?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux...
[linux-2.6/btrfs-unstable.git] / fs / btrfs / inode.c
blob06e9a4152b1419c6e4afbd4fd3580615ffc4eadd
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/aio.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include <linux/slab.h>
41 #include <linux/ratelimit.h>
42 #include <linux/mount.h>
43 #include <linux/btrfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/posix_acl_xattr.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
63 struct btrfs_iget_args {
64 struct btrfs_key *location;
65 struct btrfs_root *root;
66 };
67
68 static const struct inode_operations btrfs_dir_inode_operations;
69 static const struct inode_operations btrfs_symlink_inode_operations;
70 static const struct inode_operations btrfs_dir_ro_inode_operations;
71 static const struct inode_operations btrfs_special_inode_operations;
72 static const struct inode_operations btrfs_file_inode_operations;
73 static const struct address_space_operations btrfs_aops;
74 static const struct address_space_operations btrfs_symlink_aops;
75 static const struct file_operations btrfs_dir_file_operations;
76 static struct extent_io_ops btrfs_extent_io_ops;
77
78 static struct kmem_cache *btrfs_inode_cachep;
79 static struct kmem_cache *btrfs_delalloc_work_cachep;
80 struct kmem_cache *btrfs_trans_handle_cachep;
81 struct kmem_cache *btrfs_transaction_cachep;
82 struct kmem_cache *btrfs_path_cachep;
83 struct kmem_cache *btrfs_free_space_cachep;
84
85 #define S_SHIFT 12
86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
87 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
88 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
89 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
90 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
91 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
92 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
93 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
94 };
95
96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
97 static int btrfs_truncate(struct inode *inode);
98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
99 static noinline int cow_file_range(struct inode *inode,
100 struct page *locked_page,
101 u64 start, u64 end, int *page_started,
102 unsigned long *nr_written, int unlock);
103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
104 u64 len, u64 orig_start,
105 u64 block_start, u64 block_len,
106 u64 orig_block_len, u64 ram_bytes,
107 int type);
108
109 static int btrfs_dirty_inode(struct inode *inode);
110
111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
112 struct inode *inode, struct inode *dir,
113 const struct qstr *qstr)
114 {
115 int err;
116
117 err = btrfs_init_acl(trans, inode, dir);
118 if (!err)
119 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
120 return err;
121 }
122
123 /*
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
127 */
128 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
129 struct btrfs_path *path, int extent_inserted,
130 struct btrfs_root *root, struct inode *inode,
131 u64 start, size_t size, size_t compressed_size,
132 int compress_type,
133 struct page **compressed_pages)
134 {
135 struct extent_buffer *leaf;
136 struct page *page = NULL;
137 char *kaddr;
138 unsigned long ptr;
139 struct btrfs_file_extent_item *ei;
140 int err = 0;
141 int ret;
142 size_t cur_size = size;
143 unsigned long offset;
144
145 if (compressed_size && compressed_pages)
146 cur_size = compressed_size;
147
148 inode_add_bytes(inode, size);
149
150 if (!extent_inserted) {
151 struct btrfs_key key;
152 size_t datasize;
153
154 key.objectid = btrfs_ino(inode);
155 key.offset = start;
156 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
157
158 datasize = btrfs_file_extent_calc_inline_size(cur_size);
159 path->leave_spinning = 1;
160 ret = btrfs_insert_empty_item(trans, root, path, &key,
161 datasize);
162 if (ret) {
163 err = ret;
164 goto fail;
165 }
166 }
167 leaf = path->nodes[0];
168 ei = btrfs_item_ptr(leaf, path->slots[0],
169 struct btrfs_file_extent_item);
170 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
171 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
172 btrfs_set_file_extent_encryption(leaf, ei, 0);
173 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
174 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
175 ptr = btrfs_file_extent_inline_start(ei);
176
177 if (compress_type != BTRFS_COMPRESS_NONE) {
178 struct page *cpage;
179 int i = 0;
180 while (compressed_size > 0) {
181 cpage = compressed_pages[i];
182 cur_size = min_t(unsigned long, compressed_size,
183 PAGE_CACHE_SIZE);
184
185 kaddr = kmap_atomic(cpage);
186 write_extent_buffer(leaf, kaddr, ptr, cur_size);
187 kunmap_atomic(kaddr);
188
189 i++;
190 ptr += cur_size;
191 compressed_size -= cur_size;
192 }
193 btrfs_set_file_extent_compression(leaf, ei,
194 compress_type);
195 } else {
196 page = find_get_page(inode->i_mapping,
197 start >> PAGE_CACHE_SHIFT);
198 btrfs_set_file_extent_compression(leaf, ei, 0);
199 kaddr = kmap_atomic(page);
200 offset = start & (PAGE_CACHE_SIZE - 1);
201 write_extent_buffer(leaf, kaddr + offset, ptr, size);
202 kunmap_atomic(kaddr);
203 page_cache_release(page);
204 }
205 btrfs_mark_buffer_dirty(leaf);
206 btrfs_release_path(path);
207
208 /*
209 * we're an inline extent, so nobody can
210 * extend the file past i_size without locking
211 * a page we already have locked.
212 *
213 * We must do any isize and inode updates
214 * before we unlock the pages. Otherwise we
215 * could end up racing with unlink.
216 */
217 BTRFS_I(inode)->disk_i_size = inode->i_size;
218 ret = btrfs_update_inode(trans, root, inode);
219
220 return ret;
221 fail:
222 return err;
223 }
224
225
226 /*
227 * conditionally insert an inline extent into the file. This
228 * does the checks required to make sure the data is small enough
229 * to fit as an inline extent.
230 */
231 static noinline int cow_file_range_inline(struct btrfs_root *root,
232 struct inode *inode, u64 start,
233 u64 end, size_t compressed_size,
234 int compress_type,
235 struct page **compressed_pages)
236 {
237 struct btrfs_trans_handle *trans;
238 u64 isize = i_size_read(inode);
239 u64 actual_end = min(end + 1, isize);
240 u64 inline_len = actual_end - start;
241 u64 aligned_end = ALIGN(end, root->sectorsize);
242 u64 data_len = inline_len;
243 int ret;
244 struct btrfs_path *path;
245 int extent_inserted = 0;
246 u32 extent_item_size;
247
248 if (compressed_size)
249 data_len = compressed_size;
250
251 if (start > 0 ||
252 actual_end >= PAGE_CACHE_SIZE ||
253 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
254 (!compressed_size &&
255 (actual_end & (root->sectorsize - 1)) == 0) ||
256 end + 1 < isize ||
257 data_len > root->fs_info->max_inline) {
258 return 1;
259 }
260
261 path = btrfs_alloc_path();
262 if (!path)
263 return -ENOMEM;
264
265 trans = btrfs_join_transaction(root);
266 if (IS_ERR(trans)) {
267 btrfs_free_path(path);
268 return PTR_ERR(trans);
269 }
270 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
271
272 if (compressed_size && compressed_pages)
273 extent_item_size = btrfs_file_extent_calc_inline_size(
274 compressed_size);
275 else
276 extent_item_size = btrfs_file_extent_calc_inline_size(
277 inline_len);
278
279 ret = __btrfs_drop_extents(trans, root, inode, path,
280 start, aligned_end, NULL,
281 1, 1, extent_item_size, &extent_inserted);
282 if (ret) {
283 btrfs_abort_transaction(trans, root, ret);
284 goto out;
285 }
286
287 if (isize > actual_end)
288 inline_len = min_t(u64, isize, actual_end);
289 ret = insert_inline_extent(trans, path, extent_inserted,
290 root, inode, start,
291 inline_len, compressed_size,
292 compress_type, compressed_pages);
293 if (ret && ret != -ENOSPC) {
294 btrfs_abort_transaction(trans, root, ret);
295 goto out;
296 } else if (ret == -ENOSPC) {
297 ret = 1;
298 goto out;
299 }
300
301 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
302 btrfs_delalloc_release_metadata(inode, end + 1 - start);
303 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
304 out:
305 btrfs_free_path(path);
306 btrfs_end_transaction(trans, root);
307 return ret;
308 }
309
310 struct async_extent {
311 u64 start;
312 u64 ram_size;
313 u64 compressed_size;
314 struct page **pages;
315 unsigned long nr_pages;
316 int compress_type;
317 struct list_head list;
318 };
319
320 struct async_cow {
321 struct inode *inode;
322 struct btrfs_root *root;
323 struct page *locked_page;
324 u64 start;
325 u64 end;
326 struct list_head extents;
327 struct btrfs_work work;
328 };
329
330 static noinline int add_async_extent(struct async_cow *cow,
331 u64 start, u64 ram_size,
332 u64 compressed_size,
333 struct page **pages,
334 unsigned long nr_pages,
335 int compress_type)
336 {
337 struct async_extent *async_extent;
338
339 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
340 BUG_ON(!async_extent); /* -ENOMEM */
341 async_extent->start = start;
342 async_extent->ram_size = ram_size;
343 async_extent->compressed_size = compressed_size;
344 async_extent->pages = pages;
345 async_extent->nr_pages = nr_pages;
346 async_extent->compress_type = compress_type;
347 list_add_tail(&async_extent->list, &cow->extents);
348 return 0;
349 }
350
351 /*
352 * we create compressed extents in two phases. The first
353 * phase compresses a range of pages that have already been
354 * locked (both pages and state bits are locked).
355 *
356 * This is done inside an ordered work queue, and the compression
357 * is spread across many cpus. The actual IO submission is step
358 * two, and the ordered work queue takes care of making sure that
359 * happens in the same order things were put onto the queue by
360 * writepages and friends.
361 *
362 * If this code finds it can't get good compression, it puts an
363 * entry onto the work queue to write the uncompressed bytes. This
364 * makes sure that both compressed inodes and uncompressed inodes
365 * are written in the same order that the flusher thread sent them
366 * down.
367 */
368 static noinline int compress_file_range(struct inode *inode,
369 struct page *locked_page,
370 u64 start, u64 end,
371 struct async_cow *async_cow,
372 int *num_added)
373 {
374 struct btrfs_root *root = BTRFS_I(inode)->root;
375 u64 num_bytes;
376 u64 blocksize = root->sectorsize;
377 u64 actual_end;
378 u64 isize = i_size_read(inode);
379 int ret = 0;
380 struct page **pages = NULL;
381 unsigned long nr_pages;
382 unsigned long nr_pages_ret = 0;
383 unsigned long total_compressed = 0;
384 unsigned long total_in = 0;
385 unsigned long max_compressed = 128 * 1024;
386 unsigned long max_uncompressed = 128 * 1024;
387 int i;
388 int will_compress;
389 int compress_type = root->fs_info->compress_type;
390 int redirty = 0;
391
392 /* if this is a small write inside eof, kick off a defrag */
393 if ((end - start + 1) < 16 * 1024 &&
394 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
395 btrfs_add_inode_defrag(NULL, inode);
396
397 actual_end = min_t(u64, isize, end + 1);
398 again:
399 will_compress = 0;
400 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
401 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
402
403 /*
404 * we don't want to send crud past the end of i_size through
405 * compression, that's just a waste of CPU time. So, if the
406 * end of the file is before the start of our current
407 * requested range of bytes, we bail out to the uncompressed
408 * cleanup code that can deal with all of this.
409 *
410 * It isn't really the fastest way to fix things, but this is a
411 * very uncommon corner.
412 */
413 if (actual_end <= start)
414 goto cleanup_and_bail_uncompressed;
415
416 total_compressed = actual_end - start;
417
418 /* we want to make sure that amount of ram required to uncompress
419 * an extent is reasonable, so we limit the total size in ram
420 * of a compressed extent to 128k. This is a crucial number
421 * because it also controls how easily we can spread reads across
422 * cpus for decompression.
423 *
424 * We also want to make sure the amount of IO required to do
425 * a random read is reasonably small, so we limit the size of
426 * a compressed extent to 128k.
427 */
428 total_compressed = min(total_compressed, max_uncompressed);
429 num_bytes = ALIGN(end - start + 1, blocksize);
430 num_bytes = max(blocksize, num_bytes);
431 total_in = 0;
432 ret = 0;
433
434 /*
435 * we do compression for mount -o compress and when the
436 * inode has not been flagged as nocompress. This flag can
437 * change at any time if we discover bad compression ratios.
438 */
439 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
440 (btrfs_test_opt(root, COMPRESS) ||
441 (BTRFS_I(inode)->force_compress) ||
442 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
443 WARN_ON(pages);
444 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
445 if (!pages) {
446 /* just bail out to the uncompressed code */
447 goto cont;
448 }
449
450 if (BTRFS_I(inode)->force_compress)
451 compress_type = BTRFS_I(inode)->force_compress;
452
453 /*
454 * we need to call clear_page_dirty_for_io on each
455 * page in the range. Otherwise applications with the file
456 * mmap'd can wander in and change the page contents while
457 * we are compressing them.
458 *
459 * If the compression fails for any reason, we set the pages
460 * dirty again later on.
461 */
462 extent_range_clear_dirty_for_io(inode, start, end);
463 redirty = 1;
464 ret = btrfs_compress_pages(compress_type,
465 inode->i_mapping, start,
466 total_compressed, pages,
467 nr_pages, &nr_pages_ret,
468 &total_in,
469 &total_compressed,
470 max_compressed);
471
472 if (!ret) {
473 unsigned long offset = total_compressed &
474 (PAGE_CACHE_SIZE - 1);
475 struct page *page = pages[nr_pages_ret - 1];
476 char *kaddr;
477
478 /* zero the tail end of the last page, we might be
479 * sending it down to disk
480 */
481 if (offset) {
482 kaddr = kmap_atomic(page);
483 memset(kaddr + offset, 0,
484 PAGE_CACHE_SIZE - offset);
485 kunmap_atomic(kaddr);
486 }
487 will_compress = 1;
488 }
489 }
490 cont:
491 if (start == 0) {
492 /* lets try to make an inline extent */
493 if (ret || total_in < (actual_end - start)) {
494 /* we didn't compress the entire range, try
495 * to make an uncompressed inline extent.
496 */
497 ret = cow_file_range_inline(root, inode, start, end,
498 0, 0, NULL);
499 } else {
500 /* try making a compressed inline extent */
501 ret = cow_file_range_inline(root, inode, start, end,
502 total_compressed,
503 compress_type, pages);
504 }
505 if (ret <= 0) {
506 unsigned long clear_flags = EXTENT_DELALLOC |
507 EXTENT_DEFRAG;
508 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
509
510 /*
511 * inline extent creation worked or returned error,
512 * we don't need to create any more async work items.
513 * Unlock and free up our temp pages.
514 */
515 extent_clear_unlock_delalloc(inode, start, end, NULL,
516 clear_flags, PAGE_UNLOCK |
517 PAGE_CLEAR_DIRTY |
518 PAGE_SET_WRITEBACK |
519 PAGE_END_WRITEBACK);
520 goto free_pages_out;
521 }
522 }
523
524 if (will_compress) {
525 /*
526 * we aren't doing an inline extent round the compressed size
527 * up to a block size boundary so the allocator does sane
528 * things
529 */
530 total_compressed = ALIGN(total_compressed, blocksize);
531
532 /*
533 * one last check to make sure the compression is really a
534 * win, compare the page count read with the blocks on disk
535 */
536 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
537 if (total_compressed >= total_in) {
538 will_compress = 0;
539 } else {
540 num_bytes = total_in;
541 }
542 }
543 if (!will_compress && pages) {
544 /*
545 * the compression code ran but failed to make things smaller,
546 * free any pages it allocated and our page pointer array
547 */
548 for (i = 0; i < nr_pages_ret; i++) {
549 WARN_ON(pages[i]->mapping);
550 page_cache_release(pages[i]);
551 }
552 kfree(pages);
553 pages = NULL;
554 total_compressed = 0;
555 nr_pages_ret = 0;
556
557 /* flag the file so we don't compress in the future */
558 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
559 !(BTRFS_I(inode)->force_compress)) {
560 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
561 }
562 }
563 if (will_compress) {
564 *num_added += 1;
565
566 /* the async work queues will take care of doing actual
567 * allocation on disk for these compressed pages,
568 * and will submit them to the elevator.
569 */
570 add_async_extent(async_cow, start, num_bytes,
571 total_compressed, pages, nr_pages_ret,
572 compress_type);
573
574 if (start + num_bytes < end) {
575 start += num_bytes;
576 pages = NULL;
577 cond_resched();
578 goto again;
579 }
580 } else {
581 cleanup_and_bail_uncompressed:
582 /*
583 * No compression, but we still need to write the pages in
584 * the file we've been given so far. redirty the locked
585 * page if it corresponds to our extent and set things up
586 * for the async work queue to run cow_file_range to do
587 * the normal delalloc dance
588 */
589 if (page_offset(locked_page) >= start &&
590 page_offset(locked_page) <= end) {
591 __set_page_dirty_nobuffers(locked_page);
592 /* unlocked later on in the async handlers */
593 }
594 if (redirty)
595 extent_range_redirty_for_io(inode, start, end);
596 add_async_extent(async_cow, start, end - start + 1,
597 0, NULL, 0, BTRFS_COMPRESS_NONE);
598 *num_added += 1;
599 }
600
601 out:
602 return ret;
603
604 free_pages_out:
605 for (i = 0; i < nr_pages_ret; i++) {
606 WARN_ON(pages[i]->mapping);
607 page_cache_release(pages[i]);
608 }
609 kfree(pages);
610
611 goto out;
612 }
613
614 /*
615 * phase two of compressed writeback. This is the ordered portion
616 * of the code, which only gets called in the order the work was
617 * queued. We walk all the async extents created by compress_file_range
618 * and send them down to the disk.
619 */
620 static noinline int submit_compressed_extents(struct inode *inode,
621 struct async_cow *async_cow)
622 {
623 struct async_extent *async_extent;
624 u64 alloc_hint = 0;
625 struct btrfs_key ins;
626 struct extent_map *em;
627 struct btrfs_root *root = BTRFS_I(inode)->root;
628 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
629 struct extent_io_tree *io_tree;
630 int ret = 0;
631
632 if (list_empty(&async_cow->extents))
633 return 0;
634
635 again:
636 while (!list_empty(&async_cow->extents)) {
637 async_extent = list_entry(async_cow->extents.next,
638 struct async_extent, list);
639 list_del(&async_extent->list);
640
641 io_tree = &BTRFS_I(inode)->io_tree;
642
643 retry:
644 /* did the compression code fall back to uncompressed IO? */
645 if (!async_extent->pages) {
646 int page_started = 0;
647 unsigned long nr_written = 0;
648
649 lock_extent(io_tree, async_extent->start,
650 async_extent->start +
651 async_extent->ram_size - 1);
652
653 /* allocate blocks */
654 ret = cow_file_range(inode, async_cow->locked_page,
655 async_extent->start,
656 async_extent->start +
657 async_extent->ram_size - 1,
658 &page_started, &nr_written, 0);
659
660 /* JDM XXX */
661
662 /*
663 * if page_started, cow_file_range inserted an
664 * inline extent and took care of all the unlocking
665 * and IO for us. Otherwise, we need to submit
666 * all those pages down to the drive.
667 */
668 if (!page_started && !ret)
669 extent_write_locked_range(io_tree,
670 inode, async_extent->start,
671 async_extent->start +
672 async_extent->ram_size - 1,
673 btrfs_get_extent,
674 WB_SYNC_ALL);
675 else if (ret)
676 unlock_page(async_cow->locked_page);
677 kfree(async_extent);
678 cond_resched();
679 continue;
680 }
681
682 lock_extent(io_tree, async_extent->start,
683 async_extent->start + async_extent->ram_size - 1);
684
685 ret = btrfs_reserve_extent(root,
686 async_extent->compressed_size,
687 async_extent->compressed_size,
688 0, alloc_hint, &ins, 1);
689 if (ret) {
690 int i;
691
692 for (i = 0; i < async_extent->nr_pages; i++) {
693 WARN_ON(async_extent->pages[i]->mapping);
694 page_cache_release(async_extent->pages[i]);
695 }
696 kfree(async_extent->pages);
697 async_extent->nr_pages = 0;
698 async_extent->pages = NULL;
699
700 if (ret == -ENOSPC) {
701 unlock_extent(io_tree, async_extent->start,
702 async_extent->start +
703 async_extent->ram_size - 1);
704 goto retry;
705 }
706 goto out_free;
707 }
708
709 /*
710 * here we're doing allocation and writeback of the
711 * compressed pages
712 */
713 btrfs_drop_extent_cache(inode, async_extent->start,
714 async_extent->start +
715 async_extent->ram_size - 1, 0);
716
717 em = alloc_extent_map();
718 if (!em) {
719 ret = -ENOMEM;
720 goto out_free_reserve;
721 }
722 em->start = async_extent->start;
723 em->len = async_extent->ram_size;
724 em->orig_start = em->start;
725 em->mod_start = em->start;
726 em->mod_len = em->len;
727
728 em->block_start = ins.objectid;
729 em->block_len = ins.offset;
730 em->orig_block_len = ins.offset;
731 em->ram_bytes = async_extent->ram_size;
732 em->bdev = root->fs_info->fs_devices->latest_bdev;
733 em->compress_type = async_extent->compress_type;
734 set_bit(EXTENT_FLAG_PINNED, &em->flags);
735 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
736 em->generation = -1;
737
738 while (1) {
739 write_lock(&em_tree->lock);
740 ret = add_extent_mapping(em_tree, em, 1);
741 write_unlock(&em_tree->lock);
742 if (ret != -EEXIST) {
743 free_extent_map(em);
744 break;
745 }
746 btrfs_drop_extent_cache(inode, async_extent->start,
747 async_extent->start +
748 async_extent->ram_size - 1, 0);
749 }
750
751 if (ret)
752 goto out_free_reserve;
753
754 ret = btrfs_add_ordered_extent_compress(inode,
755 async_extent->start,
756 ins.objectid,
757 async_extent->ram_size,
758 ins.offset,
759 BTRFS_ORDERED_COMPRESSED,
760 async_extent->compress_type);
761 if (ret)
762 goto out_free_reserve;
763
764 /*
765 * clear dirty, set writeback and unlock the pages.
766 */
767 extent_clear_unlock_delalloc(inode, async_extent->start,
768 async_extent->start +
769 async_extent->ram_size - 1,
770 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
771 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
772 PAGE_SET_WRITEBACK);
773 ret = btrfs_submit_compressed_write(inode,
774 async_extent->start,
775 async_extent->ram_size,
776 ins.objectid,
777 ins.offset, async_extent->pages,
778 async_extent->nr_pages);
779 alloc_hint = ins.objectid + ins.offset;
780 kfree(async_extent);
781 if (ret)
782 goto out;
783 cond_resched();
784 }
785 ret = 0;
786 out:
787 return ret;
788 out_free_reserve:
789 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
790 out_free:
791 extent_clear_unlock_delalloc(inode, async_extent->start,
792 async_extent->start +
793 async_extent->ram_size - 1,
794 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
795 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
796 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
797 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
798 kfree(async_extent);
799 goto again;
800 }
801
802 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
803 u64 num_bytes)
804 {
805 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
806 struct extent_map *em;
807 u64 alloc_hint = 0;
808
809 read_lock(&em_tree->lock);
810 em = search_extent_mapping(em_tree, start, num_bytes);
811 if (em) {
812 /*
813 * if block start isn't an actual block number then find the
814 * first block in this inode and use that as a hint. If that
815 * block is also bogus then just don't worry about it.
816 */
817 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
818 free_extent_map(em);
819 em = search_extent_mapping(em_tree, 0, 0);
820 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
821 alloc_hint = em->block_start;
822 if (em)
823 free_extent_map(em);
824 } else {
825 alloc_hint = em->block_start;
826 free_extent_map(em);
827 }
828 }
829 read_unlock(&em_tree->lock);
830
831 return alloc_hint;
832 }
833
834 /*
835 * when extent_io.c finds a delayed allocation range in the file,
836 * the call backs end up in this code. The basic idea is to
837 * allocate extents on disk for the range, and create ordered data structs
838 * in ram to track those extents.
839 *
840 * locked_page is the page that writepage had locked already. We use
841 * it to make sure we don't do extra locks or unlocks.
842 *
843 * *page_started is set to one if we unlock locked_page and do everything
844 * required to start IO on it. It may be clean and already done with
845 * IO when we return.
846 */
847 static noinline int cow_file_range(struct inode *inode,
848 struct page *locked_page,
849 u64 start, u64 end, int *page_started,
850 unsigned long *nr_written,
851 int unlock)
852 {
853 struct btrfs_root *root = BTRFS_I(inode)->root;
854 u64 alloc_hint = 0;
855 u64 num_bytes;
856 unsigned long ram_size;
857 u64 disk_num_bytes;
858 u64 cur_alloc_size;
859 u64 blocksize = root->sectorsize;
860 struct btrfs_key ins;
861 struct extent_map *em;
862 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
863 int ret = 0;
864
865 if (btrfs_is_free_space_inode(inode)) {
866 WARN_ON_ONCE(1);
867 ret = -EINVAL;
868 goto out_unlock;
869 }
870
871 num_bytes = ALIGN(end - start + 1, blocksize);
872 num_bytes = max(blocksize, num_bytes);
873 disk_num_bytes = num_bytes;
874
875 /* if this is a small write inside eof, kick off defrag */
876 if (num_bytes < 64 * 1024 &&
877 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
878 btrfs_add_inode_defrag(NULL, inode);
879
880 if (start == 0) {
881 /* lets try to make an inline extent */
882 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
883 NULL);
884 if (ret == 0) {
885 extent_clear_unlock_delalloc(inode, start, end, NULL,
886 EXTENT_LOCKED | EXTENT_DELALLOC |
887 EXTENT_DEFRAG, PAGE_UNLOCK |
888 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
889 PAGE_END_WRITEBACK);
890
891 *nr_written = *nr_written +
892 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
893 *page_started = 1;
894 goto out;
895 } else if (ret < 0) {
896 goto out_unlock;
897 }
898 }
899
900 BUG_ON(disk_num_bytes >
901 btrfs_super_total_bytes(root->fs_info->super_copy));
902
903 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
904 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
905
906 while (disk_num_bytes > 0) {
907 unsigned long op;
908
909 cur_alloc_size = disk_num_bytes;
910 ret = btrfs_reserve_extent(root, cur_alloc_size,
911 root->sectorsize, 0, alloc_hint,
912 &ins, 1);
913 if (ret < 0)
914 goto out_unlock;
915
916 em = alloc_extent_map();
917 if (!em) {
918 ret = -ENOMEM;
919 goto out_reserve;
920 }
921 em->start = start;
922 em->orig_start = em->start;
923 ram_size = ins.offset;
924 em->len = ins.offset;
925 em->mod_start = em->start;
926 em->mod_len = em->len;
927
928 em->block_start = ins.objectid;
929 em->block_len = ins.offset;
930 em->orig_block_len = ins.offset;
931 em->ram_bytes = ram_size;
932 em->bdev = root->fs_info->fs_devices->latest_bdev;
933 set_bit(EXTENT_FLAG_PINNED, &em->flags);
934 em->generation = -1;
935
936 while (1) {
937 write_lock(&em_tree->lock);
938 ret = add_extent_mapping(em_tree, em, 1);
939 write_unlock(&em_tree->lock);
940 if (ret != -EEXIST) {
941 free_extent_map(em);
942 break;
943 }
944 btrfs_drop_extent_cache(inode, start,
945 start + ram_size - 1, 0);
946 }
947 if (ret)
948 goto out_reserve;
949
950 cur_alloc_size = ins.offset;
951 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
952 ram_size, cur_alloc_size, 0);
953 if (ret)
954 goto out_reserve;
955
956 if (root->root_key.objectid ==
957 BTRFS_DATA_RELOC_TREE_OBJECTID) {
958 ret = btrfs_reloc_clone_csums(inode, start,
959 cur_alloc_size);
960 if (ret)
961 goto out_reserve;
962 }
963
964 if (disk_num_bytes < cur_alloc_size)
965 break;
966
967 /* we're not doing compressed IO, don't unlock the first
968 * page (which the caller expects to stay locked), don't
969 * clear any dirty bits and don't set any writeback bits
970 *
971 * Do set the Private2 bit so we know this page was properly
972 * setup for writepage
973 */
974 op = unlock ? PAGE_UNLOCK : 0;
975 op |= PAGE_SET_PRIVATE2;
976
977 extent_clear_unlock_delalloc(inode, start,
978 start + ram_size - 1, locked_page,
979 EXTENT_LOCKED | EXTENT_DELALLOC,
980 op);
981 disk_num_bytes -= cur_alloc_size;
982 num_bytes -= cur_alloc_size;
983 alloc_hint = ins.objectid + ins.offset;
984 start += cur_alloc_size;
985 }
986 out:
987 return ret;
988
989 out_reserve:
990 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
991 out_unlock:
992 extent_clear_unlock_delalloc(inode, start, end, locked_page,
993 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
994 EXTENT_DELALLOC | EXTENT_DEFRAG,
995 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
996 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
997 goto out;
998 }
999
1000 /*
1001 * work queue call back to started compression on a file and pages
1002 */
1003 static noinline void async_cow_start(struct btrfs_work *work)
1004 {
1005 struct async_cow *async_cow;
1006 int num_added = 0;
1007 async_cow = container_of(work, struct async_cow, work);
1008
1009 compress_file_range(async_cow->inode, async_cow->locked_page,
1010 async_cow->start, async_cow->end, async_cow,
1011 &num_added);
1012 if (num_added == 0) {
1013 btrfs_add_delayed_iput(async_cow->inode);
1014 async_cow->inode = NULL;
1015 }
1016 }
1017
1018 /*
1019 * work queue call back to submit previously compressed pages
1020 */
1021 static noinline void async_cow_submit(struct btrfs_work *work)
1022 {
1023 struct async_cow *async_cow;
1024 struct btrfs_root *root;
1025 unsigned long nr_pages;
1026
1027 async_cow = container_of(work, struct async_cow, work);
1028
1029 root = async_cow->root;
1030 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1031 PAGE_CACHE_SHIFT;
1032
1033 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1034 5 * 1024 * 1024 &&
1035 waitqueue_active(&root->fs_info->async_submit_wait))
1036 wake_up(&root->fs_info->async_submit_wait);
1037
1038 if (async_cow->inode)
1039 submit_compressed_extents(async_cow->inode, async_cow);
1040 }
1041
1042 static noinline void async_cow_free(struct btrfs_work *work)
1043 {
1044 struct async_cow *async_cow;
1045 async_cow = container_of(work, struct async_cow, work);
1046 if (async_cow->inode)
1047 btrfs_add_delayed_iput(async_cow->inode);
1048 kfree(async_cow);
1049 }
1050
1051 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1052 u64 start, u64 end, int *page_started,
1053 unsigned long *nr_written)
1054 {
1055 struct async_cow *async_cow;
1056 struct btrfs_root *root = BTRFS_I(inode)->root;
1057 unsigned long nr_pages;
1058 u64 cur_end;
1059 int limit = 10 * 1024 * 1024;
1060
1061 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1062 1, 0, NULL, GFP_NOFS);
1063 while (start < end) {
1064 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1065 BUG_ON(!async_cow); /* -ENOMEM */
1066 async_cow->inode = igrab(inode);
1067 async_cow->root = root;
1068 async_cow->locked_page = locked_page;
1069 async_cow->start = start;
1070
1071 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1072 cur_end = end;
1073 else
1074 cur_end = min(end, start + 512 * 1024 - 1);
1075
1076 async_cow->end = cur_end;
1077 INIT_LIST_HEAD(&async_cow->extents);
1078
1079 btrfs_init_work(&async_cow->work, async_cow_start,
1080 async_cow_submit, async_cow_free);
1081
1082 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1083 PAGE_CACHE_SHIFT;
1084 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1085
1086 btrfs_queue_work(root->fs_info->delalloc_workers,
1087 &async_cow->work);
1088
1089 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1090 wait_event(root->fs_info->async_submit_wait,
1091 (atomic_read(&root->fs_info->async_delalloc_pages) <
1092 limit));
1093 }
1094
1095 while (atomic_read(&root->fs_info->async_submit_draining) &&
1096 atomic_read(&root->fs_info->async_delalloc_pages)) {
1097 wait_event(root->fs_info->async_submit_wait,
1098 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1099 0));
1100 }
1101
1102 *nr_written += nr_pages;
1103 start = cur_end + 1;
1104 }
1105 *page_started = 1;
1106 return 0;
1107 }
1108
1109 static noinline int csum_exist_in_range(struct btrfs_root *root,
1110 u64 bytenr, u64 num_bytes)
1111 {
1112 int ret;
1113 struct btrfs_ordered_sum *sums;
1114 LIST_HEAD(list);
1115
1116 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1117 bytenr + num_bytes - 1, &list, 0);
1118 if (ret == 0 && list_empty(&list))
1119 return 0;
1120
1121 while (!list_empty(&list)) {
1122 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1123 list_del(&sums->list);
1124 kfree(sums);
1125 }
1126 return 1;
1127 }
1128
1129 /*
1130 * when nowcow writeback call back. This checks for snapshots or COW copies
1131 * of the extents that exist in the file, and COWs the file as required.
1132 *
1133 * If no cow copies or snapshots exist, we write directly to the existing
1134 * blocks on disk
1135 */
1136 static noinline int run_delalloc_nocow(struct inode *inode,
1137 struct page *locked_page,
1138 u64 start, u64 end, int *page_started, int force,
1139 unsigned long *nr_written)
1140 {
1141 struct btrfs_root *root = BTRFS_I(inode)->root;
1142 struct btrfs_trans_handle *trans;
1143 struct extent_buffer *leaf;
1144 struct btrfs_path *path;
1145 struct btrfs_file_extent_item *fi;
1146 struct btrfs_key found_key;
1147 u64 cow_start;
1148 u64 cur_offset;
1149 u64 extent_end;
1150 u64 extent_offset;
1151 u64 disk_bytenr;
1152 u64 num_bytes;
1153 u64 disk_num_bytes;
1154 u64 ram_bytes;
1155 int extent_type;
1156 int ret, err;
1157 int type;
1158 int nocow;
1159 int check_prev = 1;
1160 bool nolock;
1161 u64 ino = btrfs_ino(inode);
1162
1163 path = btrfs_alloc_path();
1164 if (!path) {
1165 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1166 EXTENT_LOCKED | EXTENT_DELALLOC |
1167 EXTENT_DO_ACCOUNTING |
1168 EXTENT_DEFRAG, PAGE_UNLOCK |
1169 PAGE_CLEAR_DIRTY |
1170 PAGE_SET_WRITEBACK |
1171 PAGE_END_WRITEBACK);
1172 return -ENOMEM;
1173 }
1174
1175 nolock = btrfs_is_free_space_inode(inode);
1176
1177 if (nolock)
1178 trans = btrfs_join_transaction_nolock(root);
1179 else
1180 trans = btrfs_join_transaction(root);
1181
1182 if (IS_ERR(trans)) {
1183 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1184 EXTENT_LOCKED | EXTENT_DELALLOC |
1185 EXTENT_DO_ACCOUNTING |
1186 EXTENT_DEFRAG, PAGE_UNLOCK |
1187 PAGE_CLEAR_DIRTY |
1188 PAGE_SET_WRITEBACK |
1189 PAGE_END_WRITEBACK);
1190 btrfs_free_path(path);
1191 return PTR_ERR(trans);
1192 }
1193
1194 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1195
1196 cow_start = (u64)-1;
1197 cur_offset = start;
1198 while (1) {
1199 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1200 cur_offset, 0);
1201 if (ret < 0)
1202 goto error;
1203 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1204 leaf = path->nodes[0];
1205 btrfs_item_key_to_cpu(leaf, &found_key,
1206 path->slots[0] - 1);
1207 if (found_key.objectid == ino &&
1208 found_key.type == BTRFS_EXTENT_DATA_KEY)
1209 path->slots[0]--;
1210 }
1211 check_prev = 0;
1212 next_slot:
1213 leaf = path->nodes[0];
1214 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1215 ret = btrfs_next_leaf(root, path);
1216 if (ret < 0)
1217 goto error;
1218 if (ret > 0)
1219 break;
1220 leaf = path->nodes[0];
1221 }
1222
1223 nocow = 0;
1224 disk_bytenr = 0;
1225 num_bytes = 0;
1226 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1227
1228 if (found_key.objectid > ino ||
1229 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1230 found_key.offset > end)
1231 break;
1232
1233 if (found_key.offset > cur_offset) {
1234 extent_end = found_key.offset;
1235 extent_type = 0;
1236 goto out_check;
1237 }
1238
1239 fi = btrfs_item_ptr(leaf, path->slots[0],
1240 struct btrfs_file_extent_item);
1241 extent_type = btrfs_file_extent_type(leaf, fi);
1242
1243 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1244 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1245 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1246 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1247 extent_offset = btrfs_file_extent_offset(leaf, fi);
1248 extent_end = found_key.offset +
1249 btrfs_file_extent_num_bytes(leaf, fi);
1250 disk_num_bytes =
1251 btrfs_file_extent_disk_num_bytes(leaf, fi);
1252 if (extent_end <= start) {
1253 path->slots[0]++;
1254 goto next_slot;
1255 }
1256 if (disk_bytenr == 0)
1257 goto out_check;
1258 if (btrfs_file_extent_compression(leaf, fi) ||
1259 btrfs_file_extent_encryption(leaf, fi) ||
1260 btrfs_file_extent_other_encoding(leaf, fi))
1261 goto out_check;
1262 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1263 goto out_check;
1264 if (btrfs_extent_readonly(root, disk_bytenr))
1265 goto out_check;
1266 if (btrfs_cross_ref_exist(trans, root, ino,
1267 found_key.offset -
1268 extent_offset, disk_bytenr))
1269 goto out_check;
1270 disk_bytenr += extent_offset;
1271 disk_bytenr += cur_offset - found_key.offset;
1272 num_bytes = min(end + 1, extent_end) - cur_offset;
1273 /*
1274 * force cow if csum exists in the range.
1275 * this ensure that csum for a given extent are
1276 * either valid or do not exist.
1277 */
1278 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1279 goto out_check;
1280 nocow = 1;
1281 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1282 extent_end = found_key.offset +
1283 btrfs_file_extent_inline_len(leaf,
1284 path->slots[0], fi);
1285 extent_end = ALIGN(extent_end, root->sectorsize);
1286 } else {
1287 BUG_ON(1);
1288 }
1289 out_check:
1290 if (extent_end <= start) {
1291 path->slots[0]++;
1292 goto next_slot;
1293 }
1294 if (!nocow) {
1295 if (cow_start == (u64)-1)
1296 cow_start = cur_offset;
1297 cur_offset = extent_end;
1298 if (cur_offset > end)
1299 break;
1300 path->slots[0]++;
1301 goto next_slot;
1302 }
1303
1304 btrfs_release_path(path);
1305 if (cow_start != (u64)-1) {
1306 ret = cow_file_range(inode, locked_page,
1307 cow_start, found_key.offset - 1,
1308 page_started, nr_written, 1);
1309 if (ret)
1310 goto error;
1311 cow_start = (u64)-1;
1312 }
1313
1314 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1315 struct extent_map *em;
1316 struct extent_map_tree *em_tree;
1317 em_tree = &BTRFS_I(inode)->extent_tree;
1318 em = alloc_extent_map();
1319 BUG_ON(!em); /* -ENOMEM */
1320 em->start = cur_offset;
1321 em->orig_start = found_key.offset - extent_offset;
1322 em->len = num_bytes;
1323 em->block_len = num_bytes;
1324 em->block_start = disk_bytenr;
1325 em->orig_block_len = disk_num_bytes;
1326 em->ram_bytes = ram_bytes;
1327 em->bdev = root->fs_info->fs_devices->latest_bdev;
1328 em->mod_start = em->start;
1329 em->mod_len = em->len;
1330 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1331 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1332 em->generation = -1;
1333 while (1) {
1334 write_lock(&em_tree->lock);
1335 ret = add_extent_mapping(em_tree, em, 1);
1336 write_unlock(&em_tree->lock);
1337 if (ret != -EEXIST) {
1338 free_extent_map(em);
1339 break;
1340 }
1341 btrfs_drop_extent_cache(inode, em->start,
1342 em->start + em->len - 1, 0);
1343 }
1344 type = BTRFS_ORDERED_PREALLOC;
1345 } else {
1346 type = BTRFS_ORDERED_NOCOW;
1347 }
1348
1349 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1350 num_bytes, num_bytes, type);
1351 BUG_ON(ret); /* -ENOMEM */
1352
1353 if (root->root_key.objectid ==
1354 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1355 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1356 num_bytes);
1357 if (ret)
1358 goto error;
1359 }
1360
1361 extent_clear_unlock_delalloc(inode, cur_offset,
1362 cur_offset + num_bytes - 1,
1363 locked_page, EXTENT_LOCKED |
1364 EXTENT_DELALLOC, PAGE_UNLOCK |
1365 PAGE_SET_PRIVATE2);
1366 cur_offset = extent_end;
1367 if (cur_offset > end)
1368 break;
1369 }
1370 btrfs_release_path(path);
1371
1372 if (cur_offset <= end && cow_start == (u64)-1) {
1373 cow_start = cur_offset;
1374 cur_offset = end;
1375 }
1376
1377 if (cow_start != (u64)-1) {
1378 ret = cow_file_range(inode, locked_page, cow_start, end,
1379 page_started, nr_written, 1);
1380 if (ret)
1381 goto error;
1382 }
1383
1384 error:
1385 err = btrfs_end_transaction(trans, root);
1386 if (!ret)
1387 ret = err;
1388
1389 if (ret && cur_offset < end)
1390 extent_clear_unlock_delalloc(inode, cur_offset, end,
1391 locked_page, EXTENT_LOCKED |
1392 EXTENT_DELALLOC | EXTENT_DEFRAG |
1393 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1394 PAGE_CLEAR_DIRTY |
1395 PAGE_SET_WRITEBACK |
1396 PAGE_END_WRITEBACK);
1397 btrfs_free_path(path);
1398 return ret;
1399 }
1400
1401 /*
1402 * extent_io.c call back to do delayed allocation processing
1403 */
1404 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1405 u64 start, u64 end, int *page_started,
1406 unsigned long *nr_written)
1407 {
1408 int ret;
1409 struct btrfs_root *root = BTRFS_I(inode)->root;
1410
1411 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1412 ret = run_delalloc_nocow(inode, locked_page, start, end,
1413 page_started, 1, nr_written);
1414 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1415 ret = run_delalloc_nocow(inode, locked_page, start, end,
1416 page_started, 0, nr_written);
1417 } else if (!btrfs_test_opt(root, COMPRESS) &&
1418 !(BTRFS_I(inode)->force_compress) &&
1419 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1420 ret = cow_file_range(inode, locked_page, start, end,
1421 page_started, nr_written, 1);
1422 } else {
1423 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1424 &BTRFS_I(inode)->runtime_flags);
1425 ret = cow_file_range_async(inode, locked_page, start, end,
1426 page_started, nr_written);
1427 }
1428 return ret;
1429 }
1430
1431 static void btrfs_split_extent_hook(struct inode *inode,
1432 struct extent_state *orig, u64 split)
1433 {
1434 /* not delalloc, ignore it */
1435 if (!(orig->state & EXTENT_DELALLOC))
1436 return;
1437
1438 spin_lock(&BTRFS_I(inode)->lock);
1439 BTRFS_I(inode)->outstanding_extents++;
1440 spin_unlock(&BTRFS_I(inode)->lock);
1441 }
1442
1443 /*
1444 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1445 * extents so we can keep track of new extents that are just merged onto old
1446 * extents, such as when we are doing sequential writes, so we can properly
1447 * account for the metadata space we'll need.
1448 */
1449 static void btrfs_merge_extent_hook(struct inode *inode,
1450 struct extent_state *new,
1451 struct extent_state *other)
1452 {
1453 /* not delalloc, ignore it */
1454 if (!(other->state & EXTENT_DELALLOC))
1455 return;
1456
1457 spin_lock(&BTRFS_I(inode)->lock);
1458 BTRFS_I(inode)->outstanding_extents--;
1459 spin_unlock(&BTRFS_I(inode)->lock);
1460 }
1461
1462 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1463 struct inode *inode)
1464 {
1465 spin_lock(&root->delalloc_lock);
1466 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1467 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1468 &root->delalloc_inodes);
1469 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1470 &BTRFS_I(inode)->runtime_flags);
1471 root->nr_delalloc_inodes++;
1472 if (root->nr_delalloc_inodes == 1) {
1473 spin_lock(&root->fs_info->delalloc_root_lock);
1474 BUG_ON(!list_empty(&root->delalloc_root));
1475 list_add_tail(&root->delalloc_root,
1476 &root->fs_info->delalloc_roots);
1477 spin_unlock(&root->fs_info->delalloc_root_lock);
1478 }
1479 }
1480 spin_unlock(&root->delalloc_lock);
1481 }
1482
1483 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1484 struct inode *inode)
1485 {
1486 spin_lock(&root->delalloc_lock);
1487 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1488 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1489 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1490 &BTRFS_I(inode)->runtime_flags);
1491 root->nr_delalloc_inodes--;
1492 if (!root->nr_delalloc_inodes) {
1493 spin_lock(&root->fs_info->delalloc_root_lock);
1494 BUG_ON(list_empty(&root->delalloc_root));
1495 list_del_init(&root->delalloc_root);
1496 spin_unlock(&root->fs_info->delalloc_root_lock);
1497 }
1498 }
1499 spin_unlock(&root->delalloc_lock);
1500 }
1501
1502 /*
1503 * extent_io.c set_bit_hook, used to track delayed allocation
1504 * bytes in this file, and to maintain the list of inodes that
1505 * have pending delalloc work to be done.
1506 */
1507 static void btrfs_set_bit_hook(struct inode *inode,
1508 struct extent_state *state, unsigned long *bits)
1509 {
1510
1511 /*
1512 * set_bit and clear bit hooks normally require _irqsave/restore
1513 * but in this case, we are only testing for the DELALLOC
1514 * bit, which is only set or cleared with irqs on
1515 */
1516 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1517 struct btrfs_root *root = BTRFS_I(inode)->root;
1518 u64 len = state->end + 1 - state->start;
1519 bool do_list = !btrfs_is_free_space_inode(inode);
1520
1521 if (*bits & EXTENT_FIRST_DELALLOC) {
1522 *bits &= ~EXTENT_FIRST_DELALLOC;
1523 } else {
1524 spin_lock(&BTRFS_I(inode)->lock);
1525 BTRFS_I(inode)->outstanding_extents++;
1526 spin_unlock(&BTRFS_I(inode)->lock);
1527 }
1528
1529 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1530 root->fs_info->delalloc_batch);
1531 spin_lock(&BTRFS_I(inode)->lock);
1532 BTRFS_I(inode)->delalloc_bytes += len;
1533 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1534 &BTRFS_I(inode)->runtime_flags))
1535 btrfs_add_delalloc_inodes(root, inode);
1536 spin_unlock(&BTRFS_I(inode)->lock);
1537 }
1538 }
1539
1540 /*
1541 * extent_io.c clear_bit_hook, see set_bit_hook for why
1542 */
1543 static void btrfs_clear_bit_hook(struct inode *inode,
1544 struct extent_state *state,
1545 unsigned long *bits)
1546 {
1547 /*
1548 * set_bit and clear bit hooks normally require _irqsave/restore
1549 * but in this case, we are only testing for the DELALLOC
1550 * bit, which is only set or cleared with irqs on
1551 */
1552 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1553 struct btrfs_root *root = BTRFS_I(inode)->root;
1554 u64 len = state->end + 1 - state->start;
1555 bool do_list = !btrfs_is_free_space_inode(inode);
1556
1557 if (*bits & EXTENT_FIRST_DELALLOC) {
1558 *bits &= ~EXTENT_FIRST_DELALLOC;
1559 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1560 spin_lock(&BTRFS_I(inode)->lock);
1561 BTRFS_I(inode)->outstanding_extents--;
1562 spin_unlock(&BTRFS_I(inode)->lock);
1563 }
1564
1565 /*
1566 * We don't reserve metadata space for space cache inodes so we
1567 * don't need to call dellalloc_release_metadata if there is an
1568 * error.
1569 */
1570 if (*bits & EXTENT_DO_ACCOUNTING &&
1571 root != root->fs_info->tree_root)
1572 btrfs_delalloc_release_metadata(inode, len);
1573
1574 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1575 && do_list && !(state->state & EXTENT_NORESERVE))
1576 btrfs_free_reserved_data_space(inode, len);
1577
1578 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1579 root->fs_info->delalloc_batch);
1580 spin_lock(&BTRFS_I(inode)->lock);
1581 BTRFS_I(inode)->delalloc_bytes -= len;
1582 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1583 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1584 &BTRFS_I(inode)->runtime_flags))
1585 btrfs_del_delalloc_inode(root, inode);
1586 spin_unlock(&BTRFS_I(inode)->lock);
1587 }
1588 }
1589
1590 /*
1591 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1592 * we don't create bios that span stripes or chunks
1593 */
1594 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1595 size_t size, struct bio *bio,
1596 unsigned long bio_flags)
1597 {
1598 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1599 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1600 u64 length = 0;
1601 u64 map_length;
1602 int ret;
1603
1604 if (bio_flags & EXTENT_BIO_COMPRESSED)
1605 return 0;
1606
1607 length = bio->bi_iter.bi_size;
1608 map_length = length;
1609 ret = btrfs_map_block(root->fs_info, rw, logical,
1610 &map_length, NULL, 0);
1611 /* Will always return 0 with map_multi == NULL */
1612 BUG_ON(ret < 0);
1613 if (map_length < length + size)
1614 return 1;
1615 return 0;
1616 }
1617
1618 /*
1619 * in order to insert checksums into the metadata in large chunks,
1620 * we wait until bio submission time. All the pages in the bio are
1621 * checksummed and sums are attached onto the ordered extent record.
1622 *
1623 * At IO completion time the cums attached on the ordered extent record
1624 * are inserted into the btree
1625 */
1626 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1627 struct bio *bio, int mirror_num,
1628 unsigned long bio_flags,
1629 u64 bio_offset)
1630 {
1631 struct btrfs_root *root = BTRFS_I(inode)->root;
1632 int ret = 0;
1633
1634 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1635 BUG_ON(ret); /* -ENOMEM */
1636 return 0;
1637 }
1638
1639 /*
1640 * in order to insert checksums into the metadata in large chunks,
1641 * we wait until bio submission time. All the pages in the bio are
1642 * checksummed and sums are attached onto the ordered extent record.
1643 *
1644 * At IO completion time the cums attached on the ordered extent record
1645 * are inserted into the btree
1646 */
1647 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1648 int mirror_num, unsigned long bio_flags,
1649 u64 bio_offset)
1650 {
1651 struct btrfs_root *root = BTRFS_I(inode)->root;
1652 int ret;
1653
1654 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1655 if (ret)
1656 bio_endio(bio, ret);
1657 return ret;
1658 }
1659
1660 /*
1661 * extent_io.c submission hook. This does the right thing for csum calculation
1662 * on write, or reading the csums from the tree before a read
1663 */
1664 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1665 int mirror_num, unsigned long bio_flags,
1666 u64 bio_offset)
1667 {
1668 struct btrfs_root *root = BTRFS_I(inode)->root;
1669 int ret = 0;
1670 int skip_sum;
1671 int metadata = 0;
1672 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1673
1674 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1675
1676 if (btrfs_is_free_space_inode(inode))
1677 metadata = 2;
1678
1679 if (!(rw & REQ_WRITE)) {
1680 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1681 if (ret)
1682 goto out;
1683
1684 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1685 ret = btrfs_submit_compressed_read(inode, bio,
1686 mirror_num,
1687 bio_flags);
1688 goto out;
1689 } else if (!skip_sum) {
1690 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1691 if (ret)
1692 goto out;
1693 }
1694 goto mapit;
1695 } else if (async && !skip_sum) {
1696 /* csum items have already been cloned */
1697 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1698 goto mapit;
1699 /* we're doing a write, do the async checksumming */
1700 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1701 inode, rw, bio, mirror_num,
1702 bio_flags, bio_offset,
1703 __btrfs_submit_bio_start,
1704 __btrfs_submit_bio_done);
1705 goto out;
1706 } else if (!skip_sum) {
1707 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1708 if (ret)
1709 goto out;
1710 }
1711
1712 mapit:
1713 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1714
1715 out:
1716 if (ret < 0)
1717 bio_endio(bio, ret);
1718 return ret;
1719 }
1720
1721 /*
1722 * given a list of ordered sums record them in the inode. This happens
1723 * at IO completion time based on sums calculated at bio submission time.
1724 */
1725 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1726 struct inode *inode, u64 file_offset,
1727 struct list_head *list)
1728 {
1729 struct btrfs_ordered_sum *sum;
1730
1731 list_for_each_entry(sum, list, list) {
1732 trans->adding_csums = 1;
1733 btrfs_csum_file_blocks(trans,
1734 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1735 trans->adding_csums = 0;
1736 }
1737 return 0;
1738 }
1739
1740 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1741 struct extent_state **cached_state)
1742 {
1743 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1744 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1745 cached_state, GFP_NOFS);
1746 }
1747
1748 /* see btrfs_writepage_start_hook for details on why this is required */
1749 struct btrfs_writepage_fixup {
1750 struct page *page;
1751 struct btrfs_work work;
1752 };
1753
1754 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1755 {
1756 struct btrfs_writepage_fixup *fixup;
1757 struct btrfs_ordered_extent *ordered;
1758 struct extent_state *cached_state = NULL;
1759 struct page *page;
1760 struct inode *inode;
1761 u64 page_start;
1762 u64 page_end;
1763 int ret;
1764
1765 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1766 page = fixup->page;
1767 again:
1768 lock_page(page);
1769 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1770 ClearPageChecked(page);
1771 goto out_page;
1772 }
1773
1774 inode = page->mapping->host;
1775 page_start = page_offset(page);
1776 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1777
1778 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1779 &cached_state);
1780
1781 /* already ordered? We're done */
1782 if (PagePrivate2(page))
1783 goto out;
1784
1785 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1786 if (ordered) {
1787 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1788 page_end, &cached_state, GFP_NOFS);
1789 unlock_page(page);
1790 btrfs_start_ordered_extent(inode, ordered, 1);
1791 btrfs_put_ordered_extent(ordered);
1792 goto again;
1793 }
1794
1795 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1796 if (ret) {
1797 mapping_set_error(page->mapping, ret);
1798 end_extent_writepage(page, ret, page_start, page_end);
1799 ClearPageChecked(page);
1800 goto out;
1801 }
1802
1803 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1804 ClearPageChecked(page);
1805 set_page_dirty(page);
1806 out:
1807 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1808 &cached_state, GFP_NOFS);
1809 out_page:
1810 unlock_page(page);
1811 page_cache_release(page);
1812 kfree(fixup);
1813 }
1814
1815 /*
1816 * There are a few paths in the higher layers of the kernel that directly
1817 * set the page dirty bit without asking the filesystem if it is a
1818 * good idea. This causes problems because we want to make sure COW
1819 * properly happens and the data=ordered rules are followed.
1820 *
1821 * In our case any range that doesn't have the ORDERED bit set
1822 * hasn't been properly setup for IO. We kick off an async process
1823 * to fix it up. The async helper will wait for ordered extents, set
1824 * the delalloc bit and make it safe to write the page.
1825 */
1826 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1827 {
1828 struct inode *inode = page->mapping->host;
1829 struct btrfs_writepage_fixup *fixup;
1830 struct btrfs_root *root = BTRFS_I(inode)->root;
1831
1832 /* this page is properly in the ordered list */
1833 if (TestClearPagePrivate2(page))
1834 return 0;
1835
1836 if (PageChecked(page))
1837 return -EAGAIN;
1838
1839 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1840 if (!fixup)
1841 return -EAGAIN;
1842
1843 SetPageChecked(page);
1844 page_cache_get(page);
1845 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
1846 fixup->page = page;
1847 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
1848 return -EBUSY;
1849 }
1850
1851 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1852 struct inode *inode, u64 file_pos,
1853 u64 disk_bytenr, u64 disk_num_bytes,
1854 u64 num_bytes, u64 ram_bytes,
1855 u8 compression, u8 encryption,
1856 u16 other_encoding, int extent_type)
1857 {
1858 struct btrfs_root *root = BTRFS_I(inode)->root;
1859 struct btrfs_file_extent_item *fi;
1860 struct btrfs_path *path;
1861 struct extent_buffer *leaf;
1862 struct btrfs_key ins;
1863 int extent_inserted = 0;
1864 int ret;
1865
1866 path = btrfs_alloc_path();
1867 if (!path)
1868 return -ENOMEM;
1869
1870 /*
1871 * we may be replacing one extent in the tree with another.
1872 * The new extent is pinned in the extent map, and we don't want
1873 * to drop it from the cache until it is completely in the btree.
1874 *
1875 * So, tell btrfs_drop_extents to leave this extent in the cache.
1876 * the caller is expected to unpin it and allow it to be merged
1877 * with the others.
1878 */
1879 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1880 file_pos + num_bytes, NULL, 0,
1881 1, sizeof(*fi), &extent_inserted);
1882 if (ret)
1883 goto out;
1884
1885 if (!extent_inserted) {
1886 ins.objectid = btrfs_ino(inode);
1887 ins.offset = file_pos;
1888 ins.type = BTRFS_EXTENT_DATA_KEY;
1889
1890 path->leave_spinning = 1;
1891 ret = btrfs_insert_empty_item(trans, root, path, &ins,
1892 sizeof(*fi));
1893 if (ret)
1894 goto out;
1895 }
1896 leaf = path->nodes[0];
1897 fi = btrfs_item_ptr(leaf, path->slots[0],
1898 struct btrfs_file_extent_item);
1899 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1900 btrfs_set_file_extent_type(leaf, fi, extent_type);
1901 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1902 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1903 btrfs_set_file_extent_offset(leaf, fi, 0);
1904 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1905 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1906 btrfs_set_file_extent_compression(leaf, fi, compression);
1907 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1908 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1909
1910 btrfs_mark_buffer_dirty(leaf);
1911 btrfs_release_path(path);
1912
1913 inode_add_bytes(inode, num_bytes);
1914
1915 ins.objectid = disk_bytenr;
1916 ins.offset = disk_num_bytes;
1917 ins.type = BTRFS_EXTENT_ITEM_KEY;
1918 ret = btrfs_alloc_reserved_file_extent(trans, root,
1919 root->root_key.objectid,
1920 btrfs_ino(inode), file_pos, &ins);
1921 out:
1922 btrfs_free_path(path);
1923
1924 return ret;
1925 }
1926
1927 /* snapshot-aware defrag */
1928 struct sa_defrag_extent_backref {
1929 struct rb_node node;
1930 struct old_sa_defrag_extent *old;
1931 u64 root_id;
1932 u64 inum;
1933 u64 file_pos;
1934 u64 extent_offset;
1935 u64 num_bytes;
1936 u64 generation;
1937 };
1938
1939 struct old_sa_defrag_extent {
1940 struct list_head list;
1941 struct new_sa_defrag_extent *new;
1942
1943 u64 extent_offset;
1944 u64 bytenr;
1945 u64 offset;
1946 u64 len;
1947 int count;
1948 };
1949
1950 struct new_sa_defrag_extent {
1951 struct rb_root root;
1952 struct list_head head;
1953 struct btrfs_path *path;
1954 struct inode *inode;
1955 u64 file_pos;
1956 u64 len;
1957 u64 bytenr;
1958 u64 disk_len;
1959 u8 compress_type;
1960 };
1961
1962 static int backref_comp(struct sa_defrag_extent_backref *b1,
1963 struct sa_defrag_extent_backref *b2)
1964 {
1965 if (b1->root_id < b2->root_id)
1966 return -1;
1967 else if (b1->root_id > b2->root_id)
1968 return 1;
1969
1970 if (b1->inum < b2->inum)
1971 return -1;
1972 else if (b1->inum > b2->inum)
1973 return 1;
1974
1975 if (b1->file_pos < b2->file_pos)
1976 return -1;
1977 else if (b1->file_pos > b2->file_pos)
1978 return 1;
1979
1980 /*
1981 * [------------------------------] ===> (a range of space)
1982 * |<--->| |<---->| =============> (fs/file tree A)
1983 * |<---------------------------->| ===> (fs/file tree B)
1984 *
1985 * A range of space can refer to two file extents in one tree while
1986 * refer to only one file extent in another tree.
1987 *
1988 * So we may process a disk offset more than one time(two extents in A)
1989 * and locate at the same extent(one extent in B), then insert two same
1990 * backrefs(both refer to the extent in B).
1991 */
1992 return 0;
1993 }
1994
1995 static void backref_insert(struct rb_root *root,
1996 struct sa_defrag_extent_backref *backref)
1997 {
1998 struct rb_node **p = &root->rb_node;
1999 struct rb_node *parent = NULL;
2000 struct sa_defrag_extent_backref *entry;
2001 int ret;
2002
2003 while (*p) {
2004 parent = *p;
2005 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2006
2007 ret = backref_comp(backref, entry);
2008 if (ret < 0)
2009 p = &(*p)->rb_left;
2010 else
2011 p = &(*p)->rb_right;
2012 }
2013
2014 rb_link_node(&backref->node, parent, p);
2015 rb_insert_color(&backref->node, root);
2016 }
2017
2018 /*
2019 * Note the backref might has changed, and in this case we just return 0.
2020 */
2021 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2022 void *ctx)
2023 {
2024 struct btrfs_file_extent_item *extent;
2025 struct btrfs_fs_info *fs_info;
2026 struct old_sa_defrag_extent *old = ctx;
2027 struct new_sa_defrag_extent *new = old->new;
2028 struct btrfs_path *path = new->path;
2029 struct btrfs_key key;
2030 struct btrfs_root *root;
2031 struct sa_defrag_extent_backref *backref;
2032 struct extent_buffer *leaf;
2033 struct inode *inode = new->inode;
2034 int slot;
2035 int ret;
2036 u64 extent_offset;
2037 u64 num_bytes;
2038
2039 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2040 inum == btrfs_ino(inode))
2041 return 0;
2042
2043 key.objectid = root_id;
2044 key.type = BTRFS_ROOT_ITEM_KEY;
2045 key.offset = (u64)-1;
2046
2047 fs_info = BTRFS_I(inode)->root->fs_info;
2048 root = btrfs_read_fs_root_no_name(fs_info, &key);
2049 if (IS_ERR(root)) {
2050 if (PTR_ERR(root) == -ENOENT)
2051 return 0;
2052 WARN_ON(1);
2053 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2054 inum, offset, root_id);
2055 return PTR_ERR(root);
2056 }
2057
2058 key.objectid = inum;
2059 key.type = BTRFS_EXTENT_DATA_KEY;
2060 if (offset > (u64)-1 << 32)
2061 key.offset = 0;
2062 else
2063 key.offset = offset;
2064
2065 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2066 if (WARN_ON(ret < 0))
2067 return ret;
2068 ret = 0;
2069
2070 while (1) {
2071 cond_resched();
2072
2073 leaf = path->nodes[0];
2074 slot = path->slots[0];
2075
2076 if (slot >= btrfs_header_nritems(leaf)) {
2077 ret = btrfs_next_leaf(root, path);
2078 if (ret < 0) {
2079 goto out;
2080 } else if (ret > 0) {
2081 ret = 0;
2082 goto out;
2083 }
2084 continue;
2085 }
2086
2087 path->slots[0]++;
2088
2089 btrfs_item_key_to_cpu(leaf, &key, slot);
2090
2091 if (key.objectid > inum)
2092 goto out;
2093
2094 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2095 continue;
2096
2097 extent = btrfs_item_ptr(leaf, slot,
2098 struct btrfs_file_extent_item);
2099
2100 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2101 continue;
2102
2103 /*
2104 * 'offset' refers to the exact key.offset,
2105 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2106 * (key.offset - extent_offset).
2107 */
2108 if (key.offset != offset)
2109 continue;
2110
2111 extent_offset = btrfs_file_extent_offset(leaf, extent);
2112 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2113
2114 if (extent_offset >= old->extent_offset + old->offset +
2115 old->len || extent_offset + num_bytes <=
2116 old->extent_offset + old->offset)
2117 continue;
2118 break;
2119 }
2120
2121 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2122 if (!backref) {
2123 ret = -ENOENT;
2124 goto out;
2125 }
2126
2127 backref->root_id = root_id;
2128 backref->inum = inum;
2129 backref->file_pos = offset;
2130 backref->num_bytes = num_bytes;
2131 backref->extent_offset = extent_offset;
2132 backref->generation = btrfs_file_extent_generation(leaf, extent);
2133 backref->old = old;
2134 backref_insert(&new->root, backref);
2135 old->count++;
2136 out:
2137 btrfs_release_path(path);
2138 WARN_ON(ret);
2139 return ret;
2140 }
2141
2142 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2143 struct new_sa_defrag_extent *new)
2144 {
2145 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2146 struct old_sa_defrag_extent *old, *tmp;
2147 int ret;
2148
2149 new->path = path;
2150
2151 list_for_each_entry_safe(old, tmp, &new->head, list) {
2152 ret = iterate_inodes_from_logical(old->bytenr +
2153 old->extent_offset, fs_info,
2154 path, record_one_backref,
2155 old);
2156 if (ret < 0 && ret != -ENOENT)
2157 return false;
2158
2159 /* no backref to be processed for this extent */
2160 if (!old->count) {
2161 list_del(&old->list);
2162 kfree(old);
2163 }
2164 }
2165
2166 if (list_empty(&new->head))
2167 return false;
2168
2169 return true;
2170 }
2171
2172 static int relink_is_mergable(struct extent_buffer *leaf,
2173 struct btrfs_file_extent_item *fi,
2174 struct new_sa_defrag_extent *new)
2175 {
2176 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2177 return 0;
2178
2179 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2180 return 0;
2181
2182 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2183 return 0;
2184
2185 if (btrfs_file_extent_encryption(leaf, fi) ||
2186 btrfs_file_extent_other_encoding(leaf, fi))
2187 return 0;
2188
2189 return 1;
2190 }
2191
2192 /*
2193 * Note the backref might has changed, and in this case we just return 0.
2194 */
2195 static noinline int relink_extent_backref(struct btrfs_path *path,
2196 struct sa_defrag_extent_backref *prev,
2197 struct sa_defrag_extent_backref *backref)
2198 {
2199 struct btrfs_file_extent_item *extent;
2200 struct btrfs_file_extent_item *item;
2201 struct btrfs_ordered_extent *ordered;
2202 struct btrfs_trans_handle *trans;
2203 struct btrfs_fs_info *fs_info;
2204 struct btrfs_root *root;
2205 struct btrfs_key key;
2206 struct extent_buffer *leaf;
2207 struct old_sa_defrag_extent *old = backref->old;
2208 struct new_sa_defrag_extent *new = old->new;
2209 struct inode *src_inode = new->inode;
2210 struct inode *inode;
2211 struct extent_state *cached = NULL;
2212 int ret = 0;
2213 u64 start;
2214 u64 len;
2215 u64 lock_start;
2216 u64 lock_end;
2217 bool merge = false;
2218 int index;
2219
2220 if (prev && prev->root_id == backref->root_id &&
2221 prev->inum == backref->inum &&
2222 prev->file_pos + prev->num_bytes == backref->file_pos)
2223 merge = true;
2224
2225 /* step 1: get root */
2226 key.objectid = backref->root_id;
2227 key.type = BTRFS_ROOT_ITEM_KEY;
2228 key.offset = (u64)-1;
2229
2230 fs_info = BTRFS_I(src_inode)->root->fs_info;
2231 index = srcu_read_lock(&fs_info->subvol_srcu);
2232
2233 root = btrfs_read_fs_root_no_name(fs_info, &key);
2234 if (IS_ERR(root)) {
2235 srcu_read_unlock(&fs_info->subvol_srcu, index);
2236 if (PTR_ERR(root) == -ENOENT)
2237 return 0;
2238 return PTR_ERR(root);
2239 }
2240
2241 if (btrfs_root_readonly(root)) {
2242 srcu_read_unlock(&fs_info->subvol_srcu, index);
2243 return 0;
2244 }
2245
2246 /* step 2: get inode */
2247 key.objectid = backref->inum;
2248 key.type = BTRFS_INODE_ITEM_KEY;
2249 key.offset = 0;
2250
2251 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2252 if (IS_ERR(inode)) {
2253 srcu_read_unlock(&fs_info->subvol_srcu, index);
2254 return 0;
2255 }
2256
2257 srcu_read_unlock(&fs_info->subvol_srcu, index);
2258
2259 /* step 3: relink backref */
2260 lock_start = backref->file_pos;
2261 lock_end = backref->file_pos + backref->num_bytes - 1;
2262 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2263 0, &cached);
2264
2265 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2266 if (ordered) {
2267 btrfs_put_ordered_extent(ordered);
2268 goto out_unlock;
2269 }
2270
2271 trans = btrfs_join_transaction(root);
2272 if (IS_ERR(trans)) {
2273 ret = PTR_ERR(trans);
2274 goto out_unlock;
2275 }
2276
2277 key.objectid = backref->inum;
2278 key.type = BTRFS_EXTENT_DATA_KEY;
2279 key.offset = backref->file_pos;
2280
2281 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2282 if (ret < 0) {
2283 goto out_free_path;
2284 } else if (ret > 0) {
2285 ret = 0;
2286 goto out_free_path;
2287 }
2288
2289 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2290 struct btrfs_file_extent_item);
2291
2292 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2293 backref->generation)
2294 goto out_free_path;
2295
2296 btrfs_release_path(path);
2297
2298 start = backref->file_pos;
2299 if (backref->extent_offset < old->extent_offset + old->offset)
2300 start += old->extent_offset + old->offset -
2301 backref->extent_offset;
2302
2303 len = min(backref->extent_offset + backref->num_bytes,
2304 old->extent_offset + old->offset + old->len);
2305 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2306
2307 ret = btrfs_drop_extents(trans, root, inode, start,
2308 start + len, 1);
2309 if (ret)
2310 goto out_free_path;
2311 again:
2312 key.objectid = btrfs_ino(inode);
2313 key.type = BTRFS_EXTENT_DATA_KEY;
2314 key.offset = start;
2315
2316 path->leave_spinning = 1;
2317 if (merge) {
2318 struct btrfs_file_extent_item *fi;
2319 u64 extent_len;
2320 struct btrfs_key found_key;
2321
2322 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2323 if (ret < 0)
2324 goto out_free_path;
2325
2326 path->slots[0]--;
2327 leaf = path->nodes[0];
2328 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2329
2330 fi = btrfs_item_ptr(leaf, path->slots[0],
2331 struct btrfs_file_extent_item);
2332 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2333
2334 if (extent_len + found_key.offset == start &&
2335 relink_is_mergable(leaf, fi, new)) {
2336 btrfs_set_file_extent_num_bytes(leaf, fi,
2337 extent_len + len);
2338 btrfs_mark_buffer_dirty(leaf);
2339 inode_add_bytes(inode, len);
2340
2341 ret = 1;
2342 goto out_free_path;
2343 } else {
2344 merge = false;
2345 btrfs_release_path(path);
2346 goto again;
2347 }
2348 }
2349
2350 ret = btrfs_insert_empty_item(trans, root, path, &key,
2351 sizeof(*extent));
2352 if (ret) {
2353 btrfs_abort_transaction(trans, root, ret);
2354 goto out_free_path;
2355 }
2356
2357 leaf = path->nodes[0];
2358 item = btrfs_item_ptr(leaf, path->slots[0],
2359 struct btrfs_file_extent_item);
2360 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2361 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2362 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2363 btrfs_set_file_extent_num_bytes(leaf, item, len);
2364 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2365 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2366 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2367 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2368 btrfs_set_file_extent_encryption(leaf, item, 0);
2369 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2370
2371 btrfs_mark_buffer_dirty(leaf);
2372 inode_add_bytes(inode, len);
2373 btrfs_release_path(path);
2374
2375 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2376 new->disk_len, 0,
2377 backref->root_id, backref->inum,
2378 new->file_pos, 0); /* start - extent_offset */
2379 if (ret) {
2380 btrfs_abort_transaction(trans, root, ret);
2381 goto out_free_path;
2382 }
2383
2384 ret = 1;
2385 out_free_path:
2386 btrfs_release_path(path);
2387 path->leave_spinning = 0;
2388 btrfs_end_transaction(trans, root);
2389 out_unlock:
2390 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2391 &cached, GFP_NOFS);
2392 iput(inode);
2393 return ret;
2394 }
2395
2396 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2397 {
2398 struct old_sa_defrag_extent *old, *tmp;
2399
2400 if (!new)
2401 return;
2402
2403 list_for_each_entry_safe(old, tmp, &new->head, list) {
2404 list_del(&old->list);
2405 kfree(old);
2406 }
2407 kfree(new);
2408 }
2409
2410 static void relink_file_extents(struct new_sa_defrag_extent *new)
2411 {
2412 struct btrfs_path *path;
2413 struct sa_defrag_extent_backref *backref;
2414 struct sa_defrag_extent_backref *prev = NULL;
2415 struct inode *inode;
2416 struct btrfs_root *root;
2417 struct rb_node *node;
2418 int ret;
2419
2420 inode = new->inode;
2421 root = BTRFS_I(inode)->root;
2422
2423 path = btrfs_alloc_path();
2424 if (!path)
2425 return;
2426
2427 if (!record_extent_backrefs(path, new)) {
2428 btrfs_free_path(path);
2429 goto out;
2430 }
2431 btrfs_release_path(path);
2432
2433 while (1) {
2434 node = rb_first(&new->root);
2435 if (!node)
2436 break;
2437 rb_erase(node, &new->root);
2438
2439 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2440
2441 ret = relink_extent_backref(path, prev, backref);
2442 WARN_ON(ret < 0);
2443
2444 kfree(prev);
2445
2446 if (ret == 1)
2447 prev = backref;
2448 else
2449 prev = NULL;
2450 cond_resched();
2451 }
2452 kfree(prev);
2453
2454 btrfs_free_path(path);
2455 out:
2456 free_sa_defrag_extent(new);
2457
2458 atomic_dec(&root->fs_info->defrag_running);
2459 wake_up(&root->fs_info->transaction_wait);
2460 }
2461
2462 static struct new_sa_defrag_extent *
2463 record_old_file_extents(struct inode *inode,
2464 struct btrfs_ordered_extent *ordered)
2465 {
2466 struct btrfs_root *root = BTRFS_I(inode)->root;
2467 struct btrfs_path *path;
2468 struct btrfs_key key;
2469 struct old_sa_defrag_extent *old;
2470 struct new_sa_defrag_extent *new;
2471 int ret;
2472
2473 new = kmalloc(sizeof(*new), GFP_NOFS);
2474 if (!new)
2475 return NULL;
2476
2477 new->inode = inode;
2478 new->file_pos = ordered->file_offset;
2479 new->len = ordered->len;
2480 new->bytenr = ordered->start;
2481 new->disk_len = ordered->disk_len;
2482 new->compress_type = ordered->compress_type;
2483 new->root = RB_ROOT;
2484 INIT_LIST_HEAD(&new->head);
2485
2486 path = btrfs_alloc_path();
2487 if (!path)
2488 goto out_kfree;
2489
2490 key.objectid = btrfs_ino(inode);
2491 key.type = BTRFS_EXTENT_DATA_KEY;
2492 key.offset = new->file_pos;
2493
2494 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2495 if (ret < 0)
2496 goto out_free_path;
2497 if (ret > 0 && path->slots[0] > 0)
2498 path->slots[0]--;
2499
2500 /* find out all the old extents for the file range */
2501 while (1) {
2502 struct btrfs_file_extent_item *extent;
2503 struct extent_buffer *l;
2504 int slot;
2505 u64 num_bytes;
2506 u64 offset;
2507 u64 end;
2508 u64 disk_bytenr;
2509 u64 extent_offset;
2510
2511 l = path->nodes[0];
2512 slot = path->slots[0];
2513
2514 if (slot >= btrfs_header_nritems(l)) {
2515 ret = btrfs_next_leaf(root, path);
2516 if (ret < 0)
2517 goto out_free_path;
2518 else if (ret > 0)
2519 break;
2520 continue;
2521 }
2522
2523 btrfs_item_key_to_cpu(l, &key, slot);
2524
2525 if (key.objectid != btrfs_ino(inode))
2526 break;
2527 if (key.type != BTRFS_EXTENT_DATA_KEY)
2528 break;
2529 if (key.offset >= new->file_pos + new->len)
2530 break;
2531
2532 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2533
2534 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2535 if (key.offset + num_bytes < new->file_pos)
2536 goto next;
2537
2538 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2539 if (!disk_bytenr)
2540 goto next;
2541
2542 extent_offset = btrfs_file_extent_offset(l, extent);
2543
2544 old = kmalloc(sizeof(*old), GFP_NOFS);
2545 if (!old)
2546 goto out_free_path;
2547
2548 offset = max(new->file_pos, key.offset);
2549 end = min(new->file_pos + new->len, key.offset + num_bytes);
2550
2551 old->bytenr = disk_bytenr;
2552 old->extent_offset = extent_offset;
2553 old->offset = offset - key.offset;
2554 old->len = end - offset;
2555 old->new = new;
2556 old->count = 0;
2557 list_add_tail(&old->list, &new->head);
2558 next:
2559 path->slots[0]++;
2560 cond_resched();
2561 }
2562
2563 btrfs_free_path(path);
2564 atomic_inc(&root->fs_info->defrag_running);
2565
2566 return new;
2567
2568 out_free_path:
2569 btrfs_free_path(path);
2570 out_kfree:
2571 free_sa_defrag_extent(new);
2572 return NULL;
2573 }
2574
2575 /* as ordered data IO finishes, this gets called so we can finish
2576 * an ordered extent if the range of bytes in the file it covers are
2577 * fully written.
2578 */
2579 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2580 {
2581 struct inode *inode = ordered_extent->inode;
2582 struct btrfs_root *root = BTRFS_I(inode)->root;
2583 struct btrfs_trans_handle *trans = NULL;
2584 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2585 struct extent_state *cached_state = NULL;
2586 struct new_sa_defrag_extent *new = NULL;
2587 int compress_type = 0;
2588 int ret = 0;
2589 u64 logical_len = ordered_extent->len;
2590 bool nolock;
2591 bool truncated = false;
2592
2593 nolock = btrfs_is_free_space_inode(inode);
2594
2595 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2596 ret = -EIO;
2597 goto out;
2598 }
2599
2600 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2601 truncated = true;
2602 logical_len = ordered_extent->truncated_len;
2603 /* Truncated the entire extent, don't bother adding */
2604 if (!logical_len)
2605 goto out;
2606 }
2607
2608 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2609 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2610 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2611 if (nolock)
2612 trans = btrfs_join_transaction_nolock(root);
2613 else
2614 trans = btrfs_join_transaction(root);
2615 if (IS_ERR(trans)) {
2616 ret = PTR_ERR(trans);
2617 trans = NULL;
2618 goto out;
2619 }
2620 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2621 ret = btrfs_update_inode_fallback(trans, root, inode);
2622 if (ret) /* -ENOMEM or corruption */
2623 btrfs_abort_transaction(trans, root, ret);
2624 goto out;
2625 }
2626
2627 lock_extent_bits(io_tree, ordered_extent->file_offset,
2628 ordered_extent->file_offset + ordered_extent->len - 1,
2629 0, &cached_state);
2630
2631 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2632 ordered_extent->file_offset + ordered_extent->len - 1,
2633 EXTENT_DEFRAG, 1, cached_state);
2634 if (ret) {
2635 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2636 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2637 /* the inode is shared */
2638 new = record_old_file_extents(inode, ordered_extent);
2639
2640 clear_extent_bit(io_tree, ordered_extent->file_offset,
2641 ordered_extent->file_offset + ordered_extent->len - 1,
2642 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2643 }
2644
2645 if (nolock)
2646 trans = btrfs_join_transaction_nolock(root);
2647 else
2648 trans = btrfs_join_transaction(root);
2649 if (IS_ERR(trans)) {
2650 ret = PTR_ERR(trans);
2651 trans = NULL;
2652 goto out_unlock;
2653 }
2654 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2655
2656 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2657 compress_type = ordered_extent->compress_type;
2658 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2659 BUG_ON(compress_type);
2660 ret = btrfs_mark_extent_written(trans, inode,
2661 ordered_extent->file_offset,
2662 ordered_extent->file_offset +
2663 logical_len);
2664 } else {
2665 BUG_ON(root == root->fs_info->tree_root);
2666 ret = insert_reserved_file_extent(trans, inode,
2667 ordered_extent->file_offset,
2668 ordered_extent->start,
2669 ordered_extent->disk_len,
2670 logical_len, logical_len,
2671 compress_type, 0, 0,
2672 BTRFS_FILE_EXTENT_REG);
2673 }
2674 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2675 ordered_extent->file_offset, ordered_extent->len,
2676 trans->transid);
2677 if (ret < 0) {
2678 btrfs_abort_transaction(trans, root, ret);
2679 goto out_unlock;
2680 }
2681
2682 add_pending_csums(trans, inode, ordered_extent->file_offset,
2683 &ordered_extent->list);
2684
2685 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2686 ret = btrfs_update_inode_fallback(trans, root, inode);
2687 if (ret) { /* -ENOMEM or corruption */
2688 btrfs_abort_transaction(trans, root, ret);
2689 goto out_unlock;
2690 }
2691 ret = 0;
2692 out_unlock:
2693 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2694 ordered_extent->file_offset +
2695 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2696 out:
2697 if (root != root->fs_info->tree_root)
2698 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2699 if (trans)
2700 btrfs_end_transaction(trans, root);
2701
2702 if (ret || truncated) {
2703 u64 start, end;
2704
2705 if (truncated)
2706 start = ordered_extent->file_offset + logical_len;
2707 else
2708 start = ordered_extent->file_offset;
2709 end = ordered_extent->file_offset + ordered_extent->len - 1;
2710 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2711
2712 /* Drop the cache for the part of the extent we didn't write. */
2713 btrfs_drop_extent_cache(inode, start, end, 0);
2714
2715 /*
2716 * If the ordered extent had an IOERR or something else went
2717 * wrong we need to return the space for this ordered extent
2718 * back to the allocator. We only free the extent in the
2719 * truncated case if we didn't write out the extent at all.
2720 */
2721 if ((ret || !logical_len) &&
2722 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2723 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2724 btrfs_free_reserved_extent(root, ordered_extent->start,
2725 ordered_extent->disk_len);
2726 }
2727
2728
2729 /*
2730 * This needs to be done to make sure anybody waiting knows we are done
2731 * updating everything for this ordered extent.
2732 */
2733 btrfs_remove_ordered_extent(inode, ordered_extent);
2734
2735 /* for snapshot-aware defrag */
2736 if (new) {
2737 if (ret) {
2738 free_sa_defrag_extent(new);
2739 atomic_dec(&root->fs_info->defrag_running);
2740 } else {
2741 relink_file_extents(new);
2742 }
2743 }
2744
2745 /* once for us */
2746 btrfs_put_ordered_extent(ordered_extent);
2747 /* once for the tree */
2748 btrfs_put_ordered_extent(ordered_extent);
2749
2750 return ret;
2751 }
2752
2753 static void finish_ordered_fn(struct btrfs_work *work)
2754 {
2755 struct btrfs_ordered_extent *ordered_extent;
2756 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2757 btrfs_finish_ordered_io(ordered_extent);
2758 }
2759
2760 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2761 struct extent_state *state, int uptodate)
2762 {
2763 struct inode *inode = page->mapping->host;
2764 struct btrfs_root *root = BTRFS_I(inode)->root;
2765 struct btrfs_ordered_extent *ordered_extent = NULL;
2766 struct btrfs_workqueue *workers;
2767
2768 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2769
2770 ClearPagePrivate2(page);
2771 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2772 end - start + 1, uptodate))
2773 return 0;
2774
2775 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2776
2777 if (btrfs_is_free_space_inode(inode))
2778 workers = root->fs_info->endio_freespace_worker;
2779 else
2780 workers = root->fs_info->endio_write_workers;
2781 btrfs_queue_work(workers, &ordered_extent->work);
2782
2783 return 0;
2784 }
2785
2786 /*
2787 * when reads are done, we need to check csums to verify the data is correct
2788 * if there's a match, we allow the bio to finish. If not, the code in
2789 * extent_io.c will try to find good copies for us.
2790 */
2791 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2792 u64 phy_offset, struct page *page,
2793 u64 start, u64 end, int mirror)
2794 {
2795 size_t offset = start - page_offset(page);
2796 struct inode *inode = page->mapping->host;
2797 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2798 char *kaddr;
2799 struct btrfs_root *root = BTRFS_I(inode)->root;
2800 u32 csum_expected;
2801 u32 csum = ~(u32)0;
2802 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2803 DEFAULT_RATELIMIT_BURST);
2804
2805 if (PageChecked(page)) {
2806 ClearPageChecked(page);
2807 goto good;
2808 }
2809
2810 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2811 goto good;
2812
2813 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2814 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2815 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2816 GFP_NOFS);
2817 return 0;
2818 }
2819
2820 phy_offset >>= inode->i_sb->s_blocksize_bits;
2821 csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2822
2823 kaddr = kmap_atomic(page);
2824 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
2825 btrfs_csum_final(csum, (char *)&csum);
2826 if (csum != csum_expected)
2827 goto zeroit;
2828
2829 kunmap_atomic(kaddr);
2830 good:
2831 return 0;
2832
2833 zeroit:
2834 if (__ratelimit(&_rs))
2835 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2836 btrfs_ino(page->mapping->host), start, csum, csum_expected);
2837 memset(kaddr + offset, 1, end - start + 1);
2838 flush_dcache_page(page);
2839 kunmap_atomic(kaddr);
2840 if (csum_expected == 0)
2841 return 0;
2842 return -EIO;
2843 }
2844
2845 struct delayed_iput {
2846 struct list_head list;
2847 struct inode *inode;
2848 };
2849
2850 /* JDM: If this is fs-wide, why can't we add a pointer to
2851 * btrfs_inode instead and avoid the allocation? */
2852 void btrfs_add_delayed_iput(struct inode *inode)
2853 {
2854 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2855 struct delayed_iput *delayed;
2856
2857 if (atomic_add_unless(&inode->i_count, -1, 1))
2858 return;
2859
2860 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2861 delayed->inode = inode;
2862
2863 spin_lock(&fs_info->delayed_iput_lock);
2864 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2865 spin_unlock(&fs_info->delayed_iput_lock);
2866 }
2867
2868 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2869 {
2870 LIST_HEAD(list);
2871 struct btrfs_fs_info *fs_info = root->fs_info;
2872 struct delayed_iput *delayed;
2873 int empty;
2874
2875 spin_lock(&fs_info->delayed_iput_lock);
2876 empty = list_empty(&fs_info->delayed_iputs);
2877 spin_unlock(&fs_info->delayed_iput_lock);
2878 if (empty)
2879 return;
2880
2881 spin_lock(&fs_info->delayed_iput_lock);
2882 list_splice_init(&fs_info->delayed_iputs, &list);
2883 spin_unlock(&fs_info->delayed_iput_lock);
2884
2885 while (!list_empty(&list)) {
2886 delayed = list_entry(list.next, struct delayed_iput, list);
2887 list_del(&delayed->list);
2888 iput(delayed->inode);
2889 kfree(delayed);
2890 }
2891 }
2892
2893 /*
2894 * This is called in transaction commit time. If there are no orphan
2895 * files in the subvolume, it removes orphan item and frees block_rsv
2896 * structure.
2897 */
2898 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2899 struct btrfs_root *root)
2900 {
2901 struct btrfs_block_rsv *block_rsv;
2902 int ret;
2903
2904 if (atomic_read(&root->orphan_inodes) ||
2905 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2906 return;
2907
2908 spin_lock(&root->orphan_lock);
2909 if (atomic_read(&root->orphan_inodes)) {
2910 spin_unlock(&root->orphan_lock);
2911 return;
2912 }
2913
2914 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2915 spin_unlock(&root->orphan_lock);
2916 return;
2917 }
2918
2919 block_rsv = root->orphan_block_rsv;
2920 root->orphan_block_rsv = NULL;
2921 spin_unlock(&root->orphan_lock);
2922
2923 if (root->orphan_item_inserted &&
2924 btrfs_root_refs(&root->root_item) > 0) {
2925 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2926 root->root_key.objectid);
2927 if (ret)
2928 btrfs_abort_transaction(trans, root, ret);
2929 else
2930 root->orphan_item_inserted = 0;
2931 }
2932
2933 if (block_rsv) {
2934 WARN_ON(block_rsv->size > 0);
2935 btrfs_free_block_rsv(root, block_rsv);
2936 }
2937 }
2938
2939 /*
2940 * This creates an orphan entry for the given inode in case something goes
2941 * wrong in the middle of an unlink/truncate.
2942 *
2943 * NOTE: caller of this function should reserve 5 units of metadata for
2944 * this function.
2945 */
2946 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2947 {
2948 struct btrfs_root *root = BTRFS_I(inode)->root;
2949 struct btrfs_block_rsv *block_rsv = NULL;
2950 int reserve = 0;
2951 int insert = 0;
2952 int ret;
2953
2954 if (!root->orphan_block_rsv) {
2955 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2956 if (!block_rsv)
2957 return -ENOMEM;
2958 }
2959
2960 spin_lock(&root->orphan_lock);
2961 if (!root->orphan_block_rsv) {
2962 root->orphan_block_rsv = block_rsv;
2963 } else if (block_rsv) {
2964 btrfs_free_block_rsv(root, block_rsv);
2965 block_rsv = NULL;
2966 }
2967
2968 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2969 &BTRFS_I(inode)->runtime_flags)) {
2970 #if 0
2971 /*
2972 * For proper ENOSPC handling, we should do orphan
2973 * cleanup when mounting. But this introduces backward
2974 * compatibility issue.
2975 */
2976 if (!xchg(&root->orphan_item_inserted, 1))
2977 insert = 2;
2978 else
2979 insert = 1;
2980 #endif
2981 insert = 1;
2982 atomic_inc(&root->orphan_inodes);
2983 }
2984
2985 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2986 &BTRFS_I(inode)->runtime_flags))
2987 reserve = 1;
2988 spin_unlock(&root->orphan_lock);
2989
2990 /* grab metadata reservation from transaction handle */
2991 if (reserve) {
2992 ret = btrfs_orphan_reserve_metadata(trans, inode);
2993 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2994 }
2995
2996 /* insert an orphan item to track this unlinked/truncated file */
2997 if (insert >= 1) {
2998 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2999 if (ret) {
3000 atomic_dec(&root->orphan_inodes);
3001 if (reserve) {
3002 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3003 &BTRFS_I(inode)->runtime_flags);
3004 btrfs_orphan_release_metadata(inode);
3005 }
3006 if (ret != -EEXIST) {
3007 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3008 &BTRFS_I(inode)->runtime_flags);
3009 btrfs_abort_transaction(trans, root, ret);
3010 return ret;
3011 }
3012 }
3013 ret = 0;
3014 }
3015
3016 /* insert an orphan item to track subvolume contains orphan files */
3017 if (insert >= 2) {
3018 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3019 root->root_key.objectid);
3020 if (ret && ret != -EEXIST) {
3021 btrfs_abort_transaction(trans, root, ret);
3022 return ret;
3023 }
3024 }
3025 return 0;
3026 }
3027
3028 /*
3029 * We have done the truncate/delete so we can go ahead and remove the orphan
3030 * item for this particular inode.
3031 */
3032 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3033 struct inode *inode)
3034 {
3035 struct btrfs_root *root = BTRFS_I(inode)->root;
3036 int delete_item = 0;
3037 int release_rsv = 0;
3038 int ret = 0;
3039
3040 spin_lock(&root->orphan_lock);
3041 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3042 &BTRFS_I(inode)->runtime_flags))
3043 delete_item = 1;
3044
3045 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3046 &BTRFS_I(inode)->runtime_flags))
3047 release_rsv = 1;
3048 spin_unlock(&root->orphan_lock);
3049
3050 if (delete_item) {
3051 atomic_dec(&root->orphan_inodes);
3052 if (trans)
3053 ret = btrfs_del_orphan_item(trans, root,
3054 btrfs_ino(inode));
3055 }
3056
3057 if (release_rsv)
3058 btrfs_orphan_release_metadata(inode);
3059
3060 return ret;
3061 }
3062
3063 /*
3064 * this cleans up any orphans that may be left on the list from the last use
3065 * of this root.
3066 */
3067 int btrfs_orphan_cleanup(struct btrfs_root *root)
3068 {
3069 struct btrfs_path *path;
3070 struct extent_buffer *leaf;
3071 struct btrfs_key key, found_key;
3072 struct btrfs_trans_handle *trans;
3073 struct inode *inode;
3074 u64 last_objectid = 0;
3075 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3076
3077 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3078 return 0;
3079
3080 path = btrfs_alloc_path();
3081 if (!path) {
3082 ret = -ENOMEM;
3083 goto out;
3084 }
3085 path->reada = -1;
3086
3087 key.objectid = BTRFS_ORPHAN_OBJECTID;
3088 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3089 key.offset = (u64)-1;
3090
3091 while (1) {
3092 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3093 if (ret < 0)
3094 goto out;
3095
3096 /*
3097 * if ret == 0 means we found what we were searching for, which
3098 * is weird, but possible, so only screw with path if we didn't
3099 * find the key and see if we have stuff that matches
3100 */
3101 if (ret > 0) {
3102 ret = 0;
3103 if (path->slots[0] == 0)
3104 break;
3105 path->slots[0]--;
3106 }
3107
3108 /* pull out the item */
3109 leaf = path->nodes[0];
3110 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3111
3112 /* make sure the item matches what we want */
3113 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3114 break;
3115 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3116 break;
3117
3118 /* release the path since we're done with it */
3119 btrfs_release_path(path);
3120
3121 /*
3122 * this is where we are basically btrfs_lookup, without the
3123 * crossing root thing. we store the inode number in the
3124 * offset of the orphan item.
3125 */
3126
3127 if (found_key.offset == last_objectid) {
3128 btrfs_err(root->fs_info,
3129 "Error removing orphan entry, stopping orphan cleanup");
3130 ret = -EINVAL;
3131 goto out;
3132 }
3133
3134 last_objectid = found_key.offset;
3135
3136 found_key.objectid = found_key.offset;
3137 found_key.type = BTRFS_INODE_ITEM_KEY;
3138 found_key.offset = 0;
3139 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3140 ret = PTR_ERR_OR_ZERO(inode);
3141 if (ret && ret != -ESTALE)
3142 goto out;
3143
3144 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3145 struct btrfs_root *dead_root;
3146 struct btrfs_fs_info *fs_info = root->fs_info;
3147 int is_dead_root = 0;
3148
3149 /*
3150 * this is an orphan in the tree root. Currently these
3151 * could come from 2 sources:
3152 * a) a snapshot deletion in progress
3153 * b) a free space cache inode
3154 * We need to distinguish those two, as the snapshot
3155 * orphan must not get deleted.
3156 * find_dead_roots already ran before us, so if this
3157 * is a snapshot deletion, we should find the root
3158 * in the dead_roots list
3159 */
3160 spin_lock(&fs_info->trans_lock);
3161 list_for_each_entry(dead_root, &fs_info->dead_roots,
3162 root_list) {
3163 if (dead_root->root_key.objectid ==
3164 found_key.objectid) {
3165 is_dead_root = 1;
3166 break;
3167 }
3168 }
3169 spin_unlock(&fs_info->trans_lock);
3170 if (is_dead_root) {
3171 /* prevent this orphan from being found again */
3172 key.offset = found_key.objectid - 1;
3173 continue;
3174 }
3175 }
3176 /*
3177 * Inode is already gone but the orphan item is still there,
3178 * kill the orphan item.
3179 */
3180 if (ret == -ESTALE) {
3181 trans = btrfs_start_transaction(root, 1);
3182 if (IS_ERR(trans)) {
3183 ret = PTR_ERR(trans);
3184 goto out;
3185 }
3186 btrfs_debug(root->fs_info, "auto deleting %Lu",
3187 found_key.objectid);
3188 ret = btrfs_del_orphan_item(trans, root,
3189 found_key.objectid);
3190 btrfs_end_transaction(trans, root);
3191 if (ret)
3192 goto out;
3193 continue;
3194 }
3195
3196 /*
3197 * add this inode to the orphan list so btrfs_orphan_del does
3198 * the proper thing when we hit it
3199 */
3200 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3201 &BTRFS_I(inode)->runtime_flags);
3202 atomic_inc(&root->orphan_inodes);
3203
3204 /* if we have links, this was a truncate, lets do that */
3205 if (inode->i_nlink) {
3206 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3207 iput(inode);
3208 continue;
3209 }
3210 nr_truncate++;
3211
3212 /* 1 for the orphan item deletion. */
3213 trans = btrfs_start_transaction(root, 1);
3214 if (IS_ERR(trans)) {
3215 iput(inode);
3216 ret = PTR_ERR(trans);
3217 goto out;
3218 }
3219 ret = btrfs_orphan_add(trans, inode);
3220 btrfs_end_transaction(trans, root);
3221 if (ret) {
3222 iput(inode);
3223 goto out;
3224 }
3225
3226 ret = btrfs_truncate(inode);
3227 if (ret)
3228 btrfs_orphan_del(NULL, inode);
3229 } else {
3230 nr_unlink++;
3231 }
3232
3233 /* this will do delete_inode and everything for us */
3234 iput(inode);
3235 if (ret)
3236 goto out;
3237 }
3238 /* release the path since we're done with it */
3239 btrfs_release_path(path);
3240
3241 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3242
3243 if (root->orphan_block_rsv)
3244 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3245 (u64)-1);
3246
3247 if (root->orphan_block_rsv || root->orphan_item_inserted) {
3248 trans = btrfs_join_transaction(root);
3249 if (!IS_ERR(trans))
3250 btrfs_end_transaction(trans, root);
3251 }
3252
3253 if (nr_unlink)
3254 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3255 if (nr_truncate)
3256 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3257
3258 out:
3259 if (ret)
3260 btrfs_crit(root->fs_info,
3261 "could not do orphan cleanup %d", ret);
3262 btrfs_free_path(path);
3263 return ret;
3264 }
3265
3266 /*
3267 * very simple check to peek ahead in the leaf looking for xattrs. If we
3268 * don't find any xattrs, we know there can't be any acls.
3269 *
3270 * slot is the slot the inode is in, objectid is the objectid of the inode
3271 */
3272 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3273 int slot, u64 objectid,
3274 int *first_xattr_slot)
3275 {
3276 u32 nritems = btrfs_header_nritems(leaf);
3277 struct btrfs_key found_key;
3278 static u64 xattr_access = 0;
3279 static u64 xattr_default = 0;
3280 int scanned = 0;
3281
3282 if (!xattr_access) {
3283 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3284 strlen(POSIX_ACL_XATTR_ACCESS));
3285 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3286 strlen(POSIX_ACL_XATTR_DEFAULT));
3287 }
3288
3289 slot++;
3290 *first_xattr_slot = -1;
3291 while (slot < nritems) {
3292 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3293
3294 /* we found a different objectid, there must not be acls */
3295 if (found_key.objectid != objectid)
3296 return 0;
3297
3298 /* we found an xattr, assume we've got an acl */
3299 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3300 if (*first_xattr_slot == -1)
3301 *first_xattr_slot = slot;
3302 if (found_key.offset == xattr_access ||
3303 found_key.offset == xattr_default)
3304 return 1;
3305 }
3306
3307 /*
3308 * we found a key greater than an xattr key, there can't
3309 * be any acls later on
3310 */
3311 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3312 return 0;
3313
3314 slot++;
3315 scanned++;
3316
3317 /*
3318 * it goes inode, inode backrefs, xattrs, extents,
3319 * so if there are a ton of hard links to an inode there can
3320 * be a lot of backrefs. Don't waste time searching too hard,
3321 * this is just an optimization
3322 */
3323 if (scanned >= 8)
3324 break;
3325 }
3326 /* we hit the end of the leaf before we found an xattr or
3327 * something larger than an xattr. We have to assume the inode
3328 * has acls
3329 */
3330 if (*first_xattr_slot == -1)
3331 *first_xattr_slot = slot;
3332 return 1;
3333 }
3334
3335 /*
3336 * read an inode from the btree into the in-memory inode
3337 */
3338 static void btrfs_read_locked_inode(struct inode *inode)
3339 {
3340 struct btrfs_path *path;
3341 struct extent_buffer *leaf;
3342 struct btrfs_inode_item *inode_item;
3343 struct btrfs_timespec *tspec;
3344 struct btrfs_root *root = BTRFS_I(inode)->root;
3345 struct btrfs_key location;
3346 unsigned long ptr;
3347 int maybe_acls;
3348 u32 rdev;
3349 int ret;
3350 bool filled = false;
3351 int first_xattr_slot;
3352
3353 ret = btrfs_fill_inode(inode, &rdev);
3354 if (!ret)
3355 filled = true;
3356
3357 path = btrfs_alloc_path();
3358 if (!path)
3359 goto make_bad;
3360
3361 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3362
3363 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3364 if (ret)
3365 goto make_bad;
3366
3367 leaf = path->nodes[0];
3368
3369 if (filled)
3370 goto cache_index;
3371
3372 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3373 struct btrfs_inode_item);
3374 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3375 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3376 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3377 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3378 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3379
3380 tspec = btrfs_inode_atime(inode_item);
3381 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3382 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3383
3384 tspec = btrfs_inode_mtime(inode_item);
3385 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3386 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3387
3388 tspec = btrfs_inode_ctime(inode_item);
3389 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3390 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3391
3392 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3393 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3394 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3395
3396 /*
3397 * If we were modified in the current generation and evicted from memory
3398 * and then re-read we need to do a full sync since we don't have any
3399 * idea about which extents were modified before we were evicted from
3400 * cache.
3401 */
3402 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3403 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3404 &BTRFS_I(inode)->runtime_flags);
3405
3406 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3407 inode->i_generation = BTRFS_I(inode)->generation;
3408 inode->i_rdev = 0;
3409 rdev = btrfs_inode_rdev(leaf, inode_item);
3410
3411 BTRFS_I(inode)->index_cnt = (u64)-1;
3412 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3413
3414 cache_index:
3415 path->slots[0]++;
3416 if (inode->i_nlink != 1 ||
3417 path->slots[0] >= btrfs_header_nritems(leaf))
3418 goto cache_acl;
3419
3420 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3421 if (location.objectid != btrfs_ino(inode))
3422 goto cache_acl;
3423
3424 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3425 if (location.type == BTRFS_INODE_REF_KEY) {
3426 struct btrfs_inode_ref *ref;
3427
3428 ref = (struct btrfs_inode_ref *)ptr;
3429 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3430 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3431 struct btrfs_inode_extref *extref;
3432
3433 extref = (struct btrfs_inode_extref *)ptr;
3434 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3435 extref);
3436 }
3437 cache_acl:
3438 /*
3439 * try to precache a NULL acl entry for files that don't have
3440 * any xattrs or acls
3441 */
3442 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3443 btrfs_ino(inode), &first_xattr_slot);
3444 if (first_xattr_slot != -1) {
3445 path->slots[0] = first_xattr_slot;
3446 ret = btrfs_load_inode_props(inode, path);
3447 if (ret)
3448 btrfs_err(root->fs_info,
3449 "error loading props for ino %llu (root %llu): %d\n",
3450 btrfs_ino(inode),
3451 root->root_key.objectid, ret);
3452 }
3453 btrfs_free_path(path);
3454
3455 if (!maybe_acls)
3456 cache_no_acl(inode);
3457
3458 switch (inode->i_mode & S_IFMT) {
3459 case S_IFREG:
3460 inode->i_mapping->a_ops = &btrfs_aops;
3461 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3462 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3463 inode->i_fop = &btrfs_file_operations;
3464 inode->i_op = &btrfs_file_inode_operations;
3465 break;
3466 case S_IFDIR:
3467 inode->i_fop = &btrfs_dir_file_operations;
3468 if (root == root->fs_info->tree_root)
3469 inode->i_op = &btrfs_dir_ro_inode_operations;
3470 else
3471 inode->i_op = &btrfs_dir_inode_operations;
3472 break;
3473 case S_IFLNK:
3474 inode->i_op = &btrfs_symlink_inode_operations;
3475 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3476 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3477 break;
3478 default:
3479 inode->i_op = &btrfs_special_inode_operations;
3480 init_special_inode(inode, inode->i_mode, rdev);
3481 break;
3482 }
3483
3484 btrfs_update_iflags(inode);
3485 return;
3486
3487 make_bad:
3488 btrfs_free_path(path);
3489 make_bad_inode(inode);
3490 }
3491
3492 /*
3493 * given a leaf and an inode, copy the inode fields into the leaf
3494 */
3495 static void fill_inode_item(struct btrfs_trans_handle *trans,
3496 struct extent_buffer *leaf,
3497 struct btrfs_inode_item *item,
3498 struct inode *inode)
3499 {
3500 struct btrfs_map_token token;
3501
3502 btrfs_init_map_token(&token);
3503
3504 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3505 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3506 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3507 &token);
3508 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3509 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3510
3511 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3512 inode->i_atime.tv_sec, &token);
3513 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3514 inode->i_atime.tv_nsec, &token);
3515
3516 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3517 inode->i_mtime.tv_sec, &token);
3518 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3519 inode->i_mtime.tv_nsec, &token);
3520
3521 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3522 inode->i_ctime.tv_sec, &token);
3523 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3524 inode->i_ctime.tv_nsec, &token);
3525
3526 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3527 &token);
3528 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3529 &token);
3530 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3531 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3532 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3533 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3534 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3535 }
3536
3537 /*
3538 * copy everything in the in-memory inode into the btree.
3539 */
3540 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3541 struct btrfs_root *root, struct inode *inode)
3542 {
3543 struct btrfs_inode_item *inode_item;
3544 struct btrfs_path *path;
3545 struct extent_buffer *leaf;
3546 int ret;
3547
3548 path = btrfs_alloc_path();
3549 if (!path)
3550 return -ENOMEM;
3551
3552 path->leave_spinning = 1;
3553 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3554 1);
3555 if (ret) {
3556 if (ret > 0)
3557 ret = -ENOENT;
3558 goto failed;
3559 }
3560
3561 leaf = path->nodes[0];
3562 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3563 struct btrfs_inode_item);
3564
3565 fill_inode_item(trans, leaf, inode_item, inode);
3566 btrfs_mark_buffer_dirty(leaf);
3567 btrfs_set_inode_last_trans(trans, inode);
3568 ret = 0;
3569 failed:
3570 btrfs_free_path(path);
3571 return ret;
3572 }
3573
3574 /*
3575 * copy everything in the in-memory inode into the btree.
3576 */
3577 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3578 struct btrfs_root *root, struct inode *inode)
3579 {
3580 int ret;
3581
3582 /*
3583 * If the inode is a free space inode, we can deadlock during commit
3584 * if we put it into the delayed code.
3585 *
3586 * The data relocation inode should also be directly updated
3587 * without delay
3588 */
3589 if (!btrfs_is_free_space_inode(inode)
3590 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3591 btrfs_update_root_times(trans, root);
3592
3593 ret = btrfs_delayed_update_inode(trans, root, inode);
3594 if (!ret)
3595 btrfs_set_inode_last_trans(trans, inode);
3596 return ret;
3597 }
3598
3599 return btrfs_update_inode_item(trans, root, inode);
3600 }
3601
3602 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3603 struct btrfs_root *root,
3604 struct inode *inode)
3605 {
3606 int ret;
3607
3608 ret = btrfs_update_inode(trans, root, inode);
3609 if (ret == -ENOSPC)
3610 return btrfs_update_inode_item(trans, root, inode);
3611 return ret;
3612 }
3613
3614 /*
3615 * unlink helper that gets used here in inode.c and in the tree logging
3616 * recovery code. It remove a link in a directory with a given name, and
3617 * also drops the back refs in the inode to the directory
3618 */
3619 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3620 struct btrfs_root *root,
3621 struct inode *dir, struct inode *inode,
3622 const char *name, int name_len)
3623 {
3624 struct btrfs_path *path;
3625 int ret = 0;
3626 struct extent_buffer *leaf;
3627 struct btrfs_dir_item *di;
3628 struct btrfs_key key;
3629 u64 index;
3630 u64 ino = btrfs_ino(inode);
3631 u64 dir_ino = btrfs_ino(dir);
3632
3633 path = btrfs_alloc_path();
3634 if (!path) {
3635 ret = -ENOMEM;
3636 goto out;
3637 }
3638
3639 path->leave_spinning = 1;
3640 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3641 name, name_len, -1);
3642 if (IS_ERR(di)) {
3643 ret = PTR_ERR(di);
3644 goto err;
3645 }
3646 if (!di) {
3647 ret = -ENOENT;
3648 goto err;
3649 }
3650 leaf = path->nodes[0];
3651 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3652 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3653 if (ret)
3654 goto err;
3655 btrfs_release_path(path);
3656
3657 /*
3658 * If we don't have dir index, we have to get it by looking up
3659 * the inode ref, since we get the inode ref, remove it directly,
3660 * it is unnecessary to do delayed deletion.
3661 *
3662 * But if we have dir index, needn't search inode ref to get it.
3663 * Since the inode ref is close to the inode item, it is better
3664 * that we delay to delete it, and just do this deletion when
3665 * we update the inode item.
3666 */
3667 if (BTRFS_I(inode)->dir_index) {
3668 ret = btrfs_delayed_delete_inode_ref(inode);
3669 if (!ret) {
3670 index = BTRFS_I(inode)->dir_index;
3671 goto skip_backref;
3672 }
3673 }
3674
3675 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3676 dir_ino, &index);
3677 if (ret) {
3678 btrfs_info(root->fs_info,
3679 "failed to delete reference to %.*s, inode %llu parent %llu",
3680 name_len, name, ino, dir_ino);
3681 btrfs_abort_transaction(trans, root, ret);
3682 goto err;
3683 }
3684 skip_backref:
3685 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3686 if (ret) {
3687 btrfs_abort_transaction(trans, root, ret);
3688 goto err;
3689 }
3690
3691 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3692 inode, dir_ino);
3693 if (ret != 0 && ret != -ENOENT) {
3694 btrfs_abort_transaction(trans, root, ret);
3695 goto err;
3696 }
3697
3698 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3699 dir, index);
3700 if (ret == -ENOENT)
3701 ret = 0;
3702 else if (ret)
3703 btrfs_abort_transaction(trans, root, ret);
3704 err:
3705 btrfs_free_path(path);
3706 if (ret)
3707 goto out;
3708
3709 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3710 inode_inc_iversion(inode);
3711 inode_inc_iversion(dir);
3712 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3713 ret = btrfs_update_inode(trans, root, dir);
3714 out:
3715 return ret;
3716 }
3717
3718 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3719 struct btrfs_root *root,
3720 struct inode *dir, struct inode *inode,
3721 const char *name, int name_len)
3722 {
3723 int ret;
3724 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3725 if (!ret) {
3726 drop_nlink(inode);
3727 ret = btrfs_update_inode(trans, root, inode);
3728 }
3729 return ret;
3730 }
3731
3732 /*
3733 * helper to start transaction for unlink and rmdir.
3734 *
3735 * unlink and rmdir are special in btrfs, they do not always free space, so
3736 * if we cannot make our reservations the normal way try and see if there is
3737 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3738 * allow the unlink to occur.
3739 */
3740 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3741 {
3742 struct btrfs_trans_handle *trans;
3743 struct btrfs_root *root = BTRFS_I(dir)->root;
3744 int ret;
3745
3746 /*
3747 * 1 for the possible orphan item
3748 * 1 for the dir item
3749 * 1 for the dir index
3750 * 1 for the inode ref
3751 * 1 for the inode
3752 */
3753 trans = btrfs_start_transaction(root, 5);
3754 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3755 return trans;
3756
3757 if (PTR_ERR(trans) == -ENOSPC) {
3758 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3759
3760 trans = btrfs_start_transaction(root, 0);
3761 if (IS_ERR(trans))
3762 return trans;
3763 ret = btrfs_cond_migrate_bytes(root->fs_info,
3764 &root->fs_info->trans_block_rsv,
3765 num_bytes, 5);
3766 if (ret) {
3767 btrfs_end_transaction(trans, root);
3768 return ERR_PTR(ret);
3769 }
3770 trans->block_rsv = &root->fs_info->trans_block_rsv;
3771 trans->bytes_reserved = num_bytes;
3772 }
3773 return trans;
3774 }
3775
3776 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3777 {
3778 struct btrfs_root *root = BTRFS_I(dir)->root;
3779 struct btrfs_trans_handle *trans;
3780 struct inode *inode = dentry->d_inode;
3781 int ret;
3782
3783 trans = __unlink_start_trans(dir);
3784 if (IS_ERR(trans))
3785 return PTR_ERR(trans);
3786
3787 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3788
3789 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3790 dentry->d_name.name, dentry->d_name.len);
3791 if (ret)
3792 goto out;
3793
3794 if (inode->i_nlink == 0) {
3795 ret = btrfs_orphan_add(trans, inode);
3796 if (ret)
3797 goto out;
3798 }
3799
3800 out:
3801 btrfs_end_transaction(trans, root);
3802 btrfs_btree_balance_dirty(root);
3803 return ret;
3804 }
3805
3806 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3807 struct btrfs_root *root,
3808 struct inode *dir, u64 objectid,
3809 const char *name, int name_len)
3810 {
3811 struct btrfs_path *path;
3812 struct extent_buffer *leaf;
3813 struct btrfs_dir_item *di;
3814 struct btrfs_key key;
3815 u64 index;
3816 int ret;
3817 u64 dir_ino = btrfs_ino(dir);
3818
3819 path = btrfs_alloc_path();
3820 if (!path)
3821 return -ENOMEM;
3822
3823 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3824 name, name_len, -1);
3825 if (IS_ERR_OR_NULL(di)) {
3826 if (!di)
3827 ret = -ENOENT;
3828 else
3829 ret = PTR_ERR(di);
3830 goto out;
3831 }
3832
3833 leaf = path->nodes[0];
3834 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3835 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3836 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3837 if (ret) {
3838 btrfs_abort_transaction(trans, root, ret);
3839 goto out;
3840 }
3841 btrfs_release_path(path);
3842
3843 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3844 objectid, root->root_key.objectid,
3845 dir_ino, &index, name, name_len);
3846 if (ret < 0) {
3847 if (ret != -ENOENT) {
3848 btrfs_abort_transaction(trans, root, ret);
3849 goto out;
3850 }
3851 di = btrfs_search_dir_index_item(root, path, dir_ino,
3852 name, name_len);
3853 if (IS_ERR_OR_NULL(di)) {
3854 if (!di)
3855 ret = -ENOENT;
3856 else
3857 ret = PTR_ERR(di);
3858 btrfs_abort_transaction(trans, root, ret);
3859 goto out;
3860 }
3861
3862 leaf = path->nodes[0];
3863 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3864 btrfs_release_path(path);
3865 index = key.offset;
3866 }
3867 btrfs_release_path(path);
3868
3869 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3870 if (ret) {
3871 btrfs_abort_transaction(trans, root, ret);
3872 goto out;
3873 }
3874
3875 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3876 inode_inc_iversion(dir);
3877 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3878 ret = btrfs_update_inode_fallback(trans, root, dir);
3879 if (ret)
3880 btrfs_abort_transaction(trans, root, ret);
3881 out:
3882 btrfs_free_path(path);
3883 return ret;
3884 }
3885
3886 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3887 {
3888 struct inode *inode = dentry->d_inode;
3889 int err = 0;
3890 struct btrfs_root *root = BTRFS_I(dir)->root;
3891 struct btrfs_trans_handle *trans;
3892
3893 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3894 return -ENOTEMPTY;
3895 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3896 return -EPERM;
3897
3898 trans = __unlink_start_trans(dir);
3899 if (IS_ERR(trans))
3900 return PTR_ERR(trans);
3901
3902 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3903 err = btrfs_unlink_subvol(trans, root, dir,
3904 BTRFS_I(inode)->location.objectid,
3905 dentry->d_name.name,
3906 dentry->d_name.len);
3907 goto out;
3908 }
3909
3910 err = btrfs_orphan_add(trans, inode);
3911 if (err)
3912 goto out;
3913
3914 /* now the directory is empty */
3915 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3916 dentry->d_name.name, dentry->d_name.len);
3917 if (!err)
3918 btrfs_i_size_write(inode, 0);
3919 out:
3920 btrfs_end_transaction(trans, root);
3921 btrfs_btree_balance_dirty(root);
3922
3923 return err;
3924 }
3925
3926 /*
3927 * this can truncate away extent items, csum items and directory items.
3928 * It starts at a high offset and removes keys until it can't find
3929 * any higher than new_size
3930 *
3931 * csum items that cross the new i_size are truncated to the new size
3932 * as well.
3933 *
3934 * min_type is the minimum key type to truncate down to. If set to 0, this
3935 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3936 */
3937 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3938 struct btrfs_root *root,
3939 struct inode *inode,
3940 u64 new_size, u32 min_type)
3941 {
3942 struct btrfs_path *path;
3943 struct extent_buffer *leaf;
3944 struct btrfs_file_extent_item *fi;
3945 struct btrfs_key key;
3946 struct btrfs_key found_key;
3947 u64 extent_start = 0;
3948 u64 extent_num_bytes = 0;
3949 u64 extent_offset = 0;
3950 u64 item_end = 0;
3951 u64 last_size = (u64)-1;
3952 u32 found_type = (u8)-1;
3953 int found_extent;
3954 int del_item;
3955 int pending_del_nr = 0;
3956 int pending_del_slot = 0;
3957 int extent_type = -1;
3958 int ret;
3959 int err = 0;
3960 u64 ino = btrfs_ino(inode);
3961
3962 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3963
3964 path = btrfs_alloc_path();
3965 if (!path)
3966 return -ENOMEM;
3967 path->reada = -1;
3968
3969 /*
3970 * We want to drop from the next block forward in case this new size is
3971 * not block aligned since we will be keeping the last block of the
3972 * extent just the way it is.
3973 */
3974 if (root->ref_cows || root == root->fs_info->tree_root)
3975 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3976 root->sectorsize), (u64)-1, 0);
3977
3978 /*
3979 * This function is also used to drop the items in the log tree before
3980 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3981 * it is used to drop the loged items. So we shouldn't kill the delayed
3982 * items.
3983 */
3984 if (min_type == 0 && root == BTRFS_I(inode)->root)
3985 btrfs_kill_delayed_inode_items(inode);
3986
3987 key.objectid = ino;
3988 key.offset = (u64)-1;
3989 key.type = (u8)-1;
3990
3991 search_again:
3992 path->leave_spinning = 1;
3993 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3994 if (ret < 0) {
3995 err = ret;
3996 goto out;
3997 }
3998
3999 if (ret > 0) {
4000 /* there are no items in the tree for us to truncate, we're
4001 * done
4002 */
4003 if (path->slots[0] == 0)
4004 goto out;
4005 path->slots[0]--;
4006 }
4007
4008 while (1) {
4009 fi = NULL;
4010 leaf = path->nodes[0];
4011 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4012 found_type = btrfs_key_type(&found_key);
4013
4014 if (found_key.objectid != ino)
4015 break;
4016
4017 if (found_type < min_type)
4018 break;
4019
4020 item_end = found_key.offset;
4021 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4022 fi = btrfs_item_ptr(leaf, path->slots[0],
4023 struct btrfs_file_extent_item);
4024 extent_type = btrfs_file_extent_type(leaf, fi);
4025 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4026 item_end +=
4027 btrfs_file_extent_num_bytes(leaf, fi);
4028 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4029 item_end += btrfs_file_extent_inline_len(leaf,
4030 path->slots[0], fi);
4031 }
4032 item_end--;
4033 }
4034 if (found_type > min_type) {
4035 del_item = 1;
4036 } else {
4037 if (item_end < new_size)
4038 break;
4039 if (found_key.offset >= new_size)
4040 del_item = 1;
4041 else
4042 del_item = 0;
4043 }
4044 found_extent = 0;
4045 /* FIXME, shrink the extent if the ref count is only 1 */
4046 if (found_type != BTRFS_EXTENT_DATA_KEY)
4047 goto delete;
4048
4049 if (del_item)
4050 last_size = found_key.offset;
4051 else
4052 last_size = new_size;
4053
4054 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4055 u64 num_dec;
4056 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4057 if (!del_item) {
4058 u64 orig_num_bytes =
4059 btrfs_file_extent_num_bytes(leaf, fi);
4060 extent_num_bytes = ALIGN(new_size -
4061 found_key.offset,
4062 root->sectorsize);
4063 btrfs_set_file_extent_num_bytes(leaf, fi,
4064 extent_num_bytes);
4065 num_dec = (orig_num_bytes -
4066 extent_num_bytes);
4067 if (root->ref_cows && extent_start != 0)
4068 inode_sub_bytes(inode, num_dec);
4069 btrfs_mark_buffer_dirty(leaf);
4070 } else {
4071 extent_num_bytes =
4072 btrfs_file_extent_disk_num_bytes(leaf,
4073 fi);
4074 extent_offset = found_key.offset -
4075 btrfs_file_extent_offset(leaf, fi);
4076
4077 /* FIXME blocksize != 4096 */
4078 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4079 if (extent_start != 0) {
4080 found_extent = 1;
4081 if (root->ref_cows)
4082 inode_sub_bytes(inode, num_dec);
4083 }
4084 }
4085 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4086 /*
4087 * we can't truncate inline items that have had
4088 * special encodings
4089 */
4090 if (!del_item &&
4091 btrfs_file_extent_compression(leaf, fi) == 0 &&
4092 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4093 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4094 u32 size = new_size - found_key.offset;
4095
4096 if (root->ref_cows) {
4097 inode_sub_bytes(inode, item_end + 1 -
4098 new_size);
4099 }
4100
4101 /*
4102 * update the ram bytes to properly reflect
4103 * the new size of our item
4104 */
4105 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4106 size =
4107 btrfs_file_extent_calc_inline_size(size);
4108 btrfs_truncate_item(root, path, size, 1);
4109 } else if (root->ref_cows) {
4110 inode_sub_bytes(inode, item_end + 1 -
4111 found_key.offset);
4112 }
4113 }
4114 delete:
4115 if (del_item) {
4116 if (!pending_del_nr) {
4117 /* no pending yet, add ourselves */
4118 pending_del_slot = path->slots[0];
4119 pending_del_nr = 1;
4120 } else if (pending_del_nr &&
4121 path->slots[0] + 1 == pending_del_slot) {
4122 /* hop on the pending chunk */
4123 pending_del_nr++;
4124 pending_del_slot = path->slots[0];
4125 } else {
4126 BUG();
4127 }
4128 } else {
4129 break;
4130 }
4131 if (found_extent && (root->ref_cows ||
4132 root == root->fs_info->tree_root)) {
4133 btrfs_set_path_blocking(path);
4134 ret = btrfs_free_extent(trans, root, extent_start,
4135 extent_num_bytes, 0,
4136 btrfs_header_owner(leaf),
4137 ino, extent_offset, 0);
4138 BUG_ON(ret);
4139 }
4140
4141 if (found_type == BTRFS_INODE_ITEM_KEY)
4142 break;
4143
4144 if (path->slots[0] == 0 ||
4145 path->slots[0] != pending_del_slot) {
4146 if (pending_del_nr) {
4147 ret = btrfs_del_items(trans, root, path,
4148 pending_del_slot,
4149 pending_del_nr);
4150 if (ret) {
4151 btrfs_abort_transaction(trans,
4152 root, ret);
4153 goto error;
4154 }
4155 pending_del_nr = 0;
4156 }
4157 btrfs_release_path(path);
4158 goto search_again;
4159 } else {
4160 path->slots[0]--;
4161 }
4162 }
4163 out:
4164 if (pending_del_nr) {
4165 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4166 pending_del_nr);
4167 if (ret)
4168 btrfs_abort_transaction(trans, root, ret);
4169 }
4170 error:
4171 if (last_size != (u64)-1)
4172 btrfs_ordered_update_i_size(inode, last_size, NULL);
4173 btrfs_free_path(path);
4174 return err;
4175 }
4176
4177 /*
4178 * btrfs_truncate_page - read, zero a chunk and write a page
4179 * @inode - inode that we're zeroing
4180 * @from - the offset to start zeroing
4181 * @len - the length to zero, 0 to zero the entire range respective to the
4182 * offset
4183 * @front - zero up to the offset instead of from the offset on
4184 *
4185 * This will find the page for the "from" offset and cow the page and zero the
4186 * part we want to zero. This is used with truncate and hole punching.
4187 */
4188 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4189 int front)
4190 {
4191 struct address_space *mapping = inode->i_mapping;
4192 struct btrfs_root *root = BTRFS_I(inode)->root;
4193 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4194 struct btrfs_ordered_extent *ordered;
4195 struct extent_state *cached_state = NULL;
4196 char *kaddr;
4197 u32 blocksize = root->sectorsize;
4198 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4199 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4200 struct page *page;
4201 gfp_t mask = btrfs_alloc_write_mask(mapping);
4202 int ret = 0;
4203 u64 page_start;
4204 u64 page_end;
4205
4206 if ((offset & (blocksize - 1)) == 0 &&
4207 (!len || ((len & (blocksize - 1)) == 0)))
4208 goto out;
4209 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4210 if (ret)
4211 goto out;
4212
4213 again:
4214 page = find_or_create_page(mapping, index, mask);
4215 if (!page) {
4216 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4217 ret = -ENOMEM;
4218 goto out;
4219 }
4220
4221 page_start = page_offset(page);
4222 page_end = page_start + PAGE_CACHE_SIZE - 1;
4223
4224 if (!PageUptodate(page)) {
4225 ret = btrfs_readpage(NULL, page);
4226 lock_page(page);
4227 if (page->mapping != mapping) {
4228 unlock_page(page);
4229 page_cache_release(page);
4230 goto again;
4231 }
4232 if (!PageUptodate(page)) {
4233 ret = -EIO;
4234 goto out_unlock;
4235 }
4236 }
4237 wait_on_page_writeback(page);
4238
4239 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4240 set_page_extent_mapped(page);
4241
4242 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4243 if (ordered) {
4244 unlock_extent_cached(io_tree, page_start, page_end,
4245 &cached_state, GFP_NOFS);
4246 unlock_page(page);
4247 page_cache_release(page);
4248 btrfs_start_ordered_extent(inode, ordered, 1);
4249 btrfs_put_ordered_extent(ordered);
4250 goto again;
4251 }
4252
4253 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4254 EXTENT_DIRTY | EXTENT_DELALLOC |
4255 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4256 0, 0, &cached_state, GFP_NOFS);
4257
4258 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4259 &cached_state);
4260 if (ret) {
4261 unlock_extent_cached(io_tree, page_start, page_end,
4262 &cached_state, GFP_NOFS);
4263 goto out_unlock;
4264 }
4265
4266 if (offset != PAGE_CACHE_SIZE) {
4267 if (!len)
4268 len = PAGE_CACHE_SIZE - offset;
4269 kaddr = kmap(page);
4270 if (front)
4271 memset(kaddr, 0, offset);
4272 else
4273 memset(kaddr + offset, 0, len);
4274 flush_dcache_page(page);
4275 kunmap(page);
4276 }
4277 ClearPageChecked(page);
4278 set_page_dirty(page);
4279 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4280 GFP_NOFS);
4281
4282 out_unlock:
4283 if (ret)
4284 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4285 unlock_page(page);
4286 page_cache_release(page);
4287 out:
4288 return ret;
4289 }
4290
4291 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4292 u64 offset, u64 len)
4293 {
4294 struct btrfs_trans_handle *trans;
4295 int ret;
4296
4297 /*
4298 * Still need to make sure the inode looks like it's been updated so
4299 * that any holes get logged if we fsync.
4300 */
4301 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4302 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4303 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4304 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4305 return 0;
4306 }
4307
4308 /*
4309 * 1 - for the one we're dropping
4310 * 1 - for the one we're adding
4311 * 1 - for updating the inode.
4312 */
4313 trans = btrfs_start_transaction(root, 3);
4314 if (IS_ERR(trans))
4315 return PTR_ERR(trans);
4316
4317 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4318 if (ret) {
4319 btrfs_abort_transaction(trans, root, ret);
4320 btrfs_end_transaction(trans, root);
4321 return ret;
4322 }
4323
4324 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4325 0, 0, len, 0, len, 0, 0, 0);
4326 if (ret)
4327 btrfs_abort_transaction(trans, root, ret);
4328 else
4329 btrfs_update_inode(trans, root, inode);
4330 btrfs_end_transaction(trans, root);
4331 return ret;
4332 }
4333
4334 /*
4335 * This function puts in dummy file extents for the area we're creating a hole
4336 * for. So if we are truncating this file to a larger size we need to insert
4337 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4338 * the range between oldsize and size
4339 */
4340 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4341 {
4342 struct btrfs_root *root = BTRFS_I(inode)->root;
4343 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4344 struct extent_map *em = NULL;
4345 struct extent_state *cached_state = NULL;
4346 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4347 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4348 u64 block_end = ALIGN(size, root->sectorsize);
4349 u64 last_byte;
4350 u64 cur_offset;
4351 u64 hole_size;
4352 int err = 0;
4353
4354 /*
4355 * If our size started in the middle of a page we need to zero out the
4356 * rest of the page before we expand the i_size, otherwise we could
4357 * expose stale data.
4358 */
4359 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4360 if (err)
4361 return err;
4362
4363 if (size <= hole_start)
4364 return 0;
4365
4366 while (1) {
4367 struct btrfs_ordered_extent *ordered;
4368
4369 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4370 &cached_state);
4371 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4372 block_end - hole_start);
4373 if (!ordered)
4374 break;
4375 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4376 &cached_state, GFP_NOFS);
4377 btrfs_start_ordered_extent(inode, ordered, 1);
4378 btrfs_put_ordered_extent(ordered);
4379 }
4380
4381 cur_offset = hole_start;
4382 while (1) {
4383 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4384 block_end - cur_offset, 0);
4385 if (IS_ERR(em)) {
4386 err = PTR_ERR(em);
4387 em = NULL;
4388 break;
4389 }
4390 last_byte = min(extent_map_end(em), block_end);
4391 last_byte = ALIGN(last_byte , root->sectorsize);
4392 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4393 struct extent_map *hole_em;
4394 hole_size = last_byte - cur_offset;
4395
4396 err = maybe_insert_hole(root, inode, cur_offset,
4397 hole_size);
4398 if (err)
4399 break;
4400 btrfs_drop_extent_cache(inode, cur_offset,
4401 cur_offset + hole_size - 1, 0);
4402 hole_em = alloc_extent_map();
4403 if (!hole_em) {
4404 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4405 &BTRFS_I(inode)->runtime_flags);
4406 goto next;
4407 }
4408 hole_em->start = cur_offset;
4409 hole_em->len = hole_size;
4410 hole_em->orig_start = cur_offset;
4411
4412 hole_em->block_start = EXTENT_MAP_HOLE;
4413 hole_em->block_len = 0;
4414 hole_em->orig_block_len = 0;
4415 hole_em->ram_bytes = hole_size;
4416 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4417 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4418 hole_em->generation = root->fs_info->generation;
4419
4420 while (1) {
4421 write_lock(&em_tree->lock);
4422 err = add_extent_mapping(em_tree, hole_em, 1);
4423 write_unlock(&em_tree->lock);
4424 if (err != -EEXIST)
4425 break;
4426 btrfs_drop_extent_cache(inode, cur_offset,
4427 cur_offset +
4428 hole_size - 1, 0);
4429 }
4430 free_extent_map(hole_em);
4431 }
4432 next:
4433 free_extent_map(em);
4434 em = NULL;
4435 cur_offset = last_byte;
4436 if (cur_offset >= block_end)
4437 break;
4438 }
4439 free_extent_map(em);
4440 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4441 GFP_NOFS);
4442 return err;
4443 }
4444
4445 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4446 {
4447 struct btrfs_root *root = BTRFS_I(inode)->root;
4448 struct btrfs_trans_handle *trans;
4449 loff_t oldsize = i_size_read(inode);
4450 loff_t newsize = attr->ia_size;
4451 int mask = attr->ia_valid;
4452 int ret;
4453
4454 /*
4455 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4456 * special case where we need to update the times despite not having
4457 * these flags set. For all other operations the VFS set these flags
4458 * explicitly if it wants a timestamp update.
4459 */
4460 if (newsize != oldsize) {
4461 inode_inc_iversion(inode);
4462 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4463 inode->i_ctime = inode->i_mtime =
4464 current_fs_time(inode->i_sb);
4465 }
4466
4467 if (newsize > oldsize) {
4468 truncate_pagecache(inode, newsize);
4469 ret = btrfs_cont_expand(inode, oldsize, newsize);
4470 if (ret)
4471 return ret;
4472
4473 trans = btrfs_start_transaction(root, 1);
4474 if (IS_ERR(trans))
4475 return PTR_ERR(trans);
4476
4477 i_size_write(inode, newsize);
4478 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4479 ret = btrfs_update_inode(trans, root, inode);
4480 btrfs_end_transaction(trans, root);
4481 } else {
4482
4483 /*
4484 * We're truncating a file that used to have good data down to
4485 * zero. Make sure it gets into the ordered flush list so that
4486 * any new writes get down to disk quickly.
4487 */
4488 if (newsize == 0)
4489 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4490 &BTRFS_I(inode)->runtime_flags);
4491
4492 /*
4493 * 1 for the orphan item we're going to add
4494 * 1 for the orphan item deletion.
4495 */
4496 trans = btrfs_start_transaction(root, 2);
4497 if (IS_ERR(trans))
4498 return PTR_ERR(trans);
4499
4500 /*
4501 * We need to do this in case we fail at _any_ point during the
4502 * actual truncate. Once we do the truncate_setsize we could
4503 * invalidate pages which forces any outstanding ordered io to
4504 * be instantly completed which will give us extents that need
4505 * to be truncated. If we fail to get an orphan inode down we
4506 * could have left over extents that were never meant to live,
4507 * so we need to garuntee from this point on that everything
4508 * will be consistent.
4509 */
4510 ret = btrfs_orphan_add(trans, inode);
4511 btrfs_end_transaction(trans, root);
4512 if (ret)
4513 return ret;
4514
4515 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4516 truncate_setsize(inode, newsize);
4517
4518 /* Disable nonlocked read DIO to avoid the end less truncate */
4519 btrfs_inode_block_unlocked_dio(inode);
4520 inode_dio_wait(inode);
4521 btrfs_inode_resume_unlocked_dio(inode);
4522
4523 ret = btrfs_truncate(inode);
4524 if (ret && inode->i_nlink) {
4525 int err;
4526
4527 /*
4528 * failed to truncate, disk_i_size is only adjusted down
4529 * as we remove extents, so it should represent the true
4530 * size of the inode, so reset the in memory size and
4531 * delete our orphan entry.
4532 */
4533 trans = btrfs_join_transaction(root);
4534 if (IS_ERR(trans)) {
4535 btrfs_orphan_del(NULL, inode);
4536 return ret;
4537 }
4538 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4539 err = btrfs_orphan_del(trans, inode);
4540 if (err)
4541 btrfs_abort_transaction(trans, root, err);
4542 btrfs_end_transaction(trans, root);
4543 }
4544 }
4545
4546 return ret;
4547 }
4548
4549 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4550 {
4551 struct inode *inode = dentry->d_inode;
4552 struct btrfs_root *root = BTRFS_I(inode)->root;
4553 int err;
4554
4555 if (btrfs_root_readonly(root))
4556 return -EROFS;
4557
4558 err = inode_change_ok(inode, attr);
4559 if (err)
4560 return err;
4561
4562 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4563 err = btrfs_setsize(inode, attr);
4564 if (err)
4565 return err;
4566 }
4567
4568 if (attr->ia_valid) {
4569 setattr_copy(inode, attr);
4570 inode_inc_iversion(inode);
4571 err = btrfs_dirty_inode(inode);
4572
4573 if (!err && attr->ia_valid & ATTR_MODE)
4574 err = posix_acl_chmod(inode, inode->i_mode);
4575 }
4576
4577 return err;
4578 }
4579
4580 /*
4581 * While truncating the inode pages during eviction, we get the VFS calling
4582 * btrfs_invalidatepage() against each page of the inode. This is slow because
4583 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4584 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4585 * extent_state structures over and over, wasting lots of time.
4586 *
4587 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4588 * those expensive operations on a per page basis and do only the ordered io
4589 * finishing, while we release here the extent_map and extent_state structures,
4590 * without the excessive merging and splitting.
4591 */
4592 static void evict_inode_truncate_pages(struct inode *inode)
4593 {
4594 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4595 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4596 struct rb_node *node;
4597
4598 ASSERT(inode->i_state & I_FREEING);
4599 truncate_inode_pages_final(&inode->i_data);
4600
4601 write_lock(&map_tree->lock);
4602 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4603 struct extent_map *em;
4604
4605 node = rb_first(&map_tree->map);
4606 em = rb_entry(node, struct extent_map, rb_node);
4607 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4608 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4609 remove_extent_mapping(map_tree, em);
4610 free_extent_map(em);
4611 }
4612 write_unlock(&map_tree->lock);
4613
4614 spin_lock(&io_tree->lock);
4615 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4616 struct extent_state *state;
4617 struct extent_state *cached_state = NULL;
4618
4619 node = rb_first(&io_tree->state);
4620 state = rb_entry(node, struct extent_state, rb_node);
4621 atomic_inc(&state->refs);
4622 spin_unlock(&io_tree->lock);
4623
4624 lock_extent_bits(io_tree, state->start, state->end,
4625 0, &cached_state);
4626 clear_extent_bit(io_tree, state->start, state->end,
4627 EXTENT_LOCKED | EXTENT_DIRTY |
4628 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4629 EXTENT_DEFRAG, 1, 1,
4630 &cached_state, GFP_NOFS);
4631 free_extent_state(state);
4632
4633 spin_lock(&io_tree->lock);
4634 }
4635 spin_unlock(&io_tree->lock);
4636 }
4637
4638 void btrfs_evict_inode(struct inode *inode)
4639 {
4640 struct btrfs_trans_handle *trans;
4641 struct btrfs_root *root = BTRFS_I(inode)->root;
4642 struct btrfs_block_rsv *rsv, *global_rsv;
4643 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4644 int ret;
4645
4646 trace_btrfs_inode_evict(inode);
4647
4648 evict_inode_truncate_pages(inode);
4649
4650 if (inode->i_nlink &&
4651 ((btrfs_root_refs(&root->root_item) != 0 &&
4652 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4653 btrfs_is_free_space_inode(inode)))
4654 goto no_delete;
4655
4656 if (is_bad_inode(inode)) {
4657 btrfs_orphan_del(NULL, inode);
4658 goto no_delete;
4659 }
4660 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4661 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4662
4663 if (root->fs_info->log_root_recovering) {
4664 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4665 &BTRFS_I(inode)->runtime_flags));
4666 goto no_delete;
4667 }
4668
4669 if (inode->i_nlink > 0) {
4670 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4671 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4672 goto no_delete;
4673 }
4674
4675 ret = btrfs_commit_inode_delayed_inode(inode);
4676 if (ret) {
4677 btrfs_orphan_del(NULL, inode);
4678 goto no_delete;
4679 }
4680
4681 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4682 if (!rsv) {
4683 btrfs_orphan_del(NULL, inode);
4684 goto no_delete;
4685 }
4686 rsv->size = min_size;
4687 rsv->failfast = 1;
4688 global_rsv = &root->fs_info->global_block_rsv;
4689
4690 btrfs_i_size_write(inode, 0);
4691
4692 /*
4693 * This is a bit simpler than btrfs_truncate since we've already
4694 * reserved our space for our orphan item in the unlink, so we just
4695 * need to reserve some slack space in case we add bytes and update
4696 * inode item when doing the truncate.
4697 */
4698 while (1) {
4699 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4700 BTRFS_RESERVE_FLUSH_LIMIT);
4701
4702 /*
4703 * Try and steal from the global reserve since we will
4704 * likely not use this space anyway, we want to try as
4705 * hard as possible to get this to work.
4706 */
4707 if (ret)
4708 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4709
4710 if (ret) {
4711 btrfs_warn(root->fs_info,
4712 "Could not get space for a delete, will truncate on mount %d",
4713 ret);
4714 btrfs_orphan_del(NULL, inode);
4715 btrfs_free_block_rsv(root, rsv);
4716 goto no_delete;
4717 }
4718
4719 trans = btrfs_join_transaction(root);
4720 if (IS_ERR(trans)) {
4721 btrfs_orphan_del(NULL, inode);
4722 btrfs_free_block_rsv(root, rsv);
4723 goto no_delete;
4724 }
4725
4726 trans->block_rsv = rsv;
4727
4728 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4729 if (ret != -ENOSPC)
4730 break;
4731
4732 trans->block_rsv = &root->fs_info->trans_block_rsv;
4733 btrfs_end_transaction(trans, root);
4734 trans = NULL;
4735 btrfs_btree_balance_dirty(root);
4736 }
4737
4738 btrfs_free_block_rsv(root, rsv);
4739
4740 /*
4741 * Errors here aren't a big deal, it just means we leave orphan items
4742 * in the tree. They will be cleaned up on the next mount.
4743 */
4744 if (ret == 0) {
4745 trans->block_rsv = root->orphan_block_rsv;
4746 btrfs_orphan_del(trans, inode);
4747 } else {
4748 btrfs_orphan_del(NULL, inode);
4749 }
4750
4751 trans->block_rsv = &root->fs_info->trans_block_rsv;
4752 if (!(root == root->fs_info->tree_root ||
4753 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4754 btrfs_return_ino(root, btrfs_ino(inode));
4755
4756 btrfs_end_transaction(trans, root);
4757 btrfs_btree_balance_dirty(root);
4758 no_delete:
4759 btrfs_remove_delayed_node(inode);
4760 clear_inode(inode);
4761 return;
4762 }
4763
4764 /*
4765 * this returns the key found in the dir entry in the location pointer.
4766 * If no dir entries were found, location->objectid is 0.
4767 */
4768 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4769 struct btrfs_key *location)
4770 {
4771 const char *name = dentry->d_name.name;
4772 int namelen = dentry->d_name.len;
4773 struct btrfs_dir_item *di;
4774 struct btrfs_path *path;
4775 struct btrfs_root *root = BTRFS_I(dir)->root;
4776 int ret = 0;
4777
4778 path = btrfs_alloc_path();
4779 if (!path)
4780 return -ENOMEM;
4781
4782 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4783 namelen, 0);
4784 if (IS_ERR(di))
4785 ret = PTR_ERR(di);
4786
4787 if (IS_ERR_OR_NULL(di))
4788 goto out_err;
4789
4790 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4791 out:
4792 btrfs_free_path(path);
4793 return ret;
4794 out_err:
4795 location->objectid = 0;
4796 goto out;
4797 }
4798
4799 /*
4800 * when we hit a tree root in a directory, the btrfs part of the inode
4801 * needs to be changed to reflect the root directory of the tree root. This
4802 * is kind of like crossing a mount point.
4803 */
4804 static int fixup_tree_root_location(struct btrfs_root *root,
4805 struct inode *dir,
4806 struct dentry *dentry,
4807 struct btrfs_key *location,
4808 struct btrfs_root **sub_root)
4809 {
4810 struct btrfs_path *path;
4811 struct btrfs_root *new_root;
4812 struct btrfs_root_ref *ref;
4813 struct extent_buffer *leaf;
4814 int ret;
4815 int err = 0;
4816
4817 path = btrfs_alloc_path();
4818 if (!path) {
4819 err = -ENOMEM;
4820 goto out;
4821 }
4822
4823 err = -ENOENT;
4824 ret = btrfs_find_item(root->fs_info->tree_root, path,
4825 BTRFS_I(dir)->root->root_key.objectid,
4826 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4827 if (ret) {
4828 if (ret < 0)
4829 err = ret;
4830 goto out;
4831 }
4832
4833 leaf = path->nodes[0];
4834 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4835 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4836 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4837 goto out;
4838
4839 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4840 (unsigned long)(ref + 1),
4841 dentry->d_name.len);
4842 if (ret)
4843 goto out;
4844
4845 btrfs_release_path(path);
4846
4847 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4848 if (IS_ERR(new_root)) {
4849 err = PTR_ERR(new_root);
4850 goto out;
4851 }
4852
4853 *sub_root = new_root;
4854 location->objectid = btrfs_root_dirid(&new_root->root_item);
4855 location->type = BTRFS_INODE_ITEM_KEY;
4856 location->offset = 0;
4857 err = 0;
4858 out:
4859 btrfs_free_path(path);
4860 return err;
4861 }
4862
4863 static void inode_tree_add(struct inode *inode)
4864 {
4865 struct btrfs_root *root = BTRFS_I(inode)->root;
4866 struct btrfs_inode *entry;
4867 struct rb_node **p;
4868 struct rb_node *parent;
4869 struct rb_node *new = &BTRFS_I(inode)->rb_node;
4870 u64 ino = btrfs_ino(inode);
4871
4872 if (inode_unhashed(inode))
4873 return;
4874 parent = NULL;
4875 spin_lock(&root->inode_lock);
4876 p = &root->inode_tree.rb_node;
4877 while (*p) {
4878 parent = *p;
4879 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4880
4881 if (ino < btrfs_ino(&entry->vfs_inode))
4882 p = &parent->rb_left;
4883 else if (ino > btrfs_ino(&entry->vfs_inode))
4884 p = &parent->rb_right;
4885 else {
4886 WARN_ON(!(entry->vfs_inode.i_state &
4887 (I_WILL_FREE | I_FREEING)));
4888 rb_replace_node(parent, new, &root->inode_tree);
4889 RB_CLEAR_NODE(parent);
4890 spin_unlock(&root->inode_lock);
4891 return;
4892 }
4893 }
4894 rb_link_node(new, parent, p);
4895 rb_insert_color(new, &root->inode_tree);
4896 spin_unlock(&root->inode_lock);
4897 }
4898
4899 static void inode_tree_del(struct inode *inode)
4900 {
4901 struct btrfs_root *root = BTRFS_I(inode)->root;
4902 int empty = 0;
4903
4904 spin_lock(&root->inode_lock);
4905 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4906 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4907 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4908 empty = RB_EMPTY_ROOT(&root->inode_tree);
4909 }
4910 spin_unlock(&root->inode_lock);
4911
4912 if (empty && btrfs_root_refs(&root->root_item) == 0) {
4913 synchronize_srcu(&root->fs_info->subvol_srcu);
4914 spin_lock(&root->inode_lock);
4915 empty = RB_EMPTY_ROOT(&root->inode_tree);
4916 spin_unlock(&root->inode_lock);
4917 if (empty)
4918 btrfs_add_dead_root(root);
4919 }
4920 }
4921
4922 void btrfs_invalidate_inodes(struct btrfs_root *root)
4923 {
4924 struct rb_node *node;
4925 struct rb_node *prev;
4926 struct btrfs_inode *entry;
4927 struct inode *inode;
4928 u64 objectid = 0;
4929
4930 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
4931 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4932
4933 spin_lock(&root->inode_lock);
4934 again:
4935 node = root->inode_tree.rb_node;
4936 prev = NULL;
4937 while (node) {
4938 prev = node;
4939 entry = rb_entry(node, struct btrfs_inode, rb_node);
4940
4941 if (objectid < btrfs_ino(&entry->vfs_inode))
4942 node = node->rb_left;
4943 else if (objectid > btrfs_ino(&entry->vfs_inode))
4944 node = node->rb_right;
4945 else
4946 break;
4947 }
4948 if (!node) {
4949 while (prev) {
4950 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4951 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4952 node = prev;
4953 break;
4954 }
4955 prev = rb_next(prev);
4956 }
4957 }
4958 while (node) {
4959 entry = rb_entry(node, struct btrfs_inode, rb_node);
4960 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4961 inode = igrab(&entry->vfs_inode);
4962 if (inode) {
4963 spin_unlock(&root->inode_lock);
4964 if (atomic_read(&inode->i_count) > 1)
4965 d_prune_aliases(inode);
4966 /*
4967 * btrfs_drop_inode will have it removed from
4968 * the inode cache when its usage count
4969 * hits zero.
4970 */
4971 iput(inode);
4972 cond_resched();
4973 spin_lock(&root->inode_lock);
4974 goto again;
4975 }
4976
4977 if (cond_resched_lock(&root->inode_lock))
4978 goto again;
4979
4980 node = rb_next(node);
4981 }
4982 spin_unlock(&root->inode_lock);
4983 }
4984
4985 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4986 {
4987 struct btrfs_iget_args *args = p;
4988 inode->i_ino = args->location->objectid;
4989 memcpy(&BTRFS_I(inode)->location, args->location,
4990 sizeof(*args->location));
4991 BTRFS_I(inode)->root = args->root;
4992 return 0;
4993 }
4994
4995 static int btrfs_find_actor(struct inode *inode, void *opaque)
4996 {
4997 struct btrfs_iget_args *args = opaque;
4998 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
4999 args->root == BTRFS_I(inode)->root;
5000 }
5001
5002 static struct inode *btrfs_iget_locked(struct super_block *s,
5003 struct btrfs_key *location,
5004 struct btrfs_root *root)
5005 {
5006 struct inode *inode;
5007 struct btrfs_iget_args args;
5008 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5009
5010 args.location = location;
5011 args.root = root;
5012
5013 inode = iget5_locked(s, hashval, btrfs_find_actor,
5014 btrfs_init_locked_inode,
5015 (void *)&args);
5016 return inode;
5017 }
5018
5019 /* Get an inode object given its location and corresponding root.
5020 * Returns in *is_new if the inode was read from disk
5021 */
5022 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5023 struct btrfs_root *root, int *new)
5024 {
5025 struct inode *inode;
5026
5027 inode = btrfs_iget_locked(s, location, root);
5028 if (!inode)
5029 return ERR_PTR(-ENOMEM);
5030
5031 if (inode->i_state & I_NEW) {
5032 btrfs_read_locked_inode(inode);
5033 if (!is_bad_inode(inode)) {
5034 inode_tree_add(inode);
5035 unlock_new_inode(inode);
5036 if (new)
5037 *new = 1;
5038 } else {
5039 unlock_new_inode(inode);
5040 iput(inode);
5041 inode = ERR_PTR(-ESTALE);
5042 }
5043 }
5044
5045 return inode;
5046 }
5047
5048 static struct inode *new_simple_dir(struct super_block *s,
5049 struct btrfs_key *key,
5050 struct btrfs_root *root)
5051 {
5052 struct inode *inode = new_inode(s);
5053
5054 if (!inode)
5055 return ERR_PTR(-ENOMEM);
5056
5057 BTRFS_I(inode)->root = root;
5058 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5059 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5060
5061 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5062 inode->i_op = &btrfs_dir_ro_inode_operations;
5063 inode->i_fop = &simple_dir_operations;
5064 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5065 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5066
5067 return inode;
5068 }
5069
5070 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5071 {
5072 struct inode *inode;
5073 struct btrfs_root *root = BTRFS_I(dir)->root;
5074 struct btrfs_root *sub_root = root;
5075 struct btrfs_key location;
5076 int index;
5077 int ret = 0;
5078
5079 if (dentry->d_name.len > BTRFS_NAME_LEN)
5080 return ERR_PTR(-ENAMETOOLONG);
5081
5082 ret = btrfs_inode_by_name(dir, dentry, &location);
5083 if (ret < 0)
5084 return ERR_PTR(ret);
5085
5086 if (location.objectid == 0)
5087 return ERR_PTR(-ENOENT);
5088
5089 if (location.type == BTRFS_INODE_ITEM_KEY) {
5090 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5091 return inode;
5092 }
5093
5094 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5095
5096 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5097 ret = fixup_tree_root_location(root, dir, dentry,
5098 &location, &sub_root);
5099 if (ret < 0) {
5100 if (ret != -ENOENT)
5101 inode = ERR_PTR(ret);
5102 else
5103 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5104 } else {
5105 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5106 }
5107 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5108
5109 if (!IS_ERR(inode) && root != sub_root) {
5110 down_read(&root->fs_info->cleanup_work_sem);
5111 if (!(inode->i_sb->s_flags & MS_RDONLY))
5112 ret = btrfs_orphan_cleanup(sub_root);
5113 up_read(&root->fs_info->cleanup_work_sem);
5114 if (ret) {
5115 iput(inode);
5116 inode = ERR_PTR(ret);
5117 }
5118 }
5119
5120 return inode;
5121 }
5122
5123 static int btrfs_dentry_delete(const struct dentry *dentry)
5124 {
5125 struct btrfs_root *root;
5126 struct inode *inode = dentry->d_inode;
5127
5128 if (!inode && !IS_ROOT(dentry))
5129 inode = dentry->d_parent->d_inode;
5130
5131 if (inode) {
5132 root = BTRFS_I(inode)->root;
5133 if (btrfs_root_refs(&root->root_item) == 0)
5134 return 1;
5135
5136 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5137 return 1;
5138 }
5139 return 0;
5140 }
5141
5142 static void btrfs_dentry_release(struct dentry *dentry)
5143 {
5144 if (dentry->d_fsdata)
5145 kfree(dentry->d_fsdata);
5146 }
5147
5148 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5149 unsigned int flags)
5150 {
5151 struct inode *inode;
5152
5153 inode = btrfs_lookup_dentry(dir, dentry);
5154 if (IS_ERR(inode)) {
5155 if (PTR_ERR(inode) == -ENOENT)
5156 inode = NULL;
5157 else
5158 return ERR_CAST(inode);
5159 }
5160
5161 return d_materialise_unique(dentry, inode);
5162 }
5163
5164 unsigned char btrfs_filetype_table[] = {
5165 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5166 };
5167
5168 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5169 {
5170 struct inode *inode = file_inode(file);
5171 struct btrfs_root *root = BTRFS_I(inode)->root;
5172 struct btrfs_item *item;
5173 struct btrfs_dir_item *di;
5174 struct btrfs_key key;
5175 struct btrfs_key found_key;
5176 struct btrfs_path *path;
5177 struct list_head ins_list;
5178 struct list_head del_list;
5179 int ret;
5180 struct extent_buffer *leaf;
5181 int slot;
5182 unsigned char d_type;
5183 int over = 0;
5184 u32 di_cur;
5185 u32 di_total;
5186 u32 di_len;
5187 int key_type = BTRFS_DIR_INDEX_KEY;
5188 char tmp_name[32];
5189 char *name_ptr;
5190 int name_len;
5191 int is_curr = 0; /* ctx->pos points to the current index? */
5192
5193 /* FIXME, use a real flag for deciding about the key type */
5194 if (root->fs_info->tree_root == root)
5195 key_type = BTRFS_DIR_ITEM_KEY;
5196
5197 if (!dir_emit_dots(file, ctx))
5198 return 0;
5199
5200 path = btrfs_alloc_path();
5201 if (!path)
5202 return -ENOMEM;
5203
5204 path->reada = 1;
5205
5206 if (key_type == BTRFS_DIR_INDEX_KEY) {
5207 INIT_LIST_HEAD(&ins_list);
5208 INIT_LIST_HEAD(&del_list);
5209 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5210 }
5211
5212 btrfs_set_key_type(&key, key_type);
5213 key.offset = ctx->pos;
5214 key.objectid = btrfs_ino(inode);
5215
5216 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5217 if (ret < 0)
5218 goto err;
5219
5220 while (1) {
5221 leaf = path->nodes[0];
5222 slot = path->slots[0];
5223 if (slot >= btrfs_header_nritems(leaf)) {
5224 ret = btrfs_next_leaf(root, path);
5225 if (ret < 0)
5226 goto err;
5227 else if (ret > 0)
5228 break;
5229 continue;
5230 }
5231
5232 item = btrfs_item_nr(slot);
5233 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5234
5235 if (found_key.objectid != key.objectid)
5236 break;
5237 if (btrfs_key_type(&found_key) != key_type)
5238 break;
5239 if (found_key.offset < ctx->pos)
5240 goto next;
5241 if (key_type == BTRFS_DIR_INDEX_KEY &&
5242 btrfs_should_delete_dir_index(&del_list,
5243 found_key.offset))
5244 goto next;
5245
5246 ctx->pos = found_key.offset;
5247 is_curr = 1;
5248
5249 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5250 di_cur = 0;
5251 di_total = btrfs_item_size(leaf, item);
5252
5253 while (di_cur < di_total) {
5254 struct btrfs_key location;
5255
5256 if (verify_dir_item(root, leaf, di))
5257 break;
5258
5259 name_len = btrfs_dir_name_len(leaf, di);
5260 if (name_len <= sizeof(tmp_name)) {
5261 name_ptr = tmp_name;
5262 } else {
5263 name_ptr = kmalloc(name_len, GFP_NOFS);
5264 if (!name_ptr) {
5265 ret = -ENOMEM;
5266 goto err;
5267 }
5268 }
5269 read_extent_buffer(leaf, name_ptr,
5270 (unsigned long)(di + 1), name_len);
5271
5272 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5273 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5274
5275
5276 /* is this a reference to our own snapshot? If so
5277 * skip it.
5278 *
5279 * In contrast to old kernels, we insert the snapshot's
5280 * dir item and dir index after it has been created, so
5281 * we won't find a reference to our own snapshot. We
5282 * still keep the following code for backward
5283 * compatibility.
5284 */
5285 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5286 location.objectid == root->root_key.objectid) {
5287 over = 0;
5288 goto skip;
5289 }
5290 over = !dir_emit(ctx, name_ptr, name_len,
5291 location.objectid, d_type);
5292
5293 skip:
5294 if (name_ptr != tmp_name)
5295 kfree(name_ptr);
5296
5297 if (over)
5298 goto nopos;
5299 di_len = btrfs_dir_name_len(leaf, di) +
5300 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5301 di_cur += di_len;
5302 di = (struct btrfs_dir_item *)((char *)di + di_len);
5303 }
5304 next:
5305 path->slots[0]++;
5306 }
5307
5308 if (key_type == BTRFS_DIR_INDEX_KEY) {
5309 if (is_curr)
5310 ctx->pos++;
5311 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5312 if (ret)
5313 goto nopos;
5314 }
5315
5316 /* Reached end of directory/root. Bump pos past the last item. */
5317 ctx->pos++;
5318
5319 /*
5320 * Stop new entries from being returned after we return the last
5321 * entry.
5322 *
5323 * New directory entries are assigned a strictly increasing
5324 * offset. This means that new entries created during readdir
5325 * are *guaranteed* to be seen in the future by that readdir.
5326 * This has broken buggy programs which operate on names as
5327 * they're returned by readdir. Until we re-use freed offsets
5328 * we have this hack to stop new entries from being returned
5329 * under the assumption that they'll never reach this huge
5330 * offset.
5331 *
5332 * This is being careful not to overflow 32bit loff_t unless the
5333 * last entry requires it because doing so has broken 32bit apps
5334 * in the past.
5335 */
5336 if (key_type == BTRFS_DIR_INDEX_KEY) {
5337 if (ctx->pos >= INT_MAX)
5338 ctx->pos = LLONG_MAX;
5339 else
5340 ctx->pos = INT_MAX;
5341 }
5342 nopos:
5343 ret = 0;
5344 err:
5345 if (key_type == BTRFS_DIR_INDEX_KEY)
5346 btrfs_put_delayed_items(&ins_list, &del_list);
5347 btrfs_free_path(path);
5348 return ret;
5349 }
5350
5351 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5352 {
5353 struct btrfs_root *root = BTRFS_I(inode)->root;
5354 struct btrfs_trans_handle *trans;
5355 int ret = 0;
5356 bool nolock = false;
5357
5358 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5359 return 0;
5360
5361 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5362 nolock = true;
5363
5364 if (wbc->sync_mode == WB_SYNC_ALL) {
5365 if (nolock)
5366 trans = btrfs_join_transaction_nolock(root);
5367 else
5368 trans = btrfs_join_transaction(root);
5369 if (IS_ERR(trans))
5370 return PTR_ERR(trans);
5371 ret = btrfs_commit_transaction(trans, root);
5372 }
5373 return ret;
5374 }
5375
5376 /*
5377 * This is somewhat expensive, updating the tree every time the
5378 * inode changes. But, it is most likely to find the inode in cache.
5379 * FIXME, needs more benchmarking...there are no reasons other than performance
5380 * to keep or drop this code.
5381 */
5382 static int btrfs_dirty_inode(struct inode *inode)
5383 {
5384 struct btrfs_root *root = BTRFS_I(inode)->root;
5385 struct btrfs_trans_handle *trans;
5386 int ret;
5387
5388 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5389 return 0;
5390
5391 trans = btrfs_join_transaction(root);
5392 if (IS_ERR(trans))
5393 return PTR_ERR(trans);
5394
5395 ret = btrfs_update_inode(trans, root, inode);
5396 if (ret && ret == -ENOSPC) {
5397 /* whoops, lets try again with the full transaction */
5398 btrfs_end_transaction(trans, root);
5399 trans = btrfs_start_transaction(root, 1);
5400 if (IS_ERR(trans))
5401 return PTR_ERR(trans);
5402
5403 ret = btrfs_update_inode(trans, root, inode);
5404 }
5405 btrfs_end_transaction(trans, root);
5406 if (BTRFS_I(inode)->delayed_node)
5407 btrfs_balance_delayed_items(root);
5408
5409 return ret;
5410 }
5411
5412 /*
5413 * This is a copy of file_update_time. We need this so we can return error on
5414 * ENOSPC for updating the inode in the case of file write and mmap writes.
5415 */
5416 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5417 int flags)
5418 {
5419 struct btrfs_root *root = BTRFS_I(inode)->root;
5420
5421 if (btrfs_root_readonly(root))
5422 return -EROFS;
5423
5424 if (flags & S_VERSION)
5425 inode_inc_iversion(inode);
5426 if (flags & S_CTIME)
5427 inode->i_ctime = *now;
5428 if (flags & S_MTIME)
5429 inode->i_mtime = *now;
5430 if (flags & S_ATIME)
5431 inode->i_atime = *now;
5432 return btrfs_dirty_inode(inode);
5433 }
5434
5435 /*
5436 * find the highest existing sequence number in a directory
5437 * and then set the in-memory index_cnt variable to reflect
5438 * free sequence numbers
5439 */
5440 static int btrfs_set_inode_index_count(struct inode *inode)
5441 {
5442 struct btrfs_root *root = BTRFS_I(inode)->root;
5443 struct btrfs_key key, found_key;
5444 struct btrfs_path *path;
5445 struct extent_buffer *leaf;
5446 int ret;
5447
5448 key.objectid = btrfs_ino(inode);
5449 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5450 key.offset = (u64)-1;
5451
5452 path = btrfs_alloc_path();
5453 if (!path)
5454 return -ENOMEM;
5455
5456 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5457 if (ret < 0)
5458 goto out;
5459 /* FIXME: we should be able to handle this */
5460 if (ret == 0)
5461 goto out;
5462 ret = 0;
5463
5464 /*
5465 * MAGIC NUMBER EXPLANATION:
5466 * since we search a directory based on f_pos we have to start at 2
5467 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5468 * else has to start at 2
5469 */
5470 if (path->slots[0] == 0) {
5471 BTRFS_I(inode)->index_cnt = 2;
5472 goto out;
5473 }
5474
5475 path->slots[0]--;
5476
5477 leaf = path->nodes[0];
5478 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5479
5480 if (found_key.objectid != btrfs_ino(inode) ||
5481 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5482 BTRFS_I(inode)->index_cnt = 2;
5483 goto out;
5484 }
5485
5486 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5487 out:
5488 btrfs_free_path(path);
5489 return ret;
5490 }
5491
5492 /*
5493 * helper to find a free sequence number in a given directory. This current
5494 * code is very simple, later versions will do smarter things in the btree
5495 */
5496 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5497 {
5498 int ret = 0;
5499
5500 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5501 ret = btrfs_inode_delayed_dir_index_count(dir);
5502 if (ret) {
5503 ret = btrfs_set_inode_index_count(dir);
5504 if (ret)
5505 return ret;
5506 }
5507 }
5508
5509 *index = BTRFS_I(dir)->index_cnt;
5510 BTRFS_I(dir)->index_cnt++;
5511
5512 return ret;
5513 }
5514
5515 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5516 struct btrfs_root *root,
5517 struct inode *dir,
5518 const char *name, int name_len,
5519 u64 ref_objectid, u64 objectid,
5520 umode_t mode, u64 *index)
5521 {
5522 struct inode *inode;
5523 struct btrfs_inode_item *inode_item;
5524 struct btrfs_key *location;
5525 struct btrfs_path *path;
5526 struct btrfs_inode_ref *ref;
5527 struct btrfs_key key[2];
5528 u32 sizes[2];
5529 unsigned long ptr;
5530 int ret;
5531
5532 path = btrfs_alloc_path();
5533 if (!path)
5534 return ERR_PTR(-ENOMEM);
5535
5536 inode = new_inode(root->fs_info->sb);
5537 if (!inode) {
5538 btrfs_free_path(path);
5539 return ERR_PTR(-ENOMEM);
5540 }
5541
5542 /*
5543 * we have to initialize this early, so we can reclaim the inode
5544 * number if we fail afterwards in this function.
5545 */
5546 inode->i_ino = objectid;
5547
5548 if (dir) {
5549 trace_btrfs_inode_request(dir);
5550
5551 ret = btrfs_set_inode_index(dir, index);
5552 if (ret) {
5553 btrfs_free_path(path);
5554 iput(inode);
5555 return ERR_PTR(ret);
5556 }
5557 }
5558 /*
5559 * index_cnt is ignored for everything but a dir,
5560 * btrfs_get_inode_index_count has an explanation for the magic
5561 * number
5562 */
5563 BTRFS_I(inode)->index_cnt = 2;
5564 BTRFS_I(inode)->dir_index = *index;
5565 BTRFS_I(inode)->root = root;
5566 BTRFS_I(inode)->generation = trans->transid;
5567 inode->i_generation = BTRFS_I(inode)->generation;
5568
5569 /*
5570 * We could have gotten an inode number from somebody who was fsynced
5571 * and then removed in this same transaction, so let's just set full
5572 * sync since it will be a full sync anyway and this will blow away the
5573 * old info in the log.
5574 */
5575 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5576
5577 key[0].objectid = objectid;
5578 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5579 key[0].offset = 0;
5580
5581 /*
5582 * Start new inodes with an inode_ref. This is slightly more
5583 * efficient for small numbers of hard links since they will
5584 * be packed into one item. Extended refs will kick in if we
5585 * add more hard links than can fit in the ref item.
5586 */
5587 key[1].objectid = objectid;
5588 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5589 key[1].offset = ref_objectid;
5590
5591 sizes[0] = sizeof(struct btrfs_inode_item);
5592 sizes[1] = name_len + sizeof(*ref);
5593
5594 path->leave_spinning = 1;
5595 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5596 if (ret != 0)
5597 goto fail;
5598
5599 inode_init_owner(inode, dir, mode);
5600 inode_set_bytes(inode, 0);
5601 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5602 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5603 struct btrfs_inode_item);
5604 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5605 sizeof(*inode_item));
5606 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5607
5608 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5609 struct btrfs_inode_ref);
5610 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5611 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5612 ptr = (unsigned long)(ref + 1);
5613 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5614
5615 btrfs_mark_buffer_dirty(path->nodes[0]);
5616 btrfs_free_path(path);
5617
5618 location = &BTRFS_I(inode)->location;
5619 location->objectid = objectid;
5620 location->offset = 0;
5621 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5622
5623 btrfs_inherit_iflags(inode, dir);
5624
5625 if (S_ISREG(mode)) {
5626 if (btrfs_test_opt(root, NODATASUM))
5627 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5628 if (btrfs_test_opt(root, NODATACOW))
5629 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5630 BTRFS_INODE_NODATASUM;
5631 }
5632
5633 btrfs_insert_inode_hash(inode);
5634 inode_tree_add(inode);
5635
5636 trace_btrfs_inode_new(inode);
5637 btrfs_set_inode_last_trans(trans, inode);
5638
5639 btrfs_update_root_times(trans, root);
5640
5641 ret = btrfs_inode_inherit_props(trans, inode, dir);
5642 if (ret)
5643 btrfs_err(root->fs_info,
5644 "error inheriting props for ino %llu (root %llu): %d",
5645 btrfs_ino(inode), root->root_key.objectid, ret);
5646
5647 return inode;
5648 fail:
5649 if (dir)
5650 BTRFS_I(dir)->index_cnt--;
5651 btrfs_free_path(path);
5652 iput(inode);
5653 return ERR_PTR(ret);
5654 }
5655
5656 static inline u8 btrfs_inode_type(struct inode *inode)
5657 {
5658 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5659 }
5660
5661 /*
5662 * utility function to add 'inode' into 'parent_inode' with
5663 * a give name and a given sequence number.
5664 * if 'add_backref' is true, also insert a backref from the
5665 * inode to the parent directory.
5666 */
5667 int btrfs_add_link(struct btrfs_trans_handle *trans,
5668 struct inode *parent_inode, struct inode *inode,
5669 const char *name, int name_len, int add_backref, u64 index)
5670 {
5671 int ret = 0;
5672 struct btrfs_key key;
5673 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5674 u64 ino = btrfs_ino(inode);
5675 u64 parent_ino = btrfs_ino(parent_inode);
5676
5677 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5678 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5679 } else {
5680 key.objectid = ino;
5681 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5682 key.offset = 0;
5683 }
5684
5685 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5686 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5687 key.objectid, root->root_key.objectid,
5688 parent_ino, index, name, name_len);
5689 } else if (add_backref) {
5690 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5691 parent_ino, index);
5692 }
5693
5694 /* Nothing to clean up yet */
5695 if (ret)
5696 return ret;
5697
5698 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5699 parent_inode, &key,
5700 btrfs_inode_type(inode), index);
5701 if (ret == -EEXIST || ret == -EOVERFLOW)
5702 goto fail_dir_item;
5703 else if (ret) {
5704 btrfs_abort_transaction(trans, root, ret);
5705 return ret;
5706 }
5707
5708 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5709 name_len * 2);
5710 inode_inc_iversion(parent_inode);
5711 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5712 ret = btrfs_update_inode(trans, root, parent_inode);
5713 if (ret)
5714 btrfs_abort_transaction(trans, root, ret);
5715 return ret;
5716
5717 fail_dir_item:
5718 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5719 u64 local_index;
5720 int err;
5721 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5722 key.objectid, root->root_key.objectid,
5723 parent_ino, &local_index, name, name_len);
5724
5725 } else if (add_backref) {
5726 u64 local_index;
5727 int err;
5728
5729 err = btrfs_del_inode_ref(trans, root, name, name_len,
5730 ino, parent_ino, &local_index);
5731 }
5732 return ret;
5733 }
5734
5735 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5736 struct inode *dir, struct dentry *dentry,
5737 struct inode *inode, int backref, u64 index)
5738 {
5739 int err = btrfs_add_link(trans, dir, inode,
5740 dentry->d_name.name, dentry->d_name.len,
5741 backref, index);
5742 if (err > 0)
5743 err = -EEXIST;
5744 return err;
5745 }
5746
5747 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5748 umode_t mode, dev_t rdev)
5749 {
5750 struct btrfs_trans_handle *trans;
5751 struct btrfs_root *root = BTRFS_I(dir)->root;
5752 struct inode *inode = NULL;
5753 int err;
5754 int drop_inode = 0;
5755 u64 objectid;
5756 u64 index = 0;
5757
5758 if (!new_valid_dev(rdev))
5759 return -EINVAL;
5760
5761 /*
5762 * 2 for inode item and ref
5763 * 2 for dir items
5764 * 1 for xattr if selinux is on
5765 */
5766 trans = btrfs_start_transaction(root, 5);
5767 if (IS_ERR(trans))
5768 return PTR_ERR(trans);
5769
5770 err = btrfs_find_free_ino(root, &objectid);
5771 if (err)
5772 goto out_unlock;
5773
5774 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5775 dentry->d_name.len, btrfs_ino(dir), objectid,
5776 mode, &index);
5777 if (IS_ERR(inode)) {
5778 err = PTR_ERR(inode);
5779 goto out_unlock;
5780 }
5781
5782 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5783 if (err) {
5784 drop_inode = 1;
5785 goto out_unlock;
5786 }
5787
5788 /*
5789 * If the active LSM wants to access the inode during
5790 * d_instantiate it needs these. Smack checks to see
5791 * if the filesystem supports xattrs by looking at the
5792 * ops vector.
5793 */
5794
5795 inode->i_op = &btrfs_special_inode_operations;
5796 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5797 if (err)
5798 drop_inode = 1;
5799 else {
5800 init_special_inode(inode, inode->i_mode, rdev);
5801 btrfs_update_inode(trans, root, inode);
5802 d_instantiate(dentry, inode);
5803 }
5804 out_unlock:
5805 btrfs_end_transaction(trans, root);
5806 btrfs_balance_delayed_items(root);
5807 btrfs_btree_balance_dirty(root);
5808 if (drop_inode) {
5809 inode_dec_link_count(inode);
5810 iput(inode);
5811 }
5812 return err;
5813 }
5814
5815 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5816 umode_t mode, bool excl)
5817 {
5818 struct btrfs_trans_handle *trans;
5819 struct btrfs_root *root = BTRFS_I(dir)->root;
5820 struct inode *inode = NULL;
5821 int drop_inode_on_err = 0;
5822 int err;
5823 u64 objectid;
5824 u64 index = 0;
5825
5826 /*
5827 * 2 for inode item and ref
5828 * 2 for dir items
5829 * 1 for xattr if selinux is on
5830 */
5831 trans = btrfs_start_transaction(root, 5);
5832 if (IS_ERR(trans))
5833 return PTR_ERR(trans);
5834
5835 err = btrfs_find_free_ino(root, &objectid);
5836 if (err)
5837 goto out_unlock;
5838
5839 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5840 dentry->d_name.len, btrfs_ino(dir), objectid,
5841 mode, &index);
5842 if (IS_ERR(inode)) {
5843 err = PTR_ERR(inode);
5844 goto out_unlock;
5845 }
5846 drop_inode_on_err = 1;
5847
5848 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5849 if (err)
5850 goto out_unlock;
5851
5852 err = btrfs_update_inode(trans, root, inode);
5853 if (err)
5854 goto out_unlock;
5855
5856 /*
5857 * If the active LSM wants to access the inode during
5858 * d_instantiate it needs these. Smack checks to see
5859 * if the filesystem supports xattrs by looking at the
5860 * ops vector.
5861 */
5862 inode->i_fop = &btrfs_file_operations;
5863 inode->i_op = &btrfs_file_inode_operations;
5864
5865 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5866 if (err)
5867 goto out_unlock;
5868
5869 inode->i_mapping->a_ops = &btrfs_aops;
5870 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5871 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5872 d_instantiate(dentry, inode);
5873
5874 out_unlock:
5875 btrfs_end_transaction(trans, root);
5876 if (err && drop_inode_on_err) {
5877 inode_dec_link_count(inode);
5878 iput(inode);
5879 }
5880 btrfs_balance_delayed_items(root);
5881 btrfs_btree_balance_dirty(root);
5882 return err;
5883 }
5884
5885 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5886 struct dentry *dentry)
5887 {
5888 struct btrfs_trans_handle *trans;
5889 struct btrfs_root *root = BTRFS_I(dir)->root;
5890 struct inode *inode = old_dentry->d_inode;
5891 u64 index;
5892 int err;
5893 int drop_inode = 0;
5894
5895 /* do not allow sys_link's with other subvols of the same device */
5896 if (root->objectid != BTRFS_I(inode)->root->objectid)
5897 return -EXDEV;
5898
5899 if (inode->i_nlink >= BTRFS_LINK_MAX)
5900 return -EMLINK;
5901
5902 err = btrfs_set_inode_index(dir, &index);
5903 if (err)
5904 goto fail;
5905
5906 /*
5907 * 2 items for inode and inode ref
5908 * 2 items for dir items
5909 * 1 item for parent inode
5910 */
5911 trans = btrfs_start_transaction(root, 5);
5912 if (IS_ERR(trans)) {
5913 err = PTR_ERR(trans);
5914 goto fail;
5915 }
5916
5917 /* There are several dir indexes for this inode, clear the cache. */
5918 BTRFS_I(inode)->dir_index = 0ULL;
5919 inc_nlink(inode);
5920 inode_inc_iversion(inode);
5921 inode->i_ctime = CURRENT_TIME;
5922 ihold(inode);
5923 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5924
5925 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5926
5927 if (err) {
5928 drop_inode = 1;
5929 } else {
5930 struct dentry *parent = dentry->d_parent;
5931 err = btrfs_update_inode(trans, root, inode);
5932 if (err)
5933 goto fail;
5934 d_instantiate(dentry, inode);
5935 btrfs_log_new_name(trans, inode, NULL, parent);
5936 }
5937
5938 btrfs_end_transaction(trans, root);
5939 btrfs_balance_delayed_items(root);
5940 fail:
5941 if (drop_inode) {
5942 inode_dec_link_count(inode);
5943 iput(inode);
5944 }
5945 btrfs_btree_balance_dirty(root);
5946 return err;
5947 }
5948
5949 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5950 {
5951 struct inode *inode = NULL;
5952 struct btrfs_trans_handle *trans;
5953 struct btrfs_root *root = BTRFS_I(dir)->root;
5954 int err = 0;
5955 int drop_on_err = 0;
5956 u64 objectid = 0;
5957 u64 index = 0;
5958
5959 /*
5960 * 2 items for inode and ref
5961 * 2 items for dir items
5962 * 1 for xattr if selinux is on
5963 */
5964 trans = btrfs_start_transaction(root, 5);
5965 if (IS_ERR(trans))
5966 return PTR_ERR(trans);
5967
5968 err = btrfs_find_free_ino(root, &objectid);
5969 if (err)
5970 goto out_fail;
5971
5972 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5973 dentry->d_name.len, btrfs_ino(dir), objectid,
5974 S_IFDIR | mode, &index);
5975 if (IS_ERR(inode)) {
5976 err = PTR_ERR(inode);
5977 goto out_fail;
5978 }
5979
5980 drop_on_err = 1;
5981
5982 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5983 if (err)
5984 goto out_fail;
5985
5986 inode->i_op = &btrfs_dir_inode_operations;
5987 inode->i_fop = &btrfs_dir_file_operations;
5988
5989 btrfs_i_size_write(inode, 0);
5990 err = btrfs_update_inode(trans, root, inode);
5991 if (err)
5992 goto out_fail;
5993
5994 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5995 dentry->d_name.len, 0, index);
5996 if (err)
5997 goto out_fail;
5998
5999 d_instantiate(dentry, inode);
6000 drop_on_err = 0;
6001
6002 out_fail:
6003 btrfs_end_transaction(trans, root);
6004 if (drop_on_err)
6005 iput(inode);
6006 btrfs_balance_delayed_items(root);
6007 btrfs_btree_balance_dirty(root);
6008 return err;
6009 }
6010
6011 /* helper for btfs_get_extent. Given an existing extent in the tree,
6012 * and an extent that you want to insert, deal with overlap and insert
6013 * the new extent into the tree.
6014 */
6015 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6016 struct extent_map *existing,
6017 struct extent_map *em,
6018 u64 map_start, u64 map_len)
6019 {
6020 u64 start_diff;
6021
6022 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6023 start_diff = map_start - em->start;
6024 em->start = map_start;
6025 em->len = map_len;
6026 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6027 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6028 em->block_start += start_diff;
6029 em->block_len -= start_diff;
6030 }
6031 return add_extent_mapping(em_tree, em, 0);
6032 }
6033
6034 static noinline int uncompress_inline(struct btrfs_path *path,
6035 struct inode *inode, struct page *page,
6036 size_t pg_offset, u64 extent_offset,
6037 struct btrfs_file_extent_item *item)
6038 {
6039 int ret;
6040 struct extent_buffer *leaf = path->nodes[0];
6041 char *tmp;
6042 size_t max_size;
6043 unsigned long inline_size;
6044 unsigned long ptr;
6045 int compress_type;
6046
6047 WARN_ON(pg_offset != 0);
6048 compress_type = btrfs_file_extent_compression(leaf, item);
6049 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6050 inline_size = btrfs_file_extent_inline_item_len(leaf,
6051 btrfs_item_nr(path->slots[0]));
6052 tmp = kmalloc(inline_size, GFP_NOFS);
6053 if (!tmp)
6054 return -ENOMEM;
6055 ptr = btrfs_file_extent_inline_start(item);
6056
6057 read_extent_buffer(leaf, tmp, ptr, inline_size);
6058
6059 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6060 ret = btrfs_decompress(compress_type, tmp, page,
6061 extent_offset, inline_size, max_size);
6062 if (ret) {
6063 char *kaddr = kmap_atomic(page);
6064 unsigned long copy_size = min_t(u64,
6065 PAGE_CACHE_SIZE - pg_offset,
6066 max_size - extent_offset);
6067 memset(kaddr + pg_offset, 0, copy_size);
6068 kunmap_atomic(kaddr);
6069 }
6070 kfree(tmp);
6071 return 0;
6072 }
6073
6074 /*
6075 * a bit scary, this does extent mapping from logical file offset to the disk.
6076 * the ugly parts come from merging extents from the disk with the in-ram
6077 * representation. This gets more complex because of the data=ordered code,
6078 * where the in-ram extents might be locked pending data=ordered completion.
6079 *
6080 * This also copies inline extents directly into the page.
6081 */
6082
6083 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6084 size_t pg_offset, u64 start, u64 len,
6085 int create)
6086 {
6087 int ret;
6088 int err = 0;
6089 u64 bytenr;
6090 u64 extent_start = 0;
6091 u64 extent_end = 0;
6092 u64 objectid = btrfs_ino(inode);
6093 u32 found_type;
6094 struct btrfs_path *path = NULL;
6095 struct btrfs_root *root = BTRFS_I(inode)->root;
6096 struct btrfs_file_extent_item *item;
6097 struct extent_buffer *leaf;
6098 struct btrfs_key found_key;
6099 struct extent_map *em = NULL;
6100 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6101 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6102 struct btrfs_trans_handle *trans = NULL;
6103 int compress_type;
6104
6105 again:
6106 read_lock(&em_tree->lock);
6107 em = lookup_extent_mapping(em_tree, start, len);
6108 if (em)
6109 em->bdev = root->fs_info->fs_devices->latest_bdev;
6110 read_unlock(&em_tree->lock);
6111
6112 if (em) {
6113 if (em->start > start || em->start + em->len <= start)
6114 free_extent_map(em);
6115 else if (em->block_start == EXTENT_MAP_INLINE && page)
6116 free_extent_map(em);
6117 else
6118 goto out;
6119 }
6120 em = alloc_extent_map();
6121 if (!em) {
6122 err = -ENOMEM;
6123 goto out;
6124 }
6125 em->bdev = root->fs_info->fs_devices->latest_bdev;
6126 em->start = EXTENT_MAP_HOLE;
6127 em->orig_start = EXTENT_MAP_HOLE;
6128 em->len = (u64)-1;
6129 em->block_len = (u64)-1;
6130
6131 if (!path) {
6132 path = btrfs_alloc_path();
6133 if (!path) {
6134 err = -ENOMEM;
6135 goto out;
6136 }
6137 /*
6138 * Chances are we'll be called again, so go ahead and do
6139 * readahead
6140 */
6141 path->reada = 1;
6142 }
6143
6144 ret = btrfs_lookup_file_extent(trans, root, path,
6145 objectid, start, trans != NULL);
6146 if (ret < 0) {
6147 err = ret;
6148 goto out;
6149 }
6150
6151 if (ret != 0) {
6152 if (path->slots[0] == 0)
6153 goto not_found;
6154 path->slots[0]--;
6155 }
6156
6157 leaf = path->nodes[0];
6158 item = btrfs_item_ptr(leaf, path->slots[0],
6159 struct btrfs_file_extent_item);
6160 /* are we inside the extent that was found? */
6161 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6162 found_type = btrfs_key_type(&found_key);
6163 if (found_key.objectid != objectid ||
6164 found_type != BTRFS_EXTENT_DATA_KEY) {
6165 /*
6166 * If we backup past the first extent we want to move forward
6167 * and see if there is an extent in front of us, otherwise we'll
6168 * say there is a hole for our whole search range which can
6169 * cause problems.
6170 */
6171 extent_end = start;
6172 goto next;
6173 }
6174
6175 found_type = btrfs_file_extent_type(leaf, item);
6176 extent_start = found_key.offset;
6177 compress_type = btrfs_file_extent_compression(leaf, item);
6178 if (found_type == BTRFS_FILE_EXTENT_REG ||
6179 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6180 extent_end = extent_start +
6181 btrfs_file_extent_num_bytes(leaf, item);
6182 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6183 size_t size;
6184 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6185 extent_end = ALIGN(extent_start + size, root->sectorsize);
6186 }
6187 next:
6188 if (start >= extent_end) {
6189 path->slots[0]++;
6190 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6191 ret = btrfs_next_leaf(root, path);
6192 if (ret < 0) {
6193 err = ret;
6194 goto out;
6195 }
6196 if (ret > 0)
6197 goto not_found;
6198 leaf = path->nodes[0];
6199 }
6200 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6201 if (found_key.objectid != objectid ||
6202 found_key.type != BTRFS_EXTENT_DATA_KEY)
6203 goto not_found;
6204 if (start + len <= found_key.offset)
6205 goto not_found;
6206 em->start = start;
6207 em->orig_start = start;
6208 em->len = found_key.offset - start;
6209 goto not_found_em;
6210 }
6211
6212 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6213 if (found_type == BTRFS_FILE_EXTENT_REG ||
6214 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6215 em->start = extent_start;
6216 em->len = extent_end - extent_start;
6217 em->orig_start = extent_start -
6218 btrfs_file_extent_offset(leaf, item);
6219 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6220 item);
6221 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6222 if (bytenr == 0) {
6223 em->block_start = EXTENT_MAP_HOLE;
6224 goto insert;
6225 }
6226 if (compress_type != BTRFS_COMPRESS_NONE) {
6227 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6228 em->compress_type = compress_type;
6229 em->block_start = bytenr;
6230 em->block_len = em->orig_block_len;
6231 } else {
6232 bytenr += btrfs_file_extent_offset(leaf, item);
6233 em->block_start = bytenr;
6234 em->block_len = em->len;
6235 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6236 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6237 }
6238 goto insert;
6239 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6240 unsigned long ptr;
6241 char *map;
6242 size_t size;
6243 size_t extent_offset;
6244 size_t copy_size;
6245
6246 em->block_start = EXTENT_MAP_INLINE;
6247 if (!page || create) {
6248 em->start = extent_start;
6249 em->len = extent_end - extent_start;
6250 goto out;
6251 }
6252
6253 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6254 extent_offset = page_offset(page) + pg_offset - extent_start;
6255 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6256 size - extent_offset);
6257 em->start = extent_start + extent_offset;
6258 em->len = ALIGN(copy_size, root->sectorsize);
6259 em->orig_block_len = em->len;
6260 em->orig_start = em->start;
6261 if (compress_type) {
6262 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6263 em->compress_type = compress_type;
6264 }
6265 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6266 if (create == 0 && !PageUptodate(page)) {
6267 if (btrfs_file_extent_compression(leaf, item) !=
6268 BTRFS_COMPRESS_NONE) {
6269 ret = uncompress_inline(path, inode, page,
6270 pg_offset,
6271 extent_offset, item);
6272 BUG_ON(ret); /* -ENOMEM */
6273 } else {
6274 map = kmap(page);
6275 read_extent_buffer(leaf, map + pg_offset, ptr,
6276 copy_size);
6277 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6278 memset(map + pg_offset + copy_size, 0,
6279 PAGE_CACHE_SIZE - pg_offset -
6280 copy_size);
6281 }
6282 kunmap(page);
6283 }
6284 flush_dcache_page(page);
6285 } else if (create && PageUptodate(page)) {
6286 BUG();
6287 if (!trans) {
6288 kunmap(page);
6289 free_extent_map(em);
6290 em = NULL;
6291
6292 btrfs_release_path(path);
6293 trans = btrfs_join_transaction(root);
6294
6295 if (IS_ERR(trans))
6296 return ERR_CAST(trans);
6297 goto again;
6298 }
6299 map = kmap(page);
6300 write_extent_buffer(leaf, map + pg_offset, ptr,
6301 copy_size);
6302 kunmap(page);
6303 btrfs_mark_buffer_dirty(leaf);
6304 }
6305 set_extent_uptodate(io_tree, em->start,
6306 extent_map_end(em) - 1, NULL, GFP_NOFS);
6307 goto insert;
6308 } else {
6309 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6310 }
6311 not_found:
6312 em->start = start;
6313 em->orig_start = start;
6314 em->len = len;
6315 not_found_em:
6316 em->block_start = EXTENT_MAP_HOLE;
6317 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6318 insert:
6319 btrfs_release_path(path);
6320 if (em->start > start || extent_map_end(em) <= start) {
6321 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6322 em->start, em->len, start, len);
6323 err = -EIO;
6324 goto out;
6325 }
6326
6327 err = 0;
6328 write_lock(&em_tree->lock);
6329 ret = add_extent_mapping(em_tree, em, 0);
6330 /* it is possible that someone inserted the extent into the tree
6331 * while we had the lock dropped. It is also possible that
6332 * an overlapping map exists in the tree
6333 */
6334 if (ret == -EEXIST) {
6335 struct extent_map *existing;
6336
6337 ret = 0;
6338
6339 existing = lookup_extent_mapping(em_tree, start, len);
6340 if (existing && (existing->start > start ||
6341 existing->start + existing->len <= start)) {
6342 free_extent_map(existing);
6343 existing = NULL;
6344 }
6345 if (!existing) {
6346 existing = lookup_extent_mapping(em_tree, em->start,
6347 em->len);
6348 if (existing) {
6349 err = merge_extent_mapping(em_tree, existing,
6350 em, start,
6351 root->sectorsize);
6352 free_extent_map(existing);
6353 if (err) {
6354 free_extent_map(em);
6355 em = NULL;
6356 }
6357 } else {
6358 err = -EIO;
6359 free_extent_map(em);
6360 em = NULL;
6361 }
6362 } else {
6363 free_extent_map(em);
6364 em = existing;
6365 err = 0;
6366 }
6367 }
6368 write_unlock(&em_tree->lock);
6369 out:
6370
6371 trace_btrfs_get_extent(root, em);
6372
6373 if (path)
6374 btrfs_free_path(path);
6375 if (trans) {
6376 ret = btrfs_end_transaction(trans, root);
6377 if (!err)
6378 err = ret;
6379 }
6380 if (err) {
6381 free_extent_map(em);
6382 return ERR_PTR(err);
6383 }
6384 BUG_ON(!em); /* Error is always set */
6385 return em;
6386 }
6387
6388 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6389 size_t pg_offset, u64 start, u64 len,
6390 int create)
6391 {
6392 struct extent_map *em;
6393 struct extent_map *hole_em = NULL;
6394 u64 range_start = start;
6395 u64 end;
6396 u64 found;
6397 u64 found_end;
6398 int err = 0;
6399
6400 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6401 if (IS_ERR(em))
6402 return em;
6403 if (em) {
6404 /*
6405 * if our em maps to
6406 * - a hole or
6407 * - a pre-alloc extent,
6408 * there might actually be delalloc bytes behind it.
6409 */
6410 if (em->block_start != EXTENT_MAP_HOLE &&
6411 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6412 return em;
6413 else
6414 hole_em = em;
6415 }
6416
6417 /* check to see if we've wrapped (len == -1 or similar) */
6418 end = start + len;
6419 if (end < start)
6420 end = (u64)-1;
6421 else
6422 end -= 1;
6423
6424 em = NULL;
6425
6426 /* ok, we didn't find anything, lets look for delalloc */
6427 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6428 end, len, EXTENT_DELALLOC, 1);
6429 found_end = range_start + found;
6430 if (found_end < range_start)
6431 found_end = (u64)-1;
6432
6433 /*
6434 * we didn't find anything useful, return
6435 * the original results from get_extent()
6436 */
6437 if (range_start > end || found_end <= start) {
6438 em = hole_em;
6439 hole_em = NULL;
6440 goto out;
6441 }
6442
6443 /* adjust the range_start to make sure it doesn't
6444 * go backwards from the start they passed in
6445 */
6446 range_start = max(start, range_start);
6447 found = found_end - range_start;
6448
6449 if (found > 0) {
6450 u64 hole_start = start;
6451 u64 hole_len = len;
6452
6453 em = alloc_extent_map();
6454 if (!em) {
6455 err = -ENOMEM;
6456 goto out;
6457 }
6458 /*
6459 * when btrfs_get_extent can't find anything it
6460 * returns one huge hole
6461 *
6462 * make sure what it found really fits our range, and
6463 * adjust to make sure it is based on the start from
6464 * the caller
6465 */
6466 if (hole_em) {
6467 u64 calc_end = extent_map_end(hole_em);
6468
6469 if (calc_end <= start || (hole_em->start > end)) {
6470 free_extent_map(hole_em);
6471 hole_em = NULL;
6472 } else {
6473 hole_start = max(hole_em->start, start);
6474 hole_len = calc_end - hole_start;
6475 }
6476 }
6477 em->bdev = NULL;
6478 if (hole_em && range_start > hole_start) {
6479 /* our hole starts before our delalloc, so we
6480 * have to return just the parts of the hole
6481 * that go until the delalloc starts
6482 */
6483 em->len = min(hole_len,
6484 range_start - hole_start);
6485 em->start = hole_start;
6486 em->orig_start = hole_start;
6487 /*
6488 * don't adjust block start at all,
6489 * it is fixed at EXTENT_MAP_HOLE
6490 */
6491 em->block_start = hole_em->block_start;
6492 em->block_len = hole_len;
6493 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6494 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6495 } else {
6496 em->start = range_start;
6497 em->len = found;
6498 em->orig_start = range_start;
6499 em->block_start = EXTENT_MAP_DELALLOC;
6500 em->block_len = found;
6501 }
6502 } else if (hole_em) {
6503 return hole_em;
6504 }
6505 out:
6506
6507 free_extent_map(hole_em);
6508 if (err) {
6509 free_extent_map(em);
6510 return ERR_PTR(err);
6511 }
6512 return em;
6513 }
6514
6515 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6516 u64 start, u64 len)
6517 {
6518 struct btrfs_root *root = BTRFS_I(inode)->root;
6519 struct extent_map *em;
6520 struct btrfs_key ins;
6521 u64 alloc_hint;
6522 int ret;
6523
6524 alloc_hint = get_extent_allocation_hint(inode, start, len);
6525 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6526 alloc_hint, &ins, 1);
6527 if (ret)
6528 return ERR_PTR(ret);
6529
6530 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6531 ins.offset, ins.offset, ins.offset, 0);
6532 if (IS_ERR(em)) {
6533 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6534 return em;
6535 }
6536
6537 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6538 ins.offset, ins.offset, 0);
6539 if (ret) {
6540 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6541 free_extent_map(em);
6542 return ERR_PTR(ret);
6543 }
6544
6545 return em;
6546 }
6547
6548 /*
6549 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6550 * block must be cow'd
6551 */
6552 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6553 u64 *orig_start, u64 *orig_block_len,
6554 u64 *ram_bytes)
6555 {
6556 struct btrfs_trans_handle *trans;
6557 struct btrfs_path *path;
6558 int ret;
6559 struct extent_buffer *leaf;
6560 struct btrfs_root *root = BTRFS_I(inode)->root;
6561 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6562 struct btrfs_file_extent_item *fi;
6563 struct btrfs_key key;
6564 u64 disk_bytenr;
6565 u64 backref_offset;
6566 u64 extent_end;
6567 u64 num_bytes;
6568 int slot;
6569 int found_type;
6570 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6571
6572 path = btrfs_alloc_path();
6573 if (!path)
6574 return -ENOMEM;
6575
6576 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6577 offset, 0);
6578 if (ret < 0)
6579 goto out;
6580
6581 slot = path->slots[0];
6582 if (ret == 1) {
6583 if (slot == 0) {
6584 /* can't find the item, must cow */
6585 ret = 0;
6586 goto out;
6587 }
6588 slot--;
6589 }
6590 ret = 0;
6591 leaf = path->nodes[0];
6592 btrfs_item_key_to_cpu(leaf, &key, slot);
6593 if (key.objectid != btrfs_ino(inode) ||
6594 key.type != BTRFS_EXTENT_DATA_KEY) {
6595 /* not our file or wrong item type, must cow */
6596 goto out;
6597 }
6598
6599 if (key.offset > offset) {
6600 /* Wrong offset, must cow */
6601 goto out;
6602 }
6603
6604 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6605 found_type = btrfs_file_extent_type(leaf, fi);
6606 if (found_type != BTRFS_FILE_EXTENT_REG &&
6607 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6608 /* not a regular extent, must cow */
6609 goto out;
6610 }
6611
6612 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6613 goto out;
6614
6615 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6616 if (extent_end <= offset)
6617 goto out;
6618
6619 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6620 if (disk_bytenr == 0)
6621 goto out;
6622
6623 if (btrfs_file_extent_compression(leaf, fi) ||
6624 btrfs_file_extent_encryption(leaf, fi) ||
6625 btrfs_file_extent_other_encoding(leaf, fi))
6626 goto out;
6627
6628 backref_offset = btrfs_file_extent_offset(leaf, fi);
6629
6630 if (orig_start) {
6631 *orig_start = key.offset - backref_offset;
6632 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6633 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6634 }
6635
6636 if (btrfs_extent_readonly(root, disk_bytenr))
6637 goto out;
6638
6639 num_bytes = min(offset + *len, extent_end) - offset;
6640 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6641 u64 range_end;
6642
6643 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6644 ret = test_range_bit(io_tree, offset, range_end,
6645 EXTENT_DELALLOC, 0, NULL);
6646 if (ret) {
6647 ret = -EAGAIN;
6648 goto out;
6649 }
6650 }
6651
6652 btrfs_release_path(path);
6653
6654 /*
6655 * look for other files referencing this extent, if we
6656 * find any we must cow
6657 */
6658 trans = btrfs_join_transaction(root);
6659 if (IS_ERR(trans)) {
6660 ret = 0;
6661 goto out;
6662 }
6663
6664 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6665 key.offset - backref_offset, disk_bytenr);
6666 btrfs_end_transaction(trans, root);
6667 if (ret) {
6668 ret = 0;
6669 goto out;
6670 }
6671
6672 /*
6673 * adjust disk_bytenr and num_bytes to cover just the bytes
6674 * in this extent we are about to write. If there
6675 * are any csums in that range we have to cow in order
6676 * to keep the csums correct
6677 */
6678 disk_bytenr += backref_offset;
6679 disk_bytenr += offset - key.offset;
6680 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6681 goto out;
6682 /*
6683 * all of the above have passed, it is safe to overwrite this extent
6684 * without cow
6685 */
6686 *len = num_bytes;
6687 ret = 1;
6688 out:
6689 btrfs_free_path(path);
6690 return ret;
6691 }
6692
6693 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6694 struct extent_state **cached_state, int writing)
6695 {
6696 struct btrfs_ordered_extent *ordered;
6697 int ret = 0;
6698
6699 while (1) {
6700 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6701 0, cached_state);
6702 /*
6703 * We're concerned with the entire range that we're going to be
6704 * doing DIO to, so we need to make sure theres no ordered
6705 * extents in this range.
6706 */
6707 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6708 lockend - lockstart + 1);
6709
6710 /*
6711 * We need to make sure there are no buffered pages in this
6712 * range either, we could have raced between the invalidate in
6713 * generic_file_direct_write and locking the extent. The
6714 * invalidate needs to happen so that reads after a write do not
6715 * get stale data.
6716 */
6717 if (!ordered && (!writing ||
6718 !test_range_bit(&BTRFS_I(inode)->io_tree,
6719 lockstart, lockend, EXTENT_UPTODATE, 0,
6720 *cached_state)))
6721 break;
6722
6723 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6724 cached_state, GFP_NOFS);
6725
6726 if (ordered) {
6727 btrfs_start_ordered_extent(inode, ordered, 1);
6728 btrfs_put_ordered_extent(ordered);
6729 } else {
6730 /* Screw you mmap */
6731 ret = filemap_write_and_wait_range(inode->i_mapping,
6732 lockstart,
6733 lockend);
6734 if (ret)
6735 break;
6736
6737 /*
6738 * If we found a page that couldn't be invalidated just
6739 * fall back to buffered.
6740 */
6741 ret = invalidate_inode_pages2_range(inode->i_mapping,
6742 lockstart >> PAGE_CACHE_SHIFT,
6743 lockend >> PAGE_CACHE_SHIFT);
6744 if (ret)
6745 break;
6746 }
6747
6748 cond_resched();
6749 }
6750
6751 return ret;
6752 }
6753
6754 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6755 u64 len, u64 orig_start,
6756 u64 block_start, u64 block_len,
6757 u64 orig_block_len, u64 ram_bytes,
6758 int type)
6759 {
6760 struct extent_map_tree *em_tree;
6761 struct extent_map *em;
6762 struct btrfs_root *root = BTRFS_I(inode)->root;
6763 int ret;
6764
6765 em_tree = &BTRFS_I(inode)->extent_tree;
6766 em = alloc_extent_map();
6767 if (!em)
6768 return ERR_PTR(-ENOMEM);
6769
6770 em->start = start;
6771 em->orig_start = orig_start;
6772 em->mod_start = start;
6773 em->mod_len = len;
6774 em->len = len;
6775 em->block_len = block_len;
6776 em->block_start = block_start;
6777 em->bdev = root->fs_info->fs_devices->latest_bdev;
6778 em->orig_block_len = orig_block_len;
6779 em->ram_bytes = ram_bytes;
6780 em->generation = -1;
6781 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6782 if (type == BTRFS_ORDERED_PREALLOC)
6783 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6784
6785 do {
6786 btrfs_drop_extent_cache(inode, em->start,
6787 em->start + em->len - 1, 0);
6788 write_lock(&em_tree->lock);
6789 ret = add_extent_mapping(em_tree, em, 1);
6790 write_unlock(&em_tree->lock);
6791 } while (ret == -EEXIST);
6792
6793 if (ret) {
6794 free_extent_map(em);
6795 return ERR_PTR(ret);
6796 }
6797
6798 return em;
6799 }
6800
6801
6802 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6803 struct buffer_head *bh_result, int create)
6804 {
6805 struct extent_map *em;
6806 struct btrfs_root *root = BTRFS_I(inode)->root;
6807 struct extent_state *cached_state = NULL;
6808 u64 start = iblock << inode->i_blkbits;
6809 u64 lockstart, lockend;
6810 u64 len = bh_result->b_size;
6811 int unlock_bits = EXTENT_LOCKED;
6812 int ret = 0;
6813
6814 if (create)
6815 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6816 else
6817 len = min_t(u64, len, root->sectorsize);
6818
6819 lockstart = start;
6820 lockend = start + len - 1;
6821
6822 /*
6823 * If this errors out it's because we couldn't invalidate pagecache for
6824 * this range and we need to fallback to buffered.
6825 */
6826 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6827 return -ENOTBLK;
6828
6829 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6830 if (IS_ERR(em)) {
6831 ret = PTR_ERR(em);
6832 goto unlock_err;
6833 }
6834
6835 /*
6836 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6837 * io. INLINE is special, and we could probably kludge it in here, but
6838 * it's still buffered so for safety lets just fall back to the generic
6839 * buffered path.
6840 *
6841 * For COMPRESSED we _have_ to read the entire extent in so we can
6842 * decompress it, so there will be buffering required no matter what we
6843 * do, so go ahead and fallback to buffered.
6844 *
6845 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6846 * to buffered IO. Don't blame me, this is the price we pay for using
6847 * the generic code.
6848 */
6849 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6850 em->block_start == EXTENT_MAP_INLINE) {
6851 free_extent_map(em);
6852 ret = -ENOTBLK;
6853 goto unlock_err;
6854 }
6855
6856 /* Just a good old fashioned hole, return */
6857 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6858 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6859 free_extent_map(em);
6860 goto unlock_err;
6861 }
6862
6863 /*
6864 * We don't allocate a new extent in the following cases
6865 *
6866 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6867 * existing extent.
6868 * 2) The extent is marked as PREALLOC. We're good to go here and can
6869 * just use the extent.
6870 *
6871 */
6872 if (!create) {
6873 len = min(len, em->len - (start - em->start));
6874 lockstart = start + len;
6875 goto unlock;
6876 }
6877
6878 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6879 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6880 em->block_start != EXTENT_MAP_HOLE)) {
6881 int type;
6882 int ret;
6883 u64 block_start, orig_start, orig_block_len, ram_bytes;
6884
6885 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6886 type = BTRFS_ORDERED_PREALLOC;
6887 else
6888 type = BTRFS_ORDERED_NOCOW;
6889 len = min(len, em->len - (start - em->start));
6890 block_start = em->block_start + (start - em->start);
6891
6892 if (can_nocow_extent(inode, start, &len, &orig_start,
6893 &orig_block_len, &ram_bytes) == 1) {
6894 if (type == BTRFS_ORDERED_PREALLOC) {
6895 free_extent_map(em);
6896 em = create_pinned_em(inode, start, len,
6897 orig_start,
6898 block_start, len,
6899 orig_block_len,
6900 ram_bytes, type);
6901 if (IS_ERR(em))
6902 goto unlock_err;
6903 }
6904
6905 ret = btrfs_add_ordered_extent_dio(inode, start,
6906 block_start, len, len, type);
6907 if (ret) {
6908 free_extent_map(em);
6909 goto unlock_err;
6910 }
6911 goto unlock;
6912 }
6913 }
6914
6915 /*
6916 * this will cow the extent, reset the len in case we changed
6917 * it above
6918 */
6919 len = bh_result->b_size;
6920 free_extent_map(em);
6921 em = btrfs_new_extent_direct(inode, start, len);
6922 if (IS_ERR(em)) {
6923 ret = PTR_ERR(em);
6924 goto unlock_err;
6925 }
6926 len = min(len, em->len - (start - em->start));
6927 unlock:
6928 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6929 inode->i_blkbits;
6930 bh_result->b_size = len;
6931 bh_result->b_bdev = em->bdev;
6932 set_buffer_mapped(bh_result);
6933 if (create) {
6934 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6935 set_buffer_new(bh_result);
6936
6937 /*
6938 * Need to update the i_size under the extent lock so buffered
6939 * readers will get the updated i_size when we unlock.
6940 */
6941 if (start + len > i_size_read(inode))
6942 i_size_write(inode, start + len);
6943
6944 spin_lock(&BTRFS_I(inode)->lock);
6945 BTRFS_I(inode)->outstanding_extents++;
6946 spin_unlock(&BTRFS_I(inode)->lock);
6947
6948 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6949 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6950 &cached_state, GFP_NOFS);
6951 BUG_ON(ret);
6952 }
6953
6954 /*
6955 * In the case of write we need to clear and unlock the entire range,
6956 * in the case of read we need to unlock only the end area that we
6957 * aren't using if there is any left over space.
6958 */
6959 if (lockstart < lockend) {
6960 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6961 lockend, unlock_bits, 1, 0,
6962 &cached_state, GFP_NOFS);
6963 } else {
6964 free_extent_state(cached_state);
6965 }
6966
6967 free_extent_map(em);
6968
6969 return 0;
6970
6971 unlock_err:
6972 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6973 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6974 return ret;
6975 }
6976
6977 static void btrfs_endio_direct_read(struct bio *bio, int err)
6978 {
6979 struct btrfs_dio_private *dip = bio->bi_private;
6980 struct bio_vec *bvec;
6981 struct inode *inode = dip->inode;
6982 struct btrfs_root *root = BTRFS_I(inode)->root;
6983 struct bio *dio_bio;
6984 u32 *csums = (u32 *)dip->csum;
6985 u64 start;
6986 int i;
6987
6988 start = dip->logical_offset;
6989 bio_for_each_segment_all(bvec, bio, i) {
6990 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6991 struct page *page = bvec->bv_page;
6992 char *kaddr;
6993 u32 csum = ~(u32)0;
6994 unsigned long flags;
6995
6996 local_irq_save(flags);
6997 kaddr = kmap_atomic(page);
6998 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6999 csum, bvec->bv_len);
7000 btrfs_csum_final(csum, (char *)&csum);
7001 kunmap_atomic(kaddr);
7002 local_irq_restore(flags);
7003
7004 flush_dcache_page(bvec->bv_page);
7005 if (csum != csums[i]) {
7006 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
7007 btrfs_ino(inode), start, csum,
7008 csums[i]);
7009 err = -EIO;
7010 }
7011 }
7012
7013 start += bvec->bv_len;
7014 }
7015
7016 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7017 dip->logical_offset + dip->bytes - 1);
7018 dio_bio = dip->dio_bio;
7019
7020 kfree(dip);
7021
7022 /* If we had a csum failure make sure to clear the uptodate flag */
7023 if (err)
7024 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7025 dio_end_io(dio_bio, err);
7026 bio_put(bio);
7027 }
7028
7029 static void btrfs_endio_direct_write(struct bio *bio, int err)
7030 {
7031 struct btrfs_dio_private *dip = bio->bi_private;
7032 struct inode *inode = dip->inode;
7033 struct btrfs_root *root = BTRFS_I(inode)->root;
7034 struct btrfs_ordered_extent *ordered = NULL;
7035 u64 ordered_offset = dip->logical_offset;
7036 u64 ordered_bytes = dip->bytes;
7037 struct bio *dio_bio;
7038 int ret;
7039
7040 if (err)
7041 goto out_done;
7042 again:
7043 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7044 &ordered_offset,
7045 ordered_bytes, !err);
7046 if (!ret)
7047 goto out_test;
7048
7049 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL, NULL);
7050 btrfs_queue_work(root->fs_info->endio_write_workers,
7051 &ordered->work);
7052 out_test:
7053 /*
7054 * our bio might span multiple ordered extents. If we haven't
7055 * completed the accounting for the whole dio, go back and try again
7056 */
7057 if (ordered_offset < dip->logical_offset + dip->bytes) {
7058 ordered_bytes = dip->logical_offset + dip->bytes -
7059 ordered_offset;
7060 ordered = NULL;
7061 goto again;
7062 }
7063 out_done:
7064 dio_bio = dip->dio_bio;
7065
7066 kfree(dip);
7067
7068 /* If we had an error make sure to clear the uptodate flag */
7069 if (err)
7070 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7071 dio_end_io(dio_bio, err);
7072 bio_put(bio);
7073 }
7074
7075 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7076 struct bio *bio, int mirror_num,
7077 unsigned long bio_flags, u64 offset)
7078 {
7079 int ret;
7080 struct btrfs_root *root = BTRFS_I(inode)->root;
7081 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7082 BUG_ON(ret); /* -ENOMEM */
7083 return 0;
7084 }
7085
7086 static void btrfs_end_dio_bio(struct bio *bio, int err)
7087 {
7088 struct btrfs_dio_private *dip = bio->bi_private;
7089
7090 if (err) {
7091 btrfs_err(BTRFS_I(dip->inode)->root->fs_info,
7092 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7093 btrfs_ino(dip->inode), bio->bi_rw,
7094 (unsigned long long)bio->bi_iter.bi_sector,
7095 bio->bi_iter.bi_size, err);
7096 dip->errors = 1;
7097
7098 /*
7099 * before atomic variable goto zero, we must make sure
7100 * dip->errors is perceived to be set.
7101 */
7102 smp_mb__before_atomic_dec();
7103 }
7104
7105 /* if there are more bios still pending for this dio, just exit */
7106 if (!atomic_dec_and_test(&dip->pending_bios))
7107 goto out;
7108
7109 if (dip->errors) {
7110 bio_io_error(dip->orig_bio);
7111 } else {
7112 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7113 bio_endio(dip->orig_bio, 0);
7114 }
7115 out:
7116 bio_put(bio);
7117 }
7118
7119 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7120 u64 first_sector, gfp_t gfp_flags)
7121 {
7122 int nr_vecs = bio_get_nr_vecs(bdev);
7123 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7124 }
7125
7126 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7127 int rw, u64 file_offset, int skip_sum,
7128 int async_submit)
7129 {
7130 struct btrfs_dio_private *dip = bio->bi_private;
7131 int write = rw & REQ_WRITE;
7132 struct btrfs_root *root = BTRFS_I(inode)->root;
7133 int ret;
7134
7135 if (async_submit)
7136 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7137
7138 bio_get(bio);
7139
7140 if (!write) {
7141 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7142 if (ret)
7143 goto err;
7144 }
7145
7146 if (skip_sum)
7147 goto map;
7148
7149 if (write && async_submit) {
7150 ret = btrfs_wq_submit_bio(root->fs_info,
7151 inode, rw, bio, 0, 0,
7152 file_offset,
7153 __btrfs_submit_bio_start_direct_io,
7154 __btrfs_submit_bio_done);
7155 goto err;
7156 } else if (write) {
7157 /*
7158 * If we aren't doing async submit, calculate the csum of the
7159 * bio now.
7160 */
7161 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7162 if (ret)
7163 goto err;
7164 } else if (!skip_sum) {
7165 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7166 file_offset);
7167 if (ret)
7168 goto err;
7169 }
7170
7171 map:
7172 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7173 err:
7174 bio_put(bio);
7175 return ret;
7176 }
7177
7178 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7179 int skip_sum)
7180 {
7181 struct inode *inode = dip->inode;
7182 struct btrfs_root *root = BTRFS_I(inode)->root;
7183 struct bio *bio;
7184 struct bio *orig_bio = dip->orig_bio;
7185 struct bio_vec *bvec = orig_bio->bi_io_vec;
7186 u64 start_sector = orig_bio->bi_iter.bi_sector;
7187 u64 file_offset = dip->logical_offset;
7188 u64 submit_len = 0;
7189 u64 map_length;
7190 int nr_pages = 0;
7191 int ret = 0;
7192 int async_submit = 0;
7193
7194 map_length = orig_bio->bi_iter.bi_size;
7195 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7196 &map_length, NULL, 0);
7197 if (ret) {
7198 bio_put(orig_bio);
7199 return -EIO;
7200 }
7201
7202 if (map_length >= orig_bio->bi_iter.bi_size) {
7203 bio = orig_bio;
7204 goto submit;
7205 }
7206
7207 /* async crcs make it difficult to collect full stripe writes. */
7208 if (btrfs_get_alloc_profile(root, 1) &
7209 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7210 async_submit = 0;
7211 else
7212 async_submit = 1;
7213
7214 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7215 if (!bio)
7216 return -ENOMEM;
7217 bio->bi_private = dip;
7218 bio->bi_end_io = btrfs_end_dio_bio;
7219 atomic_inc(&dip->pending_bios);
7220
7221 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7222 if (unlikely(map_length < submit_len + bvec->bv_len ||
7223 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7224 bvec->bv_offset) < bvec->bv_len)) {
7225 /*
7226 * inc the count before we submit the bio so
7227 * we know the end IO handler won't happen before
7228 * we inc the count. Otherwise, the dip might get freed
7229 * before we're done setting it up
7230 */
7231 atomic_inc(&dip->pending_bios);
7232 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7233 file_offset, skip_sum,
7234 async_submit);
7235 if (ret) {
7236 bio_put(bio);
7237 atomic_dec(&dip->pending_bios);
7238 goto out_err;
7239 }
7240
7241 start_sector += submit_len >> 9;
7242 file_offset += submit_len;
7243
7244 submit_len = 0;
7245 nr_pages = 0;
7246
7247 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7248 start_sector, GFP_NOFS);
7249 if (!bio)
7250 goto out_err;
7251 bio->bi_private = dip;
7252 bio->bi_end_io = btrfs_end_dio_bio;
7253
7254 map_length = orig_bio->bi_iter.bi_size;
7255 ret = btrfs_map_block(root->fs_info, rw,
7256 start_sector << 9,
7257 &map_length, NULL, 0);
7258 if (ret) {
7259 bio_put(bio);
7260 goto out_err;
7261 }
7262 } else {
7263 submit_len += bvec->bv_len;
7264 nr_pages++;
7265 bvec++;
7266 }
7267 }
7268
7269 submit:
7270 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7271 async_submit);
7272 if (!ret)
7273 return 0;
7274
7275 bio_put(bio);
7276 out_err:
7277 dip->errors = 1;
7278 /*
7279 * before atomic variable goto zero, we must
7280 * make sure dip->errors is perceived to be set.
7281 */
7282 smp_mb__before_atomic_dec();
7283 if (atomic_dec_and_test(&dip->pending_bios))
7284 bio_io_error(dip->orig_bio);
7285
7286 /* bio_end_io() will handle error, so we needn't return it */
7287 return 0;
7288 }
7289
7290 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7291 struct inode *inode, loff_t file_offset)
7292 {
7293 struct btrfs_root *root = BTRFS_I(inode)->root;
7294 struct btrfs_dio_private *dip;
7295 struct bio *io_bio;
7296 int skip_sum;
7297 int sum_len;
7298 int write = rw & REQ_WRITE;
7299 int ret = 0;
7300 u16 csum_size;
7301
7302 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7303
7304 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7305 if (!io_bio) {
7306 ret = -ENOMEM;
7307 goto free_ordered;
7308 }
7309
7310 if (!skip_sum && !write) {
7311 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7312 sum_len = dio_bio->bi_iter.bi_size >>
7313 inode->i_sb->s_blocksize_bits;
7314 sum_len *= csum_size;
7315 } else {
7316 sum_len = 0;
7317 }
7318
7319 dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7320 if (!dip) {
7321 ret = -ENOMEM;
7322 goto free_io_bio;
7323 }
7324
7325 dip->private = dio_bio->bi_private;
7326 dip->inode = inode;
7327 dip->logical_offset = file_offset;
7328 dip->bytes = dio_bio->bi_iter.bi_size;
7329 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7330 io_bio->bi_private = dip;
7331 dip->errors = 0;
7332 dip->orig_bio = io_bio;
7333 dip->dio_bio = dio_bio;
7334 atomic_set(&dip->pending_bios, 0);
7335
7336 if (write)
7337 io_bio->bi_end_io = btrfs_endio_direct_write;
7338 else
7339 io_bio->bi_end_io = btrfs_endio_direct_read;
7340
7341 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7342 if (!ret)
7343 return;
7344
7345 free_io_bio:
7346 bio_put(io_bio);
7347
7348 free_ordered:
7349 /*
7350 * If this is a write, we need to clean up the reserved space and kill
7351 * the ordered extent.
7352 */
7353 if (write) {
7354 struct btrfs_ordered_extent *ordered;
7355 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7356 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7357 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7358 btrfs_free_reserved_extent(root, ordered->start,
7359 ordered->disk_len);
7360 btrfs_put_ordered_extent(ordered);
7361 btrfs_put_ordered_extent(ordered);
7362 }
7363 bio_endio(dio_bio, ret);
7364 }
7365
7366 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7367 const struct iovec *iov, loff_t offset,
7368 unsigned long nr_segs)
7369 {
7370 int seg;
7371 int i;
7372 size_t size;
7373 unsigned long addr;
7374 unsigned blocksize_mask = root->sectorsize - 1;
7375 ssize_t retval = -EINVAL;
7376 loff_t end = offset;
7377
7378 if (offset & blocksize_mask)
7379 goto out;
7380
7381 /* Check the memory alignment. Blocks cannot straddle pages */
7382 for (seg = 0; seg < nr_segs; seg++) {
7383 addr = (unsigned long)iov[seg].iov_base;
7384 size = iov[seg].iov_len;
7385 end += size;
7386 if ((addr & blocksize_mask) || (size & blocksize_mask))
7387 goto out;
7388
7389 /* If this is a write we don't need to check anymore */
7390 if (rw & WRITE)
7391 continue;
7392
7393 /*
7394 * Check to make sure we don't have duplicate iov_base's in this
7395 * iovec, if so return EINVAL, otherwise we'll get csum errors
7396 * when reading back.
7397 */
7398 for (i = seg + 1; i < nr_segs; i++) {
7399 if (iov[seg].iov_base == iov[i].iov_base)
7400 goto out;
7401 }
7402 }
7403 retval = 0;
7404 out:
7405 return retval;
7406 }
7407
7408 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7409 const struct iovec *iov, loff_t offset,
7410 unsigned long nr_segs)
7411 {
7412 struct file *file = iocb->ki_filp;
7413 struct inode *inode = file->f_mapping->host;
7414 size_t count = 0;
7415 int flags = 0;
7416 bool wakeup = true;
7417 bool relock = false;
7418 ssize_t ret;
7419
7420 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7421 offset, nr_segs))
7422 return 0;
7423
7424 atomic_inc(&inode->i_dio_count);
7425 smp_mb__after_atomic_inc();
7426
7427 /*
7428 * The generic stuff only does filemap_write_and_wait_range, which
7429 * isn't enough if we've written compressed pages to this area, so
7430 * we need to flush the dirty pages again to make absolutely sure
7431 * that any outstanding dirty pages are on disk.
7432 */
7433 count = iov_length(iov, nr_segs);
7434 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7435 &BTRFS_I(inode)->runtime_flags))
7436 filemap_fdatawrite_range(inode->i_mapping, offset, count);
7437
7438 if (rw & WRITE) {
7439 /*
7440 * If the write DIO is beyond the EOF, we need update
7441 * the isize, but it is protected by i_mutex. So we can
7442 * not unlock the i_mutex at this case.
7443 */
7444 if (offset + count <= inode->i_size) {
7445 mutex_unlock(&inode->i_mutex);
7446 relock = true;
7447 }
7448 ret = btrfs_delalloc_reserve_space(inode, count);
7449 if (ret)
7450 goto out;
7451 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7452 &BTRFS_I(inode)->runtime_flags))) {
7453 inode_dio_done(inode);
7454 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7455 wakeup = false;
7456 }
7457
7458 ret = __blockdev_direct_IO(rw, iocb, inode,
7459 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7460 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7461 btrfs_submit_direct, flags);
7462 if (rw & WRITE) {
7463 if (ret < 0 && ret != -EIOCBQUEUED)
7464 btrfs_delalloc_release_space(inode, count);
7465 else if (ret >= 0 && (size_t)ret < count)
7466 btrfs_delalloc_release_space(inode,
7467 count - (size_t)ret);
7468 else
7469 btrfs_delalloc_release_metadata(inode, 0);
7470 }
7471 out:
7472 if (wakeup)
7473 inode_dio_done(inode);
7474 if (relock)
7475 mutex_lock(&inode->i_mutex);
7476
7477 return ret;
7478 }
7479
7480 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7481
7482 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7483 __u64 start, __u64 len)
7484 {
7485 int ret;
7486
7487 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7488 if (ret)
7489 return ret;
7490
7491 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7492 }
7493
7494 int btrfs_readpage(struct file *file, struct page *page)
7495 {
7496 struct extent_io_tree *tree;
7497 tree = &BTRFS_I(page->mapping->host)->io_tree;
7498 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7499 }
7500
7501 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7502 {
7503 struct extent_io_tree *tree;
7504
7505
7506 if (current->flags & PF_MEMALLOC) {
7507 redirty_page_for_writepage(wbc, page);
7508 unlock_page(page);
7509 return 0;
7510 }
7511 tree = &BTRFS_I(page->mapping->host)->io_tree;
7512 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7513 }
7514
7515 static int btrfs_writepages(struct address_space *mapping,
7516 struct writeback_control *wbc)
7517 {
7518 struct extent_io_tree *tree;
7519
7520 tree = &BTRFS_I(mapping->host)->io_tree;
7521 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7522 }
7523
7524 static int
7525 btrfs_readpages(struct file *file, struct address_space *mapping,
7526 struct list_head *pages, unsigned nr_pages)
7527 {
7528 struct extent_io_tree *tree;
7529 tree = &BTRFS_I(mapping->host)->io_tree;
7530 return extent_readpages(tree, mapping, pages, nr_pages,
7531 btrfs_get_extent);
7532 }
7533 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7534 {
7535 struct extent_io_tree *tree;
7536 struct extent_map_tree *map;
7537 int ret;
7538
7539 tree = &BTRFS_I(page->mapping->host)->io_tree;
7540 map = &BTRFS_I(page->mapping->host)->extent_tree;
7541 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7542 if (ret == 1) {
7543 ClearPagePrivate(page);
7544 set_page_private(page, 0);
7545 page_cache_release(page);
7546 }
7547 return ret;
7548 }
7549
7550 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7551 {
7552 if (PageWriteback(page) || PageDirty(page))
7553 return 0;
7554 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7555 }
7556
7557 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7558 unsigned int length)
7559 {
7560 struct inode *inode = page->mapping->host;
7561 struct extent_io_tree *tree;
7562 struct btrfs_ordered_extent *ordered;
7563 struct extent_state *cached_state = NULL;
7564 u64 page_start = page_offset(page);
7565 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7566 int inode_evicting = inode->i_state & I_FREEING;
7567
7568 /*
7569 * we have the page locked, so new writeback can't start,
7570 * and the dirty bit won't be cleared while we are here.
7571 *
7572 * Wait for IO on this page so that we can safely clear
7573 * the PagePrivate2 bit and do ordered accounting
7574 */
7575 wait_on_page_writeback(page);
7576
7577 tree = &BTRFS_I(inode)->io_tree;
7578 if (offset) {
7579 btrfs_releasepage(page, GFP_NOFS);
7580 return;
7581 }
7582
7583 if (!inode_evicting)
7584 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7585 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7586 if (ordered) {
7587 /*
7588 * IO on this page will never be started, so we need
7589 * to account for any ordered extents now
7590 */
7591 if (!inode_evicting)
7592 clear_extent_bit(tree, page_start, page_end,
7593 EXTENT_DIRTY | EXTENT_DELALLOC |
7594 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7595 EXTENT_DEFRAG, 1, 0, &cached_state,
7596 GFP_NOFS);
7597 /*
7598 * whoever cleared the private bit is responsible
7599 * for the finish_ordered_io
7600 */
7601 if (TestClearPagePrivate2(page)) {
7602 struct btrfs_ordered_inode_tree *tree;
7603 u64 new_len;
7604
7605 tree = &BTRFS_I(inode)->ordered_tree;
7606
7607 spin_lock_irq(&tree->lock);
7608 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7609 new_len = page_start - ordered->file_offset;
7610 if (new_len < ordered->truncated_len)
7611 ordered->truncated_len = new_len;
7612 spin_unlock_irq(&tree->lock);
7613
7614 if (btrfs_dec_test_ordered_pending(inode, &ordered,
7615 page_start,
7616 PAGE_CACHE_SIZE, 1))
7617 btrfs_finish_ordered_io(ordered);
7618 }
7619 btrfs_put_ordered_extent(ordered);
7620 if (!inode_evicting) {
7621 cached_state = NULL;
7622 lock_extent_bits(tree, page_start, page_end, 0,
7623 &cached_state);
7624 }
7625 }
7626
7627 if (!inode_evicting) {
7628 clear_extent_bit(tree, page_start, page_end,
7629 EXTENT_LOCKED | EXTENT_DIRTY |
7630 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7631 EXTENT_DEFRAG, 1, 1,
7632 &cached_state, GFP_NOFS);
7633
7634 __btrfs_releasepage(page, GFP_NOFS);
7635 }
7636
7637 ClearPageChecked(page);
7638 if (PagePrivate(page)) {
7639 ClearPagePrivate(page);
7640 set_page_private(page, 0);
7641 page_cache_release(page);
7642 }
7643 }
7644
7645 /*
7646 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7647 * called from a page fault handler when a page is first dirtied. Hence we must
7648 * be careful to check for EOF conditions here. We set the page up correctly
7649 * for a written page which means we get ENOSPC checking when writing into
7650 * holes and correct delalloc and unwritten extent mapping on filesystems that
7651 * support these features.
7652 *
7653 * We are not allowed to take the i_mutex here so we have to play games to
7654 * protect against truncate races as the page could now be beyond EOF. Because
7655 * vmtruncate() writes the inode size before removing pages, once we have the
7656 * page lock we can determine safely if the page is beyond EOF. If it is not
7657 * beyond EOF, then the page is guaranteed safe against truncation until we
7658 * unlock the page.
7659 */
7660 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7661 {
7662 struct page *page = vmf->page;
7663 struct inode *inode = file_inode(vma->vm_file);
7664 struct btrfs_root *root = BTRFS_I(inode)->root;
7665 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7666 struct btrfs_ordered_extent *ordered;
7667 struct extent_state *cached_state = NULL;
7668 char *kaddr;
7669 unsigned long zero_start;
7670 loff_t size;
7671 int ret;
7672 int reserved = 0;
7673 u64 page_start;
7674 u64 page_end;
7675
7676 sb_start_pagefault(inode->i_sb);
7677 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7678 if (!ret) {
7679 ret = file_update_time(vma->vm_file);
7680 reserved = 1;
7681 }
7682 if (ret) {
7683 if (ret == -ENOMEM)
7684 ret = VM_FAULT_OOM;
7685 else /* -ENOSPC, -EIO, etc */
7686 ret = VM_FAULT_SIGBUS;
7687 if (reserved)
7688 goto out;
7689 goto out_noreserve;
7690 }
7691
7692 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7693 again:
7694 lock_page(page);
7695 size = i_size_read(inode);
7696 page_start = page_offset(page);
7697 page_end = page_start + PAGE_CACHE_SIZE - 1;
7698
7699 if ((page->mapping != inode->i_mapping) ||
7700 (page_start >= size)) {
7701 /* page got truncated out from underneath us */
7702 goto out_unlock;
7703 }
7704 wait_on_page_writeback(page);
7705
7706 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7707 set_page_extent_mapped(page);
7708
7709 /*
7710 * we can't set the delalloc bits if there are pending ordered
7711 * extents. Drop our locks and wait for them to finish
7712 */
7713 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7714 if (ordered) {
7715 unlock_extent_cached(io_tree, page_start, page_end,
7716 &cached_state, GFP_NOFS);
7717 unlock_page(page);
7718 btrfs_start_ordered_extent(inode, ordered, 1);
7719 btrfs_put_ordered_extent(ordered);
7720 goto again;
7721 }
7722
7723 /*
7724 * XXX - page_mkwrite gets called every time the page is dirtied, even
7725 * if it was already dirty, so for space accounting reasons we need to
7726 * clear any delalloc bits for the range we are fixing to save. There
7727 * is probably a better way to do this, but for now keep consistent with
7728 * prepare_pages in the normal write path.
7729 */
7730 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7731 EXTENT_DIRTY | EXTENT_DELALLOC |
7732 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7733 0, 0, &cached_state, GFP_NOFS);
7734
7735 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7736 &cached_state);
7737 if (ret) {
7738 unlock_extent_cached(io_tree, page_start, page_end,
7739 &cached_state, GFP_NOFS);
7740 ret = VM_FAULT_SIGBUS;
7741 goto out_unlock;
7742 }
7743 ret = 0;
7744
7745 /* page is wholly or partially inside EOF */
7746 if (page_start + PAGE_CACHE_SIZE > size)
7747 zero_start = size & ~PAGE_CACHE_MASK;
7748 else
7749 zero_start = PAGE_CACHE_SIZE;
7750
7751 if (zero_start != PAGE_CACHE_SIZE) {
7752 kaddr = kmap(page);
7753 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7754 flush_dcache_page(page);
7755 kunmap(page);
7756 }
7757 ClearPageChecked(page);
7758 set_page_dirty(page);
7759 SetPageUptodate(page);
7760
7761 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7762 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7763 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7764
7765 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7766
7767 out_unlock:
7768 if (!ret) {
7769 sb_end_pagefault(inode->i_sb);
7770 return VM_FAULT_LOCKED;
7771 }
7772 unlock_page(page);
7773 out:
7774 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7775 out_noreserve:
7776 sb_end_pagefault(inode->i_sb);
7777 return ret;
7778 }
7779
7780 static int btrfs_truncate(struct inode *inode)
7781 {
7782 struct btrfs_root *root = BTRFS_I(inode)->root;
7783 struct btrfs_block_rsv *rsv;
7784 int ret = 0;
7785 int err = 0;
7786 struct btrfs_trans_handle *trans;
7787 u64 mask = root->sectorsize - 1;
7788 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7789
7790 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7791 (u64)-1);
7792 if (ret)
7793 return ret;
7794
7795 /*
7796 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7797 * 3 things going on here
7798 *
7799 * 1) We need to reserve space for our orphan item and the space to
7800 * delete our orphan item. Lord knows we don't want to have a dangling
7801 * orphan item because we didn't reserve space to remove it.
7802 *
7803 * 2) We need to reserve space to update our inode.
7804 *
7805 * 3) We need to have something to cache all the space that is going to
7806 * be free'd up by the truncate operation, but also have some slack
7807 * space reserved in case it uses space during the truncate (thank you
7808 * very much snapshotting).
7809 *
7810 * And we need these to all be seperate. The fact is we can use alot of
7811 * space doing the truncate, and we have no earthly idea how much space
7812 * we will use, so we need the truncate reservation to be seperate so it
7813 * doesn't end up using space reserved for updating the inode or
7814 * removing the orphan item. We also need to be able to stop the
7815 * transaction and start a new one, which means we need to be able to
7816 * update the inode several times, and we have no idea of knowing how
7817 * many times that will be, so we can't just reserve 1 item for the
7818 * entirety of the opration, so that has to be done seperately as well.
7819 * Then there is the orphan item, which does indeed need to be held on
7820 * to for the whole operation, and we need nobody to touch this reserved
7821 * space except the orphan code.
7822 *
7823 * So that leaves us with
7824 *
7825 * 1) root->orphan_block_rsv - for the orphan deletion.
7826 * 2) rsv - for the truncate reservation, which we will steal from the
7827 * transaction reservation.
7828 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7829 * updating the inode.
7830 */
7831 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7832 if (!rsv)
7833 return -ENOMEM;
7834 rsv->size = min_size;
7835 rsv->failfast = 1;
7836
7837 /*
7838 * 1 for the truncate slack space
7839 * 1 for updating the inode.
7840 */
7841 trans = btrfs_start_transaction(root, 2);
7842 if (IS_ERR(trans)) {
7843 err = PTR_ERR(trans);
7844 goto out;
7845 }
7846
7847 /* Migrate the slack space for the truncate to our reserve */
7848 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7849 min_size);
7850 BUG_ON(ret);
7851
7852 /*
7853 * setattr is responsible for setting the ordered_data_close flag,
7854 * but that is only tested during the last file release. That
7855 * could happen well after the next commit, leaving a great big
7856 * window where new writes may get lost if someone chooses to write
7857 * to this file after truncating to zero
7858 *
7859 * The inode doesn't have any dirty data here, and so if we commit
7860 * this is a noop. If someone immediately starts writing to the inode
7861 * it is very likely we'll catch some of their writes in this
7862 * transaction, and the commit will find this file on the ordered
7863 * data list with good things to send down.
7864 *
7865 * This is a best effort solution, there is still a window where
7866 * using truncate to replace the contents of the file will
7867 * end up with a zero length file after a crash.
7868 */
7869 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7870 &BTRFS_I(inode)->runtime_flags))
7871 btrfs_add_ordered_operation(trans, root, inode);
7872
7873 /*
7874 * So if we truncate and then write and fsync we normally would just
7875 * write the extents that changed, which is a problem if we need to
7876 * first truncate that entire inode. So set this flag so we write out
7877 * all of the extents in the inode to the sync log so we're completely
7878 * safe.
7879 */
7880 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7881 trans->block_rsv = rsv;
7882
7883 while (1) {
7884 ret = btrfs_truncate_inode_items(trans, root, inode,
7885 inode->i_size,
7886 BTRFS_EXTENT_DATA_KEY);
7887 if (ret != -ENOSPC) {
7888 err = ret;
7889 break;
7890 }
7891
7892 trans->block_rsv = &root->fs_info->trans_block_rsv;
7893 ret = btrfs_update_inode(trans, root, inode);
7894 if (ret) {
7895 err = ret;
7896 break;
7897 }
7898
7899 btrfs_end_transaction(trans, root);
7900 btrfs_btree_balance_dirty(root);
7901
7902 trans = btrfs_start_transaction(root, 2);
7903 if (IS_ERR(trans)) {
7904 ret = err = PTR_ERR(trans);
7905 trans = NULL;
7906 break;
7907 }
7908
7909 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7910 rsv, min_size);
7911 BUG_ON(ret); /* shouldn't happen */
7912 trans->block_rsv = rsv;
7913 }
7914
7915 if (ret == 0 && inode->i_nlink > 0) {
7916 trans->block_rsv = root->orphan_block_rsv;
7917 ret = btrfs_orphan_del(trans, inode);
7918 if (ret)
7919 err = ret;
7920 }
7921
7922 if (trans) {
7923 trans->block_rsv = &root->fs_info->trans_block_rsv;
7924 ret = btrfs_update_inode(trans, root, inode);
7925 if (ret && !err)
7926 err = ret;
7927
7928 ret = btrfs_end_transaction(trans, root);
7929 btrfs_btree_balance_dirty(root);
7930 }
7931
7932 out:
7933 btrfs_free_block_rsv(root, rsv);
7934
7935 if (ret && !err)
7936 err = ret;
7937
7938 return err;
7939 }
7940
7941 /*
7942 * create a new subvolume directory/inode (helper for the ioctl).
7943 */
7944 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7945 struct btrfs_root *new_root,
7946 struct btrfs_root *parent_root,
7947 u64 new_dirid)
7948 {
7949 struct inode *inode;
7950 int err;
7951 u64 index = 0;
7952
7953 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7954 new_dirid, new_dirid,
7955 S_IFDIR | (~current_umask() & S_IRWXUGO),
7956 &index);
7957 if (IS_ERR(inode))
7958 return PTR_ERR(inode);
7959 inode->i_op = &btrfs_dir_inode_operations;
7960 inode->i_fop = &btrfs_dir_file_operations;
7961
7962 set_nlink(inode, 1);
7963 btrfs_i_size_write(inode, 0);
7964
7965 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
7966 if (err)
7967 btrfs_err(new_root->fs_info,
7968 "error inheriting subvolume %llu properties: %d\n",
7969 new_root->root_key.objectid, err);
7970
7971 err = btrfs_update_inode(trans, new_root, inode);
7972
7973 iput(inode);
7974 return err;
7975 }
7976
7977 struct inode *btrfs_alloc_inode(struct super_block *sb)
7978 {
7979 struct btrfs_inode *ei;
7980 struct inode *inode;
7981
7982 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7983 if (!ei)
7984 return NULL;
7985
7986 ei->root = NULL;
7987 ei->generation = 0;
7988 ei->last_trans = 0;
7989 ei->last_sub_trans = 0;
7990 ei->logged_trans = 0;
7991 ei->delalloc_bytes = 0;
7992 ei->disk_i_size = 0;
7993 ei->flags = 0;
7994 ei->csum_bytes = 0;
7995 ei->index_cnt = (u64)-1;
7996 ei->dir_index = 0;
7997 ei->last_unlink_trans = 0;
7998 ei->last_log_commit = 0;
7999
8000 spin_lock_init(&ei->lock);
8001 ei->outstanding_extents = 0;
8002 ei->reserved_extents = 0;
8003
8004 ei->runtime_flags = 0;
8005 ei->force_compress = BTRFS_COMPRESS_NONE;
8006
8007 ei->delayed_node = NULL;
8008
8009 inode = &ei->vfs_inode;
8010 extent_map_tree_init(&ei->extent_tree);
8011 extent_io_tree_init(&ei->io_tree, &inode->i_data);
8012 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8013 ei->io_tree.track_uptodate = 1;
8014 ei->io_failure_tree.track_uptodate = 1;
8015 atomic_set(&ei->sync_writers, 0);
8016 mutex_init(&ei->log_mutex);
8017 mutex_init(&ei->delalloc_mutex);
8018 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8019 INIT_LIST_HEAD(&ei->delalloc_inodes);
8020 INIT_LIST_HEAD(&ei->ordered_operations);
8021 RB_CLEAR_NODE(&ei->rb_node);
8022
8023 return inode;
8024 }
8025
8026 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8027 void btrfs_test_destroy_inode(struct inode *inode)
8028 {
8029 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8030 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8031 }
8032 #endif
8033
8034 static void btrfs_i_callback(struct rcu_head *head)
8035 {
8036 struct inode *inode = container_of(head, struct inode, i_rcu);
8037 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8038 }
8039
8040 void btrfs_destroy_inode(struct inode *inode)
8041 {
8042 struct btrfs_ordered_extent *ordered;
8043 struct btrfs_root *root = BTRFS_I(inode)->root;
8044
8045 WARN_ON(!hlist_empty(&inode->i_dentry));
8046 WARN_ON(inode->i_data.nrpages);
8047 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8048 WARN_ON(BTRFS_I(inode)->reserved_extents);
8049 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8050 WARN_ON(BTRFS_I(inode)->csum_bytes);
8051
8052 /*
8053 * This can happen where we create an inode, but somebody else also
8054 * created the same inode and we need to destroy the one we already
8055 * created.
8056 */
8057 if (!root)
8058 goto free;
8059
8060 /*
8061 * Make sure we're properly removed from the ordered operation
8062 * lists.
8063 */
8064 smp_mb();
8065 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
8066 spin_lock(&root->fs_info->ordered_root_lock);
8067 list_del_init(&BTRFS_I(inode)->ordered_operations);
8068 spin_unlock(&root->fs_info->ordered_root_lock);
8069 }
8070
8071 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8072 &BTRFS_I(inode)->runtime_flags)) {
8073 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8074 btrfs_ino(inode));
8075 atomic_dec(&root->orphan_inodes);
8076 }
8077
8078 while (1) {
8079 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8080 if (!ordered)
8081 break;
8082 else {
8083 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8084 ordered->file_offset, ordered->len);
8085 btrfs_remove_ordered_extent(inode, ordered);
8086 btrfs_put_ordered_extent(ordered);
8087 btrfs_put_ordered_extent(ordered);
8088 }
8089 }
8090 inode_tree_del(inode);
8091 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8092 free:
8093 call_rcu(&inode->i_rcu, btrfs_i_callback);
8094 }
8095
8096 int btrfs_drop_inode(struct inode *inode)
8097 {
8098 struct btrfs_root *root = BTRFS_I(inode)->root;
8099
8100 if (root == NULL)
8101 return 1;
8102
8103 /* the snap/subvol tree is on deleting */
8104 if (btrfs_root_refs(&root->root_item) == 0)
8105 return 1;
8106 else
8107 return generic_drop_inode(inode);
8108 }
8109
8110 static void init_once(void *foo)
8111 {
8112 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8113
8114 inode_init_once(&ei->vfs_inode);
8115 }
8116
8117 void btrfs_destroy_cachep(void)
8118 {
8119 /*
8120 * Make sure all delayed rcu free inodes are flushed before we
8121 * destroy cache.
8122 */
8123 rcu_barrier();
8124 if (btrfs_inode_cachep)
8125 kmem_cache_destroy(btrfs_inode_cachep);
8126 if (btrfs_trans_handle_cachep)
8127 kmem_cache_destroy(btrfs_trans_handle_cachep);
8128 if (btrfs_transaction_cachep)
8129 kmem_cache_destroy(btrfs_transaction_cachep);
8130 if (btrfs_path_cachep)
8131 kmem_cache_destroy(btrfs_path_cachep);
8132 if (btrfs_free_space_cachep)
8133 kmem_cache_destroy(btrfs_free_space_cachep);
8134 if (btrfs_delalloc_work_cachep)
8135 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8136 }
8137
8138 int btrfs_init_cachep(void)
8139 {
8140 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8141 sizeof(struct btrfs_inode), 0,
8142 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8143 if (!btrfs_inode_cachep)
8144 goto fail;
8145
8146 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8147 sizeof(struct btrfs_trans_handle), 0,
8148 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8149 if (!btrfs_trans_handle_cachep)
8150 goto fail;
8151
8152 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8153 sizeof(struct btrfs_transaction), 0,
8154 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8155 if (!btrfs_transaction_cachep)
8156 goto fail;
8157
8158 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8159 sizeof(struct btrfs_path), 0,
8160 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8161 if (!btrfs_path_cachep)
8162 goto fail;
8163
8164 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8165 sizeof(struct btrfs_free_space), 0,
8166 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8167 if (!btrfs_free_space_cachep)
8168 goto fail;
8169
8170 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8171 sizeof(struct btrfs_delalloc_work), 0,
8172 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8173 NULL);
8174 if (!btrfs_delalloc_work_cachep)
8175 goto fail;
8176
8177 return 0;
8178 fail:
8179 btrfs_destroy_cachep();
8180 return -ENOMEM;
8181 }
8182
8183 static int btrfs_getattr(struct vfsmount *mnt,
8184 struct dentry *dentry, struct kstat *stat)
8185 {
8186 u64 delalloc_bytes;
8187 struct inode *inode = dentry->d_inode;
8188 u32 blocksize = inode->i_sb->s_blocksize;
8189
8190 generic_fillattr(inode, stat);
8191 stat->dev = BTRFS_I(inode)->root->anon_dev;
8192 stat->blksize = PAGE_CACHE_SIZE;
8193
8194 spin_lock(&BTRFS_I(inode)->lock);
8195 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8196 spin_unlock(&BTRFS_I(inode)->lock);
8197 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8198 ALIGN(delalloc_bytes, blocksize)) >> 9;
8199 return 0;
8200 }
8201
8202 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8203 struct inode *new_dir, struct dentry *new_dentry)
8204 {
8205 struct btrfs_trans_handle *trans;
8206 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8207 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8208 struct inode *new_inode = new_dentry->d_inode;
8209 struct inode *old_inode = old_dentry->d_inode;
8210 struct timespec ctime = CURRENT_TIME;
8211 u64 index = 0;
8212 u64 root_objectid;
8213 int ret;
8214 u64 old_ino = btrfs_ino(old_inode);
8215
8216 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8217 return -EPERM;
8218
8219 /* we only allow rename subvolume link between subvolumes */
8220 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8221 return -EXDEV;
8222
8223 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8224 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8225 return -ENOTEMPTY;
8226
8227 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8228 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8229 return -ENOTEMPTY;
8230
8231
8232 /* check for collisions, even if the name isn't there */
8233 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8234 new_dentry->d_name.name,
8235 new_dentry->d_name.len);
8236
8237 if (ret) {
8238 if (ret == -EEXIST) {
8239 /* we shouldn't get
8240 * eexist without a new_inode */
8241 if (WARN_ON(!new_inode)) {
8242 return ret;
8243 }
8244 } else {
8245 /* maybe -EOVERFLOW */
8246 return ret;
8247 }
8248 }
8249 ret = 0;
8250
8251 /*
8252 * we're using rename to replace one file with another.
8253 * and the replacement file is large. Start IO on it now so
8254 * we don't add too much work to the end of the transaction
8255 */
8256 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8257 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8258 filemap_flush(old_inode->i_mapping);
8259
8260 /* close the racy window with snapshot create/destroy ioctl */
8261 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8262 down_read(&root->fs_info->subvol_sem);
8263 /*
8264 * We want to reserve the absolute worst case amount of items. So if
8265 * both inodes are subvols and we need to unlink them then that would
8266 * require 4 item modifications, but if they are both normal inodes it
8267 * would require 5 item modifications, so we'll assume their normal
8268 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8269 * should cover the worst case number of items we'll modify.
8270 */
8271 trans = btrfs_start_transaction(root, 11);
8272 if (IS_ERR(trans)) {
8273 ret = PTR_ERR(trans);
8274 goto out_notrans;
8275 }
8276
8277 if (dest != root)
8278 btrfs_record_root_in_trans(trans, dest);
8279
8280 ret = btrfs_set_inode_index(new_dir, &index);
8281 if (ret)
8282 goto out_fail;
8283
8284 BTRFS_I(old_inode)->dir_index = 0ULL;
8285 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8286 /* force full log commit if subvolume involved. */
8287 root->fs_info->last_trans_log_full_commit = trans->transid;
8288 } else {
8289 ret = btrfs_insert_inode_ref(trans, dest,
8290 new_dentry->d_name.name,
8291 new_dentry->d_name.len,
8292 old_ino,
8293 btrfs_ino(new_dir), index);
8294 if (ret)
8295 goto out_fail;
8296 /*
8297 * this is an ugly little race, but the rename is required
8298 * to make sure that if we crash, the inode is either at the
8299 * old name or the new one. pinning the log transaction lets
8300 * us make sure we don't allow a log commit to come in after
8301 * we unlink the name but before we add the new name back in.
8302 */
8303 btrfs_pin_log_trans(root);
8304 }
8305 /*
8306 * make sure the inode gets flushed if it is replacing
8307 * something.
8308 */
8309 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8310 btrfs_add_ordered_operation(trans, root, old_inode);
8311
8312 inode_inc_iversion(old_dir);
8313 inode_inc_iversion(new_dir);
8314 inode_inc_iversion(old_inode);
8315 old_dir->i_ctime = old_dir->i_mtime = ctime;
8316 new_dir->i_ctime = new_dir->i_mtime = ctime;
8317 old_inode->i_ctime = ctime;
8318
8319 if (old_dentry->d_parent != new_dentry->d_parent)
8320 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8321
8322 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8323 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8324 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8325 old_dentry->d_name.name,
8326 old_dentry->d_name.len);
8327 } else {
8328 ret = __btrfs_unlink_inode(trans, root, old_dir,
8329 old_dentry->d_inode,
8330 old_dentry->d_name.name,
8331 old_dentry->d_name.len);
8332 if (!ret)
8333 ret = btrfs_update_inode(trans, root, old_inode);
8334 }
8335 if (ret) {
8336 btrfs_abort_transaction(trans, root, ret);
8337 goto out_fail;
8338 }
8339
8340 if (new_inode) {
8341 inode_inc_iversion(new_inode);
8342 new_inode->i_ctime = CURRENT_TIME;
8343 if (unlikely(btrfs_ino(new_inode) ==
8344 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8345 root_objectid = BTRFS_I(new_inode)->location.objectid;
8346 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8347 root_objectid,
8348 new_dentry->d_name.name,
8349 new_dentry->d_name.len);
8350 BUG_ON(new_inode->i_nlink == 0);
8351 } else {
8352 ret = btrfs_unlink_inode(trans, dest, new_dir,
8353 new_dentry->d_inode,
8354 new_dentry->d_name.name,
8355 new_dentry->d_name.len);
8356 }
8357 if (!ret && new_inode->i_nlink == 0)
8358 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8359 if (ret) {
8360 btrfs_abort_transaction(trans, root, ret);
8361 goto out_fail;
8362 }
8363 }
8364
8365 ret = btrfs_add_link(trans, new_dir, old_inode,
8366 new_dentry->d_name.name,
8367 new_dentry->d_name.len, 0, index);
8368 if (ret) {
8369 btrfs_abort_transaction(trans, root, ret);
8370 goto out_fail;
8371 }
8372
8373 if (old_inode->i_nlink == 1)
8374 BTRFS_I(old_inode)->dir_index = index;
8375
8376 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8377 struct dentry *parent = new_dentry->d_parent;
8378 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8379 btrfs_end_log_trans(root);
8380 }
8381 out_fail:
8382 btrfs_end_transaction(trans, root);
8383 out_notrans:
8384 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8385 up_read(&root->fs_info->subvol_sem);
8386
8387 return ret;
8388 }
8389
8390 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8391 {
8392 struct btrfs_delalloc_work *delalloc_work;
8393 struct inode *inode;
8394
8395 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8396 work);
8397 inode = delalloc_work->inode;
8398 if (delalloc_work->wait) {
8399 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8400 } else {
8401 filemap_flush(inode->i_mapping);
8402 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8403 &BTRFS_I(inode)->runtime_flags))
8404 filemap_flush(inode->i_mapping);
8405 }
8406
8407 if (delalloc_work->delay_iput)
8408 btrfs_add_delayed_iput(inode);
8409 else
8410 iput(inode);
8411 complete(&delalloc_work->completion);
8412 }
8413
8414 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8415 int wait, int delay_iput)
8416 {
8417 struct btrfs_delalloc_work *work;
8418
8419 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8420 if (!work)
8421 return NULL;
8422
8423 init_completion(&work->completion);
8424 INIT_LIST_HEAD(&work->list);
8425 work->inode = inode;
8426 work->wait = wait;
8427 work->delay_iput = delay_iput;
8428 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
8429
8430 return work;
8431 }
8432
8433 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8434 {
8435 wait_for_completion(&work->completion);
8436 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8437 }
8438
8439 /*
8440 * some fairly slow code that needs optimization. This walks the list
8441 * of all the inodes with pending delalloc and forces them to disk.
8442 */
8443 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
8444 int nr)
8445 {
8446 struct btrfs_inode *binode;
8447 struct inode *inode;
8448 struct btrfs_delalloc_work *work, *next;
8449 struct list_head works;
8450 struct list_head splice;
8451 int ret = 0;
8452
8453 INIT_LIST_HEAD(&works);
8454 INIT_LIST_HEAD(&splice);
8455
8456 mutex_lock(&root->delalloc_mutex);
8457 spin_lock(&root->delalloc_lock);
8458 list_splice_init(&root->delalloc_inodes, &splice);
8459 while (!list_empty(&splice)) {
8460 binode = list_entry(splice.next, struct btrfs_inode,
8461 delalloc_inodes);
8462
8463 list_move_tail(&binode->delalloc_inodes,
8464 &root->delalloc_inodes);
8465 inode = igrab(&binode->vfs_inode);
8466 if (!inode) {
8467 cond_resched_lock(&root->delalloc_lock);
8468 continue;
8469 }
8470 spin_unlock(&root->delalloc_lock);
8471
8472 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8473 if (unlikely(!work)) {
8474 if (delay_iput)
8475 btrfs_add_delayed_iput(inode);
8476 else
8477 iput(inode);
8478 ret = -ENOMEM;
8479 break;
8480 }
8481 list_add_tail(&work->list, &works);
8482 btrfs_queue_work(root->fs_info->flush_workers,
8483 &work->work);
8484 ret++;
8485 if (nr != -1 && ret >= nr)
8486 break;
8487 cond_resched();
8488 spin_lock(&root->delalloc_lock);
8489 }
8490 spin_unlock(&root->delalloc_lock);
8491
8492 list_for_each_entry_safe(work, next, &works, list) {
8493 list_del_init(&work->list);
8494 btrfs_wait_and_free_delalloc_work(work);
8495 }
8496
8497 if (!list_empty_careful(&splice)) {
8498 spin_lock(&root->delalloc_lock);
8499 list_splice_tail(&splice, &root->delalloc_inodes);
8500 spin_unlock(&root->delalloc_lock);
8501 }
8502 mutex_unlock(&root->delalloc_mutex);
8503 return ret;
8504 }
8505
8506 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8507 {
8508 int ret;
8509
8510 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
8511 return -EROFS;
8512
8513 ret = __start_delalloc_inodes(root, delay_iput, -1);
8514 if (ret > 0)
8515 ret = 0;
8516 /*
8517 * the filemap_flush will queue IO into the worker threads, but
8518 * we have to make sure the IO is actually started and that
8519 * ordered extents get created before we return
8520 */
8521 atomic_inc(&root->fs_info->async_submit_draining);
8522 while (atomic_read(&root->fs_info->nr_async_submits) ||
8523 atomic_read(&root->fs_info->async_delalloc_pages)) {
8524 wait_event(root->fs_info->async_submit_wait,
8525 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8526 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8527 }
8528 atomic_dec(&root->fs_info->async_submit_draining);
8529 return ret;
8530 }
8531
8532 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
8533 int nr)
8534 {
8535 struct btrfs_root *root;
8536 struct list_head splice;
8537 int ret;
8538
8539 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
8540 return -EROFS;
8541
8542 INIT_LIST_HEAD(&splice);
8543
8544 mutex_lock(&fs_info->delalloc_root_mutex);
8545 spin_lock(&fs_info->delalloc_root_lock);
8546 list_splice_init(&fs_info->delalloc_roots, &splice);
8547 while (!list_empty(&splice) && nr) {
8548 root = list_first_entry(&splice, struct btrfs_root,
8549 delalloc_root);
8550 root = btrfs_grab_fs_root(root);
8551 BUG_ON(!root);
8552 list_move_tail(&root->delalloc_root,
8553 &fs_info->delalloc_roots);
8554 spin_unlock(&fs_info->delalloc_root_lock);
8555
8556 ret = __start_delalloc_inodes(root, delay_iput, nr);
8557 btrfs_put_fs_root(root);
8558 if (ret < 0)
8559 goto out;
8560
8561 if (nr != -1) {
8562 nr -= ret;
8563 WARN_ON(nr < 0);
8564 }
8565 spin_lock(&fs_info->delalloc_root_lock);
8566 }
8567 spin_unlock(&fs_info->delalloc_root_lock);
8568
8569 ret = 0;
8570 atomic_inc(&fs_info->async_submit_draining);
8571 while (atomic_read(&fs_info->nr_async_submits) ||
8572 atomic_read(&fs_info->async_delalloc_pages)) {
8573 wait_event(fs_info->async_submit_wait,
8574 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8575 atomic_read(&fs_info->async_delalloc_pages) == 0));
8576 }
8577 atomic_dec(&fs_info->async_submit_draining);
8578 out:
8579 if (!list_empty_careful(&splice)) {
8580 spin_lock(&fs_info->delalloc_root_lock);
8581 list_splice_tail(&splice, &fs_info->delalloc_roots);
8582 spin_unlock(&fs_info->delalloc_root_lock);
8583 }
8584 mutex_unlock(&fs_info->delalloc_root_mutex);
8585 return ret;
8586 }
8587
8588 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8589 const char *symname)
8590 {
8591 struct btrfs_trans_handle *trans;
8592 struct btrfs_root *root = BTRFS_I(dir)->root;
8593 struct btrfs_path *path;
8594 struct btrfs_key key;
8595 struct inode *inode = NULL;
8596 int err;
8597 int drop_inode = 0;
8598 u64 objectid;
8599 u64 index = 0;
8600 int name_len;
8601 int datasize;
8602 unsigned long ptr;
8603 struct btrfs_file_extent_item *ei;
8604 struct extent_buffer *leaf;
8605
8606 name_len = strlen(symname);
8607 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8608 return -ENAMETOOLONG;
8609
8610 /*
8611 * 2 items for inode item and ref
8612 * 2 items for dir items
8613 * 1 item for xattr if selinux is on
8614 */
8615 trans = btrfs_start_transaction(root, 5);
8616 if (IS_ERR(trans))
8617 return PTR_ERR(trans);
8618
8619 err = btrfs_find_free_ino(root, &objectid);
8620 if (err)
8621 goto out_unlock;
8622
8623 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8624 dentry->d_name.len, btrfs_ino(dir), objectid,
8625 S_IFLNK|S_IRWXUGO, &index);
8626 if (IS_ERR(inode)) {
8627 err = PTR_ERR(inode);
8628 goto out_unlock;
8629 }
8630
8631 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8632 if (err) {
8633 drop_inode = 1;
8634 goto out_unlock;
8635 }
8636
8637 /*
8638 * If the active LSM wants to access the inode during
8639 * d_instantiate it needs these. Smack checks to see
8640 * if the filesystem supports xattrs by looking at the
8641 * ops vector.
8642 */
8643 inode->i_fop = &btrfs_file_operations;
8644 inode->i_op = &btrfs_file_inode_operations;
8645
8646 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8647 if (err)
8648 drop_inode = 1;
8649 else {
8650 inode->i_mapping->a_ops = &btrfs_aops;
8651 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8652 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8653 }
8654 if (drop_inode)
8655 goto out_unlock;
8656
8657 path = btrfs_alloc_path();
8658 if (!path) {
8659 err = -ENOMEM;
8660 drop_inode = 1;
8661 goto out_unlock;
8662 }
8663 key.objectid = btrfs_ino(inode);
8664 key.offset = 0;
8665 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8666 datasize = btrfs_file_extent_calc_inline_size(name_len);
8667 err = btrfs_insert_empty_item(trans, root, path, &key,
8668 datasize);
8669 if (err) {
8670 drop_inode = 1;
8671 btrfs_free_path(path);
8672 goto out_unlock;
8673 }
8674 leaf = path->nodes[0];
8675 ei = btrfs_item_ptr(leaf, path->slots[0],
8676 struct btrfs_file_extent_item);
8677 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8678 btrfs_set_file_extent_type(leaf, ei,
8679 BTRFS_FILE_EXTENT_INLINE);
8680 btrfs_set_file_extent_encryption(leaf, ei, 0);
8681 btrfs_set_file_extent_compression(leaf, ei, 0);
8682 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8683 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8684
8685 ptr = btrfs_file_extent_inline_start(ei);
8686 write_extent_buffer(leaf, symname, ptr, name_len);
8687 btrfs_mark_buffer_dirty(leaf);
8688 btrfs_free_path(path);
8689
8690 inode->i_op = &btrfs_symlink_inode_operations;
8691 inode->i_mapping->a_ops = &btrfs_symlink_aops;
8692 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8693 inode_set_bytes(inode, name_len);
8694 btrfs_i_size_write(inode, name_len);
8695 err = btrfs_update_inode(trans, root, inode);
8696 if (err)
8697 drop_inode = 1;
8698
8699 out_unlock:
8700 if (!err)
8701 d_instantiate(dentry, inode);
8702 btrfs_end_transaction(trans, root);
8703 if (drop_inode) {
8704 inode_dec_link_count(inode);
8705 iput(inode);
8706 }
8707 btrfs_btree_balance_dirty(root);
8708 return err;
8709 }
8710
8711 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8712 u64 start, u64 num_bytes, u64 min_size,
8713 loff_t actual_len, u64 *alloc_hint,
8714 struct btrfs_trans_handle *trans)
8715 {
8716 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8717 struct extent_map *em;
8718 struct btrfs_root *root = BTRFS_I(inode)->root;
8719 struct btrfs_key ins;
8720 u64 cur_offset = start;
8721 u64 i_size;
8722 u64 cur_bytes;
8723 int ret = 0;
8724 bool own_trans = true;
8725
8726 if (trans)
8727 own_trans = false;
8728 while (num_bytes > 0) {
8729 if (own_trans) {
8730 trans = btrfs_start_transaction(root, 3);
8731 if (IS_ERR(trans)) {
8732 ret = PTR_ERR(trans);
8733 break;
8734 }
8735 }
8736
8737 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8738 cur_bytes = max(cur_bytes, min_size);
8739 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8740 *alloc_hint, &ins, 1);
8741 if (ret) {
8742 if (own_trans)
8743 btrfs_end_transaction(trans, root);
8744 break;
8745 }
8746
8747 ret = insert_reserved_file_extent(trans, inode,
8748 cur_offset, ins.objectid,
8749 ins.offset, ins.offset,
8750 ins.offset, 0, 0, 0,
8751 BTRFS_FILE_EXTENT_PREALLOC);
8752 if (ret) {
8753 btrfs_free_reserved_extent(root, ins.objectid,
8754 ins.offset);
8755 btrfs_abort_transaction(trans, root, ret);
8756 if (own_trans)
8757 btrfs_end_transaction(trans, root);
8758 break;
8759 }
8760 btrfs_drop_extent_cache(inode, cur_offset,
8761 cur_offset + ins.offset -1, 0);
8762
8763 em = alloc_extent_map();
8764 if (!em) {
8765 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8766 &BTRFS_I(inode)->runtime_flags);
8767 goto next;
8768 }
8769
8770 em->start = cur_offset;
8771 em->orig_start = cur_offset;
8772 em->len = ins.offset;
8773 em->block_start = ins.objectid;
8774 em->block_len = ins.offset;
8775 em->orig_block_len = ins.offset;
8776 em->ram_bytes = ins.offset;
8777 em->bdev = root->fs_info->fs_devices->latest_bdev;
8778 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8779 em->generation = trans->transid;
8780
8781 while (1) {
8782 write_lock(&em_tree->lock);
8783 ret = add_extent_mapping(em_tree, em, 1);
8784 write_unlock(&em_tree->lock);
8785 if (ret != -EEXIST)
8786 break;
8787 btrfs_drop_extent_cache(inode, cur_offset,
8788 cur_offset + ins.offset - 1,
8789 0);
8790 }
8791 free_extent_map(em);
8792 next:
8793 num_bytes -= ins.offset;
8794 cur_offset += ins.offset;
8795 *alloc_hint = ins.objectid + ins.offset;
8796
8797 inode_inc_iversion(inode);
8798 inode->i_ctime = CURRENT_TIME;
8799 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8800 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8801 (actual_len > inode->i_size) &&
8802 (cur_offset > inode->i_size)) {
8803 if (cur_offset > actual_len)
8804 i_size = actual_len;
8805 else
8806 i_size = cur_offset;
8807 i_size_write(inode, i_size);
8808 btrfs_ordered_update_i_size(inode, i_size, NULL);
8809 }
8810
8811 ret = btrfs_update_inode(trans, root, inode);
8812
8813 if (ret) {
8814 btrfs_abort_transaction(trans, root, ret);
8815 if (own_trans)
8816 btrfs_end_transaction(trans, root);
8817 break;
8818 }
8819
8820 if (own_trans)
8821 btrfs_end_transaction(trans, root);
8822 }
8823 return ret;
8824 }
8825
8826 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8827 u64 start, u64 num_bytes, u64 min_size,
8828 loff_t actual_len, u64 *alloc_hint)
8829 {
8830 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8831 min_size, actual_len, alloc_hint,
8832 NULL);
8833 }
8834
8835 int btrfs_prealloc_file_range_trans(struct inode *inode,
8836 struct btrfs_trans_handle *trans, int mode,
8837 u64 start, u64 num_bytes, u64 min_size,
8838 loff_t actual_len, u64 *alloc_hint)
8839 {
8840 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8841 min_size, actual_len, alloc_hint, trans);
8842 }
8843
8844 static int btrfs_set_page_dirty(struct page *page)
8845 {
8846 return __set_page_dirty_nobuffers(page);
8847 }
8848
8849 static int btrfs_permission(struct inode *inode, int mask)
8850 {
8851 struct btrfs_root *root = BTRFS_I(inode)->root;
8852 umode_t mode = inode->i_mode;
8853
8854 if (mask & MAY_WRITE &&
8855 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8856 if (btrfs_root_readonly(root))
8857 return -EROFS;
8858 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8859 return -EACCES;
8860 }
8861 return generic_permission(inode, mask);
8862 }
8863
8864 static const struct inode_operations btrfs_dir_inode_operations = {
8865 .getattr = btrfs_getattr,
8866 .lookup = btrfs_lookup,
8867 .create = btrfs_create,
8868 .unlink = btrfs_unlink,
8869 .link = btrfs_link,
8870 .mkdir = btrfs_mkdir,
8871 .rmdir = btrfs_rmdir,
8872 .rename = btrfs_rename,
8873 .symlink = btrfs_symlink,
8874 .setattr = btrfs_setattr,
8875 .mknod = btrfs_mknod,
8876 .setxattr = btrfs_setxattr,
8877 .getxattr = btrfs_getxattr,
8878 .listxattr = btrfs_listxattr,
8879 .removexattr = btrfs_removexattr,
8880 .permission = btrfs_permission,
8881 .get_acl = btrfs_get_acl,
8882 .set_acl = btrfs_set_acl,
8883 .update_time = btrfs_update_time,
8884 };
8885 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8886 .lookup = btrfs_lookup,
8887 .permission = btrfs_permission,
8888 .get_acl = btrfs_get_acl,
8889 .set_acl = btrfs_set_acl,
8890 .update_time = btrfs_update_time,
8891 };
8892
8893 static const struct file_operations btrfs_dir_file_operations = {
8894 .llseek = generic_file_llseek,
8895 .read = generic_read_dir,
8896 .iterate = btrfs_real_readdir,
8897 .unlocked_ioctl = btrfs_ioctl,
8898 #ifdef CONFIG_COMPAT
8899 .compat_ioctl = btrfs_ioctl,
8900 #endif
8901 .release = btrfs_release_file,
8902 .fsync = btrfs_sync_file,
8903 };
8904
8905 static struct extent_io_ops btrfs_extent_io_ops = {
8906 .fill_delalloc = run_delalloc_range,
8907 .submit_bio_hook = btrfs_submit_bio_hook,
8908 .merge_bio_hook = btrfs_merge_bio_hook,
8909 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8910 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8911 .writepage_start_hook = btrfs_writepage_start_hook,
8912 .set_bit_hook = btrfs_set_bit_hook,
8913 .clear_bit_hook = btrfs_clear_bit_hook,
8914 .merge_extent_hook = btrfs_merge_extent_hook,
8915 .split_extent_hook = btrfs_split_extent_hook,
8916 };
8917
8918 /*
8919 * btrfs doesn't support the bmap operation because swapfiles
8920 * use bmap to make a mapping of extents in the file. They assume
8921 * these extents won't change over the life of the file and they
8922 * use the bmap result to do IO directly to the drive.
8923 *
8924 * the btrfs bmap call would return logical addresses that aren't
8925 * suitable for IO and they also will change frequently as COW
8926 * operations happen. So, swapfile + btrfs == corruption.
8927 *
8928 * For now we're avoiding this by dropping bmap.
8929 */
8930 static const struct address_space_operations btrfs_aops = {
8931 .readpage = btrfs_readpage,
8932 .writepage = btrfs_writepage,
8933 .writepages = btrfs_writepages,
8934 .readpages = btrfs_readpages,
8935 .direct_IO = btrfs_direct_IO,
8936 .invalidatepage = btrfs_invalidatepage,
8937 .releasepage = btrfs_releasepage,
8938 .set_page_dirty = btrfs_set_page_dirty,
8939 .error_remove_page = generic_error_remove_page,
8940 };
8941
8942 static const struct address_space_operations btrfs_symlink_aops = {
8943 .readpage = btrfs_readpage,
8944 .writepage = btrfs_writepage,
8945 .invalidatepage = btrfs_invalidatepage,
8946 .releasepage = btrfs_releasepage,
8947 };
8948
8949 static const struct inode_operations btrfs_file_inode_operations = {
8950 .getattr = btrfs_getattr,
8951 .setattr = btrfs_setattr,
8952 .setxattr = btrfs_setxattr,
8953 .getxattr = btrfs_getxattr,
8954 .listxattr = btrfs_listxattr,
8955 .removexattr = btrfs_removexattr,
8956 .permission = btrfs_permission,
8957 .fiemap = btrfs_fiemap,
8958 .get_acl = btrfs_get_acl,
8959 .set_acl = btrfs_set_acl,
8960 .update_time = btrfs_update_time,
8961 };
8962 static const struct inode_operations btrfs_special_inode_operations = {
8963 .getattr = btrfs_getattr,
8964 .setattr = btrfs_setattr,
8965 .permission = btrfs_permission,
8966 .setxattr = btrfs_setxattr,
8967 .getxattr = btrfs_getxattr,
8968 .listxattr = btrfs_listxattr,
8969 .removexattr = btrfs_removexattr,
8970 .get_acl = btrfs_get_acl,
8971 .set_acl = btrfs_set_acl,
8972 .update_time = btrfs_update_time,
8973 };
8974 static const struct inode_operations btrfs_symlink_inode_operations = {
8975 .readlink = generic_readlink,
8976 .follow_link = page_follow_link_light,
8977 .put_link = page_put_link,
8978 .getattr = btrfs_getattr,
8979 .setattr = btrfs_setattr,
8980 .permission = btrfs_permission,
8981 .setxattr = btrfs_setxattr,
8982 .getxattr = btrfs_getxattr,
8983 .listxattr = btrfs_listxattr,
8984 .removexattr = btrfs_removexattr,
8985 .update_time = btrfs_update_time,
8986 };
8987
8988 const struct dentry_operations btrfs_dentry_operations = {
8989 .d_delete = btrfs_dentry_delete,
8990 .d_release = btrfs_dentry_release,
8991 };
(deprecated) Btrfs devel tree