projects / deliverable / linux.git / blob
? search:


3b39cd8cd1be906094e19ceba2a019d446e87f1a
[deliverable/linux.git] / fs / btrfs / inode.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
42 #include <linux/btrfs.h>
43 #include <linux/blkdev.h>
44 #include <linux/posix_acl_xattr.h>
45 #include <linux/uio.h>
46 #include "ctree.h"
47 #include "disk-io.h"
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
52 #include "xattr.h"
53 #include "tree-log.h"
54 #include "volumes.h"
55 #include "compression.h"
56 #include "locking.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
59 #include "backref.h"
60 #include "hash.h"
61 #include "props.h"
62 #include "qgroup.h"
63
64 struct btrfs_iget_args {
65 struct btrfs_key *location;
66 struct btrfs_root *root;
67 };
68
69 static const struct inode_operations btrfs_dir_inode_operations;
70 static const struct inode_operations btrfs_symlink_inode_operations;
71 static const struct inode_operations btrfs_dir_ro_inode_operations;
72 static const struct inode_operations btrfs_special_inode_operations;
73 static const struct inode_operations btrfs_file_inode_operations;
74 static const struct address_space_operations btrfs_aops;
75 static const struct address_space_operations btrfs_symlink_aops;
76 static const struct file_operations btrfs_dir_file_operations;
77 static struct extent_io_ops btrfs_extent_io_ops;
78
79 static struct kmem_cache *btrfs_inode_cachep;
80 static struct kmem_cache *btrfs_delalloc_work_cachep;
81 struct kmem_cache *btrfs_trans_handle_cachep;
82 struct kmem_cache *btrfs_transaction_cachep;
83 struct kmem_cache *btrfs_path_cachep;
84 struct kmem_cache *btrfs_free_space_cachep;
85
86 #define S_SHIFT 12
87 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
88 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
89 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
90 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
91 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
92 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
93 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
94 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
95 };
96
97 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
98 static int btrfs_truncate(struct inode *inode);
99 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
100 static noinline int cow_file_range(struct inode *inode,
101 struct page *locked_page,
102 u64 start, u64 end, int *page_started,
103 unsigned long *nr_written, int unlock);
104 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
105 u64 len, u64 orig_start,
106 u64 block_start, u64 block_len,
107 u64 orig_block_len, u64 ram_bytes,
108 int type);
109
110 static int btrfs_dirty_inode(struct inode *inode);
111
112 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
113 void btrfs_test_inode_set_ops(struct inode *inode)
114 {
115 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
116 }
117 #endif
118
119 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
120 struct inode *inode, struct inode *dir,
121 const struct qstr *qstr)
122 {
123 int err;
124
125 err = btrfs_init_acl(trans, inode, dir);
126 if (!err)
127 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
128 return err;
129 }
130
131 /*
132 * this does all the hard work for inserting an inline extent into
133 * the btree. The caller should have done a btrfs_drop_extents so that
134 * no overlapping inline items exist in the btree
135 */
136 static int insert_inline_extent(struct btrfs_trans_handle *trans,
137 struct btrfs_path *path, int extent_inserted,
138 struct btrfs_root *root, struct inode *inode,
139 u64 start, size_t size, size_t compressed_size,
140 int compress_type,
141 struct page **compressed_pages)
142 {
143 struct extent_buffer *leaf;
144 struct page *page = NULL;
145 char *kaddr;
146 unsigned long ptr;
147 struct btrfs_file_extent_item *ei;
148 int err = 0;
149 int ret;
150 size_t cur_size = size;
151 unsigned long offset;
152
153 if (compressed_size && compressed_pages)
154 cur_size = compressed_size;
155
156 inode_add_bytes(inode, size);
157
158 if (!extent_inserted) {
159 struct btrfs_key key;
160 size_t datasize;
161
162 key.objectid = btrfs_ino(inode);
163 key.offset = start;
164 key.type = BTRFS_EXTENT_DATA_KEY;
165
166 datasize = btrfs_file_extent_calc_inline_size(cur_size);
167 path->leave_spinning = 1;
168 ret = btrfs_insert_empty_item(trans, root, path, &key,
169 datasize);
170 if (ret) {
171 err = ret;
172 goto fail;
173 }
174 }
175 leaf = path->nodes[0];
176 ei = btrfs_item_ptr(leaf, path->slots[0],
177 struct btrfs_file_extent_item);
178 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
179 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
180 btrfs_set_file_extent_encryption(leaf, ei, 0);
181 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
182 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
183 ptr = btrfs_file_extent_inline_start(ei);
184
185 if (compress_type != BTRFS_COMPRESS_NONE) {
186 struct page *cpage;
187 int i = 0;
188 while (compressed_size > 0) {
189 cpage = compressed_pages[i];
190 cur_size = min_t(unsigned long, compressed_size,
191 PAGE_CACHE_SIZE);
192
193 kaddr = kmap_atomic(cpage);
194 write_extent_buffer(leaf, kaddr, ptr, cur_size);
195 kunmap_atomic(kaddr);
196
197 i++;
198 ptr += cur_size;
199 compressed_size -= cur_size;
200 }
201 btrfs_set_file_extent_compression(leaf, ei,
202 compress_type);
203 } else {
204 page = find_get_page(inode->i_mapping,
205 start >> PAGE_CACHE_SHIFT);
206 btrfs_set_file_extent_compression(leaf, ei, 0);
207 kaddr = kmap_atomic(page);
208 offset = start & (PAGE_CACHE_SIZE - 1);
209 write_extent_buffer(leaf, kaddr + offset, ptr, size);
210 kunmap_atomic(kaddr);
211 page_cache_release(page);
212 }
213 btrfs_mark_buffer_dirty(leaf);
214 btrfs_release_path(path);
215
216 /*
217 * we're an inline extent, so nobody can
218 * extend the file past i_size without locking
219 * a page we already have locked.
220 *
221 * We must do any isize and inode updates
222 * before we unlock the pages. Otherwise we
223 * could end up racing with unlink.
224 */
225 BTRFS_I(inode)->disk_i_size = inode->i_size;
226 ret = btrfs_update_inode(trans, root, inode);
227
228 return ret;
229 fail:
230 return err;
231 }
232
233
234 /*
235 * conditionally insert an inline extent into the file. This
236 * does the checks required to make sure the data is small enough
237 * to fit as an inline extent.
238 */
239 static noinline int cow_file_range_inline(struct btrfs_root *root,
240 struct inode *inode, u64 start,
241 u64 end, size_t compressed_size,
242 int compress_type,
243 struct page **compressed_pages)
244 {
245 struct btrfs_trans_handle *trans;
246 u64 isize = i_size_read(inode);
247 u64 actual_end = min(end + 1, isize);
248 u64 inline_len = actual_end - start;
249 u64 aligned_end = ALIGN(end, root->sectorsize);
250 u64 data_len = inline_len;
251 int ret;
252 struct btrfs_path *path;
253 int extent_inserted = 0;
254 u32 extent_item_size;
255
256 if (compressed_size)
257 data_len = compressed_size;
258
259 if (start > 0 ||
260 actual_end > PAGE_CACHE_SIZE ||
261 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
262 (!compressed_size &&
263 (actual_end & (root->sectorsize - 1)) == 0) ||
264 end + 1 < isize ||
265 data_len > root->fs_info->max_inline) {
266 return 1;
267 }
268
269 path = btrfs_alloc_path();
270 if (!path)
271 return -ENOMEM;
272
273 trans = btrfs_join_transaction(root);
274 if (IS_ERR(trans)) {
275 btrfs_free_path(path);
276 return PTR_ERR(trans);
277 }
278 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
279
280 if (compressed_size && compressed_pages)
281 extent_item_size = btrfs_file_extent_calc_inline_size(
282 compressed_size);
283 else
284 extent_item_size = btrfs_file_extent_calc_inline_size(
285 inline_len);
286
287 ret = __btrfs_drop_extents(trans, root, inode, path,
288 start, aligned_end, NULL,
289 1, 1, extent_item_size, &extent_inserted);
290 if (ret) {
291 btrfs_abort_transaction(trans, root, ret);
292 goto out;
293 }
294
295 if (isize > actual_end)
296 inline_len = min_t(u64, isize, actual_end);
297 ret = insert_inline_extent(trans, path, extent_inserted,
298 root, inode, start,
299 inline_len, compressed_size,
300 compress_type, compressed_pages);
301 if (ret && ret != -ENOSPC) {
302 btrfs_abort_transaction(trans, root, ret);
303 goto out;
304 } else if (ret == -ENOSPC) {
305 ret = 1;
306 goto out;
307 }
308
309 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
310 btrfs_delalloc_release_metadata(inode, end + 1 - start);
311 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
312 out:
313 /*
314 * Don't forget to free the reserved space, as for inlined extent
315 * it won't count as data extent, free them directly here.
316 * And at reserve time, it's always aligned to page size, so
317 * just free one page here.
318 */
319 btrfs_qgroup_free_data(inode, 0, PAGE_CACHE_SIZE);
320 btrfs_free_path(path);
321 btrfs_end_transaction(trans, root);
322 return ret;
323 }
324
325 struct async_extent {
326 u64 start;
327 u64 ram_size;
328 u64 compressed_size;
329 struct page **pages;
330 unsigned long nr_pages;
331 int compress_type;
332 struct list_head list;
333 };
334
335 struct async_cow {
336 struct inode *inode;
337 struct btrfs_root *root;
338 struct page *locked_page;
339 u64 start;
340 u64 end;
341 struct list_head extents;
342 struct btrfs_work work;
343 };
344
345 static noinline int add_async_extent(struct async_cow *cow,
346 u64 start, u64 ram_size,
347 u64 compressed_size,
348 struct page **pages,
349 unsigned long nr_pages,
350 int compress_type)
351 {
352 struct async_extent *async_extent;
353
354 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
355 BUG_ON(!async_extent); /* -ENOMEM */
356 async_extent->start = start;
357 async_extent->ram_size = ram_size;
358 async_extent->compressed_size = compressed_size;
359 async_extent->pages = pages;
360 async_extent->nr_pages = nr_pages;
361 async_extent->compress_type = compress_type;
362 list_add_tail(&async_extent->list, &cow->extents);
363 return 0;
364 }
365
366 static inline int inode_need_compress(struct inode *inode)
367 {
368 struct btrfs_root *root = BTRFS_I(inode)->root;
369
370 /* force compress */
371 if (btrfs_test_opt(root, FORCE_COMPRESS))
372 return 1;
373 /* bad compression ratios */
374 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
375 return 0;
376 if (btrfs_test_opt(root, COMPRESS) ||
377 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
378 BTRFS_I(inode)->force_compress)
379 return 1;
380 return 0;
381 }
382
383 /*
384 * we create compressed extents in two phases. The first
385 * phase compresses a range of pages that have already been
386 * locked (both pages and state bits are locked).
387 *
388 * This is done inside an ordered work queue, and the compression
389 * is spread across many cpus. The actual IO submission is step
390 * two, and the ordered work queue takes care of making sure that
391 * happens in the same order things were put onto the queue by
392 * writepages and friends.
393 *
394 * If this code finds it can't get good compression, it puts an
395 * entry onto the work queue to write the uncompressed bytes. This
396 * makes sure that both compressed inodes and uncompressed inodes
397 * are written in the same order that the flusher thread sent them
398 * down.
399 */
400 static noinline void compress_file_range(struct inode *inode,
401 struct page *locked_page,
402 u64 start, u64 end,
403 struct async_cow *async_cow,
404 int *num_added)
405 {
406 struct btrfs_root *root = BTRFS_I(inode)->root;
407 u64 num_bytes;
408 u64 blocksize = root->sectorsize;
409 u64 actual_end;
410 u64 isize = i_size_read(inode);
411 int ret = 0;
412 struct page **pages = NULL;
413 unsigned long nr_pages;
414 unsigned long nr_pages_ret = 0;
415 unsigned long total_compressed = 0;
416 unsigned long total_in = 0;
417 unsigned long max_compressed = 128 * 1024;
418 unsigned long max_uncompressed = 128 * 1024;
419 int i;
420 int will_compress;
421 int compress_type = root->fs_info->compress_type;
422 int redirty = 0;
423
424 /* if this is a small write inside eof, kick off a defrag */
425 if ((end - start + 1) < 16 * 1024 &&
426 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
427 btrfs_add_inode_defrag(NULL, inode);
428
429 actual_end = min_t(u64, isize, end + 1);
430 again:
431 will_compress = 0;
432 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
433 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
434
435 /*
436 * we don't want to send crud past the end of i_size through
437 * compression, that's just a waste of CPU time. So, if the
438 * end of the file is before the start of our current
439 * requested range of bytes, we bail out to the uncompressed
440 * cleanup code that can deal with all of this.
441 *
442 * It isn't really the fastest way to fix things, but this is a
443 * very uncommon corner.
444 */
445 if (actual_end <= start)
446 goto cleanup_and_bail_uncompressed;
447
448 total_compressed = actual_end - start;
449
450 /*
451 * skip compression for a small file range(<=blocksize) that
452 * isn't an inline extent, since it dosen't save disk space at all.
453 */
454 if (total_compressed <= blocksize &&
455 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
456 goto cleanup_and_bail_uncompressed;
457
458 /* we want to make sure that amount of ram required to uncompress
459 * an extent is reasonable, so we limit the total size in ram
460 * of a compressed extent to 128k. This is a crucial number
461 * because it also controls how easily we can spread reads across
462 * cpus for decompression.
463 *
464 * We also want to make sure the amount of IO required to do
465 * a random read is reasonably small, so we limit the size of
466 * a compressed extent to 128k.
467 */
468 total_compressed = min(total_compressed, max_uncompressed);
469 num_bytes = ALIGN(end - start + 1, blocksize);
470 num_bytes = max(blocksize, num_bytes);
471 total_in = 0;
472 ret = 0;
473
474 /*
475 * we do compression for mount -o compress and when the
476 * inode has not been flagged as nocompress. This flag can
477 * change at any time if we discover bad compression ratios.
478 */
479 if (inode_need_compress(inode)) {
480 WARN_ON(pages);
481 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
482 if (!pages) {
483 /* just bail out to the uncompressed code */
484 goto cont;
485 }
486
487 if (BTRFS_I(inode)->force_compress)
488 compress_type = BTRFS_I(inode)->force_compress;
489
490 /*
491 * we need to call clear_page_dirty_for_io on each
492 * page in the range. Otherwise applications with the file
493 * mmap'd can wander in and change the page contents while
494 * we are compressing them.
495 *
496 * If the compression fails for any reason, we set the pages
497 * dirty again later on.
498 */
499 extent_range_clear_dirty_for_io(inode, start, end);
500 redirty = 1;
501 ret = btrfs_compress_pages(compress_type,
502 inode->i_mapping, start,
503 total_compressed, pages,
504 nr_pages, &nr_pages_ret,
505 &total_in,
506 &total_compressed,
507 max_compressed);
508
509 if (!ret) {
510 unsigned long offset = total_compressed &
511 (PAGE_CACHE_SIZE - 1);
512 struct page *page = pages[nr_pages_ret - 1];
513 char *kaddr;
514
515 /* zero the tail end of the last page, we might be
516 * sending it down to disk
517 */
518 if (offset) {
519 kaddr = kmap_atomic(page);
520 memset(kaddr + offset, 0,
521 PAGE_CACHE_SIZE - offset);
522 kunmap_atomic(kaddr);
523 }
524 will_compress = 1;
525 }
526 }
527 cont:
528 if (start == 0) {
529 /* lets try to make an inline extent */
530 if (ret || total_in < (actual_end - start)) {
531 /* we didn't compress the entire range, try
532 * to make an uncompressed inline extent.
533 */
534 ret = cow_file_range_inline(root, inode, start, end,
535 0, 0, NULL);
536 } else {
537 /* try making a compressed inline extent */
538 ret = cow_file_range_inline(root, inode, start, end,
539 total_compressed,
540 compress_type, pages);
541 }
542 if (ret <= 0) {
543 unsigned long clear_flags = EXTENT_DELALLOC |
544 EXTENT_DEFRAG;
545 unsigned long page_error_op;
546
547 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
548 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
549
550 /*
551 * inline extent creation worked or returned error,
552 * we don't need to create any more async work items.
553 * Unlock and free up our temp pages.
554 */
555 extent_clear_unlock_delalloc(inode, start, end, NULL,
556 clear_flags, PAGE_UNLOCK |
557 PAGE_CLEAR_DIRTY |
558 PAGE_SET_WRITEBACK |
559 page_error_op |
560 PAGE_END_WRITEBACK);
561 goto free_pages_out;
562 }
563 }
564
565 if (will_compress) {
566 /*
567 * we aren't doing an inline extent round the compressed size
568 * up to a block size boundary so the allocator does sane
569 * things
570 */
571 total_compressed = ALIGN(total_compressed, blocksize);
572
573 /*
574 * one last check to make sure the compression is really a
575 * win, compare the page count read with the blocks on disk
576 */
577 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
578 if (total_compressed >= total_in) {
579 will_compress = 0;
580 } else {
581 num_bytes = total_in;
582 }
583 }
584 if (!will_compress && pages) {
585 /*
586 * the compression code ran but failed to make things smaller,
587 * free any pages it allocated and our page pointer array
588 */
589 for (i = 0; i < nr_pages_ret; i++) {
590 WARN_ON(pages[i]->mapping);
591 page_cache_release(pages[i]);
592 }
593 kfree(pages);
594 pages = NULL;
595 total_compressed = 0;
596 nr_pages_ret = 0;
597
598 /* flag the file so we don't compress in the future */
599 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
600 !(BTRFS_I(inode)->force_compress)) {
601 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
602 }
603 }
604 if (will_compress) {
605 *num_added += 1;
606
607 /* the async work queues will take care of doing actual
608 * allocation on disk for these compressed pages,
609 * and will submit them to the elevator.
610 */
611 add_async_extent(async_cow, start, num_bytes,
612 total_compressed, pages, nr_pages_ret,
613 compress_type);
614
615 if (start + num_bytes < end) {
616 start += num_bytes;
617 pages = NULL;
618 cond_resched();
619 goto again;
620 }
621 } else {
622 cleanup_and_bail_uncompressed:
623 /*
624 * No compression, but we still need to write the pages in
625 * the file we've been given so far. redirty the locked
626 * page if it corresponds to our extent and set things up
627 * for the async work queue to run cow_file_range to do
628 * the normal delalloc dance
629 */
630 if (page_offset(locked_page) >= start &&
631 page_offset(locked_page) <= end) {
632 __set_page_dirty_nobuffers(locked_page);
633 /* unlocked later on in the async handlers */
634 }
635 if (redirty)
636 extent_range_redirty_for_io(inode, start, end);
637 add_async_extent(async_cow, start, end - start + 1,
638 0, NULL, 0, BTRFS_COMPRESS_NONE);
639 *num_added += 1;
640 }
641
642 return;
643
644 free_pages_out:
645 for (i = 0; i < nr_pages_ret; i++) {
646 WARN_ON(pages[i]->mapping);
647 page_cache_release(pages[i]);
648 }
649 kfree(pages);
650 }
651
652 static void free_async_extent_pages(struct async_extent *async_extent)
653 {
654 int i;
655
656 if (!async_extent->pages)
657 return;
658
659 for (i = 0; i < async_extent->nr_pages; i++) {
660 WARN_ON(async_extent->pages[i]->mapping);
661 page_cache_release(async_extent->pages[i]);
662 }
663 kfree(async_extent->pages);
664 async_extent->nr_pages = 0;
665 async_extent->pages = NULL;
666 }
667
668 /*
669 * phase two of compressed writeback. This is the ordered portion
670 * of the code, which only gets called in the order the work was
671 * queued. We walk all the async extents created by compress_file_range
672 * and send them down to the disk.
673 */
674 static noinline void submit_compressed_extents(struct inode *inode,
675 struct async_cow *async_cow)
676 {
677 struct async_extent *async_extent;
678 u64 alloc_hint = 0;
679 struct btrfs_key ins;
680 struct extent_map *em;
681 struct btrfs_root *root = BTRFS_I(inode)->root;
682 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
683 struct extent_io_tree *io_tree;
684 int ret = 0;
685
686 again:
687 while (!list_empty(&async_cow->extents)) {
688 async_extent = list_entry(async_cow->extents.next,
689 struct async_extent, list);
690 list_del(&async_extent->list);
691
692 io_tree = &BTRFS_I(inode)->io_tree;
693
694 retry:
695 /* did the compression code fall back to uncompressed IO? */
696 if (!async_extent->pages) {
697 int page_started = 0;
698 unsigned long nr_written = 0;
699
700 lock_extent(io_tree, async_extent->start,
701 async_extent->start +
702 async_extent->ram_size - 1);
703
704 /* allocate blocks */
705 ret = cow_file_range(inode, async_cow->locked_page,
706 async_extent->start,
707 async_extent->start +
708 async_extent->ram_size - 1,
709 &page_started, &nr_written, 0);
710
711 /* JDM XXX */
712
713 /*
714 * if page_started, cow_file_range inserted an
715 * inline extent and took care of all the unlocking
716 * and IO for us. Otherwise, we need to submit
717 * all those pages down to the drive.
718 */
719 if (!page_started && !ret)
720 extent_write_locked_range(io_tree,
721 inode, async_extent->start,
722 async_extent->start +
723 async_extent->ram_size - 1,
724 btrfs_get_extent,
725 WB_SYNC_ALL);
726 else if (ret)
727 unlock_page(async_cow->locked_page);
728 kfree(async_extent);
729 cond_resched();
730 continue;
731 }
732
733 lock_extent(io_tree, async_extent->start,
734 async_extent->start + async_extent->ram_size - 1);
735
736 ret = btrfs_reserve_extent(root,
737 async_extent->compressed_size,
738 async_extent->compressed_size,
739 0, alloc_hint, &ins, 1, 1);
740 if (ret) {
741 free_async_extent_pages(async_extent);
742
743 if (ret == -ENOSPC) {
744 unlock_extent(io_tree, async_extent->start,
745 async_extent->start +
746 async_extent->ram_size - 1);
747
748 /*
749 * we need to redirty the pages if we decide to
750 * fallback to uncompressed IO, otherwise we
751 * will not submit these pages down to lower
752 * layers.
753 */
754 extent_range_redirty_for_io(inode,
755 async_extent->start,
756 async_extent->start +
757 async_extent->ram_size - 1);
758
759 goto retry;
760 }
761 goto out_free;
762 }
763 /*
764 * here we're doing allocation and writeback of the
765 * compressed pages
766 */
767 btrfs_drop_extent_cache(inode, async_extent->start,
768 async_extent->start +
769 async_extent->ram_size - 1, 0);
770
771 em = alloc_extent_map();
772 if (!em) {
773 ret = -ENOMEM;
774 goto out_free_reserve;
775 }
776 em->start = async_extent->start;
777 em->len = async_extent->ram_size;
778 em->orig_start = em->start;
779 em->mod_start = em->start;
780 em->mod_len = em->len;
781
782 em->block_start = ins.objectid;
783 em->block_len = ins.offset;
784 em->orig_block_len = ins.offset;
785 em->ram_bytes = async_extent->ram_size;
786 em->bdev = root->fs_info->fs_devices->latest_bdev;
787 em->compress_type = async_extent->compress_type;
788 set_bit(EXTENT_FLAG_PINNED, &em->flags);
789 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
790 em->generation = -1;
791
792 while (1) {
793 write_lock(&em_tree->lock);
794 ret = add_extent_mapping(em_tree, em, 1);
795 write_unlock(&em_tree->lock);
796 if (ret != -EEXIST) {
797 free_extent_map(em);
798 break;
799 }
800 btrfs_drop_extent_cache(inode, async_extent->start,
801 async_extent->start +
802 async_extent->ram_size - 1, 0);
803 }
804
805 if (ret)
806 goto out_free_reserve;
807
808 ret = btrfs_add_ordered_extent_compress(inode,
809 async_extent->start,
810 ins.objectid,
811 async_extent->ram_size,
812 ins.offset,
813 BTRFS_ORDERED_COMPRESSED,
814 async_extent->compress_type);
815 if (ret) {
816 btrfs_drop_extent_cache(inode, async_extent->start,
817 async_extent->start +
818 async_extent->ram_size - 1, 0);
819 goto out_free_reserve;
820 }
821
822 /*
823 * clear dirty, set writeback and unlock the pages.
824 */
825 extent_clear_unlock_delalloc(inode, async_extent->start,
826 async_extent->start +
827 async_extent->ram_size - 1,
828 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
829 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
830 PAGE_SET_WRITEBACK);
831 ret = btrfs_submit_compressed_write(inode,
832 async_extent->start,
833 async_extent->ram_size,
834 ins.objectid,
835 ins.offset, async_extent->pages,
836 async_extent->nr_pages);
837 if (ret) {
838 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
839 struct page *p = async_extent->pages[0];
840 const u64 start = async_extent->start;
841 const u64 end = start + async_extent->ram_size - 1;
842
843 p->mapping = inode->i_mapping;
844 tree->ops->writepage_end_io_hook(p, start, end,
845 NULL, 0);
846 p->mapping = NULL;
847 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
848 PAGE_END_WRITEBACK |
849 PAGE_SET_ERROR);
850 free_async_extent_pages(async_extent);
851 }
852 alloc_hint = ins.objectid + ins.offset;
853 kfree(async_extent);
854 cond_resched();
855 }
856 return;
857 out_free_reserve:
858 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
859 out_free:
860 extent_clear_unlock_delalloc(inode, async_extent->start,
861 async_extent->start +
862 async_extent->ram_size - 1,
863 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
864 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
865 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
866 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
867 PAGE_SET_ERROR);
868 free_async_extent_pages(async_extent);
869 kfree(async_extent);
870 goto again;
871 }
872
873 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
874 u64 num_bytes)
875 {
876 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
877 struct extent_map *em;
878 u64 alloc_hint = 0;
879
880 read_lock(&em_tree->lock);
881 em = search_extent_mapping(em_tree, start, num_bytes);
882 if (em) {
883 /*
884 * if block start isn't an actual block number then find the
885 * first block in this inode and use that as a hint. If that
886 * block is also bogus then just don't worry about it.
887 */
888 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
889 free_extent_map(em);
890 em = search_extent_mapping(em_tree, 0, 0);
891 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
892 alloc_hint = em->block_start;
893 if (em)
894 free_extent_map(em);
895 } else {
896 alloc_hint = em->block_start;
897 free_extent_map(em);
898 }
899 }
900 read_unlock(&em_tree->lock);
901
902 return alloc_hint;
903 }
904
905 /*
906 * when extent_io.c finds a delayed allocation range in the file,
907 * the call backs end up in this code. The basic idea is to
908 * allocate extents on disk for the range, and create ordered data structs
909 * in ram to track those extents.
910 *
911 * locked_page is the page that writepage had locked already. We use
912 * it to make sure we don't do extra locks or unlocks.
913 *
914 * *page_started is set to one if we unlock locked_page and do everything
915 * required to start IO on it. It may be clean and already done with
916 * IO when we return.
917 */
918 static noinline int cow_file_range(struct inode *inode,
919 struct page *locked_page,
920 u64 start, u64 end, int *page_started,
921 unsigned long *nr_written,
922 int unlock)
923 {
924 struct btrfs_root *root = BTRFS_I(inode)->root;
925 u64 alloc_hint = 0;
926 u64 num_bytes;
927 unsigned long ram_size;
928 u64 disk_num_bytes;
929 u64 cur_alloc_size;
930 u64 blocksize = root->sectorsize;
931 struct btrfs_key ins;
932 struct extent_map *em;
933 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
934 int ret = 0;
935
936 if (btrfs_is_free_space_inode(inode)) {
937 WARN_ON_ONCE(1);
938 ret = -EINVAL;
939 goto out_unlock;
940 }
941
942 num_bytes = ALIGN(end - start + 1, blocksize);
943 num_bytes = max(blocksize, num_bytes);
944 disk_num_bytes = num_bytes;
945
946 /* if this is a small write inside eof, kick off defrag */
947 if (num_bytes < 64 * 1024 &&
948 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
949 btrfs_add_inode_defrag(NULL, inode);
950
951 if (start == 0) {
952 /* lets try to make an inline extent */
953 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
954 NULL);
955 if (ret == 0) {
956 extent_clear_unlock_delalloc(inode, start, end, NULL,
957 EXTENT_LOCKED | EXTENT_DELALLOC |
958 EXTENT_DEFRAG, PAGE_UNLOCK |
959 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
960 PAGE_END_WRITEBACK);
961
962 *nr_written = *nr_written +
963 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
964 *page_started = 1;
965 goto out;
966 } else if (ret < 0) {
967 goto out_unlock;
968 }
969 }
970
971 BUG_ON(disk_num_bytes >
972 btrfs_super_total_bytes(root->fs_info->super_copy));
973
974 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
975 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
976
977 while (disk_num_bytes > 0) {
978 unsigned long op;
979
980 cur_alloc_size = disk_num_bytes;
981 ret = btrfs_reserve_extent(root, cur_alloc_size,
982 root->sectorsize, 0, alloc_hint,
983 &ins, 1, 1);
984 if (ret < 0)
985 goto out_unlock;
986
987 em = alloc_extent_map();
988 if (!em) {
989 ret = -ENOMEM;
990 goto out_reserve;
991 }
992 em->start = start;
993 em->orig_start = em->start;
994 ram_size = ins.offset;
995 em->len = ins.offset;
996 em->mod_start = em->start;
997 em->mod_len = em->len;
998
999 em->block_start = ins.objectid;
1000 em->block_len = ins.offset;
1001 em->orig_block_len = ins.offset;
1002 em->ram_bytes = ram_size;
1003 em->bdev = root->fs_info->fs_devices->latest_bdev;
1004 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1005 em->generation = -1;
1006
1007 while (1) {
1008 write_lock(&em_tree->lock);
1009 ret = add_extent_mapping(em_tree, em, 1);
1010 write_unlock(&em_tree->lock);
1011 if (ret != -EEXIST) {
1012 free_extent_map(em);
1013 break;
1014 }
1015 btrfs_drop_extent_cache(inode, start,
1016 start + ram_size - 1, 0);
1017 }
1018 if (ret)
1019 goto out_reserve;
1020
1021 cur_alloc_size = ins.offset;
1022 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1023 ram_size, cur_alloc_size, 0);
1024 if (ret)
1025 goto out_drop_extent_cache;
1026
1027 if (root->root_key.objectid ==
1028 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1029 ret = btrfs_reloc_clone_csums(inode, start,
1030 cur_alloc_size);
1031 if (ret)
1032 goto out_drop_extent_cache;
1033 }
1034
1035 if (disk_num_bytes < cur_alloc_size)
1036 break;
1037
1038 /* we're not doing compressed IO, don't unlock the first
1039 * page (which the caller expects to stay locked), don't
1040 * clear any dirty bits and don't set any writeback bits
1041 *
1042 * Do set the Private2 bit so we know this page was properly
1043 * setup for writepage
1044 */
1045 op = unlock ? PAGE_UNLOCK : 0;
1046 op |= PAGE_SET_PRIVATE2;
1047
1048 extent_clear_unlock_delalloc(inode, start,
1049 start + ram_size - 1, locked_page,
1050 EXTENT_LOCKED | EXTENT_DELALLOC,
1051 op);
1052 disk_num_bytes -= cur_alloc_size;
1053 num_bytes -= cur_alloc_size;
1054 alloc_hint = ins.objectid + ins.offset;
1055 start += cur_alloc_size;
1056 }
1057 out:
1058 return ret;
1059
1060 out_drop_extent_cache:
1061 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1062 out_reserve:
1063 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1064 out_unlock:
1065 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1066 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1067 EXTENT_DELALLOC | EXTENT_DEFRAG,
1068 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1069 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1070 goto out;
1071 }
1072
1073 /*
1074 * work queue call back to started compression on a file and pages
1075 */
1076 static noinline void async_cow_start(struct btrfs_work *work)
1077 {
1078 struct async_cow *async_cow;
1079 int num_added = 0;
1080 async_cow = container_of(work, struct async_cow, work);
1081
1082 compress_file_range(async_cow->inode, async_cow->locked_page,
1083 async_cow->start, async_cow->end, async_cow,
1084 &num_added);
1085 if (num_added == 0) {
1086 btrfs_add_delayed_iput(async_cow->inode);
1087 async_cow->inode = NULL;
1088 }
1089 }
1090
1091 /*
1092 * work queue call back to submit previously compressed pages
1093 */
1094 static noinline void async_cow_submit(struct btrfs_work *work)
1095 {
1096 struct async_cow *async_cow;
1097 struct btrfs_root *root;
1098 unsigned long nr_pages;
1099
1100 async_cow = container_of(work, struct async_cow, work);
1101
1102 root = async_cow->root;
1103 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1104 PAGE_CACHE_SHIFT;
1105
1106 /*
1107 * atomic_sub_return implies a barrier for waitqueue_active
1108 */
1109 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1110 5 * 1024 * 1024 &&
1111 waitqueue_active(&root->fs_info->async_submit_wait))
1112 wake_up(&root->fs_info->async_submit_wait);
1113
1114 if (async_cow->inode)
1115 submit_compressed_extents(async_cow->inode, async_cow);
1116 }
1117
1118 static noinline void async_cow_free(struct btrfs_work *work)
1119 {
1120 struct async_cow *async_cow;
1121 async_cow = container_of(work, struct async_cow, work);
1122 if (async_cow->inode)
1123 btrfs_add_delayed_iput(async_cow->inode);
1124 kfree(async_cow);
1125 }
1126
1127 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1128 u64 start, u64 end, int *page_started,
1129 unsigned long *nr_written)
1130 {
1131 struct async_cow *async_cow;
1132 struct btrfs_root *root = BTRFS_I(inode)->root;
1133 unsigned long nr_pages;
1134 u64 cur_end;
1135 int limit = 10 * 1024 * 1024;
1136
1137 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1138 1, 0, NULL, GFP_NOFS);
1139 while (start < end) {
1140 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1141 BUG_ON(!async_cow); /* -ENOMEM */
1142 async_cow->inode = igrab(inode);
1143 async_cow->root = root;
1144 async_cow->locked_page = locked_page;
1145 async_cow->start = start;
1146
1147 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1148 !btrfs_test_opt(root, FORCE_COMPRESS))
1149 cur_end = end;
1150 else
1151 cur_end = min(end, start + 512 * 1024 - 1);
1152
1153 async_cow->end = cur_end;
1154 INIT_LIST_HEAD(&async_cow->extents);
1155
1156 btrfs_init_work(&async_cow->work,
1157 btrfs_delalloc_helper,
1158 async_cow_start, async_cow_submit,
1159 async_cow_free);
1160
1161 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1162 PAGE_CACHE_SHIFT;
1163 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1164
1165 btrfs_queue_work(root->fs_info->delalloc_workers,
1166 &async_cow->work);
1167
1168 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1169 wait_event(root->fs_info->async_submit_wait,
1170 (atomic_read(&root->fs_info->async_delalloc_pages) <
1171 limit));
1172 }
1173
1174 while (atomic_read(&root->fs_info->async_submit_draining) &&
1175 atomic_read(&root->fs_info->async_delalloc_pages)) {
1176 wait_event(root->fs_info->async_submit_wait,
1177 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1178 0));
1179 }
1180
1181 *nr_written += nr_pages;
1182 start = cur_end + 1;
1183 }
1184 *page_started = 1;
1185 return 0;
1186 }
1187
1188 static noinline int csum_exist_in_range(struct btrfs_root *root,
1189 u64 bytenr, u64 num_bytes)
1190 {
1191 int ret;
1192 struct btrfs_ordered_sum *sums;
1193 LIST_HEAD(list);
1194
1195 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1196 bytenr + num_bytes - 1, &list, 0);
1197 if (ret == 0 && list_empty(&list))
1198 return 0;
1199
1200 while (!list_empty(&list)) {
1201 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1202 list_del(&sums->list);
1203 kfree(sums);
1204 }
1205 return 1;
1206 }
1207
1208 /*
1209 * when nowcow writeback call back. This checks for snapshots or COW copies
1210 * of the extents that exist in the file, and COWs the file as required.
1211 *
1212 * If no cow copies or snapshots exist, we write directly to the existing
1213 * blocks on disk
1214 */
1215 static noinline int run_delalloc_nocow(struct inode *inode,
1216 struct page *locked_page,
1217 u64 start, u64 end, int *page_started, int force,
1218 unsigned long *nr_written)
1219 {
1220 struct btrfs_root *root = BTRFS_I(inode)->root;
1221 struct btrfs_trans_handle *trans;
1222 struct extent_buffer *leaf;
1223 struct btrfs_path *path;
1224 struct btrfs_file_extent_item *fi;
1225 struct btrfs_key found_key;
1226 u64 cow_start;
1227 u64 cur_offset;
1228 u64 extent_end;
1229 u64 extent_offset;
1230 u64 disk_bytenr;
1231 u64 num_bytes;
1232 u64 disk_num_bytes;
1233 u64 ram_bytes;
1234 int extent_type;
1235 int ret, err;
1236 int type;
1237 int nocow;
1238 int check_prev = 1;
1239 bool nolock;
1240 u64 ino = btrfs_ino(inode);
1241
1242 path = btrfs_alloc_path();
1243 if (!path) {
1244 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1245 EXTENT_LOCKED | EXTENT_DELALLOC |
1246 EXTENT_DO_ACCOUNTING |
1247 EXTENT_DEFRAG, PAGE_UNLOCK |
1248 PAGE_CLEAR_DIRTY |
1249 PAGE_SET_WRITEBACK |
1250 PAGE_END_WRITEBACK);
1251 return -ENOMEM;
1252 }
1253
1254 nolock = btrfs_is_free_space_inode(inode);
1255
1256 if (nolock)
1257 trans = btrfs_join_transaction_nolock(root);
1258 else
1259 trans = btrfs_join_transaction(root);
1260
1261 if (IS_ERR(trans)) {
1262 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1263 EXTENT_LOCKED | EXTENT_DELALLOC |
1264 EXTENT_DO_ACCOUNTING |
1265 EXTENT_DEFRAG, PAGE_UNLOCK |
1266 PAGE_CLEAR_DIRTY |
1267 PAGE_SET_WRITEBACK |
1268 PAGE_END_WRITEBACK);
1269 btrfs_free_path(path);
1270 return PTR_ERR(trans);
1271 }
1272
1273 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1274
1275 cow_start = (u64)-1;
1276 cur_offset = start;
1277 while (1) {
1278 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1279 cur_offset, 0);
1280 if (ret < 0)
1281 goto error;
1282 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1283 leaf = path->nodes[0];
1284 btrfs_item_key_to_cpu(leaf, &found_key,
1285 path->slots[0] - 1);
1286 if (found_key.objectid == ino &&
1287 found_key.type == BTRFS_EXTENT_DATA_KEY)
1288 path->slots[0]--;
1289 }
1290 check_prev = 0;
1291 next_slot:
1292 leaf = path->nodes[0];
1293 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1294 ret = btrfs_next_leaf(root, path);
1295 if (ret < 0)
1296 goto error;
1297 if (ret > 0)
1298 break;
1299 leaf = path->nodes[0];
1300 }
1301
1302 nocow = 0;
1303 disk_bytenr = 0;
1304 num_bytes = 0;
1305 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1306
1307 if (found_key.objectid > ino)
1308 break;
1309 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1310 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1311 path->slots[0]++;
1312 goto next_slot;
1313 }
1314 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1315 found_key.offset > end)
1316 break;
1317
1318 if (found_key.offset > cur_offset) {
1319 extent_end = found_key.offset;
1320 extent_type = 0;
1321 goto out_check;
1322 }
1323
1324 fi = btrfs_item_ptr(leaf, path->slots[0],
1325 struct btrfs_file_extent_item);
1326 extent_type = btrfs_file_extent_type(leaf, fi);
1327
1328 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1329 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1330 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1331 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1332 extent_offset = btrfs_file_extent_offset(leaf, fi);
1333 extent_end = found_key.offset +
1334 btrfs_file_extent_num_bytes(leaf, fi);
1335 disk_num_bytes =
1336 btrfs_file_extent_disk_num_bytes(leaf, fi);
1337 if (extent_end <= start) {
1338 path->slots[0]++;
1339 goto next_slot;
1340 }
1341 if (disk_bytenr == 0)
1342 goto out_check;
1343 if (btrfs_file_extent_compression(leaf, fi) ||
1344 btrfs_file_extent_encryption(leaf, fi) ||
1345 btrfs_file_extent_other_encoding(leaf, fi))
1346 goto out_check;
1347 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1348 goto out_check;
1349 if (btrfs_extent_readonly(root, disk_bytenr))
1350 goto out_check;
1351 if (btrfs_cross_ref_exist(trans, root, ino,
1352 found_key.offset -
1353 extent_offset, disk_bytenr))
1354 goto out_check;
1355 disk_bytenr += extent_offset;
1356 disk_bytenr += cur_offset - found_key.offset;
1357 num_bytes = min(end + 1, extent_end) - cur_offset;
1358 /*
1359 * if there are pending snapshots for this root,
1360 * we fall into common COW way.
1361 */
1362 if (!nolock) {
1363 err = btrfs_start_write_no_snapshoting(root);
1364 if (!err)
1365 goto out_check;
1366 }
1367 /*
1368 * force cow if csum exists in the range.
1369 * this ensure that csum for a given extent are
1370 * either valid or do not exist.
1371 */
1372 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1373 goto out_check;
1374 nocow = 1;
1375 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1376 extent_end = found_key.offset +
1377 btrfs_file_extent_inline_len(leaf,
1378 path->slots[0], fi);
1379 extent_end = ALIGN(extent_end, root->sectorsize);
1380 } else {
1381 BUG_ON(1);
1382 }
1383 out_check:
1384 if (extent_end <= start) {
1385 path->slots[0]++;
1386 if (!nolock && nocow)
1387 btrfs_end_write_no_snapshoting(root);
1388 goto next_slot;
1389 }
1390 if (!nocow) {
1391 if (cow_start == (u64)-1)
1392 cow_start = cur_offset;
1393 cur_offset = extent_end;
1394 if (cur_offset > end)
1395 break;
1396 path->slots[0]++;
1397 goto next_slot;
1398 }
1399
1400 btrfs_release_path(path);
1401 if (cow_start != (u64)-1) {
1402 ret = cow_file_range(inode, locked_page,
1403 cow_start, found_key.offset - 1,
1404 page_started, nr_written, 1);
1405 if (ret) {
1406 if (!nolock && nocow)
1407 btrfs_end_write_no_snapshoting(root);
1408 goto error;
1409 }
1410 cow_start = (u64)-1;
1411 }
1412
1413 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1414 struct extent_map *em;
1415 struct extent_map_tree *em_tree;
1416 em_tree = &BTRFS_I(inode)->extent_tree;
1417 em = alloc_extent_map();
1418 BUG_ON(!em); /* -ENOMEM */
1419 em->start = cur_offset;
1420 em->orig_start = found_key.offset - extent_offset;
1421 em->len = num_bytes;
1422 em->block_len = num_bytes;
1423 em->block_start = disk_bytenr;
1424 em->orig_block_len = disk_num_bytes;
1425 em->ram_bytes = ram_bytes;
1426 em->bdev = root->fs_info->fs_devices->latest_bdev;
1427 em->mod_start = em->start;
1428 em->mod_len = em->len;
1429 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1430 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1431 em->generation = -1;
1432 while (1) {
1433 write_lock(&em_tree->lock);
1434 ret = add_extent_mapping(em_tree, em, 1);
1435 write_unlock(&em_tree->lock);
1436 if (ret != -EEXIST) {
1437 free_extent_map(em);
1438 break;
1439 }
1440 btrfs_drop_extent_cache(inode, em->start,
1441 em->start + em->len - 1, 0);
1442 }
1443 type = BTRFS_ORDERED_PREALLOC;
1444 } else {
1445 type = BTRFS_ORDERED_NOCOW;
1446 }
1447
1448 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1449 num_bytes, num_bytes, type);
1450 BUG_ON(ret); /* -ENOMEM */
1451
1452 if (root->root_key.objectid ==
1453 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1454 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1455 num_bytes);
1456 if (ret) {
1457 if (!nolock && nocow)
1458 btrfs_end_write_no_snapshoting(root);
1459 goto error;
1460 }
1461 }
1462
1463 extent_clear_unlock_delalloc(inode, cur_offset,
1464 cur_offset + num_bytes - 1,
1465 locked_page, EXTENT_LOCKED |
1466 EXTENT_DELALLOC, PAGE_UNLOCK |
1467 PAGE_SET_PRIVATE2);
1468 if (!nolock && nocow)
1469 btrfs_end_write_no_snapshoting(root);
1470 cur_offset = extent_end;
1471 if (cur_offset > end)
1472 break;
1473 }
1474 btrfs_release_path(path);
1475
1476 if (cur_offset <= end && cow_start == (u64)-1) {
1477 cow_start = cur_offset;
1478 cur_offset = end;
1479 }
1480
1481 if (cow_start != (u64)-1) {
1482 ret = cow_file_range(inode, locked_page, cow_start, end,
1483 page_started, nr_written, 1);
1484 if (ret)
1485 goto error;
1486 }
1487
1488 error:
1489 err = btrfs_end_transaction(trans, root);
1490 if (!ret)
1491 ret = err;
1492
1493 if (ret && cur_offset < end)
1494 extent_clear_unlock_delalloc(inode, cur_offset, end,
1495 locked_page, EXTENT_LOCKED |
1496 EXTENT_DELALLOC | EXTENT_DEFRAG |
1497 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1498 PAGE_CLEAR_DIRTY |
1499 PAGE_SET_WRITEBACK |
1500 PAGE_END_WRITEBACK);
1501 btrfs_free_path(path);
1502 return ret;
1503 }
1504
1505 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1506 {
1507
1508 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1509 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1510 return 0;
1511
1512 /*
1513 * @defrag_bytes is a hint value, no spinlock held here,
1514 * if is not zero, it means the file is defragging.
1515 * Force cow if given extent needs to be defragged.
1516 */
1517 if (BTRFS_I(inode)->defrag_bytes &&
1518 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1519 EXTENT_DEFRAG, 0, NULL))
1520 return 1;
1521
1522 return 0;
1523 }
1524
1525 /*
1526 * extent_io.c call back to do delayed allocation processing
1527 */
1528 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1529 u64 start, u64 end, int *page_started,
1530 unsigned long *nr_written)
1531 {
1532 int ret;
1533 int force_cow = need_force_cow(inode, start, end);
1534
1535 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1536 ret = run_delalloc_nocow(inode, locked_page, start, end,
1537 page_started, 1, nr_written);
1538 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1539 ret = run_delalloc_nocow(inode, locked_page, start, end,
1540 page_started, 0, nr_written);
1541 } else if (!inode_need_compress(inode)) {
1542 ret = cow_file_range(inode, locked_page, start, end,
1543 page_started, nr_written, 1);
1544 } else {
1545 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1546 &BTRFS_I(inode)->runtime_flags);
1547 ret = cow_file_range_async(inode, locked_page, start, end,
1548 page_started, nr_written);
1549 }
1550 return ret;
1551 }
1552
1553 static void btrfs_split_extent_hook(struct inode *inode,
1554 struct extent_state *orig, u64 split)
1555 {
1556 u64 size;
1557
1558 /* not delalloc, ignore it */
1559 if (!(orig->state & EXTENT_DELALLOC))
1560 return;
1561
1562 size = orig->end - orig->start + 1;
1563 if (size > BTRFS_MAX_EXTENT_SIZE) {
1564 u64 num_extents;
1565 u64 new_size;
1566
1567 /*
1568 * See the explanation in btrfs_merge_extent_hook, the same
1569 * applies here, just in reverse.
1570 */
1571 new_size = orig->end - split + 1;
1572 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1573 BTRFS_MAX_EXTENT_SIZE);
1574 new_size = split - orig->start;
1575 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1576 BTRFS_MAX_EXTENT_SIZE);
1577 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1578 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1579 return;
1580 }
1581
1582 spin_lock(&BTRFS_I(inode)->lock);
1583 BTRFS_I(inode)->outstanding_extents++;
1584 spin_unlock(&BTRFS_I(inode)->lock);
1585 }
1586
1587 /*
1588 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1589 * extents so we can keep track of new extents that are just merged onto old
1590 * extents, such as when we are doing sequential writes, so we can properly
1591 * account for the metadata space we'll need.
1592 */
1593 static void btrfs_merge_extent_hook(struct inode *inode,
1594 struct extent_state *new,
1595 struct extent_state *other)
1596 {
1597 u64 new_size, old_size;
1598 u64 num_extents;
1599
1600 /* not delalloc, ignore it */
1601 if (!(other->state & EXTENT_DELALLOC))
1602 return;
1603
1604 if (new->start > other->start)
1605 new_size = new->end - other->start + 1;
1606 else
1607 new_size = other->end - new->start + 1;
1608
1609 /* we're not bigger than the max, unreserve the space and go */
1610 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1611 spin_lock(&BTRFS_I(inode)->lock);
1612 BTRFS_I(inode)->outstanding_extents--;
1613 spin_unlock(&BTRFS_I(inode)->lock);
1614 return;
1615 }
1616
1617 /*
1618 * We have to add up either side to figure out how many extents were
1619 * accounted for before we merged into one big extent. If the number of
1620 * extents we accounted for is <= the amount we need for the new range
1621 * then we can return, otherwise drop. Think of it like this
1622 *
1623 * [ 4k][MAX_SIZE]
1624 *
1625 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1626 * need 2 outstanding extents, on one side we have 1 and the other side
1627 * we have 1 so they are == and we can return. But in this case
1628 *
1629 * [MAX_SIZE+4k][MAX_SIZE+4k]
1630 *
1631 * Each range on their own accounts for 2 extents, but merged together
1632 * they are only 3 extents worth of accounting, so we need to drop in
1633 * this case.
1634 */
1635 old_size = other->end - other->start + 1;
1636 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1637 BTRFS_MAX_EXTENT_SIZE);
1638 old_size = new->end - new->start + 1;
1639 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1640 BTRFS_MAX_EXTENT_SIZE);
1641
1642 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1643 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1644 return;
1645
1646 spin_lock(&BTRFS_I(inode)->lock);
1647 BTRFS_I(inode)->outstanding_extents--;
1648 spin_unlock(&BTRFS_I(inode)->lock);
1649 }
1650
1651 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1652 struct inode *inode)
1653 {
1654 spin_lock(&root->delalloc_lock);
1655 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1656 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1657 &root->delalloc_inodes);
1658 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1659 &BTRFS_I(inode)->runtime_flags);
1660 root->nr_delalloc_inodes++;
1661 if (root->nr_delalloc_inodes == 1) {
1662 spin_lock(&root->fs_info->delalloc_root_lock);
1663 BUG_ON(!list_empty(&root->delalloc_root));
1664 list_add_tail(&root->delalloc_root,
1665 &root->fs_info->delalloc_roots);
1666 spin_unlock(&root->fs_info->delalloc_root_lock);
1667 }
1668 }
1669 spin_unlock(&root->delalloc_lock);
1670 }
1671
1672 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1673 struct inode *inode)
1674 {
1675 spin_lock(&root->delalloc_lock);
1676 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1677 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1678 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1679 &BTRFS_I(inode)->runtime_flags);
1680 root->nr_delalloc_inodes--;
1681 if (!root->nr_delalloc_inodes) {
1682 spin_lock(&root->fs_info->delalloc_root_lock);
1683 BUG_ON(list_empty(&root->delalloc_root));
1684 list_del_init(&root->delalloc_root);
1685 spin_unlock(&root->fs_info->delalloc_root_lock);
1686 }
1687 }
1688 spin_unlock(&root->delalloc_lock);
1689 }
1690
1691 /*
1692 * extent_io.c set_bit_hook, used to track delayed allocation
1693 * bytes in this file, and to maintain the list of inodes that
1694 * have pending delalloc work to be done.
1695 */
1696 static void btrfs_set_bit_hook(struct inode *inode,
1697 struct extent_state *state, unsigned *bits)
1698 {
1699
1700 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1701 WARN_ON(1);
1702 /*
1703 * set_bit and clear bit hooks normally require _irqsave/restore
1704 * but in this case, we are only testing for the DELALLOC
1705 * bit, which is only set or cleared with irqs on
1706 */
1707 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1708 struct btrfs_root *root = BTRFS_I(inode)->root;
1709 u64 len = state->end + 1 - state->start;
1710 bool do_list = !btrfs_is_free_space_inode(inode);
1711
1712 if (*bits & EXTENT_FIRST_DELALLOC) {
1713 *bits &= ~EXTENT_FIRST_DELALLOC;
1714 } else {
1715 spin_lock(&BTRFS_I(inode)->lock);
1716 BTRFS_I(inode)->outstanding_extents++;
1717 spin_unlock(&BTRFS_I(inode)->lock);
1718 }
1719
1720 /* For sanity tests */
1721 if (btrfs_test_is_dummy_root(root))
1722 return;
1723
1724 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1725 root->fs_info->delalloc_batch);
1726 spin_lock(&BTRFS_I(inode)->lock);
1727 BTRFS_I(inode)->delalloc_bytes += len;
1728 if (*bits & EXTENT_DEFRAG)
1729 BTRFS_I(inode)->defrag_bytes += len;
1730 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1731 &BTRFS_I(inode)->runtime_flags))
1732 btrfs_add_delalloc_inodes(root, inode);
1733 spin_unlock(&BTRFS_I(inode)->lock);
1734 }
1735 }
1736
1737 /*
1738 * extent_io.c clear_bit_hook, see set_bit_hook for why
1739 */
1740 static void btrfs_clear_bit_hook(struct inode *inode,
1741 struct extent_state *state,
1742 unsigned *bits)
1743 {
1744 u64 len = state->end + 1 - state->start;
1745 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1746 BTRFS_MAX_EXTENT_SIZE);
1747
1748 spin_lock(&BTRFS_I(inode)->lock);
1749 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1750 BTRFS_I(inode)->defrag_bytes -= len;
1751 spin_unlock(&BTRFS_I(inode)->lock);
1752
1753 /*
1754 * set_bit and clear bit hooks normally require _irqsave/restore
1755 * but in this case, we are only testing for the DELALLOC
1756 * bit, which is only set or cleared with irqs on
1757 */
1758 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1759 struct btrfs_root *root = BTRFS_I(inode)->root;
1760 bool do_list = !btrfs_is_free_space_inode(inode);
1761
1762 if (*bits & EXTENT_FIRST_DELALLOC) {
1763 *bits &= ~EXTENT_FIRST_DELALLOC;
1764 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1765 spin_lock(&BTRFS_I(inode)->lock);
1766 BTRFS_I(inode)->outstanding_extents -= num_extents;
1767 spin_unlock(&BTRFS_I(inode)->lock);
1768 }
1769
1770 /*
1771 * We don't reserve metadata space for space cache inodes so we
1772 * don't need to call dellalloc_release_metadata if there is an
1773 * error.
1774 */
1775 if (*bits & EXTENT_DO_ACCOUNTING &&
1776 root != root->fs_info->tree_root)
1777 btrfs_delalloc_release_metadata(inode, len);
1778
1779 /* For sanity tests. */
1780 if (btrfs_test_is_dummy_root(root))
1781 return;
1782
1783 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1784 && do_list && !(state->state & EXTENT_NORESERVE))
1785 btrfs_free_reserved_data_space_noquota(inode,
1786 state->start, len);
1787
1788 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1789 root->fs_info->delalloc_batch);
1790 spin_lock(&BTRFS_I(inode)->lock);
1791 BTRFS_I(inode)->delalloc_bytes -= len;
1792 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1793 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1794 &BTRFS_I(inode)->runtime_flags))
1795 btrfs_del_delalloc_inode(root, inode);
1796 spin_unlock(&BTRFS_I(inode)->lock);
1797 }
1798 }
1799
1800 /*
1801 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1802 * we don't create bios that span stripes or chunks
1803 */
1804 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1805 size_t size, struct bio *bio,
1806 unsigned long bio_flags)
1807 {
1808 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1809 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1810 u64 length = 0;
1811 u64 map_length;
1812 int ret;
1813
1814 if (bio_flags & EXTENT_BIO_COMPRESSED)
1815 return 0;
1816
1817 length = bio->bi_iter.bi_size;
1818 map_length = length;
1819 ret = btrfs_map_block(root->fs_info, rw, logical,
1820 &map_length, NULL, 0);
1821 /* Will always return 0 with map_multi == NULL */
1822 BUG_ON(ret < 0);
1823 if (map_length < length + size)
1824 return 1;
1825 return 0;
1826 }
1827
1828 /*
1829 * in order to insert checksums into the metadata in large chunks,
1830 * we wait until bio submission time. All the pages in the bio are
1831 * checksummed and sums are attached onto the ordered extent record.
1832 *
1833 * At IO completion time the cums attached on the ordered extent record
1834 * are inserted into the btree
1835 */
1836 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1837 struct bio *bio, int mirror_num,
1838 unsigned long bio_flags,
1839 u64 bio_offset)
1840 {
1841 struct btrfs_root *root = BTRFS_I(inode)->root;
1842 int ret = 0;
1843
1844 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1845 BUG_ON(ret); /* -ENOMEM */
1846 return 0;
1847 }
1848
1849 /*
1850 * in order to insert checksums into the metadata in large chunks,
1851 * we wait until bio submission time. All the pages in the bio are
1852 * checksummed and sums are attached onto the ordered extent record.
1853 *
1854 * At IO completion time the cums attached on the ordered extent record
1855 * are inserted into the btree
1856 */
1857 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1858 int mirror_num, unsigned long bio_flags,
1859 u64 bio_offset)
1860 {
1861 struct btrfs_root *root = BTRFS_I(inode)->root;
1862 int ret;
1863
1864 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1865 if (ret) {
1866 bio->bi_error = ret;
1867 bio_endio(bio);
1868 }
1869 return ret;
1870 }
1871
1872 /*
1873 * extent_io.c submission hook. This does the right thing for csum calculation
1874 * on write, or reading the csums from the tree before a read
1875 */
1876 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1877 int mirror_num, unsigned long bio_flags,
1878 u64 bio_offset)
1879 {
1880 struct btrfs_root *root = BTRFS_I(inode)->root;
1881 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1882 int ret = 0;
1883 int skip_sum;
1884 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1885
1886 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1887
1888 if (btrfs_is_free_space_inode(inode))
1889 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1890
1891 if (!(rw & REQ_WRITE)) {
1892 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1893 if (ret)
1894 goto out;
1895
1896 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1897 ret = btrfs_submit_compressed_read(inode, bio,
1898 mirror_num,
1899 bio_flags);
1900 goto out;
1901 } else if (!skip_sum) {
1902 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1903 if (ret)
1904 goto out;
1905 }
1906 goto mapit;
1907 } else if (async && !skip_sum) {
1908 /* csum items have already been cloned */
1909 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1910 goto mapit;
1911 /* we're doing a write, do the async checksumming */
1912 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1913 inode, rw, bio, mirror_num,
1914 bio_flags, bio_offset,
1915 __btrfs_submit_bio_start,
1916 __btrfs_submit_bio_done);
1917 goto out;
1918 } else if (!skip_sum) {
1919 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1920 if (ret)
1921 goto out;
1922 }
1923
1924 mapit:
1925 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1926
1927 out:
1928 if (ret < 0) {
1929 bio->bi_error = ret;
1930 bio_endio(bio);
1931 }
1932 return ret;
1933 }
1934
1935 /*
1936 * given a list of ordered sums record them in the inode. This happens
1937 * at IO completion time based on sums calculated at bio submission time.
1938 */
1939 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1940 struct inode *inode, u64 file_offset,
1941 struct list_head *list)
1942 {
1943 struct btrfs_ordered_sum *sum;
1944
1945 list_for_each_entry(sum, list, list) {
1946 trans->adding_csums = 1;
1947 btrfs_csum_file_blocks(trans,
1948 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1949 trans->adding_csums = 0;
1950 }
1951 return 0;
1952 }
1953
1954 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1955 struct extent_state **cached_state)
1956 {
1957 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1958 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1959 cached_state, GFP_NOFS);
1960 }
1961
1962 /* see btrfs_writepage_start_hook for details on why this is required */
1963 struct btrfs_writepage_fixup {
1964 struct page *page;
1965 struct btrfs_work work;
1966 };
1967
1968 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1969 {
1970 struct btrfs_writepage_fixup *fixup;
1971 struct btrfs_ordered_extent *ordered;
1972 struct extent_state *cached_state = NULL;
1973 struct page *page;
1974 struct inode *inode;
1975 u64 page_start;
1976 u64 page_end;
1977 int ret;
1978
1979 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1980 page = fixup->page;
1981 again:
1982 lock_page(page);
1983 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1984 ClearPageChecked(page);
1985 goto out_page;
1986 }
1987
1988 inode = page->mapping->host;
1989 page_start = page_offset(page);
1990 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1991
1992 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1993 &cached_state);
1994
1995 /* already ordered? We're done */
1996 if (PagePrivate2(page))
1997 goto out;
1998
1999 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2000 if (ordered) {
2001 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2002 page_end, &cached_state, GFP_NOFS);
2003 unlock_page(page);
2004 btrfs_start_ordered_extent(inode, ordered, 1);
2005 btrfs_put_ordered_extent(ordered);
2006 goto again;
2007 }
2008
2009 ret = btrfs_delalloc_reserve_space(inode, page_start,
2010 PAGE_CACHE_SIZE);
2011 if (ret) {
2012 mapping_set_error(page->mapping, ret);
2013 end_extent_writepage(page, ret, page_start, page_end);
2014 ClearPageChecked(page);
2015 goto out;
2016 }
2017
2018 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2019 ClearPageChecked(page);
2020 set_page_dirty(page);
2021 out:
2022 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2023 &cached_state, GFP_NOFS);
2024 out_page:
2025 unlock_page(page);
2026 page_cache_release(page);
2027 kfree(fixup);
2028 }
2029
2030 /*
2031 * There are a few paths in the higher layers of the kernel that directly
2032 * set the page dirty bit without asking the filesystem if it is a
2033 * good idea. This causes problems because we want to make sure COW
2034 * properly happens and the data=ordered rules are followed.
2035 *
2036 * In our case any range that doesn't have the ORDERED bit set
2037 * hasn't been properly setup for IO. We kick off an async process
2038 * to fix it up. The async helper will wait for ordered extents, set
2039 * the delalloc bit and make it safe to write the page.
2040 */
2041 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2042 {
2043 struct inode *inode = page->mapping->host;
2044 struct btrfs_writepage_fixup *fixup;
2045 struct btrfs_root *root = BTRFS_I(inode)->root;
2046
2047 /* this page is properly in the ordered list */
2048 if (TestClearPagePrivate2(page))
2049 return 0;
2050
2051 if (PageChecked(page))
2052 return -EAGAIN;
2053
2054 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2055 if (!fixup)
2056 return -EAGAIN;
2057
2058 SetPageChecked(page);
2059 page_cache_get(page);
2060 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2061 btrfs_writepage_fixup_worker, NULL, NULL);
2062 fixup->page = page;
2063 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2064 return -EBUSY;
2065 }
2066
2067 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2068 struct inode *inode, u64 file_pos,
2069 u64 disk_bytenr, u64 disk_num_bytes,
2070 u64 num_bytes, u64 ram_bytes,
2071 u8 compression, u8 encryption,
2072 u16 other_encoding, int extent_type)
2073 {
2074 struct btrfs_root *root = BTRFS_I(inode)->root;
2075 struct btrfs_file_extent_item *fi;
2076 struct btrfs_path *path;
2077 struct extent_buffer *leaf;
2078 struct btrfs_key ins;
2079 int extent_inserted = 0;
2080 int ret;
2081
2082 path = btrfs_alloc_path();
2083 if (!path)
2084 return -ENOMEM;
2085
2086 /*
2087 * we may be replacing one extent in the tree with another.
2088 * The new extent is pinned in the extent map, and we don't want
2089 * to drop it from the cache until it is completely in the btree.
2090 *
2091 * So, tell btrfs_drop_extents to leave this extent in the cache.
2092 * the caller is expected to unpin it and allow it to be merged
2093 * with the others.
2094 */
2095 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2096 file_pos + num_bytes, NULL, 0,
2097 1, sizeof(*fi), &extent_inserted);
2098 if (ret)
2099 goto out;
2100
2101 if (!extent_inserted) {
2102 ins.objectid = btrfs_ino(inode);
2103 ins.offset = file_pos;
2104 ins.type = BTRFS_EXTENT_DATA_KEY;
2105
2106 path->leave_spinning = 1;
2107 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2108 sizeof(*fi));
2109 if (ret)
2110 goto out;
2111 }
2112 leaf = path->nodes[0];
2113 fi = btrfs_item_ptr(leaf, path->slots[0],
2114 struct btrfs_file_extent_item);
2115 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2116 btrfs_set_file_extent_type(leaf, fi, extent_type);
2117 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2118 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2119 btrfs_set_file_extent_offset(leaf, fi, 0);
2120 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2121 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2122 btrfs_set_file_extent_compression(leaf, fi, compression);
2123 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2124 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2125
2126 btrfs_mark_buffer_dirty(leaf);
2127 btrfs_release_path(path);
2128
2129 inode_add_bytes(inode, num_bytes);
2130
2131 ins.objectid = disk_bytenr;
2132 ins.offset = disk_num_bytes;
2133 ins.type = BTRFS_EXTENT_ITEM_KEY;
2134 ret = btrfs_alloc_reserved_file_extent(trans, root,
2135 root->root_key.objectid,
2136 btrfs_ino(inode), file_pos,
2137 ram_bytes, &ins);
2138 /*
2139 * Release the reserved range from inode dirty range map, as it is
2140 * already moved into delayed_ref_head
2141 */
2142 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2143 out:
2144 btrfs_free_path(path);
2145
2146 return ret;
2147 }
2148
2149 /* snapshot-aware defrag */
2150 struct sa_defrag_extent_backref {
2151 struct rb_node node;
2152 struct old_sa_defrag_extent *old;
2153 u64 root_id;
2154 u64 inum;
2155 u64 file_pos;
2156 u64 extent_offset;
2157 u64 num_bytes;
2158 u64 generation;
2159 };
2160
2161 struct old_sa_defrag_extent {
2162 struct list_head list;
2163 struct new_sa_defrag_extent *new;
2164
2165 u64 extent_offset;
2166 u64 bytenr;
2167 u64 offset;
2168 u64 len;
2169 int count;
2170 };
2171
2172 struct new_sa_defrag_extent {
2173 struct rb_root root;
2174 struct list_head head;
2175 struct btrfs_path *path;
2176 struct inode *inode;
2177 u64 file_pos;
2178 u64 len;
2179 u64 bytenr;
2180 u64 disk_len;
2181 u8 compress_type;
2182 };
2183
2184 static int backref_comp(struct sa_defrag_extent_backref *b1,
2185 struct sa_defrag_extent_backref *b2)
2186 {
2187 if (b1->root_id < b2->root_id)
2188 return -1;
2189 else if (b1->root_id > b2->root_id)
2190 return 1;
2191
2192 if (b1->inum < b2->inum)
2193 return -1;
2194 else if (b1->inum > b2->inum)
2195 return 1;
2196
2197 if (b1->file_pos < b2->file_pos)
2198 return -1;
2199 else if (b1->file_pos > b2->file_pos)
2200 return 1;
2201
2202 /*
2203 * [------------------------------] ===> (a range of space)
2204 * |<--->| |<---->| =============> (fs/file tree A)
2205 * |<---------------------------->| ===> (fs/file tree B)
2206 *
2207 * A range of space can refer to two file extents in one tree while
2208 * refer to only one file extent in another tree.
2209 *
2210 * So we may process a disk offset more than one time(two extents in A)
2211 * and locate at the same extent(one extent in B), then insert two same
2212 * backrefs(both refer to the extent in B).
2213 */
2214 return 0;
2215 }
2216
2217 static void backref_insert(struct rb_root *root,
2218 struct sa_defrag_extent_backref *backref)
2219 {
2220 struct rb_node **p = &root->rb_node;
2221 struct rb_node *parent = NULL;
2222 struct sa_defrag_extent_backref *entry;
2223 int ret;
2224
2225 while (*p) {
2226 parent = *p;
2227 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2228
2229 ret = backref_comp(backref, entry);
2230 if (ret < 0)
2231 p = &(*p)->rb_left;
2232 else
2233 p = &(*p)->rb_right;
2234 }
2235
2236 rb_link_node(&backref->node, parent, p);
2237 rb_insert_color(&backref->node, root);
2238 }
2239
2240 /*
2241 * Note the backref might has changed, and in this case we just return 0.
2242 */
2243 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2244 void *ctx)
2245 {
2246 struct btrfs_file_extent_item *extent;
2247 struct btrfs_fs_info *fs_info;
2248 struct old_sa_defrag_extent *old = ctx;
2249 struct new_sa_defrag_extent *new = old->new;
2250 struct btrfs_path *path = new->path;
2251 struct btrfs_key key;
2252 struct btrfs_root *root;
2253 struct sa_defrag_extent_backref *backref;
2254 struct extent_buffer *leaf;
2255 struct inode *inode = new->inode;
2256 int slot;
2257 int ret;
2258 u64 extent_offset;
2259 u64 num_bytes;
2260
2261 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2262 inum == btrfs_ino(inode))
2263 return 0;
2264
2265 key.objectid = root_id;
2266 key.type = BTRFS_ROOT_ITEM_KEY;
2267 key.offset = (u64)-1;
2268
2269 fs_info = BTRFS_I(inode)->root->fs_info;
2270 root = btrfs_read_fs_root_no_name(fs_info, &key);
2271 if (IS_ERR(root)) {
2272 if (PTR_ERR(root) == -ENOENT)
2273 return 0;
2274 WARN_ON(1);
2275 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2276 inum, offset, root_id);
2277 return PTR_ERR(root);
2278 }
2279
2280 key.objectid = inum;
2281 key.type = BTRFS_EXTENT_DATA_KEY;
2282 if (offset > (u64)-1 << 32)
2283 key.offset = 0;
2284 else
2285 key.offset = offset;
2286
2287 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2288 if (WARN_ON(ret < 0))
2289 return ret;
2290 ret = 0;
2291
2292 while (1) {
2293 cond_resched();
2294
2295 leaf = path->nodes[0];
2296 slot = path->slots[0];
2297
2298 if (slot >= btrfs_header_nritems(leaf)) {
2299 ret = btrfs_next_leaf(root, path);
2300 if (ret < 0) {
2301 goto out;
2302 } else if (ret > 0) {
2303 ret = 0;
2304 goto out;
2305 }
2306 continue;
2307 }
2308
2309 path->slots[0]++;
2310
2311 btrfs_item_key_to_cpu(leaf, &key, slot);
2312
2313 if (key.objectid > inum)
2314 goto out;
2315
2316 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2317 continue;
2318
2319 extent = btrfs_item_ptr(leaf, slot,
2320 struct btrfs_file_extent_item);
2321
2322 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2323 continue;
2324
2325 /*
2326 * 'offset' refers to the exact key.offset,
2327 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2328 * (key.offset - extent_offset).
2329 */
2330 if (key.offset != offset)
2331 continue;
2332
2333 extent_offset = btrfs_file_extent_offset(leaf, extent);
2334 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2335
2336 if (extent_offset >= old->extent_offset + old->offset +
2337 old->len || extent_offset + num_bytes <=
2338 old->extent_offset + old->offset)
2339 continue;
2340 break;
2341 }
2342
2343 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2344 if (!backref) {
2345 ret = -ENOENT;
2346 goto out;
2347 }
2348
2349 backref->root_id = root_id;
2350 backref->inum = inum;
2351 backref->file_pos = offset;
2352 backref->num_bytes = num_bytes;
2353 backref->extent_offset = extent_offset;
2354 backref->generation = btrfs_file_extent_generation(leaf, extent);
2355 backref->old = old;
2356 backref_insert(&new->root, backref);
2357 old->count++;
2358 out:
2359 btrfs_release_path(path);
2360 WARN_ON(ret);
2361 return ret;
2362 }
2363
2364 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2365 struct new_sa_defrag_extent *new)
2366 {
2367 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2368 struct old_sa_defrag_extent *old, *tmp;
2369 int ret;
2370
2371 new->path = path;
2372
2373 list_for_each_entry_safe(old, tmp, &new->head, list) {
2374 ret = iterate_inodes_from_logical(old->bytenr +
2375 old->extent_offset, fs_info,
2376 path, record_one_backref,
2377 old);
2378 if (ret < 0 && ret != -ENOENT)
2379 return false;
2380
2381 /* no backref to be processed for this extent */
2382 if (!old->count) {
2383 list_del(&old->list);
2384 kfree(old);
2385 }
2386 }
2387
2388 if (list_empty(&new->head))
2389 return false;
2390
2391 return true;
2392 }
2393
2394 static int relink_is_mergable(struct extent_buffer *leaf,
2395 struct btrfs_file_extent_item *fi,
2396 struct new_sa_defrag_extent *new)
2397 {
2398 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2399 return 0;
2400
2401 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2402 return 0;
2403
2404 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2405 return 0;
2406
2407 if (btrfs_file_extent_encryption(leaf, fi) ||
2408 btrfs_file_extent_other_encoding(leaf, fi))
2409 return 0;
2410
2411 return 1;
2412 }
2413
2414 /*
2415 * Note the backref might has changed, and in this case we just return 0.
2416 */
2417 static noinline int relink_extent_backref(struct btrfs_path *path,
2418 struct sa_defrag_extent_backref *prev,
2419 struct sa_defrag_extent_backref *backref)
2420 {
2421 struct btrfs_file_extent_item *extent;
2422 struct btrfs_file_extent_item *item;
2423 struct btrfs_ordered_extent *ordered;
2424 struct btrfs_trans_handle *trans;
2425 struct btrfs_fs_info *fs_info;
2426 struct btrfs_root *root;
2427 struct btrfs_key key;
2428 struct extent_buffer *leaf;
2429 struct old_sa_defrag_extent *old = backref->old;
2430 struct new_sa_defrag_extent *new = old->new;
2431 struct inode *src_inode = new->inode;
2432 struct inode *inode;
2433 struct extent_state *cached = NULL;
2434 int ret = 0;
2435 u64 start;
2436 u64 len;
2437 u64 lock_start;
2438 u64 lock_end;
2439 bool merge = false;
2440 int index;
2441
2442 if (prev && prev->root_id == backref->root_id &&
2443 prev->inum == backref->inum &&
2444 prev->file_pos + prev->num_bytes == backref->file_pos)
2445 merge = true;
2446
2447 /* step 1: get root */
2448 key.objectid = backref->root_id;
2449 key.type = BTRFS_ROOT_ITEM_KEY;
2450 key.offset = (u64)-1;
2451
2452 fs_info = BTRFS_I(src_inode)->root->fs_info;
2453 index = srcu_read_lock(&fs_info->subvol_srcu);
2454
2455 root = btrfs_read_fs_root_no_name(fs_info, &key);
2456 if (IS_ERR(root)) {
2457 srcu_read_unlock(&fs_info->subvol_srcu, index);
2458 if (PTR_ERR(root) == -ENOENT)
2459 return 0;
2460 return PTR_ERR(root);
2461 }
2462
2463 if (btrfs_root_readonly(root)) {
2464 srcu_read_unlock(&fs_info->subvol_srcu, index);
2465 return 0;
2466 }
2467
2468 /* step 2: get inode */
2469 key.objectid = backref->inum;
2470 key.type = BTRFS_INODE_ITEM_KEY;
2471 key.offset = 0;
2472
2473 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2474 if (IS_ERR(inode)) {
2475 srcu_read_unlock(&fs_info->subvol_srcu, index);
2476 return 0;
2477 }
2478
2479 srcu_read_unlock(&fs_info->subvol_srcu, index);
2480
2481 /* step 3: relink backref */
2482 lock_start = backref->file_pos;
2483 lock_end = backref->file_pos + backref->num_bytes - 1;
2484 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2485 0, &cached);
2486
2487 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2488 if (ordered) {
2489 btrfs_put_ordered_extent(ordered);
2490 goto out_unlock;
2491 }
2492
2493 trans = btrfs_join_transaction(root);
2494 if (IS_ERR(trans)) {
2495 ret = PTR_ERR(trans);
2496 goto out_unlock;
2497 }
2498
2499 key.objectid = backref->inum;
2500 key.type = BTRFS_EXTENT_DATA_KEY;
2501 key.offset = backref->file_pos;
2502
2503 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2504 if (ret < 0) {
2505 goto out_free_path;
2506 } else if (ret > 0) {
2507 ret = 0;
2508 goto out_free_path;
2509 }
2510
2511 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2512 struct btrfs_file_extent_item);
2513
2514 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2515 backref->generation)
2516 goto out_free_path;
2517
2518 btrfs_release_path(path);
2519
2520 start = backref->file_pos;
2521 if (backref->extent_offset < old->extent_offset + old->offset)
2522 start += old->extent_offset + old->offset -
2523 backref->extent_offset;
2524
2525 len = min(backref->extent_offset + backref->num_bytes,
2526 old->extent_offset + old->offset + old->len);
2527 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2528
2529 ret = btrfs_drop_extents(trans, root, inode, start,
2530 start + len, 1);
2531 if (ret)
2532 goto out_free_path;
2533 again:
2534 key.objectid = btrfs_ino(inode);
2535 key.type = BTRFS_EXTENT_DATA_KEY;
2536 key.offset = start;
2537
2538 path->leave_spinning = 1;
2539 if (merge) {
2540 struct btrfs_file_extent_item *fi;
2541 u64 extent_len;
2542 struct btrfs_key found_key;
2543
2544 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2545 if (ret < 0)
2546 goto out_free_path;
2547
2548 path->slots[0]--;
2549 leaf = path->nodes[0];
2550 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2551
2552 fi = btrfs_item_ptr(leaf, path->slots[0],
2553 struct btrfs_file_extent_item);
2554 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2555
2556 if (extent_len + found_key.offset == start &&
2557 relink_is_mergable(leaf, fi, new)) {
2558 btrfs_set_file_extent_num_bytes(leaf, fi,
2559 extent_len + len);
2560 btrfs_mark_buffer_dirty(leaf);
2561 inode_add_bytes(inode, len);
2562
2563 ret = 1;
2564 goto out_free_path;
2565 } else {
2566 merge = false;
2567 btrfs_release_path(path);
2568 goto again;
2569 }
2570 }
2571
2572 ret = btrfs_insert_empty_item(trans, root, path, &key,
2573 sizeof(*extent));
2574 if (ret) {
2575 btrfs_abort_transaction(trans, root, ret);
2576 goto out_free_path;
2577 }
2578
2579 leaf = path->nodes[0];
2580 item = btrfs_item_ptr(leaf, path->slots[0],
2581 struct btrfs_file_extent_item);
2582 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2583 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2584 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2585 btrfs_set_file_extent_num_bytes(leaf, item, len);
2586 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2587 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2588 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2589 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2590 btrfs_set_file_extent_encryption(leaf, item, 0);
2591 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2592
2593 btrfs_mark_buffer_dirty(leaf);
2594 inode_add_bytes(inode, len);
2595 btrfs_release_path(path);
2596
2597 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2598 new->disk_len, 0,
2599 backref->root_id, backref->inum,
2600 new->file_pos); /* start - extent_offset */
2601 if (ret) {
2602 btrfs_abort_transaction(trans, root, ret);
2603 goto out_free_path;
2604 }
2605
2606 ret = 1;
2607 out_free_path:
2608 btrfs_release_path(path);
2609 path->leave_spinning = 0;
2610 btrfs_end_transaction(trans, root);
2611 out_unlock:
2612 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2613 &cached, GFP_NOFS);
2614 iput(inode);
2615 return ret;
2616 }
2617
2618 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2619 {
2620 struct old_sa_defrag_extent *old, *tmp;
2621
2622 if (!new)
2623 return;
2624
2625 list_for_each_entry_safe(old, tmp, &new->head, list) {
2626 kfree(old);
2627 }
2628 kfree(new);
2629 }
2630
2631 static void relink_file_extents(struct new_sa_defrag_extent *new)
2632 {
2633 struct btrfs_path *path;
2634 struct sa_defrag_extent_backref *backref;
2635 struct sa_defrag_extent_backref *prev = NULL;
2636 struct inode *inode;
2637 struct btrfs_root *root;
2638 struct rb_node *node;
2639 int ret;
2640
2641 inode = new->inode;
2642 root = BTRFS_I(inode)->root;
2643
2644 path = btrfs_alloc_path();
2645 if (!path)
2646 return;
2647
2648 if (!record_extent_backrefs(path, new)) {
2649 btrfs_free_path(path);
2650 goto out;
2651 }
2652 btrfs_release_path(path);
2653
2654 while (1) {
2655 node = rb_first(&new->root);
2656 if (!node)
2657 break;
2658 rb_erase(node, &new->root);
2659
2660 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2661
2662 ret = relink_extent_backref(path, prev, backref);
2663 WARN_ON(ret < 0);
2664
2665 kfree(prev);
2666
2667 if (ret == 1)
2668 prev = backref;
2669 else
2670 prev = NULL;
2671 cond_resched();
2672 }
2673 kfree(prev);
2674
2675 btrfs_free_path(path);
2676 out:
2677 free_sa_defrag_extent(new);
2678
2679 atomic_dec(&root->fs_info->defrag_running);
2680 wake_up(&root->fs_info->transaction_wait);
2681 }
2682
2683 static struct new_sa_defrag_extent *
2684 record_old_file_extents(struct inode *inode,
2685 struct btrfs_ordered_extent *ordered)
2686 {
2687 struct btrfs_root *root = BTRFS_I(inode)->root;
2688 struct btrfs_path *path;
2689 struct btrfs_key key;
2690 struct old_sa_defrag_extent *old;
2691 struct new_sa_defrag_extent *new;
2692 int ret;
2693
2694 new = kmalloc(sizeof(*new), GFP_NOFS);
2695 if (!new)
2696 return NULL;
2697
2698 new->inode = inode;
2699 new->file_pos = ordered->file_offset;
2700 new->len = ordered->len;
2701 new->bytenr = ordered->start;
2702 new->disk_len = ordered->disk_len;
2703 new->compress_type = ordered->compress_type;
2704 new->root = RB_ROOT;
2705 INIT_LIST_HEAD(&new->head);
2706
2707 path = btrfs_alloc_path();
2708 if (!path)
2709 goto out_kfree;
2710
2711 key.objectid = btrfs_ino(inode);
2712 key.type = BTRFS_EXTENT_DATA_KEY;
2713 key.offset = new->file_pos;
2714
2715 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2716 if (ret < 0)
2717 goto out_free_path;
2718 if (ret > 0 && path->slots[0] > 0)
2719 path->slots[0]--;
2720
2721 /* find out all the old extents for the file range */
2722 while (1) {
2723 struct btrfs_file_extent_item *extent;
2724 struct extent_buffer *l;
2725 int slot;
2726 u64 num_bytes;
2727 u64 offset;
2728 u64 end;
2729 u64 disk_bytenr;
2730 u64 extent_offset;
2731
2732 l = path->nodes[0];
2733 slot = path->slots[0];
2734
2735 if (slot >= btrfs_header_nritems(l)) {
2736 ret = btrfs_next_leaf(root, path);
2737 if (ret < 0)
2738 goto out_free_path;
2739 else if (ret > 0)
2740 break;
2741 continue;
2742 }
2743
2744 btrfs_item_key_to_cpu(l, &key, slot);
2745
2746 if (key.objectid != btrfs_ino(inode))
2747 break;
2748 if (key.type != BTRFS_EXTENT_DATA_KEY)
2749 break;
2750 if (key.offset >= new->file_pos + new->len)
2751 break;
2752
2753 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2754
2755 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2756 if (key.offset + num_bytes < new->file_pos)
2757 goto next;
2758
2759 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2760 if (!disk_bytenr)
2761 goto next;
2762
2763 extent_offset = btrfs_file_extent_offset(l, extent);
2764
2765 old = kmalloc(sizeof(*old), GFP_NOFS);
2766 if (!old)
2767 goto out_free_path;
2768
2769 offset = max(new->file_pos, key.offset);
2770 end = min(new->file_pos + new->len, key.offset + num_bytes);
2771
2772 old->bytenr = disk_bytenr;
2773 old->extent_offset = extent_offset;
2774 old->offset = offset - key.offset;
2775 old->len = end - offset;
2776 old->new = new;
2777 old->count = 0;
2778 list_add_tail(&old->list, &new->head);
2779 next:
2780 path->slots[0]++;
2781 cond_resched();
2782 }
2783
2784 btrfs_free_path(path);
2785 atomic_inc(&root->fs_info->defrag_running);
2786
2787 return new;
2788
2789 out_free_path:
2790 btrfs_free_path(path);
2791 out_kfree:
2792 free_sa_defrag_extent(new);
2793 return NULL;
2794 }
2795
2796 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2797 u64 start, u64 len)
2798 {
2799 struct btrfs_block_group_cache *cache;
2800
2801 cache = btrfs_lookup_block_group(root->fs_info, start);
2802 ASSERT(cache);
2803
2804 spin_lock(&cache->lock);
2805 cache->delalloc_bytes -= len;
2806 spin_unlock(&cache->lock);
2807
2808 btrfs_put_block_group(cache);
2809 }
2810
2811 /* as ordered data IO finishes, this gets called so we can finish
2812 * an ordered extent if the range of bytes in the file it covers are
2813 * fully written.
2814 */
2815 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2816 {
2817 struct inode *inode = ordered_extent->inode;
2818 struct btrfs_root *root = BTRFS_I(inode)->root;
2819 struct btrfs_trans_handle *trans = NULL;
2820 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2821 struct extent_state *cached_state = NULL;
2822 struct new_sa_defrag_extent *new = NULL;
2823 int compress_type = 0;
2824 int ret = 0;
2825 u64 logical_len = ordered_extent->len;
2826 bool nolock;
2827 bool truncated = false;
2828
2829 nolock = btrfs_is_free_space_inode(inode);
2830
2831 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2832 ret = -EIO;
2833 goto out;
2834 }
2835
2836 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2837 ordered_extent->file_offset +
2838 ordered_extent->len - 1);
2839
2840 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2841 truncated = true;
2842 logical_len = ordered_extent->truncated_len;
2843 /* Truncated the entire extent, don't bother adding */
2844 if (!logical_len)
2845 goto out;
2846 }
2847
2848 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2849 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2850
2851 /*
2852 * For mwrite(mmap + memset to write) case, we still reserve
2853 * space for NOCOW range.
2854 * As NOCOW won't cause a new delayed ref, just free the space
2855 */
2856 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2857 ordered_extent->len);
2858 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2859 if (nolock)
2860 trans = btrfs_join_transaction_nolock(root);
2861 else
2862 trans = btrfs_join_transaction(root);
2863 if (IS_ERR(trans)) {
2864 ret = PTR_ERR(trans);
2865 trans = NULL;
2866 goto out;
2867 }
2868 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2869 ret = btrfs_update_inode_fallback(trans, root, inode);
2870 if (ret) /* -ENOMEM or corruption */
2871 btrfs_abort_transaction(trans, root, ret);
2872 goto out;
2873 }
2874
2875 lock_extent_bits(io_tree, ordered_extent->file_offset,
2876 ordered_extent->file_offset + ordered_extent->len - 1,
2877 0, &cached_state);
2878
2879 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2880 ordered_extent->file_offset + ordered_extent->len - 1,
2881 EXTENT_DEFRAG, 1, cached_state);
2882 if (ret) {
2883 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2884 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2885 /* the inode is shared */
2886 new = record_old_file_extents(inode, ordered_extent);
2887
2888 clear_extent_bit(io_tree, ordered_extent->file_offset,
2889 ordered_extent->file_offset + ordered_extent->len - 1,
2890 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2891 }
2892
2893 if (nolock)
2894 trans = btrfs_join_transaction_nolock(root);
2895 else
2896 trans = btrfs_join_transaction(root);
2897 if (IS_ERR(trans)) {
2898 ret = PTR_ERR(trans);
2899 trans = NULL;
2900 goto out_unlock;
2901 }
2902
2903 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2904
2905 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2906 compress_type = ordered_extent->compress_type;
2907 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2908 BUG_ON(compress_type);
2909 ret = btrfs_mark_extent_written(trans, inode,
2910 ordered_extent->file_offset,
2911 ordered_extent->file_offset +
2912 logical_len);
2913 } else {
2914 BUG_ON(root == root->fs_info->tree_root);
2915 ret = insert_reserved_file_extent(trans, inode,
2916 ordered_extent->file_offset,
2917 ordered_extent->start,
2918 ordered_extent->disk_len,
2919 logical_len, logical_len,
2920 compress_type, 0, 0,
2921 BTRFS_FILE_EXTENT_REG);
2922 if (!ret)
2923 btrfs_release_delalloc_bytes(root,
2924 ordered_extent->start,
2925 ordered_extent->disk_len);
2926 }
2927 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2928 ordered_extent->file_offset, ordered_extent->len,
2929 trans->transid);
2930 if (ret < 0) {
2931 btrfs_abort_transaction(trans, root, ret);
2932 goto out_unlock;
2933 }
2934
2935 add_pending_csums(trans, inode, ordered_extent->file_offset,
2936 &ordered_extent->list);
2937
2938 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2939 ret = btrfs_update_inode_fallback(trans, root, inode);
2940 if (ret) { /* -ENOMEM or corruption */
2941 btrfs_abort_transaction(trans, root, ret);
2942 goto out_unlock;
2943 }
2944 ret = 0;
2945 out_unlock:
2946 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2947 ordered_extent->file_offset +
2948 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2949 out:
2950 if (root != root->fs_info->tree_root)
2951 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2952 if (trans)
2953 btrfs_end_transaction(trans, root);
2954
2955 if (ret || truncated) {
2956 u64 start, end;
2957
2958 if (truncated)
2959 start = ordered_extent->file_offset + logical_len;
2960 else
2961 start = ordered_extent->file_offset;
2962 end = ordered_extent->file_offset + ordered_extent->len - 1;
2963 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2964
2965 /* Drop the cache for the part of the extent we didn't write. */
2966 btrfs_drop_extent_cache(inode, start, end, 0);
2967
2968 /*
2969 * If the ordered extent had an IOERR or something else went
2970 * wrong we need to return the space for this ordered extent
2971 * back to the allocator. We only free the extent in the
2972 * truncated case if we didn't write out the extent at all.
2973 */
2974 if ((ret || !logical_len) &&
2975 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2976 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2977 btrfs_free_reserved_extent(root, ordered_extent->start,
2978 ordered_extent->disk_len, 1);
2979 }
2980
2981
2982 /*
2983 * This needs to be done to make sure anybody waiting knows we are done
2984 * updating everything for this ordered extent.
2985 */
2986 btrfs_remove_ordered_extent(inode, ordered_extent);
2987
2988 /* for snapshot-aware defrag */
2989 if (new) {
2990 if (ret) {
2991 free_sa_defrag_extent(new);
2992 atomic_dec(&root->fs_info->defrag_running);
2993 } else {
2994 relink_file_extents(new);
2995 }
2996 }
2997
2998 /* once for us */
2999 btrfs_put_ordered_extent(ordered_extent);
3000 /* once for the tree */
3001 btrfs_put_ordered_extent(ordered_extent);
3002
3003 return ret;
3004 }
3005
3006 static void finish_ordered_fn(struct btrfs_work *work)
3007 {
3008 struct btrfs_ordered_extent *ordered_extent;
3009 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3010 btrfs_finish_ordered_io(ordered_extent);
3011 }
3012
3013 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3014 struct extent_state *state, int uptodate)
3015 {
3016 struct inode *inode = page->mapping->host;
3017 struct btrfs_root *root = BTRFS_I(inode)->root;
3018 struct btrfs_ordered_extent *ordered_extent = NULL;
3019 struct btrfs_workqueue *wq;
3020 btrfs_work_func_t func;
3021
3022 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3023
3024 ClearPagePrivate2(page);
3025 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3026 end - start + 1, uptodate))
3027 return 0;
3028
3029 if (btrfs_is_free_space_inode(inode)) {
3030 wq = root->fs_info->endio_freespace_worker;
3031 func = btrfs_freespace_write_helper;
3032 } else {
3033 wq = root->fs_info->endio_write_workers;
3034 func = btrfs_endio_write_helper;
3035 }
3036
3037 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3038 NULL);
3039 btrfs_queue_work(wq, &ordered_extent->work);
3040
3041 return 0;
3042 }
3043
3044 static int __readpage_endio_check(struct inode *inode,
3045 struct btrfs_io_bio *io_bio,
3046 int icsum, struct page *page,
3047 int pgoff, u64 start, size_t len)
3048 {
3049 char *kaddr;
3050 u32 csum_expected;
3051 u32 csum = ~(u32)0;
3052
3053 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3054
3055 kaddr = kmap_atomic(page);
3056 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3057 btrfs_csum_final(csum, (char *)&csum);
3058 if (csum != csum_expected)
3059 goto zeroit;
3060
3061 kunmap_atomic(kaddr);
3062 return 0;
3063 zeroit:
3064 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3065 "csum failed ino %llu off %llu csum %u expected csum %u",
3066 btrfs_ino(inode), start, csum, csum_expected);
3067 memset(kaddr + pgoff, 1, len);
3068 flush_dcache_page(page);
3069 kunmap_atomic(kaddr);
3070 if (csum_expected == 0)
3071 return 0;
3072 return -EIO;
3073 }
3074
3075 /*
3076 * when reads are done, we need to check csums to verify the data is correct
3077 * if there's a match, we allow the bio to finish. If not, the code in
3078 * extent_io.c will try to find good copies for us.
3079 */
3080 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3081 u64 phy_offset, struct page *page,
3082 u64 start, u64 end, int mirror)
3083 {
3084 size_t offset = start - page_offset(page);
3085 struct inode *inode = page->mapping->host;
3086 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3087 struct btrfs_root *root = BTRFS_I(inode)->root;
3088
3089 if (PageChecked(page)) {
3090 ClearPageChecked(page);
3091 return 0;
3092 }
3093
3094 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3095 return 0;
3096
3097 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3098 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3099 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3100 GFP_NOFS);
3101 return 0;
3102 }
3103
3104 phy_offset >>= inode->i_sb->s_blocksize_bits;
3105 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3106 start, (size_t)(end - start + 1));
3107 }
3108
3109 struct delayed_iput {
3110 struct list_head list;
3111 struct inode *inode;
3112 };
3113
3114 /* JDM: If this is fs-wide, why can't we add a pointer to
3115 * btrfs_inode instead and avoid the allocation? */
3116 void btrfs_add_delayed_iput(struct inode *inode)
3117 {
3118 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3119 struct delayed_iput *delayed;
3120
3121 if (atomic_add_unless(&inode->i_count, -1, 1))
3122 return;
3123
3124 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3125 delayed->inode = inode;
3126
3127 spin_lock(&fs_info->delayed_iput_lock);
3128 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3129 spin_unlock(&fs_info->delayed_iput_lock);
3130 }
3131
3132 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3133 {
3134 LIST_HEAD(list);
3135 struct btrfs_fs_info *fs_info = root->fs_info;
3136 struct delayed_iput *delayed;
3137 int empty;
3138
3139 spin_lock(&fs_info->delayed_iput_lock);
3140 empty = list_empty(&fs_info->delayed_iputs);
3141 spin_unlock(&fs_info->delayed_iput_lock);
3142 if (empty)
3143 return;
3144
3145 down_read(&fs_info->delayed_iput_sem);
3146
3147 spin_lock(&fs_info->delayed_iput_lock);
3148 list_splice_init(&fs_info->delayed_iputs, &list);
3149 spin_unlock(&fs_info->delayed_iput_lock);
3150
3151 while (!list_empty(&list)) {
3152 delayed = list_entry(list.next, struct delayed_iput, list);
3153 list_del(&delayed->list);
3154 iput(delayed->inode);
3155 kfree(delayed);
3156 }
3157
3158 up_read(&root->fs_info->delayed_iput_sem);
3159 }
3160
3161 /*
3162 * This is called in transaction commit time. If there are no orphan
3163 * files in the subvolume, it removes orphan item and frees block_rsv
3164 * structure.
3165 */
3166 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3167 struct btrfs_root *root)
3168 {
3169 struct btrfs_block_rsv *block_rsv;
3170 int ret;
3171
3172 if (atomic_read(&root->orphan_inodes) ||
3173 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3174 return;
3175
3176 spin_lock(&root->orphan_lock);
3177 if (atomic_read(&root->orphan_inodes)) {
3178 spin_unlock(&root->orphan_lock);
3179 return;
3180 }
3181
3182 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3183 spin_unlock(&root->orphan_lock);
3184 return;
3185 }
3186
3187 block_rsv = root->orphan_block_rsv;
3188 root->orphan_block_rsv = NULL;
3189 spin_unlock(&root->orphan_lock);
3190
3191 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3192 btrfs_root_refs(&root->root_item) > 0) {
3193 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3194 root->root_key.objectid);
3195 if (ret)
3196 btrfs_abort_transaction(trans, root, ret);
3197 else
3198 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3199 &root->state);
3200 }
3201
3202 if (block_rsv) {
3203 WARN_ON(block_rsv->size > 0);
3204 btrfs_free_block_rsv(root, block_rsv);
3205 }
3206 }
3207
3208 /*
3209 * This creates an orphan entry for the given inode in case something goes
3210 * wrong in the middle of an unlink/truncate.
3211 *
3212 * NOTE: caller of this function should reserve 5 units of metadata for
3213 * this function.
3214 */
3215 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3216 {
3217 struct btrfs_root *root = BTRFS_I(inode)->root;
3218 struct btrfs_block_rsv *block_rsv = NULL;
3219 int reserve = 0;
3220 int insert = 0;
3221 int ret;
3222
3223 if (!root->orphan_block_rsv) {
3224 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3225 if (!block_rsv)
3226 return -ENOMEM;
3227 }
3228
3229 spin_lock(&root->orphan_lock);
3230 if (!root->orphan_block_rsv) {
3231 root->orphan_block_rsv = block_rsv;
3232 } else if (block_rsv) {
3233 btrfs_free_block_rsv(root, block_rsv);
3234 block_rsv = NULL;
3235 }
3236
3237 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3238 &BTRFS_I(inode)->runtime_flags)) {
3239 #if 0
3240 /*
3241 * For proper ENOSPC handling, we should do orphan
3242 * cleanup when mounting. But this introduces backward
3243 * compatibility issue.
3244 */
3245 if (!xchg(&root->orphan_item_inserted, 1))
3246 insert = 2;
3247 else
3248 insert = 1;
3249 #endif
3250 insert = 1;
3251 atomic_inc(&root->orphan_inodes);
3252 }
3253
3254 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3255 &BTRFS_I(inode)->runtime_flags))
3256 reserve = 1;
3257 spin_unlock(&root->orphan_lock);
3258
3259 /* grab metadata reservation from transaction handle */
3260 if (reserve) {
3261 ret = btrfs_orphan_reserve_metadata(trans, inode);
3262 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3263 }
3264
3265 /* insert an orphan item to track this unlinked/truncated file */
3266 if (insert >= 1) {
3267 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3268 if (ret) {
3269 atomic_dec(&root->orphan_inodes);
3270 if (reserve) {
3271 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3272 &BTRFS_I(inode)->runtime_flags);
3273 btrfs_orphan_release_metadata(inode);
3274 }
3275 if (ret != -EEXIST) {
3276 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3277 &BTRFS_I(inode)->runtime_flags);
3278 btrfs_abort_transaction(trans, root, ret);
3279 return ret;
3280 }
3281 }
3282 ret = 0;
3283 }
3284
3285 /* insert an orphan item to track subvolume contains orphan files */
3286 if (insert >= 2) {
3287 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3288 root->root_key.objectid);
3289 if (ret && ret != -EEXIST) {
3290 btrfs_abort_transaction(trans, root, ret);
3291 return ret;
3292 }
3293 }
3294 return 0;
3295 }
3296
3297 /*
3298 * We have done the truncate/delete so we can go ahead and remove the orphan
3299 * item for this particular inode.
3300 */
3301 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3302 struct inode *inode)
3303 {
3304 struct btrfs_root *root = BTRFS_I(inode)->root;
3305 int delete_item = 0;
3306 int release_rsv = 0;
3307 int ret = 0;
3308
3309 spin_lock(&root->orphan_lock);
3310 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3311 &BTRFS_I(inode)->runtime_flags))
3312 delete_item = 1;
3313
3314 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3315 &BTRFS_I(inode)->runtime_flags))
3316 release_rsv = 1;
3317 spin_unlock(&root->orphan_lock);
3318
3319 if (delete_item) {
3320 atomic_dec(&root->orphan_inodes);
3321 if (trans)
3322 ret = btrfs_del_orphan_item(trans, root,
3323 btrfs_ino(inode));
3324 }
3325
3326 if (release_rsv)
3327 btrfs_orphan_release_metadata(inode);
3328
3329 return ret;
3330 }
3331
3332 /*
3333 * this cleans up any orphans that may be left on the list from the last use
3334 * of this root.
3335 */
3336 int btrfs_orphan_cleanup(struct btrfs_root *root)
3337 {
3338 struct btrfs_path *path;
3339 struct extent_buffer *leaf;
3340 struct btrfs_key key, found_key;
3341 struct btrfs_trans_handle *trans;
3342 struct inode *inode;
3343 u64 last_objectid = 0;
3344 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3345
3346 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3347 return 0;
3348
3349 path = btrfs_alloc_path();
3350 if (!path) {
3351 ret = -ENOMEM;
3352 goto out;
3353 }
3354 path->reada = -1;
3355
3356 key.objectid = BTRFS_ORPHAN_OBJECTID;
3357 key.type = BTRFS_ORPHAN_ITEM_KEY;
3358 key.offset = (u64)-1;
3359
3360 while (1) {
3361 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3362 if (ret < 0)
3363 goto out;
3364
3365 /*
3366 * if ret == 0 means we found what we were searching for, which
3367 * is weird, but possible, so only screw with path if we didn't
3368 * find the key and see if we have stuff that matches
3369 */
3370 if (ret > 0) {
3371 ret = 0;
3372 if (path->slots[0] == 0)
3373 break;
3374 path->slots[0]--;
3375 }
3376
3377 /* pull out the item */
3378 leaf = path->nodes[0];
3379 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3380
3381 /* make sure the item matches what we want */
3382 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3383 break;
3384 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3385 break;
3386
3387 /* release the path since we're done with it */
3388 btrfs_release_path(path);
3389
3390 /*
3391 * this is where we are basically btrfs_lookup, without the
3392 * crossing root thing. we store the inode number in the
3393 * offset of the orphan item.
3394 */
3395
3396 if (found_key.offset == last_objectid) {
3397 btrfs_err(root->fs_info,
3398 "Error removing orphan entry, stopping orphan cleanup");
3399 ret = -EINVAL;
3400 goto out;
3401 }
3402
3403 last_objectid = found_key.offset;
3404
3405 found_key.objectid = found_key.offset;
3406 found_key.type = BTRFS_INODE_ITEM_KEY;
3407 found_key.offset = 0;
3408 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3409 ret = PTR_ERR_OR_ZERO(inode);
3410 if (ret && ret != -ESTALE)
3411 goto out;
3412
3413 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3414 struct btrfs_root *dead_root;
3415 struct btrfs_fs_info *fs_info = root->fs_info;
3416 int is_dead_root = 0;
3417
3418 /*
3419 * this is an orphan in the tree root. Currently these
3420 * could come from 2 sources:
3421 * a) a snapshot deletion in progress
3422 * b) a free space cache inode
3423 * We need to distinguish those two, as the snapshot
3424 * orphan must not get deleted.
3425 * find_dead_roots already ran before us, so if this
3426 * is a snapshot deletion, we should find the root
3427 * in the dead_roots list
3428 */
3429 spin_lock(&fs_info->trans_lock);
3430 list_for_each_entry(dead_root, &fs_info->dead_roots,
3431 root_list) {
3432 if (dead_root->root_key.objectid ==
3433 found_key.objectid) {
3434 is_dead_root = 1;
3435 break;
3436 }
3437 }
3438 spin_unlock(&fs_info->trans_lock);
3439 if (is_dead_root) {
3440 /* prevent this orphan from being found again */
3441 key.offset = found_key.objectid - 1;
3442 continue;
3443 }
3444 }
3445 /*
3446 * Inode is already gone but the orphan item is still there,
3447 * kill the orphan item.
3448 */
3449 if (ret == -ESTALE) {
3450 trans = btrfs_start_transaction(root, 1);
3451 if (IS_ERR(trans)) {
3452 ret = PTR_ERR(trans);
3453 goto out;
3454 }
3455 btrfs_debug(root->fs_info, "auto deleting %Lu",
3456 found_key.objectid);
3457 ret = btrfs_del_orphan_item(trans, root,
3458 found_key.objectid);
3459 btrfs_end_transaction(trans, root);
3460 if (ret)
3461 goto out;
3462 continue;
3463 }
3464
3465 /*
3466 * add this inode to the orphan list so btrfs_orphan_del does
3467 * the proper thing when we hit it
3468 */
3469 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3470 &BTRFS_I(inode)->runtime_flags);
3471 atomic_inc(&root->orphan_inodes);
3472
3473 /* if we have links, this was a truncate, lets do that */
3474 if (inode->i_nlink) {
3475 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3476 iput(inode);
3477 continue;
3478 }
3479 nr_truncate++;
3480
3481 /* 1 for the orphan item deletion. */
3482 trans = btrfs_start_transaction(root, 1);
3483 if (IS_ERR(trans)) {
3484 iput(inode);
3485 ret = PTR_ERR(trans);
3486 goto out;
3487 }
3488 ret = btrfs_orphan_add(trans, inode);
3489 btrfs_end_transaction(trans, root);
3490 if (ret) {
3491 iput(inode);
3492 goto out;
3493 }
3494
3495 ret = btrfs_truncate(inode);
3496 if (ret)
3497 btrfs_orphan_del(NULL, inode);
3498 } else {
3499 nr_unlink++;
3500 }
3501
3502 /* this will do delete_inode and everything for us */
3503 iput(inode);
3504 if (ret)
3505 goto out;
3506 }
3507 /* release the path since we're done with it */
3508 btrfs_release_path(path);
3509
3510 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3511
3512 if (root->orphan_block_rsv)
3513 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3514 (u64)-1);
3515
3516 if (root->orphan_block_rsv ||
3517 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3518 trans = btrfs_join_transaction(root);
3519 if (!IS_ERR(trans))
3520 btrfs_end_transaction(trans, root);
3521 }
3522
3523 if (nr_unlink)
3524 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3525 if (nr_truncate)
3526 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3527
3528 out:
3529 if (ret)
3530 btrfs_err(root->fs_info,
3531 "could not do orphan cleanup %d", ret);
3532 btrfs_free_path(path);
3533 return ret;
3534 }
3535
3536 /*
3537 * very simple check to peek ahead in the leaf looking for xattrs. If we
3538 * don't find any xattrs, we know there can't be any acls.
3539 *
3540 * slot is the slot the inode is in, objectid is the objectid of the inode
3541 */
3542 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3543 int slot, u64 objectid,
3544 int *first_xattr_slot)
3545 {
3546 u32 nritems = btrfs_header_nritems(leaf);
3547 struct btrfs_key found_key;
3548 static u64 xattr_access = 0;
3549 static u64 xattr_default = 0;
3550 int scanned = 0;
3551
3552 if (!xattr_access) {
3553 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3554 strlen(POSIX_ACL_XATTR_ACCESS));
3555 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3556 strlen(POSIX_ACL_XATTR_DEFAULT));
3557 }
3558
3559 slot++;
3560 *first_xattr_slot = -1;
3561 while (slot < nritems) {
3562 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3563
3564 /* we found a different objectid, there must not be acls */
3565 if (found_key.objectid != objectid)
3566 return 0;
3567
3568 /* we found an xattr, assume we've got an acl */
3569 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3570 if (*first_xattr_slot == -1)
3571 *first_xattr_slot = slot;
3572 if (found_key.offset == xattr_access ||
3573 found_key.offset == xattr_default)
3574 return 1;
3575 }
3576
3577 /*
3578 * we found a key greater than an xattr key, there can't
3579 * be any acls later on
3580 */
3581 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3582 return 0;
3583
3584 slot++;
3585 scanned++;
3586
3587 /*
3588 * it goes inode, inode backrefs, xattrs, extents,
3589 * so if there are a ton of hard links to an inode there can
3590 * be a lot of backrefs. Don't waste time searching too hard,
3591 * this is just an optimization
3592 */
3593 if (scanned >= 8)
3594 break;
3595 }
3596 /* we hit the end of the leaf before we found an xattr or
3597 * something larger than an xattr. We have to assume the inode
3598 * has acls
3599 */
3600 if (*first_xattr_slot == -1)
3601 *first_xattr_slot = slot;
3602 return 1;
3603 }
3604
3605 /*
3606 * read an inode from the btree into the in-memory inode
3607 */
3608 static void btrfs_read_locked_inode(struct inode *inode)
3609 {
3610 struct btrfs_path *path;
3611 struct extent_buffer *leaf;
3612 struct btrfs_inode_item *inode_item;
3613 struct btrfs_root *root = BTRFS_I(inode)->root;
3614 struct btrfs_key location;
3615 unsigned long ptr;
3616 int maybe_acls;
3617 u32 rdev;
3618 int ret;
3619 bool filled = false;
3620 int first_xattr_slot;
3621
3622 ret = btrfs_fill_inode(inode, &rdev);
3623 if (!ret)
3624 filled = true;
3625
3626 path = btrfs_alloc_path();
3627 if (!path)
3628 goto make_bad;
3629
3630 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3631
3632 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3633 if (ret)
3634 goto make_bad;
3635
3636 leaf = path->nodes[0];
3637
3638 if (filled)
3639 goto cache_index;
3640
3641 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3642 struct btrfs_inode_item);
3643 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3644 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3645 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3646 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3647 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3648
3649 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3650 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3651
3652 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3653 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3654
3655 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3656 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3657
3658 BTRFS_I(inode)->i_otime.tv_sec =
3659 btrfs_timespec_sec(leaf, &inode_item->otime);
3660 BTRFS_I(inode)->i_otime.tv_nsec =
3661 btrfs_timespec_nsec(leaf, &inode_item->otime);
3662
3663 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3664 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3665 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3666
3667 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3668 inode->i_generation = BTRFS_I(inode)->generation;
3669 inode->i_rdev = 0;
3670 rdev = btrfs_inode_rdev(leaf, inode_item);
3671
3672 BTRFS_I(inode)->index_cnt = (u64)-1;
3673 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3674
3675 cache_index:
3676 /*
3677 * If we were modified in the current generation and evicted from memory
3678 * and then re-read we need to do a full sync since we don't have any
3679 * idea about which extents were modified before we were evicted from
3680 * cache.
3681 *
3682 * This is required for both inode re-read from disk and delayed inode
3683 * in delayed_nodes_tree.
3684 */
3685 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3686 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3687 &BTRFS_I(inode)->runtime_flags);
3688
3689 /*
3690 * We don't persist the id of the transaction where an unlink operation
3691 * against the inode was last made. So here we assume the inode might
3692 * have been evicted, and therefore the exact value of last_unlink_trans
3693 * lost, and set it to last_trans to avoid metadata inconsistencies
3694 * between the inode and its parent if the inode is fsync'ed and the log
3695 * replayed. For example, in the scenario:
3696 *
3697 * touch mydir/foo
3698 * ln mydir/foo mydir/bar
3699 * sync
3700 * unlink mydir/bar
3701 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3702 * xfs_io -c fsync mydir/foo
3703 * <power failure>
3704 * mount fs, triggers fsync log replay
3705 *
3706 * We must make sure that when we fsync our inode foo we also log its
3707 * parent inode, otherwise after log replay the parent still has the
3708 * dentry with the "bar" name but our inode foo has a link count of 1
3709 * and doesn't have an inode ref with the name "bar" anymore.
3710 *
3711 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3712 * but it guarantees correctness at the expense of ocassional full
3713 * transaction commits on fsync if our inode is a directory, or if our
3714 * inode is not a directory, logging its parent unnecessarily.
3715 */
3716 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3717
3718 path->slots[0]++;
3719 if (inode->i_nlink != 1 ||
3720 path->slots[0] >= btrfs_header_nritems(leaf))
3721 goto cache_acl;
3722
3723 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3724 if (location.objectid != btrfs_ino(inode))
3725 goto cache_acl;
3726
3727 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3728 if (location.type == BTRFS_INODE_REF_KEY) {
3729 struct btrfs_inode_ref *ref;
3730
3731 ref = (struct btrfs_inode_ref *)ptr;
3732 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3733 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3734 struct btrfs_inode_extref *extref;
3735
3736 extref = (struct btrfs_inode_extref *)ptr;
3737 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3738 extref);
3739 }
3740 cache_acl:
3741 /*
3742 * try to precache a NULL acl entry for files that don't have
3743 * any xattrs or acls
3744 */
3745 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3746 btrfs_ino(inode), &first_xattr_slot);
3747 if (first_xattr_slot != -1) {
3748 path->slots[0] = first_xattr_slot;
3749 ret = btrfs_load_inode_props(inode, path);
3750 if (ret)
3751 btrfs_err(root->fs_info,
3752 "error loading props for ino %llu (root %llu): %d",
3753 btrfs_ino(inode),
3754 root->root_key.objectid, ret);
3755 }
3756 btrfs_free_path(path);
3757
3758 if (!maybe_acls)
3759 cache_no_acl(inode);
3760
3761 switch (inode->i_mode & S_IFMT) {
3762 case S_IFREG:
3763 inode->i_mapping->a_ops = &btrfs_aops;
3764 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3765 inode->i_fop = &btrfs_file_operations;
3766 inode->i_op = &btrfs_file_inode_operations;
3767 break;
3768 case S_IFDIR:
3769 inode->i_fop = &btrfs_dir_file_operations;
3770 if (root == root->fs_info->tree_root)
3771 inode->i_op = &btrfs_dir_ro_inode_operations;
3772 else
3773 inode->i_op = &btrfs_dir_inode_operations;
3774 break;
3775 case S_IFLNK:
3776 inode->i_op = &btrfs_symlink_inode_operations;
3777 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3778 break;
3779 default:
3780 inode->i_op = &btrfs_special_inode_operations;
3781 init_special_inode(inode, inode->i_mode, rdev);
3782 break;
3783 }
3784
3785 btrfs_update_iflags(inode);
3786 return;
3787
3788 make_bad:
3789 btrfs_free_path(path);
3790 make_bad_inode(inode);
3791 }
3792
3793 /*
3794 * given a leaf and an inode, copy the inode fields into the leaf
3795 */
3796 static void fill_inode_item(struct btrfs_trans_handle *trans,
3797 struct extent_buffer *leaf,
3798 struct btrfs_inode_item *item,
3799 struct inode *inode)
3800 {
3801 struct btrfs_map_token token;
3802
3803 btrfs_init_map_token(&token);
3804
3805 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3806 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3807 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3808 &token);
3809 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3810 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3811
3812 btrfs_set_token_timespec_sec(leaf, &item->atime,
3813 inode->i_atime.tv_sec, &token);
3814 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3815 inode->i_atime.tv_nsec, &token);
3816
3817 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3818 inode->i_mtime.tv_sec, &token);
3819 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3820 inode->i_mtime.tv_nsec, &token);
3821
3822 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3823 inode->i_ctime.tv_sec, &token);
3824 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3825 inode->i_ctime.tv_nsec, &token);
3826
3827 btrfs_set_token_timespec_sec(leaf, &item->otime,
3828 BTRFS_I(inode)->i_otime.tv_sec, &token);
3829 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3830 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3831
3832 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3833 &token);
3834 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3835 &token);
3836 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3837 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3838 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3839 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3840 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3841 }
3842
3843 /*
3844 * copy everything in the in-memory inode into the btree.
3845 */
3846 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3847 struct btrfs_root *root, struct inode *inode)
3848 {
3849 struct btrfs_inode_item *inode_item;
3850 struct btrfs_path *path;
3851 struct extent_buffer *leaf;
3852 int ret;
3853
3854 path = btrfs_alloc_path();
3855 if (!path)
3856 return -ENOMEM;
3857
3858 path->leave_spinning = 1;
3859 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3860 1);
3861 if (ret) {
3862 if (ret > 0)
3863 ret = -ENOENT;
3864 goto failed;
3865 }
3866
3867 leaf = path->nodes[0];
3868 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3869 struct btrfs_inode_item);
3870
3871 fill_inode_item(trans, leaf, inode_item, inode);
3872 btrfs_mark_buffer_dirty(leaf);
3873 btrfs_set_inode_last_trans(trans, inode);
3874 ret = 0;
3875 failed:
3876 btrfs_free_path(path);
3877 return ret;
3878 }
3879
3880 /*
3881 * copy everything in the in-memory inode into the btree.
3882 */
3883 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3884 struct btrfs_root *root, struct inode *inode)
3885 {
3886 int ret;
3887
3888 /*
3889 * If the inode is a free space inode, we can deadlock during commit
3890 * if we put it into the delayed code.
3891 *
3892 * The data relocation inode should also be directly updated
3893 * without delay
3894 */
3895 if (!btrfs_is_free_space_inode(inode)
3896 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3897 && !root->fs_info->log_root_recovering) {
3898 btrfs_update_root_times(trans, root);
3899
3900 ret = btrfs_delayed_update_inode(trans, root, inode);
3901 if (!ret)
3902 btrfs_set_inode_last_trans(trans, inode);
3903 return ret;
3904 }
3905
3906 return btrfs_update_inode_item(trans, root, inode);
3907 }
3908
3909 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3910 struct btrfs_root *root,
3911 struct inode *inode)
3912 {
3913 int ret;
3914
3915 ret = btrfs_update_inode(trans, root, inode);
3916 if (ret == -ENOSPC)
3917 return btrfs_update_inode_item(trans, root, inode);
3918 return ret;
3919 }
3920
3921 /*
3922 * unlink helper that gets used here in inode.c and in the tree logging
3923 * recovery code. It remove a link in a directory with a given name, and
3924 * also drops the back refs in the inode to the directory
3925 */
3926 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3927 struct btrfs_root *root,
3928 struct inode *dir, struct inode *inode,
3929 const char *name, int name_len)
3930 {
3931 struct btrfs_path *path;
3932 int ret = 0;
3933 struct extent_buffer *leaf;
3934 struct btrfs_dir_item *di;
3935 struct btrfs_key key;
3936 u64 index;
3937 u64 ino = btrfs_ino(inode);
3938 u64 dir_ino = btrfs_ino(dir);
3939
3940 path = btrfs_alloc_path();
3941 if (!path) {
3942 ret = -ENOMEM;
3943 goto out;
3944 }
3945
3946 path->leave_spinning = 1;
3947 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3948 name, name_len, -1);
3949 if (IS_ERR(di)) {
3950 ret = PTR_ERR(di);
3951 goto err;
3952 }
3953 if (!di) {
3954 ret = -ENOENT;
3955 goto err;
3956 }
3957 leaf = path->nodes[0];
3958 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3959 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3960 if (ret)
3961 goto err;
3962 btrfs_release_path(path);
3963
3964 /*
3965 * If we don't have dir index, we have to get it by looking up
3966 * the inode ref, since we get the inode ref, remove it directly,
3967 * it is unnecessary to do delayed deletion.
3968 *
3969 * But if we have dir index, needn't search inode ref to get it.
3970 * Since the inode ref is close to the inode item, it is better
3971 * that we delay to delete it, and just do this deletion when
3972 * we update the inode item.
3973 */
3974 if (BTRFS_I(inode)->dir_index) {
3975 ret = btrfs_delayed_delete_inode_ref(inode);
3976 if (!ret) {
3977 index = BTRFS_I(inode)->dir_index;
3978 goto skip_backref;
3979 }
3980 }
3981
3982 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3983 dir_ino, &index);
3984 if (ret) {
3985 btrfs_info(root->fs_info,
3986 "failed to delete reference to %.*s, inode %llu parent %llu",
3987 name_len, name, ino, dir_ino);
3988 btrfs_abort_transaction(trans, root, ret);
3989 goto err;
3990 }
3991 skip_backref:
3992 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3993 if (ret) {
3994 btrfs_abort_transaction(trans, root, ret);
3995 goto err;
3996 }
3997
3998 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3999 inode, dir_ino);
4000 if (ret != 0 && ret != -ENOENT) {
4001 btrfs_abort_transaction(trans, root, ret);
4002 goto err;
4003 }
4004
4005 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4006 dir, index);
4007 if (ret == -ENOENT)
4008 ret = 0;
4009 else if (ret)
4010 btrfs_abort_transaction(trans, root, ret);
4011 err:
4012 btrfs_free_path(path);
4013 if (ret)
4014 goto out;
4015
4016 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4017 inode_inc_iversion(inode);
4018 inode_inc_iversion(dir);
4019 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4020 ret = btrfs_update_inode(trans, root, dir);
4021 out:
4022 return ret;
4023 }
4024
4025 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4026 struct btrfs_root *root,
4027 struct inode *dir, struct inode *inode,
4028 const char *name, int name_len)
4029 {
4030 int ret;
4031 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4032 if (!ret) {
4033 drop_nlink(inode);
4034 ret = btrfs_update_inode(trans, root, inode);
4035 }
4036 return ret;
4037 }
4038
4039 /*
4040 * helper to start transaction for unlink and rmdir.
4041 *
4042 * unlink and rmdir are special in btrfs, they do not always free space, so
4043 * if we cannot make our reservations the normal way try and see if there is
4044 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4045 * allow the unlink to occur.
4046 */
4047 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4048 {
4049 struct btrfs_root *root = BTRFS_I(dir)->root;
4050
4051 /*
4052 * 1 for the possible orphan item
4053 * 1 for the dir item
4054 * 1 for the dir index
4055 * 1 for the inode ref
4056 * 1 for the inode
4057 */
4058 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4059 }
4060
4061 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4062 {
4063 struct btrfs_root *root = BTRFS_I(dir)->root;
4064 struct btrfs_trans_handle *trans;
4065 struct inode *inode = d_inode(dentry);
4066 int ret;
4067
4068 trans = __unlink_start_trans(dir);
4069 if (IS_ERR(trans))
4070 return PTR_ERR(trans);
4071
4072 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4073
4074 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4075 dentry->d_name.name, dentry->d_name.len);
4076 if (ret)
4077 goto out;
4078
4079 if (inode->i_nlink == 0) {
4080 ret = btrfs_orphan_add(trans, inode);
4081 if (ret)
4082 goto out;
4083 }
4084
4085 out:
4086 btrfs_end_transaction(trans, root);
4087 btrfs_btree_balance_dirty(root);
4088 return ret;
4089 }
4090
4091 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4092 struct btrfs_root *root,
4093 struct inode *dir, u64 objectid,
4094 const char *name, int name_len)
4095 {
4096 struct btrfs_path *path;
4097 struct extent_buffer *leaf;
4098 struct btrfs_dir_item *di;
4099 struct btrfs_key key;
4100 u64 index;
4101 int ret;
4102 u64 dir_ino = btrfs_ino(dir);
4103
4104 path = btrfs_alloc_path();
4105 if (!path)
4106 return -ENOMEM;
4107
4108 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4109 name, name_len, -1);
4110 if (IS_ERR_OR_NULL(di)) {
4111 if (!di)
4112 ret = -ENOENT;
4113 else
4114 ret = PTR_ERR(di);
4115 goto out;
4116 }
4117
4118 leaf = path->nodes[0];
4119 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4120 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4121 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4122 if (ret) {
4123 btrfs_abort_transaction(trans, root, ret);
4124 goto out;
4125 }
4126 btrfs_release_path(path);
4127
4128 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4129 objectid, root->root_key.objectid,
4130 dir_ino, &index, name, name_len);
4131 if (ret < 0) {
4132 if (ret != -ENOENT) {
4133 btrfs_abort_transaction(trans, root, ret);
4134 goto out;
4135 }
4136 di = btrfs_search_dir_index_item(root, path, dir_ino,
4137 name, name_len);
4138 if (IS_ERR_OR_NULL(di)) {
4139 if (!di)
4140 ret = -ENOENT;
4141 else
4142 ret = PTR_ERR(di);
4143 btrfs_abort_transaction(trans, root, ret);
4144 goto out;
4145 }
4146
4147 leaf = path->nodes[0];
4148 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4149 btrfs_release_path(path);
4150 index = key.offset;
4151 }
4152 btrfs_release_path(path);
4153
4154 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4155 if (ret) {
4156 btrfs_abort_transaction(trans, root, ret);
4157 goto out;
4158 }
4159
4160 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4161 inode_inc_iversion(dir);
4162 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4163 ret = btrfs_update_inode_fallback(trans, root, dir);
4164 if (ret)
4165 btrfs_abort_transaction(trans, root, ret);
4166 out:
4167 btrfs_free_path(path);
4168 return ret;
4169 }
4170
4171 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4172 {
4173 struct inode *inode = d_inode(dentry);
4174 int err = 0;
4175 struct btrfs_root *root = BTRFS_I(dir)->root;
4176 struct btrfs_trans_handle *trans;
4177
4178 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4179 return -ENOTEMPTY;
4180 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4181 return -EPERM;
4182
4183 trans = __unlink_start_trans(dir);
4184 if (IS_ERR(trans))
4185 return PTR_ERR(trans);
4186
4187 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4188 err = btrfs_unlink_subvol(trans, root, dir,
4189 BTRFS_I(inode)->location.objectid,
4190 dentry->d_name.name,
4191 dentry->d_name.len);
4192 goto out;
4193 }
4194
4195 err = btrfs_orphan_add(trans, inode);
4196 if (err)
4197 goto out;
4198
4199 /* now the directory is empty */
4200 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4201 dentry->d_name.name, dentry->d_name.len);
4202 if (!err)
4203 btrfs_i_size_write(inode, 0);
4204 out:
4205 btrfs_end_transaction(trans, root);
4206 btrfs_btree_balance_dirty(root);
4207
4208 return err;
4209 }
4210
4211 static int truncate_space_check(struct btrfs_trans_handle *trans,
4212 struct btrfs_root *root,
4213 u64 bytes_deleted)
4214 {
4215 int ret;
4216
4217 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4218 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4219 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4220 if (!ret)
4221 trans->bytes_reserved += bytes_deleted;
4222 return ret;
4223
4224 }
4225
4226 static int truncate_inline_extent(struct inode *inode,
4227 struct btrfs_path *path,
4228 struct btrfs_key *found_key,
4229 const u64 item_end,
4230 const u64 new_size)
4231 {
4232 struct extent_buffer *leaf = path->nodes[0];
4233 int slot = path->slots[0];
4234 struct btrfs_file_extent_item *fi;
4235 u32 size = (u32)(new_size - found_key->offset);
4236 struct btrfs_root *root = BTRFS_I(inode)->root;
4237
4238 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4239
4240 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4241 loff_t offset = new_size;
4242 loff_t page_end = ALIGN(offset, PAGE_CACHE_SIZE);
4243
4244 /*
4245 * Zero out the remaining of the last page of our inline extent,
4246 * instead of directly truncating our inline extent here - that
4247 * would be much more complex (decompressing all the data, then
4248 * compressing the truncated data, which might be bigger than
4249 * the size of the inline extent, resize the extent, etc).
4250 * We release the path because to get the page we might need to
4251 * read the extent item from disk (data not in the page cache).
4252 */
4253 btrfs_release_path(path);
4254 return btrfs_truncate_page(inode, offset, page_end - offset, 0);
4255 }
4256
4257 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4258 size = btrfs_file_extent_calc_inline_size(size);
4259 btrfs_truncate_item(root, path, size, 1);
4260
4261 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4262 inode_sub_bytes(inode, item_end + 1 - new_size);
4263
4264 return 0;
4265 }
4266
4267 /*
4268 * this can truncate away extent items, csum items and directory items.
4269 * It starts at a high offset and removes keys until it can't find
4270 * any higher than new_size
4271 *
4272 * csum items that cross the new i_size are truncated to the new size
4273 * as well.
4274 *
4275 * min_type is the minimum key type to truncate down to. If set to 0, this
4276 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4277 */
4278 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4279 struct btrfs_root *root,
4280 struct inode *inode,
4281 u64 new_size, u32 min_type)
4282 {
4283 struct btrfs_path *path;
4284 struct extent_buffer *leaf;
4285 struct btrfs_file_extent_item *fi;
4286 struct btrfs_key key;
4287 struct btrfs_key found_key;
4288 u64 extent_start = 0;
4289 u64 extent_num_bytes = 0;
4290 u64 extent_offset = 0;
4291 u64 item_end = 0;
4292 u64 last_size = new_size;
4293 u32 found_type = (u8)-1;
4294 int found_extent;
4295 int del_item;
4296 int pending_del_nr = 0;
4297 int pending_del_slot = 0;
4298 int extent_type = -1;
4299 int ret;
4300 int err = 0;
4301 u64 ino = btrfs_ino(inode);
4302 u64 bytes_deleted = 0;
4303 bool be_nice = 0;
4304 bool should_throttle = 0;
4305 bool should_end = 0;
4306
4307 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4308
4309 /*
4310 * for non-free space inodes and ref cows, we want to back off from
4311 * time to time
4312 */
4313 if (!btrfs_is_free_space_inode(inode) &&
4314 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4315 be_nice = 1;
4316
4317 path = btrfs_alloc_path();
4318 if (!path)
4319 return -ENOMEM;
4320 path->reada = -1;
4321
4322 /*
4323 * We want to drop from the next block forward in case this new size is
4324 * not block aligned since we will be keeping the last block of the
4325 * extent just the way it is.
4326 */
4327 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4328 root == root->fs_info->tree_root)
4329 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4330 root->sectorsize), (u64)-1, 0);
4331
4332 /*
4333 * This function is also used to drop the items in the log tree before
4334 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4335 * it is used to drop the loged items. So we shouldn't kill the delayed
4336 * items.
4337 */
4338 if (min_type == 0 && root == BTRFS_I(inode)->root)
4339 btrfs_kill_delayed_inode_items(inode);
4340
4341 key.objectid = ino;
4342 key.offset = (u64)-1;
4343 key.type = (u8)-1;
4344
4345 search_again:
4346 /*
4347 * with a 16K leaf size and 128MB extents, you can actually queue
4348 * up a huge file in a single leaf. Most of the time that
4349 * bytes_deleted is > 0, it will be huge by the time we get here
4350 */
4351 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4352 if (btrfs_should_end_transaction(trans, root)) {
4353 err = -EAGAIN;
4354 goto error;
4355 }
4356 }
4357
4358
4359 path->leave_spinning = 1;
4360 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4361 if (ret < 0) {
4362 err = ret;
4363 goto out;
4364 }
4365
4366 if (ret > 0) {
4367 /* there are no items in the tree for us to truncate, we're
4368 * done
4369 */
4370 if (path->slots[0] == 0)
4371 goto out;
4372 path->slots[0]--;
4373 }
4374
4375 while (1) {
4376 fi = NULL;
4377 leaf = path->nodes[0];
4378 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4379 found_type = found_key.type;
4380
4381 if (found_key.objectid != ino)
4382 break;
4383
4384 if (found_type < min_type)
4385 break;
4386
4387 item_end = found_key.offset;
4388 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4389 fi = btrfs_item_ptr(leaf, path->slots[0],
4390 struct btrfs_file_extent_item);
4391 extent_type = btrfs_file_extent_type(leaf, fi);
4392 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4393 item_end +=
4394 btrfs_file_extent_num_bytes(leaf, fi);
4395 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4396 item_end += btrfs_file_extent_inline_len(leaf,
4397 path->slots[0], fi);
4398 }
4399 item_end--;
4400 }
4401 if (found_type > min_type) {
4402 del_item = 1;
4403 } else {
4404 if (item_end < new_size)
4405 break;
4406 if (found_key.offset >= new_size)
4407 del_item = 1;
4408 else
4409 del_item = 0;
4410 }
4411 found_extent = 0;
4412 /* FIXME, shrink the extent if the ref count is only 1 */
4413 if (found_type != BTRFS_EXTENT_DATA_KEY)
4414 goto delete;
4415
4416 if (del_item)
4417 last_size = found_key.offset;
4418 else
4419 last_size = new_size;
4420
4421 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4422 u64 num_dec;
4423 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4424 if (!del_item) {
4425 u64 orig_num_bytes =
4426 btrfs_file_extent_num_bytes(leaf, fi);
4427 extent_num_bytes = ALIGN(new_size -
4428 found_key.offset,
4429 root->sectorsize);
4430 btrfs_set_file_extent_num_bytes(leaf, fi,
4431 extent_num_bytes);
4432 num_dec = (orig_num_bytes -
4433 extent_num_bytes);
4434 if (test_bit(BTRFS_ROOT_REF_COWS,
4435 &root->state) &&
4436 extent_start != 0)
4437 inode_sub_bytes(inode, num_dec);
4438 btrfs_mark_buffer_dirty(leaf);
4439 } else {
4440 extent_num_bytes =
4441 btrfs_file_extent_disk_num_bytes(leaf,
4442 fi);
4443 extent_offset = found_key.offset -
4444 btrfs_file_extent_offset(leaf, fi);
4445
4446 /* FIXME blocksize != 4096 */
4447 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4448 if (extent_start != 0) {
4449 found_extent = 1;
4450 if (test_bit(BTRFS_ROOT_REF_COWS,
4451 &root->state))
4452 inode_sub_bytes(inode, num_dec);
4453 }
4454 }
4455 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4456 /*
4457 * we can't truncate inline items that have had
4458 * special encodings
4459 */
4460 if (!del_item &&
4461 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4462 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4463
4464 /*
4465 * Need to release path in order to truncate a
4466 * compressed extent. So delete any accumulated
4467 * extent items so far.
4468 */
4469 if (btrfs_file_extent_compression(leaf, fi) !=
4470 BTRFS_COMPRESS_NONE && pending_del_nr) {
4471 err = btrfs_del_items(trans, root, path,
4472 pending_del_slot,
4473 pending_del_nr);
4474 if (err) {
4475 btrfs_abort_transaction(trans,
4476 root,
4477 err);
4478 goto error;
4479 }
4480 pending_del_nr = 0;
4481 }
4482
4483 err = truncate_inline_extent(inode, path,
4484 &found_key,
4485 item_end,
4486 new_size);
4487 if (err) {
4488 btrfs_abort_transaction(trans,
4489 root, err);
4490 goto error;
4491 }
4492 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4493 &root->state)) {
4494 inode_sub_bytes(inode, item_end + 1 - new_size);
4495 }
4496 }
4497 delete:
4498 if (del_item) {
4499 if (!pending_del_nr) {
4500 /* no pending yet, add ourselves */
4501 pending_del_slot = path->slots[0];
4502 pending_del_nr = 1;
4503 } else if (pending_del_nr &&
4504 path->slots[0] + 1 == pending_del_slot) {
4505 /* hop on the pending chunk */
4506 pending_del_nr++;
4507 pending_del_slot = path->slots[0];
4508 } else {
4509 BUG();
4510 }
4511 } else {
4512 break;
4513 }
4514 should_throttle = 0;
4515
4516 if (found_extent &&
4517 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4518 root == root->fs_info->tree_root)) {
4519 btrfs_set_path_blocking(path);
4520 bytes_deleted += extent_num_bytes;
4521 ret = btrfs_free_extent(trans, root, extent_start,
4522 extent_num_bytes, 0,
4523 btrfs_header_owner(leaf),
4524 ino, extent_offset);
4525 BUG_ON(ret);
4526 if (btrfs_should_throttle_delayed_refs(trans, root))
4527 btrfs_async_run_delayed_refs(root,
4528 trans->delayed_ref_updates * 2, 0);
4529 if (be_nice) {
4530 if (truncate_space_check(trans, root,
4531 extent_num_bytes)) {
4532 should_end = 1;
4533 }
4534 if (btrfs_should_throttle_delayed_refs(trans,
4535 root)) {
4536 should_throttle = 1;
4537 }
4538 }
4539 }
4540
4541 if (found_type == BTRFS_INODE_ITEM_KEY)
4542 break;
4543
4544 if (path->slots[0] == 0 ||
4545 path->slots[0] != pending_del_slot ||
4546 should_throttle || should_end) {
4547 if (pending_del_nr) {
4548 ret = btrfs_del_items(trans, root, path,
4549 pending_del_slot,
4550 pending_del_nr);
4551 if (ret) {
4552 btrfs_abort_transaction(trans,
4553 root, ret);
4554 goto error;
4555 }
4556 pending_del_nr = 0;
4557 }
4558 btrfs_release_path(path);
4559 if (should_throttle) {
4560 unsigned long updates = trans->delayed_ref_updates;
4561 if (updates) {
4562 trans->delayed_ref_updates = 0;
4563 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4564 if (ret && !err)
4565 err = ret;
4566 }
4567 }
4568 /*
4569 * if we failed to refill our space rsv, bail out
4570 * and let the transaction restart
4571 */
4572 if (should_end) {
4573 err = -EAGAIN;
4574 goto error;
4575 }
4576 goto search_again;
4577 } else {
4578 path->slots[0]--;
4579 }
4580 }
4581 out:
4582 if (pending_del_nr) {
4583 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4584 pending_del_nr);
4585 if (ret)
4586 btrfs_abort_transaction(trans, root, ret);
4587 }
4588 error:
4589 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4590 btrfs_ordered_update_i_size(inode, last_size, NULL);
4591
4592 btrfs_free_path(path);
4593
4594 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4595 unsigned long updates = trans->delayed_ref_updates;
4596 if (updates) {
4597 trans->delayed_ref_updates = 0;
4598 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4599 if (ret && !err)
4600 err = ret;
4601 }
4602 }
4603 return err;
4604 }
4605
4606 /*
4607 * btrfs_truncate_page - read, zero a chunk and write a page
4608 * @inode - inode that we're zeroing
4609 * @from - the offset to start zeroing
4610 * @len - the length to zero, 0 to zero the entire range respective to the
4611 * offset
4612 * @front - zero up to the offset instead of from the offset on
4613 *
4614 * This will find the page for the "from" offset and cow the page and zero the
4615 * part we want to zero. This is used with truncate and hole punching.
4616 */
4617 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4618 int front)
4619 {
4620 struct address_space *mapping = inode->i_mapping;
4621 struct btrfs_root *root = BTRFS_I(inode)->root;
4622 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4623 struct btrfs_ordered_extent *ordered;
4624 struct extent_state *cached_state = NULL;
4625 char *kaddr;
4626 u32 blocksize = root->sectorsize;
4627 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4628 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4629 struct page *page;
4630 gfp_t mask = btrfs_alloc_write_mask(mapping);
4631 int ret = 0;
4632 u64 page_start;
4633 u64 page_end;
4634
4635 if ((offset & (blocksize - 1)) == 0 &&
4636 (!len || ((len & (blocksize - 1)) == 0)))
4637 goto out;
4638 ret = btrfs_delalloc_reserve_space(inode,
4639 round_down(from, PAGE_CACHE_SIZE), PAGE_CACHE_SIZE);
4640 if (ret)
4641 goto out;
4642
4643 again:
4644 page = find_or_create_page(mapping, index, mask);
4645 if (!page) {
4646 btrfs_delalloc_release_space(inode,
4647 round_down(from, PAGE_CACHE_SIZE),
4648 PAGE_CACHE_SIZE);
4649 ret = -ENOMEM;
4650 goto out;
4651 }
4652
4653 page_start = page_offset(page);
4654 page_end = page_start + PAGE_CACHE_SIZE - 1;
4655
4656 if (!PageUptodate(page)) {
4657 ret = btrfs_readpage(NULL, page);
4658 lock_page(page);
4659 if (page->mapping != mapping) {
4660 unlock_page(page);
4661 page_cache_release(page);
4662 goto again;
4663 }
4664 if (!PageUptodate(page)) {
4665 ret = -EIO;
4666 goto out_unlock;
4667 }
4668 }
4669 wait_on_page_writeback(page);
4670
4671 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4672 set_page_extent_mapped(page);
4673
4674 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4675 if (ordered) {
4676 unlock_extent_cached(io_tree, page_start, page_end,
4677 &cached_state, GFP_NOFS);
4678 unlock_page(page);
4679 page_cache_release(page);
4680 btrfs_start_ordered_extent(inode, ordered, 1);
4681 btrfs_put_ordered_extent(ordered);
4682 goto again;
4683 }
4684
4685 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4686 EXTENT_DIRTY | EXTENT_DELALLOC |
4687 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4688 0, 0, &cached_state, GFP_NOFS);
4689
4690 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4691 &cached_state);
4692 if (ret) {
4693 unlock_extent_cached(io_tree, page_start, page_end,
4694 &cached_state, GFP_NOFS);
4695 goto out_unlock;
4696 }
4697
4698 if (offset != PAGE_CACHE_SIZE) {
4699 if (!len)
4700 len = PAGE_CACHE_SIZE - offset;
4701 kaddr = kmap(page);
4702 if (front)
4703 memset(kaddr, 0, offset);
4704 else
4705 memset(kaddr + offset, 0, len);
4706 flush_dcache_page(page);
4707 kunmap(page);
4708 }
4709 ClearPageChecked(page);
4710 set_page_dirty(page);
4711 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4712 GFP_NOFS);
4713
4714 out_unlock:
4715 if (ret)
4716 btrfs_delalloc_release_space(inode, page_start,
4717 PAGE_CACHE_SIZE);
4718 unlock_page(page);
4719 page_cache_release(page);
4720 out:
4721 return ret;
4722 }
4723
4724 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4725 u64 offset, u64 len)
4726 {
4727 struct btrfs_trans_handle *trans;
4728 int ret;
4729
4730 /*
4731 * Still need to make sure the inode looks like it's been updated so
4732 * that any holes get logged if we fsync.
4733 */
4734 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4735 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4736 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4737 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4738 return 0;
4739 }
4740
4741 /*
4742 * 1 - for the one we're dropping
4743 * 1 - for the one we're adding
4744 * 1 - for updating the inode.
4745 */
4746 trans = btrfs_start_transaction(root, 3);
4747 if (IS_ERR(trans))
4748 return PTR_ERR(trans);
4749
4750 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4751 if (ret) {
4752 btrfs_abort_transaction(trans, root, ret);
4753 btrfs_end_transaction(trans, root);
4754 return ret;
4755 }
4756
4757 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4758 0, 0, len, 0, len, 0, 0, 0);
4759 if (ret)
4760 btrfs_abort_transaction(trans, root, ret);
4761 else
4762 btrfs_update_inode(trans, root, inode);
4763 btrfs_end_transaction(trans, root);
4764 return ret;
4765 }
4766
4767 /*
4768 * This function puts in dummy file extents for the area we're creating a hole
4769 * for. So if we are truncating this file to a larger size we need to insert
4770 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4771 * the range between oldsize and size
4772 */
4773 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4774 {
4775 struct btrfs_root *root = BTRFS_I(inode)->root;
4776 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4777 struct extent_map *em = NULL;
4778 struct extent_state *cached_state = NULL;
4779 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4780 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4781 u64 block_end = ALIGN(size, root->sectorsize);
4782 u64 last_byte;
4783 u64 cur_offset;
4784 u64 hole_size;
4785 int err = 0;
4786
4787 /*
4788 * If our size started in the middle of a page we need to zero out the
4789 * rest of the page before we expand the i_size, otherwise we could
4790 * expose stale data.
4791 */
4792 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4793 if (err)
4794 return err;
4795
4796 if (size <= hole_start)
4797 return 0;
4798
4799 while (1) {
4800 struct btrfs_ordered_extent *ordered;
4801
4802 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4803 &cached_state);
4804 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4805 block_end - hole_start);
4806 if (!ordered)
4807 break;
4808 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4809 &cached_state, GFP_NOFS);
4810 btrfs_start_ordered_extent(inode, ordered, 1);
4811 btrfs_put_ordered_extent(ordered);
4812 }
4813
4814 cur_offset = hole_start;
4815 while (1) {
4816 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4817 block_end - cur_offset, 0);
4818 if (IS_ERR(em)) {
4819 err = PTR_ERR(em);
4820 em = NULL;
4821 break;
4822 }
4823 last_byte = min(extent_map_end(em), block_end);
4824 last_byte = ALIGN(last_byte , root->sectorsize);
4825 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4826 struct extent_map *hole_em;
4827 hole_size = last_byte - cur_offset;
4828
4829 err = maybe_insert_hole(root, inode, cur_offset,
4830 hole_size);
4831 if (err)
4832 break;
4833 btrfs_drop_extent_cache(inode, cur_offset,
4834 cur_offset + hole_size - 1, 0);
4835 hole_em = alloc_extent_map();
4836 if (!hole_em) {
4837 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4838 &BTRFS_I(inode)->runtime_flags);
4839 goto next;
4840 }
4841 hole_em->start = cur_offset;
4842 hole_em->len = hole_size;
4843 hole_em->orig_start = cur_offset;
4844
4845 hole_em->block_start = EXTENT_MAP_HOLE;
4846 hole_em->block_len = 0;
4847 hole_em->orig_block_len = 0;
4848 hole_em->ram_bytes = hole_size;
4849 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4850 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4851 hole_em->generation = root->fs_info->generation;
4852
4853 while (1) {
4854 write_lock(&em_tree->lock);
4855 err = add_extent_mapping(em_tree, hole_em, 1);
4856 write_unlock(&em_tree->lock);
4857 if (err != -EEXIST)
4858 break;
4859 btrfs_drop_extent_cache(inode, cur_offset,
4860 cur_offset +
4861 hole_size - 1, 0);
4862 }
4863 free_extent_map(hole_em);
4864 }
4865 next:
4866 free_extent_map(em);
4867 em = NULL;
4868 cur_offset = last_byte;
4869 if (cur_offset >= block_end)
4870 break;
4871 }
4872 free_extent_map(em);
4873 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4874 GFP_NOFS);
4875 return err;
4876 }
4877
4878 static int wait_snapshoting_atomic_t(atomic_t *a)
4879 {
4880 schedule();
4881 return 0;
4882 }
4883
4884 static void wait_for_snapshot_creation(struct btrfs_root *root)
4885 {
4886 while (true) {
4887 int ret;
4888
4889 ret = btrfs_start_write_no_snapshoting(root);
4890 if (ret)
4891 break;
4892 wait_on_atomic_t(&root->will_be_snapshoted,
4893 wait_snapshoting_atomic_t,
4894 TASK_UNINTERRUPTIBLE);
4895 }
4896 }
4897
4898 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4899 {
4900 struct btrfs_root *root = BTRFS_I(inode)->root;
4901 struct btrfs_trans_handle *trans;
4902 loff_t oldsize = i_size_read(inode);
4903 loff_t newsize = attr->ia_size;
4904 int mask = attr->ia_valid;
4905 int ret;
4906
4907 /*
4908 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4909 * special case where we need to update the times despite not having
4910 * these flags set. For all other operations the VFS set these flags
4911 * explicitly if it wants a timestamp update.
4912 */
4913 if (newsize != oldsize) {
4914 inode_inc_iversion(inode);
4915 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4916 inode->i_ctime = inode->i_mtime =
4917 current_fs_time(inode->i_sb);
4918 }
4919
4920 if (newsize > oldsize) {
4921 truncate_pagecache(inode, newsize);
4922 /*
4923 * Don't do an expanding truncate while snapshoting is ongoing.
4924 * This is to ensure the snapshot captures a fully consistent
4925 * state of this file - if the snapshot captures this expanding
4926 * truncation, it must capture all writes that happened before
4927 * this truncation.
4928 */
4929 wait_for_snapshot_creation(root);
4930 ret = btrfs_cont_expand(inode, oldsize, newsize);
4931 if (ret) {
4932 btrfs_end_write_no_snapshoting(root);
4933 return ret;
4934 }
4935
4936 trans = btrfs_start_transaction(root, 1);
4937 if (IS_ERR(trans)) {
4938 btrfs_end_write_no_snapshoting(root);
4939 return PTR_ERR(trans);
4940 }
4941
4942 i_size_write(inode, newsize);
4943 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4944 ret = btrfs_update_inode(trans, root, inode);
4945 btrfs_end_write_no_snapshoting(root);
4946 btrfs_end_transaction(trans, root);
4947 } else {
4948
4949 /*
4950 * We're truncating a file that used to have good data down to
4951 * zero. Make sure it gets into the ordered flush list so that
4952 * any new writes get down to disk quickly.
4953 */
4954 if (newsize == 0)
4955 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4956 &BTRFS_I(inode)->runtime_flags);
4957
4958 /*
4959 * 1 for the orphan item we're going to add
4960 * 1 for the orphan item deletion.
4961 */
4962 trans = btrfs_start_transaction(root, 2);
4963 if (IS_ERR(trans))
4964 return PTR_ERR(trans);
4965
4966 /*
4967 * We need to do this in case we fail at _any_ point during the
4968 * actual truncate. Once we do the truncate_setsize we could
4969 * invalidate pages which forces any outstanding ordered io to
4970 * be instantly completed which will give us extents that need
4971 * to be truncated. If we fail to get an orphan inode down we
4972 * could have left over extents that were never meant to live,
4973 * so we need to garuntee from this point on that everything
4974 * will be consistent.
4975 */
4976 ret = btrfs_orphan_add(trans, inode);
4977 btrfs_end_transaction(trans, root);
4978 if (ret)
4979 return ret;
4980
4981 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4982 truncate_setsize(inode, newsize);
4983
4984 /* Disable nonlocked read DIO to avoid the end less truncate */
4985 btrfs_inode_block_unlocked_dio(inode);
4986 inode_dio_wait(inode);
4987 btrfs_inode_resume_unlocked_dio(inode);
4988
4989 ret = btrfs_truncate(inode);
4990 if (ret && inode->i_nlink) {
4991 int err;
4992
4993 /*
4994 * failed to truncate, disk_i_size is only adjusted down
4995 * as we remove extents, so it should represent the true
4996 * size of the inode, so reset the in memory size and
4997 * delete our orphan entry.
4998 */
4999 trans = btrfs_join_transaction(root);
5000 if (IS_ERR(trans)) {
5001 btrfs_orphan_del(NULL, inode);
5002 return ret;
5003 }
5004 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5005 err = btrfs_orphan_del(trans, inode);
5006 if (err)
5007 btrfs_abort_transaction(trans, root, err);
5008 btrfs_end_transaction(trans, root);
5009 }
5010 }
5011
5012 return ret;
5013 }
5014
5015 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5016 {
5017 struct inode *inode = d_inode(dentry);
5018 struct btrfs_root *root = BTRFS_I(inode)->root;
5019 int err;
5020
5021 if (btrfs_root_readonly(root))
5022 return -EROFS;
5023
5024 err = inode_change_ok(inode, attr);
5025 if (err)
5026 return err;
5027
5028 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5029 err = btrfs_setsize(inode, attr);
5030 if (err)
5031 return err;
5032 }
5033
5034 if (attr->ia_valid) {
5035 setattr_copy(inode, attr);
5036 inode_inc_iversion(inode);
5037 err = btrfs_dirty_inode(inode);
5038
5039 if (!err && attr->ia_valid & ATTR_MODE)
5040 err = posix_acl_chmod(inode, inode->i_mode);
5041 }
5042
5043 return err;
5044 }
5045
5046 /*
5047 * While truncating the inode pages during eviction, we get the VFS calling
5048 * btrfs_invalidatepage() against each page of the inode. This is slow because
5049 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5050 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5051 * extent_state structures over and over, wasting lots of time.
5052 *
5053 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5054 * those expensive operations on a per page basis and do only the ordered io
5055 * finishing, while we release here the extent_map and extent_state structures,
5056 * without the excessive merging and splitting.
5057 */
5058 static void evict_inode_truncate_pages(struct inode *inode)
5059 {
5060 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5061 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5062 struct rb_node *node;
5063
5064 ASSERT(inode->i_state & I_FREEING);
5065 truncate_inode_pages_final(&inode->i_data);
5066
5067 write_lock(&map_tree->lock);
5068 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5069 struct extent_map *em;
5070
5071 node = rb_first(&map_tree->map);
5072 em = rb_entry(node, struct extent_map, rb_node);
5073 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5074 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5075 remove_extent_mapping(map_tree, em);
5076 free_extent_map(em);
5077 if (need_resched()) {
5078 write_unlock(&map_tree->lock);
5079 cond_resched();
5080 write_lock(&map_tree->lock);
5081 }
5082 }
5083 write_unlock(&map_tree->lock);
5084
5085 /*
5086 * Keep looping until we have no more ranges in the io tree.
5087 * We can have ongoing bios started by readpages (called from readahead)
5088 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5089 * still in progress (unlocked the pages in the bio but did not yet
5090 * unlocked the ranges in the io tree). Therefore this means some
5091 * ranges can still be locked and eviction started because before
5092 * submitting those bios, which are executed by a separate task (work
5093 * queue kthread), inode references (inode->i_count) were not taken
5094 * (which would be dropped in the end io callback of each bio).
5095 * Therefore here we effectively end up waiting for those bios and
5096 * anyone else holding locked ranges without having bumped the inode's
5097 * reference count - if we don't do it, when they access the inode's
5098 * io_tree to unlock a range it may be too late, leading to an
5099 * use-after-free issue.
5100 */
5101 spin_lock(&io_tree->lock);
5102 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5103 struct extent_state *state;
5104 struct extent_state *cached_state = NULL;
5105 u64 start;
5106 u64 end;
5107
5108 node = rb_first(&io_tree->state);
5109 state = rb_entry(node, struct extent_state, rb_node);
5110 start = state->start;
5111 end = state->end;
5112 spin_unlock(&io_tree->lock);
5113
5114 lock_extent_bits(io_tree, start, end, 0, &cached_state);
5115
5116 /*
5117 * If still has DELALLOC flag, the extent didn't reach disk,
5118 * and its reserved space won't be freed by delayed_ref.
5119 * So we need to free its reserved space here.
5120 * (Refer to comment in btrfs_invalidatepage, case 2)
5121 *
5122 * Note, end is the bytenr of last byte, so we need + 1 here.
5123 */
5124 if (state->state & EXTENT_DELALLOC)
5125 btrfs_qgroup_free_data(inode, start, end - start + 1);
5126
5127 clear_extent_bit(io_tree, start, end,
5128 EXTENT_LOCKED | EXTENT_DIRTY |
5129 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5130 EXTENT_DEFRAG, 1, 1,
5131 &cached_state, GFP_NOFS);
5132
5133 cond_resched();
5134 spin_lock(&io_tree->lock);
5135 }
5136 spin_unlock(&io_tree->lock);
5137 }
5138
5139 void btrfs_evict_inode(struct inode *inode)
5140 {
5141 struct btrfs_trans_handle *trans;
5142 struct btrfs_root *root = BTRFS_I(inode)->root;
5143 struct btrfs_block_rsv *rsv, *global_rsv;
5144 int steal_from_global = 0;
5145 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5146 int ret;
5147
5148 trace_btrfs_inode_evict(inode);
5149
5150 evict_inode_truncate_pages(inode);
5151
5152 if (inode->i_nlink &&
5153 ((btrfs_root_refs(&root->root_item) != 0 &&
5154 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5155 btrfs_is_free_space_inode(inode)))
5156 goto no_delete;
5157
5158 if (is_bad_inode(inode)) {
5159 btrfs_orphan_del(NULL, inode);
5160 goto no_delete;
5161 }
5162 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5163 if (!special_file(inode->i_mode))
5164 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5165
5166 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5167
5168 if (root->fs_info->log_root_recovering) {
5169 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5170 &BTRFS_I(inode)->runtime_flags));
5171 goto no_delete;
5172 }
5173
5174 if (inode->i_nlink > 0) {
5175 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5176 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5177 goto no_delete;
5178 }
5179
5180 ret = btrfs_commit_inode_delayed_inode(inode);
5181 if (ret) {
5182 btrfs_orphan_del(NULL, inode);
5183 goto no_delete;
5184 }
5185
5186 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5187 if (!rsv) {
5188 btrfs_orphan_del(NULL, inode);
5189 goto no_delete;
5190 }
5191 rsv->size = min_size;
5192 rsv->failfast = 1;
5193 global_rsv = &root->fs_info->global_block_rsv;
5194
5195 btrfs_i_size_write(inode, 0);
5196
5197 /*
5198 * This is a bit simpler than btrfs_truncate since we've already
5199 * reserved our space for our orphan item in the unlink, so we just
5200 * need to reserve some slack space in case we add bytes and update
5201 * inode item when doing the truncate.
5202 */
5203 while (1) {
5204 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5205 BTRFS_RESERVE_FLUSH_LIMIT);
5206
5207 /*
5208 * Try and steal from the global reserve since we will
5209 * likely not use this space anyway, we want to try as
5210 * hard as possible to get this to work.
5211 */
5212 if (ret)
5213 steal_from_global++;
5214 else
5215 steal_from_global = 0;
5216 ret = 0;
5217
5218 /*
5219 * steal_from_global == 0: we reserved stuff, hooray!
5220 * steal_from_global == 1: we didn't reserve stuff, boo!
5221 * steal_from_global == 2: we've committed, still not a lot of
5222 * room but maybe we'll have room in the global reserve this
5223 * time.
5224 * steal_from_global == 3: abandon all hope!
5225 */
5226 if (steal_from_global > 2) {
5227 btrfs_warn(root->fs_info,
5228 "Could not get space for a delete, will truncate on mount %d",
5229 ret);
5230 btrfs_orphan_del(NULL, inode);
5231 btrfs_free_block_rsv(root, rsv);
5232 goto no_delete;
5233 }
5234
5235 trans = btrfs_join_transaction(root);
5236 if (IS_ERR(trans)) {
5237 btrfs_orphan_del(NULL, inode);
5238 btrfs_free_block_rsv(root, rsv);
5239 goto no_delete;
5240 }
5241
5242 /*
5243 * We can't just steal from the global reserve, we need tomake
5244 * sure there is room to do it, if not we need to commit and try
5245 * again.
5246 */
5247 if (steal_from_global) {
5248 if (!btrfs_check_space_for_delayed_refs(trans, root))
5249 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5250 min_size);
5251 else
5252 ret = -ENOSPC;
5253 }
5254
5255 /*
5256 * Couldn't steal from the global reserve, we have too much
5257 * pending stuff built up, commit the transaction and try it
5258 * again.
5259 */
5260 if (ret) {
5261 ret = btrfs_commit_transaction(trans, root);
5262 if (ret) {
5263 btrfs_orphan_del(NULL, inode);
5264 btrfs_free_block_rsv(root, rsv);
5265 goto no_delete;
5266 }
5267 continue;
5268 } else {
5269 steal_from_global = 0;
5270 }
5271
5272 trans->block_rsv = rsv;
5273
5274 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5275 if (ret != -ENOSPC && ret != -EAGAIN)
5276 break;
5277
5278 trans->block_rsv = &root->fs_info->trans_block_rsv;
5279 btrfs_end_transaction(trans, root);
5280 trans = NULL;
5281 btrfs_btree_balance_dirty(root);
5282 }
5283
5284 btrfs_free_block_rsv(root, rsv);
5285
5286 /*
5287 * Errors here aren't a big deal, it just means we leave orphan items
5288 * in the tree. They will be cleaned up on the next mount.
5289 */
5290 if (ret == 0) {
5291 trans->block_rsv = root->orphan_block_rsv;
5292 btrfs_orphan_del(trans, inode);
5293 } else {
5294 btrfs_orphan_del(NULL, inode);
5295 }
5296
5297 trans->block_rsv = &root->fs_info->trans_block_rsv;
5298 if (!(root == root->fs_info->tree_root ||
5299 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5300 btrfs_return_ino(root, btrfs_ino(inode));
5301
5302 btrfs_end_transaction(trans, root);
5303 btrfs_btree_balance_dirty(root);
5304 no_delete:
5305 btrfs_remove_delayed_node(inode);
5306 clear_inode(inode);
5307 }
5308
5309 /*
5310 * this returns the key found in the dir entry in the location pointer.
5311 * If no dir entries were found, location->objectid is 0.
5312 */
5313 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5314 struct btrfs_key *location)
5315 {
5316 const char *name = dentry->d_name.name;
5317 int namelen = dentry->d_name.len;
5318 struct btrfs_dir_item *di;
5319 struct btrfs_path *path;
5320 struct btrfs_root *root = BTRFS_I(dir)->root;
5321 int ret = 0;
5322
5323 path = btrfs_alloc_path();
5324 if (!path)
5325 return -ENOMEM;
5326
5327 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5328 namelen, 0);
5329 if (IS_ERR(di))
5330 ret = PTR_ERR(di);
5331
5332 if (IS_ERR_OR_NULL(di))
5333 goto out_err;
5334
5335 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5336 out:
5337 btrfs_free_path(path);
5338 return ret;
5339 out_err:
5340 location->objectid = 0;
5341 goto out;
5342 }
5343
5344 /*
5345 * when we hit a tree root in a directory, the btrfs part of the inode
5346 * needs to be changed to reflect the root directory of the tree root. This
5347 * is kind of like crossing a mount point.
5348 */
5349 static int fixup_tree_root_location(struct btrfs_root *root,
5350 struct inode *dir,
5351 struct dentry *dentry,
5352 struct btrfs_key *location,
5353 struct btrfs_root **sub_root)
5354 {
5355 struct btrfs_path *path;
5356 struct btrfs_root *new_root;
5357 struct btrfs_root_ref *ref;
5358 struct extent_buffer *leaf;
5359 struct btrfs_key key;
5360 int ret;
5361 int err = 0;
5362
5363 path = btrfs_alloc_path();
5364 if (!path) {
5365 err = -ENOMEM;
5366 goto out;
5367 }
5368
5369 err = -ENOENT;
5370 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5371 key.type = BTRFS_ROOT_REF_KEY;
5372 key.offset = location->objectid;
5373
5374 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5375 0, 0);
5376 if (ret) {
5377 if (ret < 0)
5378 err = ret;
5379 goto out;
5380 }
5381
5382 leaf = path->nodes[0];
5383 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5384 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5385 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5386 goto out;
5387
5388 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5389 (unsigned long)(ref + 1),
5390 dentry->d_name.len);
5391 if (ret)
5392 goto out;
5393
5394 btrfs_release_path(path);
5395
5396 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5397 if (IS_ERR(new_root)) {
5398 err = PTR_ERR(new_root);
5399 goto out;
5400 }
5401
5402 *sub_root = new_root;
5403 location->objectid = btrfs_root_dirid(&new_root->root_item);
5404 location->type = BTRFS_INODE_ITEM_KEY;
5405 location->offset = 0;
5406 err = 0;
5407 out:
5408 btrfs_free_path(path);
5409 return err;
5410 }
5411
5412 static void inode_tree_add(struct inode *inode)
5413 {
5414 struct btrfs_root *root = BTRFS_I(inode)->root;
5415 struct btrfs_inode *entry;
5416 struct rb_node **p;
5417 struct rb_node *parent;
5418 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5419 u64 ino = btrfs_ino(inode);
5420
5421 if (inode_unhashed(inode))
5422 return;
5423 parent = NULL;
5424 spin_lock(&root->inode_lock);
5425 p = &root->inode_tree.rb_node;
5426 while (*p) {
5427 parent = *p;
5428 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5429
5430 if (ino < btrfs_ino(&entry->vfs_inode))
5431 p = &parent->rb_left;
5432 else if (ino > btrfs_ino(&entry->vfs_inode))
5433 p = &parent->rb_right;
5434 else {
5435 WARN_ON(!(entry->vfs_inode.i_state &
5436 (I_WILL_FREE | I_FREEING)));
5437 rb_replace_node(parent, new, &root->inode_tree);
5438 RB_CLEAR_NODE(parent);
5439 spin_unlock(&root->inode_lock);
5440 return;
5441 }
5442 }
5443 rb_link_node(new, parent, p);
5444 rb_insert_color(new, &root->inode_tree);
5445 spin_unlock(&root->inode_lock);
5446 }
5447
5448 static void inode_tree_del(struct inode *inode)
5449 {
5450 struct btrfs_root *root = BTRFS_I(inode)->root;
5451 int empty = 0;
5452
5453 spin_lock(&root->inode_lock);
5454 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5455 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5456 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5457 empty = RB_EMPTY_ROOT(&root->inode_tree);
5458 }
5459 spin_unlock(&root->inode_lock);
5460
5461 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5462 synchronize_srcu(&root->fs_info->subvol_srcu);
5463 spin_lock(&root->inode_lock);
5464 empty = RB_EMPTY_ROOT(&root->inode_tree);
5465 spin_unlock(&root->inode_lock);
5466 if (empty)
5467 btrfs_add_dead_root(root);
5468 }
5469 }
5470
5471 void btrfs_invalidate_inodes(struct btrfs_root *root)
5472 {
5473 struct rb_node *node;
5474 struct rb_node *prev;
5475 struct btrfs_inode *entry;
5476 struct inode *inode;
5477 u64 objectid = 0;
5478
5479 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5480 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5481
5482 spin_lock(&root->inode_lock);
5483 again:
5484 node = root->inode_tree.rb_node;
5485 prev = NULL;
5486 while (node) {
5487 prev = node;
5488 entry = rb_entry(node, struct btrfs_inode, rb_node);
5489
5490 if (objectid < btrfs_ino(&entry->vfs_inode))
5491 node = node->rb_left;
5492 else if (objectid > btrfs_ino(&entry->vfs_inode))
5493 node = node->rb_right;
5494 else
5495 break;
5496 }
5497 if (!node) {
5498 while (prev) {
5499 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5500 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5501 node = prev;
5502 break;
5503 }
5504 prev = rb_next(prev);
5505 }
5506 }
5507 while (node) {
5508 entry = rb_entry(node, struct btrfs_inode, rb_node);
5509 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5510 inode = igrab(&entry->vfs_inode);
5511 if (inode) {
5512 spin_unlock(&root->inode_lock);
5513 if (atomic_read(&inode->i_count) > 1)
5514 d_prune_aliases(inode);
5515 /*
5516 * btrfs_drop_inode will have it removed from
5517 * the inode cache when its usage count
5518 * hits zero.
5519 */
5520 iput(inode);
5521 cond_resched();
5522 spin_lock(&root->inode_lock);
5523 goto again;
5524 }
5525
5526 if (cond_resched_lock(&root->inode_lock))
5527 goto again;
5528
5529 node = rb_next(node);
5530 }
5531 spin_unlock(&root->inode_lock);
5532 }
5533
5534 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5535 {
5536 struct btrfs_iget_args *args = p;
5537 inode->i_ino = args->location->objectid;
5538 memcpy(&BTRFS_I(inode)->location, args->location,
5539 sizeof(*args->location));
5540 BTRFS_I(inode)->root = args->root;
5541 return 0;
5542 }
5543
5544 static int btrfs_find_actor(struct inode *inode, void *opaque)
5545 {
5546 struct btrfs_iget_args *args = opaque;
5547 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5548 args->root == BTRFS_I(inode)->root;
5549 }
5550
5551 static struct inode *btrfs_iget_locked(struct super_block *s,
5552 struct btrfs_key *location,
5553 struct btrfs_root *root)
5554 {
5555 struct inode *inode;
5556 struct btrfs_iget_args args;
5557 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5558
5559 args.location = location;
5560 args.root = root;
5561
5562 inode = iget5_locked(s, hashval, btrfs_find_actor,
5563 btrfs_init_locked_inode,
5564 (void *)&args);
5565 return inode;
5566 }
5567
5568 /* Get an inode object given its location and corresponding root.
5569 * Returns in *is_new if the inode was read from disk
5570 */
5571 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5572 struct btrfs_root *root, int *new)
5573 {
5574 struct inode *inode;
5575
5576 inode = btrfs_iget_locked(s, location, root);
5577 if (!inode)
5578 return ERR_PTR(-ENOMEM);
5579
5580 if (inode->i_state & I_NEW) {
5581 btrfs_read_locked_inode(inode);
5582 if (!is_bad_inode(inode)) {
5583 inode_tree_add(inode);
5584 unlock_new_inode(inode);
5585 if (new)
5586 *new = 1;
5587 } else {
5588 unlock_new_inode(inode);
5589 iput(inode);
5590 inode = ERR_PTR(-ESTALE);
5591 }
5592 }
5593
5594 return inode;
5595 }
5596
5597 static struct inode *new_simple_dir(struct super_block *s,
5598 struct btrfs_key *key,
5599 struct btrfs_root *root)
5600 {
5601 struct inode *inode = new_inode(s);
5602
5603 if (!inode)
5604 return ERR_PTR(-ENOMEM);
5605
5606 BTRFS_I(inode)->root = root;
5607 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5608 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5609
5610 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5611 inode->i_op = &btrfs_dir_ro_inode_operations;
5612 inode->i_fop = &simple_dir_operations;
5613 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5614 inode->i_mtime = CURRENT_TIME;
5615 inode->i_atime = inode->i_mtime;
5616 inode->i_ctime = inode->i_mtime;
5617 BTRFS_I(inode)->i_otime = inode->i_mtime;
5618
5619 return inode;
5620 }
5621
5622 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5623 {
5624 struct inode *inode;
5625 struct btrfs_root *root = BTRFS_I(dir)->root;
5626 struct btrfs_root *sub_root = root;
5627 struct btrfs_key location;
5628 int index;
5629 int ret = 0;
5630
5631 if (dentry->d_name.len > BTRFS_NAME_LEN)
5632 return ERR_PTR(-ENAMETOOLONG);
5633
5634 ret = btrfs_inode_by_name(dir, dentry, &location);
5635 if (ret < 0)
5636 return ERR_PTR(ret);
5637
5638 if (location.objectid == 0)
5639 return ERR_PTR(-ENOENT);
5640
5641 if (location.type == BTRFS_INODE_ITEM_KEY) {
5642 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5643 return inode;
5644 }
5645
5646 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5647
5648 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5649 ret = fixup_tree_root_location(root, dir, dentry,
5650 &location, &sub_root);
5651 if (ret < 0) {
5652 if (ret != -ENOENT)
5653 inode = ERR_PTR(ret);
5654 else
5655 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5656 } else {
5657 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5658 }
5659 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5660
5661 if (!IS_ERR(inode) && root != sub_root) {
5662 down_read(&root->fs_info->cleanup_work_sem);
5663 if (!(inode->i_sb->s_flags & MS_RDONLY))
5664 ret = btrfs_orphan_cleanup(sub_root);
5665 up_read(&root->fs_info->cleanup_work_sem);
5666 if (ret) {
5667 iput(inode);
5668 inode = ERR_PTR(ret);
5669 }
5670 }
5671
5672 return inode;
5673 }
5674
5675 static int btrfs_dentry_delete(const struct dentry *dentry)
5676 {
5677 struct btrfs_root *root;
5678 struct inode *inode = d_inode(dentry);
5679
5680 if (!inode && !IS_ROOT(dentry))
5681 inode = d_inode(dentry->d_parent);
5682
5683 if (inode) {
5684 root = BTRFS_I(inode)->root;
5685 if (btrfs_root_refs(&root->root_item) == 0)
5686 return 1;
5687
5688 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5689 return 1;
5690 }
5691 return 0;
5692 }
5693
5694 static void btrfs_dentry_release(struct dentry *dentry)
5695 {
5696 kfree(dentry->d_fsdata);
5697 }
5698
5699 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5700 unsigned int flags)
5701 {
5702 struct inode *inode;
5703
5704 inode = btrfs_lookup_dentry(dir, dentry);
5705 if (IS_ERR(inode)) {
5706 if (PTR_ERR(inode) == -ENOENT)
5707 inode = NULL;
5708 else
5709 return ERR_CAST(inode);
5710 }
5711
5712 return d_splice_alias(inode, dentry);
5713 }
5714
5715 unsigned char btrfs_filetype_table[] = {
5716 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5717 };
5718
5719 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5720 {
5721 struct inode *inode = file_inode(file);
5722 struct btrfs_root *root = BTRFS_I(inode)->root;
5723 struct btrfs_item *item;
5724 struct btrfs_dir_item *di;
5725 struct btrfs_key key;
5726 struct btrfs_key found_key;
5727 struct btrfs_path *path;
5728 struct list_head ins_list;
5729 struct list_head del_list;
5730 int ret;
5731 struct extent_buffer *leaf;
5732 int slot;
5733 unsigned char d_type;
5734 int over = 0;
5735 u32 di_cur;
5736 u32 di_total;
5737 u32 di_len;
5738 int key_type = BTRFS_DIR_INDEX_KEY;
5739 char tmp_name[32];
5740 char *name_ptr;
5741 int name_len;
5742 int is_curr = 0; /* ctx->pos points to the current index? */
5743
5744 /* FIXME, use a real flag for deciding about the key type */
5745 if (root->fs_info->tree_root == root)
5746 key_type = BTRFS_DIR_ITEM_KEY;
5747
5748 if (!dir_emit_dots(file, ctx))
5749 return 0;
5750
5751 path = btrfs_alloc_path();
5752 if (!path)
5753 return -ENOMEM;
5754
5755 path->reada = 1;
5756
5757 if (key_type == BTRFS_DIR_INDEX_KEY) {
5758 INIT_LIST_HEAD(&ins_list);
5759 INIT_LIST_HEAD(&del_list);
5760 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5761 }
5762
5763 key.type = key_type;
5764 key.offset = ctx->pos;
5765 key.objectid = btrfs_ino(inode);
5766
5767 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5768 if (ret < 0)
5769 goto err;
5770
5771 while (1) {
5772 leaf = path->nodes[0];
5773 slot = path->slots[0];
5774 if (slot >= btrfs_header_nritems(leaf)) {
5775 ret = btrfs_next_leaf(root, path);
5776 if (ret < 0)
5777 goto err;
5778 else if (ret > 0)
5779 break;
5780 continue;
5781 }
5782
5783 item = btrfs_item_nr(slot);
5784 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5785
5786 if (found_key.objectid != key.objectid)
5787 break;
5788 if (found_key.type != key_type)
5789 break;
5790 if (found_key.offset < ctx->pos)
5791 goto next;
5792 if (key_type == BTRFS_DIR_INDEX_KEY &&
5793 btrfs_should_delete_dir_index(&del_list,
5794 found_key.offset))
5795 goto next;
5796
5797 ctx->pos = found_key.offset;
5798 is_curr = 1;
5799
5800 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5801 di_cur = 0;
5802 di_total = btrfs_item_size(leaf, item);
5803
5804 while (di_cur < di_total) {
5805 struct btrfs_key location;
5806
5807 if (verify_dir_item(root, leaf, di))
5808 break;
5809
5810 name_len = btrfs_dir_name_len(leaf, di);
5811 if (name_len <= sizeof(tmp_name)) {
5812 name_ptr = tmp_name;
5813 } else {
5814 name_ptr = kmalloc(name_len, GFP_NOFS);
5815 if (!name_ptr) {
5816 ret = -ENOMEM;
5817 goto err;
5818 }
5819 }
5820 read_extent_buffer(leaf, name_ptr,
5821 (unsigned long)(di + 1), name_len);
5822
5823 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5824 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5825
5826
5827 /* is this a reference to our own snapshot? If so
5828 * skip it.
5829 *
5830 * In contrast to old kernels, we insert the snapshot's
5831 * dir item and dir index after it has been created, so
5832 * we won't find a reference to our own snapshot. We
5833 * still keep the following code for backward
5834 * compatibility.
5835 */
5836 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5837 location.objectid == root->root_key.objectid) {
5838 over = 0;
5839 goto skip;
5840 }
5841 over = !dir_emit(ctx, name_ptr, name_len,
5842 location.objectid, d_type);
5843
5844 skip:
5845 if (name_ptr != tmp_name)
5846 kfree(name_ptr);
5847
5848 if (over)
5849 goto nopos;
5850 di_len = btrfs_dir_name_len(leaf, di) +
5851 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5852 di_cur += di_len;
5853 di = (struct btrfs_dir_item *)((char *)di + di_len);
5854 }
5855 next:
5856 path->slots[0]++;
5857 }
5858
5859 if (key_type == BTRFS_DIR_INDEX_KEY) {
5860 if (is_curr)
5861 ctx->pos++;
5862 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5863 if (ret)
5864 goto nopos;
5865 }
5866
5867 /* Reached end of directory/root. Bump pos past the last item. */
5868 ctx->pos++;
5869
5870 /*
5871 * Stop new entries from being returned after we return the last
5872 * entry.
5873 *
5874 * New directory entries are assigned a strictly increasing
5875 * offset. This means that new entries created during readdir
5876 * are *guaranteed* to be seen in the future by that readdir.
5877 * This has broken buggy programs which operate on names as
5878 * they're returned by readdir. Until we re-use freed offsets
5879 * we have this hack to stop new entries from being returned
5880 * under the assumption that they'll never reach this huge
5881 * offset.
5882 *
5883 * This is being careful not to overflow 32bit loff_t unless the
5884 * last entry requires it because doing so has broken 32bit apps
5885 * in the past.
5886 */
5887 if (key_type == BTRFS_DIR_INDEX_KEY) {
5888 if (ctx->pos >= INT_MAX)
5889 ctx->pos = LLONG_MAX;
5890 else
5891 ctx->pos = INT_MAX;
5892 }
5893 nopos:
5894 ret = 0;
5895 err:
5896 if (key_type == BTRFS_DIR_INDEX_KEY)
5897 btrfs_put_delayed_items(&ins_list, &del_list);
5898 btrfs_free_path(path);
5899 return ret;
5900 }
5901
5902 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5903 {
5904 struct btrfs_root *root = BTRFS_I(inode)->root;
5905 struct btrfs_trans_handle *trans;
5906 int ret = 0;
5907 bool nolock = false;
5908
5909 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5910 return 0;
5911
5912 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5913 nolock = true;
5914
5915 if (wbc->sync_mode == WB_SYNC_ALL) {
5916 if (nolock)
5917 trans = btrfs_join_transaction_nolock(root);
5918 else
5919 trans = btrfs_join_transaction(root);
5920 if (IS_ERR(trans))
5921 return PTR_ERR(trans);
5922 ret = btrfs_commit_transaction(trans, root);
5923 }
5924 return ret;
5925 }
5926
5927 /*
5928 * This is somewhat expensive, updating the tree every time the
5929 * inode changes. But, it is most likely to find the inode in cache.
5930 * FIXME, needs more benchmarking...there are no reasons other than performance
5931 * to keep or drop this code.
5932 */
5933 static int btrfs_dirty_inode(struct inode *inode)
5934 {
5935 struct btrfs_root *root = BTRFS_I(inode)->root;
5936 struct btrfs_trans_handle *trans;
5937 int ret;
5938
5939 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5940 return 0;
5941
5942 trans = btrfs_join_transaction(root);
5943 if (IS_ERR(trans))
5944 return PTR_ERR(trans);
5945
5946 ret = btrfs_update_inode(trans, root, inode);
5947 if (ret && ret == -ENOSPC) {
5948 /* whoops, lets try again with the full transaction */
5949 btrfs_end_transaction(trans, root);
5950 trans = btrfs_start_transaction(root, 1);
5951 if (IS_ERR(trans))
5952 return PTR_ERR(trans);
5953
5954 ret = btrfs_update_inode(trans, root, inode);
5955 }
5956 btrfs_end_transaction(trans, root);
5957 if (BTRFS_I(inode)->delayed_node)
5958 btrfs_balance_delayed_items(root);
5959
5960 return ret;
5961 }
5962
5963 /*
5964 * This is a copy of file_update_time. We need this so we can return error on
5965 * ENOSPC for updating the inode in the case of file write and mmap writes.
5966 */
5967 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5968 int flags)
5969 {
5970 struct btrfs_root *root = BTRFS_I(inode)->root;
5971
5972 if (btrfs_root_readonly(root))
5973 return -EROFS;
5974
5975 if (flags & S_VERSION)
5976 inode_inc_iversion(inode);
5977 if (flags & S_CTIME)
5978 inode->i_ctime = *now;
5979 if (flags & S_MTIME)
5980 inode->i_mtime = *now;
5981 if (flags & S_ATIME)
5982 inode->i_atime = *now;
5983 return btrfs_dirty_inode(inode);
5984 }
5985
5986 /*
5987 * find the highest existing sequence number in a directory
5988 * and then set the in-memory index_cnt variable to reflect
5989 * free sequence numbers
5990 */
5991 static int btrfs_set_inode_index_count(struct inode *inode)
5992 {
5993 struct btrfs_root *root = BTRFS_I(inode)->root;
5994 struct btrfs_key key, found_key;
5995 struct btrfs_path *path;
5996 struct extent_buffer *leaf;
5997 int ret;
5998
5999 key.objectid = btrfs_ino(inode);
6000 key.type = BTRFS_DIR_INDEX_KEY;
6001 key.offset = (u64)-1;
6002
6003 path = btrfs_alloc_path();
6004 if (!path)
6005 return -ENOMEM;
6006
6007 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6008 if (ret < 0)
6009 goto out;
6010 /* FIXME: we should be able to handle this */
6011 if (ret == 0)
6012 goto out;
6013 ret = 0;
6014
6015 /*
6016 * MAGIC NUMBER EXPLANATION:
6017 * since we search a directory based on f_pos we have to start at 2
6018 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6019 * else has to start at 2
6020 */
6021 if (path->slots[0] == 0) {
6022 BTRFS_I(inode)->index_cnt = 2;
6023 goto out;
6024 }
6025
6026 path->slots[0]--;
6027
6028 leaf = path->nodes[0];
6029 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6030
6031 if (found_key.objectid != btrfs_ino(inode) ||
6032 found_key.type != BTRFS_DIR_INDEX_KEY) {
6033 BTRFS_I(inode)->index_cnt = 2;
6034 goto out;
6035 }
6036
6037 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6038 out:
6039 btrfs_free_path(path);
6040 return ret;
6041 }
6042
6043 /*
6044 * helper to find a free sequence number in a given directory. This current
6045 * code is very simple, later versions will do smarter things in the btree
6046 */
6047 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6048 {
6049 int ret = 0;
6050
6051 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6052 ret = btrfs_inode_delayed_dir_index_count(dir);
6053 if (ret) {
6054 ret = btrfs_set_inode_index_count(dir);
6055 if (ret)
6056 return ret;
6057 }
6058 }
6059
6060 *index = BTRFS_I(dir)->index_cnt;
6061 BTRFS_I(dir)->index_cnt++;
6062
6063 return ret;
6064 }
6065
6066 static int btrfs_insert_inode_locked(struct inode *inode)
6067 {
6068 struct btrfs_iget_args args;
6069 args.location = &BTRFS_I(inode)->location;
6070 args.root = BTRFS_I(inode)->root;
6071
6072 return insert_inode_locked4(inode,
6073 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6074 btrfs_find_actor, &args);
6075 }
6076
6077 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6078 struct btrfs_root *root,
6079 struct inode *dir,
6080 const char *name, int name_len,
6081 u64 ref_objectid, u64 objectid,
6082 umode_t mode, u64 *index)
6083 {
6084 struct inode *inode;
6085 struct btrfs_inode_item *inode_item;
6086 struct btrfs_key *location;
6087 struct btrfs_path *path;
6088 struct btrfs_inode_ref *ref;
6089 struct btrfs_key key[2];
6090 u32 sizes[2];
6091 int nitems = name ? 2 : 1;
6092 unsigned long ptr;
6093 int ret;
6094
6095 path = btrfs_alloc_path();
6096 if (!path)
6097 return ERR_PTR(-ENOMEM);
6098
6099 inode = new_inode(root->fs_info->sb);
6100 if (!inode) {
6101 btrfs_free_path(path);
6102 return ERR_PTR(-ENOMEM);
6103 }
6104
6105 /*
6106 * O_TMPFILE, set link count to 0, so that after this point,
6107 * we fill in an inode item with the correct link count.
6108 */
6109 if (!name)
6110 set_nlink(inode, 0);
6111
6112 /*
6113 * we have to initialize this early, so we can reclaim the inode
6114 * number if we fail afterwards in this function.
6115 */
6116 inode->i_ino = objectid;
6117
6118 if (dir && name) {
6119 trace_btrfs_inode_request(dir);
6120
6121 ret = btrfs_set_inode_index(dir, index);
6122 if (ret) {
6123 btrfs_free_path(path);
6124 iput(inode);
6125 return ERR_PTR(ret);
6126 }
6127 } else if (dir) {
6128 *index = 0;
6129 }
6130 /*
6131 * index_cnt is ignored for everything but a dir,
6132 * btrfs_get_inode_index_count has an explanation for the magic
6133 * number
6134 */
6135 BTRFS_I(inode)->index_cnt = 2;
6136 BTRFS_I(inode)->dir_index = *index;
6137 BTRFS_I(inode)->root = root;
6138 BTRFS_I(inode)->generation = trans->transid;
6139 inode->i_generation = BTRFS_I(inode)->generation;
6140
6141 /*
6142 * We could have gotten an inode number from somebody who was fsynced
6143 * and then removed in this same transaction, so let's just set full
6144 * sync since it will be a full sync anyway and this will blow away the
6145 * old info in the log.
6146 */
6147 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6148
6149 key[0].objectid = objectid;
6150 key[0].type = BTRFS_INODE_ITEM_KEY;
6151 key[0].offset = 0;
6152
6153 sizes[0] = sizeof(struct btrfs_inode_item);
6154
6155 if (name) {
6156 /*
6157 * Start new inodes with an inode_ref. This is slightly more
6158 * efficient for small numbers of hard links since they will
6159 * be packed into one item. Extended refs will kick in if we
6160 * add more hard links than can fit in the ref item.
6161 */
6162 key[1].objectid = objectid;
6163 key[1].type = BTRFS_INODE_REF_KEY;
6164 key[1].offset = ref_objectid;
6165
6166 sizes[1] = name_len + sizeof(*ref);
6167 }
6168
6169 location = &BTRFS_I(inode)->location;
6170 location->objectid = objectid;
6171 location->offset = 0;
6172 location->type = BTRFS_INODE_ITEM_KEY;
6173
6174 ret = btrfs_insert_inode_locked(inode);
6175 if (ret < 0)
6176 goto fail;
6177
6178 path->leave_spinning = 1;
6179 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6180 if (ret != 0)
6181 goto fail_unlock;
6182
6183 inode_init_owner(inode, dir, mode);
6184 inode_set_bytes(inode, 0);
6185
6186 inode->i_mtime = CURRENT_TIME;
6187 inode->i_atime = inode->i_mtime;
6188 inode->i_ctime = inode->i_mtime;
6189 BTRFS_I(inode)->i_otime = inode->i_mtime;
6190
6191 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6192 struct btrfs_inode_item);
6193 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6194 sizeof(*inode_item));
6195 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6196
6197 if (name) {
6198 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6199 struct btrfs_inode_ref);
6200 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6201 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6202 ptr = (unsigned long)(ref + 1);
6203 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6204 }
6205
6206 btrfs_mark_buffer_dirty(path->nodes[0]);
6207 btrfs_free_path(path);
6208
6209 btrfs_inherit_iflags(inode, dir);
6210
6211 if (S_ISREG(mode)) {
6212 if (btrfs_test_opt(root, NODATASUM))
6213 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6214 if (btrfs_test_opt(root, NODATACOW))
6215 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6216 BTRFS_INODE_NODATASUM;
6217 }
6218
6219 inode_tree_add(inode);
6220
6221 trace_btrfs_inode_new(inode);
6222 btrfs_set_inode_last_trans(trans, inode);
6223
6224 btrfs_update_root_times(trans, root);
6225
6226 ret = btrfs_inode_inherit_props(trans, inode, dir);
6227 if (ret)
6228 btrfs_err(root->fs_info,
6229 "error inheriting props for ino %llu (root %llu): %d",
6230 btrfs_ino(inode), root->root_key.objectid, ret);
6231
6232 return inode;
6233
6234 fail_unlock:
6235 unlock_new_inode(inode);
6236 fail:
6237 if (dir && name)
6238 BTRFS_I(dir)->index_cnt--;
6239 btrfs_free_path(path);
6240 iput(inode);
6241 return ERR_PTR(ret);
6242 }
6243
6244 static inline u8 btrfs_inode_type(struct inode *inode)
6245 {
6246 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6247 }
6248
6249 /*
6250 * utility function to add 'inode' into 'parent_inode' with
6251 * a give name and a given sequence number.
6252 * if 'add_backref' is true, also insert a backref from the
6253 * inode to the parent directory.
6254 */
6255 int btrfs_add_link(struct btrfs_trans_handle *trans,
6256 struct inode *parent_inode, struct inode *inode,
6257 const char *name, int name_len, int add_backref, u64 index)
6258 {
6259 int ret = 0;
6260 struct btrfs_key key;
6261 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6262 u64 ino = btrfs_ino(inode);
6263 u64 parent_ino = btrfs_ino(parent_inode);
6264
6265 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6266 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6267 } else {
6268 key.objectid = ino;
6269 key.type = BTRFS_INODE_ITEM_KEY;
6270 key.offset = 0;
6271 }
6272
6273 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6274 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6275 key.objectid, root->root_key.objectid,
6276 parent_ino, index, name, name_len);
6277 } else if (add_backref) {
6278 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6279 parent_ino, index);
6280 }
6281
6282 /* Nothing to clean up yet */
6283 if (ret)
6284 return ret;
6285
6286 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6287 parent_inode, &key,
6288 btrfs_inode_type(inode), index);
6289 if (ret == -EEXIST || ret == -EOVERFLOW)
6290 goto fail_dir_item;
6291 else if (ret) {
6292 btrfs_abort_transaction(trans, root, ret);
6293 return ret;
6294 }
6295
6296 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6297 name_len * 2);
6298 inode_inc_iversion(parent_inode);
6299 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6300 ret = btrfs_update_inode(trans, root, parent_inode);
6301 if (ret)
6302 btrfs_abort_transaction(trans, root, ret);
6303 return ret;
6304
6305 fail_dir_item:
6306 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6307 u64 local_index;
6308 int err;
6309 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6310 key.objectid, root->root_key.objectid,
6311 parent_ino, &local_index, name, name_len);
6312
6313 } else if (add_backref) {
6314 u64 local_index;
6315 int err;
6316
6317 err = btrfs_del_inode_ref(trans, root, name, name_len,
6318 ino, parent_ino, &local_index);
6319 }
6320 return ret;
6321 }
6322
6323 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6324 struct inode *dir, struct dentry *dentry,
6325 struct inode *inode, int backref, u64 index)
6326 {
6327 int err = btrfs_add_link(trans, dir, inode,
6328 dentry->d_name.name, dentry->d_name.len,
6329 backref, index);
6330 if (err > 0)
6331 err = -EEXIST;
6332 return err;
6333 }
6334
6335 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6336 umode_t mode, dev_t rdev)
6337 {
6338 struct btrfs_trans_handle *trans;
6339 struct btrfs_root *root = BTRFS_I(dir)->root;
6340 struct inode *inode = NULL;
6341 int err;
6342 int drop_inode = 0;
6343 u64 objectid;
6344 u64 index = 0;
6345
6346 /*
6347 * 2 for inode item and ref
6348 * 2 for dir items
6349 * 1 for xattr if selinux is on
6350 */
6351 trans = btrfs_start_transaction(root, 5);
6352 if (IS_ERR(trans))
6353 return PTR_ERR(trans);
6354
6355 err = btrfs_find_free_ino(root, &objectid);
6356 if (err)
6357 goto out_unlock;
6358
6359 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6360 dentry->d_name.len, btrfs_ino(dir), objectid,
6361 mode, &index);
6362 if (IS_ERR(inode)) {
6363 err = PTR_ERR(inode);
6364 goto out_unlock;
6365 }
6366
6367 /*
6368 * If the active LSM wants to access the inode during
6369 * d_instantiate it needs these. Smack checks to see
6370 * if the filesystem supports xattrs by looking at the
6371 * ops vector.
6372 */
6373 inode->i_op = &btrfs_special_inode_operations;
6374 init_special_inode(inode, inode->i_mode, rdev);
6375
6376 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6377 if (err)
6378 goto out_unlock_inode;
6379
6380 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6381 if (err) {
6382 goto out_unlock_inode;
6383 } else {
6384 btrfs_update_inode(trans, root, inode);
6385 unlock_new_inode(inode);
6386 d_instantiate(dentry, inode);
6387 }
6388
6389 out_unlock:
6390 btrfs_end_transaction(trans, root);
6391 btrfs_balance_delayed_items(root);
6392 btrfs_btree_balance_dirty(root);
6393 if (drop_inode) {
6394 inode_dec_link_count(inode);
6395 iput(inode);
6396 }
6397 return err;
6398
6399 out_unlock_inode:
6400 drop_inode = 1;
6401 unlock_new_inode(inode);
6402 goto out_unlock;
6403
6404 }
6405
6406 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6407 umode_t mode, bool excl)
6408 {
6409 struct btrfs_trans_handle *trans;
6410 struct btrfs_root *root = BTRFS_I(dir)->root;
6411 struct inode *inode = NULL;
6412 int drop_inode_on_err = 0;
6413 int err;
6414 u64 objectid;
6415 u64 index = 0;
6416
6417 /*
6418 * 2 for inode item and ref
6419 * 2 for dir items
6420 * 1 for xattr if selinux is on
6421 */
6422 trans = btrfs_start_transaction(root, 5);
6423 if (IS_ERR(trans))
6424 return PTR_ERR(trans);
6425
6426 err = btrfs_find_free_ino(root, &objectid);
6427 if (err)
6428 goto out_unlock;
6429
6430 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6431 dentry->d_name.len, btrfs_ino(dir), objectid,
6432 mode, &index);
6433 if (IS_ERR(inode)) {
6434 err = PTR_ERR(inode);
6435 goto out_unlock;
6436 }
6437 drop_inode_on_err = 1;
6438 /*
6439 * If the active LSM wants to access the inode during
6440 * d_instantiate it needs these. Smack checks to see
6441 * if the filesystem supports xattrs by looking at the
6442 * ops vector.
6443 */
6444 inode->i_fop = &btrfs_file_operations;
6445 inode->i_op = &btrfs_file_inode_operations;
6446 inode->i_mapping->a_ops = &btrfs_aops;
6447
6448 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6449 if (err)
6450 goto out_unlock_inode;
6451
6452 err = btrfs_update_inode(trans, root, inode);
6453 if (err)
6454 goto out_unlock_inode;
6455
6456 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6457 if (err)
6458 goto out_unlock_inode;
6459
6460 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6461 unlock_new_inode(inode);
6462 d_instantiate(dentry, inode);
6463
6464 out_unlock:
6465 btrfs_end_transaction(trans, root);
6466 if (err && drop_inode_on_err) {
6467 inode_dec_link_count(inode);
6468 iput(inode);
6469 }
6470 btrfs_balance_delayed_items(root);
6471 btrfs_btree_balance_dirty(root);
6472 return err;
6473
6474 out_unlock_inode:
6475 unlock_new_inode(inode);
6476 goto out_unlock;
6477
6478 }
6479
6480 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6481 struct dentry *dentry)
6482 {
6483 struct btrfs_trans_handle *trans;
6484 struct btrfs_root *root = BTRFS_I(dir)->root;
6485 struct inode *inode = d_inode(old_dentry);
6486 u64 index;
6487 int err;
6488 int drop_inode = 0;
6489
6490 /* do not allow sys_link's with other subvols of the same device */
6491 if (root->objectid != BTRFS_I(inode)->root->objectid)
6492 return -EXDEV;
6493
6494 if (inode->i_nlink >= BTRFS_LINK_MAX)
6495 return -EMLINK;
6496
6497 err = btrfs_set_inode_index(dir, &index);
6498 if (err)
6499 goto fail;
6500
6501 /*
6502 * 2 items for inode and inode ref
6503 * 2 items for dir items
6504 * 1 item for parent inode
6505 */
6506 trans = btrfs_start_transaction(root, 5);
6507 if (IS_ERR(trans)) {
6508 err = PTR_ERR(trans);
6509 goto fail;
6510 }
6511
6512 /* There are several dir indexes for this inode, clear the cache. */
6513 BTRFS_I(inode)->dir_index = 0ULL;
6514 inc_nlink(inode);
6515 inode_inc_iversion(inode);
6516 inode->i_ctime = CURRENT_TIME;
6517 ihold(inode);
6518 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6519
6520 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6521
6522 if (err) {
6523 drop_inode = 1;
6524 } else {
6525 struct dentry *parent = dentry->d_parent;
6526 err = btrfs_update_inode(trans, root, inode);
6527 if (err)
6528 goto fail;
6529 if (inode->i_nlink == 1) {
6530 /*
6531 * If new hard link count is 1, it's a file created
6532 * with open(2) O_TMPFILE flag.
6533 */
6534 err = btrfs_orphan_del(trans, inode);
6535 if (err)
6536 goto fail;
6537 }
6538 d_instantiate(dentry, inode);
6539 btrfs_log_new_name(trans, inode, NULL, parent);
6540 }
6541
6542 btrfs_end_transaction(trans, root);
6543 btrfs_balance_delayed_items(root);
6544 fail:
6545 if (drop_inode) {
6546 inode_dec_link_count(inode);
6547 iput(inode);
6548 }
6549 btrfs_btree_balance_dirty(root);
6550 return err;
6551 }
6552
6553 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6554 {
6555 struct inode *inode = NULL;
6556 struct btrfs_trans_handle *trans;
6557 struct btrfs_root *root = BTRFS_I(dir)->root;
6558 int err = 0;
6559 int drop_on_err = 0;
6560 u64 objectid = 0;
6561 u64 index = 0;
6562
6563 /*
6564 * 2 items for inode and ref
6565 * 2 items for dir items
6566 * 1 for xattr if selinux is on
6567 */
6568 trans = btrfs_start_transaction(root, 5);
6569 if (IS_ERR(trans))
6570 return PTR_ERR(trans);
6571
6572 err = btrfs_find_free_ino(root, &objectid);
6573 if (err)
6574 goto out_fail;
6575
6576 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6577 dentry->d_name.len, btrfs_ino(dir), objectid,
6578 S_IFDIR | mode, &index);
6579 if (IS_ERR(inode)) {
6580 err = PTR_ERR(inode);
6581 goto out_fail;
6582 }
6583
6584 drop_on_err = 1;
6585 /* these must be set before we unlock the inode */
6586 inode->i_op = &btrfs_dir_inode_operations;
6587 inode->i_fop = &btrfs_dir_file_operations;
6588
6589 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6590 if (err)
6591 goto out_fail_inode;
6592
6593 btrfs_i_size_write(inode, 0);
6594 err = btrfs_update_inode(trans, root, inode);
6595 if (err)
6596 goto out_fail_inode;
6597
6598 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6599 dentry->d_name.len, 0, index);
6600 if (err)
6601 goto out_fail_inode;
6602
6603 d_instantiate(dentry, inode);
6604 /*
6605 * mkdir is special. We're unlocking after we call d_instantiate
6606 * to avoid a race with nfsd calling d_instantiate.
6607 */
6608 unlock_new_inode(inode);
6609 drop_on_err = 0;
6610
6611 out_fail:
6612 btrfs_end_transaction(trans, root);
6613 if (drop_on_err) {
6614 inode_dec_link_count(inode);
6615 iput(inode);
6616 }
6617 btrfs_balance_delayed_items(root);
6618 btrfs_btree_balance_dirty(root);
6619 return err;
6620
6621 out_fail_inode:
6622 unlock_new_inode(inode);
6623 goto out_fail;
6624 }
6625
6626 /* Find next extent map of a given extent map, caller needs to ensure locks */
6627 static struct extent_map *next_extent_map(struct extent_map *em)
6628 {
6629 struct rb_node *next;
6630
6631 next = rb_next(&em->rb_node);
6632 if (!next)
6633 return NULL;
6634 return container_of(next, struct extent_map, rb_node);
6635 }
6636
6637 static struct extent_map *prev_extent_map(struct extent_map *em)
6638 {
6639 struct rb_node *prev;
6640
6641 prev = rb_prev(&em->rb_node);
6642 if (!prev)
6643 return NULL;
6644 return container_of(prev, struct extent_map, rb_node);
6645 }
6646
6647 /* helper for btfs_get_extent. Given an existing extent in the tree,
6648 * the existing extent is the nearest extent to map_start,
6649 * and an extent that you want to insert, deal with overlap and insert
6650 * the best fitted new extent into the tree.
6651 */
6652 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6653 struct extent_map *existing,
6654 struct extent_map *em,
6655 u64 map_start)
6656 {
6657 struct extent_map *prev;
6658 struct extent_map *next;
6659 u64 start;
6660 u64 end;
6661 u64 start_diff;
6662
6663 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6664
6665 if (existing->start > map_start) {
6666 next = existing;
6667 prev = prev_extent_map(next);
6668 } else {
6669 prev = existing;
6670 next = next_extent_map(prev);
6671 }
6672
6673 start = prev ? extent_map_end(prev) : em->start;
6674 start = max_t(u64, start, em->start);
6675 end = next ? next->start : extent_map_end(em);
6676 end = min_t(u64, end, extent_map_end(em));
6677 start_diff = start - em->start;
6678 em->start = start;
6679 em->len = end - start;
6680 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6681 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6682 em->block_start += start_diff;
6683 em->block_len -= start_diff;
6684 }
6685 return add_extent_mapping(em_tree, em, 0);
6686 }
6687
6688 static noinline int uncompress_inline(struct btrfs_path *path,
6689 struct page *page,
6690 size_t pg_offset, u64 extent_offset,
6691 struct btrfs_file_extent_item *item)
6692 {
6693 int ret;
6694 struct extent_buffer *leaf = path->nodes[0];
6695 char *tmp;
6696 size_t max_size;
6697 unsigned long inline_size;
6698 unsigned long ptr;
6699 int compress_type;
6700
6701 WARN_ON(pg_offset != 0);
6702 compress_type = btrfs_file_extent_compression(leaf, item);
6703 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6704 inline_size = btrfs_file_extent_inline_item_len(leaf,
6705 btrfs_item_nr(path->slots[0]));
6706 tmp = kmalloc(inline_size, GFP_NOFS);
6707 if (!tmp)
6708 return -ENOMEM;
6709 ptr = btrfs_file_extent_inline_start(item);
6710
6711 read_extent_buffer(leaf, tmp, ptr, inline_size);
6712
6713 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6714 ret = btrfs_decompress(compress_type, tmp, page,
6715 extent_offset, inline_size, max_size);
6716 kfree(tmp);
6717 return ret;
6718 }
6719
6720 /*
6721 * a bit scary, this does extent mapping from logical file offset to the disk.
6722 * the ugly parts come from merging extents from the disk with the in-ram
6723 * representation. This gets more complex because of the data=ordered code,
6724 * where the in-ram extents might be locked pending data=ordered completion.
6725 *
6726 * This also copies inline extents directly into the page.
6727 */
6728
6729 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6730 size_t pg_offset, u64 start, u64 len,
6731 int create)
6732 {
6733 int ret;
6734 int err = 0;
6735 u64 extent_start = 0;
6736 u64 extent_end = 0;
6737 u64 objectid = btrfs_ino(inode);
6738 u32 found_type;
6739 struct btrfs_path *path = NULL;
6740 struct btrfs_root *root = BTRFS_I(inode)->root;
6741 struct btrfs_file_extent_item *item;
6742 struct extent_buffer *leaf;
6743 struct btrfs_key found_key;
6744 struct extent_map *em = NULL;
6745 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6746 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6747 struct btrfs_trans_handle *trans = NULL;
6748 const bool new_inline = !page || create;
6749
6750 again:
6751 read_lock(&em_tree->lock);
6752 em = lookup_extent_mapping(em_tree, start, len);
6753 if (em)
6754 em->bdev = root->fs_info->fs_devices->latest_bdev;
6755 read_unlock(&em_tree->lock);
6756
6757 if (em) {
6758 if (em->start > start || em->start + em->len <= start)
6759 free_extent_map(em);
6760 else if (em->block_start == EXTENT_MAP_INLINE && page)
6761 free_extent_map(em);
6762 else
6763 goto out;
6764 }
6765 em = alloc_extent_map();
6766 if (!em) {
6767 err = -ENOMEM;
6768 goto out;
6769 }
6770 em->bdev = root->fs_info->fs_devices->latest_bdev;
6771 em->start = EXTENT_MAP_HOLE;
6772 em->orig_start = EXTENT_MAP_HOLE;
6773 em->len = (u64)-1;
6774 em->block_len = (u64)-1;
6775
6776 if (!path) {
6777 path = btrfs_alloc_path();
6778 if (!path) {
6779 err = -ENOMEM;
6780 goto out;
6781 }
6782 /*
6783 * Chances are we'll be called again, so go ahead and do
6784 * readahead
6785 */
6786 path->reada = 1;
6787 }
6788
6789 ret = btrfs_lookup_file_extent(trans, root, path,
6790 objectid, start, trans != NULL);
6791 if (ret < 0) {
6792 err = ret;
6793 goto out;
6794 }
6795
6796 if (ret != 0) {
6797 if (path->slots[0] == 0)
6798 goto not_found;
6799 path->slots[0]--;
6800 }
6801
6802 leaf = path->nodes[0];
6803 item = btrfs_item_ptr(leaf, path->slots[0],
6804 struct btrfs_file_extent_item);
6805 /* are we inside the extent that was found? */
6806 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6807 found_type = found_key.type;
6808 if (found_key.objectid != objectid ||
6809 found_type != BTRFS_EXTENT_DATA_KEY) {
6810 /*
6811 * If we backup past the first extent we want to move forward
6812 * and see if there is an extent in front of us, otherwise we'll
6813 * say there is a hole for our whole search range which can
6814 * cause problems.
6815 */
6816 extent_end = start;
6817 goto next;
6818 }
6819
6820 found_type = btrfs_file_extent_type(leaf, item);
6821 extent_start = found_key.offset;
6822 if (found_type == BTRFS_FILE_EXTENT_REG ||
6823 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6824 extent_end = extent_start +
6825 btrfs_file_extent_num_bytes(leaf, item);
6826 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6827 size_t size;
6828 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6829 extent_end = ALIGN(extent_start + size, root->sectorsize);
6830 }
6831 next:
6832 if (start >= extent_end) {
6833 path->slots[0]++;
6834 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6835 ret = btrfs_next_leaf(root, path);
6836 if (ret < 0) {
6837 err = ret;
6838 goto out;
6839 }
6840 if (ret > 0)
6841 goto not_found;
6842 leaf = path->nodes[0];
6843 }
6844 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6845 if (found_key.objectid != objectid ||
6846 found_key.type != BTRFS_EXTENT_DATA_KEY)
6847 goto not_found;
6848 if (start + len <= found_key.offset)
6849 goto not_found;
6850 if (start > found_key.offset)
6851 goto next;
6852 em->start = start;
6853 em->orig_start = start;
6854 em->len = found_key.offset - start;
6855 goto not_found_em;
6856 }
6857
6858 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6859
6860 if (found_type == BTRFS_FILE_EXTENT_REG ||
6861 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6862 goto insert;
6863 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6864 unsigned long ptr;
6865 char *map;
6866 size_t size;
6867 size_t extent_offset;
6868 size_t copy_size;
6869
6870 if (new_inline)
6871 goto out;
6872
6873 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6874 extent_offset = page_offset(page) + pg_offset - extent_start;
6875 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6876 size - extent_offset);
6877 em->start = extent_start + extent_offset;
6878 em->len = ALIGN(copy_size, root->sectorsize);
6879 em->orig_block_len = em->len;
6880 em->orig_start = em->start;
6881 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6882 if (create == 0 && !PageUptodate(page)) {
6883 if (btrfs_file_extent_compression(leaf, item) !=
6884 BTRFS_COMPRESS_NONE) {
6885 ret = uncompress_inline(path, page, pg_offset,
6886 extent_offset, item);
6887 if (ret) {
6888 err = ret;
6889 goto out;
6890 }
6891 } else {
6892 map = kmap(page);
6893 read_extent_buffer(leaf, map + pg_offset, ptr,
6894 copy_size);
6895 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6896 memset(map + pg_offset + copy_size, 0,
6897 PAGE_CACHE_SIZE - pg_offset -
6898 copy_size);
6899 }
6900 kunmap(page);
6901 }
6902 flush_dcache_page(page);
6903 } else if (create && PageUptodate(page)) {
6904 BUG();
6905 if (!trans) {
6906 kunmap(page);
6907 free_extent_map(em);
6908 em = NULL;
6909
6910 btrfs_release_path(path);
6911 trans = btrfs_join_transaction(root);
6912
6913 if (IS_ERR(trans))
6914 return ERR_CAST(trans);
6915 goto again;
6916 }
6917 map = kmap(page);
6918 write_extent_buffer(leaf, map + pg_offset, ptr,
6919 copy_size);
6920 kunmap(page);
6921 btrfs_mark_buffer_dirty(leaf);
6922 }
6923 set_extent_uptodate(io_tree, em->start,
6924 extent_map_end(em) - 1, NULL, GFP_NOFS);
6925 goto insert;
6926 }
6927 not_found:
6928 em->start = start;
6929 em->orig_start = start;
6930 em->len = len;
6931 not_found_em:
6932 em->block_start = EXTENT_MAP_HOLE;
6933 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6934 insert:
6935 btrfs_release_path(path);
6936 if (em->start > start || extent_map_end(em) <= start) {
6937 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6938 em->start, em->len, start, len);
6939 err = -EIO;
6940 goto out;
6941 }
6942
6943 err = 0;
6944 write_lock(&em_tree->lock);
6945 ret = add_extent_mapping(em_tree, em, 0);
6946 /* it is possible that someone inserted the extent into the tree
6947 * while we had the lock dropped. It is also possible that
6948 * an overlapping map exists in the tree
6949 */
6950 if (ret == -EEXIST) {
6951 struct extent_map *existing;
6952
6953 ret = 0;
6954
6955 existing = search_extent_mapping(em_tree, start, len);
6956 /*
6957 * existing will always be non-NULL, since there must be
6958 * extent causing the -EEXIST.
6959 */
6960 if (start >= extent_map_end(existing) ||
6961 start <= existing->start) {
6962 /*
6963 * The existing extent map is the one nearest to
6964 * the [start, start + len) range which overlaps
6965 */
6966 err = merge_extent_mapping(em_tree, existing,
6967 em, start);
6968 free_extent_map(existing);
6969 if (err) {
6970 free_extent_map(em);
6971 em = NULL;
6972 }
6973 } else {
6974 free_extent_map(em);
6975 em = existing;
6976 err = 0;
6977 }
6978 }
6979 write_unlock(&em_tree->lock);
6980 out:
6981
6982 trace_btrfs_get_extent(root, em);
6983
6984 btrfs_free_path(path);
6985 if (trans) {
6986 ret = btrfs_end_transaction(trans, root);
6987 if (!err)
6988 err = ret;
6989 }
6990 if (err) {
6991 free_extent_map(em);
6992 return ERR_PTR(err);
6993 }
6994 BUG_ON(!em); /* Error is always set */
6995 return em;
6996 }
6997
6998 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6999 size_t pg_offset, u64 start, u64 len,
7000 int create)
7001 {
7002 struct extent_map *em;
7003 struct extent_map *hole_em = NULL;
7004 u64 range_start = start;
7005 u64 end;
7006 u64 found;
7007 u64 found_end;
7008 int err = 0;
7009
7010 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7011 if (IS_ERR(em))
7012 return em;
7013 if (em) {
7014 /*
7015 * if our em maps to
7016 * - a hole or
7017 * - a pre-alloc extent,
7018 * there might actually be delalloc bytes behind it.
7019 */
7020 if (em->block_start != EXTENT_MAP_HOLE &&
7021 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7022 return em;
7023 else
7024 hole_em = em;
7025 }
7026
7027 /* check to see if we've wrapped (len == -1 or similar) */
7028 end = start + len;
7029 if (end < start)
7030 end = (u64)-1;
7031 else
7032 end -= 1;
7033
7034 em = NULL;
7035
7036 /* ok, we didn't find anything, lets look for delalloc */
7037 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7038 end, len, EXTENT_DELALLOC, 1);
7039 found_end = range_start + found;
7040 if (found_end < range_start)
7041 found_end = (u64)-1;
7042
7043 /*
7044 * we didn't find anything useful, return
7045 * the original results from get_extent()
7046 */
7047 if (range_start > end || found_end <= start) {
7048 em = hole_em;
7049 hole_em = NULL;
7050 goto out;
7051 }
7052
7053 /* adjust the range_start to make sure it doesn't
7054 * go backwards from the start they passed in
7055 */
7056 range_start = max(start, range_start);
7057 found = found_end - range_start;
7058
7059 if (found > 0) {
7060 u64 hole_start = start;
7061 u64 hole_len = len;
7062
7063 em = alloc_extent_map();
7064 if (!em) {
7065 err = -ENOMEM;
7066 goto out;
7067 }
7068 /*
7069 * when btrfs_get_extent can't find anything it
7070 * returns one huge hole
7071 *
7072 * make sure what it found really fits our range, and
7073 * adjust to make sure it is based on the start from
7074 * the caller
7075 */
7076 if (hole_em) {
7077 u64 calc_end = extent_map_end(hole_em);
7078
7079 if (calc_end <= start || (hole_em->start > end)) {
7080 free_extent_map(hole_em);
7081 hole_em = NULL;
7082 } else {
7083 hole_start = max(hole_em->start, start);
7084 hole_len = calc_end - hole_start;
7085 }
7086 }
7087 em->bdev = NULL;
7088 if (hole_em && range_start > hole_start) {
7089 /* our hole starts before our delalloc, so we
7090 * have to return just the parts of the hole
7091 * that go until the delalloc starts
7092 */
7093 em->len = min(hole_len,
7094 range_start - hole_start);
7095 em->start = hole_start;
7096 em->orig_start = hole_start;
7097 /*
7098 * don't adjust block start at all,
7099 * it is fixed at EXTENT_MAP_HOLE
7100 */
7101 em->block_start = hole_em->block_start;
7102 em->block_len = hole_len;
7103 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7104 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7105 } else {
7106 em->start = range_start;
7107 em->len = found;
7108 em->orig_start = range_start;
7109 em->block_start = EXTENT_MAP_DELALLOC;
7110 em->block_len = found;
7111 }
7112 } else if (hole_em) {
7113 return hole_em;
7114 }
7115 out:
7116
7117 free_extent_map(hole_em);
7118 if (err) {
7119 free_extent_map(em);
7120 return ERR_PTR(err);
7121 }
7122 return em;
7123 }
7124
7125 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7126 u64 start, u64 len)
7127 {
7128 struct btrfs_root *root = BTRFS_I(inode)->root;
7129 struct extent_map *em;
7130 struct btrfs_key ins;
7131 u64 alloc_hint;
7132 int ret;
7133
7134 alloc_hint = get_extent_allocation_hint(inode, start, len);
7135 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7136 alloc_hint, &ins, 1, 1);
7137 if (ret)
7138 return ERR_PTR(ret);
7139
7140 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7141 ins.offset, ins.offset, ins.offset, 0);
7142 if (IS_ERR(em)) {
7143 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7144 return em;
7145 }
7146
7147 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7148 ins.offset, ins.offset, 0);
7149 if (ret) {
7150 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7151 free_extent_map(em);
7152 return ERR_PTR(ret);
7153 }
7154
7155 return em;
7156 }
7157
7158 /*
7159 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7160 * block must be cow'd
7161 */
7162 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7163 u64 *orig_start, u64 *orig_block_len,
7164 u64 *ram_bytes)
7165 {
7166 struct btrfs_trans_handle *trans;
7167 struct btrfs_path *path;
7168 int ret;
7169 struct extent_buffer *leaf;
7170 struct btrfs_root *root = BTRFS_I(inode)->root;
7171 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7172 struct btrfs_file_extent_item *fi;
7173 struct btrfs_key key;
7174 u64 disk_bytenr;
7175 u64 backref_offset;
7176 u64 extent_end;
7177 u64 num_bytes;
7178 int slot;
7179 int found_type;
7180 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7181
7182 path = btrfs_alloc_path();
7183 if (!path)
7184 return -ENOMEM;
7185
7186 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7187 offset, 0);
7188 if (ret < 0)
7189 goto out;
7190
7191 slot = path->slots[0];
7192 if (ret == 1) {
7193 if (slot == 0) {
7194 /* can't find the item, must cow */
7195 ret = 0;
7196 goto out;
7197 }
7198 slot--;
7199 }
7200 ret = 0;
7201 leaf = path->nodes[0];
7202 btrfs_item_key_to_cpu(leaf, &key, slot);
7203 if (key.objectid != btrfs_ino(inode) ||
7204 key.type != BTRFS_EXTENT_DATA_KEY) {
7205 /* not our file or wrong item type, must cow */
7206 goto out;
7207 }
7208
7209 if (key.offset > offset) {
7210 /* Wrong offset, must cow */
7211 goto out;
7212 }
7213
7214 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7215 found_type = btrfs_file_extent_type(leaf, fi);
7216 if (found_type != BTRFS_FILE_EXTENT_REG &&
7217 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7218 /* not a regular extent, must cow */
7219 goto out;
7220 }
7221
7222 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7223 goto out;
7224
7225 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7226 if (extent_end <= offset)
7227 goto out;
7228
7229 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7230 if (disk_bytenr == 0)
7231 goto out;
7232
7233 if (btrfs_file_extent_compression(leaf, fi) ||
7234 btrfs_file_extent_encryption(leaf, fi) ||
7235 btrfs_file_extent_other_encoding(leaf, fi))
7236 goto out;
7237
7238 backref_offset = btrfs_file_extent_offset(leaf, fi);
7239
7240 if (orig_start) {
7241 *orig_start = key.offset - backref_offset;
7242 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7243 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7244 }
7245
7246 if (btrfs_extent_readonly(root, disk_bytenr))
7247 goto out;
7248
7249 num_bytes = min(offset + *len, extent_end) - offset;
7250 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7251 u64 range_end;
7252
7253 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7254 ret = test_range_bit(io_tree, offset, range_end,
7255 EXTENT_DELALLOC, 0, NULL);
7256 if (ret) {
7257 ret = -EAGAIN;
7258 goto out;
7259 }
7260 }
7261
7262 btrfs_release_path(path);
7263
7264 /*
7265 * look for other files referencing this extent, if we
7266 * find any we must cow
7267 */
7268 trans = btrfs_join_transaction(root);
7269 if (IS_ERR(trans)) {
7270 ret = 0;
7271 goto out;
7272 }
7273
7274 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7275 key.offset - backref_offset, disk_bytenr);
7276 btrfs_end_transaction(trans, root);
7277 if (ret) {
7278 ret = 0;
7279 goto out;
7280 }
7281
7282 /*
7283 * adjust disk_bytenr and num_bytes to cover just the bytes
7284 * in this extent we are about to write. If there
7285 * are any csums in that range we have to cow in order
7286 * to keep the csums correct
7287 */
7288 disk_bytenr += backref_offset;
7289 disk_bytenr += offset - key.offset;
7290 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7291 goto out;
7292 /*
7293 * all of the above have passed, it is safe to overwrite this extent
7294 * without cow
7295 */
7296 *len = num_bytes;
7297 ret = 1;
7298 out:
7299 btrfs_free_path(path);
7300 return ret;
7301 }
7302
7303 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7304 {
7305 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7306 int found = false;
7307 void **pagep = NULL;
7308 struct page *page = NULL;
7309 int start_idx;
7310 int end_idx;
7311
7312 start_idx = start >> PAGE_CACHE_SHIFT;
7313
7314 /*
7315 * end is the last byte in the last page. end == start is legal
7316 */
7317 end_idx = end >> PAGE_CACHE_SHIFT;
7318
7319 rcu_read_lock();
7320
7321 /* Most of the code in this while loop is lifted from
7322 * find_get_page. It's been modified to begin searching from a
7323 * page and return just the first page found in that range. If the
7324 * found idx is less than or equal to the end idx then we know that
7325 * a page exists. If no pages are found or if those pages are
7326 * outside of the range then we're fine (yay!) */
7327 while (page == NULL &&
7328 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7329 page = radix_tree_deref_slot(pagep);
7330 if (unlikely(!page))
7331 break;
7332
7333 if (radix_tree_exception(page)) {
7334 if (radix_tree_deref_retry(page)) {
7335 page = NULL;
7336 continue;
7337 }
7338 /*
7339 * Otherwise, shmem/tmpfs must be storing a swap entry
7340 * here as an exceptional entry: so return it without
7341 * attempting to raise page count.
7342 */
7343 page = NULL;
7344 break; /* TODO: Is this relevant for this use case? */
7345 }
7346
7347 if (!page_cache_get_speculative(page)) {
7348 page = NULL;
7349 continue;
7350 }
7351
7352 /*
7353 * Has the page moved?
7354 * This is part of the lockless pagecache protocol. See
7355 * include/linux/pagemap.h for details.
7356 */
7357 if (unlikely(page != *pagep)) {
7358 page_cache_release(page);
7359 page = NULL;
7360 }
7361 }
7362
7363 if (page) {
7364 if (page->index <= end_idx)
7365 found = true;
7366 page_cache_release(page);
7367 }
7368
7369 rcu_read_unlock();
7370 return found;
7371 }
7372
7373 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7374 struct extent_state **cached_state, int writing)
7375 {
7376 struct btrfs_ordered_extent *ordered;
7377 int ret = 0;
7378
7379 while (1) {
7380 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7381 0, cached_state);
7382 /*
7383 * We're concerned with the entire range that we're going to be
7384 * doing DIO to, so we need to make sure theres no ordered
7385 * extents in this range.
7386 */
7387 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7388 lockend - lockstart + 1);
7389
7390 /*
7391 * We need to make sure there are no buffered pages in this
7392 * range either, we could have raced between the invalidate in
7393 * generic_file_direct_write and locking the extent. The
7394 * invalidate needs to happen so that reads after a write do not
7395 * get stale data.
7396 */
7397 if (!ordered &&
7398 (!writing ||
7399 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7400 break;
7401
7402 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7403 cached_state, GFP_NOFS);
7404
7405 if (ordered) {
7406 btrfs_start_ordered_extent(inode, ordered, 1);
7407 btrfs_put_ordered_extent(ordered);
7408 } else {
7409 /* Screw you mmap */
7410 ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
7411 if (ret)
7412 break;
7413 ret = filemap_fdatawait_range(inode->i_mapping,
7414 lockstart,
7415 lockend);
7416 if (ret)
7417 break;
7418
7419 /*
7420 * If we found a page that couldn't be invalidated just
7421 * fall back to buffered.
7422 */
7423 ret = invalidate_inode_pages2_range(inode->i_mapping,
7424 lockstart >> PAGE_CACHE_SHIFT,
7425 lockend >> PAGE_CACHE_SHIFT);
7426 if (ret)
7427 break;
7428 }
7429
7430 cond_resched();
7431 }
7432
7433 return ret;
7434 }
7435
7436 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7437 u64 len, u64 orig_start,
7438 u64 block_start, u64 block_len,
7439 u64 orig_block_len, u64 ram_bytes,
7440 int type)
7441 {
7442 struct extent_map_tree *em_tree;
7443 struct extent_map *em;
7444 struct btrfs_root *root = BTRFS_I(inode)->root;
7445 int ret;
7446
7447 em_tree = &BTRFS_I(inode)->extent_tree;
7448 em = alloc_extent_map();
7449 if (!em)
7450 return ERR_PTR(-ENOMEM);
7451
7452 em->start = start;
7453 em->orig_start = orig_start;
7454 em->mod_start = start;
7455 em->mod_len = len;
7456 em->len = len;
7457 em->block_len = block_len;
7458 em->block_start = block_start;
7459 em->bdev = root->fs_info->fs_devices->latest_bdev;
7460 em->orig_block_len = orig_block_len;
7461 em->ram_bytes = ram_bytes;
7462 em->generation = -1;
7463 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7464 if (type == BTRFS_ORDERED_PREALLOC)
7465 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7466
7467 do {
7468 btrfs_drop_extent_cache(inode, em->start,
7469 em->start + em->len - 1, 0);
7470 write_lock(&em_tree->lock);
7471 ret = add_extent_mapping(em_tree, em, 1);
7472 write_unlock(&em_tree->lock);
7473 } while (ret == -EEXIST);
7474
7475 if (ret) {
7476 free_extent_map(em);
7477 return ERR_PTR(ret);
7478 }
7479
7480 return em;
7481 }
7482
7483 struct btrfs_dio_data {
7484 u64 outstanding_extents;
7485 u64 reserve;
7486 };
7487
7488 static void adjust_dio_outstanding_extents(struct inode *inode,
7489 struct btrfs_dio_data *dio_data,
7490 const u64 len)
7491 {
7492 unsigned num_extents;
7493
7494 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7495 BTRFS_MAX_EXTENT_SIZE);
7496 /*
7497 * If we have an outstanding_extents count still set then we're
7498 * within our reservation, otherwise we need to adjust our inode
7499 * counter appropriately.
7500 */
7501 if (dio_data->outstanding_extents) {
7502 dio_data->outstanding_extents -= num_extents;
7503 } else {
7504 spin_lock(&BTRFS_I(inode)->lock);
7505 BTRFS_I(inode)->outstanding_extents += num_extents;
7506 spin_unlock(&BTRFS_I(inode)->lock);
7507 }
7508 }
7509
7510 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7511 struct buffer_head *bh_result, int create)
7512 {
7513 struct extent_map *em;
7514 struct btrfs_root *root = BTRFS_I(inode)->root;
7515 struct extent_state *cached_state = NULL;
7516 struct btrfs_dio_data *dio_data = NULL;
7517 u64 start = iblock << inode->i_blkbits;
7518 u64 lockstart, lockend;
7519 u64 len = bh_result->b_size;
7520 int unlock_bits = EXTENT_LOCKED;
7521 int ret = 0;
7522
7523 if (create)
7524 unlock_bits |= EXTENT_DIRTY;
7525 else
7526 len = min_t(u64, len, root->sectorsize);
7527
7528 lockstart = start;
7529 lockend = start + len - 1;
7530
7531 if (current->journal_info) {
7532 /*
7533 * Need to pull our outstanding extents and set journal_info to NULL so
7534 * that anything that needs to check if there's a transction doesn't get
7535 * confused.
7536 */
7537 dio_data = current->journal_info;
7538 current->journal_info = NULL;
7539 }
7540
7541 /*
7542 * If this errors out it's because we couldn't invalidate pagecache for
7543 * this range and we need to fallback to buffered.
7544 */
7545 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7546 create)) {
7547 ret = -ENOTBLK;
7548 goto err;
7549 }
7550
7551 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7552 if (IS_ERR(em)) {
7553 ret = PTR_ERR(em);
7554 goto unlock_err;
7555 }
7556
7557 /*
7558 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7559 * io. INLINE is special, and we could probably kludge it in here, but
7560 * it's still buffered so for safety lets just fall back to the generic
7561 * buffered path.
7562 *
7563 * For COMPRESSED we _have_ to read the entire extent in so we can
7564 * decompress it, so there will be buffering required no matter what we
7565 * do, so go ahead and fallback to buffered.
7566 *
7567 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7568 * to buffered IO. Don't blame me, this is the price we pay for using
7569 * the generic code.
7570 */
7571 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7572 em->block_start == EXTENT_MAP_INLINE) {
7573 free_extent_map(em);
7574 ret = -ENOTBLK;
7575 goto unlock_err;
7576 }
7577
7578 /* Just a good old fashioned hole, return */
7579 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7580 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7581 free_extent_map(em);
7582 goto unlock_err;
7583 }
7584
7585 /*
7586 * We don't allocate a new extent in the following cases
7587 *
7588 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7589 * existing extent.
7590 * 2) The extent is marked as PREALLOC. We're good to go here and can
7591 * just use the extent.
7592 *
7593 */
7594 if (!create) {
7595 len = min(len, em->len - (start - em->start));
7596 lockstart = start + len;
7597 goto unlock;
7598 }
7599
7600 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7601 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7602 em->block_start != EXTENT_MAP_HOLE)) {
7603 int type;
7604 u64 block_start, orig_start, orig_block_len, ram_bytes;
7605
7606 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7607 type = BTRFS_ORDERED_PREALLOC;
7608 else
7609 type = BTRFS_ORDERED_NOCOW;
7610 len = min(len, em->len - (start - em->start));
7611 block_start = em->block_start + (start - em->start);
7612
7613 if (can_nocow_extent(inode, start, &len, &orig_start,
7614 &orig_block_len, &ram_bytes) == 1) {
7615 if (type == BTRFS_ORDERED_PREALLOC) {
7616 free_extent_map(em);
7617 em = create_pinned_em(inode, start, len,
7618 orig_start,
7619 block_start, len,
7620 orig_block_len,
7621 ram_bytes, type);
7622 if (IS_ERR(em)) {
7623 ret = PTR_ERR(em);
7624 goto unlock_err;
7625 }
7626 }
7627
7628 ret = btrfs_add_ordered_extent_dio(inode, start,
7629 block_start, len, len, type);
7630 if (ret) {
7631 free_extent_map(em);
7632 goto unlock_err;
7633 }
7634 goto unlock;
7635 }
7636 }
7637
7638 /*
7639 * this will cow the extent, reset the len in case we changed
7640 * it above
7641 */
7642 len = bh_result->b_size;
7643 free_extent_map(em);
7644 em = btrfs_new_extent_direct(inode, start, len);
7645 if (IS_ERR(em)) {
7646 ret = PTR_ERR(em);
7647 goto unlock_err;
7648 }
7649 len = min(len, em->len - (start - em->start));
7650 unlock:
7651 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7652 inode->i_blkbits;
7653 bh_result->b_size = len;
7654 bh_result->b_bdev = em->bdev;
7655 set_buffer_mapped(bh_result);
7656 if (create) {
7657 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7658 set_buffer_new(bh_result);
7659
7660 /*
7661 * Need to update the i_size under the extent lock so buffered
7662 * readers will get the updated i_size when we unlock.
7663 */
7664 if (start + len > i_size_read(inode))
7665 i_size_write(inode, start + len);
7666
7667 adjust_dio_outstanding_extents(inode, dio_data, len);
7668 btrfs_free_reserved_data_space(inode, start, len);
7669 WARN_ON(dio_data->reserve < len);
7670 dio_data->reserve -= len;
7671 current->journal_info = dio_data;
7672 }
7673
7674 /*
7675 * In the case of write we need to clear and unlock the entire range,
7676 * in the case of read we need to unlock only the end area that we
7677 * aren't using if there is any left over space.
7678 */
7679 if (lockstart < lockend) {
7680 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7681 lockend, unlock_bits, 1, 0,
7682 &cached_state, GFP_NOFS);
7683 } else {
7684 free_extent_state(cached_state);
7685 }
7686
7687 free_extent_map(em);
7688
7689 return 0;
7690
7691 unlock_err:
7692 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7693 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7694 err:
7695 if (dio_data)
7696 current->journal_info = dio_data;
7697 /*
7698 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7699 * write less data then expected, so that we don't underflow our inode's
7700 * outstanding extents counter.
7701 */
7702 if (create && dio_data)
7703 adjust_dio_outstanding_extents(inode, dio_data, len);
7704
7705 return ret;
7706 }
7707
7708 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7709 int rw, int mirror_num)
7710 {
7711 struct btrfs_root *root = BTRFS_I(inode)->root;
7712 int ret;
7713
7714 BUG_ON(rw & REQ_WRITE);
7715
7716 bio_get(bio);
7717
7718 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7719 BTRFS_WQ_ENDIO_DIO_REPAIR);
7720 if (ret)
7721 goto err;
7722
7723 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7724 err:
7725 bio_put(bio);
7726 return ret;
7727 }
7728
7729 static int btrfs_check_dio_repairable(struct inode *inode,
7730 struct bio *failed_bio,
7731 struct io_failure_record *failrec,
7732 int failed_mirror)
7733 {
7734 int num_copies;
7735
7736 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7737 failrec->logical, failrec->len);
7738 if (num_copies == 1) {
7739 /*
7740 * we only have a single copy of the data, so don't bother with
7741 * all the retry and error correction code that follows. no
7742 * matter what the error is, it is very likely to persist.
7743 */
7744 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7745 num_copies, failrec->this_mirror, failed_mirror);
7746 return 0;
7747 }
7748
7749 failrec->failed_mirror = failed_mirror;
7750 failrec->this_mirror++;
7751 if (failrec->this_mirror == failed_mirror)
7752 failrec->this_mirror++;
7753
7754 if (failrec->this_mirror > num_copies) {
7755 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7756 num_copies, failrec->this_mirror, failed_mirror);
7757 return 0;
7758 }
7759
7760 return 1;
7761 }
7762
7763 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7764 struct page *page, u64 start, u64 end,
7765 int failed_mirror, bio_end_io_t *repair_endio,
7766 void *repair_arg)
7767 {
7768 struct io_failure_record *failrec;
7769 struct bio *bio;
7770 int isector;
7771 int read_mode;
7772 int ret;
7773
7774 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7775
7776 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7777 if (ret)
7778 return ret;
7779
7780 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7781 failed_mirror);
7782 if (!ret) {
7783 free_io_failure(inode, failrec);
7784 return -EIO;
7785 }
7786
7787 if (failed_bio->bi_vcnt > 1)
7788 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7789 else
7790 read_mode = READ_SYNC;
7791
7792 isector = start - btrfs_io_bio(failed_bio)->logical;
7793 isector >>= inode->i_sb->s_blocksize_bits;
7794 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7795 0, isector, repair_endio, repair_arg);
7796 if (!bio) {
7797 free_io_failure(inode, failrec);
7798 return -EIO;
7799 }
7800
7801 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7802 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7803 read_mode, failrec->this_mirror, failrec->in_validation);
7804
7805 ret = submit_dio_repair_bio(inode, bio, read_mode,
7806 failrec->this_mirror);
7807 if (ret) {
7808 free_io_failure(inode, failrec);
7809 bio_put(bio);
7810 }
7811
7812 return ret;
7813 }
7814
7815 struct btrfs_retry_complete {
7816 struct completion done;
7817 struct inode *inode;
7818 u64 start;
7819 int uptodate;
7820 };
7821
7822 static void btrfs_retry_endio_nocsum(struct bio *bio)
7823 {
7824 struct btrfs_retry_complete *done = bio->bi_private;
7825 struct bio_vec *bvec;
7826 int i;
7827
7828 if (bio->bi_error)
7829 goto end;
7830
7831 done->uptodate = 1;
7832 bio_for_each_segment_all(bvec, bio, i)
7833 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7834 end:
7835 complete(&done->done);
7836 bio_put(bio);
7837 }
7838
7839 static int __btrfs_correct_data_nocsum(struct inode *inode,
7840 struct btrfs_io_bio *io_bio)
7841 {
7842 struct bio_vec *bvec;
7843 struct btrfs_retry_complete done;
7844 u64 start;
7845 int i;
7846 int ret;
7847
7848 start = io_bio->logical;
7849 done.inode = inode;
7850
7851 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7852 try_again:
7853 done.uptodate = 0;
7854 done.start = start;
7855 init_completion(&done.done);
7856
7857 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7858 start + bvec->bv_len - 1,
7859 io_bio->mirror_num,
7860 btrfs_retry_endio_nocsum, &done);
7861 if (ret)
7862 return ret;
7863
7864 wait_for_completion(&done.done);
7865
7866 if (!done.uptodate) {
7867 /* We might have another mirror, so try again */
7868 goto try_again;
7869 }
7870
7871 start += bvec->bv_len;
7872 }
7873
7874 return 0;
7875 }
7876
7877 static void btrfs_retry_endio(struct bio *bio)
7878 {
7879 struct btrfs_retry_complete *done = bio->bi_private;
7880 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7881 struct bio_vec *bvec;
7882 int uptodate;
7883 int ret;
7884 int i;
7885
7886 if (bio->bi_error)
7887 goto end;
7888
7889 uptodate = 1;
7890 bio_for_each_segment_all(bvec, bio, i) {
7891 ret = __readpage_endio_check(done->inode, io_bio, i,
7892 bvec->bv_page, 0,
7893 done->start, bvec->bv_len);
7894 if (!ret)
7895 clean_io_failure(done->inode, done->start,
7896 bvec->bv_page, 0);
7897 else
7898 uptodate = 0;
7899 }
7900
7901 done->uptodate = uptodate;
7902 end:
7903 complete(&done->done);
7904 bio_put(bio);
7905 }
7906
7907 static int __btrfs_subio_endio_read(struct inode *inode,
7908 struct btrfs_io_bio *io_bio, int err)
7909 {
7910 struct bio_vec *bvec;
7911 struct btrfs_retry_complete done;
7912 u64 start;
7913 u64 offset = 0;
7914 int i;
7915 int ret;
7916
7917 err = 0;
7918 start = io_bio->logical;
7919 done.inode = inode;
7920
7921 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7922 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7923 0, start, bvec->bv_len);
7924 if (likely(!ret))
7925 goto next;
7926 try_again:
7927 done.uptodate = 0;
7928 done.start = start;
7929 init_completion(&done.done);
7930
7931 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7932 start + bvec->bv_len - 1,
7933 io_bio->mirror_num,
7934 btrfs_retry_endio, &done);
7935 if (ret) {
7936 err = ret;
7937 goto next;
7938 }
7939
7940 wait_for_completion(&done.done);
7941
7942 if (!done.uptodate) {
7943 /* We might have another mirror, so try again */
7944 goto try_again;
7945 }
7946 next:
7947 offset += bvec->bv_len;
7948 start += bvec->bv_len;
7949 }
7950
7951 return err;
7952 }
7953
7954 static int btrfs_subio_endio_read(struct inode *inode,
7955 struct btrfs_io_bio *io_bio, int err)
7956 {
7957 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7958
7959 if (skip_csum) {
7960 if (unlikely(err))
7961 return __btrfs_correct_data_nocsum(inode, io_bio);
7962 else
7963 return 0;
7964 } else {
7965 return __btrfs_subio_endio_read(inode, io_bio, err);
7966 }
7967 }
7968
7969 static void btrfs_endio_direct_read(struct bio *bio)
7970 {
7971 struct btrfs_dio_private *dip = bio->bi_private;
7972 struct inode *inode = dip->inode;
7973 struct bio *dio_bio;
7974 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7975 int err = bio->bi_error;
7976
7977 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7978 err = btrfs_subio_endio_read(inode, io_bio, err);
7979
7980 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7981 dip->logical_offset + dip->bytes - 1);
7982 dio_bio = dip->dio_bio;
7983
7984 kfree(dip);
7985
7986 dio_end_io(dio_bio, bio->bi_error);
7987
7988 if (io_bio->end_io)
7989 io_bio->end_io(io_bio, err);
7990 bio_put(bio);
7991 }
7992
7993 static void btrfs_endio_direct_write(struct bio *bio)
7994 {
7995 struct btrfs_dio_private *dip = bio->bi_private;
7996 struct inode *inode = dip->inode;
7997 struct btrfs_root *root = BTRFS_I(inode)->root;
7998 struct btrfs_ordered_extent *ordered = NULL;
7999 u64 ordered_offset = dip->logical_offset;
8000 u64 ordered_bytes = dip->bytes;
8001 struct bio *dio_bio;
8002 int ret;
8003
8004 again:
8005 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8006 &ordered_offset,
8007 ordered_bytes,
8008 !bio->bi_error);
8009 if (!ret)
8010 goto out_test;
8011
8012 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8013 finish_ordered_fn, NULL, NULL);
8014 btrfs_queue_work(root->fs_info->endio_write_workers,
8015 &ordered->work);
8016 out_test:
8017 /*
8018 * our bio might span multiple ordered extents. If we haven't
8019 * completed the accounting for the whole dio, go back and try again
8020 */
8021 if (ordered_offset < dip->logical_offset + dip->bytes) {
8022 ordered_bytes = dip->logical_offset + dip->bytes -
8023 ordered_offset;
8024 ordered = NULL;
8025 goto again;
8026 }
8027 dio_bio = dip->dio_bio;
8028
8029 kfree(dip);
8030
8031 dio_end_io(dio_bio, bio->bi_error);
8032 bio_put(bio);
8033 }
8034
8035 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8036 struct bio *bio, int mirror_num,
8037 unsigned long bio_flags, u64 offset)
8038 {
8039 int ret;
8040 struct btrfs_root *root = BTRFS_I(inode)->root;
8041 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8042 BUG_ON(ret); /* -ENOMEM */
8043 return 0;
8044 }
8045
8046 static void btrfs_end_dio_bio(struct bio *bio)
8047 {
8048 struct btrfs_dio_private *dip = bio->bi_private;
8049 int err = bio->bi_error;
8050
8051 if (err)
8052 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8053 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8054 btrfs_ino(dip->inode), bio->bi_rw,
8055 (unsigned long long)bio->bi_iter.bi_sector,
8056 bio->bi_iter.bi_size, err);
8057
8058 if (dip->subio_endio)
8059 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8060
8061 if (err) {
8062 dip->errors = 1;
8063
8064 /*
8065 * before atomic variable goto zero, we must make sure
8066 * dip->errors is perceived to be set.
8067 */
8068 smp_mb__before_atomic();
8069 }
8070
8071 /* if there are more bios still pending for this dio, just exit */
8072 if (!atomic_dec_and_test(&dip->pending_bios))
8073 goto out;
8074
8075 if (dip->errors) {
8076 bio_io_error(dip->orig_bio);
8077 } else {
8078 dip->dio_bio->bi_error = 0;
8079 bio_endio(dip->orig_bio);
8080 }
8081 out:
8082 bio_put(bio);
8083 }
8084
8085 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8086 u64 first_sector, gfp_t gfp_flags)
8087 {
8088 struct bio *bio;
8089 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8090 if (bio)
8091 bio_associate_current(bio);
8092 return bio;
8093 }
8094
8095 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8096 struct inode *inode,
8097 struct btrfs_dio_private *dip,
8098 struct bio *bio,
8099 u64 file_offset)
8100 {
8101 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8102 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8103 int ret;
8104
8105 /*
8106 * We load all the csum data we need when we submit
8107 * the first bio to reduce the csum tree search and
8108 * contention.
8109 */
8110 if (dip->logical_offset == file_offset) {
8111 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8112 file_offset);
8113 if (ret)
8114 return ret;
8115 }
8116
8117 if (bio == dip->orig_bio)
8118 return 0;
8119
8120 file_offset -= dip->logical_offset;
8121 file_offset >>= inode->i_sb->s_blocksize_bits;
8122 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8123
8124 return 0;
8125 }
8126
8127 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8128 int rw, u64 file_offset, int skip_sum,
8129 int async_submit)
8130 {
8131 struct btrfs_dio_private *dip = bio->bi_private;
8132 int write = rw & REQ_WRITE;
8133 struct btrfs_root *root = BTRFS_I(inode)->root;
8134 int ret;
8135
8136 if (async_submit)
8137 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8138
8139 bio_get(bio);
8140
8141 if (!write) {
8142 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8143 BTRFS_WQ_ENDIO_DATA);
8144 if (ret)
8145 goto err;
8146 }
8147
8148 if (skip_sum)
8149 goto map;
8150
8151 if (write && async_submit) {
8152 ret = btrfs_wq_submit_bio(root->fs_info,
8153 inode, rw, bio, 0, 0,
8154 file_offset,
8155 __btrfs_submit_bio_start_direct_io,
8156 __btrfs_submit_bio_done);
8157 goto err;
8158 } else if (write) {
8159 /*
8160 * If we aren't doing async submit, calculate the csum of the
8161 * bio now.
8162 */
8163 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8164 if (ret)
8165 goto err;
8166 } else {
8167 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8168 file_offset);
8169 if (ret)
8170 goto err;
8171 }
8172 map:
8173 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8174 err:
8175 bio_put(bio);
8176 return ret;
8177 }
8178
8179 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8180 int skip_sum)
8181 {
8182 struct inode *inode = dip->inode;
8183 struct btrfs_root *root = BTRFS_I(inode)->root;
8184 struct bio *bio;
8185 struct bio *orig_bio = dip->orig_bio;
8186 struct bio_vec *bvec = orig_bio->bi_io_vec;
8187 u64 start_sector = orig_bio->bi_iter.bi_sector;
8188 u64 file_offset = dip->logical_offset;
8189 u64 submit_len = 0;
8190 u64 map_length;
8191 int nr_pages = 0;
8192 int ret;
8193 int async_submit = 0;
8194
8195 map_length = orig_bio->bi_iter.bi_size;
8196 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8197 &map_length, NULL, 0);
8198 if (ret)
8199 return -EIO;
8200
8201 if (map_length >= orig_bio->bi_iter.bi_size) {
8202 bio = orig_bio;
8203 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8204 goto submit;
8205 }
8206
8207 /* async crcs make it difficult to collect full stripe writes. */
8208 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8209 async_submit = 0;
8210 else
8211 async_submit = 1;
8212
8213 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8214 if (!bio)
8215 return -ENOMEM;
8216
8217 bio->bi_private = dip;
8218 bio->bi_end_io = btrfs_end_dio_bio;
8219 btrfs_io_bio(bio)->logical = file_offset;
8220 atomic_inc(&dip->pending_bios);
8221
8222 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8223 if (map_length < submit_len + bvec->bv_len ||
8224 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
8225 bvec->bv_offset) < bvec->bv_len) {
8226 /*
8227 * inc the count before we submit the bio so
8228 * we know the end IO handler won't happen before
8229 * we inc the count. Otherwise, the dip might get freed
8230 * before we're done setting it up
8231 */
8232 atomic_inc(&dip->pending_bios);
8233 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8234 file_offset, skip_sum,
8235 async_submit);
8236 if (ret) {
8237 bio_put(bio);
8238 atomic_dec(&dip->pending_bios);
8239 goto out_err;
8240 }
8241
8242 start_sector += submit_len >> 9;
8243 file_offset += submit_len;
8244
8245 submit_len = 0;
8246 nr_pages = 0;
8247
8248 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8249 start_sector, GFP_NOFS);
8250 if (!bio)
8251 goto out_err;
8252 bio->bi_private = dip;
8253 bio->bi_end_io = btrfs_end_dio_bio;
8254 btrfs_io_bio(bio)->logical = file_offset;
8255
8256 map_length = orig_bio->bi_iter.bi_size;
8257 ret = btrfs_map_block(root->fs_info, rw,
8258 start_sector << 9,
8259 &map_length, NULL, 0);
8260 if (ret) {
8261 bio_put(bio);
8262 goto out_err;
8263 }
8264 } else {
8265 submit_len += bvec->bv_len;
8266 nr_pages++;
8267 bvec++;
8268 }
8269 }
8270
8271 submit:
8272 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8273 async_submit);
8274 if (!ret)
8275 return 0;
8276
8277 bio_put(bio);
8278 out_err:
8279 dip->errors = 1;
8280 /*
8281 * before atomic variable goto zero, we must
8282 * make sure dip->errors is perceived to be set.
8283 */
8284 smp_mb__before_atomic();
8285 if (atomic_dec_and_test(&dip->pending_bios))
8286 bio_io_error(dip->orig_bio);
8287
8288 /* bio_end_io() will handle error, so we needn't return it */
8289 return 0;
8290 }
8291
8292 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8293 struct inode *inode, loff_t file_offset)
8294 {
8295 struct btrfs_dio_private *dip = NULL;
8296 struct bio *io_bio = NULL;
8297 struct btrfs_io_bio *btrfs_bio;
8298 int skip_sum;
8299 int write = rw & REQ_WRITE;
8300 int ret = 0;
8301
8302 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8303
8304 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8305 if (!io_bio) {
8306 ret = -ENOMEM;
8307 goto free_ordered;
8308 }
8309
8310 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8311 if (!dip) {
8312 ret = -ENOMEM;
8313 goto free_ordered;
8314 }
8315
8316 dip->private = dio_bio->bi_private;
8317 dip->inode = inode;
8318 dip->logical_offset = file_offset;
8319 dip->bytes = dio_bio->bi_iter.bi_size;
8320 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8321 io_bio->bi_private = dip;
8322 dip->orig_bio = io_bio;
8323 dip->dio_bio = dio_bio;
8324 atomic_set(&dip->pending_bios, 0);
8325 btrfs_bio = btrfs_io_bio(io_bio);
8326 btrfs_bio->logical = file_offset;
8327
8328 if (write) {
8329 io_bio->bi_end_io = btrfs_endio_direct_write;
8330 } else {
8331 io_bio->bi_end_io = btrfs_endio_direct_read;
8332 dip->subio_endio = btrfs_subio_endio_read;
8333 }
8334
8335 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8336 if (!ret)
8337 return;
8338
8339 if (btrfs_bio->end_io)
8340 btrfs_bio->end_io(btrfs_bio, ret);
8341
8342 free_ordered:
8343 /*
8344 * If we arrived here it means either we failed to submit the dip
8345 * or we either failed to clone the dio_bio or failed to allocate the
8346 * dip. If we cloned the dio_bio and allocated the dip, we can just
8347 * call bio_endio against our io_bio so that we get proper resource
8348 * cleanup if we fail to submit the dip, otherwise, we must do the
8349 * same as btrfs_endio_direct_[write|read] because we can't call these
8350 * callbacks - they require an allocated dip and a clone of dio_bio.
8351 */
8352 if (io_bio && dip) {
8353 io_bio->bi_error = -EIO;
8354 bio_endio(io_bio);
8355 /*
8356 * The end io callbacks free our dip, do the final put on io_bio
8357 * and all the cleanup and final put for dio_bio (through
8358 * dio_end_io()).
8359 */
8360 dip = NULL;
8361 io_bio = NULL;
8362 } else {
8363 if (write) {
8364 struct btrfs_ordered_extent *ordered;
8365
8366 ordered = btrfs_lookup_ordered_extent(inode,
8367 file_offset);
8368 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
8369 /*
8370 * Decrements our ref on the ordered extent and removes
8371 * the ordered extent from the inode's ordered tree,
8372 * doing all the proper resource cleanup such as for the
8373 * reserved space and waking up any waiters for this
8374 * ordered extent (through btrfs_remove_ordered_extent).
8375 */
8376 btrfs_finish_ordered_io(ordered);
8377 } else {
8378 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8379 file_offset + dio_bio->bi_iter.bi_size - 1);
8380 }
8381 dio_bio->bi_error = -EIO;
8382 /*
8383 * Releases and cleans up our dio_bio, no need to bio_put()
8384 * nor bio_endio()/bio_io_error() against dio_bio.
8385 */
8386 dio_end_io(dio_bio, ret);
8387 }
8388 if (io_bio)
8389 bio_put(io_bio);
8390 kfree(dip);
8391 }
8392
8393 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8394 const struct iov_iter *iter, loff_t offset)
8395 {
8396 int seg;
8397 int i;
8398 unsigned blocksize_mask = root->sectorsize - 1;
8399 ssize_t retval = -EINVAL;
8400
8401 if (offset & blocksize_mask)
8402 goto out;
8403
8404 if (iov_iter_alignment(iter) & blocksize_mask)
8405 goto out;
8406
8407 /* If this is a write we don't need to check anymore */
8408 if (iov_iter_rw(iter) == WRITE)
8409 return 0;
8410 /*
8411 * Check to make sure we don't have duplicate iov_base's in this
8412 * iovec, if so return EINVAL, otherwise we'll get csum errors
8413 * when reading back.
8414 */
8415 for (seg = 0; seg < iter->nr_segs; seg++) {
8416 for (i = seg + 1; i < iter->nr_segs; i++) {
8417 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8418 goto out;
8419 }
8420 }
8421 retval = 0;
8422 out:
8423 return retval;
8424 }
8425
8426 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8427 loff_t offset)
8428 {
8429 struct file *file = iocb->ki_filp;
8430 struct inode *inode = file->f_mapping->host;
8431 struct btrfs_root *root = BTRFS_I(inode)->root;
8432 struct btrfs_dio_data dio_data = { 0 };
8433 size_t count = 0;
8434 int flags = 0;
8435 bool wakeup = true;
8436 bool relock = false;
8437 ssize_t ret;
8438
8439 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8440 return 0;
8441
8442 inode_dio_begin(inode);
8443 smp_mb__after_atomic();
8444
8445 /*
8446 * The generic stuff only does filemap_write_and_wait_range, which
8447 * isn't enough if we've written compressed pages to this area, so
8448 * we need to flush the dirty pages again to make absolutely sure
8449 * that any outstanding dirty pages are on disk.
8450 */
8451 count = iov_iter_count(iter);
8452 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8453 &BTRFS_I(inode)->runtime_flags))
8454 filemap_fdatawrite_range(inode->i_mapping, offset,
8455 offset + count - 1);
8456
8457 if (iov_iter_rw(iter) == WRITE) {
8458 /*
8459 * If the write DIO is beyond the EOF, we need update
8460 * the isize, but it is protected by i_mutex. So we can
8461 * not unlock the i_mutex at this case.
8462 */
8463 if (offset + count <= inode->i_size) {
8464 mutex_unlock(&inode->i_mutex);
8465 relock = true;
8466 }
8467 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8468 if (ret)
8469 goto out;
8470 dio_data.outstanding_extents = div64_u64(count +
8471 BTRFS_MAX_EXTENT_SIZE - 1,
8472 BTRFS_MAX_EXTENT_SIZE);
8473
8474 /*
8475 * We need to know how many extents we reserved so that we can
8476 * do the accounting properly if we go over the number we
8477 * originally calculated. Abuse current->journal_info for this.
8478 */
8479 dio_data.reserve = round_up(count, root->sectorsize);
8480 current->journal_info = &dio_data;
8481 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8482 &BTRFS_I(inode)->runtime_flags)) {
8483 inode_dio_end(inode);
8484 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8485 wakeup = false;
8486 }
8487
8488 ret = __blockdev_direct_IO(iocb, inode,
8489 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8490 iter, offset, btrfs_get_blocks_direct, NULL,
8491 btrfs_submit_direct, flags);
8492 if (iov_iter_rw(iter) == WRITE) {
8493 current->journal_info = NULL;
8494 if (ret < 0 && ret != -EIOCBQUEUED) {
8495 if (dio_data.reserve)
8496 btrfs_delalloc_release_space(inode, offset,
8497 dio_data.reserve);
8498 } else if (ret >= 0 && (size_t)ret < count)
8499 btrfs_delalloc_release_space(inode, offset,
8500 count - (size_t)ret);
8501 }
8502 out:
8503 if (wakeup)
8504 inode_dio_end(inode);
8505 if (relock)
8506 mutex_lock(&inode->i_mutex);
8507
8508 return ret;
8509 }
8510
8511 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8512
8513 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8514 __u64 start, __u64 len)
8515 {
8516 int ret;
8517
8518 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8519 if (ret)
8520 return ret;
8521
8522 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8523 }
8524
8525 int btrfs_readpage(struct file *file, struct page *page)
8526 {
8527 struct extent_io_tree *tree;
8528 tree = &BTRFS_I(page->mapping->host)->io_tree;
8529 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8530 }
8531
8532 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8533 {
8534 struct extent_io_tree *tree;
8535
8536
8537 if (current->flags & PF_MEMALLOC) {
8538 redirty_page_for_writepage(wbc, page);
8539 unlock_page(page);
8540 return 0;
8541 }
8542 tree = &BTRFS_I(page->mapping->host)->io_tree;
8543 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8544 }
8545
8546 static int btrfs_writepages(struct address_space *mapping,
8547 struct writeback_control *wbc)
8548 {
8549 struct extent_io_tree *tree;
8550
8551 tree = &BTRFS_I(mapping->host)->io_tree;
8552 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8553 }
8554
8555 static int
8556 btrfs_readpages(struct file *file, struct address_space *mapping,
8557 struct list_head *pages, unsigned nr_pages)
8558 {
8559 struct extent_io_tree *tree;
8560 tree = &BTRFS_I(mapping->host)->io_tree;
8561 return extent_readpages(tree, mapping, pages, nr_pages,
8562 btrfs_get_extent);
8563 }
8564 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8565 {
8566 struct extent_io_tree *tree;
8567 struct extent_map_tree *map;
8568 int ret;
8569
8570 tree = &BTRFS_I(page->mapping->host)->io_tree;
8571 map = &BTRFS_I(page->mapping->host)->extent_tree;
8572 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8573 if (ret == 1) {
8574 ClearPagePrivate(page);
8575 set_page_private(page, 0);
8576 page_cache_release(page);
8577 }
8578 return ret;
8579 }
8580
8581 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8582 {
8583 if (PageWriteback(page) || PageDirty(page))
8584 return 0;
8585 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8586 }
8587
8588 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8589 unsigned int length)
8590 {
8591 struct inode *inode = page->mapping->host;
8592 struct extent_io_tree *tree;
8593 struct btrfs_ordered_extent *ordered;
8594 struct extent_state *cached_state = NULL;
8595 u64 page_start = page_offset(page);
8596 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8597 int inode_evicting = inode->i_state & I_FREEING;
8598
8599 /*
8600 * we have the page locked, so new writeback can't start,
8601 * and the dirty bit won't be cleared while we are here.
8602 *
8603 * Wait for IO on this page so that we can safely clear
8604 * the PagePrivate2 bit and do ordered accounting
8605 */
8606 wait_on_page_writeback(page);
8607
8608 tree = &BTRFS_I(inode)->io_tree;
8609 if (offset) {
8610 btrfs_releasepage(page, GFP_NOFS);
8611 return;
8612 }
8613
8614 if (!inode_evicting)
8615 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8616 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8617 if (ordered) {
8618 /*
8619 * IO on this page will never be started, so we need
8620 * to account for any ordered extents now
8621 */
8622 if (!inode_evicting)
8623 clear_extent_bit(tree, page_start, page_end,
8624 EXTENT_DIRTY | EXTENT_DELALLOC |
8625 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8626 EXTENT_DEFRAG, 1, 0, &cached_state,
8627 GFP_NOFS);
8628 /*
8629 * whoever cleared the private bit is responsible
8630 * for the finish_ordered_io
8631 */
8632 if (TestClearPagePrivate2(page)) {
8633 struct btrfs_ordered_inode_tree *tree;
8634 u64 new_len;
8635
8636 tree = &BTRFS_I(inode)->ordered_tree;
8637
8638 spin_lock_irq(&tree->lock);
8639 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8640 new_len = page_start - ordered->file_offset;
8641 if (new_len < ordered->truncated_len)
8642 ordered->truncated_len = new_len;
8643 spin_unlock_irq(&tree->lock);
8644
8645 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8646 page_start,
8647 PAGE_CACHE_SIZE, 1))
8648 btrfs_finish_ordered_io(ordered);
8649 }
8650 btrfs_put_ordered_extent(ordered);
8651 if (!inode_evicting) {
8652 cached_state = NULL;
8653 lock_extent_bits(tree, page_start, page_end, 0,
8654 &cached_state);
8655 }
8656 }
8657
8658 /*
8659 * Qgroup reserved space handler
8660 * Page here will be either
8661 * 1) Already written to disk
8662 * In this case, its reserved space is released from data rsv map
8663 * and will be freed by delayed_ref handler finally.
8664 * So even we call qgroup_free_data(), it won't decrease reserved
8665 * space.
8666 * 2) Not written to disk
8667 * This means the reserved space should be freed here.
8668 */
8669 btrfs_qgroup_free_data(inode, page_start, PAGE_CACHE_SIZE);
8670 if (!inode_evicting) {
8671 clear_extent_bit(tree, page_start, page_end,
8672 EXTENT_LOCKED | EXTENT_DIRTY |
8673 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8674 EXTENT_DEFRAG, 1, 1,
8675 &cached_state, GFP_NOFS);
8676
8677 __btrfs_releasepage(page, GFP_NOFS);
8678 }
8679
8680 ClearPageChecked(page);
8681 if (PagePrivate(page)) {
8682 ClearPagePrivate(page);
8683 set_page_private(page, 0);
8684 page_cache_release(page);
8685 }
8686 }
8687
8688 /*
8689 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8690 * called from a page fault handler when a page is first dirtied. Hence we must
8691 * be careful to check for EOF conditions here. We set the page up correctly
8692 * for a written page which means we get ENOSPC checking when writing into
8693 * holes and correct delalloc and unwritten extent mapping on filesystems that
8694 * support these features.
8695 *
8696 * We are not allowed to take the i_mutex here so we have to play games to
8697 * protect against truncate races as the page could now be beyond EOF. Because
8698 * vmtruncate() writes the inode size before removing pages, once we have the
8699 * page lock we can determine safely if the page is beyond EOF. If it is not
8700 * beyond EOF, then the page is guaranteed safe against truncation until we
8701 * unlock the page.
8702 */
8703 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8704 {
8705 struct page *page = vmf->page;
8706 struct inode *inode = file_inode(vma->vm_file);
8707 struct btrfs_root *root = BTRFS_I(inode)->root;
8708 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8709 struct btrfs_ordered_extent *ordered;
8710 struct extent_state *cached_state = NULL;
8711 char *kaddr;
8712 unsigned long zero_start;
8713 loff_t size;
8714 int ret;
8715 int reserved = 0;
8716 u64 page_start;
8717 u64 page_end;
8718
8719 sb_start_pagefault(inode->i_sb);
8720 page_start = page_offset(page);
8721 page_end = page_start + PAGE_CACHE_SIZE - 1;
8722
8723 ret = btrfs_delalloc_reserve_space(inode, page_start,
8724 PAGE_CACHE_SIZE);
8725 if (!ret) {
8726 ret = file_update_time(vma->vm_file);
8727 reserved = 1;
8728 }
8729 if (ret) {
8730 if (ret == -ENOMEM)
8731 ret = VM_FAULT_OOM;
8732 else /* -ENOSPC, -EIO, etc */
8733 ret = VM_FAULT_SIGBUS;
8734 if (reserved)
8735 goto out;
8736 goto out_noreserve;
8737 }
8738
8739 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8740 again:
8741 lock_page(page);
8742 size = i_size_read(inode);
8743
8744 if ((page->mapping != inode->i_mapping) ||
8745 (page_start >= size)) {
8746 /* page got truncated out from underneath us */
8747 goto out_unlock;
8748 }
8749 wait_on_page_writeback(page);
8750
8751 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
8752 set_page_extent_mapped(page);
8753
8754 /*
8755 * we can't set the delalloc bits if there are pending ordered
8756 * extents. Drop our locks and wait for them to finish
8757 */
8758 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8759 if (ordered) {
8760 unlock_extent_cached(io_tree, page_start, page_end,
8761 &cached_state, GFP_NOFS);
8762 unlock_page(page);
8763 btrfs_start_ordered_extent(inode, ordered, 1);
8764 btrfs_put_ordered_extent(ordered);
8765 goto again;
8766 }
8767
8768 /*
8769 * XXX - page_mkwrite gets called every time the page is dirtied, even
8770 * if it was already dirty, so for space accounting reasons we need to
8771 * clear any delalloc bits for the range we are fixing to save. There
8772 * is probably a better way to do this, but for now keep consistent with
8773 * prepare_pages in the normal write path.
8774 */
8775 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8776 EXTENT_DIRTY | EXTENT_DELALLOC |
8777 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8778 0, 0, &cached_state, GFP_NOFS);
8779
8780 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8781 &cached_state);
8782 if (ret) {
8783 unlock_extent_cached(io_tree, page_start, page_end,
8784 &cached_state, GFP_NOFS);
8785 ret = VM_FAULT_SIGBUS;
8786 goto out_unlock;
8787 }
8788 ret = 0;
8789
8790 /* page is wholly or partially inside EOF */
8791 if (page_start + PAGE_CACHE_SIZE > size)
8792 zero_start = size & ~PAGE_CACHE_MASK;
8793 else
8794 zero_start = PAGE_CACHE_SIZE;
8795
8796 if (zero_start != PAGE_CACHE_SIZE) {
8797 kaddr = kmap(page);
8798 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8799 flush_dcache_page(page);
8800 kunmap(page);
8801 }
8802 ClearPageChecked(page);
8803 set_page_dirty(page);
8804 SetPageUptodate(page);
8805
8806 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8807 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8808 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8809
8810 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8811
8812 out_unlock:
8813 if (!ret) {
8814 sb_end_pagefault(inode->i_sb);
8815 return VM_FAULT_LOCKED;
8816 }
8817 unlock_page(page);
8818 out:
8819 btrfs_delalloc_release_space(inode, page_start, PAGE_CACHE_SIZE);
8820 out_noreserve:
8821 sb_end_pagefault(inode->i_sb);
8822 return ret;
8823 }
8824
8825 static int btrfs_truncate(struct inode *inode)
8826 {
8827 struct btrfs_root *root = BTRFS_I(inode)->root;
8828 struct btrfs_block_rsv *rsv;
8829 int ret = 0;
8830 int err = 0;
8831 struct btrfs_trans_handle *trans;
8832 u64 mask = root->sectorsize - 1;
8833 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8834
8835 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8836 (u64)-1);
8837 if (ret)
8838 return ret;
8839
8840 /*
8841 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8842 * 3 things going on here
8843 *
8844 * 1) We need to reserve space for our orphan item and the space to
8845 * delete our orphan item. Lord knows we don't want to have a dangling
8846 * orphan item because we didn't reserve space to remove it.
8847 *
8848 * 2) We need to reserve space to update our inode.
8849 *
8850 * 3) We need to have something to cache all the space that is going to
8851 * be free'd up by the truncate operation, but also have some slack
8852 * space reserved in case it uses space during the truncate (thank you
8853 * very much snapshotting).
8854 *
8855 * And we need these to all be seperate. The fact is we can use alot of
8856 * space doing the truncate, and we have no earthly idea how much space
8857 * we will use, so we need the truncate reservation to be seperate so it
8858 * doesn't end up using space reserved for updating the inode or
8859 * removing the orphan item. We also need to be able to stop the
8860 * transaction and start a new one, which means we need to be able to
8861 * update the inode several times, and we have no idea of knowing how
8862 * many times that will be, so we can't just reserve 1 item for the
8863 * entirety of the opration, so that has to be done seperately as well.
8864 * Then there is the orphan item, which does indeed need to be held on
8865 * to for the whole operation, and we need nobody to touch this reserved
8866 * space except the orphan code.
8867 *
8868 * So that leaves us with
8869 *
8870 * 1) root->orphan_block_rsv - for the orphan deletion.
8871 * 2) rsv - for the truncate reservation, which we will steal from the
8872 * transaction reservation.
8873 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8874 * updating the inode.
8875 */
8876 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8877 if (!rsv)
8878 return -ENOMEM;
8879 rsv->size = min_size;
8880 rsv->failfast = 1;
8881
8882 /*
8883 * 1 for the truncate slack space
8884 * 1 for updating the inode.
8885 */
8886 trans = btrfs_start_transaction(root, 2);
8887 if (IS_ERR(trans)) {
8888 err = PTR_ERR(trans);
8889 goto out;
8890 }
8891
8892 /* Migrate the slack space for the truncate to our reserve */
8893 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8894 min_size);
8895 BUG_ON(ret);
8896
8897 /*
8898 * So if we truncate and then write and fsync we normally would just
8899 * write the extents that changed, which is a problem if we need to
8900 * first truncate that entire inode. So set this flag so we write out
8901 * all of the extents in the inode to the sync log so we're completely
8902 * safe.
8903 */
8904 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8905 trans->block_rsv = rsv;
8906
8907 while (1) {
8908 ret = btrfs_truncate_inode_items(trans, root, inode,
8909 inode->i_size,
8910 BTRFS_EXTENT_DATA_KEY);
8911 if (ret != -ENOSPC && ret != -EAGAIN) {
8912 err = ret;
8913 break;
8914 }
8915
8916 trans->block_rsv = &root->fs_info->trans_block_rsv;
8917 ret = btrfs_update_inode(trans, root, inode);
8918 if (ret) {
8919 err = ret;
8920 break;
8921 }
8922
8923 btrfs_end_transaction(trans, root);
8924 btrfs_btree_balance_dirty(root);
8925
8926 trans = btrfs_start_transaction(root, 2);
8927 if (IS_ERR(trans)) {
8928 ret = err = PTR_ERR(trans);
8929 trans = NULL;
8930 break;
8931 }
8932
8933 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8934 rsv, min_size);
8935 BUG_ON(ret); /* shouldn't happen */
8936 trans->block_rsv = rsv;
8937 }
8938
8939 if (ret == 0 && inode->i_nlink > 0) {
8940 trans->block_rsv = root->orphan_block_rsv;
8941 ret = btrfs_orphan_del(trans, inode);
8942 if (ret)
8943 err = ret;
8944 }
8945
8946 if (trans) {
8947 trans->block_rsv = &root->fs_info->trans_block_rsv;
8948 ret = btrfs_update_inode(trans, root, inode);
8949 if (ret && !err)
8950 err = ret;
8951
8952 ret = btrfs_end_transaction(trans, root);
8953 btrfs_btree_balance_dirty(root);
8954 }
8955
8956 out:
8957 btrfs_free_block_rsv(root, rsv);
8958
8959 if (ret && !err)
8960 err = ret;
8961
8962 return err;
8963 }
8964
8965 /*
8966 * create a new subvolume directory/inode (helper for the ioctl).
8967 */
8968 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8969 struct btrfs_root *new_root,
8970 struct btrfs_root *parent_root,
8971 u64 new_dirid)
8972 {
8973 struct inode *inode;
8974 int err;
8975 u64 index = 0;
8976
8977 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8978 new_dirid, new_dirid,
8979 S_IFDIR | (~current_umask() & S_IRWXUGO),
8980 &index);
8981 if (IS_ERR(inode))
8982 return PTR_ERR(inode);
8983 inode->i_op = &btrfs_dir_inode_operations;
8984 inode->i_fop = &btrfs_dir_file_operations;
8985
8986 set_nlink(inode, 1);
8987 btrfs_i_size_write(inode, 0);
8988 unlock_new_inode(inode);
8989
8990 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8991 if (err)
8992 btrfs_err(new_root->fs_info,
8993 "error inheriting subvolume %llu properties: %d",
8994 new_root->root_key.objectid, err);
8995
8996 err = btrfs_update_inode(trans, new_root, inode);
8997
8998 iput(inode);
8999 return err;
9000 }
9001
9002 struct inode *btrfs_alloc_inode(struct super_block *sb)
9003 {
9004 struct btrfs_inode *ei;
9005 struct inode *inode;
9006
9007 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9008 if (!ei)
9009 return NULL;
9010
9011 ei->root = NULL;
9012 ei->generation = 0;
9013 ei->last_trans = 0;
9014 ei->last_sub_trans = 0;
9015 ei->logged_trans = 0;
9016 ei->delalloc_bytes = 0;
9017 ei->defrag_bytes = 0;
9018 ei->disk_i_size = 0;
9019 ei->flags = 0;
9020 ei->csum_bytes = 0;
9021 ei->index_cnt = (u64)-1;
9022 ei->dir_index = 0;
9023 ei->last_unlink_trans = 0;
9024 ei->last_log_commit = 0;
9025
9026 spin_lock_init(&ei->lock);
9027 ei->outstanding_extents = 0;
9028 ei->reserved_extents = 0;
9029
9030 ei->runtime_flags = 0;
9031 ei->force_compress = BTRFS_COMPRESS_NONE;
9032
9033 ei->delayed_node = NULL;
9034
9035 ei->i_otime.tv_sec = 0;
9036 ei->i_otime.tv_nsec = 0;
9037
9038 inode = &ei->vfs_inode;
9039 extent_map_tree_init(&ei->extent_tree);
9040 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9041 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9042 ei->io_tree.track_uptodate = 1;
9043 ei->io_failure_tree.track_uptodate = 1;
9044 atomic_set(&ei->sync_writers, 0);
9045 mutex_init(&ei->log_mutex);
9046 mutex_init(&ei->delalloc_mutex);
9047 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9048 INIT_LIST_HEAD(&ei->delalloc_inodes);
9049 RB_CLEAR_NODE(&ei->rb_node);
9050
9051 return inode;
9052 }
9053
9054 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9055 void btrfs_test_destroy_inode(struct inode *inode)
9056 {
9057 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9058 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9059 }
9060 #endif
9061
9062 static void btrfs_i_callback(struct rcu_head *head)
9063 {
9064 struct inode *inode = container_of(head, struct inode, i_rcu);
9065 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9066 }
9067
9068 void btrfs_destroy_inode(struct inode *inode)
9069 {
9070 struct btrfs_ordered_extent *ordered;
9071 struct btrfs_root *root = BTRFS_I(inode)->root;
9072
9073 WARN_ON(!hlist_empty(&inode->i_dentry));
9074 WARN_ON(inode->i_data.nrpages);
9075 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9076 WARN_ON(BTRFS_I(inode)->reserved_extents);
9077 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9078 WARN_ON(BTRFS_I(inode)->csum_bytes);
9079 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9080
9081 /*
9082 * This can happen where we create an inode, but somebody else also
9083 * created the same inode and we need to destroy the one we already
9084 * created.
9085 */
9086 if (!root)
9087 goto free;
9088
9089 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9090 &BTRFS_I(inode)->runtime_flags)) {
9091 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9092 btrfs_ino(inode));
9093 atomic_dec(&root->orphan_inodes);
9094 }
9095
9096 while (1) {
9097 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9098 if (!ordered)
9099 break;
9100 else {
9101 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9102 ordered->file_offset, ordered->len);
9103 btrfs_remove_ordered_extent(inode, ordered);
9104 btrfs_put_ordered_extent(ordered);
9105 btrfs_put_ordered_extent(ordered);
9106 }
9107 }
9108 btrfs_qgroup_check_reserved_leak(inode);
9109 inode_tree_del(inode);
9110 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9111 free:
9112 call_rcu(&inode->i_rcu, btrfs_i_callback);
9113 }
9114
9115 int btrfs_drop_inode(struct inode *inode)
9116 {
9117 struct btrfs_root *root = BTRFS_I(inode)->root;
9118
9119 if (root == NULL)
9120 return 1;
9121
9122 /* the snap/subvol tree is on deleting */
9123 if (btrfs_root_refs(&root->root_item) == 0)
9124 return 1;
9125 else
9126 return generic_drop_inode(inode);
9127 }
9128
9129 static void init_once(void *foo)
9130 {
9131 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9132
9133 inode_init_once(&ei->vfs_inode);
9134 }
9135
9136 void btrfs_destroy_cachep(void)
9137 {
9138 /*
9139 * Make sure all delayed rcu free inodes are flushed before we
9140 * destroy cache.
9141 */
9142 rcu_barrier();
9143 if (btrfs_inode_cachep)
9144 kmem_cache_destroy(btrfs_inode_cachep);
9145 if (btrfs_trans_handle_cachep)
9146 kmem_cache_destroy(btrfs_trans_handle_cachep);
9147 if (btrfs_transaction_cachep)
9148 kmem_cache_destroy(btrfs_transaction_cachep);
9149 if (btrfs_path_cachep)
9150 kmem_cache_destroy(btrfs_path_cachep);
9151 if (btrfs_free_space_cachep)
9152 kmem_cache_destroy(btrfs_free_space_cachep);
9153 if (btrfs_delalloc_work_cachep)
9154 kmem_cache_destroy(btrfs_delalloc_work_cachep);
9155 }
9156
9157 int btrfs_init_cachep(void)
9158 {
9159 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9160 sizeof(struct btrfs_inode), 0,
9161 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
9162 if (!btrfs_inode_cachep)
9163 goto fail;
9164
9165 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9166 sizeof(struct btrfs_trans_handle), 0,
9167 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9168 if (!btrfs_trans_handle_cachep)
9169 goto fail;
9170
9171 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9172 sizeof(struct btrfs_transaction), 0,
9173 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9174 if (!btrfs_transaction_cachep)
9175 goto fail;
9176
9177 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9178 sizeof(struct btrfs_path), 0,
9179 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9180 if (!btrfs_path_cachep)
9181 goto fail;
9182
9183 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9184 sizeof(struct btrfs_free_space), 0,
9185 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9186 if (!btrfs_free_space_cachep)
9187 goto fail;
9188
9189 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
9190 sizeof(struct btrfs_delalloc_work), 0,
9191 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
9192 NULL);
9193 if (!btrfs_delalloc_work_cachep)
9194 goto fail;
9195
9196 return 0;
9197 fail:
9198 btrfs_destroy_cachep();
9199 return -ENOMEM;
9200 }
9201
9202 static int btrfs_getattr(struct vfsmount *mnt,
9203 struct dentry *dentry, struct kstat *stat)
9204 {
9205 u64 delalloc_bytes;
9206 struct inode *inode = d_inode(dentry);
9207 u32 blocksize = inode->i_sb->s_blocksize;
9208
9209 generic_fillattr(inode, stat);
9210 stat->dev = BTRFS_I(inode)->root->anon_dev;
9211 stat->blksize = PAGE_CACHE_SIZE;
9212
9213 spin_lock(&BTRFS_I(inode)->lock);
9214 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9215 spin_unlock(&BTRFS_I(inode)->lock);
9216 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9217 ALIGN(delalloc_bytes, blocksize)) >> 9;
9218 return 0;
9219 }
9220
9221 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9222 struct inode *new_dir, struct dentry *new_dentry)
9223 {
9224 struct btrfs_trans_handle *trans;
9225 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9226 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9227 struct inode *new_inode = d_inode(new_dentry);
9228 struct inode *old_inode = d_inode(old_dentry);
9229 struct timespec ctime = CURRENT_TIME;
9230 u64 index = 0;
9231 u64 root_objectid;
9232 int ret;
9233 u64 old_ino = btrfs_ino(old_inode);
9234
9235 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9236 return -EPERM;
9237
9238 /* we only allow rename subvolume link between subvolumes */
9239 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9240 return -EXDEV;
9241
9242 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9243 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9244 return -ENOTEMPTY;
9245
9246 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9247 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9248 return -ENOTEMPTY;
9249
9250
9251 /* check for collisions, even if the name isn't there */
9252 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9253 new_dentry->d_name.name,
9254 new_dentry->d_name.len);
9255
9256 if (ret) {
9257 if (ret == -EEXIST) {
9258 /* we shouldn't get
9259 * eexist without a new_inode */
9260 if (WARN_ON(!new_inode)) {
9261 return ret;
9262 }
9263 } else {
9264 /* maybe -EOVERFLOW */
9265 return ret;
9266 }
9267 }
9268 ret = 0;
9269
9270 /*
9271 * we're using rename to replace one file with another. Start IO on it
9272 * now so we don't add too much work to the end of the transaction
9273 */
9274 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9275 filemap_flush(old_inode->i_mapping);
9276
9277 /* close the racy window with snapshot create/destroy ioctl */
9278 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9279 down_read(&root->fs_info->subvol_sem);
9280 /*
9281 * We want to reserve the absolute worst case amount of items. So if
9282 * both inodes are subvols and we need to unlink them then that would
9283 * require 4 item modifications, but if they are both normal inodes it
9284 * would require 5 item modifications, so we'll assume their normal
9285 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9286 * should cover the worst case number of items we'll modify.
9287 */
9288 trans = btrfs_start_transaction(root, 11);
9289 if (IS_ERR(trans)) {
9290 ret = PTR_ERR(trans);
9291 goto out_notrans;
9292 }
9293
9294 if (dest != root)
9295 btrfs_record_root_in_trans(trans, dest);
9296
9297 ret = btrfs_set_inode_index(new_dir, &index);
9298 if (ret)
9299 goto out_fail;
9300
9301 BTRFS_I(old_inode)->dir_index = 0ULL;
9302 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9303 /* force full log commit if subvolume involved. */
9304 btrfs_set_log_full_commit(root->fs_info, trans);
9305 } else {
9306 ret = btrfs_insert_inode_ref(trans, dest,
9307 new_dentry->d_name.name,
9308 new_dentry->d_name.len,
9309 old_ino,
9310 btrfs_ino(new_dir), index);
9311 if (ret)
9312 goto out_fail;
9313 /*
9314 * this is an ugly little race, but the rename is required
9315 * to make sure that if we crash, the inode is either at the
9316 * old name or the new one. pinning the log transaction lets
9317 * us make sure we don't allow a log commit to come in after
9318 * we unlink the name but before we add the new name back in.
9319 */
9320 btrfs_pin_log_trans(root);
9321 }
9322
9323 inode_inc_iversion(old_dir);
9324 inode_inc_iversion(new_dir);
9325 inode_inc_iversion(old_inode);
9326 old_dir->i_ctime = old_dir->i_mtime = ctime;
9327 new_dir->i_ctime = new_dir->i_mtime = ctime;
9328 old_inode->i_ctime = ctime;
9329
9330 if (old_dentry->d_parent != new_dentry->d_parent)
9331 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9332
9333 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9334 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9335 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9336 old_dentry->d_name.name,
9337 old_dentry->d_name.len);
9338 } else {
9339 ret = __btrfs_unlink_inode(trans, root, old_dir,
9340 d_inode(old_dentry),
9341 old_dentry->d_name.name,
9342 old_dentry->d_name.len);
9343 if (!ret)
9344 ret = btrfs_update_inode(trans, root, old_inode);
9345 }
9346 if (ret) {
9347 btrfs_abort_transaction(trans, root, ret);
9348 goto out_fail;
9349 }
9350
9351 if (new_inode) {
9352 inode_inc_iversion(new_inode);
9353 new_inode->i_ctime = CURRENT_TIME;
9354 if (unlikely(btrfs_ino(new_inode) ==
9355 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9356 root_objectid = BTRFS_I(new_inode)->location.objectid;
9357 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9358 root_objectid,
9359 new_dentry->d_name.name,
9360 new_dentry->d_name.len);
9361 BUG_ON(new_inode->i_nlink == 0);
9362 } else {
9363 ret = btrfs_unlink_inode(trans, dest, new_dir,
9364 d_inode(new_dentry),
9365 new_dentry->d_name.name,
9366 new_dentry->d_name.len);
9367 }
9368 if (!ret && new_inode->i_nlink == 0)
9369 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9370 if (ret) {
9371 btrfs_abort_transaction(trans, root, ret);
9372 goto out_fail;
9373 }
9374 }
9375
9376 ret = btrfs_add_link(trans, new_dir, old_inode,
9377 new_dentry->d_name.name,
9378 new_dentry->d_name.len, 0, index);
9379 if (ret) {
9380 btrfs_abort_transaction(trans, root, ret);
9381 goto out_fail;
9382 }
9383
9384 if (old_inode->i_nlink == 1)
9385 BTRFS_I(old_inode)->dir_index = index;
9386
9387 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9388 struct dentry *parent = new_dentry->d_parent;
9389 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9390 btrfs_end_log_trans(root);
9391 }
9392 out_fail:
9393 btrfs_end_transaction(trans, root);
9394 out_notrans:
9395 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9396 up_read(&root->fs_info->subvol_sem);
9397
9398 return ret;
9399 }
9400
9401 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9402 struct inode *new_dir, struct dentry *new_dentry,
9403 unsigned int flags)
9404 {
9405 if (flags & ~RENAME_NOREPLACE)
9406 return -EINVAL;
9407
9408 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9409 }
9410
9411 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9412 {
9413 struct btrfs_delalloc_work *delalloc_work;
9414 struct inode *inode;
9415
9416 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9417 work);
9418 inode = delalloc_work->inode;
9419 if (delalloc_work->wait) {
9420 btrfs_wait_ordered_range(inode, 0, (u64)-1);
9421 } else {
9422 filemap_flush(inode->i_mapping);
9423 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9424 &BTRFS_I(inode)->runtime_flags))
9425 filemap_flush(inode->i_mapping);
9426 }
9427
9428 if (delalloc_work->delay_iput)
9429 btrfs_add_delayed_iput(inode);
9430 else
9431 iput(inode);
9432 complete(&delalloc_work->completion);
9433 }
9434
9435 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9436 int wait, int delay_iput)
9437 {
9438 struct btrfs_delalloc_work *work;
9439
9440 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
9441 if (!work)
9442 return NULL;
9443
9444 init_completion(&work->completion);
9445 INIT_LIST_HEAD(&work->list);
9446 work->inode = inode;
9447 work->wait = wait;
9448 work->delay_iput = delay_iput;
9449 WARN_ON_ONCE(!inode);
9450 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9451 btrfs_run_delalloc_work, NULL, NULL);
9452
9453 return work;
9454 }
9455
9456 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9457 {
9458 wait_for_completion(&work->completion);
9459 kmem_cache_free(btrfs_delalloc_work_cachep, work);
9460 }
9461
9462 /*
9463 * some fairly slow code that needs optimization. This walks the list
9464 * of all the inodes with pending delalloc and forces them to disk.
9465 */
9466 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9467 int nr)
9468 {
9469 struct btrfs_inode *binode;
9470 struct inode *inode;
9471 struct btrfs_delalloc_work *work, *next;
9472 struct list_head works;
9473 struct list_head splice;
9474 int ret = 0;
9475
9476 INIT_LIST_HEAD(&works);
9477 INIT_LIST_HEAD(&splice);
9478
9479 mutex_lock(&root->delalloc_mutex);
9480 spin_lock(&root->delalloc_lock);
9481 list_splice_init(&root->delalloc_inodes, &splice);
9482 while (!list_empty(&splice)) {
9483 binode = list_entry(splice.next, struct btrfs_inode,
9484 delalloc_inodes);
9485
9486 list_move_tail(&binode->delalloc_inodes,
9487 &root->delalloc_inodes);
9488 inode = igrab(&binode->vfs_inode);
9489 if (!inode) {
9490 cond_resched_lock(&root->delalloc_lock);
9491 continue;
9492 }
9493 spin_unlock(&root->delalloc_lock);
9494
9495 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
9496 if (!work) {
9497 if (delay_iput)
9498 btrfs_add_delayed_iput(inode);
9499 else
9500 iput(inode);
9501 ret = -ENOMEM;
9502 goto out;
9503 }
9504 list_add_tail(&work->list, &works);
9505 btrfs_queue_work(root->fs_info->flush_workers,
9506 &work->work);
9507 ret++;
9508 if (nr != -1 && ret >= nr)
9509 goto out;
9510 cond_resched();
9511 spin_lock(&root->delalloc_lock);
9512 }
9513 spin_unlock(&root->delalloc_lock);
9514
9515 out:
9516 list_for_each_entry_safe(work, next, &works, list) {
9517 list_del_init(&work->list);
9518 btrfs_wait_and_free_delalloc_work(work);
9519 }
9520
9521 if (!list_empty_careful(&splice)) {
9522 spin_lock(&root->delalloc_lock);
9523 list_splice_tail(&splice, &root->delalloc_inodes);
9524 spin_unlock(&root->delalloc_lock);
9525 }
9526 mutex_unlock(&root->delalloc_mutex);
9527 return ret;
9528 }
9529
9530 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9531 {
9532 int ret;
9533
9534 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9535 return -EROFS;
9536
9537 ret = __start_delalloc_inodes(root, delay_iput, -1);
9538 if (ret > 0)
9539 ret = 0;
9540 /*
9541 * the filemap_flush will queue IO into the worker threads, but
9542 * we have to make sure the IO is actually started and that
9543 * ordered extents get created before we return
9544 */
9545 atomic_inc(&root->fs_info->async_submit_draining);
9546 while (atomic_read(&root->fs_info->nr_async_submits) ||
9547 atomic_read(&root->fs_info->async_delalloc_pages)) {
9548 wait_event(root->fs_info->async_submit_wait,
9549 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9550 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9551 }
9552 atomic_dec(&root->fs_info->async_submit_draining);
9553 return ret;
9554 }
9555
9556 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9557 int nr)
9558 {
9559 struct btrfs_root *root;
9560 struct list_head splice;
9561 int ret;
9562
9563 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9564 return -EROFS;
9565
9566 INIT_LIST_HEAD(&splice);
9567
9568 mutex_lock(&fs_info->delalloc_root_mutex);
9569 spin_lock(&fs_info->delalloc_root_lock);
9570 list_splice_init(&fs_info->delalloc_roots, &splice);
9571 while (!list_empty(&splice) && nr) {
9572 root = list_first_entry(&splice, struct btrfs_root,
9573 delalloc_root);
9574 root = btrfs_grab_fs_root(root);
9575 BUG_ON(!root);
9576 list_move_tail(&root->delalloc_root,
9577 &fs_info->delalloc_roots);
9578 spin_unlock(&fs_info->delalloc_root_lock);
9579
9580 ret = __start_delalloc_inodes(root, delay_iput, nr);
9581 btrfs_put_fs_root(root);
9582 if (ret < 0)
9583 goto out;
9584
9585 if (nr != -1) {
9586 nr -= ret;
9587 WARN_ON(nr < 0);
9588 }
9589 spin_lock(&fs_info->delalloc_root_lock);
9590 }
9591 spin_unlock(&fs_info->delalloc_root_lock);
9592
9593 ret = 0;
9594 atomic_inc(&fs_info->async_submit_draining);
9595 while (atomic_read(&fs_info->nr_async_submits) ||
9596 atomic_read(&fs_info->async_delalloc_pages)) {
9597 wait_event(fs_info->async_submit_wait,
9598 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9599 atomic_read(&fs_info->async_delalloc_pages) == 0));
9600 }
9601 atomic_dec(&fs_info->async_submit_draining);
9602 out:
9603 if (!list_empty_careful(&splice)) {
9604 spin_lock(&fs_info->delalloc_root_lock);
9605 list_splice_tail(&splice, &fs_info->delalloc_roots);
9606 spin_unlock(&fs_info->delalloc_root_lock);
9607 }
9608 mutex_unlock(&fs_info->delalloc_root_mutex);
9609 return ret;
9610 }
9611
9612 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9613 const char *symname)
9614 {
9615 struct btrfs_trans_handle *trans;
9616 struct btrfs_root *root = BTRFS_I(dir)->root;
9617 struct btrfs_path *path;
9618 struct btrfs_key key;
9619 struct inode *inode = NULL;
9620 int err;
9621 int drop_inode = 0;
9622 u64 objectid;
9623 u64 index = 0;
9624 int name_len;
9625 int datasize;
9626 unsigned long ptr;
9627 struct btrfs_file_extent_item *ei;
9628 struct extent_buffer *leaf;
9629
9630 name_len = strlen(symname);
9631 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9632 return -ENAMETOOLONG;
9633
9634 /*
9635 * 2 items for inode item and ref
9636 * 2 items for dir items
9637 * 1 item for xattr if selinux is on
9638 */
9639 trans = btrfs_start_transaction(root, 5);
9640 if (IS_ERR(trans))
9641 return PTR_ERR(trans);
9642
9643 err = btrfs_find_free_ino(root, &objectid);
9644 if (err)
9645 goto out_unlock;
9646
9647 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9648 dentry->d_name.len, btrfs_ino(dir), objectid,
9649 S_IFLNK|S_IRWXUGO, &index);
9650 if (IS_ERR(inode)) {
9651 err = PTR_ERR(inode);
9652 goto out_unlock;
9653 }
9654
9655 /*
9656 * If the active LSM wants to access the inode during
9657 * d_instantiate it needs these. Smack checks to see
9658 * if the filesystem supports xattrs by looking at the
9659 * ops vector.
9660 */
9661 inode->i_fop = &btrfs_file_operations;
9662 inode->i_op = &btrfs_file_inode_operations;
9663 inode->i_mapping->a_ops = &btrfs_aops;
9664 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9665
9666 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9667 if (err)
9668 goto out_unlock_inode;
9669
9670 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9671 if (err)
9672 goto out_unlock_inode;
9673
9674 path = btrfs_alloc_path();
9675 if (!path) {
9676 err = -ENOMEM;
9677 goto out_unlock_inode;
9678 }
9679 key.objectid = btrfs_ino(inode);
9680 key.offset = 0;
9681 key.type = BTRFS_EXTENT_DATA_KEY;
9682 datasize = btrfs_file_extent_calc_inline_size(name_len);
9683 err = btrfs_insert_empty_item(trans, root, path, &key,
9684 datasize);
9685 if (err) {
9686 btrfs_free_path(path);
9687 goto out_unlock_inode;
9688 }
9689 leaf = path->nodes[0];
9690 ei = btrfs_item_ptr(leaf, path->slots[0],
9691 struct btrfs_file_extent_item);
9692 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9693 btrfs_set_file_extent_type(leaf, ei,
9694 BTRFS_FILE_EXTENT_INLINE);
9695 btrfs_set_file_extent_encryption(leaf, ei, 0);
9696 btrfs_set_file_extent_compression(leaf, ei, 0);
9697 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9698 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9699
9700 ptr = btrfs_file_extent_inline_start(ei);
9701 write_extent_buffer(leaf, symname, ptr, name_len);
9702 btrfs_mark_buffer_dirty(leaf);
9703 btrfs_free_path(path);
9704
9705 inode->i_op = &btrfs_symlink_inode_operations;
9706 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9707 inode_set_bytes(inode, name_len);
9708 btrfs_i_size_write(inode, name_len);
9709 err = btrfs_update_inode(trans, root, inode);
9710 if (err) {
9711 drop_inode = 1;
9712 goto out_unlock_inode;
9713 }
9714
9715 unlock_new_inode(inode);
9716 d_instantiate(dentry, inode);
9717
9718 out_unlock:
9719 btrfs_end_transaction(trans, root);
9720 if (drop_inode) {
9721 inode_dec_link_count(inode);
9722 iput(inode);
9723 }
9724 btrfs_btree_balance_dirty(root);
9725 return err;
9726
9727 out_unlock_inode:
9728 drop_inode = 1;
9729 unlock_new_inode(inode);
9730 goto out_unlock;
9731 }
9732
9733 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9734 u64 start, u64 num_bytes, u64 min_size,
9735 loff_t actual_len, u64 *alloc_hint,
9736 struct btrfs_trans_handle *trans)
9737 {
9738 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9739 struct extent_map *em;
9740 struct btrfs_root *root = BTRFS_I(inode)->root;
9741 struct btrfs_key ins;
9742 u64 cur_offset = start;
9743 u64 i_size;
9744 u64 cur_bytes;
9745 u64 last_alloc = (u64)-1;
9746 int ret = 0;
9747 bool own_trans = true;
9748
9749 if (trans)
9750 own_trans = false;
9751 while (num_bytes > 0) {
9752 if (own_trans) {
9753 trans = btrfs_start_transaction(root, 3);
9754 if (IS_ERR(trans)) {
9755 ret = PTR_ERR(trans);
9756 break;
9757 }
9758 }
9759
9760 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9761 cur_bytes = max(cur_bytes, min_size);
9762 /*
9763 * If we are severely fragmented we could end up with really
9764 * small allocations, so if the allocator is returning small
9765 * chunks lets make its job easier by only searching for those
9766 * sized chunks.
9767 */
9768 cur_bytes = min(cur_bytes, last_alloc);
9769 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9770 *alloc_hint, &ins, 1, 0);
9771 if (ret) {
9772 if (own_trans)
9773 btrfs_end_transaction(trans, root);
9774 break;
9775 }
9776
9777 last_alloc = ins.offset;
9778 ret = insert_reserved_file_extent(trans, inode,
9779 cur_offset, ins.objectid,
9780 ins.offset, ins.offset,
9781 ins.offset, 0, 0, 0,
9782 BTRFS_FILE_EXTENT_PREALLOC);
9783 if (ret) {
9784 btrfs_free_reserved_extent(root, ins.objectid,
9785 ins.offset, 0);
9786 btrfs_abort_transaction(trans, root, ret);
9787 if (own_trans)
9788 btrfs_end_transaction(trans, root);
9789 break;
9790 }
9791
9792 btrfs_drop_extent_cache(inode, cur_offset,
9793 cur_offset + ins.offset -1, 0);
9794
9795 em = alloc_extent_map();
9796 if (!em) {
9797 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9798 &BTRFS_I(inode)->runtime_flags);
9799 goto next;
9800 }
9801
9802 em->start = cur_offset;
9803 em->orig_start = cur_offset;
9804 em->len = ins.offset;
9805 em->block_start = ins.objectid;
9806 em->block_len = ins.offset;
9807 em->orig_block_len = ins.offset;
9808 em->ram_bytes = ins.offset;
9809 em->bdev = root->fs_info->fs_devices->latest_bdev;
9810 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9811 em->generation = trans->transid;
9812
9813 while (1) {
9814 write_lock(&em_tree->lock);
9815 ret = add_extent_mapping(em_tree, em, 1);
9816 write_unlock(&em_tree->lock);
9817 if (ret != -EEXIST)
9818 break;
9819 btrfs_drop_extent_cache(inode, cur_offset,
9820 cur_offset + ins.offset - 1,
9821 0);
9822 }
9823 free_extent_map(em);
9824 next:
9825 num_bytes -= ins.offset;
9826 cur_offset += ins.offset;
9827 *alloc_hint = ins.objectid + ins.offset;
9828
9829 inode_inc_iversion(inode);
9830 inode->i_ctime = CURRENT_TIME;
9831 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9832 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9833 (actual_len > inode->i_size) &&
9834 (cur_offset > inode->i_size)) {
9835 if (cur_offset > actual_len)
9836 i_size = actual_len;
9837 else
9838 i_size = cur_offset;
9839 i_size_write(inode, i_size);
9840 btrfs_ordered_update_i_size(inode, i_size, NULL);
9841 }
9842
9843 ret = btrfs_update_inode(trans, root, inode);
9844
9845 if (ret) {
9846 btrfs_abort_transaction(trans, root, ret);
9847 if (own_trans)
9848 btrfs_end_transaction(trans, root);
9849 break;
9850 }
9851
9852 if (own_trans)
9853 btrfs_end_transaction(trans, root);
9854 }
9855 return ret;
9856 }
9857
9858 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9859 u64 start, u64 num_bytes, u64 min_size,
9860 loff_t actual_len, u64 *alloc_hint)
9861 {
9862 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9863 min_size, actual_len, alloc_hint,
9864 NULL);
9865 }
9866
9867 int btrfs_prealloc_file_range_trans(struct inode *inode,
9868 struct btrfs_trans_handle *trans, int mode,
9869 u64 start, u64 num_bytes, u64 min_size,
9870 loff_t actual_len, u64 *alloc_hint)
9871 {
9872 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9873 min_size, actual_len, alloc_hint, trans);
9874 }
9875
9876 static int btrfs_set_page_dirty(struct page *page)
9877 {
9878 return __set_page_dirty_nobuffers(page);
9879 }
9880
9881 static int btrfs_permission(struct inode *inode, int mask)
9882 {
9883 struct btrfs_root *root = BTRFS_I(inode)->root;
9884 umode_t mode = inode->i_mode;
9885
9886 if (mask & MAY_WRITE &&
9887 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9888 if (btrfs_root_readonly(root))
9889 return -EROFS;
9890 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9891 return -EACCES;
9892 }
9893 return generic_permission(inode, mask);
9894 }
9895
9896 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9897 {
9898 struct btrfs_trans_handle *trans;
9899 struct btrfs_root *root = BTRFS_I(dir)->root;
9900 struct inode *inode = NULL;
9901 u64 objectid;
9902 u64 index;
9903 int ret = 0;
9904
9905 /*
9906 * 5 units required for adding orphan entry
9907 */
9908 trans = btrfs_start_transaction(root, 5);
9909 if (IS_ERR(trans))
9910 return PTR_ERR(trans);
9911
9912 ret = btrfs_find_free_ino(root, &objectid);
9913 if (ret)
9914 goto out;
9915
9916 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9917 btrfs_ino(dir), objectid, mode, &index);
9918 if (IS_ERR(inode)) {
9919 ret = PTR_ERR(inode);
9920 inode = NULL;
9921 goto out;
9922 }
9923
9924 inode->i_fop = &btrfs_file_operations;
9925 inode->i_op = &btrfs_file_inode_operations;
9926
9927 inode->i_mapping->a_ops = &btrfs_aops;
9928 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9929
9930 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9931 if (ret)
9932 goto out_inode;
9933
9934 ret = btrfs_update_inode(trans, root, inode);
9935 if (ret)
9936 goto out_inode;
9937 ret = btrfs_orphan_add(trans, inode);
9938 if (ret)
9939 goto out_inode;
9940
9941 /*
9942 * We set number of links to 0 in btrfs_new_inode(), and here we set
9943 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9944 * through:
9945 *
9946 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9947 */
9948 set_nlink(inode, 1);
9949 unlock_new_inode(inode);
9950 d_tmpfile(dentry, inode);
9951 mark_inode_dirty(inode);
9952
9953 out:
9954 btrfs_end_transaction(trans, root);
9955 if (ret)
9956 iput(inode);
9957 btrfs_balance_delayed_items(root);
9958 btrfs_btree_balance_dirty(root);
9959 return ret;
9960
9961 out_inode:
9962 unlock_new_inode(inode);
9963 goto out;
9964
9965 }
9966
9967 /* Inspired by filemap_check_errors() */
9968 int btrfs_inode_check_errors(struct inode *inode)
9969 {
9970 int ret = 0;
9971
9972 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9973 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9974 ret = -ENOSPC;
9975 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9976 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9977 ret = -EIO;
9978
9979 return ret;
9980 }
9981
9982 static const struct inode_operations btrfs_dir_inode_operations = {
9983 .getattr = btrfs_getattr,
9984 .lookup = btrfs_lookup,
9985 .create = btrfs_create,
9986 .unlink = btrfs_unlink,
9987 .link = btrfs_link,
9988 .mkdir = btrfs_mkdir,
9989 .rmdir = btrfs_rmdir,
9990 .rename2 = btrfs_rename2,
9991 .symlink = btrfs_symlink,
9992 .setattr = btrfs_setattr,
9993 .mknod = btrfs_mknod,
9994 .setxattr = btrfs_setxattr,
9995 .getxattr = btrfs_getxattr,
9996 .listxattr = btrfs_listxattr,
9997 .removexattr = btrfs_removexattr,
9998 .permission = btrfs_permission,
9999 .get_acl = btrfs_get_acl,
10000 .set_acl = btrfs_set_acl,
10001 .update_time = btrfs_update_time,
10002 .tmpfile = btrfs_tmpfile,
10003 };
10004 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10005 .lookup = btrfs_lookup,
10006 .permission = btrfs_permission,
10007 .get_acl = btrfs_get_acl,
10008 .set_acl = btrfs_set_acl,
10009 .update_time = btrfs_update_time,
10010 };
10011
10012 static const struct file_operations btrfs_dir_file_operations = {
10013 .llseek = generic_file_llseek,
10014 .read = generic_read_dir,
10015 .iterate = btrfs_real_readdir,
10016 .unlocked_ioctl = btrfs_ioctl,
10017 #ifdef CONFIG_COMPAT
10018 .compat_ioctl = btrfs_ioctl,
10019 #endif
10020 .release = btrfs_release_file,
10021 .fsync = btrfs_sync_file,
10022 };
10023
10024 static struct extent_io_ops btrfs_extent_io_ops = {
10025 .fill_delalloc = run_delalloc_range,
10026 .submit_bio_hook = btrfs_submit_bio_hook,
10027 .merge_bio_hook = btrfs_merge_bio_hook,
10028 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10029 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10030 .writepage_start_hook = btrfs_writepage_start_hook,
10031 .set_bit_hook = btrfs_set_bit_hook,
10032 .clear_bit_hook = btrfs_clear_bit_hook,
10033 .merge_extent_hook = btrfs_merge_extent_hook,
10034 .split_extent_hook = btrfs_split_extent_hook,
10035 };
10036
10037 /*
10038 * btrfs doesn't support the bmap operation because swapfiles
10039 * use bmap to make a mapping of extents in the file. They assume
10040 * these extents won't change over the life of the file and they
10041 * use the bmap result to do IO directly to the drive.
10042 *
10043 * the btrfs bmap call would return logical addresses that aren't
10044 * suitable for IO and they also will change frequently as COW
10045 * operations happen. So, swapfile + btrfs == corruption.
10046 *
10047 * For now we're avoiding this by dropping bmap.
10048 */
10049 static const struct address_space_operations btrfs_aops = {
10050 .readpage = btrfs_readpage,
10051 .writepage = btrfs_writepage,
10052 .writepages = btrfs_writepages,
10053 .readpages = btrfs_readpages,
10054 .direct_IO = btrfs_direct_IO,
10055 .invalidatepage = btrfs_invalidatepage,
10056 .releasepage = btrfs_releasepage,
10057 .set_page_dirty = btrfs_set_page_dirty,
10058 .error_remove_page = generic_error_remove_page,
10059 };
10060
10061 static const struct address_space_operations btrfs_symlink_aops = {
10062 .readpage = btrfs_readpage,
10063 .writepage = btrfs_writepage,
10064 .invalidatepage = btrfs_invalidatepage,
10065 .releasepage = btrfs_releasepage,
10066 };
10067
10068 static const struct inode_operations btrfs_file_inode_operations = {
10069 .getattr = btrfs_getattr,
10070 .setattr = btrfs_setattr,
10071 .setxattr = btrfs_setxattr,
10072 .getxattr = btrfs_getxattr,
10073 .listxattr = btrfs_listxattr,
10074 .removexattr = btrfs_removexattr,
10075 .permission = btrfs_permission,
10076 .fiemap = btrfs_fiemap,
10077 .get_acl = btrfs_get_acl,
10078 .set_acl = btrfs_set_acl,
10079 .update_time = btrfs_update_time,
10080 };
10081 static const struct inode_operations btrfs_special_inode_operations = {
10082 .getattr = btrfs_getattr,
10083 .setattr = btrfs_setattr,
10084 .permission = btrfs_permission,
10085 .setxattr = btrfs_setxattr,
10086 .getxattr = btrfs_getxattr,
10087 .listxattr = btrfs_listxattr,
10088 .removexattr = btrfs_removexattr,
10089 .get_acl = btrfs_get_acl,
10090 .set_acl = btrfs_set_acl,
10091 .update_time = btrfs_update_time,
10092 };
10093 static const struct inode_operations btrfs_symlink_inode_operations = {
10094 .readlink = generic_readlink,
10095 .follow_link = page_follow_link_light,
10096 .put_link = page_put_link,
10097 .getattr = btrfs_getattr,
10098 .setattr = btrfs_setattr,
10099 .permission = btrfs_permission,
10100 .setxattr = btrfs_setxattr,
10101 .getxattr = btrfs_getxattr,
10102 .listxattr = btrfs_listxattr,
10103 .removexattr = btrfs_removexattr,
10104 .update_time = btrfs_update_time,
10105 };
10106
10107 const struct dentry_operations btrfs_dentry_operations = {
10108 .d_delete = btrfs_dentry_delete,
10109 .d_release = btrfs_dentry_release,
10110 };
Linux kernel - Deliverables
This page took 0.973546 seconds and 5 git commands to generate.