 p = &parent->rb_right;
5582 else {
5583 WARN_ON(!(entry->vfs_inode.i_state &
5584 (I_WILL_FREE | I_FREEING)));
5585 rb_replace_node(parent, new, &root->inode_tree);
5586 RB_CLEAR_NODE(parent);
5587 spin_unlock(&root->inode_lock);
5588 return;
5589 }
5590 }
5591 rb_link_node(new, parent, p);
5592 rb_insert_color(new, &root->inode_tree);
5593 spin_unlock(&root->inode_lock);
5594 }
5595
5596 static void inode_tree_del(struct inode *inode)
5597 {
5598 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5599 struct btrfs_root *root = BTRFS_I(inode)->root;
5600 int empty = 0;
5601
5602 spin_lock(&root->inode_lock);
5603 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5604 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5605 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5606 empty = RB_EMPTY_ROOT(&root->inode_tree);
5607 }
5608 spin_unlock(&root->inode_lock);
5609
5610 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5611 synchronize_srcu(&fs_info->subvol_srcu);
5612 spin_lock(&root->inode_lock);
5613 empty = RB_EMPTY_ROOT(&root->inode_tree);
5614 spin_unlock(&root->inode_lock);
5615 if (empty)
5616 btrfs_add_dead_root(root);
5617 }
5618 }
5619
5620 void btrfs_invalidate_inodes(struct btrfs_root *root)
5621 {
5622 struct btrfs_fs_info *fs_info = root->fs_info;
5623 struct rb_node *node;
5624 struct rb_node *prev;
5625 struct btrfs_inode *entry;
5626 struct inode *inode;
5627 u64 objectid = 0;
5628
5629 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5630 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5631
5632 spin_lock(&root->inode_lock);
5633 again:
5634 node = root->inode_tree.rb_node;
5635 prev = NULL;
5636 while (node) {
5637 prev = node;
5638 entry = rb_entry(node, struct btrfs_inode, rb_node);
5639
5640 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5641 node = node->rb_left;
5642 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5643 node = node->rb_right;
5644 else
5645 break;
5646 }
5647 if (!node) {
5648 while (prev) {
5649 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5650 if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5651 node = prev;
5652 break;
5653 }
5654 prev = rb_next(prev);
5655 }
5656 }
5657 while (node) {
5658 entry = rb_entry(node, struct btrfs_inode, rb_node);
5659 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5660 inode = igrab(&entry->vfs_inode);
5661 if (inode) {
5662 spin_unlock(&root->inode_lock);
5663 if (atomic_read(&inode->i_count) > 1)
5664 d_prune_aliases(inode);
5665 /*
5666 * btrfs_drop_inode will have it removed from
5667 * the inode cache when its usage count
5668 * hits zero.
5669 */
5670 iput(inode);
5671 cond_resched();
5672 spin_lock(&root->inode_lock);
5673 goto again;
5674 }
5675
5676 if (cond_resched_lock(&root->inode_lock))
5677 goto again;
5678
5679 node = rb_next(node);
5680 }
5681 spin_unlock(&root->inode_lock);
5682 }
5683
5684 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5685 {
5686 struct btrfs_iget_args *args = p;
5687 inode->i_ino = args->location->objectid;
5688 memcpy(&BTRFS_I(inode)->location, args->location,
5689 sizeof(*args->location));
5690 BTRFS_I(inode)->root = args->root;
5691 return 0;
5692 }
5693
5694 static int btrfs_find_actor(struct inode *inode, void *opaque)
5695 {
5696 struct btrfs_iget_args *args = opaque;
5697 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5698 args->root == BTRFS_I(inode)->root;
5699 }
5700
5701 static struct inode *btrfs_iget_locked(struct super_block *s,
5702 struct btrfs_key *location,
5703 struct btrfs_root *root)
5704 {
5705 struct inode *inode;
5706 struct btrfs_iget_args args;
5707 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5708
5709 args.location = location;
5710 args.root = root;
5711
5712 inode = iget5_locked(s, hashval, btrfs_find_actor,
5713 btrfs_init_locked_inode,
5714 (void *)&args);
5715 return inode;
5716 }
5717
5718 /* Get an inode object given its location and corresponding root.
5719 * Returns in *is_new if the inode was read from disk
5720 */
5721 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5722 struct btrfs_root *root, int *new)
5723 {
5724 struct inode *inode;
5725
5726 inode = btrfs_iget_locked(s, location, root);
5727 if (!inode)
5728 return ERR_PTR(-ENOMEM);
5729
5730 if (inode->i_state & I_NEW) {
5731 int ret;
5732
5733 ret = btrfs_read_locked_inode(inode);
5734 if (!is_bad_inode(inode)) {
5735 inode_tree_add(inode);
5736 unlock_new_inode(inode);
5737 if (new)
5738 *new = 1;
5739 } else {
5740 unlock_new_inode(inode);
5741 iput(inode);
5742 ASSERT(ret < 0);
5743 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5744 }
5745 }
5746
5747 return inode;
5748 }
5749
5750 static struct inode *new_simple_dir(struct super_block *s,
5751 struct btrfs_key *key,
5752 struct btrfs_root *root)
5753 {
5754 struct inode *inode = new_inode(s);
5755
5756 if (!inode)
5757 return ERR_PTR(-ENOMEM);
5758
5759 BTRFS_I(inode)->root = root;
5760 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5761 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5762
5763 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5764 inode->i_op = &btrfs_dir_ro_inode_operations;
5765 inode->i_opflags &= ~IOP_XATTR;
5766 inode->i_fop = &simple_dir_operations;
5767 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5768 inode->i_mtime = current_time(inode);
5769 inode->i_atime = inode->i_mtime;
5770 inode->i_ctime = inode->i_mtime;
5771 BTRFS_I(inode)->i_otime = inode->i_mtime;
5772
5773 return inode;
5774 }
5775
5776 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5777 {
5778 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5779 struct inode *inode;
5780 struct btrfs_root *root = BTRFS_I(dir)->root;
5781 struct btrfs_root *sub_root = root;
5782 struct btrfs_key location;
5783 int index;
5784 int ret = 0;
5785
5786 if (dentry->d_name.len > BTRFS_NAME_LEN)
5787 return ERR_PTR(-ENAMETOOLONG);
5788
5789 ret = btrfs_inode_by_name(dir, dentry, &location);
5790 if (ret < 0)
5791 return ERR_PTR(ret);
5792
5793 if (location.objectid == 0)
5794 return ERR_PTR(-ENOENT);
5795
5796 if (location.type == BTRFS_INODE_ITEM_KEY) {
5797 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5798 return inode;
5799 }
5800
5801 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5802
5803 index = srcu_read_lock(&fs_info->subvol_srcu);
5804 ret = fixup_tree_root_location(fs_info, dir, dentry,
5805 &location, &sub_root);
5806 if (ret < 0) {
5807 if (ret != -ENOENT)
5808 inode = ERR_PTR(ret);
5809 else
5810 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5811 } else {
5812 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5813 }
5814 srcu_read_unlock(&fs_info->subvol_srcu, index);
5815
5816 if (!IS_ERR(inode) && root != sub_root) {
5817 down_read(&fs_info->cleanup_work_sem);
5818 if (!(inode->i_sb->s_flags & MS_RDONLY))
5819 ret = btrfs_orphan_cleanup(sub_root);
5820 up_read(&fs_info->cleanup_work_sem);
5821 if (ret) {
5822 iput(inode);
5823 inode = ERR_PTR(ret);
5824 }
5825 }
5826
5827 return inode;
5828 }
5829
5830 static int btrfs_dentry_delete(const struct dentry *dentry)
5831 {
5832 struct btrfs_root *root;
5833 struct inode *inode = d_inode(dentry);
5834
5835 if (!inode && !IS_ROOT(dentry))
5836 inode = d_inode(dentry->d_parent);
5837
5838 if (inode) {
5839 root = BTRFS_I(inode)->root;
5840 if (btrfs_root_refs(&root->root_item) == 0)
5841 return 1;
5842
5843 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5844 return 1;
5845 }
5846 return 0;
5847 }
5848
5849 static void btrfs_dentry_release(struct dentry *dentry)
5850 {
5851 kfree(dentry->d_fsdata);
5852 }
5853
5854 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5855 unsigned int flags)
5856 {
5857 struct inode *inode;
5858
5859 inode = btrfs_lookup_dentry(dir, dentry);
5860 if (IS_ERR(inode)) {
5861 if (PTR_ERR(inode) == -ENOENT)
5862 inode = NULL;
5863 else
5864 return ERR_CAST(inode);
5865 }
5866
5867 return d_splice_alias(inode, dentry);
5868 }
5869
5870 unsigned char btrfs_filetype_table[] = {
5871 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5872 };
5873
5874 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5875 {
5876 struct inode *inode = file_inode(file);
5877 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5878 struct btrfs_root *root = BTRFS_I(inode)->root;
5879 struct btrfs_dir_item *di;
5880 struct btrfs_key key;
5881 struct btrfs_key found_key;
5882 struct btrfs_path *path;
5883 struct list_head ins_list;
5884 struct list_head del_list;
5885 int ret;
5886 struct extent_buffer *leaf;
5887 int slot;
5888 unsigned char d_type;
5889 int over = 0;
5890 char tmp_name[32];
5891 char *name_ptr;
5892 int name_len;
5893 bool put = false;
5894 struct btrfs_key location;
5895
5896 if (!dir_emit_dots(file, ctx))
5897 return 0;
5898
5899 path = btrfs_alloc_path();
5900 if (!path)
5901 return -ENOMEM;
5902
5903 path->reada = READA_FORWARD;
5904
5905 INIT_LIST_HEAD(&ins_list);
5906 INIT_LIST_HEAD(&del_list);
5907 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5908
5909 key.type = BTRFS_DIR_INDEX_KEY;
5910 key.offset = ctx->pos;
5911 key.objectid = btrfs_ino(BTRFS_I(inode));
5912
5913 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5914 if (ret < 0)
5915 goto err;
5916
5917 while (1) {
5918 leaf = path->nodes[0];
5919 slot = path->slots[0];
5920 if (slot >= btrfs_header_nritems(leaf)) {
5921 ret = btrfs_next_leaf(root, path);
5922 if (ret < 0)
5923 goto err;
5924 else if (ret > 0)
5925 break;
5926 continue;
5927 }
5928
5929 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5930
5931 if (found_key.objectid != key.objectid)
5932 break;
5933 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5934 break;
5935 if (found_key.offset < ctx->pos)
5936 goto next;
5937 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5938 goto next;
5939
5940 ctx->pos = found_key.offset;
5941
5942 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5943 if (verify_dir_item(fs_info, leaf, slot, di))
5944 goto next;
5945
5946 name_len = btrfs_dir_name_len(leaf, di);
5947 if (name_len <= sizeof(tmp_name)) {
5948 name_ptr = tmp_name;
5949 } else {
5950 name_ptr = kmalloc(name_len, GFP_KERNEL);
5951 if (!name_ptr) {
5952 ret = -ENOMEM;
5953 goto err;
5954 }
5955 }
5956 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5957 name_len);
5958
5959 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5960 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5961
5962 over = !dir_emit(ctx, name_ptr, name_len, location.objectid,
5963 d_type);
5964
5965 if (name_ptr != tmp_name)
5966 kfree(name_ptr);
5967
5968 if (over)
5969 goto nopos;
5970 ctx->pos++;
5971 next:
5972 path->slots[0]++;
5973 }
5974
5975 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5976 if (ret)
5977 goto nopos;
5978
5979 /*
5980 * Stop new entries from being returned after we return the last
5981 * entry.
5982 *
5983 * New directory entries are assigned a strictly increasing
5984 * offset. This means that new entries created during readdir
5985 * are *guaranteed* to be seen in the future by that readdir.
5986 * This has broken buggy programs which operate on names as
5987 * they're returned by readdir. Until we re-use freed offsets
5988 * we have this hack to stop new entries from being returned
5989 * under the assumption that they'll never reach this huge
5990 * offset.
5991 *
5992 * This is being careful not to overflow 32bit loff_t unless the
5993 * last entry requires it because doing so has broken 32bit apps
5994 * in the past.
5995 */
5996 if (ctx->pos >= INT_MAX)
5997 ctx->pos = LLONG_MAX;
5998 else
5999 ctx->pos = INT_MAX;
6000 nopos:
6001 ret = 0;
6002 err:
6003 if (put)
6004 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6005 btrfs_free_path(path);
6006 return ret;
6007 }
6008
6009 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6010 {
6011 struct btrfs_root *root = BTRFS_I(inode)->root;
6012 struct btrfs_trans_handle *trans;
6013 int ret = 0;
6014 bool nolock = false;
6015
6016 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6017 return 0;
6018
6019 if (btrfs_fs_closing(root->fs_info) &&
6020 btrfs_is_free_space_inode(BTRFS_I(inode)))
6021 nolock = true;
6022
6023 if (wbc->sync_mode == WB_SYNC_ALL) {
6024 if (nolock)
6025 trans = btrfs_join_transaction_nolock(root);
6026 else
6027 trans = btrfs_join_transaction(root);
6028 if (IS_ERR(trans))
6029 return PTR_ERR(trans);
6030 ret = btrfs_commit_transaction(trans);
6031 }
6032 return ret;
6033 }
6034
6035 /*
6036 * This is somewhat expensive, updating the tree every time the
6037 * inode changes. But, it is most likely to find the inode in cache.
6038 * FIXME, needs more benchmarking...there are no reasons other than performance
6039 * to keep or drop this code.
6040 */
6041 static int btrfs_dirty_inode(struct inode *inode)
6042 {
6043 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6044 struct btrfs_root *root = BTRFS_I(inode)->root;
6045 struct btrfs_trans_handle *trans;
6046 int ret;
6047
6048 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6049 return 0;
6050
6051 trans = btrfs_join_transaction(root);
6052 if (IS_ERR(trans))
6053 return PTR_ERR(trans);
6054
6055 ret = btrfs_update_inode(trans, root, inode);
6056 if (ret && ret == -ENOSPC) {
6057 /* whoops, lets try again with the full transaction */
6058 btrfs_end_transaction(trans);
6059 trans = btrfs_start_transaction(root, 1);
6060 if (IS_ERR(trans))
6061 return PTR_ERR(trans);
6062
6063 ret = btrfs_update_inode(trans, root, inode);
6064 }
6065 btrfs_end_transaction(trans);
6066 if (BTRFS_I(inode)->delayed_node)
6067 btrfs_balance_delayed_items(fs_info);
6068
6069 return ret;
6070 }
6071
6072 /*
6073 * This is a copy of file_update_time. We need this so we can return error on
6074 * ENOSPC for updating the inode in the case of file write and mmap writes.
6075 */
6076 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6077 int flags)
6078 {
6079 struct btrfs_root *root = BTRFS_I(inode)->root;
6080
6081 if (btrfs_root_readonly(root))
6082 return -EROFS;
6083
6084 if (flags & S_VERSION)
6085 inode_inc_iversion(inode);
6086 if (flags & S_CTIME)
6087 inode->i_ctime = *now;
6088 if (flags & S_MTIME)
6089 inode->i_mtime = *now;
6090 if (flags & S_ATIME)
6091 inode->i_atime = *now;
6092 return btrfs_dirty_inode(inode);
6093 }
6094
6095 /*
6096 * find the highest existing sequence number in a directory
6097 * and then set the in-memory index_cnt variable to reflect
6098 * free sequence numbers
6099 */
6100 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6101 {
6102 struct btrfs_root *root = inode->root;
6103 struct btrfs_key key, found_key;
6104 struct btrfs_path *path;
6105 struct extent_buffer *leaf;
6106 int ret;
6107
6108 key.objectid = btrfs_ino(inode);
6109 key.type = BTRFS_DIR_INDEX_KEY;
6110 key.offset = (u64)-1;
6111
6112 path = btrfs_alloc_path();
6113 if (!path)
6114 return -ENOMEM;
6115
6116 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6117 if (ret < 0)
6118 goto out;
6119 /* FIXME: we should be able to handle this */
6120 if (ret == 0)
6121 goto out;
6122 ret = 0;
6123
6124 /*
6125 * MAGIC NUMBER EXPLANATION:
6126 * since we search a directory based on f_pos we have to start at 2
6127 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6128 * else has to start at 2
6129 */
6130 if (path->slots[0] == 0) {
6131 inode->index_cnt = 2;
6132 goto out;
6133 }
6134
6135 path->slots[0]--;
6136
6137 leaf = path->nodes[0];
6138 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6139
6140 if (found_key.objectid != btrfs_ino(inode) ||
6141 found_key.type != BTRFS_DIR_INDEX_KEY) {
6142 inode->index_cnt = 2;
6143 goto out;
6144 }
6145
6146 inode->index_cnt = found_key.offset + 1;
6147 out:
6148 btrfs_free_path(path);
6149 return ret;
6150 }
6151
6152 /*
6153 * helper to find a free sequence number in a given directory. This current
6154 * code is very simple, later versions will do smarter things in the btree
6155 */
6156 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6157 {
6158 int ret = 0;
6159
6160 if (dir->index_cnt == (u64)-1) {
6161 ret = btrfs_inode_delayed_dir_index_count(dir);
6162 if (ret) {
6163 ret = btrfs_set_inode_index_count(dir);
6164 if (ret)
6165 return ret;
6166 }
6167 }
6168
6169 *index = dir->index_cnt;
6170 dir->index_cnt++;
6171
6172 return ret;
6173 }
6174
6175 static int btrfs_insert_inode_locked(struct inode *inode)
6176 {
6177 struct btrfs_iget_args args;
6178 args.location = &BTRFS_I(inode)->location;
6179 args.root = BTRFS_I(inode)->root;
6180
6181 return insert_inode_locked4(inode,
6182 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6183 btrfs_find_actor, &args);
6184 }
6185
6186 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6187 struct btrfs_root *root,
6188 struct inode *dir,
6189 const char *name, int name_len,
6190 u64 ref_objectid, u64 objectid,
6191 umode_t mode, u64 *index)
6192 {
6193 struct btrfs_fs_info *fs_info = root->fs_info;
6194 struct inode *inode;
6195 struct btrfs_inode_item *inode_item;
6196 struct btrfs_key *location;
6197 struct btrfs_path *path;
6198 struct btrfs_inode_ref *ref;
6199 struct btrfs_key key[2];
6200 u32 sizes[2];
6201 int nitems = name ? 2 : 1;
6202 unsigned long ptr;
6203 int ret;
6204
6205 path = btrfs_alloc_path();
6206 if (!path)
6207 return ERR_PTR(-ENOMEM);
6208
6209 inode = new_inode(fs_info->sb);
6210 if (!inode) {
6211 btrfs_free_path(path);
6212 return ERR_PTR(-ENOMEM);
6213 }
6214
6215 /*
6216 * O_TMPFILE, set link count to 0, so that after this point,
6217 * we fill in an inode item with the correct link count.
6218 */
6219 if (!name)
6220 set_nlink(inode, 0);
6221
6222 /*
6223 * we have to initialize this early, so we can reclaim the inode
6224 * number if we fail afterwards in this function.
6225 */
6226 inode->i_ino = objectid;
6227
6228 if (dir && name) {
6229 trace_btrfs_inode_request(dir);
6230
6231 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6232 if (ret) {
6233 btrfs_free_path(path);
6234 iput(inode);
6235 return ERR_PTR(ret);
6236 }
6237 } else if (dir) {
6238 *index = 0;
6239 }
6240 /*
6241 * index_cnt is ignored for everything but a dir,
6242 * btrfs_get_inode_index_count has an explanation for the magic
6243 * number
6244 */
6245 BTRFS_I(inode)->index_cnt = 2;
6246 BTRFS_I(inode)->dir_index = *index;
6247 BTRFS_I(inode)->root = root;
6248 BTRFS_I(inode)->generation = trans->transid;
6249 inode->i_generation = BTRFS_I(inode)->generation;
6250
6251 /*
6252 * We could have gotten an inode number from somebody who was fsynced
6253 * and then removed in this same transaction, so let's just set full
6254 * sync since it will be a full sync anyway and this will blow away the
6255 * old info in the log.
6256 */
6257 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6258
6259 key[0].objectid = objectid;
6260 key[0].type = BTRFS_INODE_ITEM_KEY;
6261 key[0].offset = 0;
6262
6263 sizes[0] = sizeof(struct btrfs_inode_item);
6264
6265 if (name) {
6266 /*
6267 * Start new inodes with an inode_ref. This is slightly more
6268 * efficient for small numbers of hard links since they will
6269 * be packed into one item. Extended refs will kick in if we
6270 * add more hard links than can fit in the ref item.
6271 */
6272 key[1].objectid = objectid;
6273 key[1].type = BTRFS_INODE_REF_KEY;
6274 key[1].offset = ref_objectid;
6275
6276 sizes[1] = name_len + sizeof(*ref);
6277 }
6278
6279 location = &BTRFS_I(inode)->location;
6280 location->objectid = objectid;
6281 location->offset = 0;
6282 location->type = BTRFS_INODE_ITEM_KEY;
6283
6284 ret = btrfs_insert_inode_locked(inode);
6285 if (ret < 0)
6286 goto fail;
6287
6288 path->leave_spinning = 1;
6289 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6290 if (ret != 0)
6291 goto fail_unlock;
6292
6293 inode_init_owner(inode, dir, mode);
6294 inode_set_bytes(inode, 0);
6295
6296 inode->i_mtime = current_time(inode);
6297 inode->i_atime = inode->i_mtime;
6298 inode->i_ctime = inode->i_mtime;
6299 BTRFS_I(inode)->i_otime = inode->i_mtime;
6300
6301 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6302 struct btrfs_inode_item);
6303 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6304 sizeof(*inode_item));
6305 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6306
6307 if (name) {
6308 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6309 struct btrfs_inode_ref);
6310 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6311 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6312 ptr = (unsigned long)(ref + 1);
6313 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6314 }
6315
6316 btrfs_mark_buffer_dirty(path->nodes[0]);
6317 btrfs_free_path(path);
6318
6319 btrfs_inherit_iflags(inode, dir);
6320
6321 if (S_ISREG(mode)) {
6322 if (btrfs_test_opt(fs_info, NODATASUM))
6323 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6324 if (btrfs_test_opt(fs_info, NODATACOW))
6325 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6326 BTRFS_INODE_NODATASUM;
6327 }
6328
6329 inode_tree_add(inode);
6330
6331 trace_btrfs_inode_new(inode);
6332 btrfs_set_inode_last_trans(trans, inode);
6333
6334 btrfs_update_root_times(trans, root);
6335
6336 ret = btrfs_inode_inherit_props(trans, inode, dir);
6337 if (ret)
6338 btrfs_err(fs_info,
6339 "error inheriting props for ino %llu (root %llu): %d",
6340 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6341
6342 return inode;
6343
6344 fail_unlock:
6345 unlock_new_inode(inode);
6346 fail:
6347 if (dir && name)
6348 BTRFS_I(dir)->index_cnt--;
6349 btrfs_free_path(path);
6350 iput(inode);
6351 return ERR_PTR(ret);
6352 }
6353
6354 static inline u8 btrfs_inode_type(struct inode *inode)
6355 {
6356 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6357 }
6358
6359 /*
6360 * utility function to add 'inode' into 'parent_inode' with
6361 * a give name and a given sequence number.
6362 * if 'add_backref' is true, also insert a backref from the
6363 * inode to the parent directory.
6364 */
6365 int btrfs_add_link(struct btrfs_trans_handle *trans,
6366 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6367 const char *name, int name_len, int add_backref, u64 index)
6368 {
6369 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6370 int ret = 0;
6371 struct btrfs_key key;
6372 struct btrfs_root *root = parent_inode->root;
6373 u64 ino = btrfs_ino(inode);
6374 u64 parent_ino = btrfs_ino(parent_inode);
6375
6376 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6377 memcpy(&key, &inode->root->root_key, sizeof(key));
6378 } else {
6379 key.objectid = ino;
6380 key.type = BTRFS_INODE_ITEM_KEY;
6381 key.offset = 0;
6382 }
6383
6384 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6385 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6386 root->root_key.objectid, parent_ino,
6387 index, name, name_len);
6388 } else if (add_backref) {
6389 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6390 parent_ino, index);
6391 }
6392
6393 /* Nothing to clean up yet */
6394 if (ret)
6395 return ret;
6396
6397 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6398 parent_inode, &key,
6399 btrfs_inode_type(&inode->vfs_inode), index);
6400 if (ret == -EEXIST || ret == -EOVERFLOW)
6401 goto fail_dir_item;
6402 else if (ret) {
6403 btrfs_abort_transaction(trans, ret);
6404 return ret;
6405 }
6406
6407 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6408 name_len * 2);
6409 inode_inc_iversion(&parent_inode->vfs_inode);
6410 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6411 current_time(&parent_inode->vfs_inode);
6412 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6413 if (ret)
6414 btrfs_abort_transaction(trans, ret);
6415 return ret;
6416
6417 fail_dir_item:
6418 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6419 u64 local_index;
6420 int err;
6421 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6422 root->root_key.objectid, parent_ino,
6423 &local_index, name, name_len);
6424
6425 } else if (add_backref) {
6426 u64 local_index;
6427 int err;
6428
6429 err = btrfs_del_inode_ref(trans, root, name, name_len,
6430 ino, parent_ino, &local_index);
6431 }
6432 return ret;
6433 }
6434
6435 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6436 struct btrfs_inode *dir, struct dentry *dentry,
6437 struct btrfs_inode *inode, int backref, u64 index)
6438 {
6439 int err = btrfs_add_link(trans, dir, inode,
6440 dentry->d_name.name, dentry->d_name.len,
6441 backref, index);
6442 if (err > 0)
6443 err = -EEXIST;
6444 return err;
6445 }
6446
6447 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6448 umode_t mode, dev_t rdev)
6449 {
6450 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6451 struct btrfs_trans_handle *trans;
6452 struct btrfs_root *root = BTRFS_I(dir)->root;
6453 struct inode *inode = NULL;
6454 int err;
6455 int drop_inode = 0;
6456 u64 objectid;
6457 u64 index = 0;
6458
6459 /*
6460 * 2 for inode item and ref
6461 * 2 for dir items
6462 * 1 for xattr if selinux is on
6463 */
6464 trans = btrfs_start_transaction(root, 5);
6465 if (IS_ERR(trans))
6466 return PTR_ERR(trans);
6467
6468 err = btrfs_find_free_ino(root, &objectid);
6469 if (err)
6470 goto out_unlock;
6471
6472 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6473 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6474 mode, &index);
6475 if (IS_ERR(inode)) {
6476 err = PTR_ERR(inode);
6477 goto out_unlock;
6478 }
6479
6480 /*
6481 * If the active LSM wants to access the inode during
6482 * d_instantiate it needs these. Smack checks to see
6483 * if the filesystem supports xattrs by looking at the
6484 * ops vector.
6485 */
6486 inode->i_op = &btrfs_special_inode_operations;
6487 init_special_inode(inode, inode->i_mode, rdev);
6488
6489 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6490 if (err)
6491 goto out_unlock_inode;
6492
6493 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6494 0, index);
6495 if (err) {
6496 goto out_unlock_inode;
6497 } else {
6498 btrfs_update_inode(trans, root, inode);
6499 unlock_new_inode(inode);
6500 d_instantiate(dentry, inode);
6501 }
6502
6503 out_unlock:
6504 btrfs_end_transaction(trans);
6505 btrfs_balance_delayed_items(fs_info);
6506 btrfs_btree_balance_dirty(fs_info);
6507 if (drop_inode) {
6508 inode_dec_link_count(inode);
6509 iput(inode);
6510 }
6511 return err;
6512
6513 out_unlock_inode:
6514 drop_inode = 1;
6515 unlock_new_inode(inode);
6516 goto out_unlock;
6517
6518 }
6519
6520 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6521 umode_t mode, bool excl)
6522 {
6523 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6524 struct btrfs_trans_handle *trans;
6525 struct btrfs_root *root = BTRFS_I(dir)->root;
6526 struct inode *inode = NULL;
6527 int drop_inode_on_err = 0;
6528 int err;
6529 u64 objectid;
6530 u64 index = 0;
6531
6532 /*
6533 * 2 for inode item and ref
6534 * 2 for dir items
6535 * 1 for xattr if selinux is on
6536 */
6537 trans = btrfs_start_transaction(root, 5);
6538 if (IS_ERR(trans))
6539 return PTR_ERR(trans);
6540
6541 err = btrfs_find_free_ino(root, &objectid);
6542 if (err)
6543 goto out_unlock;
6544
6545 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6546 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6547 mode, &index);
6548 if (IS_ERR(inode)) {
6549 err = PTR_ERR(inode);
6550 goto out_unlock;
6551 }
6552 drop_inode_on_err = 1;
6553 /*
6554 * If the active LSM wants to access the inode during
6555 * d_instantiate it needs these. Smack checks to see
6556 * if the filesystem supports xattrs by looking at the
6557 * ops vector.
6558 */
6559 inode->i_fop = &btrfs_file_operations;
6560 inode->i_op = &btrfs_file_inode_operations;
6561 inode->i_mapping->a_ops = &btrfs_aops;
6562
6563 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6564 if (err)
6565 goto out_unlock_inode;
6566
6567 err = btrfs_update_inode(trans, root, inode);
6568 if (err)
6569 goto out_unlock_inode;
6570
6571 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6572 0, index);
6573 if (err)
6574 goto out_unlock_inode;
6575
6576 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6577 unlock_new_inode(inode);
6578 d_instantiate(dentry, inode);
6579
6580 out_unlock:
6581 btrfs_end_transaction(trans);
6582 if (err && drop_inode_on_err) {
6583 inode_dec_link_count(inode);
6584 iput(inode);
6585 }
6586 btrfs_balance_delayed_items(fs_info);
6587 btrfs_btree_balance_dirty(fs_info);
6588 return err;
6589
6590 out_unlock_inode:
6591 unlock_new_inode(inode);
6592 goto out_unlock;
6593
6594 }
6595
6596 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6597 struct dentry *dentry)
6598 {
6599 struct btrfs_trans_handle *trans = NULL;
6600 struct btrfs_root *root = BTRFS_I(dir)->root;
6601 struct inode *inode = d_inode(old_dentry);
6602 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6603 u64 index;
6604 int err;
6605 int drop_inode = 0;
6606
6607 /* do not allow sys_link's with other subvols of the same device */
6608 if (root->objectid != BTRFS_I(inode)->root->objectid)
6609 return -EXDEV;
6610
6611 if (inode->i_nlink >= BTRFS_LINK_MAX)
6612 return -EMLINK;
6613
6614 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6615 if (err)
6616 goto fail;
6617
6618 /*
6619 * 2 items for inode and inode ref
6620 * 2 items for dir items
6621 * 1 item for parent inode
6622 */
6623 trans = btrfs_start_transaction(root, 5);
6624 if (IS_ERR(trans)) {
6625 err = PTR_ERR(trans);
6626 trans = NULL;
6627 goto fail;
6628 }
6629
6630 /* There are several dir indexes for this inode, clear the cache. */
6631 BTRFS_I(inode)->dir_index = 0ULL;
6632 inc_nlink(inode);
6633 inode_inc_iversion(inode);
6634 inode->i_ctime = current_time(inode);
6635 ihold(inode);
6636 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6637
6638 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6639 1, index);
6640
6641 if (err) {
6642 drop_inode = 1;
6643 } else {
6644 struct dentry *parent = dentry->d_parent;
6645 err = btrfs_update_inode(trans, root, inode);
6646 if (err)
6647 goto fail;
6648 if (inode->i_nlink == 1) {
6649 /*
6650 * If new hard link count is 1, it's a file created
6651 * with open(2) O_TMPFILE flag.
6652 */
6653 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6654 if (err)
6655 goto fail;
6656 }
6657 d_instantiate(dentry, inode);
6658 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6659 }
6660
6661 btrfs_balance_delayed_items(fs_info);
6662 fail:
6663 if (trans)
6664 btrfs_end_transaction(trans);
6665 if (drop_inode) {
6666 inode_dec_link_count(inode);
6667 iput(inode);
6668 }
6669 btrfs_btree_balance_dirty(fs_info);
6670 return err;
6671 }
6672
6673 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6674 {
6675 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6676 struct inode *inode = NULL;
6677 struct btrfs_trans_handle *trans;
6678 struct btrfs_root *root = BTRFS_I(dir)->root;
6679 int err = 0;
6680 int drop_on_err = 0;
6681 u64 objectid = 0;
6682 u64 index = 0;
6683
6684 /*
6685 * 2 items for inode and ref
6686 * 2 items for dir items
6687 * 1 for xattr if selinux is on
6688 */
6689 trans = btrfs_start_transaction(root, 5);
6690 if (IS_ERR(trans))
6691 return PTR_ERR(trans);
6692
6693 err = btrfs_find_free_ino(root, &objectid);
6694 if (err)
6695 goto out_fail;
6696
6697 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6698 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6699 S_IFDIR | mode, &index);
6700 if (IS_ERR(inode)) {
6701 err = PTR_ERR(inode);
6702 goto out_fail;
6703 }
6704
6705 drop_on_err = 1;
6706 /* these must be set before we unlock the inode */
6707 inode->i_op = &btrfs_dir_inode_operations;
6708 inode->i_fop = &btrfs_dir_file_operations;
6709
6710 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6711 if (err)
6712 goto out_fail_inode;
6713
6714 btrfs_i_size_write(BTRFS_I(inode), 0);
6715 err = btrfs_update_inode(trans, root, inode);
6716 if (err)
6717 goto out_fail_inode;
6718
6719 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6720 dentry->d_name.name,
6721 dentry->d_name.len, 0, index);
6722 if (err)
6723 goto out_fail_inode;
6724
6725 d_instantiate(dentry, inode);
6726 /*
6727 * mkdir is special. We're unlocking after we call d_instantiate
6728 * to avoid a race with nfsd calling d_instantiate.
6729 */
6730 unlock_new_inode(inode);
6731 drop_on_err = 0;
6732
6733 out_fail:
6734 btrfs_end_transaction(trans);
6735 if (drop_on_err) {
6736 inode_dec_link_count(inode);
6737 iput(inode);
6738 }
6739 btrfs_balance_delayed_items(fs_info);
6740 btrfs_btree_balance_dirty(fs_info);
6741 return err;
6742
6743 out_fail_inode:
6744 unlock_new_inode(inode);
6745 goto out_fail;
6746 }
6747
6748 /* Find next extent map of a given extent map, caller needs to ensure locks */
6749 static struct extent_map *next_extent_map(struct extent_map *em)
6750 {
6751 struct rb_node *next;
6752
6753 next = rb_next(&em->rb_node);
6754 if (!next)
6755 return NULL;
6756 return container_of(next, struct extent_map, rb_node);
6757 }
6758
6759 static struct extent_map *prev_extent_map(struct extent_map *em)
6760 {
6761 struct rb_node *prev;
6762
6763 prev = rb_prev(&em->rb_node);
6764 if (!prev)
6765 return NULL;
6766 return container_of(prev, struct extent_map, rb_node);
6767 }
6768
6769 /* helper for btfs_get_extent. Given an existing extent in the tree,
6770 * the existing extent is the nearest extent to map_start,
6771 * and an extent that you want to insert, deal with overlap and insert
6772 * the best fitted new extent into the tree.
6773 */
6774 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6775 struct extent_map *existing,
6776 struct extent_map *em,
6777 u64 map_start)
6778 {
6779 struct extent_map *prev;
6780 struct extent_map *next;
6781 u64 start;
6782 u64 end;
6783 u64 start_diff;
6784
6785 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6786
6787 if (existing->start > map_start) {
6788 next = existing;
6789 prev = prev_extent_map(next);
6790 } else {
6791 prev = existing;
6792 next = next_extent_map(prev);
6793 }
6794
6795 start = prev ? extent_map_end(prev) : em->start;
6796 start = max_t(u64, start, em->start);
6797 end = next ? next->start : extent_map_end(em);
6798 end = min_t(u64, end, extent_map_end(em));
6799 start_diff = start - em->start;
6800 em->start = start;
6801 em->len = end - start;
6802 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6803 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6804 em->block_start += start_diff;
6805 em->block_len -= start_diff;
6806 }
6807 return add_extent_mapping(em_tree, em, 0);
6808 }
6809
6810 static noinline int uncompress_inline(struct btrfs_path *path,
6811 struct page *page,
6812 size_t pg_offset, u64 extent_offset,
6813 struct btrfs_file_extent_item *item)
6814 {
6815 int ret;
6816 struct extent_buffer *leaf = path->nodes[0];
6817 char *tmp;
6818 size_t max_size;
6819 unsigned long inline_size;
6820 unsigned long ptr;
6821 int compress_type;
6822
6823 WARN_ON(pg_offset != 0);
6824 compress_type = btrfs_file_extent_compression(leaf, item);
6825 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6826 inline_size = btrfs_file_extent_inline_item_len(leaf,
6827 btrfs_item_nr(path->slots[0]));
6828 tmp = kmalloc(inline_size, GFP_NOFS);
6829 if (!tmp)
6830 return -ENOMEM;
6831 ptr = btrfs_file_extent_inline_start(item);
6832
6833 read_extent_buffer(leaf, tmp, ptr, inline_size);
6834
6835 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6836 ret = btrfs_decompress(compress_type, tmp, page,
6837 extent_offset, inline_size, max_size);
6838
6839 /*
6840 * decompression code contains a memset to fill in any space between the end
6841 * of the uncompressed data and the end of max_size in case the decompressed
6842 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6843 * the end of an inline extent and the beginning of the next block, so we
6844 * cover that region here.
6845 */
6846
6847 if (max_size + pg_offset < PAGE_SIZE) {
6848 char *map = kmap(page);
6849 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6850 kunmap(page);
6851 }
6852 kfree(tmp);
6853 return ret;
6854 }
6855
6856 /*
6857 * a bit scary, this does extent mapping from logical file offset to the disk.
6858 * the ugly parts come from merging extents from the disk with the in-ram
6859 * representation. This gets more complex because of the data=ordered code,
6860 * where the in-ram extents might be locked pending data=ordered completion.
6861 *
6862 * This also copies inline extents directly into the page.
6863 */
6864 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6865 struct page *page,
6866 size_t pg_offset, u64 start, u64 len,
6867 int create)
6868 {
6869 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6870 int ret;
6871 int err = 0;
6872 u64 extent_start = 0;
6873 u64 extent_end = 0;
6874 u64 objectid = btrfs_ino(inode);
6875 u32 found_type;
6876 struct btrfs_path *path = NULL;
6877 struct btrfs_root *root = inode->root;
6878 struct btrfs_file_extent_item *item;
6879 struct extent_buffer *leaf;
6880 struct btrfs_key found_key;
6881 struct extent_map *em = NULL;
6882 struct extent_map_tree *em_tree = &inode->extent_tree;
6883 struct extent_io_tree *io_tree = &inode->io_tree;
6884 struct btrfs_trans_handle *trans = NULL;
6885 const bool new_inline = !page || create;
6886
6887 again:
6888 read_lock(&em_tree->lock);
6889 em = lookup_extent_mapping(em_tree, start, len);
6890 if (em)
6891 em->bdev = fs_info->fs_devices->latest_bdev;
6892 read_unlock(&em_tree->lock);
6893
6894 if (em) {
6895 if (em->start > start || em->start + em->len <= start)
6896 free_extent_map(em);
6897 else if (em->block_start == EXTENT_MAP_INLINE && page)
6898 free_extent_map(em);
6899 else
6900 goto out;
6901 }
6902 em = alloc_extent_map();
6903 if (!em) {
6904 err = -ENOMEM;
6905 goto out;
6906 }
6907 em->bdev = fs_info->fs_devices->latest_bdev;
6908 em->start = EXTENT_MAP_HOLE;
6909 em->orig_start = EXTENT_MAP_HOLE;
6910 em->len = (u64)-1;
6911 em->block_len = (u64)-1;
6912
6913 if (!path) {
6914 path = btrfs_alloc_path();
6915 if (!path) {
6916 err = -ENOMEM;
6917 goto out;
6918 }
6919 /*
6920 * Chances are we'll be called again, so go ahead and do
6921 * readahead
6922 */
6923 path->reada = READA_FORWARD;
6924 }
6925
6926 ret = btrfs_lookup_file_extent(trans, root, path,
6927 objectid, start, trans != NULL);
6928 if (ret < 0) {
6929 err = ret;
6930 goto out;
6931 }
6932
6933 if (ret != 0) {
6934 if (path->slots[0] == 0)
6935 goto not_found;
6936 path->slots[0]--;
6937 }
6938
6939 leaf = path->nodes[0];
6940 item = btrfs_item_ptr(leaf, path->slots[0],
6941 struct btrfs_file_extent_item);
6942 /* are we inside the extent that was found? */
6943 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6944 found_type = found_key.type;
6945 if (found_key.objectid != objectid ||
6946 found_type != BTRFS_EXTENT_DATA_KEY) {
6947 /*
6948 * If we backup past the first extent we want to move forward
6949 * and see if there is an extent in front of us, otherwise we'll
6950 * say there is a hole for our whole search range which can
6951 * cause problems.
6952 */
6953 extent_end = start;
6954 goto next;
6955 }
6956
6957 found_type = btrfs_file_extent_type(leaf, item);
6958 extent_start = found_key.offset;
6959 if (found_type == BTRFS_FILE_EXTENT_REG ||
6960 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6961 extent_end = extent_start +
6962 btrfs_file_extent_num_bytes(leaf, item);
6963
6964 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6965 extent_start);
6966 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6967 size_t size;
6968 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6969 extent_end = ALIGN(extent_start + size,
6970 fs_info->sectorsize);
6971
6972 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6973 path->slots[0],
6974 extent_start);
6975 }
6976 next:
6977 if (start >= extent_end) {
6978 path->slots[0]++;
6979 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6980 ret = btrfs_next_leaf(root, path);
6981 if (ret < 0) {
6982 err = ret;
6983 goto out;
6984 }
6985 if (ret > 0)
6986 goto not_found;
6987 leaf = path->nodes[0];
6988 }
6989 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6990 if (found_key.objectid != objectid ||
6991 found_key.type != BTRFS_EXTENT_DATA_KEY)
6992 goto not_found;
6993 if (start + len <= found_key.offset)
6994 goto not_found;
6995 if (start > found_key.offset)
6996 goto next;
6997 em->start = start;
6998 em->orig_start = start;
6999 em->len = found_key.offset - start;
7000 goto not_found_em;
7001 }
7002
7003 btrfs_extent_item_to_extent_map(inode, path, item,
7004 new_inline, em);
7005
7006 if (found_type == BTRFS_FILE_EXTENT_REG ||
7007 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7008 goto insert;
7009 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7010 unsigned long ptr;
7011 char *map;
7012 size_t size;
7013 size_t extent_offset;
7014 size_t copy_size;
7015
7016 if (new_inline)
7017 goto out;
7018
7019 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7020 extent_offset = page_offset(page) + pg_offset - extent_start;
7021 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7022 size - extent_offset);
7023 em->start = extent_start + extent_offset;
7024 em->len = ALIGN(copy_size, fs_info->sectorsize);
7025 em->orig_block_len = em->len;
7026 em->orig_start = em->start;
7027 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7028 if (create == 0 && !PageUptodate(page)) {
7029 if (btrfs_file_extent_compression(leaf, item) !=
7030 BTRFS_COMPRESS_NONE) {
7031 ret = uncompress_inline(path, page, pg_offset,
7032 extent_offset, item);
7033 if (ret) {
7034 err = ret;
7035 goto out;
7036 }
7037 } else {
7038 map = kmap(page);
7039 read_extent_buffer(leaf, map + pg_offset, ptr,
7040 copy_size);
7041 if (pg_offset + copy_size < PAGE_SIZE) {
7042 memset(map + pg_offset + copy_size, 0,
7043 PAGE_SIZE - pg_offset -
7044 copy_size);
7045 }
7046 kunmap(page);
7047 }
7048 flush_dcache_page(page);
7049 } else if (create && PageUptodate(page)) {
7050 BUG();
7051 if (!trans) {
7052 kunmap(page);
7053 free_extent_map(em);
7054 em = NULL;
7055
7056 btrfs_release_path(path);
7057 trans = btrfs_join_transaction(root);
7058
7059 if (IS_ERR(trans))
7060 return ERR_CAST(trans);
7061 goto again;
7062 }
7063 map = kmap(page);
7064 write_extent_buffer(leaf, map + pg_offset, ptr,
7065 copy_size);
7066 kunmap(page);
7067 btrfs_mark_buffer_dirty(leaf);
7068 }
7069 set_extent_uptodate(io_tree, em->start,
7070 extent_map_end(em) - 1, NULL, GFP_NOFS);
7071 goto insert;
7072 }
7073 not_found:
7074 em->start = start;
7075 em->orig_start = start;
7076 em->len = len;
7077 not_found_em:
7078 em->block_start = EXTENT_MAP_HOLE;
7079 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
7080 insert:
7081 btrfs_release_path(path);
7082 if (em->start > start || extent_map_end(em) <= start) {
7083 btrfs_err(fs_info,
7084 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7085 em->start, em->len, start, len);
7086 err = -EIO;
7087 goto out;
7088 }
7089
7090 err = 0;
7091 write_lock(&em_tree->lock);
7092 ret = add_extent_mapping(em_tree, em, 0);
7093 /* it is possible that someone inserted the extent into the tree
7094 * while we had the lock dropped. It is also possible that
7095 * an overlapping map exists in the tree
7096 */
7097 if (ret == -EEXIST) {
7098 struct extent_map *existing;
7099
7100 ret = 0;
7101
7102 existing = search_extent_mapping(em_tree, start, len);
7103 /*
7104 * existing will always be non-NULL, since there must be
7105 * extent causing the -EEXIST.
7106 */
7107 if (existing->start == em->start &&
7108 extent_map_end(existing) >= extent_map_end(em) &&
7109 em->block_start == existing->block_start) {
7110 /*
7111 * The existing extent map already encompasses the
7112 * entire extent map we tried to add.
7113 */
7114 free_extent_map(em);
7115 em = existing;
7116 err = 0;
7117
7118 } else if (start >= extent_map_end(existing) ||
7119 start <= existing->start) {
7120 /*
7121 * The existing extent map is the one nearest to
7122 * the [start, start + len) range which overlaps
7123 */
7124 err = merge_extent_mapping(em_tree, existing,
7125 em, start);
7126 free_extent_map(existing);
7127 if (err) {
7128 free_extent_map(em);
7129 em = NULL;
7130 }
7131 } else {
7132 free_extent_map(em);
7133 em = existing;
7134 err = 0;
7135 }
7136 }
7137 write_unlock(&em_tree->lock);
7138 out:
7139
7140 trace_btrfs_get_extent(root, inode, em);
7141
7142 btrfs_free_path(path);
7143 if (trans) {
7144 ret = btrfs_end_transaction(trans);
7145 if (!err)
7146 err = ret;
7147 }
7148 if (err) {
7149 free_extent_map(em);
7150 return ERR_PTR(err);
7151 }
7152 BUG_ON(!em); /* Error is always set */
7153 return em;
7154 }
7155
7156 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7157 struct page *page,
7158 size_t pg_offset, u64 start, u64 len,
7159 int create)
7160 {
7161 struct extent_map *em;
7162 struct extent_map *hole_em = NULL;
7163 u64 range_start = start;
7164 u64 end;
7165 u64 found;
7166 u64 found_end;
7167 int err = 0;
7168
7169 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7170 if (IS_ERR(em))
7171 return em;
7172 /*
7173 * If our em maps to:
7174 * - a hole or
7175 * - a pre-alloc extent,
7176 * there might actually be delalloc bytes behind it.
7177 */
7178 if (em->block_start != EXTENT_MAP_HOLE &&
7179 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7180 return em;
7181 else
7182 hole_em = em;
7183
7184 /* check to see if we've wrapped (len == -1 or similar) */
7185 end = start + len;
7186 if (end < start)
7187 end = (u64)-1;
7188 else
7189 end -= 1;
7190
7191 em = NULL;
7192
7193 /* ok, we didn't find anything, lets look for delalloc */
7194 found = count_range_bits(&inode->io_tree, &range_start,
7195 end, len, EXTENT_DELALLOC, 1);
7196 found_end = range_start + found;
7197 if (found_end < range_start)
7198 found_end = (u64)-1;
7199
7200 /*
7201 * we didn't find anything useful, return
7202 * the original results from get_extent()
7203 */
7204 if (range_start > end || found_end <= start) {
7205 em = hole_em;
7206 hole_em = NULL;
7207 goto out;
7208 }
7209
7210 /* adjust the range_start to make sure it doesn't
7211 * go backwards from the start they passed in
7212 */
7213 range_start = max(start, range_start);
7214 found = found_end - range_start;
7215
7216 if (found > 0) {
7217 u64 hole_start = start;
7218 u64 hole_len = len;
7219
7220 em = alloc_extent_map();
7221 if (!em) {
7222 err = -ENOMEM;
7223 goto out;
7224 }
7225 /*
7226 * when btrfs_get_extent can't find anything it
7227 * returns one huge hole
7228 *
7229 * make sure what it found really fits our range, and
7230 * adjust to make sure it is based on the start from
7231 * the caller
7232 */
7233 if (hole_em) {
7234 u64 calc_end = extent_map_end(hole_em);
7235
7236 if (calc_end <= start || (hole_em->start > end)) {
7237 free_extent_map(hole_em);
7238 hole_em = NULL;
7239 } else {
7240 hole_start = max(hole_em->start, start);
7241 hole_len = calc_end - hole_start;
7242 }
7243 }
7244 em->bdev = NULL;
7245 if (hole_em && range_start > hole_start) {
7246 /* our hole starts before our delalloc, so we
7247 * have to return just the parts of the hole
7248 * that go until the delalloc starts
7249 */
7250 em->len = min(hole_len,
7251 range_start - hole_start);
7252 em->start = hole_start;
7253 em->orig_start = hole_start;
7254 /*
7255 * don't adjust block start at all,
7256 * it is fixed at EXTENT_MAP_HOLE
7257 */
7258 em->block_start = hole_em->block_start;
7259 em->block_len = hole_len;
7260 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7261 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7262 } else {
7263 em->start = range_start;
7264 em->len = found;
7265 em->orig_start = range_start;
7266 em->block_start = EXTENT_MAP_DELALLOC;
7267 em->block_len = found;
7268 }
7269 } else if (hole_em) {
7270 return hole_em;
7271 }
7272 out:
7273
7274 free_extent_map(hole_em);
7275 if (err) {
7276 free_extent_map(em);
7277 return ERR_PTR(err);
7278 }
7279 return em;
7280 }
7281
7282 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7283 const u64 start,
7284 const u64 len,
7285 const u64 orig_start,
7286 const u64 block_start,
7287 const u64 block_len,
7288 const u64 orig_block_len,
7289 const u64 ram_bytes,
7290 const int type)
7291 {
7292 struct extent_map *em = NULL;
7293 int ret;
7294
7295 if (type != BTRFS_ORDERED_NOCOW) {
7296 em = create_io_em(inode, start, len, orig_start,
7297 block_start, block_len, orig_block_len,
7298 ram_bytes,
7299 BTRFS_COMPRESS_NONE, /* compress_type */
7300 type);
7301 if (IS_ERR(em))
7302 goto out;
7303 }
7304 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7305 len, block_len, type);
7306 if (ret) {
7307 if (em) {
7308 free_extent_map(em);
7309 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7310 start + len - 1, 0);
7311 }
7312 em = ERR_PTR(ret);
7313 }
7314 out:
7315
7316 return em;
7317 }
7318
7319 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7320 u64 start, u64 len)
7321 {
7322 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7323 struct btrfs_root *root = BTRFS_I(inode)->root;
7324 struct extent_map *em;
7325 struct btrfs_key ins;
7326 u64 alloc_hint;
7327 int ret;
7328
7329 alloc_hint = get_extent_allocation_hint(inode, start, len);
7330 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7331 0, alloc_hint, &ins, 1, 1);
7332 if (ret)
7333 return ERR_PTR(ret);
7334
7335 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7336 ins.objectid, ins.offset, ins.offset,
7337 ins.offset, BTRFS_ORDERED_REGULAR);
7338 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7339 if (IS_ERR(em))
7340 btrfs_free_reserved_extent(fs_info, ins.objectid,
7341 ins.offset, 1);
7342
7343 return em;
7344 }
7345
7346 /*
7347 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7348 * block must be cow'd
7349 */
7350 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7351 u64 *orig_start, u64 *orig_block_len,
7352 u64 *ram_bytes)
7353 {
7354 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7355 struct btrfs_path *path;
7356 int ret;
7357 struct extent_buffer *leaf;
7358 struct btrfs_root *root = BTRFS_I(inode)->root;
7359 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7360 struct btrfs_file_extent_item *fi;
7361 struct btrfs_key key;
7362 u64 disk_bytenr;
7363 u64 backref_offset;
7364 u64 extent_end;
7365 u64 num_bytes;
7366 int slot;
7367 int found_type;
7368 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7369
7370 path = btrfs_alloc_path();
7371 if (!path)
7372 return -ENOMEM;
7373
7374 ret = btrfs_lookup_file_extent(NULL, root, path,
7375 btrfs_ino(BTRFS_I(inode)), offset, 0);
7376 if (ret < 0)
7377 goto out;
7378
7379 slot = path->slots[0];
7380 if (ret == 1) {
7381 if (slot == 0) {
7382 /* can't find the item, must cow */
7383 ret = 0;
7384 goto out;
7385 }
7386 slot--;
7387 }
7388 ret = 0;
7389 leaf = path->nodes[0];
7390 btrfs_item_key_to_cpu(leaf, &key, slot);
7391 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7392 key.type != BTRFS_EXTENT_DATA_KEY) {
7393 /* not our file or wrong item type, must cow */
7394 goto out;
7395 }
7396
7397 if (key.offset > offset) {
7398 /* Wrong offset, must cow */
7399 goto out;
7400 }
7401
7402 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7403 found_type = btrfs_file_extent_type(leaf, fi);
7404 if (found_type != BTRFS_FILE_EXTENT_REG &&
7405 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7406 /* not a regular extent, must cow */
7407 goto out;
7408 }
7409
7410 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7411 goto out;
7412
7413 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7414 if (extent_end <= offset)
7415 goto out;
7416
7417 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7418 if (disk_bytenr == 0)
7419 goto out;
7420
7421 if (btrfs_file_extent_compression(leaf, fi) ||
7422 btrfs_file_extent_encryption(leaf, fi) ||
7423 btrfs_file_extent_other_encoding(leaf, fi))
7424 goto out;
7425
7426 backref_offset = btrfs_file_extent_offset(leaf, fi);
7427
7428 if (orig_start) {
7429 *orig_start = key.offset - backref_offset;
7430 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7431 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7432 }
7433
7434 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7435 goto out;
7436
7437 num_bytes = min(offset + *len, extent_end) - offset;
7438 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7439 u64 range_end;
7440
7441 range_end = round_up(offset + num_bytes,
7442 root->fs_info->sectorsize) - 1;
7443 ret = test_range_bit(io_tree, offset, range_end,
7444 EXTENT_DELALLOC, 0, NULL);
7445 if (ret) {
7446 ret = -EAGAIN;
7447 goto out;
7448 }
7449 }
7450
7451 btrfs_release_path(path);
7452
7453 /*
7454 * look for other files referencing this extent, if we
7455 * find any we must cow
7456 */
7457
7458 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7459 key.offset - backref_offset, disk_bytenr);
7460 if (ret) {
7461 ret = 0;
7462 goto out;
7463 }
7464
7465 /*
7466 * adjust disk_bytenr and num_bytes to cover just the bytes
7467 * in this extent we are about to write. If there
7468 * are any csums in that range we have to cow in order
7469 * to keep the csums correct
7470 */
7471 disk_bytenr += backref_offset;
7472 disk_bytenr += offset - key.offset;
7473 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7474 goto out;
7475 /*
7476 * all of the above have passed, it is safe to overwrite this extent
7477 * without cow
7478 */
7479 *len = num_bytes;
7480 ret = 1;
7481 out:
7482 btrfs_free_path(path);
7483 return ret;
7484 }
7485
7486 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7487 {
7488 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7489 bool found = false;
7490 void **pagep = NULL;
7491 struct page *page = NULL;
7492 unsigned long start_idx;
7493 unsigned long end_idx;
7494
7495 start_idx = start >> PAGE_SHIFT;
7496
7497 /*
7498 * end is the last byte in the last page. end == start is legal
7499 */
7500 end_idx = end >> PAGE_SHIFT;
7501
7502 rcu_read_lock();
7503
7504 /* Most of the code in this while loop is lifted from
7505 * find_get_page. It's been modified to begin searching from a
7506 * page and return just the first page found in that range. If the
7507 * found idx is less than or equal to the end idx then we know that
7508 * a page exists. If no pages are found or if those pages are
7509 * outside of the range then we're fine (yay!) */
7510 while (page == NULL &&
7511 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7512 page = radix_tree_deref_slot(pagep);
7513 if (unlikely(!page))
7514 break;
7515
7516 if (radix_tree_exception(page)) {
7517 if (radix_tree_deref_retry(page)) {
7518 page = NULL;
7519 continue;
7520 }
7521 /*
7522 * Otherwise, shmem/tmpfs must be storing a swap entry
7523 * here as an exceptional entry: so return it without
7524 * attempting to raise page count.
7525 */
7526 page = NULL;
7527 break; /* TODO: Is this relevant for this use case? */
7528 }
7529
7530 if (!page_cache_get_speculative(page)) {
7531 page = NULL;
7532 continue;
7533 }
7534
7535 /*
7536 * Has the page moved?
7537 * This is part of the lockless pagecache protocol. See
7538 * include/linux/pagemap.h for details.
7539 */
7540 if (unlikely(page != *pagep)) {
7541 put_page(page);
7542 page = NULL;
7543 }
7544 }
7545
7546 if (page) {
7547 if (page->index <= end_idx)
7548 found = true;
7549 put_page(page);
7550 }
7551
7552 rcu_read_unlock();
7553 return found;
7554 }
7555
7556 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7557 struct extent_state **cached_state, int writing)
7558 {
7559 struct btrfs_ordered_extent *ordered;
7560 int ret = 0;
7561
7562 while (1) {
7563 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7564 cached_state);
7565 /*
7566 * We're concerned with the entire range that we're going to be
7567 * doing DIO to, so we need to make sure there's no ordered
7568 * extents in this range.
7569 */
7570 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7571 lockend - lockstart + 1);
7572
7573 /*
7574 * We need to make sure there are no buffered pages in this
7575 * range either, we could have raced between the invalidate in
7576 * generic_file_direct_write and locking the extent. The
7577 * invalidate needs to happen so that reads after a write do not
7578 * get stale data.
7579 */
7580 if (!ordered &&
7581 (!writing ||
7582 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7583 break;
7584
7585 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7586 cached_state, GFP_NOFS);
7587
7588 if (ordered) {
7589 /*
7590 * If we are doing a DIO read and the ordered extent we
7591 * found is for a buffered write, we can not wait for it
7592 * to complete and retry, because if we do so we can
7593 * deadlock with concurrent buffered writes on page
7594 * locks. This happens only if our DIO read covers more
7595 * than one extent map, if at this point has already
7596 * created an ordered extent for a previous extent map
7597 * and locked its range in the inode's io tree, and a
7598 * concurrent write against that previous extent map's
7599 * range and this range started (we unlock the ranges
7600 * in the io tree only when the bios complete and
7601 * buffered writes always lock pages before attempting
7602 * to lock range in the io tree).
7603 */
7604 if (writing ||
7605 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7606 btrfs_start_ordered_extent(inode, ordered, 1);
7607 else
7608 ret = -ENOTBLK;
7609 btrfs_put_ordered_extent(ordered);
7610 } else {
7611 /*
7612 * We could trigger writeback for this range (and wait
7613 * for it to complete) and then invalidate the pages for
7614 * this range (through invalidate_inode_pages2_range()),
7615 * but that can lead us to a deadlock with a concurrent
7616 * call to readpages() (a buffered read or a defrag call
7617 * triggered a readahead) on a page lock due to an
7618 * ordered dio extent we created before but did not have
7619 * yet a corresponding bio submitted (whence it can not
7620 * complete), which makes readpages() wait for that
7621 * ordered extent to complete while holding a lock on
7622 * that page.
7623 */
7624 ret = -ENOTBLK;
7625 }
7626
7627 if (ret)
7628 break;
7629
7630 cond_resched();
7631 }
7632
7633 return ret;
7634 }
7635
7636 /* The callers of this must take lock_extent() */
7637 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7638 u64 orig_start, u64 block_start,
7639 u64 block_len, u64 orig_block_len,
7640 u64 ram_bytes, int compress_type,
7641 int type)
7642 {
7643 struct extent_map_tree *em_tree;
7644 struct extent_map *em;
7645 struct btrfs_root *root = BTRFS_I(inode)->root;
7646 int ret;
7647
7648 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7649 type == BTRFS_ORDERED_COMPRESSED ||
7650 type == BTRFS_ORDERED_NOCOW ||
7651 type == BTRFS_ORDERED_REGULAR);
7652
7653 em_tree = &BTRFS_I(inode)->extent_tree;
7654 em = alloc_extent_map();
7655 if (!em)
7656 return ERR_PTR(-ENOMEM);
7657
7658 em->start = start;
7659 em->orig_start = orig_start;
7660 em->len = len;
7661 em->block_len = block_len;
7662 em->block_start = block_start;
7663 em->bdev = root->fs_info->fs_devices->latest_bdev;
7664 em->orig_block_len = orig_block_len;
7665 em->ram_bytes = ram_bytes;
7666 em->generation = -1;
7667 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7668 if (type == BTRFS_ORDERED_PREALLOC) {
7669 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7670 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7671 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7672 em->compress_type = compress_type;
7673 }
7674
7675 do {
7676 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7677 em->start + em->len - 1, 0);
7678 write_lock(&em_tree->lock);
7679 ret = add_extent_mapping(em_tree, em, 1);
7680 write_unlock(&em_tree->lock);
7681 /*
7682 * The caller has taken lock_extent(), who could race with us
7683 * to add em?
7684 */
7685 } while (ret == -EEXIST);
7686
7687 if (ret) {
7688 free_extent_map(em);
7689 return ERR_PTR(ret);
7690 }
7691
7692 /* em got 2 refs now, callers needs to do free_extent_map once. */
7693 return em;
7694 }
7695
7696 static void adjust_dio_outstanding_extents(struct inode *inode,
7697 struct btrfs_dio_data *dio_data,
7698 const u64 len)
7699 {
7700 unsigned num_extents = count_max_extents(len);
7701
7702 /*
7703 * If we have an outstanding_extents count still set then we're
7704 * within our reservation, otherwise we need to adjust our inode
7705 * counter appropriately.
7706 */
7707 if (dio_data->outstanding_extents >= num_extents) {
7708 dio_data->outstanding_extents -= num_extents;
7709 } else {
7710 /*
7711 * If dio write length has been split due to no large enough
7712 * contiguous space, we need to compensate our inode counter
7713 * appropriately.
7714 */
7715 u64 num_needed = num_extents - dio_data->outstanding_extents;
7716
7717 spin_lock(&BTRFS_I(inode)->lock);
7718 BTRFS_I(inode)->outstanding_extents += num_needed;
7719 spin_unlock(&BTRFS_I(inode)->lock);
7720 }
7721 }
7722
7723 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7724 struct buffer_head *bh_result, int create)
7725 {
7726 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7727 struct extent_map *em;
7728 struct extent_state *cached_state = NULL;
7729 struct btrfs_dio_data *dio_data = NULL;
7730 u64 start = iblock << inode->i_blkbits;
7731 u64 lockstart, lockend;
7732 u64 len = bh_result->b_size;
7733 int unlock_bits = EXTENT_LOCKED;
7734 int ret = 0;
7735
7736 if (create)
7737 unlock_bits |= EXTENT_DIRTY;
7738 else
7739 len = min_t(u64, len, fs_info->sectorsize);
7740
7741 lockstart = start;
7742 lockend = start + len - 1;
7743
7744 if (current->journal_info) {
7745 /*
7746 * Need to pull our outstanding extents and set journal_info to NULL so
7747 * that anything that needs to check if there's a transaction doesn't get
7748 * confused.
7749 */
7750 dio_data = current->journal_info;
7751 current->journal_info = NULL;
7752 }
7753
7754 /*
7755 * If this errors out it's because we couldn't invalidate pagecache for
7756 * this range and we need to fallback to buffered.
7757 */
7758 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7759 create)) {
7760 ret = -ENOTBLK;
7761 goto err;
7762 }
7763
7764 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7765 if (IS_ERR(em)) {
7766 ret = PTR_ERR(em);
7767 goto unlock_err;
7768 }
7769
7770 /*
7771 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7772 * io. INLINE is special, and we could probably kludge it in here, but
7773 * it's still buffered so for safety lets just fall back to the generic
7774 * buffered path.
7775 *
7776 * For COMPRESSED we _have_ to read the entire extent in so we can
7777 * decompress it, so there will be buffering required no matter what we
7778 * do, so go ahead and fallback to buffered.
7779 *
7780 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7781 * to buffered IO. Don't blame me, this is the price we pay for using
7782 * the generic code.
7783 */
7784 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7785 em->block_start == EXTENT_MAP_INLINE) {
7786 free_extent_map(em);
7787 ret = -ENOTBLK;
7788 goto unlock_err;
7789 }
7790
7791 /* Just a good old fashioned hole, return */
7792 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7793 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7794 free_extent_map(em);
7795 goto unlock_err;
7796 }
7797
7798 /*
7799 * We don't allocate a new extent in the following cases
7800 *
7801 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7802 * existing extent.
7803 * 2) The extent is marked as PREALLOC. We're good to go here and can
7804 * just use the extent.
7805 *
7806 */
7807 if (!create) {
7808 len = min(len, em->len - (start - em->start));
7809 lockstart = start + len;
7810 goto unlock;
7811 }
7812
7813 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7814 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7815 em->block_start != EXTENT_MAP_HOLE)) {
7816 int type;
7817 u64 block_start, orig_start, orig_block_len, ram_bytes;
7818
7819 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7820 type = BTRFS_ORDERED_PREALLOC;
7821 else
7822 type = BTRFS_ORDERED_NOCOW;
7823 len = min(len, em->len - (start - em->start));
7824 block_start = em->block_start + (start - em->start);
7825
7826 if (can_nocow_extent(inode, start, &len, &orig_start,
7827 &orig_block_len, &ram_bytes) == 1 &&
7828 btrfs_inc_nocow_writers(fs_info, block_start)) {
7829 struct extent_map *em2;
7830
7831 em2 = btrfs_create_dio_extent(inode, start, len,
7832 orig_start, block_start,
7833 len, orig_block_len,
7834 ram_bytes, type);
7835 btrfs_dec_nocow_writers(fs_info, block_start);
7836 if (type == BTRFS_ORDERED_PREALLOC) {
7837 free_extent_map(em);
7838 em = em2;
7839 }
7840 if (em2 && IS_ERR(em2)) {
7841 ret = PTR_ERR(em2);
7842 goto unlock_err;
7843 }
7844 /*
7845 * For inode marked NODATACOW or extent marked PREALLOC,
7846 * use the existing or preallocated extent, so does not
7847 * need to adjust btrfs_space_info's bytes_may_use.
7848 */
7849 btrfs_free_reserved_data_space_noquota(inode,
7850 start, len);
7851 goto unlock;
7852 }
7853 }
7854
7855 /*
7856 * this will cow the extent, reset the len in case we changed
7857 * it above
7858 */
7859 len = bh_result->b_size;
7860 free_extent_map(em);
7861 em = btrfs_new_extent_direct(inode, start, len);
7862 if (IS_ERR(em)) {
7863 ret = PTR_ERR(em);
7864 goto unlock_err;
7865 }
7866 len = min(len, em->len - (start - em->start));
7867 unlock:
7868 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7869 inode->i_blkbits;
7870 bh_result->b_size = len;
7871 bh_result->b_bdev = em->bdev;
7872 set_buffer_mapped(bh_result);
7873 if (create) {
7874 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7875 set_buffer_new(bh_result);
7876
7877 /*
7878 * Need to update the i_size under the extent lock so buffered
7879 * readers will get the updated i_size when we unlock.
7880 */
7881 if (!dio_data->overwrite && start + len > i_size_read(inode))
7882 i_size_write(inode, start + len);
7883
7884 adjust_dio_outstanding_extents(inode, dio_data, len);
7885 WARN_ON(dio_data->reserve < len);
7886 dio_data->reserve -= len;
7887 dio_data->unsubmitted_oe_range_end = start + len;
7888 current->journal_info = dio_data;
7889 }
7890
7891 /*
7892 * In the case of write we need to clear and unlock the entire range,
7893 * in the case of read we need to unlock only the end area that we
7894 * aren't using if there is any left over space.
7895 */
7896 if (lockstart < lockend) {
7897 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7898 lockend, unlock_bits, 1, 0,
7899 &cached_state, GFP_NOFS);
7900 } else {
7901 free_extent_state(cached_state);
7902 }
7903
7904 free_extent_map(em);
7905
7906 return 0;
7907
7908 unlock_err:
7909 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7910 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7911 err:
7912 if (dio_data)
7913 current->journal_info = dio_data;
7914 /*
7915 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7916 * write less data then expected, so that we don't underflow our inode's
7917 * outstanding extents counter.
7918 */
7919 if (create && dio_data)
7920 adjust_dio_outstanding_extents(inode, dio_data, len);
7921
7922 return ret;
7923 }
7924
7925 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7926 int mirror_num)
7927 {
7928 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7929 int ret;
7930
7931 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7932
7933 bio_get(bio);
7934
7935 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7936 if (ret)
7937 goto err;
7938
7939 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7940 err:
7941 bio_put(bio);
7942 return ret;
7943 }
7944
7945 static int btrfs_check_dio_repairable(struct inode *inode,
7946 struct bio *failed_bio,
7947 struct io_failure_record *failrec,
7948 int failed_mirror)
7949 {
7950 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7951 int num_copies;
7952
7953 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7954 if (num_copies == 1) {
7955 /*
7956 * we only have a single copy of the data, so don't bother with
7957 * all the retry and error correction code that follows. no
7958 * matter what the error is, it is very likely to persist.
7959 */
7960 btrfs_debug(fs_info,
7961 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7962 num_copies, failrec->this_mirror, failed_mirror);
7963 return 0;
7964 }
7965
7966 failrec->failed_mirror = failed_mirror;
7967 failrec->this_mirror++;
7968 if (failrec->this_mirror == failed_mirror)
7969 failrec->this_mirror++;
7970
7971 if (failrec->this_mirror > num_copies) {
7972 btrfs_debug(fs_info,
7973 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7974 num_copies, failrec->this_mirror, failed_mirror);
7975 return 0;
7976 }
7977
7978 return 1;
7979 }
7980
7981 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7982 struct page *page, unsigned int pgoff,
7983 u64 start, u64 end, int failed_mirror,
7984 bio_end_io_t *repair_endio, void *repair_arg)
7985 {
7986 struct io_failure_record *failrec;
7987 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7988 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7989 struct bio *bio;
7990 int isector;
7991 int read_mode = 0;
7992 int segs;
7993 int ret;
7994
7995 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7996
7997 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7998 if (ret)
7999 return ret;
8000
8001 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
8002 failed_mirror);
8003 if (!ret) {
8004 free_io_failure(failure_tree, io_tree, failrec);
8005 return -EIO;
8006 }
8007
8008 segs = bio_segments(failed_bio);
8009 if (segs > 1 ||
8010 (failed_bio->bi_io_vec->bv_len > btrfs_inode_sectorsize(inode)))
8011 read_mode |= REQ_FAILFAST_DEV;
8012
8013 isector = start - btrfs_io_bio(failed_bio)->logical;
8014 isector >>= inode->i_sb->s_blocksize_bits;
8015 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
8016 pgoff, isector, repair_endio, repair_arg);
8017 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
8018
8019 btrfs_debug(BTRFS_I(inode)->root->fs_info,
8020 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
8021 read_mode, failrec->this_mirror, failrec->in_validation);
8022
8023 ret = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
8024 if (ret) {
8025 free_io_failure(failure_tree, io_tree, failrec);
8026 bio_put(bio);
8027 }
8028
8029 return ret;
8030 }
8031
8032 struct btrfs_retry_complete {
8033 struct completion done;
8034 struct inode *inode;
8035 u64 start;
8036 int uptodate;
8037 };
8038
8039 static void btrfs_retry_endio_nocsum(struct bio *bio)
8040 {
8041 struct btrfs_retry_complete *done = bio->bi_private;
8042 struct inode *inode = done->inode;
8043 struct bio_vec *bvec;
8044 struct extent_io_tree *io_tree, *failure_tree;
8045 int i;
8046
8047 if (bio->bi_status)
8048 goto end;
8049
8050 ASSERT(bio->bi_vcnt == 1);
8051 io_tree = &BTRFS_I(inode)->io_tree;
8052 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8053 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
8054
8055 done->uptodate = 1;
8056 ASSERT(!bio_flagged(bio, BIO_CLONED));
8057 bio_for_each_segment_all(bvec, bio, i)
8058 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
8059 io_tree, done->start, bvec->bv_page,
8060 btrfs_ino(BTRFS_I(inode)), 0);
8061 end:
8062 complete(&done->done);
8063 bio_put(bio);
8064 }
8065
8066 static int __btrfs_correct_data_nocsum(struct inode *inode,
8067 struct btrfs_io_bio *io_bio)
8068 {
8069 struct btrfs_fs_info *fs_info;
8070 struct bio_vec bvec;
8071 struct bvec_iter iter;
8072 struct btrfs_retry_complete done;
8073 u64 start;
8074 unsigned int pgoff;
8075 u32 sectorsize;
8076 int nr_sectors;
8077 int ret;
8078 int err = 0;
8079
8080 fs_info = BTRFS_I(inode)->root->fs_info;
8081 sectorsize = fs_info->sectorsize;
8082
8083 start = io_bio->logical;
8084 done.inode = inode;
8085 io_bio->bio.bi_iter = io_bio->iter;
8086
8087 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8088 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8089 pgoff = bvec.bv_offset;
8090
8091 next_block_or_try_again:
8092 done.uptodate = 0;
8093 done.start = start;
8094 init_completion(&done.done);
8095
8096 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8097 pgoff, start, start + sectorsize - 1,
8098 io_bio->mirror_num,
8099 btrfs_retry_endio_nocsum, &done);
8100 if (ret) {
8101 err = ret;
8102 goto next;
8103 }
8104
8105 wait_for_completion(&done.done);
8106
8107 if (!done.uptodate) {
8108 /* We might have another mirror, so try again */
8109 goto next_block_or_try_again;
8110 }
8111
8112 next:
8113 start += sectorsize;
8114
8115 nr_sectors--;
8116 if (nr_sectors) {
8117 pgoff += sectorsize;
8118 ASSERT(pgoff < PAGE_SIZE);
8119 goto next_block_or_try_again;
8120 }
8121 }
8122
8123 return err;
8124 }
8125
8126 static void btrfs_retry_endio(struct bio *bio)
8127 {
8128 struct btrfs_retry_complete *done = bio->bi_private;
8129 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8130 struct extent_io_tree *io_tree, *failure_tree;
8131 struct inode *inode = done->inode;
8132 struct bio_vec *bvec;
8133 int uptodate;
8134 int ret;
8135 int i;
8136
8137 if (bio->bi_status)
8138 goto end;
8139
8140 uptodate = 1;
8141
8142 ASSERT(bio->bi_vcnt == 1);
8143 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(done->inode));
8144
8145 io_tree = &BTRFS_I(inode)->io_tree;
8146 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8147
8148 ASSERT(!bio_flagged(bio, BIO_CLONED));
8149 bio_for_each_segment_all(bvec, bio, i) {
8150 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8151 bvec->bv_offset, done->start,
8152 bvec->bv_len);
8153 if (!ret)
8154 clean_io_failure(BTRFS_I(inode)->root->fs_info,
8155 failure_tree, io_tree, done->start,
8156 bvec->bv_page,
8157 btrfs_ino(BTRFS_I(inode)),
8158 bvec->bv_offset);
8159 else
8160 uptodate = 0;
8161 }
8162
8163 done->uptodate = uptodate;
8164 end:
8165 complete(&done->done);
8166 bio_put(bio);
8167 }
8168
8169 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8170 struct btrfs_io_bio *io_bio, blk_status_t err)
8171 {
8172 struct btrfs_fs_info *fs_info;
8173 struct bio_vec bvec;
8174 struct bvec_iter iter;
8175 struct btrfs_retry_complete done;
8176 u64 start;
8177 u64 offset = 0;
8178 u32 sectorsize;
8179 int nr_sectors;
8180 unsigned int pgoff;
8181 int csum_pos;
8182 bool uptodate = (err == 0);
8183 int ret;
8184
8185 fs_info = BTRFS_I(inode)->root->fs_info;
8186 sectorsize = fs_info->sectorsize;
8187
8188 err = 0;
8189 start = io_bio->logical;
8190 done.inode = inode;
8191 io_bio->bio.bi_iter = io_bio->iter;
8192
8193 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8194 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8195
8196 pgoff = bvec.bv_offset;
8197 next_block:
8198 if (uptodate) {
8199 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8200 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8201 bvec.bv_page, pgoff, start, sectorsize);
8202 if (likely(!ret))
8203 goto next;
8204 }
8205 try_again:
8206 done.uptodate = 0;
8207 done.start = start;
8208 init_completion(&done.done);
8209
8210 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8211 pgoff, start, start + sectorsize - 1,
8212 io_bio->mirror_num,
8213 btrfs_retry_endio, &done);
8214 if (ret) {
8215 err = errno_to_blk_status(ret);
8216 goto next;
8217 }
8218
8219 wait_for_completion(&done.done);
8220
8221 if (!done.uptodate) {
8222 /* We might have another mirror, so try again */
8223 goto try_again;
8224 }
8225 next:
8226 offset += sectorsize;
8227 start += sectorsize;
8228
8229 ASSERT(nr_sectors);
8230
8231 nr_sectors--;
8232 if (nr_sectors) {
8233 pgoff += sectorsize;
8234 ASSERT(pgoff < PAGE_SIZE);
8235 goto next_block;
8236 }
8237 }
8238
8239 return err;
8240 }
8241
8242 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8243 struct btrfs_io_bio *io_bio, blk_status_t err)
8244 {
8245 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8246
8247 if (skip_csum) {
8248 if (unlikely(err))
8249 return __btrfs_correct_data_nocsum(inode, io_bio);
8250 else
8251 return 0;
8252 } else {
8253 return __btrfs_subio_endio_read(inode, io_bio, err);
8254 }
8255 }
8256
8257 static void btrfs_endio_direct_read(struct bio *bio)
8258 {
8259 struct btrfs_dio_private *dip = bio->bi_private;
8260 struct inode *inode = dip->inode;
8261 struct bio *dio_bio;
8262 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8263 blk_status_t err = bio->bi_status;
8264
8265 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) {
8266 err = btrfs_subio_endio_read(inode, io_bio, err);
8267 if (!err)
8268 bio->bi_status = 0;
8269 }
8270
8271 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8272 dip->logical_offset + dip->bytes - 1);
8273 dio_bio = dip->dio_bio;
8274
8275 kfree(dip);
8276
8277 dio_bio->bi_status = bio->bi_status;
8278 dio_end_io(dio_bio);
8279
8280 if (io_bio->end_io)
8281 io_bio->end_io(io_bio, blk_status_to_errno(err));
8282 bio_put(bio);
8283 }
8284
8285 static void __endio_write_update_ordered(struct inode *inode,
8286 const u64 offset, const u64 bytes,
8287 const bool uptodate)
8288 {
8289 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8290 struct btrfs_ordered_extent *ordered = NULL;
8291 struct btrfs_workqueue *wq;
8292 btrfs_work_func_t func;
8293 u64 ordered_offset = offset;
8294 u64 ordered_bytes = bytes;
8295 int ret;
8296
8297 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8298 wq = fs_info->endio_freespace_worker;
8299 func = btrfs_freespace_write_helper;
8300 } else {
8301 wq = fs_info->endio_write_workers;
8302 func = btrfs_endio_write_helper;
8303 }
8304
8305 again:
8306 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8307 &ordered_offset,
8308 ordered_bytes,
8309 uptodate);
8310 if (!ret)
8311 goto out_test;
8312
8313 btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL);
8314 btrfs_queue_work(wq, &ordered->work);
8315 out_test:
8316 /*
8317 * our bio might span multiple ordered extents. If we haven't
8318 * completed the accounting for the whole dio, go back and try again
8319 */
8320 if (ordered_offset < offset + bytes) {
8321 ordered_bytes = offset + bytes - ordered_offset;
8322 ordered = NULL;
8323 goto again;
8324 }
8325 }
8326
8327 static void btrfs_endio_direct_write(struct bio *bio)
8328 {
8329 struct btrfs_dio_private *dip = bio->bi_private;
8330 struct bio *dio_bio = dip->dio_bio;
8331
8332 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8333 dip->bytes, !bio->bi_status);
8334
8335 kfree(dip);
8336
8337 dio_bio->bi_status = bio->bi_status;
8338 dio_end_io(dio_bio);
8339 bio_put(bio);
8340 }
8341
8342 static blk_status_t __btrfs_submit_bio_start_direct_io(void *private_data,
8343 struct bio *bio, int mirror_num,
8344 unsigned long bio_flags, u64 offset)
8345 {
8346 struct inode *inode = private_data;
8347 blk_status_t ret;
8348 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8349 BUG_ON(ret); /* -ENOMEM */
8350 return 0;
8351 }
8352
8353 static void btrfs_end_dio_bio(struct bio *bio)
8354 {
8355 struct btrfs_dio_private *dip = bio->bi_private;
8356 blk_status_t err = bio->bi_status;
8357
8358 if (err)
8359 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8360 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8361 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8362 bio->bi_opf,
8363 (unsigned long long)bio->bi_iter.bi_sector,
8364 bio->bi_iter.bi_size, err);
8365
8366 if (dip->subio_endio)
8367 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8368
8369 if (err) {
8370 dip->errors = 1;
8371
8372 /*
8373 * before atomic variable goto zero, we must make sure
8374 * dip->errors is perceived to be set.
8375 */
8376 smp_mb__before_atomic();
8377 }
8378
8379 /* if there are more bios still pending for this dio, just exit */
8380 if (!atomic_dec_and_test(&dip->pending_bios))
8381 goto out;
8382
8383 if (dip->errors) {
8384 bio_io_error(dip->orig_bio);
8385 } else {
8386 dip->dio_bio->bi_status = 0;
8387 bio_endio(dip->orig_bio);
8388 }
8389 out:
8390 bio_put(bio);
8391 }
8392
8393 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8394 struct btrfs_dio_private *dip,
8395 struct bio *bio,
8396 u64 file_offset)
8397 {
8398 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8399 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8400 blk_status_t ret;
8401
8402 /*
8403 * We load all the csum data we need when we submit
8404 * the first bio to reduce the csum tree search and
8405 * contention.
8406 */
8407 if (dip->logical_offset == file_offset) {
8408 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8409 file_offset);
8410 if (ret)
8411 return ret;
8412 }
8413
8414 if (bio == dip->orig_bio)
8415 return 0;
8416
8417 file_offset -= dip->logical_offset;
8418 file_offset >>= inode->i_sb->s_blocksize_bits;
8419 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8420
8421 return 0;
8422 }
8423
8424 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8425 u64 file_offset, int skip_sum,
8426 int async_submit)
8427 {
8428 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8429 struct btrfs_dio_private *dip = bio->bi_private;
8430 bool write = bio_op(bio) == REQ_OP_WRITE;
8431 blk_status_t ret;
8432
8433 if (async_submit)
8434 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8435
8436 bio_get(bio);
8437
8438 if (!write) {
8439 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8440 if (ret)
8441 goto err;
8442 }
8443
8444 if (skip_sum)
8445 goto map;
8446
8447 if (write && async_submit) {
8448 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8449 file_offset, inode,
8450 __btrfs_submit_bio_start_direct_io,
8451 __btrfs_submit_bio_done);
8452 goto err;
8453 } else if (write) {
8454 /*
8455 * If we aren't doing async submit, calculate the csum of the
8456 * bio now.
8457 */
8458 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8459 if (ret)
8460 goto err;
8461 } else {
8462 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8463 file_offset);
8464 if (ret)
8465 goto err;
8466 }
8467 map:
8468 ret = btrfs_map_bio(fs_info, bio, 0, async_submit);
8469 err:
8470 bio_put(bio);
8471 return ret;
8472 }
8473
8474 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip,
8475 int skip_sum)
8476 {
8477 struct inode *inode = dip->inode;
8478 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8479 struct bio *bio;
8480 struct bio *orig_bio = dip->orig_bio;
8481 u64 start_sector = orig_bio->bi_iter.bi_sector;
8482 u64 file_offset = dip->logical_offset;
8483 u64 map_length;
8484 int async_submit = 0;
8485 u64 submit_len;
8486 int clone_offset = 0;
8487 int clone_len;
8488 int ret;
8489
8490 map_length = orig_bio->bi_iter.bi_size;
8491 submit_len = map_length;
8492 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8493 &map_length, NULL, 0);
8494 if (ret)
8495 return -EIO;
8496
8497 if (map_length >= submit_len) {
8498 bio = orig_bio;
8499 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8500 goto submit;
8501 }
8502
8503 /* async crcs make it difficult to collect full stripe writes. */
8504 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8505 async_submit = 0;
8506 else
8507 async_submit = 1;
8508
8509 /* bio split */
8510 ASSERT(map_length <= INT_MAX);
8511 atomic_inc(&dip->pending_bios);
8512 do {
8513 clone_len = min_t(int, submit_len, map_length);
8514
8515 /*
8516 * This will never fail as it's passing GPF_NOFS and
8517 * the allocation is backed by btrfs_bioset.
8518 */
8519 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8520 clone_len);
8521 bio->bi_private = dip;
8522 bio->bi_end_io = btrfs_end_dio_bio;
8523 btrfs_io_bio(bio)->logical = file_offset;
8524
8525 ASSERT(submit_len >= clone_len);
8526 submit_len -= clone_len;
8527 if (submit_len == 0)
8528 break;
8529
8530 /*
8531 * Increase the count before we submit the bio so we know
8532 * the end IO handler won't happen before we increase the
8533 * count. Otherwise, the dip might get freed before we're
8534 * done setting it up.
8535 */
8536 atomic_inc(&dip->pending_bios);
8537
8538 ret = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8539 async_submit);
8540 if (ret) {
8541 bio_put(bio);
8542 atomic_dec(&dip->pending_bios);
8543 goto out_err;
8544 }
8545
8546 clone_offset += clone_len;
8547 start_sector += clone_len >> 9;
8548 file_offset += clone_len;
8549
8550 map_length = submit_len;
8551 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8552 start_sector << 9, &map_length, NULL, 0);
8553 if (ret)
8554 goto out_err;
8555 } while (submit_len > 0);
8556
8557 submit:
8558 ret = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8559 async_submit);
8560 if (!ret)
8561 return 0;
8562
8563 bio_put(bio);
8564 out_err:
8565 dip->errors = 1;
8566 /*
8567 * before atomic variable goto zero, we must
8568 * make sure dip->errors is perceived to be set.
8569 */
8570 smp_mb__before_atomic();
8571 if (atomic_dec_and_test(&dip->pending_bios))
8572 bio_io_error(dip->orig_bio);
8573
8574 /* bio_end_io() will handle error, so we needn't return it */
8575 return 0;
8576 }
8577
8578 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8579 loff_t file_offset)
8580 {
8581 struct btrfs_dio_private *dip = NULL;
8582 struct bio *bio = NULL;
8583 struct btrfs_io_bio *io_bio;
8584 int skip_sum;
8585 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8586 int ret = 0;
8587
8588 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8589
8590 bio = btrfs_bio_clone(dio_bio);
8591
8592 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8593 if (!dip) {
8594 ret = -ENOMEM;
8595 goto free_ordered;
8596 }
8597
8598 dip->private = dio_bio->bi_private;
8599 dip->inode = inode;
8600 dip->logical_offset = file_offset;
8601 dip->bytes = dio_bio->bi_iter.bi_size;
8602 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8603 bio->bi_private = dip;
8604 dip->orig_bio = bio;
8605 dip->dio_bio = dio_bio;
8606 atomic_set(&dip->pending_bios, 0);
8607 io_bio = btrfs_io_bio(bio);
8608 io_bio->logical = file_offset;
8609
8610 if (write) {
8611 bio->bi_end_io = btrfs_endio_direct_write;
8612 } else {
8613 bio->bi_end_io = btrfs_endio_direct_read;
8614 dip->subio_endio = btrfs_subio_endio_read;
8615 }
8616
8617 /*
8618 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8619 * even if we fail to submit a bio, because in such case we do the
8620 * corresponding error handling below and it must not be done a second
8621 * time by btrfs_direct_IO().
8622 */
8623 if (write) {
8624 struct btrfs_dio_data *dio_data = current->journal_info;
8625
8626 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8627 dip->bytes;
8628 dio_data->unsubmitted_oe_range_start =
8629 dio_data->unsubmitted_oe_range_end;
8630 }
8631
8632 ret = btrfs_submit_direct_hook(dip, skip_sum);
8633 if (!ret)
8634 return;
8635
8636 if (io_bio->end_io)
8637 io_bio->end_io(io_bio, ret);
8638
8639 free_ordered:
8640 /*
8641 * If we arrived here it means either we failed to submit the dip
8642 * or we either failed to clone the dio_bio or failed to allocate the
8643 * dip. If we cloned the dio_bio and allocated the dip, we can just
8644 * call bio_endio against our io_bio so that we get proper resource
8645 * cleanup if we fail to submit the dip, otherwise, we must do the
8646 * same as btrfs_endio_direct_[write|read] because we can't call these
8647 * callbacks - they require an allocated dip and a clone of dio_bio.
8648 */
8649 if (bio && dip) {
8650 bio_io_error(bio);
8651 /*
8652 * The end io callbacks free our dip, do the final put on bio
8653 * and all the cleanup and final put for dio_bio (through
8654 * dio_end_io()).
8655 */
8656 dip = NULL;
8657 bio = NULL;
8658 } else {
8659 if (write)
8660 __endio_write_update_ordered(inode,
8661 file_offset,
8662 dio_bio->bi_iter.bi_size,
8663 false);
8664 else
8665 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8666 file_offset + dio_bio->bi_iter.bi_size - 1);
8667
8668 dio_bio->bi_status = BLK_STS_IOERR;
8669 /*
8670 * Releases and cleans up our dio_bio, no need to bio_put()
8671 * nor bio_endio()/bio_io_error() against dio_bio.
8672 */
8673 dio_end_io(dio_bio);
8674 }
8675 if (bio)
8676 bio_put(bio);
8677 kfree(dip);
8678 }
8679
8680 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8681 struct kiocb *iocb,
8682 const struct iov_iter *iter, loff_t offset)
8683 {
8684 int seg;
8685 int i;
8686 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8687 ssize_t retval = -EINVAL;
8688
8689 if (offset & blocksize_mask)
8690 goto out;
8691
8692 if (iov_iter_alignment(iter) & blocksize_mask)
8693 goto out;
8694
8695 /* If this is a write we don't need to check anymore */
8696 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8697 return 0;
8698 /*
8699 * Check to make sure we don't have duplicate iov_base's in this
8700 * iovec, if so return EINVAL, otherwise we'll get csum errors
8701 * when reading back.
8702 */
8703 for (seg = 0; seg < iter->nr_segs; seg++) {
8704 for (i = seg + 1; i < iter->nr_segs; i++) {
8705 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8706 goto out;
8707 }
8708 }
8709 retval = 0;
8710 out:
8711 return retval;
8712 }
8713
8714 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8715 {
8716 struct file *file = iocb->ki_filp;
8717 struct inode *inode = file->f_mapping->host;
8718 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8719 struct btrfs_dio_data dio_data = { 0 };
8720 struct extent_changeset *data_reserved = NULL;
8721 loff_t offset = iocb->ki_pos;
8722 size_t count = 0;
8723 int flags = 0;
8724 bool wakeup = true;
8725 bool relock = false;
8726 ssize_t ret;
8727
8728 if (check_direct_IO(fs_info, iocb, iter, offset))
8729 return 0;
8730
8731 inode_dio_begin(inode);
8732 smp_mb__after_atomic();
8733
8734 /*
8735 * The generic stuff only does filemap_write_and_wait_range, which
8736 * isn't enough if we've written compressed pages to this area, so
8737 * we need to flush the dirty pages again to make absolutely sure
8738 * that any outstanding dirty pages are on disk.
8739 */
8740 count = iov_iter_count(iter);
8741 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8742 &BTRFS_I(inode)->runtime_flags))
8743 filemap_fdatawrite_range(inode->i_mapping, offset,
8744 offset + count - 1);
8745
8746 if (iov_iter_rw(iter) == WRITE) {
8747 /*
8748 * If the write DIO is beyond the EOF, we need update
8749 * the isize, but it is protected by i_mutex. So we can
8750 * not unlock the i_mutex at this case.
8751 */
8752 if (offset + count <= inode->i_size) {
8753 dio_data.overwrite = 1;
8754 inode_unlock(inode);
8755 relock = true;
8756 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8757 ret = -EAGAIN;
8758 goto out;
8759 }
8760 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8761 offset, count);
8762 if (ret)
8763 goto out;
8764 dio_data.outstanding_extents = count_max_extents(count);
8765
8766 /*
8767 * We need to know how many extents we reserved so that we can
8768 * do the accounting properly if we go over the number we
8769 * originally calculated. Abuse current->journal_info for this.
8770 */
8771 dio_data.reserve = round_up(count,
8772 fs_info->sectorsize);
8773 dio_data.unsubmitted_oe_range_start = (u64)offset;
8774 dio_data.unsubmitted_oe_range_end = (u64)offset;
8775 current->journal_info = &dio_data;
8776 down_read(&BTRFS_I(inode)->dio_sem);
8777 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8778 &BTRFS_I(inode)->runtime_flags)) {
8779 inode_dio_end(inode);
8780 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8781 wakeup = false;
8782 }
8783
8784 ret = __blockdev_direct_IO(iocb, inode,
8785 fs_info->fs_devices->latest_bdev,
8786 iter, btrfs_get_blocks_direct, NULL,
8787 btrfs_submit_direct, flags);
8788 if (iov_iter_rw(iter) == WRITE) {
8789 up_read(&BTRFS_I(inode)->dio_sem);
8790 current->journal_info = NULL;
8791 if (ret < 0 && ret != -EIOCBQUEUED) {
8792 if (dio_data.reserve)
8793 btrfs_delalloc_release_space(inode, data_reserved,
8794 offset, dio_data.reserve);
8795 /*
8796 * On error we might have left some ordered extents
8797 * without submitting corresponding bios for them, so
8798 * cleanup them up to avoid other tasks getting them
8799 * and waiting for them to complete forever.
8800 */
8801 if (dio_data.unsubmitted_oe_range_start <
8802 dio_data.unsubmitted_oe_range_end)
8803 __endio_write_update_ordered(inode,
8804 dio_data.unsubmitted_oe_range_start,
8805 dio_data.unsubmitted_oe_range_end -
8806 dio_data.unsubmitted_oe_range_start,
8807 false);
8808 } else if (ret >= 0 && (size_t)ret < count)
8809 btrfs_delalloc_release_space(inode, data_reserved,
8810 offset, count - (size_t)ret);
8811 }
8812 out:
8813 if (wakeup)
8814 inode_dio_end(inode);
8815 if (relock)
8816 inode_lock(inode);
8817
8818 extent_changeset_free(data_reserved);
8819 return ret;
8820 }
8821
8822 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8823
8824 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8825 __u64 start, __u64 len)
8826 {
8827 int ret;
8828
8829 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8830 if (ret)
8831 return ret;
8832
8833 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8834 }
8835
8836 int btrfs_readpage(struct file *file, struct page *page)
8837 {
8838 struct extent_io_tree *tree;
8839 tree = &BTRFS_I(page->mapping->host)->io_tree;
8840 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8841 }
8842
8843 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8844 {
8845 struct extent_io_tree *tree;
8846 struct inode *inode = page->mapping->host;
8847 int ret;
8848
8849 if (current->flags & PF_MEMALLOC) {
8850 redirty_page_for_writepage(wbc, page);
8851 unlock_page(page);
8852 return 0;
8853 }
8854
8855 /*
8856 * If we are under memory pressure we will call this directly from the
8857 * VM, we need to make sure we have the inode referenced for the ordered
8858 * extent. If not just return like we didn't do anything.
8859 */
8860 if (!igrab(inode)) {
8861 redirty_page_for_writepage(wbc, page);
8862 return AOP_WRITEPAGE_ACTIVATE;
8863 }
8864 tree = &BTRFS_I(page->mapping->host)->io_tree;
8865 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8866 btrfs_add_delayed_iput(inode);
8867 return ret;
8868 }
8869
8870 static int btrfs_writepages(struct address_space *mapping,
8871 struct writeback_control *wbc)
8872 {
8873 struct extent_io_tree *tree;
8874
8875 tree = &BTRFS_I(mapping->host)->io_tree;
8876 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8877 }
8878
8879 static int
8880 btrfs_readpages(struct file *file, struct address_space *mapping,
8881 struct list_head *pages, unsigned nr_pages)
8882 {
8883 struct extent_io_tree *tree;
8884 tree = &BTRFS_I(mapping->host)->io_tree;
8885 return extent_readpages(tree, mapping, pages, nr_pages,
8886 btrfs_get_extent);
8887 }
8888 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8889 {
8890 struct extent_io_tree *tree;
8891 struct extent_map_tree *map;
8892 int ret;
8893
8894 tree = &BTRFS_I(page->mapping->host)->io_tree;
8895 map = &BTRFS_I(page->mapping->host)->extent_tree;
8896 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8897 if (ret == 1) {
8898 ClearPagePrivate(page);
8899 set_page_private(page, 0);
8900 put_page(page);
8901 }
8902 return ret;
8903 }
8904
8905 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8906 {
8907 if (PageWriteback(page) || PageDirty(page))
8908 return 0;
8909 return __btrfs_releasepage(page, gfp_flags);
8910 }
8911
8912 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8913 unsigned int length)
8914 {
8915 struct inode *inode = page->mapping->host;
8916 struct extent_io_tree *tree;
8917 struct btrfs_ordered_extent *ordered;
8918 struct extent_state *cached_state = NULL;
8919 u64 page_start = page_offset(page);
8920 u64 page_end = page_start + PAGE_SIZE - 1;
8921 u64 start;
8922 u64 end;
8923 int inode_evicting = inode->i_state & I_FREEING;
8924
8925 /*
8926 * we have the page locked, so new writeback can't start,
8927 * and the dirty bit won't be cleared while we are here.
8928 *
8929 * Wait for IO on this page so that we can safely clear
8930 * the PagePrivate2 bit and do ordered accounting
8931 */
8932 wait_on_page_writeback(page);
8933
8934 tree = &BTRFS_I(inode)->io_tree;
8935 if (offset) {
8936 btrfs_releasepage(page, GFP_NOFS);
8937 return;
8938 }
8939
8940 if (!inode_evicting)
8941 lock_extent_bits(tree, page_start, page_end, &cached_state);
8942 again:
8943 start = page_start;
8944 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8945 page_end - start + 1);
8946 if (ordered) {
8947 end = min(page_end, ordered->file_offset + ordered->len - 1);
8948 /*
8949 * IO on this page will never be started, so we need
8950 * to account for any ordered extents now
8951 */
8952 if (!inode_evicting)
8953 clear_extent_bit(tree, start, end,
8954 EXTENT_DIRTY | EXTENT_DELALLOC |
8955 EXTENT_DELALLOC_NEW |
8956 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8957 EXTENT_DEFRAG, 1, 0, &cached_state,
8958 GFP_NOFS);
8959 /*
8960 * whoever cleared the private bit is responsible
8961 * for the finish_ordered_io
8962 */
8963 if (TestClearPagePrivate2(page)) {
8964 struct btrfs_ordered_inode_tree *tree;
8965 u64 new_len;
8966
8967 tree = &BTRFS_I(inode)->ordered_tree;
8968
8969 spin_lock_irq(&tree->lock);
8970 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8971 new_len = start - ordered->file_offset;
8972 if (new_len < ordered->truncated_len)
8973 ordered->truncated_len = new_len;
8974 spin_unlock_irq(&tree->lock);
8975
8976 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8977 start,
8978 end - start + 1, 1))
8979 btrfs_finish_ordered_io(ordered);
8980 }
8981 btrfs_put_ordered_extent(ordered);
8982 if (!inode_evicting) {
8983 cached_state = NULL;
8984 lock_extent_bits(tree, start, end,
8985 &cached_state);
8986 }
8987
8988 start = end + 1;
8989 if (start < page_end)
8990 goto again;
8991 }
8992
8993 /*
8994 * Qgroup reserved space handler
8995 * Page here will be either
8996 * 1) Already written to disk
8997 * In this case, its reserved space is released from data rsv map
8998 * and will be freed by delayed_ref handler finally.
8999 * So even we call qgroup_free_data(), it won't decrease reserved
9000 * space.
9001 * 2) Not written to disk
9002 * This means the reserved space should be freed here. However,
9003 * if a truncate invalidates the page (by clearing PageDirty)
9004 * and the page is accounted for while allocating extent
9005 * in btrfs_check_data_free_space() we let delayed_ref to
9006 * free the entire extent.
9007 */
9008 if (PageDirty(page))
9009 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
9010 if (!inode_evicting) {
9011 clear_extent_bit(tree, page_start, page_end,
9012 EXTENT_LOCKED | EXTENT_DIRTY |
9013 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
9014 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
9015 &cached_state, GFP_NOFS);
9016
9017 __btrfs_releasepage(page, GFP_NOFS);
9018 }
9019
9020 ClearPageChecked(page);
9021 if (PagePrivate(page)) {
9022 ClearPagePrivate(page);
9023 set_page_private(page, 0);
9024 put_page(page);
9025 }
9026 }
9027
9028 /*
9029 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
9030 * called from a page fault handler when a page is first dirtied. Hence we must
9031 * be careful to check for EOF conditions here. We set the page up correctly
9032 * for a written page which means we get ENOSPC checking when writing into
9033 * holes and correct delalloc and unwritten extent mapping on filesystems that
9034 * support these features.
9035 *
9036 * We are not allowed to take the i_mutex here so we have to play games to
9037 * protect against truncate races as the page could now be beyond EOF. Because
9038 * vmtruncate() writes the inode size before removing pages, once we have the
9039 * page lock we can determine safely if the page is beyond EOF. If it is not
9040 * beyond EOF, then the page is guaranteed safe against truncation until we
9041 * unlock the page.
9042 */
9043 int btrfs_page_mkwrite(struct vm_fault *vmf)
9044 {
9045 struct page *page = vmf->page;
9046 struct inode *inode = file_inode(vmf->vma->vm_file);
9047 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9048 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
9049 struct btrfs_ordered_extent *ordered;
9050 struct extent_state *cached_state = NULL;
9051 struct extent_changeset *data_reserved = NULL;
9052 char *kaddr;
9053 unsigned long zero_start;
9054 loff_t size;
9055 int ret;
9056 int reserved = 0;
9057 u64 reserved_space;
9058 u64 page_start;
9059 u64 page_end;
9060 u64 end;
9061
9062 reserved_space = PAGE_SIZE;
9063
9064 sb_start_pagefault(inode->i_sb);
9065 page_start = page_offset(page);
9066 page_end = page_start + PAGE_SIZE - 1;
9067 end = page_end;
9068
9069 /*
9070 * Reserving delalloc space after obtaining the page lock can lead to
9071 * deadlock. For example, if a dirty page is locked by this function
9072 * and the call to btrfs_delalloc_reserve_space() ends up triggering
9073 * dirty page write out, then the btrfs_writepage() function could
9074 * end up waiting indefinitely to get a lock on the page currently
9075 * being processed by btrfs_page_mkwrite() function.
9076 */
9077 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
9078 reserved_space);
9079 if (!ret) {
9080 ret = file_update_time(vmf->vma->vm_file);
9081 reserved = 1;
9082 }
9083 if (ret) {
9084 if (ret == -ENOMEM)
9085 ret = VM_FAULT_OOM;
9086 else /* -ENOSPC, -EIO, etc */
9087 ret = VM_FAULT_SIGBUS;
9088 if (reserved)
9089 goto out;
9090 goto out_noreserve;
9091 }
9092
9093 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
9094 again:
9095 lock_page(page);
9096 size = i_size_read(inode);
9097
9098 if ((page->mapping != inode->i_mapping) ||
9099 (page_start >= size)) {
9100 /* page got truncated out from underneath us */
9101 goto out_unlock;
9102 }
9103 wait_on_page_writeback(page);
9104
9105 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
9106 set_page_extent_mapped(page);
9107
9108 /*
9109 * we can't set the delalloc bits if there are pending ordered
9110 * extents. Drop our locks and wait for them to finish
9111 */
9112 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
9113 PAGE_SIZE);
9114 if (ordered) {
9115 unlock_extent_cached(io_tree, page_start, page_end,
9116 &cached_state, GFP_NOFS);
9117 unlock_page(page);
9118 btrfs_start_ordered_extent(inode, ordered, 1);
9119 btrfs_put_ordered_extent(ordered);
9120 goto again;
9121 }
9122
9123 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
9124 reserved_space = round_up(size - page_start,
9125 fs_info->sectorsize);
9126 if (reserved_space < PAGE_SIZE) {
9127 end = page_start + reserved_space - 1;
9128 spin_lock(&BTRFS_I(inode)->lock);
9129 BTRFS_I(inode)->outstanding_extents++;
9130 spin_unlock(&BTRFS_I(inode)->lock);
9131 btrfs_delalloc_release_space(inode, data_reserved,
9132 page_start, PAGE_SIZE - reserved_space);
9133 }
9134 }
9135
9136 /*
9137 * page_mkwrite gets called when the page is firstly dirtied after it's
9138 * faulted in, but write(2) could also dirty a page and set delalloc
9139 * bits, thus in this case for space account reason, we still need to
9140 * clear any delalloc bits within this page range since we have to
9141 * reserve data&meta space before lock_page() (see above comments).
9142 */
9143 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9144 EXTENT_DIRTY | EXTENT_DELALLOC |
9145 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9146 0, 0, &cached_state, GFP_NOFS);
9147
9148 ret = btrfs_set_extent_delalloc(inode, page_start, end,
9149 &cached_state, 0);
9150 if (ret) {
9151 unlock_extent_cached(io_tree, page_start, page_end,
9152 &cached_state, GFP_NOFS);
9153 ret = VM_FAULT_SIGBUS;
9154 goto out_unlock;
9155 }
9156 ret = 0;
9157
9158 /* page is wholly or partially inside EOF */
9159 if (page_start + PAGE_SIZE > size)
9160 zero_start = size & ~PAGE_MASK;
9161 else
9162 zero_start = PAGE_SIZE;
9163
9164 if (zero_start != PAGE_SIZE) {
9165 kaddr = kmap(page);
9166 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9167 flush_dcache_page(page);
9168 kunmap(page);
9169 }
9170 ClearPageChecked(page);
9171 set_page_dirty(page);
9172 SetPageUptodate(page);
9173
9174 BTRFS_I(inode)->last_trans = fs_info->generation;
9175 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9176 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9177
9178 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9179
9180 out_unlock:
9181 if (!ret) {
9182 sb_end_pagefault(inode->i_sb);
9183 extent_changeset_free(data_reserved);
9184 return VM_FAULT_LOCKED;
9185 }
9186 unlock_page(page);
9187 out:
9188 btrfs_delalloc_release_space(inode, data_reserved, page_start,
9189 reserved_space);
9190 out_noreserve:
9191 sb_end_pagefault(inode->i_sb);
9192 extent_changeset_free(data_reserved);
9193 return ret;
9194 }
9195
9196 static int btrfs_truncate(struct inode *inode)
9197 {
9198 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9199 struct btrfs_root *root = BTRFS_I(inode)->root;
9200 struct btrfs_block_rsv *rsv;
9201 int ret = 0;
9202 int err = 0;
9203 struct btrfs_trans_handle *trans;
9204 u64 mask = fs_info->sectorsize - 1;
9205 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9206
9207 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9208 (u64)-1);
9209 if (ret)
9210 return ret;
9211
9212 /*
9213 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9214 * 3 things going on here
9215 *
9216 * 1) We need to reserve space for our orphan item and the space to
9217 * delete our orphan item. Lord knows we don't want to have a dangling
9218 * orphan item because we didn't reserve space to remove it.
9219 *
9220 * 2) We need to reserve space to update our inode.
9221 *
9222 * 3) We need to have something to cache all the space that is going to
9223 * be free'd up by the truncate operation, but also have some slack
9224 * space reserved in case it uses space during the truncate (thank you
9225 * very much snapshotting).
9226 *
9227 * And we need these to all be separate. The fact is we can use a lot of
9228 * space doing the truncate, and we have no earthly idea how much space
9229 * we will use, so we need the truncate reservation to be separate so it
9230 * doesn't end up using space reserved for updating the inode or
9231 * removing the orphan item. We also need to be able to stop the
9232 * transaction and start a new one, which means we need to be able to
9233 * update the inode several times, and we have no idea of knowing how
9234 * many times that will be, so we can't just reserve 1 item for the
9235 * entirety of the operation, so that has to be done separately as well.
9236 * Then there is the orphan item, which does indeed need to be held on
9237 * to for the whole operation, and we need nobody to touch this reserved
9238 * space except the orphan code.
9239 *
9240 * So that leaves us with
9241 *
9242 * 1) root->orphan_block_rsv - for the orphan deletion.
9243 * 2) rsv - for the truncate reservation, which we will steal from the
9244 * transaction reservation.
9245 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9246 * updating the inode.
9247 */
9248 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9249 if (!rsv)
9250 return -ENOMEM;
9251 rsv->size = min_size;
9252 rsv->failfast = 1;
9253
9254 /*
9255 * 1 for the truncate slack space
9256 * 1 for updating the inode.
9257 */
9258 trans = btrfs_start_transaction(root, 2);
9259 if (IS_ERR(trans)) {
9260 err = PTR_ERR(trans);
9261 goto out;
9262 }
9263
9264 /* Migrate the slack space for the truncate to our reserve */
9265 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9266 min_size, 0);
9267 BUG_ON(ret);
9268
9269 /*
9270 * So if we truncate and then write and fsync we normally would just
9271 * write the extents that changed, which is a problem if we need to
9272 * first truncate that entire inode. So set this flag so we write out
9273 * all of the extents in the inode to the sync log so we're completely
9274 * safe.
9275 */
9276 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9277 trans->block_rsv = rsv;
9278
9279 while (1) {
9280 ret = btrfs_truncate_inode_items(trans, root, inode,
9281 inode->i_size,
9282 BTRFS_EXTENT_DATA_KEY);
9283 if (ret != -ENOSPC && ret != -EAGAIN) {
9284 err = ret;
9285 break;
9286 }
9287
9288 trans->block_rsv = &fs_info->trans_block_rsv;
9289 ret = btrfs_update_inode(trans, root, inode);
9290 if (ret) {
9291 err = ret;
9292 break;
9293 }
9294
9295 btrfs_end_transaction(trans);
9296 btrfs_btree_balance_dirty(fs_info);
9297
9298 trans = btrfs_start_transaction(root, 2);
9299 if (IS_ERR(trans)) {
9300 ret = err = PTR_ERR(trans);
9301 trans = NULL;
9302 break;
9303 }
9304
9305 btrfs_block_rsv_release(fs_info, rsv, -1);
9306 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9307 rsv, min_size, 0);
9308 BUG_ON(ret); /* shouldn't happen */
9309 trans->block_rsv = rsv;
9310 }
9311
9312 if (ret == 0 && inode->i_nlink > 0) {
9313 trans->block_rsv = root->orphan_block_rsv;
9314 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9315 if (ret)
9316 err = ret;
9317 }
9318
9319 if (trans) {
9320 trans->block_rsv = &fs_info->trans_block_rsv;
9321 ret = btrfs_update_inode(trans, root, inode);
9322 if (ret && !err)
9323 err = ret;
9324
9325 ret = btrfs_end_transaction(trans);
9326 btrfs_btree_balance_dirty(fs_info);
9327 }
9328 out:
9329 btrfs_free_block_rsv(fs_info, rsv);
9330
9331 if (ret && !err)
9332 err = ret;
9333
9334 return err;
9335 }
9336
9337 /*
9338 * create a new subvolume directory/inode (helper for the ioctl).
9339 */
9340 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9341 struct btrfs_root *new_root,
9342 struct btrfs_root *parent_root,
9343 u64 new_dirid)
9344 {
9345 struct inode *inode;
9346 int err;
9347 u64 index = 0;
9348
9349 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9350 new_dirid, new_dirid,
9351 S_IFDIR | (~current_umask() & S_IRWXUGO),
9352 &index);
9353 if (IS_ERR(inode))
9354 return PTR_ERR(inode);
9355 inode->i_op = &btrfs_dir_inode_operations;
9356 inode->i_fop = &btrfs_dir_file_operations;
9357
9358 set_nlink(inode, 1);
9359 btrfs_i_size_write(BTRFS_I(inode), 0);
9360 unlock_new_inode(inode);
9361
9362 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9363 if (err)
9364 btrfs_err(new_root->fs_info,
9365 "error inheriting subvolume %llu properties: %d",
9366 new_root->root_key.objectid, err);
9367
9368 err = btrfs_update_inode(trans, new_root, inode);
9369
9370 iput(inode);
9371 return err;
9372 }
9373
9374 struct inode *btrfs_alloc_inode(struct super_block *sb)
9375 {
9376 struct btrfs_inode *ei;
9377 struct inode *inode;
9378
9379 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9380 if (!ei)
9381 return NULL;
9382
9383 ei->root = NULL;
9384 ei->generation = 0;
9385 ei->last_trans = 0;
9386 ei->last_sub_trans = 0;
9387 ei->logged_trans = 0;
9388 ei->delalloc_bytes = 0;
9389 ei->new_delalloc_bytes = 0;
9390 ei->defrag_bytes = 0;
9391 ei->disk_i_size = 0;
9392 ei->flags = 0;
9393 ei->csum_bytes = 0;
9394 ei->index_cnt = (u64)-1;
9395 ei->dir_index = 0;
9396 ei->last_unlink_trans = 0;
9397 ei->last_log_commit = 0;
9398 ei->delayed_iput_count = 0;
9399
9400 spin_lock_init(&ei->lock);
9401 ei->outstanding_extents = 0;
9402 ei->reserved_extents = 0;
9403
9404 ei->runtime_flags = 0;
9405 ei->force_compress = BTRFS_COMPRESS_NONE;
9406
9407 ei->delayed_node = NULL;
9408
9409 ei->i_otime.tv_sec = 0;
9410 ei->i_otime.tv_nsec = 0;
9411
9412 inode = &ei->vfs_inode;
9413 extent_map_tree_init(&ei->extent_tree);
9414 extent_io_tree_init(&ei->io_tree, inode);
9415 extent_io_tree_init(&ei->io_failure_tree, inode);
9416 ei->io_tree.track_uptodate = 1;
9417 ei->io_failure_tree.track_uptodate = 1;
9418 atomic_set(&ei->sync_writers, 0);
9419 mutex_init(&ei->log_mutex);
9420 mutex_init(&ei->delalloc_mutex);
9421 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9422 INIT_LIST_HEAD(&ei->delalloc_inodes);
9423 INIT_LIST_HEAD(&ei->delayed_iput);
9424 RB_CLEAR_NODE(&ei->rb_node);
9425 init_rwsem(&ei->dio_sem);
9426
9427 return inode;
9428 }
9429
9430 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9431 void btrfs_test_destroy_inode(struct inode *inode)
9432 {
9433 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9434 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9435 }
9436 #endif
9437
9438 static void btrfs_i_callback(struct rcu_head *head)
9439 {
9440 struct inode *inode = container_of(head, struct inode, i_rcu);
9441 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9442 }
9443
9444 void btrfs_destroy_inode(struct inode *inode)
9445 {
9446 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9447 struct btrfs_ordered_extent *ordered;
9448 struct btrfs_root *root = BTRFS_I(inode)->root;
9449
9450 WARN_ON(!hlist_empty(&inode->i_dentry));
9451 WARN_ON(inode->i_data.nrpages);
9452 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9453 WARN_ON(BTRFS_I(inode)->reserved_extents);
9454 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9455 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9456 WARN_ON(BTRFS_I(inode)->csum_bytes);
9457 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9458
9459 /*
9460 * This can happen where we create an inode, but somebody else also
9461 * created the same inode and we need to destroy the one we already
9462 * created.
9463 */
9464 if (!root)
9465 goto free;
9466
9467 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9468 &BTRFS_I(inode)->runtime_flags)) {
9469 btrfs_info(fs_info, "inode %llu still on the orphan list",
9470 btrfs_ino(BTRFS_I(inode)));
9471 atomic_dec(&root->orphan_inodes);
9472 }
9473
9474 while (1) {
9475 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9476 if (!ordered)
9477 break;
9478 else {
9479 btrfs_err(fs_info,
9480 "found ordered extent %llu %llu on inode cleanup",
9481 ordered->file_offset, ordered->len);
9482 btrfs_remove_ordered_extent(inode, ordered);
9483 btrfs_put_ordered_extent(ordered);
9484 btrfs_put_ordered_extent(ordered);
9485 }
9486 }
9487 btrfs_qgroup_check_reserved_leak(inode);
9488 inode_tree_del(inode);
9489 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9490 free:
9491 call_rcu(&inode->i_rcu, btrfs_i_callback);
9492 }
9493
9494 int btrfs_drop_inode(struct inode *inode)
9495 {
9496 struct btrfs_root *root = BTRFS_I(inode)->root;
9497
9498 if (root == NULL)
9499 return 1;
9500
9501 /* the snap/subvol tree is on deleting */
9502 if (btrfs_root_refs(&root->root_item) == 0)
9503 return 1;
9504 else
9505 return generic_drop_inode(inode);
9506 }
9507
9508 static void init_once(void *foo)
9509 {
9510 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9511
9512 inode_init_once(&ei->vfs_inode);
9513 }
9514
9515 void btrfs_destroy_cachep(void)
9516 {
9517 /*
9518 * Make sure all delayed rcu free inodes are flushed before we
9519 * destroy cache.
9520 */
9521 rcu_barrier();
9522 kmem_cache_destroy(btrfs_inode_cachep);
9523 kmem_cache_destroy(btrfs_trans_handle_cachep);
9524 kmem_cache_destroy(btrfs_path_cachep);
9525 kmem_cache_destroy(btrfs_free_space_cachep);
9526 }
9527
9528 int btrfs_init_cachep(void)
9529 {
9530 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9531 sizeof(struct btrfs_inode), 0,
9532 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9533 init_once);
9534 if (!btrfs_inode_cachep)
9535 goto fail;
9536
9537 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9538 sizeof(struct btrfs_trans_handle), 0,
9539 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9540 if (!btrfs_trans_handle_cachep)
9541 goto fail;
9542
9543 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9544 sizeof(struct btrfs_path), 0,
9545 SLAB_MEM_SPREAD, NULL);
9546 if (!btrfs_path_cachep)
9547 goto fail;
9548
9549 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9550 sizeof(struct btrfs_free_space), 0,
9551 SLAB_MEM_SPREAD, NULL);
9552 if (!btrfs_free_space_cachep)
9553 goto fail;
9554
9555 return 0;
9556 fail:
9557 btrfs_destroy_cachep();
9558 return -ENOMEM;
9559 }
9560
9561 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9562 u32 request_mask, unsigned int flags)
9563 {
9564 u64 delalloc_bytes;
9565 struct inode *inode = d_inode(path->dentry);
9566 u32 blocksize = inode->i_sb->s_blocksize;
9567 u32 bi_flags = BTRFS_I(inode)->flags;
9568
9569 stat->result_mask |= STATX_BTIME;
9570 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9571 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9572 if (bi_flags & BTRFS_INODE_APPEND)
9573 stat->attributes |= STATX_ATTR_APPEND;
9574 if (bi_flags & BTRFS_INODE_COMPRESS)
9575 stat->attributes |= STATX_ATTR_COMPRESSED;
9576 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9577 stat->attributes |= STATX_ATTR_IMMUTABLE;
9578 if (bi_flags & BTRFS_INODE_NODUMP)
9579 stat->attributes |= STATX_ATTR_NODUMP;
9580
9581 stat->attributes_mask |= (STATX_ATTR_APPEND |
9582 STATX_ATTR_COMPRESSED |
9583 STATX_ATTR_IMMUTABLE |
9584 STATX_ATTR_NODUMP);
9585
9586 generic_fillattr(inode, stat);
9587 stat->dev = BTRFS_I(inode)->root->anon_dev;
9588
9589 spin_lock(&BTRFS_I(inode)->lock);
9590 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9591 spin_unlock(&BTRFS_I(inode)->lock);
9592 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9593 ALIGN(delalloc_bytes, blocksize)) >> 9;
9594 return 0;
9595 }
9596
9597 static int btrfs_rename_exchange(struct inode *old_dir,
9598 struct dentry *old_dentry,
9599 struct inode *new_dir,
9600 struct dentry *new_dentry)
9601 {
9602 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9603 struct btrfs_trans_handle *trans;
9604 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9605 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9606 struct inode *new_inode = new_dentry->d_inode;
9607 struct inode *old_inode = old_dentry->d_inode;
9608 struct timespec ctime = current_time(old_inode);
9609 struct dentry *parent;
9610 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9611 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9612 u64 old_idx = 0;
9613 u64 new_idx = 0;
9614 u64 root_objectid;
9615 int ret;
9616 bool root_log_pinned = false;
9617 bool dest_log_pinned = false;
9618
9619 /* we only allow rename subvolume link between subvolumes */
9620 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9621 return -EXDEV;
9622
9623 /* close the race window with snapshot create/destroy ioctl */
9624 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9625 down_read(&fs_info->subvol_sem);
9626 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9627 down_read(&fs_info->subvol_sem);
9628
9629 /*
9630 * We want to reserve the absolute worst case amount of items. So if
9631 * both inodes are subvols and we need to unlink them then that would
9632 * require 4 item modifications, but if they are both normal inodes it
9633 * would require 5 item modifications, so we'll assume their normal
9634 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9635 * should cover the worst case number of items we'll modify.
9636 */
9637 trans = btrfs_start_transaction(root, 12);
9638 if (IS_ERR(trans)) {
9639 ret = PTR_ERR(trans);
9640 goto out_notrans;
9641 }
9642
9643 /*
9644 * We need to find a free sequence number both in the source and
9645 * in the destination directory for the exchange.
9646 */
9647 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9648 if (ret)
9649 goto out_fail;
9650 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9651 if (ret)
9652 goto out_fail;
9653
9654 BTRFS_I(old_inode)->dir_index = 0ULL;
9655 BTRFS_I(new_inode)->dir_index = 0ULL;
9656
9657 /* Reference for the source. */
9658 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9659 /* force full log commit if subvolume involved. */
9660 btrfs_set_log_full_commit(fs_info, trans);
9661 } else {
9662 btrfs_pin_log_trans(root);
9663 root_log_pinned = true;
9664 ret = btrfs_insert_inode_ref(trans, dest,
9665 new_dentry->d_name.name,
9666 new_dentry->d_name.len,
9667 old_ino,
9668 btrfs_ino(BTRFS_I(new_dir)),
9669 old_idx);
9670 if (ret)
9671 goto out_fail;
9672 }
9673
9674 /* And now for the dest. */
9675 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9676 /* force full log commit if subvolume involved. */
9677 btrfs_set_log_full_commit(fs_info, trans);
9678 } else {
9679 btrfs_pin_log_trans(dest);
9680 dest_log_pinned = true;
9681 ret = btrfs_insert_inode_ref(trans, root,
9682 old_dentry->d_name.name,
9683 old_dentry->d_name.len,
9684 new_ino,
9685 btrfs_ino(BTRFS_I(old_dir)),
9686 new_idx);
9687 if (ret)
9688 goto out_fail;
9689 }
9690
9691 /* Update inode version and ctime/mtime. */
9692 inode_inc_iversion(old_dir);
9693 inode_inc_iversion(new_dir);
9694 inode_inc_iversion(old_inode);
9695 inode_inc_iversion(new_inode);
9696 old_dir->i_ctime = old_dir->i_mtime = ctime;
9697 new_dir->i_ctime = new_dir->i_mtime = ctime;
9698 old_inode->i_ctime = ctime;
9699 new_inode->i_ctime = ctime;
9700
9701 if (old_dentry->d_parent != new_dentry->d_parent) {
9702 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9703 BTRFS_I(old_inode), 1);
9704 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9705 BTRFS_I(new_inode), 1);
9706 }
9707
9708 /* src is a subvolume */
9709 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9710 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9711 ret = btrfs_unlink_subvol(trans, root, old_dir,
9712 root_objectid,
9713 old_dentry->d_name.name,
9714 old_dentry->d_name.len);
9715 } else { /* src is an inode */
9716 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9717 BTRFS_I(old_dentry->d_inode),
9718 old_dentry->d_name.name,
9719 old_dentry->d_name.len);
9720 if (!ret)
9721 ret = btrfs_update_inode(trans, root, old_inode);
9722 }
9723 if (ret) {
9724 btrfs_abort_transaction(trans, ret);
9725 goto out_fail;
9726 }
9727
9728 /* dest is a subvolume */
9729 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9730 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9731 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9732 root_objectid,
9733 new_dentry->d_name.name,
9734 new_dentry->d_name.len);
9735 } else { /* dest is an inode */
9736 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9737 BTRFS_I(new_dentry->d_inode),
9738 new_dentry->d_name.name,
9739 new_dentry->d_name.len);
9740 if (!ret)
9741 ret = btrfs_update_inode(trans, dest, new_inode);
9742 }
9743 if (ret) {
9744 btrfs_abort_transaction(trans, ret);
9745 goto out_fail;
9746 }
9747
9748 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9749 new_dentry->d_name.name,
9750 new_dentry->d_name.len, 0, old_idx);
9751 if (ret) {
9752 btrfs_abort_transaction(trans, ret);
9753 goto out_fail;
9754 }
9755
9756 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9757 old_dentry->d_name.name,
9758 old_dentry->d_name.len, 0, new_idx);
9759 if (ret) {
9760 btrfs_abort_transaction(trans, ret);
9761 goto out_fail;
9762 }
9763
9764 if (old_inode->i_nlink == 1)
9765 BTRFS_I(old_inode)->dir_index = old_idx;
9766 if (new_inode->i_nlink == 1)
9767 BTRFS_I(new_inode)->dir_index = new_idx;
9768
9769 if (root_log_pinned) {
9770 parent = new_dentry->d_parent;
9771 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9772 parent);
9773 btrfs_end_log_trans(root);
9774 root_log_pinned = false;
9775 }
9776 if (dest_log_pinned) {
9777 parent = old_dentry->d_parent;
9778 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9779 parent);
9780 btrfs_end_log_trans(dest);
9781 dest_log_pinned = false;
9782 }
9783 out_fail:
9784 /*
9785 * If we have pinned a log and an error happened, we unpin tasks
9786 * trying to sync the log and force them to fallback to a transaction
9787 * commit if the log currently contains any of the inodes involved in
9788 * this rename operation (to ensure we do not persist a log with an
9789 * inconsistent state for any of these inodes or leading to any
9790 * inconsistencies when replayed). If the transaction was aborted, the
9791 * abortion reason is propagated to userspace when attempting to commit
9792 * the transaction. If the log does not contain any of these inodes, we
9793 * allow the tasks to sync it.
9794 */
9795 if (ret && (root_log_pinned || dest_log_pinned)) {
9796 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9797 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9798 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9799 (new_inode &&
9800 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9801 btrfs_set_log_full_commit(fs_info, trans);
9802
9803 if (root_log_pinned) {
9804 btrfs_end_log_trans(root);
9805 root_log_pinned = false;
9806 }
9807 if (dest_log_pinned) {
9808 btrfs_end_log_trans(dest);
9809 dest_log_pinned = false;
9810 }
9811 }
9812 ret = btrfs_end_transaction(trans);
9813 out_notrans:
9814 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9815 up_read(&fs_info->subvol_sem);
9816 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9817 up_read(&fs_info->subvol_sem);
9818
9819 return ret;
9820 }
9821
9822 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9823 struct btrfs_root *root,
9824 struct inode *dir,
9825 struct dentry *dentry)
9826 {
9827 int ret;
9828 struct inode *inode;
9829 u64 objectid;
9830 u64 index;
9831
9832 ret = btrfs_find_free_ino(root, &objectid);
9833 if (ret)
9834 return ret;
9835
9836 inode = btrfs_new_inode(trans, root, dir,
9837 dentry->d_name.name,
9838 dentry->d_name.len,
9839 btrfs_ino(BTRFS_I(dir)),
9840 objectid,
9841 S_IFCHR | WHITEOUT_MODE,
9842 &index);
9843
9844 if (IS_ERR(inode)) {
9845 ret = PTR_ERR(inode);
9846 return ret;
9847 }
9848
9849 inode->i_op = &btrfs_special_inode_operations;
9850 init_special_inode(inode, inode->i_mode,
9851 WHITEOUT_DEV);
9852
9853 ret = btrfs_init_inode_security(trans, inode, dir,
9854 &dentry->d_name);
9855 if (ret)
9856 goto out;
9857
9858 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9859 BTRFS_I(inode), 0, index);
9860 if (ret)
9861 goto out;
9862
9863 ret = btrfs_update_inode(trans, root, inode);
9864 out:
9865 unlock_new_inode(inode);
9866 if (ret)
9867 inode_dec_link_count(inode);
9868 iput(inode);
9869
9870 return ret;
9871 }
9872
9873 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9874 struct inode *new_dir, struct dentry *new_dentry,
9875 unsigned int flags)
9876 {
9877 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9878 struct btrfs_trans_handle *trans;
9879 unsigned int trans_num_items;
9880 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9881 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9882 struct inode *new_inode = d_inode(new_dentry);
9883 struct inode *old_inode = d_inode(old_dentry);
9884 u64 index = 0;
9885 u64 root_objectid;
9886 int ret;
9887 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9888 bool log_pinned = false;
9889
9890 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9891 return -EPERM;
9892
9893 /* we only allow rename subvolume link between subvolumes */
9894 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9895 return -EXDEV;
9896
9897 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9898 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9899 return -ENOTEMPTY;
9900
9901 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9902 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9903 return -ENOTEMPTY;
9904
9905
9906 /* check for collisions, even if the name isn't there */
9907 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9908 new_dentry->d_name.name,
9909 new_dentry->d_name.len);
9910
9911 if (ret) {
9912 if (ret == -EEXIST) {
9913 /* we shouldn't get
9914 * eexist without a new_inode */
9915 if (WARN_ON(!new_inode)) {
9916 return ret;
9917 }
9918 } else {
9919 /* maybe -EOVERFLOW */
9920 return ret;
9921 }
9922 }
9923 ret = 0;
9924
9925 /*
9926 * we're using rename to replace one file with another. Start IO on it
9927 * now so we don't add too much work to the end of the transaction
9928 */
9929 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9930 filemap_flush(old_inode->i_mapping);
9931
9932 /* close the racy window with snapshot create/destroy ioctl */
9933 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9934 down_read(&fs_info->subvol_sem);
9935 /*
9936 * We want to reserve the absolute worst case amount of items. So if
9937 * both inodes are subvols and we need to unlink them then that would
9938 * require 4 item modifications, but if they are both normal inodes it
9939 * would require 5 item modifications, so we'll assume they are normal
9940 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9941 * should cover the worst case number of items we'll modify.
9942 * If our rename has the whiteout flag, we need more 5 units for the
9943 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9944 * when selinux is enabled).
9945 */
9946 trans_num_items = 11;
9947 if (flags & RENAME_WHITEOUT)
9948 trans_num_items += 5;
9949 trans = btrfs_start_transaction(root, trans_num_items);
9950 if (IS_ERR(trans)) {
9951 ret = PTR_ERR(trans);
9952 goto out_notrans;
9953 }
9954
9955 if (dest != root)
9956 btrfs_record_root_in_trans(trans, dest);
9957
9958 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9959 if (ret)
9960 goto out_fail;
9961
9962 BTRFS_I(old_inode)->dir_index = 0ULL;
9963 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9964 /* force full log commit if subvolume involved. */
9965 btrfs_set_log_full_commit(fs_info, trans);
9966 } else {
9967 btrfs_pin_log_trans(root);
9968 log_pinned = true;
9969 ret = btrfs_insert_inode_ref(trans, dest,
9970 new_dentry->d_name.name,
9971 new_dentry->d_name.len,
9972 old_ino,
9973 btrfs_ino(BTRFS_I(new_dir)), index);
9974 if (ret)
9975 goto out_fail;
9976 }
9977
9978 inode_inc_iversion(old_dir);
9979 inode_inc_iversion(new_dir);
9980 inode_inc_iversion(old_inode);
9981 old_dir->i_ctime = old_dir->i_mtime =
9982 new_dir->i_ctime = new_dir->i_mtime =
9983 old_inode->i_ctime = current_time(old_dir);
9984
9985 if (old_dentry->d_parent != new_dentry->d_parent)
9986 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9987 BTRFS_I(old_inode), 1);
9988
9989 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9990 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9991 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9992 old_dentry->d_name.name,
9993 old_dentry->d_name.len);
9994 } else {
9995 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9996 BTRFS_I(d_inode(old_dentry)),
9997 old_dentry->d_name.name,
9998 old_dentry->d_name.len);
9999 if (!ret)
10000 ret = btrfs_update_inode(trans, root, old_inode);
10001 }
10002 if (ret) {
10003 btrfs_abort_transaction(trans, ret);
10004 goto out_fail;
10005 }
10006
10007 if (new_inode) {
10008 inode_inc_iversion(new_inode);
10009 new_inode->i_ctime = current_time(new_inode);
10010 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
10011 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
10012 root_objectid = BTRFS_I(new_inode)->location.objectid;
10013 ret = btrfs_unlink_subvol(trans, dest, new_dir,
10014 root_objectid,
10015 new_dentry->d_name.name,
10016 new_dentry->d_name.len);
10017 BUG_ON(new_inode->i_nlink == 0);
10018 } else {
10019 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
10020 BTRFS_I(d_inode(new_dentry)),
10021 new_dentry->d_name.name,
10022 new_dentry->d_name.len);
10023 }
10024 if (!ret && new_inode->i_nlink == 0)
10025 ret = btrfs_orphan_add(trans,
10026 BTRFS_I(d_inode(new_dentry)));
10027 if (ret) {
10028 btrfs_abort_transaction(trans, ret);
10029 goto out_fail;
10030 }
10031 }
10032
10033 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
10034 new_dentry->d_name.name,
10035 new_dentry->d_name.len, 0, index);
10036 if (ret) {
10037 btrfs_abort_transaction(trans, ret);
10038 goto out_fail;
10039 }
10040
10041 if (old_inode->i_nlink == 1)
10042 BTRFS_I(old_inode)->dir_index = index;
10043
10044 if (log_pinned) {
10045 struct dentry *parent = new_dentry->d_parent;
10046
10047 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
10048 parent);
10049 btrfs_end_log_trans(root);
10050 log_pinned = false;
10051 }
10052
10053 if (flags & RENAME_WHITEOUT) {
10054 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
10055 old_dentry);
10056
10057 if (ret) {
10058 btrfs_abort_transaction(trans, ret);
10059 goto out_fail;
10060 }
10061 }
10062 out_fail:
10063 /*
10064 * If we have pinned the log and an error happened, we unpin tasks
10065 * trying to sync the log and force them to fallback to a transaction
10066 * commit if the log currently contains any of the inodes involved in
10067 * this rename operation (to ensure we do not persist a log with an
10068 * inconsistent state for any of these inodes or leading to any
10069 * inconsistencies when replayed). If the transaction was aborted, the
10070 * abortion reason is propagated to userspace when attempting to commit
10071 * the transaction. If the log does not contain any of these inodes, we
10072 * allow the tasks to sync it.
10073 */
10074 if (ret && log_pinned) {
10075 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
10076 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
10077 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
10078 (new_inode &&
10079 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
10080 btrfs_set_log_full_commit(fs_info, trans);
10081
10082 btrfs_end_log_trans(root);
10083 log_pinned = false;
10084 }
10085 btrfs_end_transaction(trans);
10086 out_notrans:
10087 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
10088 up_read(&fs_info->subvol_sem);
10089
10090 return ret;
10091 }
10092
10093 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
10094 struct inode *new_dir, struct dentry *new_dentry,
10095 unsigned int flags)
10096 {
10097 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
10098 return -EINVAL;
10099
10100 if (flags & RENAME_EXCHANGE)
10101 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
10102 new_dentry);
10103
10104 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
10105 }
10106
10107 static void btrfs_run_delalloc_work(struct btrfs_work *work)
10108 {
10109 struct btrfs_delalloc_work *delalloc_work;
10110 struct inode *inode;
10111
10112 delalloc_work = container_of(work, struct btrfs_delalloc_work,
10113 work);
10114 inode = delalloc_work->inode;
10115 filemap_flush(inode->i_mapping);
10116 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
10117 &BTRFS_I(inode)->runtime_flags))
10118 filemap_flush(inode->i_mapping);
10119
10120 if (delalloc_work->delay_iput)
10121 btrfs_add_delayed_iput(inode);
10122 else
10123 iput(inode);
10124 complete(&delalloc_work->completion);
10125 }
10126
10127 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
10128 int delay_iput)
10129 {
10130 struct btrfs_delalloc_work *work;
10131
10132 work = kmalloc(sizeof(*work), GFP_NOFS);
10133 if (!work)
10134 return NULL;
10135
10136 init_completion(&work->completion);
10137 INIT_LIST_HEAD(&work->list);
10138 work->inode = inode;
10139 work->delay_iput = delay_iput;
10140 WARN_ON_ONCE(!inode);
10141 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10142 btrfs_run_delalloc_work, NULL, NULL);
10143
10144 return work;
10145 }
10146
10147 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10148 {
10149 wait_for_completion(&work->completion);
10150 kfree(work);
10151 }
10152
10153 /*
10154 * some fairly slow code that needs optimization. This walks the list
10155 * of all the inodes with pending delalloc and forces them to disk.
10156 */
10157 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10158 int nr)
10159 {
10160 struct btrfs_inode *binode;
10161 struct inode *inode;
10162 struct btrfs_delalloc_work *work, *next;
10163 struct list_head works;
10164 struct list_head splice;
10165 int ret = 0;
10166
10167 INIT_LIST_HEAD(&works);
10168 INIT_LIST_HEAD(&splice);
10169
10170 mutex_lock(&root->delalloc_mutex);
10171 spin_lock(&root->delalloc_lock);
10172 list_splice_init(&root->delalloc_inodes, &splice);
10173 while (!list_empty(&splice)) {
10174 binode = list_entry(splice.next, struct btrfs_inode,
10175 delalloc_inodes);
10176
10177 list_move_tail(&binode->delalloc_inodes,
10178 &root->delalloc_inodes);
10179 inode = igrab(&binode->vfs_inode);
10180 if (!inode) {
10181 cond_resched_lock(&root->delalloc_lock);
10182 continue;
10183 }
10184 spin_unlock(&root->delalloc_lock);
10185
10186 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10187 if (!work) {
10188 if (delay_iput)
10189 btrfs_add_delayed_iput(inode);
10190 else
10191 iput(inode);
10192 ret = -ENOMEM;
10193 goto out;
10194 }
10195 list_add_tail(&work->list, &works);
10196 btrfs_queue_work(root->fs_info->flush_workers,
10197 &work->work);
10198 ret++;
10199 if (nr != -1 && ret >= nr)
10200 goto out;
10201 cond_resched();
10202 spin_lock(&root->delalloc_lock);
10203 }
10204 spin_unlock(&root->delalloc_lock);
10205
10206 out:
10207 list_for_each_entry_safe(work, next, &works, list) {
10208 list_del_init(&work->list);
10209 btrfs_wait_and_free_delalloc_work(work);
10210 }
10211
10212 if (!list_empty_careful(&splice)) {
10213 spin_lock(&root->delalloc_lock);
10214 list_splice_tail(&splice, &root->delalloc_inodes);
10215 spin_unlock(&root->delalloc_lock);
10216 }
10217 mutex_unlock(&root->delalloc_mutex);
10218 return ret;
10219 }
10220
10221 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10222 {
10223 struct btrfs_fs_info *fs_info = root->fs_info;
10224 int ret;
10225
10226 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10227 return -EROFS;
10228
10229 ret = __start_delalloc_inodes(root, delay_iput, -1);
10230 if (ret > 0)
10231 ret = 0;
10232 /*
10233 * the filemap_flush will queue IO into the worker threads, but
10234 * we have to make sure the IO is actually started and that
10235 * ordered extents get created before we return
10236 */
10237 atomic_inc(&fs_info->async_submit_draining);
10238 while (atomic_read(&fs_info->nr_async_submits) ||
10239 atomic_read(&fs_info->async_delalloc_pages)) {
10240 wait_event(fs_info->async_submit_wait,
10241 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10242 atomic_read(&fs_info->async_delalloc_pages) == 0));
10243 }
10244 atomic_dec(&fs_info->async_submit_draining);
10245 return ret;
10246 }
10247
10248 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10249 int nr)
10250 {
10251 struct btrfs_root *root;
10252 struct list_head splice;
10253 int ret;
10254
10255 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10256 return -EROFS;
10257
10258 INIT_LIST_HEAD(&splice);
10259
10260 mutex_lock(&fs_info->delalloc_root_mutex);
10261 spin_lock(&fs_info->delalloc_root_lock);
10262 list_splice_init(&fs_info->delalloc_roots, &splice);
10263 while (!list_empty(&splice) && nr) {
10264 root = list_first_entry(&splice, struct btrfs_root,
10265 delalloc_root);
10266 root = btrfs_grab_fs_root(root);
10267 BUG_ON(!root);
10268 list_move_tail(&root->delalloc_root,
10269 &fs_info->delalloc_roots);
10270 spin_unlock(&fs_info->delalloc_root_lock);
10271
10272 ret = __start_delalloc_inodes(root, delay_iput, nr);
10273 btrfs_put_fs_root(root);
10274 if (ret < 0)
10275 goto out;
10276
10277 if (nr != -1) {
10278 nr -= ret;
10279 WARN_ON(nr < 0);
10280 }
10281 spin_lock(&fs_info->delalloc_root_lock);
10282 }
10283 spin_unlock(&fs_info->delalloc_root_lock);
10284
10285 ret = 0;
10286 atomic_inc(&fs_info->async_submit_draining);
10287 while (atomic_read(&fs_info->nr_async_submits) ||
10288 atomic_read(&fs_info->async_delalloc_pages)) {
10289 wait_event(fs_info->async_submit_wait,
10290 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10291 atomic_read(&fs_info->async_delalloc_pages) == 0));
10292 }
10293 atomic_dec(&fs_info->async_submit_draining);
10294 out:
10295 if (!list_empty_careful(&splice)) {
10296 spin_lock(&fs_info->delalloc_root_lock);
10297 list_splice_tail(&splice, &fs_info->delalloc_roots);
10298 spin_unlock(&fs_info->delalloc_root_lock);
10299 }
10300 mutex_unlock(&fs_info->delalloc_root_mutex);
10301 return ret;
10302 }
10303
10304 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10305 const char *symname)
10306 {
10307 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10308 struct btrfs_trans_handle *trans;
10309 struct btrfs_root *root = BTRFS_I(dir)->root;
10310 struct btrfs_path *path;
10311 struct btrfs_key key;
10312 struct inode *inode = NULL;
10313 int err;
10314 int drop_inode = 0;
10315 u64 objectid;
10316 u64 index = 0;
10317 int name_len;
10318 int datasize;
10319 unsigned long ptr;
10320 struct btrfs_file_extent_item *ei;
10321 struct extent_buffer *leaf;
10322
10323 name_len = strlen(symname);
10324 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10325 return -ENAMETOOLONG;
10326
10327 /*
10328 * 2 items for inode item and ref
10329 * 2 items for dir items
10330 * 1 item for updating parent inode item
10331 * 1 item for the inline extent item
10332 * 1 item for xattr if selinux is on
10333 */
10334 trans = btrfs_start_transaction(root, 7);
10335 if (IS_ERR(trans))
10336 return PTR_ERR(trans);
10337
10338 err = btrfs_find_free_ino(root, &objectid);
10339 if (err)
10340 goto out_unlock;
10341
10342 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10343 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10344 objectid, S_IFLNK|S_IRWXUGO, &index);
10345 if (IS_ERR(inode)) {
10346 err = PTR_ERR(inode);
10347 goto out_unlock;
10348 }
10349
10350 /*
10351 * If the active LSM wants to access the inode during
10352 * d_instantiate it needs these. Smack checks to see
10353 * if the filesystem supports xattrs by looking at the
10354 * ops vector.
10355 */
10356 inode->i_fop = &btrfs_file_operations;
10357 inode->i_op = &btrfs_file_inode_operations;
10358 inode->i_mapping->a_ops = &btrfs_aops;
10359 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10360
10361 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10362 if (err)
10363 goto out_unlock_inode;
10364
10365 path = btrfs_alloc_path();
10366 if (!path) {
10367 err = -ENOMEM;
10368 goto out_unlock_inode;
10369 }
10370 key.objectid = btrfs_ino(BTRFS_I(inode));
10371 key.offset = 0;
10372 key.type = BTRFS_EXTENT_DATA_KEY;
10373 datasize = btrfs_file_extent_calc_inline_size(name_len);
10374 err = btrfs_insert_empty_item(trans, root, path, &key,
10375 datasize);
10376 if (err) {
10377 btrfs_free_path(path);
10378 goto out_unlock_inode;
10379 }
10380 leaf = path->nodes[0];
10381 ei = btrfs_item_ptr(leaf, path->slots[0],
10382 struct btrfs_file_extent_item);
10383 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10384 btrfs_set_file_extent_type(leaf, ei,
10385 BTRFS_FILE_EXTENT_INLINE);
10386 btrfs_set_file_extent_encryption(leaf, ei, 0);
10387 btrfs_set_file_extent_compression(leaf, ei, 0);
10388 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10389 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10390
10391 ptr = btrfs_file_extent_inline_start(ei);
10392 write_extent_buffer(leaf, symname, ptr, name_len);
10393 btrfs_mark_buffer_dirty(leaf);
10394 btrfs_free_path(path);
10395
10396 inode->i_op = &btrfs_symlink_inode_operations;
10397 inode_nohighmem(inode);
10398 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10399 inode_set_bytes(inode, name_len);
10400 btrfs_i_size_write(BTRFS_I(inode), name_len);
10401 err = btrfs_update_inode(trans, root, inode);
10402 /*
10403 * Last step, add directory indexes for our symlink inode. This is the
10404 * last step to avoid extra cleanup of these indexes if an error happens
10405 * elsewhere above.
10406 */
10407 if (!err)
10408 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10409 BTRFS_I(inode), 0, index);
10410 if (err) {
10411 drop_inode = 1;
10412 goto out_unlock_inode;
10413 }
10414
10415 unlock_new_inode(inode);
10416 d_instantiate(dentry, inode);
10417
10418 out_unlock:
10419 btrfs_end_transaction(trans);
10420 if (drop_inode) {
10421 inode_dec_link_count(inode);
10422 iput(inode);
10423 }
10424 btrfs_btree_balance_dirty(fs_info);
10425 return err;
10426
10427 out_unlock_inode:
10428 drop_inode = 1;
10429 unlock_new_inode(inode);
10430 goto out_unlock;
10431 }
10432
10433 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10434 u64 start, u64 num_bytes, u64 min_size,
10435 loff_t actual_len, u64 *alloc_hint,
10436 struct btrfs_trans_handle *trans)
10437 {
10438 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10439 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10440 struct extent_map *em;
10441 struct btrfs_root *root = BTRFS_I(inode)->root;
10442 struct btrfs_key ins;
10443 u64 cur_offset = start;
10444 u64 i_size;
10445 u64 cur_bytes;
10446 u64 last_alloc = (u64)-1;
10447 int ret = 0;
10448 bool own_trans = true;
10449 u64 end = start + num_bytes - 1;
10450
10451 if (trans)
10452 own_trans = false;
10453 while (num_bytes > 0) {
10454 if (own_trans) {
10455 trans = btrfs_start_transaction(root, 3);
10456 if (IS_ERR(trans)) {
10457 ret = PTR_ERR(trans);
10458 break;
10459 }
10460 }
10461
10462 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10463 cur_bytes = max(cur_bytes, min_size);
10464 /*
10465 * If we are severely fragmented we could end up with really
10466 * small allocations, so if the allocator is returning small
10467 * chunks lets make its job easier by only searching for those
10468 * sized chunks.
10469 */
10470 cur_bytes = min(cur_bytes, last_alloc);
10471 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10472 min_size, 0, *alloc_hint, &ins, 1, 0);
10473 if (ret) {
10474 if (own_trans)
10475 btrfs_end_transaction(trans);
10476 break;
10477 }
10478 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10479
10480 last_alloc = ins.offset;
10481 ret = insert_reserved_file_extent(trans, inode,
10482 cur_offset, ins.objectid,
10483 ins.offset, ins.offset,
10484 ins.offset, 0, 0, 0,
10485 BTRFS_FILE_EXTENT_PREALLOC);
10486 if (ret) {
10487 btrfs_free_reserved_extent(fs_info, ins.objectid,
10488 ins.offset, 0);
10489 btrfs_abort_transaction(trans, ret);
10490 if (own_trans)
10491 btrfs_end_transaction(trans);
10492 break;
10493 }
10494
10495 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10496 cur_offset + ins.offset -1, 0);
10497
10498 em = alloc_extent_map();
10499 if (!em) {
10500 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10501 &BTRFS_I(inode)->runtime_flags);
10502 goto next;
10503 }
10504
10505 em->start = cur_offset;
10506 em->orig_start = cur_offset;
10507 em->len = ins.offset;
10508 em->block_start = ins.objectid;
10509 em->block_len = ins.offset;
10510 em->orig_block_len = ins.offset;
10511 em->ram_bytes = ins.offset;
10512 em->bdev = fs_info->fs_devices->latest_bdev;
10513 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10514 em->generation = trans->transid;
10515
10516 while (1) {
10517 write_lock(&em_tree->lock);
10518 ret = add_extent_mapping(em_tree, em, 1);
10519 write_unlock(&em_tree->lock);
10520 if (ret != -EEXIST)
10521 break;
10522 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10523 cur_offset + ins.offset - 1,
10524 0);
10525 }
10526 free_extent_map(em);
10527 next:
10528 num_bytes -= ins.offset;
10529 cur_offset += ins.offset;
10530 *alloc_hint = ins.objectid + ins.offset;
10531
10532 inode_inc_iversion(inode);
10533 inode->i_ctime = current_time(inode);
10534 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10535 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10536 (actual_len > inode->i_size) &&
10537 (cur_offset > inode->i_size)) {
10538 if (cur_offset > actual_len)
10539 i_size = actual_len;
10540 else
10541 i_size = cur_offset;
10542 i_size_write(inode, i_size);
10543 btrfs_ordered_update_i_size(inode, i_size, NULL);
10544 }
10545
10546 ret = btrfs_update_inode(trans, root, inode);
10547
10548 if (ret) {
10549 btrfs_abort_transaction(trans, ret);
10550 if (own_trans)
10551 btrfs_end_transaction(trans);
10552 break;
10553 }
10554
10555 if (own_trans)
10556 btrfs_end_transaction(trans);
10557 }
10558 if (cur_offset < end)
10559 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10560 end - cur_offset + 1);
10561 return ret;
10562 }
10563
10564 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10565 u64 start, u64 num_bytes, u64 min_size,
10566 loff_t actual_len, u64 *alloc_hint)
10567 {
10568 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10569 min_size, actual_len, alloc_hint,
10570 NULL);
10571 }
10572
10573 int btrfs_prealloc_file_range_trans(struct inode *inode,
10574 struct btrfs_trans_handle *trans, int mode,
10575 u64 start, u64 num_bytes, u64 min_size,
10576 loff_t actual_len, u64 *alloc_hint)
10577 {
10578 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10579 min_size, actual_len, alloc_hint, trans);
10580 }
10581
10582 static int btrfs_set_page_dirty(struct page *page)
10583 {
10584 return __set_page_dirty_nobuffers(page);
10585 }
10586
10587 static int btrfs_permission(struct inode *inode, int mask)
10588 {
10589 struct btrfs_root *root = BTRFS_I(inode)->root;
10590 umode_t mode = inode->i_mode;
10591
10592 if (mask & MAY_WRITE &&
10593 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10594 if (btrfs_root_readonly(root))
10595 return -EROFS;
10596 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10597 return -EACCES;
10598 }
10599 return generic_permission(inode, mask);
10600 }
10601
10602 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10603 {
10604 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10605 struct btrfs_trans_handle *trans;
10606 struct btrfs_root *root = BTRFS_I(dir)->root;
10607 struct inode *inode = NULL;
10608 u64 objectid;
10609 u64 index;
10610 int ret = 0;
10611
10612 /*
10613 * 5 units required for adding orphan entry
10614 */
10615 trans = btrfs_start_transaction(root, 5);
10616 if (IS_ERR(trans))
10617 return PTR_ERR(trans);
10618
10619 ret = btrfs_find_free_ino(root, &objectid);
10620 if (ret)
10621 goto out;
10622
10623 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10624 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10625 if (IS_ERR(inode)) {
10626 ret = PTR_ERR(inode);
10627 inode = NULL;
10628 goto out;
10629 }
10630
10631 inode->i_fop = &btrfs_file_operations;
10632 inode->i_op = &btrfs_file_inode_operations;
10633
10634 inode->i_mapping->a_ops = &btrfs_aops;
10635 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10636
10637 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10638 if (ret)
10639 goto out_inode;
10640
10641 ret = btrfs_update_inode(trans, root, inode);
10642 if (ret)
10643 goto out_inode;
10644 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10645 if (ret)
10646 goto out_inode;
10647
10648 /*
10649 * We set number of links to 0 in btrfs_new_inode(), and here we set
10650 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10651 * through:
10652 *
10653 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10654 */
10655 set_nlink(inode, 1);
10656 unlock_new_inode(inode);
10657 d_tmpfile(dentry, inode);
10658 mark_inode_dirty(inode);
10659
10660 out:
10661 btrfs_end_transaction(trans);
10662 if (ret)
10663 iput(inode);
10664 btrfs_balance_delayed_items(fs_info);
10665 btrfs_btree_balance_dirty(fs_info);
10666 return ret;
10667
10668 out_inode:
10669 unlock_new_inode(inode);
10670 goto out;
10671
10672 }
10673
10674 __attribute__((const))
10675 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10676 {
10677 return -EAGAIN;
10678 }
10679
10680 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10681 {
10682 struct inode *inode = private_data;
10683 return btrfs_sb(inode->i_sb);
10684 }
10685
10686 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10687 u64 start, u64 end)
10688 {
10689 struct inode *inode = private_data;
10690 u64 isize;
10691
10692 isize = i_size_read(inode);
10693 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10694 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10695 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10696 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10697 }
10698 }
10699
10700 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10701 {
10702 struct inode *inode = private_data;
10703 unsigned long index = start >> PAGE_SHIFT;
10704 unsigned long end_index = end >> PAGE_SHIFT;
10705 struct page *page;
10706
10707 while (index <= end_index) {
10708 page = find_get_page(inode->i_mapping, index);
10709 ASSERT(page); /* Pages should be in the extent_io_tree */
10710 set_page_writeback(page);
10711 put_page(page);
10712 index++;
10713 }
10714 }
10715
10716 static const struct inode_operations btrfs_dir_inode_operations = {
10717 .getattr = btrfs_getattr,
10718 .lookup = btrfs_lookup,
10719 .create = btrfs_create,
10720 .unlink = btrfs_unlink,
10721 .link = btrfs_link,
10722 .mkdir = btrfs_mkdir,
10723 .rmdir = btrfs_rmdir,
10724 .rename = btrfs_rename2,
10725 .symlink = btrfs_symlink,
10726 .setattr = btrfs_setattr,
10727 .mknod = btrfs_mknod,
10728 .listxattr = btrfs_listxattr,
10729 .permission = btrfs_permission,
10730 .get_acl = btrfs_get_acl,
10731 .set_acl = btrfs_set_acl,
10732 .update_time = btrfs_update_time,
10733 .tmpfile = btrfs_tmpfile,
10734 };
10735 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10736 .lookup = btrfs_lookup,
10737 .permission = btrfs_permission,
10738 .update_time = btrfs_update_time,
10739 };
10740
10741 static const struct file_operations btrfs_dir_file_operations = {
10742 .llseek = generic_file_llseek,
10743 .read = generic_read_dir,
10744 .iterate_shared = btrfs_real_readdir,
10745 .unlocked_ioctl = btrfs_ioctl,
10746 #ifdef CONFIG_COMPAT
10747 .compat_ioctl = btrfs_compat_ioctl,
10748 #endif
10749 .release = btrfs_release_file,
10750 .fsync = btrfs_sync_file,
10751 };
10752
10753 static const struct extent_io_ops btrfs_extent_io_ops = {
10754 /* mandatory callbacks */
10755 .submit_bio_hook = btrfs_submit_bio_hook,
10756 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10757 .merge_bio_hook = btrfs_merge_bio_hook,
10758 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10759 .tree_fs_info = iotree_fs_info,
10760 .set_range_writeback = btrfs_set_range_writeback,
10761
10762 /* optional callbacks */
10763 .fill_delalloc = run_delalloc_range,
10764 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10765 .writepage_start_hook = btrfs_writepage_start_hook,
10766 .set_bit_hook = btrfs_set_bit_hook,
10767 .clear_bit_hook = btrfs_clear_bit_hook,
10768 .merge_extent_hook = btrfs_merge_extent_hook,
10769 .split_extent_hook = btrfs_split_extent_hook,
10770 .check_extent_io_range = btrfs_check_extent_io_range,
10771 };
10772
10773 /*
10774 * btrfs doesn't support the bmap operation because swapfiles
10775 * use bmap to make a mapping of extents in the file. They assume
10776 * these extents won't change over the life of the file and they
10777 * use the bmap result to do IO directly to the drive.
10778 *
10779 * the btrfs bmap call would return logical addresses that aren't
10780 * suitable for IO and they also will change frequently as COW
10781 * operations happen. So, swapfile + btrfs == corruption.
10782 *
10783 * For now we're avoiding this by dropping bmap.
10784 */
10785 static const struct address_space_operations btrfs_aops = {
10786 .readpage = btrfs_readpage,
10787 .writepage = btrfs_writepage,
10788 .writepages = btrfs_writepages,
10789 .readpages = btrfs_readpages,
10790 .direct_IO = btrfs_direct_IO,
10791 .invalidatepage = btrfs_invalidatepage,
10792 .releasepage = btrfs_releasepage,
10793 .set_page_dirty = btrfs_set_page_dirty,
10794 .error_remove_page = generic_error_remove_page,
10795 };
10796
10797 static const struct address_space_operations btrfs_symlink_aops = {
10798 .readpage = btrfs_readpage,
10799 .writepage = btrfs_writepage,
10800 .invalidatepage = btrfs_invalidatepage,
10801 .releasepage = btrfs_releasepage,
10802 };
10803
10804 static const struct inode_operations btrfs_file_inode_operations = {
10805 .getattr = btrfs_getattr,
10806 .setattr = btrfs_setattr,
10807 .listxattr = btrfs_listxattr,
10808 .permission = btrfs_permission,
10809 .fiemap = btrfs_fiemap,
10810 .get_acl = btrfs_get_acl,
10811 .set_acl = btrfs_set_acl,
10812 .update_time = btrfs_update_time,
10813 };
10814 static const struct inode_operations btrfs_special_inode_operations = {
10815 .getattr = btrfs_getattr,
10816 .setattr = btrfs_setattr,
10817 .permission = btrfs_permission,
10818 .listxattr = btrfs_listxattr,
10819 .get_acl = btrfs_get_acl,
10820 .set_acl = btrfs_set_acl,
10821 .update_time = btrfs_update_time,
10822 };
10823 static const struct inode_operations btrfs_symlink_inode_operations = {
10824 .get_link = page_get_link,
10825 .getattr = btrfs_getattr,
10826 .setattr = btrfs_setattr,
10827 .permission = btrfs_permission,
10828 .listxattr = btrfs_listxattr,
10829 .update_time = btrfs_update_time,
10830 };
10831
10832 const struct dentry_operations btrfs_dentry_operations = {
10833 .d_delete = btrfs_dentry_delete,
10834 .d_release = btrfs_dentry_release,
10835 };
Linux with added FPC-III support